CN113810086A

CN113810086A - Channel information feedback method, communication device and storage medium

Info

Publication number: CN113810086A
Application number: CN202110181656.8A
Authority: CN
Inventors: 胡斌; 张公正; 徐晨; 王坚; 郭凯洋; 李榕
Original assignee: Huawei Technologies Co Ltd
Current assignee: Huawei Technologies Co Ltd
Priority date: 2020-06-12
Filing date: 2021-02-10
Publication date: 2021-12-17

Abstract

The application discloses a channel information feedback method, a communication device and a storage medium, and relates to the field of communication. In the application, the network device may send the CSI-RS to the terminal through the downlink channel. The terminal can measure and estimate a channel matrix of the downlink channel according to the received CSI-RS, determine channel negotiation information CNI corresponding to the downlink channel through the first neural network according to the channel matrix of the downlink channel, and send the CNI corresponding to the downlink channel to the network device. The network device may determine a modulation and coding strategy MCS corresponding to the downlink channel according to the received CNI corresponding to the downlink channel. Wherein the parameter of the first neural network is related to the historical MCS. The method and the device can reduce the quantization error of the terminal in quantizing the channel matrix, and can effectively improve the accuracy of channel information feedback. In addition, the information entropy of the CNI corresponding to the downlink channel determined by the terminal through the first neural network is smaller, and the efficiency of channel information feedback can be improved.

Description

Channel information feedback method, communication device and storage medium

Technical Field

The present invention relates to the field of communications, and in particular, to a channel information feedback method, a communication apparatus, and a storage medium.

Background

A large-scale antenna array (Massive Multiple-Input Multiple-Output, Massive MIMO) is a key technology for improving system capacity and spectrum utilization rate in a New air interface (New Radio, NR) of a fifth-generation mobile communication technology (5G). In a Massive MIMO-based communication system, a base station may enhance a signal to be transmitted to a User Equipment (UE) in a beamforming manner, so as to improve signal quality. For example, the base station may obtain channel information of a downlink channel, and determine an optimal beam and a corresponding Modulation and Coding Scheme (MCS) according to the channel information of the downlink channel.

In a Frequency-Division Duplex (FDD) communication system, an uplink channel and a downlink channel occupy different Frequency bands, and most of the uplink channel and the downlink channel are almost independent and do not satisfy channel reciprocity. Therefore, the channel information of the downlink channel can only be fed back to the base station by the UE. For example, in an FDD communication system, the base station and the UE may store one same codebook, respectively. The UE may quantize the channel matrix of the measured and estimated downlink channel to a codeword in the codebook for representation, and feed back a codebook index corresponding to the codeword to the base station. The base station can determine the corresponding code word from the stored codebook according to the received codebook index, thereby obtaining the channel information of the downlink channel.

However, in the above codebook-based channel information feedback method, when the UE quantizes the channel matrix of the downlink channel estimated by measurement to a codeword in the codebook for representation, there is inevitably a quantization error, and the determined codeword cannot be accurately matched with the downlink channel.

Disclosure of Invention

Embodiments of the present application provide a channel information feedback method, a communication device, and a storage medium, which can reduce quantization errors when a terminal quantizes a channel matrix, and can effectively improve accuracy of channel information feedback. In a first aspect, an embodiment of the present application provides a channel information feedback method, where the method includes: and the terminal receives the channel state information reference signal sent by the network equipment through a downlink channel. And the terminal measures and estimates a channel matrix of the downlink channel according to the channel state information reference signal. And the terminal determines channel negotiation information corresponding to the downlink channel through the first neural network according to the channel matrix of the downlink channel. And the terminal sends channel negotiation information corresponding to the downlink channel to the network equipment, and the channel negotiation information corresponding to the downlink channel is used for the network equipment to determine a modulation and coding strategy corresponding to the downlink channel.

Wherein the parameter of the first neural network is related to the historical modulation and coding strategy.

In the Channel Information feedback method, the terminal determines Channel Negotiation Information (CNI) corresponding to the downlink Channel through the first neural network, so that the quantization error of the terminal in quantizing the Channel matrix can be reduced, and the accuracy of Channel Information feedback can be effectively improved. In addition, because the parameters of the first neural network are related to the historical MCS, the information entropy of the CNI corresponding to the downlink channel determined by the terminal through the first neural network is smaller, the number of feedback bits can be saved, and the efficiency of channel information feedback is improved.

In one possible design, the parameter of the first neural network is related to a historical modulation and coding strategy, which may include: the parameters of the first neural network are parameters updated by the terminal according to historical decision benefits, and the historical decision benefits are obtained by the network equipment according to historical modulation and coding strategies.

In one possible design, the modulation and coding strategy corresponding to the downlink channel is determined by the network device through the second neural network according to the channel negotiation information corresponding to the downlink channel. The parameters of the second neural network are related to the historical modulation and coding strategy.

For example, the parameters of the second neural network are related to the historical modulation and coding strategy, which may include: and the parameters of the second neural network are parameters updated by the network equipment according to the historical decision benefits, and the historical decision benefits are obtained by the network equipment according to the historical modulation and coding strategy.

The network equipment determines the MCS corresponding to the downlink channel through the second neural network according to the CNI corresponding to the downlink channel, so that the possible deviation of the CNI and the MCS on probability distribution is smaller. In addition, the parameters of the second neural network are related to the historical MCS, so that the historical MCS can play a role of guiding the determination of the MCS of the downlink channel through the second neural network according to the CNI, and the possible deviation of the CNI and the MCS on the probability distribution can be further reduced.

In one possible design, the method further includes: and the terminal receives the decision benefit sent by the network equipment, wherein the decision benefit is obtained by the network equipment according to a modulation and coding strategy corresponding to the downlink channel. And the terminal updates the parameters of the first neural network according to the decision benefit.

For example, the terminal may update the parameters of the first neural network according to the first objective function according to the decision gain. The first objective function is related to the decision gain.

The terminal updates the parameters of the first neural network according to the decision benefits, so that the dynamic update of the mapping relation between the channel matrix and the CNI can be realized, and the information entropy of the CNI corresponding to the downlink channel determined by the terminal through the first neural network next time can be smaller.

In another possible design, the method further includes: the terminal receives the error of the parameter of the first layer hidden layer of the second neural network sent by the network equipment; and the error of the parameter of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameter of the second neural network according to the decision benefit, and the decision benefit is obtained by the network equipment according to a modulation and coding strategy corresponding to a downlink channel. And the terminal updates the parameters of the first neural network according to the errors of the parameters of the first hidden layer of the second neural network.

For example, the terminal may update the parameters of the first neural network according to a back propagation algorithm based on errors of the parameters of the first layer hidden layer of the second neural network. The error of the parameter of the first hidden layer of the second neural network may be an error before and after the parameter of the second neural network is updated by the network device according to the second objective function according to the decision gain, and the second objective function is related to the decision gain.

Compared with a mode that the terminal updates the parameters of the first neural network according to the decision benefits, the mode that the terminal updates the parameters of the first neural network according to the errors of the parameters of the first layer hidden layer of the second neural network is simpler.

Optionally, the decision benefit is any one of a decision code rate or throughput when the network device performs scheduling according to a modulation and coding strategy corresponding to the downlink channel; or, the network device combines any one of the decision code rate and the throughput when scheduling according to the modulation and coding strategy corresponding to the downlink channel with the evaluation value of the channel negotiation information corresponding to the downlink channel. The evaluation value of the channel negotiation information corresponding to the downlink channel by the network device is used for indicating the magnitude of the guiding effect of the channel negotiation information corresponding to the downlink channel on the modulation and coding strategy determined by the network device corresponding to the downlink channel.

In one possible design, before the terminal updates the parameters of the first neural network, the method further includes: and the terminal acquires the restored channel matrix according to the channel negotiation information corresponding to the downlink channel. And the terminal updates the parameters of the first neural network according to the third objective function according to the channel matrix of the downlink channel estimated by measurement and the restored channel matrix. And the third objective function is used for indicating the error between the channel matrix of the downlink channel estimated by the minimum measurement and the restored channel matrix.

For example, the error may be a minimum mean square error.

In a second aspect, an embodiment of the present application further provides a channel information feedback method, where the method includes: the network equipment sends the channel state information reference signal to the terminal through the downlink channel. The network equipment receives channel negotiation information corresponding to a downlink channel sent by a terminal according to a channel state information reference signal; the channel negotiation information corresponding to the downlink channel is determined by the first neural network according to the channel matrix of the downlink channel after the terminal measures and estimates the channel matrix of the downlink channel according to the channel state information reference signal. The parameters of the first neural network are related to a historical modulation and coding strategy. And the network equipment determines a modulation and coding strategy corresponding to the downlink channel according to the channel negotiation information corresponding to the downlink channel.

In one possible design, the determining, by the network device, the modulation and coding strategy corresponding to the downlink channel according to the channel negotiation information corresponding to the downlink channel includes: and the network equipment determines a modulation and coding strategy corresponding to the downlink channel through the second neural network according to the channel negotiation information corresponding to the downlink channel. The parameters of the second neural network are related to the historical modulation and coding strategy.

Optionally, the parameters of the second neural network are related to the historical modulation and coding strategy, which may include: and the parameters of the second neural network are parameters updated by the network equipment according to the historical decision benefits, and the historical decision benefits are obtained by the network equipment according to the historical modulation and coding strategy.

The parameters of the second neural network are related to the historical MCS, so that the historical MCS determined by the network equipment can play a guiding role in determining the MCS of the downlink channel through the second neural network according to the CNI, and the possible deviation of the CNI and the MCS on the probability distribution can be reduced.

In one possible design, the method further includes: and the network equipment acquires the decision benefit corresponding to the modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel. And the network equipment sends decision benefits to the terminal, and the decision benefits are used for updating the parameters of the first neural network by the terminal.

In one possible design, the method further includes: and the network equipment acquires the decision benefit corresponding to the modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel. And the network equipment updates the parameters of the second neural network according to the decision benefits.

For example, the network device may update parameters of the second neural network according to the second objective function based on the decision gain; the second objective function is related to the decision gain.

In another possible design, the method further includes: the network equipment obtains the error of the parameters of the first layer hidden layer of the second neural network before and after updating the parameters of the second neural network. The network equipment sends the error of the parameter of the first layer hidden layer of the second neural network to the terminal; and the error of the parameters of the first hidden layer of the second neural network is used for updating the parameters of the first neural network by the terminal.

In a third aspect, an embodiment of the present application provides a method for path information feedback, including: the method comprises the steps that a first communication device obtains path parameters of a first path, wherein the first path is a path for a second communication device to transmit data to the first communication device; the first communication device takes the path parameter as an input parameter and determines first negotiation information of the first path through a first network; the first network is a neural network updated according to task requirements; the first communication device sends the first negotiation information to the second communication device; the first negotiation information is used by the second communication device to obtain a transmission decision.

In the method provided by the third aspect, the first communication device inputs the path parameter of the first path into the neural network updated according to the task requirement to obtain the first negotiation information of the first path, and feeds back the first negotiation information to the second communication device. The path parameters are effectively compressed through the neural network to obtain the first negotiation information, and compared with the method of directly feeding back complete path parameters, the feedback overhead can be saved. Furthermore, the neural network is updated according to task requirements, first negotiation information matched with the task can be obtained according to actual task requirements, the high efficiency of path parameters fed back to the second communication device is guaranteed, and the accuracy of outgoing decisions is improved.

In one possible implementation, the first communication device receives a reference signal sent by the second communication device through the first path, and obtains the path parameter of the first path through measurement and estimation according to the reference signal.

In one possible implementation, the first negotiation information is used by the second communication device to obtain a transmission decision, the transmission decision including one or more of: modulation and coding strategy MCS, path selection, path weight combination, uplink scheduling and an authorization-free parameter set.

In a possible implementation, the input parameters of the first network further comprise a transmission decision of a previous round. And interacting the transmission decision information of the previous round aiming at the cooperative tasks of one or more entities, so as to accelerate decision convergence.

In a possible implementation, the first network is a neural network that is updated according to task requirements, and specifically includes: the first network is updated based on a decision benefit of the transmission decision, wherein the decision benefit is related to the task demand.

In one possible implementation, the first communication device obtains a decision gain indicated by the second communication device and updates the first network according to the decision gain.

And updating the first network according to the decision benefit, and when the decision benefit is determined according to the task requirement, enabling the first network to compress the path parameters of the first path more efficiently to match the actual task requirement, and further saving the signaling overhead of path information feedback.

In one possible implementation, the first path is a downlink channel between the network device and the terminal, and the path parameter of the first path is a channel matrix of the downlink channel.

It should be noted that, the first communication device mentioned in the third aspect and various possible implementations thereof may be a terminal, an entity a, an entity C, an entity D, an entity E, an entity F, or an entity H; the second communication device may be a network device, entity B, entity D, entity E, entity F or entity G; the first network may be a first neural network, a third neural network, a fourth neural network, a fifth neural network, a seventh neural network, an eighth neural network, a thirteenth neural network, a fourteenth neural network, or a seventeenth neural network.

In a fourth aspect, an embodiment of the present application provides a method for path information feedback, including:

the second communication device receives first negotiation information sent by the first communication device, wherein the first negotiation information is obtained by inputting the path parameters of the first path into a first network by the first communication device, and the first network is a neural network updated according to task requirements; the second communication device obtains a transmission decision according to the first negotiation information.

In the method provided in the fourth aspect, the first negotiation information used by the second communication device to obtain the transmission decision is obtained by effective compression of the first network, and compared with direct interaction path parameters, signaling overhead can be saved. Furthermore, the first network for obtaining the first negotiation information is updated according to task requirements, the first negotiation information matched with the task can be obtained according to the actual task, and accuracy of transmission decision is improved.

In a possible implementation, the second communication device sends reference information to the first communication device through the first path, and the reference information is used for the first communication device to measure and estimate to obtain the path parameter of the first path.

In one possible implementation, the obtaining, by the second communication device, a transmission decision according to the first negotiation information includes: the second communication device obtains a transmission decision through the second network according to the first negotiation information. The second network is a neural network that is updated according to task requirements. The transmission decision includes one or more of: modulation and coding strategy MCS, path selection, path weight combination, uplink scheduling and an authorization-free parameter set.

In one possible implementation, the input parameters of the second network include the first negotiation information and the transmission decision of the previous round.

In a possible implementation, the second network is a neural network that is updated according to task requirements, and specifically includes: the second network is updated based on a decision benefit of the transmission decision, wherein the decision benefit is related to the task demand.

In one possible implementation, the decision benefit is obtained according to one or more of a code rate, an evaluation value of the first negotiation information, throughput, delay, power consumption, routing hop count, and channel capacity.

In one possible implementation, the first path is a downlink channel between the network device and the terminal, the first negotiation information is negotiation information obtained by a channel matrix of the downlink channel passing through the first network, and the transmission decision is a modulation and coding strategy MCS.

It should be noted that the first communication device mentioned in the fourth aspect and various possible implementations thereof may be a terminal, an entity a, an entity C, an entity D, an entity E, an entity F, or an entity H; the second communication device may be a network device, entity B, entity D, entity E, entity F or entity G; the first network may be a first neural network, a third neural network, a fourth neural network, a fifth neural network, a seventh neural network, an eighth neural network, a thirteenth neural network, a fourteenth neural network, or a seventeenth neural network; the second network may be a second neural network, a sixth neural network, an eleventh neural network, a twelfth neural network, a fifteenth neural network, a sixteenth neural network, or an eighteenth neural network.

In a fifth aspect, an embodiment of the present application provides a communication apparatus, including: and the receiving unit is used for receiving the channel state information reference signal sent by the network equipment through a downlink channel. And the measuring unit is used for measuring and estimating a channel matrix of the downlink channel according to the channel state information reference signal. The determining unit is used for determining channel negotiation information corresponding to the downlink channel through the first neural network according to the channel matrix of the downlink channel; the parameters of the first neural network are related to a historical modulation and coding strategy. And the sending unit is used for sending channel negotiation information corresponding to the downlink channel to the network equipment, and the channel negotiation information corresponding to the downlink channel is used for the network equipment to determine a modulation and coding strategy corresponding to the downlink channel.

In one possible design, the parameter of the first neural network is related to a historical modulation and coding strategy, including: the parameter of the first neural network is the parameter updated by the determining unit according to the historical decision making benefit, and the historical decision making benefit is obtained by the network equipment according to the historical modulation and coding strategy.

In one possible design, the modulation and coding strategy corresponding to the downlink channel is determined by the network device through the second neural network according to the channel negotiation information corresponding to the downlink channel; the parameters of the second neural network are related to the historical modulation and coding strategy.

In one possible design, the parameters of the second neural network are related to a historical modulation and coding strategy, including: and the parameters of the second neural network are parameters updated by the network equipment according to the historical decision benefits, and the historical decision benefits are obtained by the network equipment according to the historical modulation and coding strategy.

In a possible design, the receiving unit is further configured to receive a decision benefit sent by the network device, where the decision benefit is obtained by the network device according to a modulation and coding strategy corresponding to a downlink channel; the determining unit is further configured to update the parameter of the first neural network according to the decision gain.

Optionally, the determining unit is specifically configured to update a parameter of the first neural network according to the decision gain and the first objective function; the first objective function is related to the decision gain.

In another possible design, the receiving unit is further configured to receive an error of a parameter of a first hidden layer of a second neural network sent by the network device; and the error of the parameter of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameter of the second neural network according to the decision benefit, and the decision benefit is obtained by the network equipment according to a modulation and coding strategy corresponding to a downlink channel. The determining unit is further configured to update the parameter of the first neural network according to an error of the parameter of the first hidden layer of the second neural network.

Optionally, the determining unit is specifically configured to update the parameter of the first neural network according to a back propagation algorithm according to the error of the parameter of the first hidden layer of the second neural network. And the error of the parameters of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameters of the second neural network according to the second objective function according to the decision gain, and the second objective function is related to the decision gain.

In a possible design, the determining unit is further configured to obtain a restored channel matrix according to channel negotiation information corresponding to the downlink channel; and updating the parameters of the first neural network according to the third objective function according to the channel matrix of the downlink channel estimated by measurement and the restored channel matrix. And the third objective function is used for indicating the error between the channel matrix of the downlink channel estimated by the minimum measurement and the restored channel matrix.

The communication apparatus according to the fifth aspect may be applied to a terminal. In a sixth aspect, an embodiment of the present application further provides a communication apparatus, including: a processor configured to execute computer instructions stored in a memory, the computer instructions, when executed, causing the apparatus to perform a method as set forth in the first aspect, the possible design of the first aspect, the third aspect, or any of the various possible implementations of the third aspect.

In a seventh aspect, an embodiment of the present application further provides a communication apparatus, including: a processor and interface circuitry, the processor being configured to communicate with other apparatus via the interface circuitry and to perform the method of the first aspect, possible designs of the first aspect, the third aspect, or any of its various possible implementations.

In an eighth aspect, an embodiment of the present application provides a communication apparatus, including: and the sending unit is used for sending the channel state information reference signal to the terminal through the downlink channel. A receiving unit, configured to receive channel negotiation information corresponding to a downlink channel sent by a terminal according to a channel state information reference signal; the channel negotiation information corresponding to the downlink channel is determined by the first neural network according to the channel matrix of the downlink channel after the terminal measures and estimates the channel matrix of the downlink channel according to the channel state information reference signal; the parameters of the first neural network are related to a historical modulation and coding strategy. And the determining unit is used for determining a modulation and coding strategy corresponding to the downlink channel according to the channel negotiation information corresponding to the downlink channel.

Optionally, the parameter of the first neural network is related to a historical modulation and coding strategy, including: the parameters of the first neural network are updated by the terminal according to the historical decision benefits, and the historical decision benefits are obtained by the determining unit according to the historical modulation and coding strategy.

In one possible design, the determining unit is specifically configured to determine, according to channel negotiation information corresponding to a downlink channel, a modulation and coding strategy corresponding to the downlink channel through the second neural network; the parameters of the second neural network are related to the historical modulation and coding strategy.

Optionally, the parameters of the second neural network are related to a historical modulation and coding strategy, including: and the parameters of the second neural network are updated by the determining unit according to the historical decision benefits, and the historical decision benefits are obtained by the determining unit according to the historical modulation and coding strategy.

In a possible design, the determining unit is further configured to obtain a decision gain corresponding to the modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel. The sending unit is also used for sending the decision benefit to the terminal; the decision gain is used for updating the parameters of the first neural network by the terminal.

In a possible design, the determining unit is further configured to obtain a decision gain corresponding to the modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel; and updating the parameters of the second neural network according to the decision gain.

Optionally, the determining unit is specifically configured to update the parameter of the second neural network according to the decision gain and the second objective function; the second objective function is related to the decision gain.

In another possible design, the determining unit is further configured to obtain an error of the parameter of the first hidden layer of the second neural network before and after the updating of the parameter of the second neural network. The sending unit is further used for sending the error of the parameter of the first layer hidden layer of the second neural network to the terminal; and the error of the parameters of the first hidden layer of the second neural network is used for updating the parameters of the first neural network by the terminal.

Optionally, the decision benefit is any one of a decision code rate or throughput when the communication device performs scheduling according to a modulation and coding strategy corresponding to the downlink channel; or, the communication device may combine any one of a decision code rate and a throughput when scheduling according to a modulation and coding strategy corresponding to the downlink channel with an evaluation value of channel negotiation information corresponding to the downlink channel. The evaluation value of the channel negotiation information corresponding to the downlink channel by the communication device is used for indicating the magnitude of the guiding effect of the channel negotiation information corresponding to the downlink channel on the modulation and coding strategy determined by the communication device corresponding to the downlink channel.

The communication apparatus according to the eighth aspect may be applied to a network device.

In a ninth aspect, an embodiment of the present application further provides a communication apparatus, including: a processor for executing computer instructions stored in the memory, which when executed, cause the apparatus to perform a method as set forth in the second aspect, the possible designs of the second aspect, the fourth aspect, or any of the various possible implementations of the fourth aspect.

In a tenth aspect, an embodiment of the present application further provides a communication apparatus, including: a processor and interface circuitry, the processor being configured to communicate with other devices via the interface circuitry and to perform the method of the second aspect, possible designs of the second aspect, the fourth aspect, or any of its various possible implementations.

In an eleventh aspect, an embodiment of the present application further provides a communication apparatus, including: a transceiving unit and a processing unit. The transceiving unit may be used for transceiving information or for communicating with other network elements. The processing unit may be adapted to process data. Such as: the apparatus may implement the method according to the first aspect, the second aspect, the third aspect or the fourth aspect by the transceiver and the processing unit.

In a twelfth aspect, an embodiment of the present application further provides a computer-readable storage medium, including: computer software instructions; the method according to the first aspect is performed when the computer software instructions are run on a processor; alternatively, the method of the second aspect is performed; alternatively, the method of the third aspect is performed; alternatively, the method of the fourth aspect is performed.

The computer software instructions may for example cause the terminal to perform the method according to the first aspect when run in the terminal or in a chip built into the terminal. Alternatively, the computer software instructions, when executed in a network device or a chip built into said network device, cause the network device to perform the method according to the second aspect. Alternatively, the computer software instructions, when executed in a network device or a chip built into said network device, cause the network device to perform the method according to the third aspect. Alternatively, the computer software instructions, when executed in a network device or a chip built into said network device, cause the network device to perform the method according to the fourth aspect.

In a thirteenth aspect, the present application further provides a computer program product, which when executed, can implement the method as described in the first aspect or any of the possible designs of the first aspect.

In a fourteenth aspect, the present application further provides a computer program product, which when executed, can implement the method according to the second aspect or any of the possible designs of the second aspect.

In a fourteenth aspect, the present application further provides a computer program product, which when executed, can implement the method as described in the third aspect or any of the possible designs of the third aspect.

In a fifteenth aspect, the present application further provides a computer program product, which when executed, can implement the method according to the fourth aspect or any of the possible designs of the fourth aspect.

In a sixteenth aspect, an embodiment of the present application further provides a chip system, where the chip system is applied to a terminal; the chip system includes one or more interface circuits and one or more processors; the interface circuit and the processor are interconnected through a line; the processor receiving and executing computer instructions from the memory of the electronic device via the interface circuit to implement the method as described in the first aspect or any one of the possible designs of the first aspect; or implementing a method as described in the third aspect or any of its possible implementations.

In a seventeenth aspect, an embodiment of the present application further provides a chip system, where the chip system is applied to a network device; the chip system includes one or more interface circuits and one or more processors; the interface circuit and the processor are interconnected through a line; the processor receiving and executing computer instructions from the memory of the electronic device via the interface circuit to implement the method as described in the second aspect or any of its possible designs; or implementing a method as described in the fourth aspect or any of its possible implementations.

It should be understood that the advantageous effects achieved by the fifth aspect to the seventeenth aspect provided above can be referred to the advantageous effects of the first aspect and any possible design manner thereof, or the advantageous effects of the second aspect and any possible design manner thereof, or the advantageous effects of the third aspect and any possible design manner thereof, or the advantageous effects of the fourth aspect and any possible design manner thereof, and thus, the detailed description thereof is omitted here.

Drawings

Fig. 1 shows a schematic diagram of a communication system based on a large-scale antenna array technology;

FIG. 2 shows a diagram of channel information quantization;

fig. 3 is a schematic diagram illustrating a communication system according to an embodiment of the present application;

fig. 4 is a schematic diagram illustrating a terminal according to an embodiment of the present application;

fig. 5 is an interaction diagram illustrating a channel information feedback method provided in an embodiment of the present application;

fig. 6 shows another interaction diagram of a channel information feedback method provided in an embodiment of the present application;

fig. 7 shows another interaction diagram of a channel information feedback method provided in an embodiment of the present application;

fig. 8 shows another interaction diagram of a channel information feedback method provided in an embodiment of the present application;

fig. 9a is a schematic view illustrating an application scenario of a path information feedback method according to an embodiment of the present application;

fig. 9b shows a further interaction diagram of the path information feedback method provided in the embodiment of the present application;

fig. 10a is a schematic diagram illustrating another application scenario of a path information feedback method provided in an embodiment of the present application;

fig. 10b shows a further interaction diagram of a path information feedback method provided in the embodiment of the present application;

fig. 11 shows another interaction diagram of a path information feedback method provided in an embodiment of the present application;

fig. 12 is a schematic diagram illustrating still another interaction of a path information feedback method provided in an embodiment of the present application;

fig. 13 is a schematic structural diagram of a communication device provided in an embodiment of the present application;

fig. 14 is a schematic structural diagram of a communication device provided in an embodiment of the present application;

fig. 15 shows a further schematic structural diagram of a communication device provided in an embodiment of the present application.

Detailed Description

Massive MIMO is a key technology for improving system capacity and spectrum utilization in 5G NR. The frequency spectrum efficiency can be greatly improved by configuring a large number of antennas at the base station. For example, as the number of antennas increases, channels between multiple users tend to be orthogonal, so that multiple users within the coverage area of the base station can communicate with the base station at the same time on the same time-frequency resource by using the spatial degree of freedom provided by massive MIMO, thereby improving the multiplexing capability of the spectrum resource among the multiple users, and greatly improving the spectrum efficiency without increasing the density and bandwidth of the base station.

In a Massive MIMO-based communication system, a base station may employ a beamforming method to enhance a signal to be transmitted to a terminal (e.g., a User Equipment (UE)) to improve signal quality. For example, the base station may modulate the signal into a narrower beam, radiating centrally in a smaller spatial area, thus making the energy efficiency on the radio transmission link between the base station and the terminal higher.

In order to perform optimal beamforming, the base station needs to acquire accurate channel information of a continuously changing downlink channel. Then, the base station may determine the optimal beam and the corresponding MCS according to the acquired channel information of the downlink channel. This process is described below in conjunction with fig. 1:

fig. 1 shows a schematic diagram of a communication system based on Massive antenna array Massive MIMO technology.

As shown in fig. 1, in a communication system based on the Massive MIMO technology, a base station is configured with a large number of antennas (e.g., antenna 1 to antenna M shown in fig. 1, M may be 100 or a number greater than 100). The terminal may be configured with one or more antennas (one is shown by way of example). A base station may communicate with a plurality of terminals (e.g., terminal 1 to terminal K as shown in fig. 1). Taking terminal 1 as an example, the downlink channel between the base station and terminal 1 is downlink channel 1. The base station may schedule data to be transmitted to the terminal 1, and transmit the data to the terminal 1 through the downlink channel 1 in a signal beam manner. In forming a signal beam, in order to perform optimal beamforming, the base station may acquire channel information 1 of a downlink channel 1, and determine an optimal beam and a corresponding MCS according to the channel information 1. Then, the base station may modulate and encode the data to be transmitted to the terminal according to the determined MCS. The communications between the base station and terminals 2 to K are the same as terminal 1, and are not described herein again.

However, for an FDD communication system based on the Massive MIMO technology, an uplink channel and a downlink channel occupy different frequency bands, and most of the uplink channel and the downlink channel are almost independent and do not satisfy channel reciprocity. Therefore, the channel information of the downlink channel can only be fed back to the base station by the terminal. For example, in an FDD communication system, a specific implementation manner for a terminal to feed back channel information of a downlink channel to a base station may be as follows:

1) the base station sends a pilot signal to the terminal through a downlink channel.

2) And the terminal measures and estimates a channel matrix of a downlink channel according to the received pilot signal.

3) And the terminal feeds back the channel matrix of the downlink channel estimated by measurement as the channel information of the downlink channel to the base station.

In the channel information feedback process, the parameters of the channel matrix of the downlink channel fed back to the base station by the terminal are in direct proportion to the number of transmitting antennas in the base station. In an FDD communication system based on the Massive MIMO technology, a base station is configured with a large number of antennas, so that a terminal includes a large number of channel matrix parameters when feeding back a channel matrix to the base station, which may cause an increase in feedback overhead of the FDD communication system and a decrease in accuracy and timeliness of fed-back channel information.

In order to solve the problems of increased feedback overhead of an FDD communication system and reduced accuracy and timeliness of fed-back channel information, codebook-based limited channel information feedback is adopted in the current 5G NR standard. The basic principle is as follows: the channel matrix is precoded and the corresponding precoding variables (e.g. beamforming vectors or matrices) are put as codewords into a codebook, which can be stored in the terminal and the base station, respectively. In practical application, the terminal may perform measurement and estimation on the downlink channel to obtain a channel matrix of the downlink channel. Then, the terminal may determine, according to the measured and estimated channel matrix and the stored codebook, a codebook index of a precoding variable that best matches the channel matrix, such as: the codebook index may be referred to as a Precoding Matrix Indicator (PMI). Meanwhile, the terminal may also calculate the Channel Quality of the downlink Channel according to the PMI to obtain a Channel Quality Indicator (CQI), and the CQI may be used to indicate the Quality of the Channel Quality of the downlink Channel. After obtaining the PMI and the CQI, the terminal may send Channel State Information (CSI) including the PMI and the CQI to the base station. And the base station can acquire the channel information of the downlink channel according to the received CSI. For example, the base station may find a corresponding precoding vector (codeword) from a stored codebook according to a PMI (i.e., the codebook index) in the received CSI, and further obtain a channel matrix of the downlink channel. Further, the base station may determine a corresponding MCS according to the channel matrix and the CQI of the downlink channel.

In the above finite channel information feedback method based on the codebook, the codebook is given in advance, and the code words in the codebook can only represent the precoding variables corresponding to the finite channel matrix. In an FDD communication system based on the Massive MIMO technology, due to the large number of antennas of the base station, there are many possible states of measuring the estimated channel matrix of the downlink channel. Therefore, when the terminal determines the codebook index of the precoding variable that most closely matches the channel matrix according to the measured and estimated channel matrix and the stored codebook, the measured and estimated channel matrix is quantized (e.g., scalar quantization) first, so that the codeword in the codebook can represent the precoding variable that most closely matches the channel matrix.

Taking CQI as an example: as described above, the CQI is calculated from the PMI of the precoding variable corresponding to the quantized channel matrix, and is related to the quantized channel matrix. Assuming that the maximum bit number of the CQI allowed to be fed back is B, the terminal can only feed back to the base station L ═ 2B different precoding variables (codewords) at most. That is, the codebook can only contain L ═ 2B different code words, and the terminal can only feed back the channel information of L ═ 2B states of the downlink channel to the base station at most through L ═ 2B different code words. This means that the channel space of the downlink channel needs to be quantized and divided into L non-overlapping regions at the terminal, and then the channel information of the corresponding region is represented by one codeword according to the aforementioned rule. Fig. 2 shows a diagram of channel information quantization.

As shown in fig. 2 (a), the probability distribution of the channel information is a continuous state that changes dynamically, and when the probability distribution is quantized into five non-overlapping regions, i.e., L0, L1, L2, L3, and L4, shown in fig. 2 (b), and the channel information of each region is represented by one codeword, the codeword corresponding to each region does not accurately reflect the channel information distribution under dynamic change.

Therefore, in the above codebook-based limited channel information feedback method, when the terminal quantizes the channel matrix (i.e., quantizes the channel information), there is a quantization precision loss, which inevitably causes a quantization error, and the determined precoding variable (codeword in the codebook) cannot be accurately matched with the downlink channel, so that the CSI fed back to the base station by the terminal cannot accurately reflect the channel information of the downlink channel under dynamic change.

In the embodiment of the present application, based on the idea of Reinforcement Learning (Reinforcement Learning), the relationship between the channel matrix and the CSI is learned through the neural network, so that feedback of the channel information is achieved through the neural network. Reinforcement learning is one area in machine learning, emphasizing that agents act on the environment (Action) according to its State (State) to get the maximum expected Reward (Reward). When the initial environment is unknown, the agent needs to interact with the environment continuously, gradually improving its policy. When the environment is known or approximately known, the calculation is only carried out according to the model (such as a neural network). In this embodiment, a first neural network may be used as an Agent, a channel matrix may be used as a State, and CSI may be used as an Action, so that the first neural network may learn a relationship between the channel matrix and the CSI. In training the first neural network, the Reward may be determined according to an MCS obtained by a network device (e.g., a base station) according to CSI. In order to distinguish from the CSI obtained by the codebook in the prior art, in the embodiment of the present application, the CSI determined by the first neural network may be defined as Channel Negotiation Information (CNI).

For example, an embodiment of the present application may provide a channel information feedback method. In the Channel Information feedback method, a network device (such as a base station) can send a Channel State Information-Reference Signal (CSI-RS) to a terminal through a downlink Channel. The terminal may measure and estimate a channel matrix of the downlink channel according to the received CSI-RS. Then, the terminal may determine, according to the Channel matrix of the downlink Channel, Channel Negotiation Information (CNI) corresponding to the downlink Channel through the first neural network, and send the CNI corresponding to the downlink Channel to the network device. The network device may determine a Modulation and Coding Scheme (MCS) corresponding to the downlink channel according to the received CNI corresponding to the downlink channel. Wherein the parameter of the first neural network is related to the historical MCS. Similar to the CSI, CNI may be used to indicate channel information of a downlink channel, such as: the CNI may include CQI.

For example, the first neural network may be obtained by performing reinforcement learning training using the sample channel matrix as an input and the sample CNI as an output. Additionally, parameters of the first neural network may be updated based on the historical MCS to correlate the parameters of the first neural network with the historical MCS.

In the channel information feedback method, the terminal determines the CNI corresponding to the downlink channel through the first neural network, and compared with a method of determining the CSI based on a codebook, the method can reduce the quantization error of the terminal when the terminal quantizes the channel matrix, and can effectively improve the accuracy of channel information feedback. In addition, because the parameters of the first neural network are related to the historical MCS, the information entropy of the CNI corresponding to the downlink channel determined by the terminal through the first neural network is smaller than that of the conventional CSI, so that the feedback bit number can be saved, and the efficiency of channel information feedback is improved.

The channel information feedback method provided by the embodiments of the present application is exemplarily described below with reference to the accompanying drawings.

It should be noted that, in the description of the present application, the words "first", "second", and the like are merely used for distinguishing between descriptions, and are not intended to limit a certain feature. In the description of the embodiment of the present application, "and/or" describes an association relationship of associated objects, which means that three relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. At least one referred to in this application means one or more; plural means two or more.

Fig. 3 is a schematic diagram illustrating a communication system according to an embodiment of the present application.

As shown in fig. 3, the communication system of the embodiment of the present application may include: at least one terminal 310 and at least one network device 320 (one network device 320 is shown in fig. 3 by way of example). The terminal 310 and the network device 320 may be communicatively connected, for example, through a wired network or a wireless network.

Optionally, the communication system may be an FDD communication system based on Massive MIMO technology, such as: a Wideband Code Division Multiple Access (WCDMA) System, a Long Term Evolution (LTE) System, an LTE Frequency Division Duplex (FDD) System, a Universal Mobile Telecommunications System (UMTS), a 5G Communication System, and other wireless Communication systems using Orthogonal Frequency Division Multiplexing (OFDM) technology, and the like, and the present application does not limit the specific type of the Communication System.

Alternatively, the terminal 310 in the communication system, or alternatively referred to as a User Equipment (UE), may be a mobile phone ("cellular" phone), a mobile phone, a computer, a cordless phone, a Session Initiation Protocol (SIP) phone, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA), a laptop, a handheld communication device, a handheld computing device, a satellite Wireless device, a Wireless modem card, a Set Top Box (STB), a Customer Premises Equipment (CPE), a wearable device (e.g., a smart watch, a smart bracelet, a pedometer, etc.), a vehicle-mounted device (e.g., an automobile, a bicycle, an electric vehicle, an airplane, a ship, a train, a high-speed rail, etc.), a Virtual Reality (VR) device, an augmented reality (augmented reality, AR) devices, wireless terminals in industrial control (industrial control), smart home devices (e.g., refrigerator, television, air conditioner, electric meter, etc.), smart robots, plant equipment, wireless terminals in self driving (self driving), wireless terminals in remote medical supply, wireless terminals in smart grid, wireless terminals in transportation safety, wireless terminals in smart city, or wireless terminals in smart home, flying devices (e.g., smart robots, hot air balloons, drones, airplanes), and other devices for communicating over a wireless system, etc., and the present application is not limited to the specific form of presentation of the terminal 310.

Network device 120 may be an access network device of the communication system, such as: and a base station. Optionally, in this embodiment, the network device 120 may include various forms of macro base stations, micro base stations (also referred to as small stations), and the like. For example, network device 120 may include: the base station in WCDMA or LTE includes a next generation base station (gNB), a next generation evolved base station (Ng-eNB), a Transmission Reception Point (TRP), an evolved Node B (eNB), a Radio Network Controller (RNC), a Node B (NB), a Base Station Controller (BSC), a Base Transceiver Station (BTS), a home base station (e.g., home Node B or home Node B, HNB), a Base Band Unit (BBU), or a wireless fidelity (Wifi) Access Point (AP).

It can be understood that the application scenario (such as the communication system shown in fig. 3) described in the embodiment of the present application is only for more clearly illustrating the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided in the embodiment of the present application. For example, other devices may also be included in the communication system, such as: network controllers, mobility management entities, and other network entities. In addition, as can be known by those skilled in the art, with the evolution of network architecture and the emergence of new service scenarios, the technical solution provided in the embodiments of the present application is also applicable to similar technical problems.

Fig. 4 shows a schematic composition diagram of a terminal provided in an embodiment of the present application. As shown in fig. 4, the terminal may include: at least one processor 41, a memory 42, a communication interface 43, a bus 44.

The following describes each constituent element of the terminal in detail with reference to fig. 4:

the processor 41 is a control center of the terminal, and may be a single processor or a collective term for a plurality of processing elements. For example, the processor 41 is a Central Processing Unit (CPU), an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits configured to implement the embodiments of the present Application, such as: one or more microprocessors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs).

The processor 41 may perform various functions of the terminal by running or executing software programs stored in the memory 42, and calling data stored in the memory 42, among others. For example, the channel information feedback method provided by the embodiment of the present application may be performed.

In particular implementations, processor 41 may include one or more CPUs such as CPU0 and CPU1 shown in fig. 4 as one example.

In a particular implementation, the terminal may include multiple processors, such as processor 41 and processor 45 shown in FIG. 4, as one example. Each of these processors may be a single-Core Processor (CPU) or a multi-Core Processor (CPU). A processor herein may refer to one or more devices, circuits, and/or processing cores for processing data (e.g., computer program instructions).

The memory 42 is used for storing software programs for executing the steps of the method executed by the terminal of the present embodiment, and is controlled by the processor 41. The Memory 42 may be a Read-Only Memory (ROM) or other type of static storage device that can store static information and instructions, a Random Access Memory (RAM) or other type of dynamic storage device that can store information and instructions, an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM) or other optical Disc storage, optical Disc storage (including Compact Disc, laser Disc, optical Disc, digital versatile Disc, blu-ray Disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such. The memory 42 may be self-contained and coupled to the processor 41 via a bus 44. The memory 42 may also be integrated with the processor 41.

The communication interface 43, using any transceiver or like device, is used to communicate with other devices or communication networks, such as: may communicate with network devices of the core network. The communication interface 43 may be an ethernet interface, a Radio Access Network (RAN) interface, a Wireless Local Area Network (WLAN) interface, or the like. The communication interface 43 may include a receiving unit implementing a receiving function and a transmitting unit implementing a transmitting function.

The bus 44 may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended ISA (EISA) bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in FIG. 4, but this does not indicate only one bus or one type of bus.

Of course, in practical applications, the processor 41, the memory 42 and the communication interface 43 in the terminal may be connected not by a bus structure, but by other structures, such as: the star structure is not particularly limited in this application.

Similar to the structure of the terminal shown in fig. 4, in this embodiment, the network device may also include: at least one processor, memory, a communication interface, and a bus, etc. The difference with the terminal is that the memory in the network device is used for storing software programs for executing the steps of the method executed by the network device of the scheme of the application and is controlled by the processor to execute. The rest of the similar or identical parts are not described in detail herein.

Fig. 5 shows a flowchart of a channel information feedback method provided in an embodiment of the present application. As shown in fig. 5, the channel information feedback method may include:

s501, the network equipment sends a channel state information reference signal CSI-RS to the terminal through a downlink channel.

Accordingly, the terminal may receive the CSI-RS transmitted by the network device.

And S502, the terminal measures and estimates a channel matrix of the downlink channel according to the CSI-RS.

For example, the CSI-RS may be a segment of a pilot sequence. After receiving the CSI-RS sent by the network device, the terminal may perform measurement and estimation on channel parameters of a downlink channel according to the CSI-RS, and obtain a channel matrix of the downlink channel.

S503, the terminal determines channel negotiation information CNI corresponding to the downlink channel through the first neural network according to the channel matrix of the downlink channel.

The first neural network may be obtained by performing reinforcement learning training using the sample channel matrix as an input and the sample CNI as an output. The terminal may input the channel matrix of the downlink channel obtained by measurement and estimation to the first neural network, and the first neural network may output the CNI corresponding to the downlink channel. The CNI may be used to indicate channel information of a downlink channel.

S504, the terminal sends the CNI corresponding to the downlink channel to the network equipment.

Accordingly, the network device may receive the CNI corresponding to the downlink channel from the terminal. After receiving the CNI corresponding to the downlink channel, the network device may determine the MCS corresponding to the downlink channel according to the received CNI corresponding to the downlink channel. The MCS may be used for the base station to process data that needs to be sent to the terminal through the downlink channel. For example, the network device may modulate and encode data to be transmitted to the terminal through the downlink channel according to the MCS.

Compared with the mode of determining CSI based on a codebook and feeding back the CSI to the network equipment, in the embodiment of the application, the terminal determines the CNI corresponding to the downlink channel through the first neural network and feeds back the CNI to the network equipment, and the channel matrix does not need to be quantized into the limited codebook for representation, so that the quantization error of the terminal to the channel matrix in the channel information feedback process can be effectively reduced, the accuracy of channel information feedback can be improved, and the CNI fed back to the base station can more accurately reflect the channel information of the downlink channel under dynamic change.

In some embodiments, a second neural network may be adopted on the network device side as Agent, CNI as State, and MCS as Action, so that the second neural network may learn the relationship between CNI and MCS. In training the second neural network, the Reward may also be determined according to an MCS obtained by a network device (e.g., a base station) according to CSI.

Continuing to refer to fig. 5, the step of the network device determining the MCS corresponding to the downlink channel according to the received CNI corresponding to the downlink channel may include S505.

And S505, the network equipment determines a modulation and coding strategy MCS corresponding to the downlink channel through the second neural network according to the CNI corresponding to the downlink channel.

For example, the second neural network may be obtained by training the neural network with the sample CNI as an input and the sample MCS as an output. The network device may input the CNI corresponding to the downlink channel into the second neural network, and the second neural network may output the MCS corresponding to the downlink channel.

In the prior art, after receiving CSI fed back by a terminal, a network device generally queries a mapping relationship table between CQI and MCS to determine an MCS corresponding to the CQI when determining the MCS according to a channel quality indicator CQI in the CSI. However, due to the limitation of spectrum efficiency, CQI and MCS are not in one-to-one correspondence in the mapping relationship table between CQI and MCS, and therefore, there may be a deviation in probability distribution between CQI and MCS.

In the embodiment of the present application, the mapping relationship between the CNI and the MCS may be implemented by the second neural network. Compared with the method of determining the MCS according to the fixed mapping relation table between the CQI and the MCS, in the method of determining the MCS corresponding to the downlink channel through the second neural network by the network device according to the CNI corresponding to the downlink channel, the deviation of the CNI and the MCS on the probability distribution is smaller, and the determined MCS can be more matched with the channel information of the downlink channel indicated by the CNI.

In one possible design, the parameters of the first neural network may be related to historical MCS. The historical MCS is an MCS determined by the network device according to the last CNI (which may be referred to as historical CNI) sent by the terminal.

For example, the network device may obtain the historical decision benefit corresponding to the historical MCS according to the historical MCS, and send the historical decision benefit to the terminal. The terminal may update the parameters of the first neural network based on the historical decision gain, such that the parameters of the first neural network are related to the historical MCS.

That is, in this design, the correlation between the parameter of the first neural network and the historical MCS may be: the parameters of the first neural network are updated by the terminal according to the historical decision gains, and the historical decision gains are obtained by the network equipment according to the historical MCS.

In the design, the parameter of the first neural network can be related to the historical MCS, so the historical MCS determined by the network equipment can play a guiding role in determining the CNI corresponding to the downlink channel through the first neural network by the terminal, the information entropy of the CNI corresponding to the downlink channel determined by the terminal through the first neural network can be smaller than that of the traditional CSI, the feedback bit number can be saved, and the efficiency of channel information feedback is improved.

Optionally, in the channel information feedback method shown in fig. 5, after the network device determines the MCS corresponding to the downlink channel according to the received CNI corresponding to the downlink channel, the network device may also obtain the decision benefit corresponding to the MCS according to the determined MCS corresponding to the downlink channel, and send the decision benefit corresponding to the MCS to the terminal. The terminal can update the parameters of the first neural network according to the decision gain corresponding to the MCS, so that the parameters of the first neural network can be dynamically updated along with the MCS determined by the network equipment each time, and the dynamic update of the mapping relation between the channel matrix and the CNI is realized.

In the embodiment of the application, the historical decision benefits and the decision benefits are the same type of data. The difference is that the historical decision benefit is obtained by the network device according to the historical MCS, and the decision benefit is obtained by the network device according to the currently determined MCS corresponding to the downlink channel.

Next, with reference to fig. 6, a process of acquiring, by the network device, the decision gain corresponding to the MCS according to the MCS corresponding to the downlink channel, and a process of updating, by the terminal, the parameter of the first neural network according to the decision gain corresponding to the MCS are described. And similarly, the network equipment obtains the historical decision benefits corresponding to the historical MCS according to the historical MCS, and the terminal updates the parameters of the first neural network according to the historical decision benefits, which is not repeated again.

Fig. 6 shows another flow chart of a channel information feedback method provided in an embodiment of the present application. As shown in fig. 6, on the basis of the channel information feedback method shown in fig. 5, after S505, the channel information feedback method may further include S601-S603, where S601-S603 are processes for updating parameters of the first neural network in the channel information feedback method.

S601, the network equipment obtains the decision gain corresponding to the MCS according to the MCS corresponding to the downlink channel.

Optionally, the network device may send the data packet to the terminal according to the MCS corresponding to the downlink channel. The terminal may reply to the network device with an Acknowledgement (ACK) or a Negative Acknowledgement (NACK) according to the received data packet. The network device may obtain the decision gain corresponding to the MCS according to the ACK or NACK replied by the terminal.

It should be noted that if there is no Uplink scheduling, the terminal may reply ACK or NACK using a Physical Uplink Control Channel (PUCCH); if there is Uplink scheduling, the ACK or NACK may be replied using a Physical Uplink Shared Channel (PUSCH) or PUCCH.

In an embodiment, the decision gain may be a decision code rate when the network device performs scheduling according to the MCS corresponding to the downlink channel. The decision code rate refers to a code rate that can be successfully received by the terminal when the network device sends a data packet to the terminal according to the MCS corresponding to the downlink channel.

In another embodiment, the decision benefit may be a code rate difference between a decision code rate and a reference code rate when the network device performs scheduling according to the MCS corresponding to the downlink channel.

In another embodiment, the decision benefit may also be throughput of the network device when performing scheduling according to the MCS corresponding to the downlink channel. For example, the throughput may be the amount of data successfully transmitted per unit time when the network device performs scheduling according to the MCS corresponding to the downlink channel.

In another embodiment, the decision gain may be a combination of any one of a decision code rate and a throughput when the network device performs scheduling according to the MCS corresponding to the downlink channel, and the evaluation value of the CNI corresponding to the downlink channel by the network device. The evaluation value of the CNI corresponding to the downlink channel by the network device may be used to indicate the magnitude of the guiding function of the CNI corresponding to the downlink channel on the determination of the MCS corresponding to the downlink channel by the network device, and may be used to measure whether the CNI can effectively cooperate with the network device to solve the scheduling task. The larger the evaluation value of the CNI corresponding to the downlink channel by the network device is, the larger the guiding function is. The smaller the evaluation value of the CNI corresponding to the downlink channel by the network device is, the smaller the guiding function is. When the network device determines the MCS corresponding to the downlink channel according to the CNI corresponding to the downlink channel, the network device may evaluate the guiding function of the CNI and give an evaluation value. For example: the evaluation value may be a value between 0 and 10, and when the MCS can be determined from the CNI, the evaluation value may be a value of 1, 3.5, 6, 10, or the like. When the MCS cannot be determined from the CNI, the evaluation value may be 0 or the like.

However, the range of the evaluation value is merely an exemplary description. In other embodiments, the evaluation value may be in other larger or smaller numerical ranges. The scope or specific implementation of the evaluation value is not limited in the present application, provided that the evaluation value can be used to indicate the magnitude of the guiding effect of CNI on MCS. For example, in some embodiments, the range of the evaluation value may be a certain percentage interval divided according to the mutual information between the CNI and the MCS.

According to any of the embodiments described above, the network device may obtain a decision gain corresponding to the MCS according to the MCS corresponding to the downlink channel, and send the decision gain to the terminal. The terminal may update the parameters of the first neural network according to the received decision gain. Such as: please refer to S602-S603.

S602, the network equipment sends the decision benefit to the terminal.

Accordingly, the terminal may receive decision revenue from the network device.

And S603, the terminal updates the parameters of the first neural network according to the decision benefits.

Optionally, the terminal may update the parameter of the first neural network according to the first objective function according to the decision gain. The parameters of the first neural network may refer to connection weights and bias values of neurons in the first neural network. The terminal updates the parameters of the first neural network according to the decision gain and the first objective function, namely: and the terminal adjusts the connection weight and the offset value of each neuron in the first neural network according to the decision gain and the first objective function, so that the error of the first neural network is smaller and smaller.

In some embodiments, the first objective function may be a policy-based objective function. For example, assuming that the parameter of the first neural network is θ, the first objective function may be as follows:

wherein J (θ) represents a first objective function; s represents a channel matrix of a downlink channel measured and estimated by the terminal; a represents a CNI determined by a terminal through a first neural network according to a channel matrix of a downlink channel estimated by measurement; pi_θ(s, a) representing a policy function containing a parameter θ of the first neural network;

means to expect all policies; r (s, a) represents a decision benefit;

the entropy of the CNI determined by the terminal through the first neural network is represented, and the entropy can be used as a regular term to increase the exploration capacity of the first objective function and improve the robustness of the first neural network;

the search weight may be controlled by a coefficient β, β being greater than or equal to 0 and less than or equal to 1. When updating the parameters of the first neural network based on the first objective function, the optimization objective (or may be referred to as an update strategy) may be to maximize the first objective function.

The first objective function may be used to increase the selection probability of CNIs with decision benefit greater than 0 and decrease the selection probability of CNIs with decision benefit less than 0.

Based on the first objective function, the terminal may update the parameter θ of the first objective function, so as to update the parameter of the first neural network.

For example, the parameter θ and the step size α may be initialized (e.g., initial values of θ and α may be randomly generated). Then, the terminal may update the parameter θ by using the following policy update function using a gradient ascent method according to the channel matrix of the downlink channel estimated by measurement (s in the first objective function), the CNI determined by the first neural network (a in the first objective function), and the decision benefit sent by the network device (R (s, a) in the first objective function):

as described above in the related explanation of S601, in different embodiments, the decision benefit may be different types of data. For different types of decision gains, different first objective functions may be employed in the embodiments of the present application. That is, the first objective function is related to the decision gain.

For example, the following steps are carried out:

1) if the decision gain (R (s, a)) is a decision code rate when the network device performs scheduling according to the MCS corresponding to the downlink channel, the first objective function may be expressed as:

wherein, rate (a) and ACK represent decision code rate.

In this case, the task requirement of the network device may be to increase the decision code rate.

2) If the decision gain (R (s, a)) is a code rate difference between a decision code rate and a reference code rate when the network device performs scheduling according to the MCS corresponding to the downlink channel, the first objective function may be represented as:

wherein, rate (a), ACK-rate (baseline) represents the code rate difference between the decision code rate and the reference code rate; rate (a) ACK denotes the decision code rate; rate (baseline) represents the reference code rate. For example, the reference code rate may be 10% block error rate.

In this case, the task requirement of the network device may be to obtain a decision code rate that is better than the reference code rate.

3) If the decision gain (R (s, a)) is a combination of a decision code rate when the network device performs scheduling according to the MCS corresponding to the downlink channel and an evaluation value of the CNI corresponding to the downlink channel by the network device, the first objective function may be expressed as:

wherein, rate (a) ACK represents a decision code rate; i (a)_BS；a_UE) The evaluation value of the CNI corresponding to the downlink channel, which may be mutual information of the CNI and the MCS corresponding to the downlink channel, is represented by the network device; p and q are both constants.

In this case, the task requirement of the network device may be to increase the code rate and simultaneously increase the evaluation value of the CNI.

Similarly, if the decision benefit is the throughput of the network device when performing scheduling according to the MCS corresponding to the downlink channel, or the combination of the throughput of the network device when performing scheduling according to the MCS corresponding to the downlink channel and the evaluation value of the CNI corresponding to the downlink channel by the network device, the first objective function may refer to the above exemplary description, and is not described in detail here.

Alternatively, the first neural network may be a convolutional neural network, a long-short term memory recurrent neural network, or the like. The specific structure and dimensions of the first neural network may be determined according to the type of input data. For example, for the channel matrix with a large number of parameters in the Massive MIMO communication system described in the foregoing embodiments, a convolutional neural network may be employed. For some scenarios of predicting channel matrix, a recurrent neural network with long and short term memory can be adopted by inputting the channel information of the previous period. The specific type of the first neural network is not limited by the embodiments of the present application.

In this embodiment, the terminal may update the parameter of the first neural network according to the decision benefit sent by the network device, so that the dynamic update of the mapping relationship between the channel matrix and the CNI may be implemented, and the information entropy of the CNI corresponding to the downlink channel determined by the terminal through the first neural network next time may be smaller.

Similarly, in one possible design, the parameters of the second neural network may also be related to historical MCS.

For example, after obtaining the historical decision gain corresponding to the historical MCS according to the historical MCS, the network device may also update the parameter of the second neural network according to the historical decision gain, so that the parameter of the second neural network is related to the historical MCS. The obtaining manner of the historical decision benefit may refer to the foregoing embodiment, and is not described herein again.

That is, in this design, the correlation between the parameters of the second neural network and the historical MCS may refer to: the parameters of the second neural network are parameters updated by the network equipment according to historical decision gains, and the historical decision gains are obtained by the network equipment according to historical MCS.

In the design, the parameters of the second neural network may be related to the historical MCS, so the historical MCS determined by the network device can guide the determination of the MCS of the downlink channel through the second neural network according to the CNI, thereby further reducing the possible deviation of the CNI and the MCS on the probability distribution.

Optionally, in the channel information feedback method, after the network device obtains the decision gain corresponding to the MCS according to the determined MCS corresponding to the downlink channel, the network device may also update the parameter of the second neural network according to the decision gain corresponding to the MCS. Therefore, the parameters of the second neural network can be dynamically updated along with the MCS determined by the network equipment each time, so as to realize dynamic update of the mapping relation between the CNI and the MCS.

For example, fig. 7 shows another flowchart of a channel information feedback method provided in an embodiment of the present application. As shown in fig. 7, on the basis of the channel information feedback method shown in fig. 5, after S505, the channel information feedback method may further include S701-S702, where S701-S702 are processes for updating parameters of the second neural network in the channel information feedback method.

S701, the network equipment obtains the decision gain corresponding to the MCS according to the MCS corresponding to the downlink channel.

The step of the network device obtaining the decision benefit corresponding to the MCS according to the MCS corresponding to the downlink channel is the same as S601 described in the foregoing embodiment, and is not described herein again.

S702, the network equipment updates the parameters of the second neural network according to the decision benefits.

Optionally, the network device may update the parameters of the second neural network according to the second objective function according to the decision gain. The parameters of the second neural network may refer to connection weights and bias values of neurons in the second neural network. The network equipment updates the parameters of the second neural network according to the decision gain and the second objective function, wherein the step of updating the parameters of the second neural network is as follows: and the network equipment adjusts the connection weight and the offset value of each neuron in the second neural network according to the decision gain and the second objective function, so that the error of the second neural network is smaller and smaller.

The form of the second objective function is similar to the first objective function described in the previous embodiments. Assuming that the parameters of the second neural network are also denoted as θ, the second objective function may be:

the difference from the first objective function is that in the second objective function: j (θ) represents a second objective function; s represents a CNI corresponding to a downlink channel received by the network equipment; a represents MCS determined by the network equipment through the second neural network according to the CNI corresponding to the downlink channel; pi₀(s, a) represents a policy function containing a parameter θ of the second neural network;

means to expect all policies; r (s, a) represents a decision benefit;

the entropy of the MCS determined by the network equipment through the second neural network is represented, and the entropy can be used as a regular item to increase the exploration capacity of the second objective function and improve the robustness of the second neural network;

the search weight may be controlled by a coefficient β, β being greater than or equal to 0 and less than or equal to 1. When updating the parameters of the second neural network based on the second objective function, the optimization objective (or may be referred to as an update strategy) may be to maximize the second objective function.

The second objective function may be used to increase the selection probability of the MCS with the decision gain greater than 0 and decrease the selection probability of the MCS with the decision gain less than 0.

Based on the second objective function, the network device may update the parameter θ of the second objective function, so as to update the parameter of the second neural network.

For example, the parameter θ and the step α may be initialized first (e.g., initial values of θ and α may be randomly generated) with reference to a manner of updating the parameter of the first neural network. Then, the terminal may update the parameter θ by using the following policy update function using a gradient ascent method according to the CNI (s in the second objective function) corresponding to the downlink channel, the MCS determined by the second neural network (a in the second objective function), and the decision gain obtained by the network device (R (s, a) in the second objective function):

in addition, as in the foregoing embodiment in which the parameters of the first neural network are updated, when the parameters of the second neural network are updated, the decision gains may be different types of data in different embodiments. Different second objective functions may also be employed for different types of decision gains. That is, the second objective function is also related to the decision gain. And will not be illustrated here.

Alternatively, the second neural network may also be a convolutional neural network, a cyclic neural network with long and short term memory, and the like, and the specific type of the second neural network is not limited in the embodiments of the present application as well.

In some embodiments, the channel information feedback method according to the embodiment of the present application may be a scheme that only includes updating the parameter of the first neural network, as shown in fig. 6.

In other embodiments, the channel information feedback method according to the embodiment of the present application may also be a scheme that only updates the parameters of the second neural network, as shown in fig. 7.

In still other embodiments, the channel information feedback method according to the embodiment of the present application may also be combined with the foregoing fig. 6 and fig. 7, and include a scheme for updating parameters of the first neural network and a scheme for updating parameters of the second neural network.

When the scheme for updating the parameters of the first neural network and the scheme for updating the parameters of the second neural network are included, the time for the network device to update the parameters of the second neural network may be before the terminal updates the parameters of the first neural network or after the terminal updates the parameters of the first neural network, and the application is not limited herein.

As can be seen from the above, in the embodiment of the present application, the terminal may determine, through the first neural network, the CNI corresponding to the downlink channel according to the channel matrix of the downlink channel measured and estimated, and send the CNI corresponding to the downlink channel to the network device. The network device may determine, by the second neural network, an MCS corresponding to the downlink channel according to the received CNI corresponding to the downlink channel. After determining the MCS corresponding to the downlink channel, the network device may obtain a decision gain corresponding to the MCS according to the determined MCS. Then, the network device may send the decision benefit to the terminal, and the terminal may update the parameter of the first neural network according to the decision benefit, so as to dynamically update the mapping relationship between the channel matrix and the CNI. In addition, the network device may also update the parameters of the second neural network according to the decision benefit, so as to dynamically update the mapping relationship between the CNI and the MCS.

For a communication system consisting of a terminal and network equipment, the first neural network and the second neural network together implement mapping between the channel matrix and the MCS. When the parameters of the first neural network and the parameters of the second neural network are updated in the above manner, it is equivalent to dynamically updating the mapping relationship between the channel matrix and the MCS.

Based on such understanding, the embodiments of the present application further provide a channel information feedback method, in which a network device may update parameters of a second neural network according to a decision gain in the manner described in the foregoing embodiments. And the terminal can update the parameters of the first neural network according to the errors of the parameters of the first layer hidden layer of the second neural network before and after the parameters of the second neural network are updated by the network equipment.

For example, fig. 8 shows another flowchart of a channel information feedback method provided in an embodiment of the present application. As shown in fig. 8, on the basis of the channel information feedback method shown in fig. 5, after S505, the channel information feedback method may further include S801-S805, where S801-S805 are processes of updating both the parameter of the first neural network and the parameter of the second neural network.

S801, the network equipment obtains a decision gain corresponding to the MCS according to the MCS corresponding to the downlink channel.

S802, the network equipment updates the parameters of the second neural network according to the decision benefits.

The step of updating, by the network device, the parameter of the second neural network according to the decision benefit is the same as that described in the foregoing embodiment, and is not described herein again.

And S803, the network equipment acquires the errors of the parameters of the first hidden layer of the second neural network before and after the parameters of the second neural network are updated.

For example, the network device may obtain pre-update and post-update errors for connection weights and bias values for neurons in a first layer hidden layer of the second neural network.

S804, the network equipment sends the error of the parameters of the first hidden layer of the second neural network to the terminal.

Accordingly, the terminal may receive an error from a parameter of a first layer hidden layer of a second neural network of the network device.

S805, the terminal updates the parameters of the first neural network according to the errors of the parameters of the first hidden layer of the second neural network.

Optionally, the terminal may update the parameter of the first neural network according to a back propagation algorithm according to the error of the parameter of the first hidden layer of the second neural network.

It should be noted that, if the terminal needs to consider the binary representation when outputting the CNI through the first neural network, discrete sampling processing needs to be performed on the error of the parameter of the first hidden layer of the second neural network when updating the parameter of the first neural network according to the back propagation algorithm and the error of the parameter of the first hidden layer of the second neural network. For example, the gradient back propagation from the network device to the terminal can be guaranteed by a Gumbel-Softmax estimator.

It can be understood that, in this embodiment, the terminal needs to update the parameter of the first neural network according to the error of the parameter of the first hidden layer of the second neural network before and after the network device updates the parameter of the second neural network. Therefore, the time for the network device to update the parameters of the second neural network must be before the terminal updates the parameters of the first neural network.

Compared with the foregoing embodiment, in which the terminal updates the parameters of the first neural network according to the decision benefit and the network device updates the parameters of the second neural network according to the decision benefit, the embodiment can simplify the updating manner of the parameters of the first neural network.

Optionally, in this embodiment of the application, when the first neural network and the second neural network are obtained through training, the first neural network and the second neural network may also be trained in combination with the method for updating the parameters of the first neural network and the second neural network described in any of the foregoing embodiments.

In some possible designs, the first objective function described in the foregoing embodiments may also be implemented based on a value function. For example, the first objective function may be implemented by the following value function:

wherein J (θ) represents a first objective function; θ represents a parameter of the first neural network; s represents a channel matrix of a downlink channel measured and estimated by the terminal; a represents a CNI corresponding to a downlink channel determined by a terminal through a first neural network; r (s, a) represents a decision benefit sent by the network device to the terminal, for example, the decision code rate, throughput, etc. described in the foregoing embodiment; q (s, a) is a cost function of the action, representing the downlink signal approximated at s (estimated measurement at the terminal)Channel matrix of the channel) to select the value of a (CNI corresponding to the downlink channel determined by the terminal through the first neural network);

indicating that all values are desired. When updating the parameters of the first neural network based on the first objective function, the optimization objective (or may be referred to as an update strategy) may be to minimize the first objective function so that the mean square error of the action cost function Q (s, a) and the decision gain R (s, a) is minimized.

Similarly, the second objective function described in the foregoing embodiments may also be implemented based on a value function. For example, the second objective function may also be implemented by the following value function:

in the second objective function, J (θ) represents a second objective function, which is different from the first objective function; θ represents a parameter of the second neural network; s represents a CNI corresponding to a downlink channel received by the network equipment; a represents MCS corresponding to a downlink channel determined by the network equipment through the second neural network; r (s, a) represents the decision gain corresponding to the MCS acquired by the network equipment; q (s, a) is an action cost function representing the value of selecting a (MCS corresponding to the downlink channel determined by the network device through the second neural network) approximately under s (CNI corresponding to the downlink channel received by the network device);

indicating that all values are desired. When updating the parameters of the second neural network based on the second objective function, the optimization objective may also (or may be referred to as an update strategy) be to minimize the second objective function so that the mean square error of the action cost function Q (s, a) and the decision gain R (s, a) is minimized.

It should be noted that, in the embodiments of the present application, the first objective function and the second objective function based on the policy function, or the first objective function and the second objective function based on the value function, are all exemplary illustrations, and the specific types of the first objective function and the second objective function are not limited in the present application.

In summary, in the embodiment of the present application, a Channel Information feedback amount facing to a task is trained in a reinforcement learning manner, and is defined as Channel Negotiation Information (CNI), so that in a Channel Information feedback process, a terminal and a network device can dynamically adjust a mapping relationship from a measured Channel matrix to an MCS according to distribution of current Channel Information (i.e., according to a decision gain corresponding to the MCS).

Wherein, the task-oriented channel information feedback means: the network equipment provides prior information for the terminal according to the requirement of the task of the network equipment, and the terminal can compress and feed back the current channel information in a targeted manner under the condition of the prior information. In the embodiment of the present application, the priori information is the decision gain or the error of the first hidden layer of the second neural network.

It can be appreciated that CNI is conceptually the same as CSI, as compared to conventional CSI feedback, and is a kind of quantization of channel information. However, in the implementation of feedback, the mapping of channel information to CSI and the mapping of CSI to MCS in the existing standard are fixed. In the embodiment of the present application, because the first neural network receives the reward information (the decision benefit or the error of the first hidden layer of the second neural network) fed back by the network device in the updating process, and the reward information can be understood as the prior information of the MCS distribution of the network device, and further adjusts the quantization of the terminal on the channel information, the information entropy included in the CNI is smaller than the CSI, which is equivalent to a higher precision under the constraint of the same bit number.

Optionally, based on the channel information feedback method described in the foregoing embodiment, an embodiment of the present application further provides a channel information feedback method. In the channel information feedback method, before the terminal updates the parameters of the first neural network and the network device updates the parameters of the second neural network, the terminal may obtain the restored channel matrix according to the CNI corresponding to the downlink channel. For example, a restoration network may be deployed in the terminal. In the terminal, the first neural network can be used as a coding network, and the CNI corresponding to the downlink channel is output according to the channel matrix of the downlink channel estimated by measurement; the reduction network can output the reduced channel matrix according to the CNI corresponding to the downlink channel. Then, the terminal may update the parameter of the first neural network according to the third objective function according to the channel matrix of the downlink channel estimated by the measurement and the restored channel matrix.

The third objective function may be as follows:

wherein J (θ, ξ) represents a third objective function; theta is a parameter of the first neural network; xi is a parameter of the reduction network; h represents a channel matrix of a downlink channel measured and estimated by the terminal; f (·; theta) represents a first neural network which can be used as an encoder to compress and quantize H; f (H; theta) represents quantization information obtained by compressing and quantizing H by the terminal through f, namely CNI corresponding to a downlink channel; g (·; ξ) represents a reduction network which can be used as a decoder to reduce f (H; theta); g (f (H; theta); xi) represents the restored channel matrix obtained by restoring f (H; theta) by the terminal through g (·; xi).

The third objective function may be used to indicate that an error between the channel matrix of the downlink channel estimated by the measurement and the restored channel matrix is minimized.

For example, the error may be a minimum mean square error, and the terminal may adjust the connection weights and the bias values of the neurons in the first neural network based on the third objective function, so that the minimum mean square error between the channel matrix of the measurement-estimated downlink channel and the restored channel matrix is minimum.

Optionally, in this embodiment of the application, an update step size for updating the first neural network based on the third objective function is far larger than an update step size for updating the first neural network based on the first objective function and an update step size for updating the second neural network based on the second objective function, so as to ensure that the measurement gradient can track the Markov chain change. For example, the update step size for updating the first neural network based on the first objective function, and the update step size for updating the second neural network based on the second objective function, may be 10 "5. The update step for updating the first neural network based on the third objective function needs to be a value much larger than the aforementioned 10 "5, such as: may be 10-2. It should be noted that, the present application does not limit specific sizes of the update step for updating the first neural network based on the third objective function, the update step for updating the first neural network based on the first objective function, and the update step for updating the second neural network based on the second objective function, and the values of the update steps are merely exemplary.

The framework for applying AI techniques such as reinforcement learning to measurement, feedback, and control of a communication system provided in the embodiments of the present application is not limited to the measurement and feedback of the downlink channel, and is also applicable to other application scenarios. The entities communicating with each other are not limited to communication between the terminal and the network device, and may be communication between the terminal and the network device, and communication between the network device and the network device. Between the network entities for communication, the first entity can provide prior information for the second entity according to the requirement of the task of the first entity, and the second entity performs compression feedback on the current channel information according to the prior information.

The embodiment of the present application provides another channel information feedback method, which is applied to a Wireless Mesh Network (Wireless Mesh Network), and is also referred to as a "multi-hop" Network.

In the wireless mesh network, any wireless device node can simultaneously serve as an access point and a router, and each node in the network can send and receive signals, so that the influence of node faults and interference on the operation of the whole network is reduced. Because the wireless mesh network is a mesh topology structure, the nodes need to interact to realize a dynamic configuration function, an optimal transmission path is selected for the service, and a plurality of tasks with different service quality requirements, such as voice service and data service, can be executed in the wireless mesh network, and the optimal path is determined according to different task requirements.

The following description will take the transmission of a wireless mesh network composed of three nodes as an example. Fig. 9a is an application scenario diagram of a path information feedback method according to an embodiment of the present application. As shown in fig. 9a, when entity a needs to send information to entity B to complete the task of a specific qos requirement at entity B, entity a may send information directly to entity B, or may forward information to entity B by using entity C as a forwarding node, thereby adjusting the efficiency of the wireless mesh network. That is, there are two paths in the network, path a-B and path a-C-B, respectively. When the information is transmitted through each path, the network entities on the paths can compress the information to be transmitted according to task requirements and prior information by using the deployed neural networks. In fig. 9a, solid arrows indicate transmission of compressed path information, and dotted lines indicate transmission of information such as decision gain.

As shown in fig. 9b, fig. 9b is an interaction diagram of a path information feedback method provided in the embodiment of the present application, which is suitable for the application scenario shown in fig. 9 a. Wherein entities A, B, C are all agents.

S1201a, the entity B sends a reference signal to the entity a, and correspondingly, the entity a receives the reference signal sent by the entity B.

In one possible implementation, the reference signal may be a CSI-RS

S1201B, the entity B sends a reference signal to the entity C, and correspondingly, the entity a receives the reference signal sent by the entity C.

In one possible implementation, the reference signal may be a CSI-RS

S1202a, entity a measures the path parameters of estimated path B-a.

The path parameters include, but are not limited to, channel matrix, bit error rate, routing hop count for path B-a.

The channel matrix of the path B-A is obtained by the entity A according to the reference signal measurement estimation after receiving the reference signal; the bit error rate is obtained by measuring the accuracy of downlink data transmission in a certain time and calculating; the channel matrix may be used as an estimate of short-term channel quality and the bit error rate may be used as an estimate of long-term channel quality.

The routing hop count is calculated by the network topology information of the routing hop count; specifically, topology information of the WMN network may be stored at each node, and each node may uniformly update the network topology information of the WMN whenever a topology structure changes (increase, decrease the availability of nodes or no communication between nodes). When the sending end and the receiving end are determined, each node can estimate the network topology information of the node.

In one possible implementation, the Routing Information Protocol (RIP) setting is used to define the number of Routing hops from the sending end to the receiving end through the router. In the scenario shown in FIG. 9a, there are 2 paths, A-B, A-C-B. The route hop number given by A in the road section A-B is 1; the route hop count of path a-C-B is given by a is 2 and the route hop count is given by C is 1. As another example, if there are 5 paths, A-C-B; A-C-D-B; A-D-C-B; A-D-B; A-B. The route hop number estimated by each node in the path A-D-C-B is respectively A: 3, D: 2, C: 1; the estimated route hop numbers of all nodes in the path A-D-B are respectively A: 2, D: 1.

S1202B, entity C measures the path parameters of estimated paths B-C.

The path parameters include, but are not limited to, channel matrix, bit error rate, routing hop count for paths B-C.

The channel matrix of the path B-C is obtained by the entity C according to the reference signal measurement estimation after receiving the reference signal; the error rate of each path is obtained by measuring the accuracy of downlink data transmission in a certain time; the number of routing hops is obtained through network topology information.

It is to be noted that the present application does not limit the order of steps S1201a, S1202a and steps S1201b, S1202 b.

S1203, the entity A determines the CNI of the path B-A through a third neural network according to the path parameters of the path B-A.

The third neural network can be obtained by performing reinforcement learning training with the sample path parameters as input and the sample CNI as output. The terminal may input the measured and estimated path parameters of the path B-C to the third neural network, and the third neural network may output the CNI of the path B-a (CNI _ BA represents the CNI of the entity B to the entity a).

The CNI of the path B-a may specifically be information obtained by compressing the parameter information of the path B-a according to the task requirement executed by the entity B.

S1204, entity A sends CNI of route B-A to entity B, correspondingly, entity B receives CNI of route B-A sent by entity A.

S1205, the entity C sends the reference signal to the entity a, and correspondingly, the entity a receives the reference signal sent by the entity C.

In one possible implementation, the reference signal may be a CSI-RS.

S1206, entity A measures path parameters of estimated path C-A

The path parameters include, but are not limited to, channel matrix, bit error rate, routing hop count for path C-a.

The channel matrix of the path C-A is obtained by the entity A according to the reference signal measurement estimation after receiving the reference signal;

the bit error rate is obtained by measuring the accuracy of downlink data transmission in a certain time and calculating;

the routing hop count is calculated by the network topology information of the routing hop count. S1207, the entity A determines the CNI of the path C-A through a fourth neural network according to the path parameters of the path C-A.

The fourth neural network can be obtained by performing reinforcement learning training with the sample path parameters as input and the sample CNI as output. The terminal may input the measured and estimated path parameters of the path C-a to the fourth neural network, and the fourth neural network may output the CNI of the path C-a (CNI _ CA represents the CNI of the entity C to the entity a).

The CNI of the path C-a may specifically be information obtained by compressing the parameter information of the path C-a according to the task requirement executed by the entity B.

S1208, the entity A sends the CNI of the path C-A to the entity C, and correspondingly, the entity C receives the CNI of the path C-A sent by the entity A.

It should be noted that the above steps S1205-S1208 may be executed after, before or in the middle of steps S1201a-S1204, and the present application is not limited thereto.

S1209, the entity C determines the CNI of the path B-C-A through a fifth neural network according to the path parameters of the path B-C and the CNI corresponding to the path C-A.

The fifth neural network can be obtained by performing reinforcement learning training with the sample path parameters as input and the sample CNI as output. Entity C may input the measured estimated path parameters of path B-C and the received CNI corresponding to path C-a into a fifth neural network, which may output the CNI of path B-C-a (CNI _ BCA represents CNI from entity B to entity a via entity C).

The CNI of the path B-C-a may specifically be compressed information obtained by integrating the B-C path parameter information and the received CNI information of the a-C according to the task requirements executed by the entity B.

S1210, the entity C sends the CNI of the path B-C-A to the entity B, and correspondingly, the entity B receives the CNI of the path B-C-A sent by the entity C.

And S1211 and the entity B determine a transmission path through a sixth neural network according to the CNI corresponding to each path.

The sixth neural network may be obtained by performing reinforcement learning training with the sample path parameters as input and the sample scheduling decisions as output. Entity B may input the CNI (e.g., CNI _ BA, CNI _ BCA) of each path into a sixth neural network, which may output scheduling decisions, e.g., the path selected by entity B, the weights of the path combinations, the precoding matrix, etc.

The precoding matrix refers to a precoding matrix of multi-user precoding fed back by a distributed (multi-path) channel, and is used for allocating transmission power and transmission rate to obtain optimal performance.

And S1212, the entity B transmits data according to the determined transmission path.

Specifically, the entity B sends data to the entity a and/or the entity C according to the path combination weights, and receives ACK or NACK replied by the entity a and/or the entity C to the entity B.

For example, where the overall Data sent by entity B is Data, it is divided into Data1 and Data 2 according to the path weights. The entity B sends Data1 to the entity A and receives ACK _ AB or NACK _ AB sent by the entity A; the entity B sends and sends Data 2 to the entity C, and the entity C forwards the Data 2 to the entity A; and the entity C receives the ACK _ AC or the NACK _ AC sent by the entity A and returns the ACK _ CB or the NACK _ CB to the entity B. Data1 or Data 2 may be null, for example, Data1 is null, which means that entity B transmits Data to entity a without going through path B-a, and the total Data is all transmitted through path B-C-a.

S1213, the entity B obtains a first decision gain corresponding to the transmission path according to the transmission path.

Specifically, the entity B obtains a first decision benefit corresponding to the transmission path according to the ACK or NACK replied by the entity a and/or the entity C.

The first decision benefit includes, but is not limited to, one or more of the following: throughput, channel capacity, power consumption. The calculation criteria for the first decision benefit are related to the specific task performed by entity B. In one possible implementation, the tasks are: by determining the weight of the path combination, the performance index under a certain scene (e.g. voice service) is maximized. The calculation criteria of the decision gain of the path combination weight may have the following relationship:

setting the throughput as T, the power consumption as P, alpha and beta not less than 0 as the demand proportion of the route hop number H and the error rate E (or other performance indexes such as time delay and bandwidth) corresponding to the task, respectively, and setting the path combination weight of the decision as lambda_iI represents the selectable path and has ∑ λ _i1, the total mission decision yield R₁Can be defined as:

throughput refers to the amount of data successfully transmitted per unit time; channel capacity refers to the minimum upper bound of achievable rates at which reliable transmission is possible in a channel; the power consumption refers to the amount of energy consumed by the device in a unit time, and the throughput, the channel capacity, the power consumption and the like can be calculated according to the selected path through channel matrix information, the self power consumption of the router on the path and the like.

In another possible implementation, the tasks are: the optimal path of the transmission service is determined, the balance performance index under various communication scenes (voice service and data service) is maximized, and the quality requirements of different services are met. For the above tasks, the calculation criteria of the decision gain may have the following relationship:

wherein R is₂For the average profit of N rounds of task decision making, the path i is a specific path selected by each task, T_n(i is the throughput, P, corresponding to the nth selected path i_n(i) For the nth selection of the power consumption, α, corresponding to path i_n,β_nNot less than 0 is the route hop number H corresponding to the task and selected for the nth time_n(i) Sum error rate E_n(i) (or delay, bandwidth, and other performance indicators).

Under the task, the CNI is path parameter information obtained by adaptively compressing the path parameter information according to communication scene changes and network topology changes, so that the quality of different services is optimal.

In one possible implementation, entity B updates the sixth neural network based on the obtained first decision benefit. Specifically, the entity B updates the parameters of the sixth neural network according to the objective function according to the first decision gain. The parameters of the sixth neural network may refer to connection weights and programmed values of neurons in the sixth neural network. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₁Or R₂. And will not be described in detail herein.

S1214a, the entity B indicates the first decision benefit to the entity a, and accordingly, the entity a obtains the first decision benefit indicated by the entity B.

S1214a, the entity B indicates the first decision benefit to the entity C, and accordingly, the entity C obtains the first decision benefit indicated by the entity B.

S1215, the entity C obtains a second decision gain according to the transmission path.

Specifically, when the transmission path determined by the entity B includes the entity C, the entity C obtains the second decision benefit.

The second decision benefit includes, but is not limited to, one or more of the following: throughput, channel capacity, power consumption. The calculation criterion of the second decision gain is similar to the calculation criterion of the first decision gain. The difference is that the first decision benefit takes into account the throughput, power consumption, etc. of the combined paths, whereas the second decision benefit takes into account only the throughput, power consumption, etc. between paths C-a.

S1216, the entity C indicates the second decision benefit to the entity a, and accordingly, the entity a obtains the second decision benefit indicated by the entity C.

S1217a, the entity A updates the neural network according to the decision gain.

In one possible implementation, entity a updates the third neural network based on the first decision yield. Specifically, the entity a updates the third neural network according to the objective function according to the first decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₁Or R₂. And will not be described in detail herein.

In one possible implementation, entity a updates the fourth neural network based on the first decision benefit and the second decision benefit. Specifically, the entity a updates the fourth neural network according to the objective function according to the first decision gain and the second decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. Wherein R (s, a) in the first objective function and the second objective function can be replaced by a combination of the first decision gain and the second decision gain according to the actual task requirement. The first decision benefit and the second decision benefit may be combined as a result of the entity a combining the first decision benefit and the second decision benefit according to the actual demand, for example, the first decision benefit and the second decision benefit are respectively multiplied by a certain proportion to obtain the combined decision benefit.

S1217b, the entity C updates the fifth neural network according to the decision benefit.

And the entity C updates the fifth neural network according to the first decision benefit and the second decision benefit. Specifically, the entity C updates the fifth neural network according to the objective function according to the first decision gain and the second decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. Wherein R (s, a) in the first objective function and the second objective function can be replaced by a combination of the first decision gain and the second decision gain according to the actual task requirement. The first decision benefit and the second decision benefit may be combined as a result of the entity C combining the first decision benefit and the second decision benefit according to the actual demand, for example, the first decision benefit and the second decision benefit are respectively multiplied by a certain proportion to obtain the combined decision benefit. And will not be described in detail herein.

In the method shown in fig. 12, an entity a and an entity C receive a reference signal of an entity B, measure and estimate path parameters of paths B-a and C-a, compress corresponding channel negotiation information CNI for the path parameters through a neural network according to different task requirements, and feed back the CNI of different paths (path B-a, B-C-a) to the entity B, so that the entity B makes a scheduling decision through the neural network according to the CNI of the different paths, determines a transmission path, performs data transmission, and calculates a decision benefit. The entity A, B, C updates the neural network according to the decision benefit, wherein the decision benefit is calculated by taking different task requirements into account, so that the neural network at each entity adaptively compresses the path parameters according to the task requirements, and reduces the overhead of path information feedback while selecting the transmission path according to the actual task requirements.

The embodiment of the present application provides another channel information feedback method, which is applied to Coordinated Multi-Point (CoMP), and includes Joint Processing (JP) Coordinated Multi-Point transmission (JP-CoMP), Coordinated Scheduling (CS), and Coordinated Beamforming (CB) Coordinated Multi-Point transmission (CS/CB-CoMP).

The CoMP technology is characterized in that several adjacent network devices serve one terminal at the same time, so that the data rate of a user is improved, the performance of an edge terminal is improved, and the overall performance of the system is improved at the same time. For JP-CoMP, all network devices in a cooperative cell have the same data packet sent by the terminal, and require that the network devices share data information and channel information, such as Non-coherent Joint Transmission (NCJT), and send data of different layers to the terminal; for CS/CB-CoMP, each terminal transmits data with only one network device at the same time, so it does not require shared data information, but requires shared channel information. For the realization of the JP-CoMP, the frequency reuse rate of the cooperative cell needs to be reduced, while the improvement of the CS/CB-CoMP on the edge terminal is limited, and only the interference is reduced and the additional diversity gain cannot be obtained.

If the plurality of network devices and the terminal are all intelligent agents, the two CoMP systems can be regarded as different interactive feedback content degrees between the intelligent agents, and in order to combine the advantages of the two schemes, the reinforcement learning training method provided by the embodiment of the application can realize the self-adaptive adjustment of the two transmission modes by adjusting the interactive signaling, so that the interactive backhaul (backhaul) overhead between the network devices and the communication overhead between each network device and the terminal are reduced.

In the following, three agents shown in fig. 10a are taken as an example for explanation, and fig. 10a is an application scenario diagram of the path information feedback method provided in the embodiment of the present application. Entity D, E is a network device responsible for transmission tasks to the same terminal, and may group these network devices into the same group (cluster). The CNI information fed back by the group interaction is different from the information fed back to the terminal, and includes not only the compressed channel information received by the terminal, but also decision information and decision rewards related to specific tasks. The decision information containing the data information is JP-CoMP; the decision information without data information is CS/CB-CoMP); and the entity F is a terminal, and the entity F compresses the channel information by adopting a neural network to obtain a corresponding CNI and feeds back the CNI. The task under the scene aims at maximizing the downlink transmission performance of the terminal for cooperation among network devices. FIG. 10a, 300, illustrates path information for inter-group compression; in a downstream communication scene, 200 represents transmission compressed path information, and 100 represents information such as feedback decision benefit; in the uplink communication scenario, 100 represents transmission decision information, and 200 represents feedback decision gain and other information.

As shown in fig. 10b, fig. 10b is an interaction diagram of a path information feedback method provided in the embodiment of the present application, which is suitable for the application scenario shown in fig. 10 a. Wherein entities D, E, F are all agents.

S1301, the entity D sends a reference signal to the entity F, and correspondingly, the entity F receives the reference signal sent by the entity D.

In one possible implementation, the reference signal may be a CSI-RS.

S1302, the entity F measures the path parameters of the estimated path D-F.

The path parameters include, but are not limited to, a channel matrix of the path D-F, a location where the entity F is located, a cell or sector to which the entity belongs, and a direct link antenna gain.

The channel matrix of the path D-F is obtained by the entity F according to the reference signal measurement estimation after receiving the reference signal; the position, the cell or the sector where the entity F is located can be obtained through the positioning service of the entity F; direct link antenna gain refers to the ability of an antenna to transmit and receive signals in a particular direction, and may be provided by a network device.

And S1303, the entity F determines the CNI of the path D-F through a seventh neural network according to the path parameters of the path D-F.

The seventh neural network can be obtained by performing reinforcement learning training with the sample path parameters as input and the sample CNI as output. Entity F may input the measured and estimated path parameters of path D-F into a seventh neural network, which may output the CNI of path D-F (CNI _ DF represents the CNI of entity D to entity F).

The CNI of the path D-F may specifically be feedback information obtained by effectively compressing the measured and estimated path parameters according to the task requirements of the network device.

S1304, the entity F sends the CNI of the path D-F to the entity D, and correspondingly, the entity D receives the CNI of the path D-F sent by the entity F.

S1305, the entity E sends a reference signal to the entity F, and correspondingly, the entity F receives the reference signal sent by the entity E.

In one possible implementation, the reference signal may be a CSI-RS

S1306, the entity F measures the path parameters of the estimated paths E-F.

The path parameters include, but are not limited to, a channel matrix of the path E-F, a location where the entity F is located, a cell or sector to which the entity belongs, and a direct link antenna gain.

The channel matrix of the path E-F is obtained by the entity F according to the reference signal measurement estimation after receiving the reference signal; the position, the cell or the sector where the entity F is located can be obtained through the positioning service of the entity F; direct link antenna gain refers to the ability of an antenna to transmit and receive signals in a particular direction, and may be provided by a network device.

S1307, the entity F determines the CNI of the path D-F through the seventh neural network according to the path parameters of the path D-F.

The eighth neural network can be obtained by performing reinforcement learning training using the sample path parameters as input and the sample CNI as output. Entity F may input the measured and estimated path parameters of path D-F to an eighth neural network, which may output the CNI of path E-F (CNI _ DF represents the CNI of entity E to entity F).

The CNI of the path E-F may specifically be feedback information obtained by effectively compressing the measured and estimated path parameters according to the task requirements of the network device.

S1308, the entity F sends the CNI of the path E-F to the entity E, and correspondingly, the entity E receives the CNI of the path E-F sent by the entity F.

In the present application, the order of steps S1301 to S1304 and steps S1305 to S1308 is not limited. That is, steps S1305-S1308 may also be performed before steps S1301-S1304, or simultaneously with steps S1301-S1304. In a possible implementation, the entity F interacts with an entity serving the location of the entity F first, or interacts with an entity corresponding to a cell or sector to which the entity F belongs first.

S1309, the entity D determines the CNI of the path D-E through a ninth neural network according to the CNI of the path D-F and the result of the last round of transmission decision.

The ninth neural network is a neural network that serves communication, and may also be referred to as an interactive neural network or a communicating neural network.

Entity D may input the received CNI for path D-F and the results of the last round of transmission decisions to a ninth neural network, which may output the CNI for path D-E (CNI _ DE denotes CNI for entity D to entity E).

The transmission decision of the previous round may be a transmission decision made by the entity D in a period of time before the current decision, and may specifically be a data table record of a certain length. And when the decision is made for the first time, the transmission decision of the upper round is empty.

The CNI of the path D-E may specifically be a result of compressing the received CNI and the decision information transmitted in the previous round, aiming at a task target of optimizing the performance of the terminal transmission.

S1310, the entity E determines the CNI of the path E-D through a tenth neural network according to the CNI of the path E-F and the result of the last round of transmission decision.

The tenth neural network is a neural network that serves interactive communications, and may also be referred to as an interactive neural network or a communications neural network.

Entity E may input the received CNI for path E-F and the results of the last round of transmission decisions to a tenth neural network, and a ninth neural network may output the CNI for path E-D (CNI _ ED represents the CNI of entity E to entity D).

The transmission decision of the previous round may be a transmission decision made by the entity E in a period of time before the current decision, and specifically may be a data table record of a certain length. And when the decision is made for the first time, the transmission decision of the upper round is empty.

The CNI of the path E-D may specifically be a result of compressing the received CNI and the decision information transmitted in the previous round, aiming at a task target of optimizing the transmission performance of the terminal.

S1311, entity D sends CNI of path D-E to entity E, and correspondingly, entity E receives CNI of path E-D sent by entity D.

S1312, the entity E sends the CNI of the path E-D to the entity D, and correspondingly, the entity E receives the CNI of the path E-D sent by the entity D.

Note that the order of steps S1311 and S1312 is not limited in the present application.

S1313, entity D determines a transmission decision through the eleventh neural network based on the CNI of path D-F and the CNI of path E-D.

The eleventh neural network is a neural network serving a task policy, which may also be referred to as a policy network or an action network, and can be obtained by performing reinforcement learning training with the sample CNI as an input and the sample transmission decision as an output.

Entity D may input the received CNI of path D-F and CNI of path E-D into an eleventh neural network, which may output a transmission decision. The transmission decision result may include adopting a JP-CoMP scheme or a CS/CB-CoMP scheme. The decision of adopting the JP-CoMP also comprises a pre-coding matrix, wherein the pre-coding matrix is used for distributing the sending power and the transmission rate so as to realize the task of optimizing the transmission performance of the terminal; the decision making method adopting the CS/CB-CoMP mode further comprises the following steps: it is determined whether entity D is in data transfer with entity F.

And S1314, the entity E determines a transmission decision through a twelfth neural network according to the CNI of the path E-F and the CNI of the path D-E.

The twelfth neural network is a neural network serving a scheduling policy, which may also be referred to as a policy network or an action network, and can be obtained by performing reinforcement learning training with the sample CNI as an input and the sample transmission decision as an output.

Entity E may input the received CNI for path E-F and CNI for path D-E into a twelfth neural network, which may output a transmission decision. The transmission decision result may include adopting a JP-CoMP scheme or a CS/CB-CoMP scheme. The case of JP-CoMP also includes a precoding matrix. The precoding matrix is used for distributing sending power and transmission rate so as to realize the task of optimizing the transmission performance of the terminal; the decision making method adopting the CS/CB-CoMP mode further comprises the following steps: it is determined whether entity E is in data transfer with entity F.

It should be noted that after steps S1311 and S1312 are completed, a timer mode may be adopted, and a transmission decision result may be output at the same time.

S1315, the entity D performs data transmission with the entity F according to the transmission strategy.

For example, entity D sends Data _ DF to entity F, entity F sends ACK _ FD or NACK _ FD to entity D after receiving Data _ DF, and entity D receives ACK _ FD or NACK _ FD.

In one possible implementation, the transmission policy is CS/CB-CoMP, and the entity D does not perform data transmission with the entity F, then the step S1315 is omitted.

S1316, the entity E transmits data with the entity F according to the transmission strategy.

For example, the entity E sends data Eata _ EF to the entity F, after receiving the data Eata _ EF, the entity F sends ACK _ FE or NACK _ FE to the entity E, and the entity E receives ACK _ FE or NACK _ FE.

In one possible implementation, the transmission policy is a CS/CB-CoMP scheme, and the entity E does not perform data transmission with the entity F, then the step S1316 is omitted.

The data transmission between the entity D and the entity E in S1315 and S1316 and the entity F according to the transmission policy includes: and when the transmission strategy decision result is in a CS/CB-CoMP mode, the entity for determining data transmission performs data transmission with the entity F, the other entity does not perform data transmission with the entity F, and the entity F and the entity for interacting data therewith jointly determine whether to restart the communication between the entity F and the other entity, namely, the communication is switched to the JP-CoMP mode.

S1317, the entity D obtains a corresponding third decision benefit according to the transmission decision result.

In one possible implementation, the entity D obtains a corresponding third decision gain according to the downlink/uplink transmission performance index based on a criterion of maximizing channel capacity. The third decision benefit includes, but is not limited to, one or more of the following: throughput, channel capacity, power consumption of the system. In this case, the task of the entity D aims to maximize the downlink transmission performance to the terminal through cooperation between the entity D and the entity E.

For example, let T be throughput and P be power consumption, and the path combining weight of the decision is set to λ_iI represents the selectable path and has ∑ λ _i1, the total mission decision yield R₁Can be defined as:

R₃＝∑λ_i[(T_i-P_i)]

and when the decision result of the transmission strategy is in a CS/CB-CoMP mode, only one path transmits data. Lambda [ alpha ]_iCan be 0 or 1, i.e. lambda_i∈{0，1}；

When the decision result of the transmission strategy is the JP-CoMP mode, a plurality of paths can be used for transmitting data. Lambda [ alpha ]_iCan take a value between 0 and 1, i.e. λ_i∈(0，1)。

S1318 and the entity E obtains a corresponding fourth decision benefit according to the transmission decision result.

In one possible implementation, the entity E obtains the corresponding fourth decision gain according to the downlink/uplink transmission performance index based on the criterion of maximizing the channel capacity. The fourth decision benefit includes, but is not limited to, one or more of the following: throughput, channel capacity, power consumption of the system. In this case, the task of the entity E aims to maximize the downlink transmission performance to the terminal through cooperation between the entity D and the entity E.

Optionally, a fourth decision gain R may be obtained by the formula in S1317₃。

S1319 and the entity D indicate the third decision benefit to the entity F, and correspondingly, the entity F obtains the third decision benefit indicated by the entity D.

In one possible implementation, entity D sends third decision revenue information to entity F.

S1320, the entity E indicates the fourth decision benefit to the entity F, and accordingly, the entity F obtains the fourth decision benefit indicated by the entity E.

In one possible implementation, entity E sends fourth decision revenue information to entity F.

S1321, the entity D updates the ninth neural network and the eleventh neural network according to the acquired decision benefit.

Specifically, the entity D updates the ninth neural network and the eleventh neural network according to the objective function according to the third decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₃. And will not be described in detail herein.

S1322 and the entity E updates the tenth neural network and the twelfth neural network according to the obtained decision benefit.

S1323, the entity F updates the seventh neural network and the eighth neural network according to the decision benefit.

And after the entity F obtains the third decision gain and the fourth decision gain indicated by the entity D and the entity E, updating the seventh neural network and the eighth neural network.

Specifically, the entity F updates the seventh neural network according to the objective function based on the third decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₃. And the entity F updates the eighth neural network according to the target function according to the fourth decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₃。

In one possible implementation, the seventh neural network and the eighth neural network are the same neural network.

In the channel information feedback method illustrated in fig. 10b, a terminal (entity F) compresses parameters of two paths through a neural network according to reference signals of the same group of network devices (entity D, E) defined by the CoMP technology, and feeds the parameters back to the two network devices, the two network devices in the same group obtain path parameters between the opposite terminal network device and the terminal and compressed information of a previous round of transmission decisions through interaction, and output transmission decisions through the neural network, perform data transmission with the terminal according to the transmission decisions, and subsequently update the neural network according to decision gains of throughput, channel capacity, power consumption, and the like of the system, so that efficiency of downlink transmission from the network devices to the terminal is improved through cooperative interaction between the network devices.

In another embodiment of the present application, a method for feeding back channel information of an uplink communication scenario is provided. In an uplink receiving CoMP scene, joint receiving and cooperative scheduling are included; the joint receiving finger can form a virtual antenna array through coordination among different network devices, so that the uplink signal receiving quality is improved; cooperative scheduling refers to reducing interference by coordinating scheduling decisions among network devices. The key point of the technology is to reduce a large amount of data transmission among network devices, and similarly, the channel information feedback of the invention can be used to reduce communication overhead.

As shown in fig. 11, fig. 11 is an interaction schematic diagram of a path information feedback method provided in the embodiment of the present application, and is applicable to the application scenario shown in fig. 10 a.

S1401, the entity F sends a reference signal to the entity D, and correspondingly, the entity D receives the reference signal sent by the entity F.

In one possible implementation, the reference signal may be a channel Sounding Reference Signal (SRS).

S1402, the entity F sends a reference signal to the entity E, and the entity E receives the reference signal sent by the entity F.

In one possible implementation, the reference signal may be a sounding reference signal, SRS.

The present application does not limit the order of steps S1401 and S1402. It is possible that entity F sends a reference signal, which entity D and entity E receive separately for channel estimation.

S1403, entity D measures the path parameters of estimated path F-D.

The path parameter may be a channel matrix between entity F and entity D. Entity D estimates the channel matrix from the SRS measurements sent by entity F.

S1404, an entity E measures path parameters of the estimated path F-E.

The path parameter may be a channel matrix between entity F and entity E. Entity D estimates the channel matrix from the SRS measurements sent by entity F.

S1405, determining the CNI corresponding to the path D-E by the entity D through a thirteenth neural network according to the path parameters of the path F-D and the decision result of the last round of transmission.

The thirteenth neural network can be obtained by performing reinforcement learning training using the sample path parameters as input and the sample CNI as output. The entity D may input the measured and estimated path parameters of the path F-D and the transmission decision result of the previous round into a thirteenth neural network, and the thirteenth neural network may output the CNI of the path D-E (CNI _ DE represents the CNI of the entity D to the entity E).

Optionally, the input parameters of the thirteenth neural network further include observed parameters on the path D-E, i.e. a process that also takes into account the interaction between the entity D and the entity E.

The CNI of the path D-E may specifically be feedback information obtained by effectively compressing the measured and estimated path parameters according to the task requirements of the network device.

S1406, the entity E determines the CNI corresponding to the path E-D through the fourteenth neural network according to the path parameters of the path F-E and the decision result of the last round of transmission.

The fourteenth neural network can be obtained by performing reinforcement learning training using the sample path parameters as inputs and the sample CNI as an output. The entity E may input the measured and estimated path parameters of the path F-E and the number of transmission decision results of the previous round into the fourteenth neural network, and the fourteenth neural network may output the CNI of the path E-D (CNI _ ED represents the CNI from the entity F to the entity D).

Optionally, the input parameters of the fourteenth neural network further include observed parameters on the path E-D, i.e. a process that also takes into account the interaction between the entity E and the entity D.

The CNI of the path E-D may specifically be feedback information obtained by effectively compressing the measured and estimated path parameters for the task requirements of the network device side.

S1407, the entity D sends the CNI of the path D-E to the entity E, and correspondingly, the entity D receives the CNI of the path D-E sent by the entity F.

S1408, the entity F sends the CNI corresponding to the path E-D to the entity E, and correspondingly, the entity E receives the CNI of the path E-D sent by the entity F.

S1409, the entity D determines a transmission decision through a fifteenth neural network according to the path parameters corresponding to the path F-D and the CNI corresponding to the path E-D.

The fifteenth neural network can be obtained by performing reinforcement learning training with the sample CNI as an input and the sample transmission decision as an output. The entity D may input the CNI corresponding to the path F-D output by the thirteenth neural network and the received CNI of the path E-D into the fifteenth neural network, and the fifteenth neural network may output the transmission decision.

The transmission decision may be a decision of uplink scheduling for determining whether to transmit an uplink (uplink) scheduling grant (UL grant), in which case one of the entities D and E receives uplink data of the entity F; or determining time-frequency resources, MCS, and the like allocated in uplink scheduling, in which case, the entity D and the entity E jointly receive uplink data of the entity F.

S1410, the entity E determines a transmission decision through a sixteenth neural network according to the CNI corresponding to the path F-E and the CNI corresponding to the path D-E

The sixteenth neural network can be obtained by performing reinforcement learning training with the sample CNI as an input and the sample transmission decision as an output. The entity D may input the CNI corresponding to the path D-E output by the fourteenth neural network and the received CNI of the path D-E into the fifteenth neural network, and the fifteenth neural network may output the transmission decision.

And S1411, the entity D and the entity F perform data transmission according to the transmission decision.

For example, entity D sends a UL grant to entity F. And the entity F transmits uplink data according to the received UL grant and receives ACK or NACK fed back by the entity D. For example, Data _ FD is sent to entity D, and entity D feeds back ACK _ DF or NACK _ DF to entity F after receiving Data _ FD.

And S1412, the entity F and the entity D perform data transmission according to the transmission decision.

For example, entity E sends a UL grant to entity F. And the entity F transmits uplink data according to the received UL grant and receives ACK or NACK fed back by the entity E. For example, Data _ FE is sent to the entity E, and the entity E feeds back ACK _ EF or NACK _ EF to the entity F after receiving the Data _ FE.

S1413, the entity D obtains a corresponding fifth decision gain according to the uplink scheduling result.

In one possible implementation, the entity D obtains a corresponding fifth decision gain according to the uplink transmission performance indicator. The fifth decision benefit includes, but is not limited to, one or more of the following: throughput, channel capacity, power consumption of the system. In this case, the task of the entity D is to optimize the uplink transmission performance of the terminal through cooperation between the entity D and the entity E.

For example, let T be throughput and P be power consumption, and the path combining weight of the decision is set to λ_iI represents the selectable path and has ∑ λ _i1, the total mission decision yield R₄Can be defined as:

R₄＝∑λ_i[(T_i-P_i)]

when the decision result is transmittedλ is used to transmit UL grant_iCan be 0 or 1, i.e. lambda_i∈{0，1}；

When the transmission decision result is the joint transmission of the entity D and the entity E, the lambda is determined_iCan take values before 0 and 1, i.e. lambda_i∈(0，1)。λ_iThe scheduling method can be obtained according to the time-frequency resources and MCS allocated in the uplink scheduling determined by the transmission decision.

S1414 and the entity E obtain a corresponding sixth decision gain according to the uplink scheduling result.

In one possible implementation, the entity E obtains a corresponding sixth decision gain according to the uplink transmission performance indicator. The sixth decision benefit includes, but is not limited to, one or more of the following: throughput, channel capacity, power consumption of the system. In this case, the task of the entity E is to optimize the uplink transmission performance of the terminal through cooperation between the entity D and the entity E.

Alternatively, a sixth decision benefit R may be obtained by the formula in S1413₄。

S1415, the entity D updates the thirteenth neural network and the fifteenth neural network according to the obtained decision benefit.

Specifically, the entity D updates the thirteenth neural network and the fifteenth neural network according to the objective function according to the fifth decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₄. And will not be described in detail herein.

S1416, the entity E updates the fourteenth neural network and the sixteenth neural network according to the obtained sixth decision benefit.

Specifically, the entity E updates the fourteenth neural network and the sixteenth neural network according to the target function according to the sixth decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₄. Here tooAnd will be described in detail.

In the channel information feedback method shown in fig. 11, a terminal sends an uplink reference signal to the same group of network devices in a CoMP scenario, the network devices estimate path parameters between the network devices and the terminal according to the reference signal measurement of the terminal, and compress the path parameters through a neural network to obtain CNIs corresponding to the paths, the CNIs of each path are interacted between the network devices in the same group, each CNI outputs a transmission decision through the neural network according to the CNIs obtained and received by itself, an uplink transmission authorization is sent to the terminal according to the transmission decision result, and the neural network is updated according to decision gains such as throughput, channel capacity, power consumption, and the like of uplink transmission of the terminal, that is, the network devices and the paths of the uplink transmission are adaptively selected according to channel changes, and the uplink transmission performance of the terminal is improved.

The embodiment of the present application provides another channel information feedback method, which is applied to an uplink grant free transmission (grant free transmission) scenario. In the uplink grant-free transmission mode, the terminal does not need to request the network device to allocate resources through scheduling, and the terminal usually performs data transmission on the pre-configured resources based on a contention mode. Therefore, related parameter adjustment or information acquisition is required between the terminal and the network equipment, so that the terminal can reasonably adjust the transmission process of the terminal. Under the condition that the terminal and the network equipment are intelligent agents, the channel information feedback method can be adopted, interaction overhead is reduced, and communication efficiency is improved.

As shown in fig. 12, fig. 12 is an interaction schematic diagram of a path information feedback method provided in the embodiment of the present application, and is applicable to a scenario of uplink unlicensed transmission. Entity G in fig. 12 may be a terminal, and entity H may be a network device, which are all intelligent agents.

S1501, the entity G sends pilot frequency and/or data to the entity H, and correspondingly, the entity H receives the pilot frequency and/or data sent by the entity G.

Specifically, entity G transmits pilot and/or data on the unlicensed transmission resources.

S1502, entity H detects the received pilot.

In particular, the pilot is used for channel estimation. The pilot may also be considered as a path parameter in the above embodiments.

And S1503, determining the CNI corresponding to the path G-H by the entity H through a seventeenth neural network according to the detected pilot frequency and/or data.

A seventeenth neural network may be obtained by performing reinforcement learning training using the sample pilot and/or data as input and the sample CNI as output. Entity D may input the detected pilot, data into a seventeenth neural network, which may output the CNI of path G-H (CNI _ GH denotes the CNI of entity G to entity H).

The CNI of the path G-H may specifically be information obtained by compressing pilot frequency, data, and the like received by the entity H for the authorization-free transmission for the upload task.

S1504, the entity H sends the CNI corresponding to the path G-H to the entity G, and correspondingly, the entity G receives the CNI corresponding to the path G-H sent by the entity H.

S1505, the entity G determines the authorization-free parameter set through the eighteenth neural network according to the CNI corresponding to the path G-H.

The set of parameters includes, but is not limited to, MCS, pilot, time-frequency resources, etc.

And S1506, the entity G performs authorization-free data transmission according to the parameters in the authorization-free parameter set.

Specifically, the entity G sends Data _ GH to the entity H according to the parameters in the authorization-exempt parameter set, the entity H receives the Data _ GH sent by the entity G and feeds back ACK _ GH or NACK _ GH to the entity G, and the entity G receives the ACK _ GH or NACK _ GH fed back by the entity G.

S1507, the entity G obtains a corresponding seventh decision benefit according to the determined set of authorization-free parameters.

And the entity G obtains a corresponding seventh decision benefit according to the uplink transmission performance index aiming at the authorization-free parameter set determined by the eighteenth neural network. The seventh decision benefit includes, but is not limited to, throughput and latency of the system, etc.

For example, let throughput be T, delay be D, and α be the demand proportion of the corresponding delay of the task.

R₅＝T-αD

S1508, the entity G indicates the seventh decision benefit to the entity H, and correspondingly, the entity H obtains the seventh decision benefit indicated by the entity G.

And S1509, the entity G updates the eighteenth neural network according to the seventh decision yield.

Specifically, the entity G updates the eighteenth neural network according to the target function according to the seventh decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₅. And will not be described in detail herein.

S1510, the entity H updates the seventeenth neural network according to the seventh decision gain.

Specifically, the entity H updates the seventeenth neural network according to the objective function according to the seventh decision gain. The form of the objective function may be any of the various possible forms of the first objective function or the second objective function referred to above. R (s, a) in the first objective function and the second objective function can be replaced by the R according to actual task requirements₅. And will not be described in detail herein.

The method for feeding back channel information in different application scenarios is described above, and may be extended to a communication interaction scenario in which, except for a terminal and a network device in an MIMO system, both the transmitting and receiving ends have an agent.

In a possible implementation, the channel information feedback method provided in the present application is applicable to generating a precoding matrix (non-orthogonal codebook) of Massive MIMO: the terminal intelligent agent inputs the estimated channel matrix into the neural network, outputs corresponding CNI information and feeds back the CNI to the network equipment, the intelligent agent at the network equipment receives the CNI information as the input state of the neural network, outputs a corresponding precoding matrix (non-orthogonal codebook), and uses the obtained channel capacity as reward for updating the network equipment and the intelligent agent of the terminal.

In another possible implementation, the channel information feedback method provided by the present application is applicable to a time-frequency resource scheduling scenario of Massive MIMO: the terminal intelligent agent inputs the estimated channel matrix and statistics such as accumulated throughput and time delay into the neural network, outputs corresponding CNI and feeds back the CNI to the network equipment; the intelligent agent at the network equipment collects the CNI information fed back by the terminals as the input of the neural network, outputs the scheduling decision, and uses the obtained combination of fairness, throughput and the like as rewards to update the intelligent agent of the network equipment and the terminals.

In another possible implementation, the channel information feedback method provided by the present application is also applicable to an interaction scenario between Massive MIMO terminals: for example, in federal learning, uplink transmission scenarios of multiple terminals to the same network device. The specific implementation is similar to that described in fig. 11, and is not described herein again.

The above-mentioned scheme provided by the embodiment of the present application is introduced mainly from the perspective of interaction between network elements. It will be understood that each network element, such as a terminal or network device, for implementing the above-described functions, includes corresponding hardware structures and/or software modules for performing the respective functions.

Such as: the embodiment of the application also can provide a communication device which can be applied to the terminal. Fig. 12 shows a schematic structural diagram of a communication device provided in an embodiment of the present application. As shown in fig. 12, the communication apparatus may include: a receiving unit 901, configured to receive a channel state information reference signal sent by a network device through a downlink channel. A measuring unit 902, configured to measure and estimate a channel matrix of a downlink channel according to the channel state information reference signal. A determining unit 903, configured to determine, according to a channel matrix of a downlink channel, channel negotiation information corresponding to the downlink channel through a first neural network; the parameters of the first neural network are related to a historical modulation and coding strategy. A sending unit 904, configured to send channel negotiation information corresponding to the downlink channel to the network device, where the channel negotiation information corresponding to the downlink channel is used by the network device to determine a modulation and coding strategy corresponding to the downlink channel.

In one possible design, the parameter of the first neural network is related to a historical modulation and coding strategy, including: the parameter of the first neural network is the parameter updated by the determining unit 903 according to the historical decision benefit, and the historical decision benefit is obtained by the network device according to the historical modulation and coding strategy.

In a possible design, the receiving unit 901 is further configured to receive a decision benefit sent by the network device, where the decision benefit is obtained by the network device according to a modulation and coding strategy corresponding to a downlink channel; the determining unit 903 is further configured to update the parameter of the first neural network according to the decision benefit.

Optionally, the determining unit 903 may be specifically configured to update a parameter of the first neural network according to the decision benefit and the first objective function; the first objective function is related to the decision gain.

In another possible design, the receiving unit 901 is further configured to receive an error of a parameter of a first hidden layer of a second neural network sent by a network device; and the error of the parameter of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameter of the second neural network according to the decision benefit, and the decision benefit is obtained by the network equipment according to a modulation and coding strategy corresponding to a downlink channel. The determining unit 903 is further configured to update the parameter of the first neural network according to the error of the parameter of the first layer hidden layer of the second neural network.

Optionally, the determining unit 903 may be specifically configured to update the parameter of the first neural network according to a back propagation algorithm according to the error of the parameter of the first hidden layer of the second neural network. And the error of the parameters of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameters of the second neural network according to the second objective function according to the decision gain, and the second objective function is related to the decision gain.

In a possible design, the determining unit 903 is further configured to obtain a reduced channel matrix according to channel negotiation information corresponding to a downlink channel; and updating the parameters of the first neural network according to the third objective function according to the channel matrix of the downlink channel estimated by measurement and the restored channel matrix. And the third objective function is used for indicating the error between the channel matrix of the downlink channel estimated by the minimum measurement and the restored channel matrix.

An embodiment of the present application further provides a communication apparatus applied to a network device, and fig. 13 shows another schematic structural diagram of the communication apparatus provided in the embodiment of the present application. As shown in fig. 13, the communication apparatus may include: a sending unit 1001, configured to send a channel state information reference signal to a terminal through a downlink channel. A receiving unit 1002, configured to receive channel negotiation information corresponding to a downlink channel sent by a terminal according to a channel state information reference signal; the channel negotiation information corresponding to the downlink channel is determined by the first neural network according to the channel matrix of the downlink channel after the terminal measures and estimates the channel matrix of the downlink channel according to the channel state information reference signal; the parameters of the first neural network are related to a historical modulation and coding strategy. A determining unit 1003, configured to determine, according to the channel negotiation information corresponding to the downlink channel, a modulation and coding strategy corresponding to the downlink channel.

In a possible design, the determining unit 1003 may be specifically configured to determine, according to channel negotiation information corresponding to a downlink channel, a modulation and coding strategy corresponding to the downlink channel through the second neural network; the parameters of the second neural network are related to the historical modulation and coding strategy.

Optionally, the parameters of the second neural network are related to a historical modulation and coding strategy, including: the parameter of the second neural network is the parameter updated by the determining unit 1003 according to the historical decision benefit, and the historical decision benefit is obtained by the determining unit 1003 according to the historical modulation and coding strategy.

In a possible design, the determining unit 1003 is further configured to obtain a decision gain corresponding to the modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel. The sending unit 1001 is further configured to send the decision benefit to the terminal; the decision gain is used for updating the parameters of the first neural network by the terminal.

In a possible design, the determining unit 1003 is further configured to obtain a decision gain corresponding to the modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel; and updating the parameters of the second neural network according to the decision gain.

Optionally, the determining unit 1003 may be specifically configured to update the parameter of the second neural network according to the decision benefit and according to the second objective function; the second objective function is related to the decision gain.

In another possible design, the determining unit 1003 is further configured to obtain an error of the parameter of the first hidden layer of the second neural network before and after the updating of the parameter of the second neural network. The sending unit 1001 is further configured to send an error of a parameter of the first layer hidden layer of the second neural network to the terminal; and the error of the parameters of the first hidden layer of the second neural network is used for updating the parameters of the first neural network by the terminal.

Optionally, an embodiment of the present application further provides a communication apparatus applied to a terminal or a network device, and fig. 14 shows another schematic structural diagram of the communication apparatus provided in the embodiment of the present application. As shown in fig. 14, the communication apparatus may include: a transceiving unit 1101 and a processing unit 1102. The transceiving unit 1101 may be used for transceiving information or for communicating with other network elements. The processing unit 1102 may be used to process data. When the apparatus is applied to a terminal, the method performed by the terminal as described in the foregoing embodiments may be implemented by the transceiving unit 1101 and the processing unit 1102. When the apparatus is applied to a network device, the method performed by the network device as described in the foregoing embodiments may be implemented by the transceiving unit 1101 and the processing unit 1102.

It should be understood that the division of the units in the above apparatus is only a division of logical functions, and the actual implementation may be wholly or partially integrated into one physical entity or may be physically separated. And the units in the device can be realized in the form of software called by the processing element; or may be implemented entirely in hardware; part of the units can also be realized in the form of software called by a processing element, and part of the units can be realized in the form of hardware.

For example, each unit may be a processing element separately set up, or may be implemented by being integrated into a chip of the apparatus, or may be stored in a memory in the form of a program, and a function of the unit may be called and executed by a processing element of the apparatus. In addition, all or part of the units can be integrated together or can be independently realized. The processing element described herein, which may also be referred to as a processor, may be an integrated circuit having signal processing capabilities. In the implementation process, the steps of the method or the units above may be implemented by integrated logic circuits of hardware in a processor element or in a form called by software through the processor element.

In one example, the units in any of the above apparatuses may be one or more integrated circuits configured to implement the above methods, such as: one or more Application Specific Integrated Circuits (ASICs), or one or more microprocessors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs), or a combination of at least two of these integrated circuit forms.

As another example, when a unit in a device may be implemented in the form of a processing element scheduler, the processing element may be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of invoking programs. As another example, these units may be integrated together and implemented in the form of a system-on-a-chip (SOC).

The above means for receiving is an interface circuit or input circuit of the device for receiving signals from other devices. For example, when the device is implemented in the form of a chip, the receiving unit is an interface circuit or an input circuit for the chip to receive signals from other chips or devices. When the communication device comprises means for transmitting, the means for transmitting is an interface circuit or an output circuit of the device for transmitting signals to other devices. For example, when the device is implemented in the form of a chip, the transmitting unit is an interface circuit or an output circuit of the chip for transmitting a signal to other chips or devices.

For example, an embodiment of the present application may also provide a communication apparatus applied to a terminal, where the communication apparatus may include: a processor and an interface circuit, the processor being configured to communicate with other devices via the interface circuit and to perform the steps performed by the terminal in the above method. The processor may include one or more.

Alternatively, an embodiment of the present application may further provide a communication apparatus applied to a network device, where the communication apparatus may also include: a processor and an interface circuit, the processor being configured to communicate with other devices through the interface circuit and to perform the steps performed by the network apparatus in the above method. The processor may also include one or more.

In one implementation, the unit for the terminal or the network device to respectively implement the corresponding steps in the above method may be implemented in the form of a processing element scheduler. For example, the apparatus for a terminal may include a processing element and a storage element, and the processing element calls a program stored in the storage element to execute the method executed by the terminal in the above method embodiment. Alternatively, the apparatus for a network device may also include a processing element and a storage element, and the processing element calls a program stored in the storage element to execute the method executed by the network device in the above method embodiment. The memory elements may be memory elements on the same chip as the processing elements, i.e. on-chip memory elements.

In another implementation, the program for performing the method performed by the terminal or the network device in the above method may be a storage element on a different chip than the processing element, i.e. an off-chip storage element. At this time, the processing element calls or loads a program from the off-chip storage element onto the on-chip storage element to call and execute the method executed by the terminal or the network device in the above method embodiment.

For example, embodiments of the present application may also provide a communication apparatus, which may include a processor for executing computer instructions stored in a memory, and when the computer instructions are executed, the apparatus is caused to perform the method performed by the above terminal or network device. The memory may be located within the communication device or external to the communication device. And the processor includes one or more.

In yet another implementation, the unit of the terminal for implementing the steps of the above method may be configured as one or more processing elements, which may be disposed on the terminal, where the processing elements may be integrated circuits, for example: one or more ASICs, or one or more DSPs, or one or more FPGAs, or a combination of these types of integrated circuits. These integrated circuits may be integrated together to form a chip.

Similarly, the unit of the network device for implementing the steps in the above method may also be configured as one or more processing elements, which may be disposed on the network device, and the processing elements herein may also be integrated circuits, for example: one or more ASICs, or one or more DSPs, or one or more FPGAs, or a combination of these types of integrated circuits. These integrated circuits may be integrated together to form a chip.

The units of the terminal or the network device for implementing the steps of the above method can be integrated together and implemented in the form of an SOC chip for implementing the corresponding method. At least one processing element and a storage element can be integrated in the chip, and the processing element calls the form of a stored program of the storage element to realize a corresponding method; alternatively, at least one integrated circuit may be integrated within the chip for implementing a corresponding method; alternatively, the above implementation modes may be combined, the functions of the partial units are implemented in the form of a processing element calling program, and the functions of the partial units are implemented in the form of an integrated circuit.

The processing elements herein, like those described above, may be a general purpose processor, such as a CPU, or one or more integrated circuits configured to implement the above methods, such as: one or more ASICs, or one or more microprocessors DSP, or one or more FPGAs, etc., or a combination of at least two of these integrated circuit forms.

The storage element may be a memory or a combination of a plurality of storage elements.

Through the above description of the embodiments, it is clear to those skilled in the art that, for convenience and simplicity of description, the foregoing division of the functional modules is merely used as an example, and in practical applications, the above function distribution may be completed by different functional modules according to needs, that is, the internal structure of the device may be divided into different functional modules to complete all or part of the above described functions.

In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described device embodiments are merely illustrative, and for example, the division of the modules or units is only one logical functional division, and there may be other divisions when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another device, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may be one physical unit or a plurality of physical units, that is, may be located in one place, or may be distributed in a plurality of different places. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application may be embodied in the form of software products, such as: and (5) programming. The software product is stored in a program product, such as a computer readable storage medium, and includes several instructions for causing a device (which may be a single chip, a chip, or the like) or a processor (processor) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

For example, embodiments of the present application may also provide a computer-readable storage medium, including: computer software instructions; the computer software instructions, when run in the terminal or in a chip built into the terminal, may cause the terminal to perform the method performed by the terminal as described in the previous embodiments. Alternatively, the computer software instructions, when executed in a network device or a chip built in the network device, cause the network device to perform the method performed by the network device as described in the foregoing embodiments.

The above description is only an embodiment of the present application, but the scope of the present application is not limited thereto, and any changes or substitutions within the technical scope of the present disclosure should be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A channel information feedback method, the method comprising:

a terminal receives a channel state information reference signal sent by network equipment through a downlink channel;

the terminal measures and estimates a channel matrix of the downlink channel according to the channel state information reference signal;

the terminal determines channel negotiation information corresponding to the downlink channel through a first neural network according to the channel matrix of the downlink channel; the parameters of the first neural network are related to historical modulation and coding strategies;

and the terminal sends channel negotiation information corresponding to the downlink channel to the network equipment, wherein the channel negotiation information corresponding to the downlink channel is used for the network equipment to determine a modulation and coding strategy corresponding to the downlink channel.

2. The method of claim 1, wherein the parameter of the first neural network is related to a historical modulation and coding strategy, comprising:

the parameters of the first neural network are parameters updated by the terminal according to historical decision making profits, and the historical decision making profits are obtained by the network equipment according to the historical modulation and coding strategy.

3. The method according to claim 1 or 2, wherein the modulation and coding strategy corresponding to the downlink channel is determined by the network device through a second neural network according to the channel negotiation information corresponding to the downlink channel; the parameters of the second neural network are related to the historical modulation and coding strategy.

4. The method of claim 3, wherein the parameters of the second neural network are related to the historical modulation and coding strategy, comprising:

the parameters of the second neural network are parameters updated by the network equipment according to historical decision benefits, and the historical decision benefits are obtained by the network equipment according to the historical modulation and coding strategy.

5. The method according to any one of claims 1-4, further comprising:

the terminal receives decision benefits sent by the network equipment, wherein the decision benefits are obtained by the network equipment according to a modulation and coding strategy corresponding to the downlink channel;

and the terminal updates the parameters of the first neural network according to the decision benefit.

6. The method of claim 5, wherein the terminal updates the parameters of the first neural network according to the decision gain, comprising:

the terminal updates the parameters of the first neural network according to the decision making income and a first objective function; the first objective function is related to the decision gain.

7. The method according to claim 3 or 4, characterized in that the method further comprises:

the terminal receives the error of the parameter of the first layer hidden layer of the second neural network sent by the network equipment; the error of the parameter of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameter of the second neural network according to the decision benefit, and the decision benefit is obtained by the network equipment according to the modulation and coding strategy corresponding to the downlink channel;

and the terminal updates the parameters of the first neural network according to the errors of the parameters of the first hidden layer of the second neural network.

8. The method of claim 7, wherein the terminal updates the parameters of the first neural network according to the error of the parameters of the first layer hidden layer of the second neural network, and comprises:

the terminal updates the parameters of the first neural network according to the back propagation algorithm and the errors of the parameters of the first hidden layer of the second neural network;

and the error of the parameter of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameter of the second neural network according to a second objective function according to the decision gain, wherein the second objective function is related to the decision gain.

9. The method according to any of claims 5-8, wherein the decision gain is any one of a decision code rate or a throughput when the network device performs scheduling according to a modulation and coding strategy corresponding to the downlink channel; or,

the network device combines any one of decision code rate or throughput when scheduling according to the modulation and coding strategy corresponding to the downlink channel with the evaluation value of the channel negotiation information corresponding to the downlink channel; the evaluation value of the channel negotiation information corresponding to the downlink channel by the network device is used to indicate the magnitude of the guiding effect of the channel negotiation information corresponding to the downlink channel on the modulation and coding strategy determined by the network device corresponding to the downlink channel.

10. The method of any one of claims 5-9, wherein prior to the terminal updating the parameters of the first neural network, the method further comprises:

the terminal acquires a restored channel matrix according to the channel negotiation information corresponding to the downlink channel;

the terminal updates the parameters of the first neural network according to a third objective function according to the channel matrix of the downlink channel estimated by measurement and the restored channel matrix;

the third objective function is used for indicating the error between the channel matrix of the downlink channel of the minimum measurement estimation and the restored channel matrix.

11. A channel information feedback method, the method comprising:

the network equipment sends a channel state information reference signal to the terminal through a downlink channel;

the network equipment receives channel negotiation information corresponding to the downlink channel, which is sent by the terminal according to the channel state information reference signal; the channel negotiation information corresponding to the downlink channel is determined by the terminal through a first neural network according to the channel matrix of the downlink channel after measuring and estimating the channel matrix of the downlink channel according to the channel state information reference signal; the parameters of the first neural network are related to historical modulation and coding strategies;

and the network equipment determines a modulation and coding strategy corresponding to the downlink channel according to the channel negotiation information corresponding to the downlink channel.

12. The method of claim 11, wherein the parameter of the first neural network is related to a historical modulation and coding strategy, comprising:

13. The method according to claim 11 or 12, wherein the determining, by the network device, the modulation and coding strategy corresponding to the downlink channel according to the channel negotiation information corresponding to the downlink channel includes:

the network equipment determines a modulation and coding strategy corresponding to the downlink channel through a second neural network according to the channel negotiation information corresponding to the downlink channel; the parameters of the second neural network are related to the historical modulation and coding strategy.

14. The method of claim 13, wherein the parameters of the second neural network are related to the historical modulation and coding strategy, comprising:

15. The method according to any one of claims 11-14, further comprising:

the network equipment acquires a decision gain corresponding to the modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel;

the network equipment sends the decision benefit to the terminal; the decision gain is used for updating the parameters of the first neural network by the terminal.

16. The method according to claim 13 or 14, characterized in that the method further comprises:

and the network equipment updates the parameters of the second neural network according to the decision benefit.

17. The method of claim 16, wherein the network device updates parameters of the second neural network based on the decision gain, comprising:

the network equipment updates the parameters of the second neural network according to the decision gain and a second objective function; the second objective function is related to the decision gain.

18. The method according to claim 16 or 17, further comprising:

the network equipment acquires errors of parameters of a first layer hidden layer of the second neural network before and after updating the parameters of the second neural network;

the network equipment sends the error of the parameters of the first layer hidden layer of the second neural network to the terminal; and the error of the parameters of the first hidden layer of the second neural network is used for updating the parameters of the first neural network by the terminal.

19. The method according to any of claims 15-18, wherein the decision gain is any one of a decision code rate or a throughput when the network device performs scheduling according to a modulation and coding strategy corresponding to the downlink channel; or,

and the network device combines any one of a decision code rate and a throughput when scheduling according to the modulation and coding strategy corresponding to the downlink channel with an evaluation value of the channel negotiation information corresponding to the downlink channel by the network device, wherein the evaluation value of the channel negotiation information corresponding to the downlink channel by the network device is used for indicating the magnitude of the guiding effect of the channel negotiation information corresponding to the downlink channel on the network device for determining the modulation and coding strategy corresponding to the downlink channel.

20. A communications apparatus, comprising:

a receiving unit, configured to receive a channel state information reference signal sent by a network device through a downlink channel;

a measuring unit, configured to measure and estimate a channel matrix of the downlink channel according to the channel state information reference signal;

a determining unit, configured to determine, according to the channel matrix of the downlink channel, channel negotiation information corresponding to the downlink channel through a first neural network; the parameters of the first neural network are related to historical modulation and coding strategies;

a sending unit, configured to send, to the network device, channel negotiation information corresponding to the downlink channel, where the channel negotiation information corresponding to the downlink channel is used for the network device to determine a modulation and coding strategy corresponding to the downlink channel.

21. The apparatus of claim 20, wherein the parameters of the first neural network are related to a historical modulation and coding strategy, comprising:

the parameter of the first neural network is the parameter updated by the determining unit according to the historical decision making benefit, and the historical decision making benefit is obtained by the network equipment according to the historical modulation and coding strategy.

22. The apparatus according to claim 20 or 21, wherein the modulation and coding strategy corresponding to the downlink channel is determined by the network device through a second neural network according to the channel negotiation information corresponding to the downlink channel; the parameters of the second neural network are related to the historical modulation and coding strategy.

23. The apparatus of claim 22, wherein the parameters of the second neural network are related to the historical modulation and coding strategy, comprising:

24. The apparatus according to any one of claims 20 to 23, wherein the receiving unit is further configured to receive a decision benefit sent by the network device, where the decision benefit is obtained by the network device according to a modulation and coding policy corresponding to the downlink channel;

the determining unit is further configured to update a parameter of the first neural network according to the decision gain.

25. The apparatus according to claim 24, wherein the determining unit is specifically configured to update parameters of the first neural network according to a first objective function based on the decision gain; the first objective function is related to the decision gain.

26. The apparatus according to claim 22 or 23, wherein the receiving unit is further configured to receive an error of the parameter of the first layer hidden layer of the second neural network sent by the network device; the error of the parameter of the first hidden layer of the second neural network is the error of the network equipment before and after updating the parameter of the second neural network according to the decision benefit, and the decision benefit is obtained by the network equipment according to the modulation and coding strategy corresponding to the downlink channel;

the determining unit is further configured to update the parameter of the first neural network according to an error of the parameter of the first hidden layer of the second neural network.

27. The apparatus according to claim 26, wherein the determining unit is specifically configured to update the parameters of the first neural network according to a back propagation algorithm based on an error of the parameters of the first layer hidden layer of the second neural network;

28. The apparatus according to any of claims 24-27, wherein the decision gain is any one of a decision code rate or a throughput when the network device performs scheduling according to a modulation and coding strategy corresponding to the downlink channel; or,

29. The apparatus according to any one of claims 24 to 28, wherein the determining unit is further configured to obtain a restored channel matrix according to channel negotiation information corresponding to the downlink channel; updating the parameters of the first neural network according to a third objective function according to the channel matrix of the downlink channel estimated by measurement and the restored channel matrix;

30. A communications apparatus, comprising:

a sending unit, configured to send a channel state information reference signal to a terminal through a downlink channel;

a receiving unit, configured to receive channel negotiation information corresponding to the downlink channel, where the channel negotiation information is sent by the terminal according to the channel state information reference signal; the channel negotiation information corresponding to the downlink channel is determined by the terminal through a first neural network according to the channel matrix of the downlink channel after measuring and estimating the channel matrix of the downlink channel according to the channel state information reference signal; the parameters of the first neural network are related to historical modulation and coding strategies;

and the determining unit is used for determining the modulation and coding strategy corresponding to the downlink channel according to the channel negotiation information corresponding to the downlink channel.

31. The apparatus of claim 30, wherein the parameters of the first neural network are related to the historical modulation and coding strategy, comprising:

the parameters of the first neural network are parameters updated by the terminal according to historical decision benefits, and the historical decision benefits are obtained by the determining unit according to the historical modulation and coding strategy.

32. The apparatus according to claim 30 or 31, wherein the determining unit is specifically configured to determine, through a second neural network, a modulation and coding strategy corresponding to the downlink channel according to the channel negotiation information corresponding to the downlink channel; the parameters of the second neural network are related to the historical modulation and coding strategy.

33. The apparatus of claim 32, wherein the parameters of the second neural network are related to the historical modulation and coding strategy, comprising:

the parameters of the second neural network are parameters updated by the determining unit according to historical decision benefits, and the historical decision benefits are obtained by the determining unit according to the historical modulation and coding strategy.

34. The apparatus according to any one of claims 30 to 33, wherein the determining unit is further configured to obtain a decision gain corresponding to a modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel;

the sending unit is further configured to send the decision benefit to the terminal; the decision gain is used for updating the parameters of the first neural network by the terminal.

35. The apparatus according to claim 32 or 33, wherein the determining unit is further configured to obtain a decision gain corresponding to a modulation and coding strategy according to the modulation and coding strategy corresponding to the downlink channel; and updating the parameters of the second neural network according to the decision gain.

36. The apparatus according to claim 35, wherein the determining unit is specifically configured to update the parameters of the second neural network according to a second objective function based on the decision gain; the second objective function is related to the decision gain.

37. The apparatus according to claim 35 or 36, wherein the determining unit is further configured to obtain an error of the parameter of the first layer hidden layer of the second neural network before and after updating the parameter of the second neural network;

the sending unit is further configured to send an error of a parameter of a first layer hidden layer of the second neural network to the terminal; and the error of the parameters of the first hidden layer of the second neural network is used for updating the parameters of the first neural network by the terminal.

38. The apparatus according to any of claims 34-37, wherein the decision gain is any one of a decision code rate or a throughput of the communication apparatus when performing scheduling according to a modulation and coding strategy corresponding to the downlink channel; or,

the communication device combines any one of a decision code rate and a throughput when scheduling according to the modulation and coding strategy corresponding to the downlink channel with an evaluation value of the channel negotiation information corresponding to the downlink channel by the communication device, wherein the evaluation value of the channel negotiation information corresponding to the downlink channel by the communication device is used for indicating the magnitude of a guiding effect of the channel negotiation information corresponding to the downlink channel on the communication device for determining the modulation and coding strategy corresponding to the downlink channel.

39. A communications apparatus, comprising: a processor for executing computer instructions stored in a memory, which when executed, cause the apparatus to perform the method of any of claims 1-10.

40. A communications apparatus, comprising: a processor for executing computer instructions stored in a memory, which when executed, cause the apparatus to perform the method of any of claims 11-19.

41. A computer-readable storage medium, comprising: computer software instructions;

when the computer software instructions are run on a processor,

the method of any one of claims 1-10 being performed; or,

the method of any of claims 11-19 is performed.