Developing A Generic Software-Defined Radar Transmitter Using GNU Radio
Developing A Generic Software-Defined Radar Transmitter Using GNU Radio
Developing A Generic Software-Defined Radar Transmitter Using GNU Radio
by
November 2012
This work contains no material which has been accepted for the award of any other degree
or diploma in any university or other tertiary institution and to the best of my knowledge
and belief, contains no material previously published or written by another person, except
where due reference has been made in the text.
I give consent to this copy of my thesis, when deposited in the University Library, being
available for loan and photocopying.
Research into the development of software defined radars (SDRs) often combines the GNU
Radio software toolkit, with the Universal Software Radio Peripheral (USRP) hardware
platform.
Studies have already demonstrated that these tools can be combined to develop and
implement versatile, low-cost, SDR systems. These studies focus on the question as to
whether or not a GNU Radio and USRP based SDR can address a specific set of
requirements for a particular radar application; but do not explore the characteristic
behaviour of the technology.
This thesis examines how effectively GNU Radio and the USRP can be combined to create
a software-defined radar transmitter. A SDR transmitter has been developed using these
tools as a subject for experimentation and implemented to produce a set of generic radar
waveforms at a frequency of 5.8GHz. This set consists of continuous wave, 1 μs pulsed
waveforms and frequency modulated continuous waveforms with sweep ranges from 0.5
to 25MHz.
GNU Radio and the USRP can be combined to effectively produce a generic radar
transmitter, however some imperfections such as intermodulation products and poor local
oscillator suppression may be unacceptable for some radar transmission applications.
Acknowledgements
I would like to express my gratitude to all whose support has made this thesis possible.
Firstly, thanks to my supervisors Dr. Said Al-Sarawi and Dr. Bevan Bates for all their
guidance and input over the course of the year. Thanks to Brian Reid for arranging
funding to make this project possible. To the staff of Electronic Warfare & Radar Division
(EWRD) who allowed me to borrow their equipment, lab space, and gave up their time to
assist with my queries; particularly Dr. Rohit Naik, Marcus Varcoe and Chris Pitcher I owe
thanks to you all for your help and support!
Finally, thanks goes to Aleksandra Golat for her love and patience throughout this year,
and to the ‘Midnight Study Sessions at the Hub’ group who provided motivation and
energy at hours where there was none.
Contents
1. INTRODUCTION ............................................................................................................. 14
1.1 Thesis Problem Statement..................................................................................... 14
1.2 Thesis Outline ......................................................................................................... 15
1.3 Background .............................................................................................................. 16
1.3.1 Software Defined Radio........................................................................ 16
1.3.2 GNU Radio ............................................................................................. 18
1.3.3 Introduction to the Universal Software Radio Peripheral ............... 19
1.4 Literature Review .................................................................................................... 20
1.4.1 Previous Work ....................................................................................... 20
1.4.2 Existing Documentation on System Behaviour ................................. 20
1.4.3 Primary Factors Limiting USRP Radar Performance ....................... 21
5. CONCLUSIONS.............................................................................................................. 117
Research into the development of software defined radars often combines the GNU Radio
software toolkit, with the Universal Software Radio Peripheral (USRP) hardware platform
[1].
Studies have already demonstrated that these tools can be combined to develop and
implement versatile, low-cost, software defined radar systems [2-7]. These studies focus on
the question as to whether or not a GNU Radio and USRP based SDR can address a
specific set of requirements for a particular radar application, but do not explore the
characteristic behaviour of the technology.
The purpose of this thesis is to determine how effectively GNU Radio and the USRP can
be combined to create a software-defined radar transmitter. This will be achieved by
developing a SDR transmitter, examining its characteristic behaviour to identify
performance limitations, and verifying the accuracy of output waveforms.
A set of generic waveforms are defined in this study. The accuracy with which the SDR
can produce these waveforms will serve as the measure of how effectively GNU Radio and
the USRP can be combined to create a SDR transmitter.
14
1.2 Thesis Outline
The thesis problem was addressed by first combining GNU Radio and the USRP to
implement a software-defined radar transmitter that can produce basic test waveforms.
Characterisation tests were then conducted to determine the performance limitations of
the system. More specifically, to understand exactly ‘what’ is being transmitted from the
device and under what operating conditions the transmitted output starts to differ from
what the radar designer is expecting.
Next a set of target radar waveforms with signal parameters representative of common
radar signals was defined. These target waveforms were then produced using the USRP
transmitter, which were then measured and assessed to determine the accuracy of the
output waveforms, hence answering how effectively these tools can be combined to
develop a SDR transmitter.
Section 2 details the design and hardware selection of the generic radar
transmitter, and the configurable variables that impact the system output.
Section 3 defines the experimental setup, test methodology and parameters of the
waveforms to be used during characterisation tests and waveform verification
tests.
Section 4 discusses the experimental testing and results. The characterisation tests
conducted are discussed along with comparisons between the expected results and
measured results. The waveform accuracy tests follow where the measurements of
the output waveforms are compared to modelled data to verify their accuracy.
15
1.3 Background
Software defined radar is an application of the same technology and concepts used for
software-defined radio, thus we begin with a discussion on software-defined radio.
A software-defined radio is a radio system that performs some or all signal processing
using software, ideally operating on a reprogrammable processor, such as a personal
computer (PC) or embedded system [8, p.1]
Moving the challenge of radio engineering design from the hardware domain to the
software domain provides several key advantages [8, p.2] [10, p.1]:
Reduced System Costs: A combination of the above two factors result in simplified
architectures, and reduces system costs.
ADC Sampling Rates: Sampling rates should ideally be twice the maximum
frequency of the signal to be digitized, as required by the Nyquist theorem [11,
p.40]. Since many RF applications use frequencies in the GHz region, the required
sampling rates are not achievable by most (COTS) analogue-to-digital converters
(ADCs).
Processor Speed: COTS processors do not have the processing speed to perform
real-time processing of signals with GHz frequencies.
16
To address these limitations, current software defined radios adopt a design similar to that
of the generic software defined radio presented in Figure 1.
This design includes a device called a RF front-end which mixes signals between the
carrier RF and lower frequencies suitable for the sampling rates of the ADC / digital-to-
analogue-converter (DAC) components. These lower frequencies may either be an
intermediate frequency (IF) or baseband (BB).
As a result the ADC / DAC components are only required to achieve sampling rates
sufficient to handle the modulation bandwidth (MBW) of the signal rather than the carrier
frequency bandwidth.
This subsystem also digitally down/up converts the signal between IF and BB (unless
direct conversion to baseband is performed by the RF-Front End.) The digital down/up
conversion may be performed by the processor itself or by separate digital-down-
converters (DDCs) or digital-up-converters (DUCs).
17
1.3.2 GNU Radio
GNU Radio consists of a toolkit of signal processing blocks for implementing software-
defined radios on external RF hardware. GNU Radio Companion (GRC) extends the
toolkit by providing a Graphical User Interface (GUI) that allows flow graphs and signal
visualisers1 to be built out of signal processing blocks in a manner similar to Matlab
Simulink. A screen show of GNU Radio Companion is shown in Figure 2.
GNU Radio is primarily used with the USRP hardware but is compatible with a number of
other available RF hardware devices. Although not intended as a simulation tool, it can be
used stand-alone to provide simulated results.
All codes for GNU Radio are copyright of the Free Software Foundation and distributed as
open-source [12] under the GNU General Public Licence (GPL). It provides the public
domain with a powerful, free, modifiable toolkit for software-defined radio development.
1 GNU Radio Companion offers a range of useful GUIs for visualising signal data, however the
output from the Fast Fourier Transform (FFT) visualiser is an artificially generated ‘realistic’ output
based upon a combination of real data values and modelled imperfections such a noise. Since these
artificially generated artefacts are undesired, the output from the GRC FFT visualiser was not used
in this study for data analysis.
18
Although initially created to run on Linux based operating systems (OS) for which there is
currently more installation support, installation packages have now been developed for
Windows and Mac operating systems.
The Universal Software Radio Peripheral products are a line of hardware platforms for
hosting software defined radios. Designed initially to support GNU Radio, the USRP can
now be used with other GUI control software such as Matlab and LabView, or can be run
from a computer command line. The USRP products are sold by Ettus Research and their
parent company National Instruments.
The two core Ettus Research products required to provide the functionality of a software-
defined radio are a USRP (with internal motherboard), and a RF daughterboard that
mounts onto the motherboard. The USRP (see Figure 3) then connects to a host computer
with GNU Radio (or other software), via USB 2.0 or Gigabit Ethernet cable. Alternatively,
some USRP models utilise an embedded computing device and may be run stand-alone
after instructions have been downloaded from a host computer.
In the context of the generic software defined radio discussed in Section 1.3.1, the USRP
motherboard consists of an FPGA with components to provide the ADC, DAC, DUC and
DDC functionality, whereas the RF daughterboard provides the RF-Front-End
functionality. A block diagram illustrating this is shown in Figure 4. A range of RF
daughterboards exist to address different frequency ranges, and can be easily
interchanged. Similarly, different series of USRP models exist with additional features to
meet different application requirements.
Due to the comparatively low cost of the Ettus Research hardware products [13] (e.g.
USRP motherboards are available for less than 1700 USD), the open-source specifications
of their sub components, and their synergy with the free GNU Radio software the two
platforms are often combined for the development of low cost software defined radios by
research and hobby groups [14].
19
Figure 4: Block diagram showing the main functions of a typical USRP
Numerous studies have demonstrated that GNU Radio software and USRP hardware can
be combined to develop software-defined radars for a variety of active and passive radar
applications. These applications include (but are not limited to) networked radars and
sonars [4], weather surveillance [7], aircraft or ship detection radars [15], measurement of
indoor human movement [16], SDR test beds, temperature sensing [17] etc .
These studies are often feasibility or demonstrator projects that illustrate the
implementation of a SDR design using these tools for a specific application. They measure
and evaluate the performance of the SDR for that application and may discuss the limiting
design factors; but often do not measure or explore in a more general sense - the
characteristic behaviour of the SDR, such as performance limitations or accuracy.
A wealth of open-source information on GNU Radio and the USRP is available on the
internet [12]. However this knowledge is poorly documented. In most cases the
information resides on community wikis / forums, is inconsistent due to contributions
from multiple authors, or out of date due to the rapid rate of software defined radio
development. The main sources of documentation and information include the following:
20
A limited number of guides exist that provide a detailed discussion of the internal
architecture of the USRP hardware [20-22] along with its implementation with
GNU Radio, however these apply to the original USRP 1 model and do not apply
to the hardware of the subsequent USRP models.
The GNU Radio Discussion list archives answers regarding specific questions on
the behaviour, limitations and implementation of GNU Radio and the USRP. Often
the core developers and pioneers directly answer questions relating to these tools.
These resources provide a useful starting point and resource for understanding the
internal architecture of the USRP along with data on the limitations of its subcomponents.
However these do not provide a compiled (or well documented) examination of the
behaviour and performance limitations of the USRP hardware as a whole system that can
be directly leveraged for radar design.
The limitations encountered in developing USRP based radars may vary depending upon
the application requirements, however the primary factors affecting the performance of
USRP transmitters are the computer processor speed, host connection bandwidth and the
choice of antenna as summarised in a 2010 study examining the current state of the
software-defined radar technology [8, pp.1-5] and as is apparent in many SDR design
studies. The low transmit power of the Ettus Research RF daughter-boards is also worth
addressing.
Processor Speed
A USRP based transmitter must have sufficient processor resources to maintain the
desired throughput rate from the host computer to the USRP. A failure to maintain this is
referred to as a transmit under-run (indicated by “U” outputs in the GNU Radio GUI).
The demands on the processor will vary depending upon the application, however GNU
Radio discussion list comments from Eric Blossom (founder of the GNU Radio Project)
indicates that from his experience it takes at least an Intel Core 2 Duo running at 3 GHz or
more to transmit at full speed to the latest USRP models [23]. Due to the small transmit
buffer size, the average transmit rate over a very small window needs to be reliable. If the
CPU is focussing processor resources on other tasks, transmit under-runs will occur.
This particularly applies if the radar is mono-static, in which case if the processing
required on the receiver side dominates resources it could cause transmit under-runs. In
some studies [3, pp.69-70] the high processor and resource requirements hindered the
performance of the radar system, causing difficulties in obtaining real time performance
and minimising packet losses during transmit and receive. If processor resources are
insufficient to balance both transmit and receive functions on a single host computer,
optimising processor resources may require performing transmission and receive on two
separate dedicated host computers.
21
Host Connection Bandwidth
The host connection bandwidth refers to the connection throughput between the USRP
and the host computer. Although the newer models of the USRP offer a Gigabit Ethernet
connection surpassing the USB 2.0 interface of earlier models, the host connection
bandwidth is still the core ‘bottle-neck’ limiting USRP system performance as shown in
many studies [2, 3, 8]. The host connection bandwidth not only limits the maximum
sampling rate achievable, but also the MBW of the system, and the samples per period for
a given intermediate frequency. This will be discussed further in 2.5.3.
Antennas
Software-defined radars that perform basic functions such as operation within a narrow
frequency range will be able to use most types of antenna. Multi-function software-defined
radars will require antennas that fulfil specific needs or functionality requirements, e.g.
that offer wideband or multi-band capacity, specific directional performance, or even fully
digital array functionality. The antennas available by Ettus Research are low-cost
antennas compatible with their RF daughterboard range and may not meet the frequency
band or signal-to-noise ratio (SNR) performance required of multi-function SDRs. Higher
performance antennas can be obtained via other suppliers but will likely exceed a low cost
budget.
Some designs [2] have used Barker codes to increase SNR with the drawback that
increasing the Barker code size increases the pulse width; lowering the range resolution
performance and increasing the radar dead zone. Other designs have adopted Frequency
Modulated Continuous Waveforms (FMCWs) [24, 25] which can offer high performance
using low transmit power [26].
A low transmit power may also require resource intensive processing at the receiver end
to obtain the desired SNR [27]. As discussed earlier, in the case of mono-static radars this
may in turn hinder transmit performance causing transmission under-runs.
22
2. Transmitter Design
This section addresses the SDR transmitter used in this thesis. It begins with a discussion
of the design requirements, then the hardware and software components selected in
implementing the transmitter.
Following this is a detailed system description of the combined hardware selected, along
with the configurable variables in GNU Radio that are used to influence the behaviour of
radar transmitter.
2.1 Requirements
As the purpose of this work is to examine how effectively the USRP and GNU Radio can
be combined to create a generic radar transmitter, this study is concerned with
investigating the achievable radar transmitter performance of the technology rather than
meeting a specific set of performance requirements such as carrier frequency etc. Radars
typically operate at frequencies, ranging from 1GHz potentially up to 100 GHz or more
[28, pp.83-85]. As such it is desirable to explore the highest frequency limits achievable. As
experiments will be conducted in a laboratory environment, an antenna will not be
addressed as part of this design.
Operating System: A self imposed constraint was to use a Debian based Linux
distribution, as it is a widely used and well supported in the public domain.
23
2.2 Hardware Selection
2.2.1 USRP
As identified in the literature review, the host connection bandwidth is the key bottleneck
in USRP performance. The Networked series (N210/N200) models offer the highest host
connection bandwidth out of the available USRP range as shown in Table 2.
The USRP N210 connects to a host computer using a high speed Gigabit Ethernet cable.
Although theoretically this could allow 1000 Mbps data throughput, a maximum of 800
Mbps is utilised. Signal processing functions from the host computer are loaded onto the
FPGA for faster execution. Although the Embedded series (E100/E110) allows for all
processing to occur on an embedded computing device on a standalone USRP, the data
throughput between the embedded computing device and the FPGA is limited to 40 MB/s
(e.g. 320 Mbps) and further limited by the processor which can only process 16-bit samples
at 8 MSps. As such, the Networked series provides the highest host connection bandwidth.
Additionally the Networked Series offers faster DAC sampling rates and hence a higher
dynamic range than the other models. The N210 model was selected over the N200 models
since it offers a larger FPGA then the N200 model. The model received was an N210
revision 4.0.
24
Table 2 USRP models currently available from Ettus Research [1]
NOTE:
This table has been removed due to copyright.
Alternatively, the item is available from the referenced links
25
2.2.2 RF Daughterboard
The range of RF daughterboards available at the time of writing is shown in Table 3. The
XCVR2450 model was selected so that operation at the highest possible frequency band
could be achieved.
The XCVR2450 operates within a frequency range of 2400 to 2500 MHz, and a frequency
range of 4900 to 5900 GHz. The Ettus product information is inconsistent, with some
documents claiming its high frequency band extends to 6000 MHz rather than 5900 MHz.
This daughterboard is a half-duplex transceiver, thus it cannot transmit and receive at the
same time. Since this study is only concerned with the transmit side of operation this
limitation is not an issue.
NOTE:
This table has been removed due to copyright.
Alternatively, the item is available from the referenced links
26
2.2.3 GPSDO Reference Clock
The standard onboard reference clock of the USRP N210 is specified to an accuracy of 2.5
parts per million (ppm), meaning that for a maximum clocking rate of 100 MHz2, the
frequency of the clock could theoretically deviate by up to a maximum of 250 Hz.
To achieve a higher clocking accuracy, the optional Ettus Research GPS Disciplined
Oscillator (GPSDO) was purchased. This internal reference oscillator connects to the USRP
board and provides a GPS referenced accuracy of 0.01 ppm, thus reducing the maximum
theoretical deviation to 1 Hz.
As Ettus do not provide GPS antennas for their GPSDO kit, a range of generic, low-cost
GPS antennas were investigated (see Table 4). The ROJONE Low Cost Antenna was
selected on the basis that it had low noise, and provided the largest gain and frequency
coverage when compared to the other antennas.
The host computer was loaned from DSTO, and offered the fastest processor performance
out of those available. The computer had the following performance specifications which
met the minimum recommended processor performance identified in the literature review
(Section 1.4.3).
As mentioned earlier, a self imposed constraint was to use a Debian based Linux
distribution. Ettus Research only provides firmware releases using officially supported
and maintained Linux distributions. This limited the selection to a version of Ubuntu. A
64-bit installation of Linux Ubuntu version 10.04 (Lucid Lynx) was selected on the basis
that it was an earlier release and has more support available.
At the time of selection, GNU Radio version 3.6 was the latest stable release of GNU Radio
available and supports Ubuntu 10.04.
The ‘Universal Software Radio Peripheral’ Hardware Driver (UHD) provides a host driver
and API for Ettus Research products. A variety of USRP Hardware Driver (UHD)
firmware releases are available for download from the Ettus Research website. At the time
of selection, UHD_003.004.001-23 was the latest version of the UHD driver rated as stable
by Ettus Research and is compatible with GNU Radio version 3.6, and Ubuntu 10.04.
28
NOTE:
This figure has been removed due to copyright.
Alternatively, the item is available from the referenced links
Figure 5: Block diagram of the USRP N210 with XCVR2450 daughterboard, modified from a block
diagram of the functionally similar National Instruments USRP-2921 [30]
29
2.4 System Description
This section provides a discussion of the signal transmit path through the subcomponents
of the USRP system and the signal processing steps performed [31]. A detailed block
diagram illustrating the core signal processing steps is provided in Figure 5 that should be
examined in conjunction with this description.
Although the focus of this thesis is on the behaviour of the radar transmitter, the receive
behaviour is described as well to provide a full understanding of the hardware.
The host computer running GNU Radio is used to develop the signal processing software,
which is transmitted and loaded onto the FPGA’S volatile memory prior to operation via
the Gigabit Ethernet interface. Samples for transmit are produced at baseband frequency
in a complex valued format (i.e. in phase and quadrature). This is termed ‘complex
baseband’. The user selects to use either 16-bit or 8-bit complex samples.
These are collected in a first–in first-out (FIFO) buffer, then interleaved into data packets
which are transmitted from the host computer to the FPGA via the Ethernet cable. The
FPGA collects received packets in a FIFO buffer, which are then de-interleaved and
transmitted to the AD9777 module [32].
The AD9777 module receives 16-bit complex samples across dual channels. It applies a
DUC process (see Figure 6) that involves filtering and interpolating the input to a user-
specified factor, then complex mixing the input with a complex modulator. By default the
complex modulator does not mix the signal up to an IF unless specified by the user (as the
XCVR2450 RF-daughterboard can shift signals up to the carrier RF directly from
baseband) [33]. Dual DACs convert the signal to analogue at 400 MSps, before passing it to
the MAX2829 integrated circuit (IC).
Figure 6: Block diagram of a digital up converter from the AD9777 module in the transmit path.
Selectable filters offer interpolation factors of 2, 4 or 8 [34].
30
The MAX2829 transceiver IC is the main component of the XCVR2450 RF-daughterboard.
The analogue signal received is sent through a low-pass filter, before branching off to both
the low band (2.4 to 2.5 GHz) and high band (4.9 to 6 GHz) signal paths, which are almost
identical. These paths are indicated in Figure 5 by the numbers ‘2450’ and ‘5’ respectively.
Next, the signal is mixed to the desired RF using a voltage controlled oscillator (VCO) with
a phase locked loop (PLL) [34]. As the MAX2829 is a direct conversion transceiver it is
capable of direct conversion between baseband and the desired RF without requiring an IF
stage. The signal then passes through a Monolithic Microwave Integrated Circuit (MMIC)
power amplifier.
At this point one variation in the transmit paths occurs; a signal travelling along the high
band path enters a bandpass filter, whereas a transmit signal on the low band path
doesn’t. The signal then undergoes power amplification based upon a user specified gain
value at which point both paths converge at a diplexer which determines the frequency
band used for transmission. The daughterboard may transmit the signal through either of
the two RF ports as specified by the user.
The receive path is, with a few exceptions, the reverse of the transmit path. The input
received from the RF port passes through a diplexer then branches off to the low band or
high band path as appropriate. The signal progressing along the low band path passes
through a band pass filter whereas a signal on the high band path does not.
The signal on either path enters the MAX2829 IC, where it undergoes power amplification.
It is then mixed down to a baseband frequency (or user specified IF) and passed through a
low pass filter and passed to theADS62P4X module.
The ADS62P4X ADCs digitise the analogue input at 100 MSps using 14-bit samples across
dual channels. A DDC process (see Figure 7) is then applied that involves mixing the
signal with a complex modulator down to baseband frequency if not done so already, then
decimating the signal by a user-specified factor. The complex valued signal is then passed
to the FPGA.
The FPGA collects the received samples in a FIFO buffer then interleaves them in data
blocks that are transmitted via a Gigabit Ethernet cable to the host computer. These are
then collected in a FIFO buffer on the host computer, de-interleaved and processed as
required.
31
Figure 7: Block diagram of a digital down converter from the ADS62P4X module in the transmit
path. Selectable filters offer decimation factors of 2, 4 or 8, and may function as low, high or pass
band filters.
The user can select between signed 16-bit complex samples (32-bits total) and signed 8-bit
complex samples (16-bits total). The implementation for 8-bit samples involves only taking
the 8 most significant bits of each sample. The use of 8-bit complex samples trades off
dynamic range to achieve a higher sampling rate.
Despite the USRP N210 incorporating a Gigabit Ethernet connection a maximum data
throughput of 800Mbps is used. Therefore during half-duplex operation 16-bit complex
samples and 8-bit complex samples can be transmitted at a maximum sampling rate of 25
MSps3 and 50 MSps4 respectively.
According to Nyquist criteria, the maximum frequency that avoids aliasing is equal to half
the sampling frequency . As we are using complex samples, our range of non-aliased
frequencies extends to from to centred at zero. This range is our MBW, and
limits our frequency modulation of complex samples to a maximum of from one end of
the MBW to the other.
Further consideration must be given to the desired number of samples per period of the
signal. Although a high value will maintain signal structure it will further limit the MBW
and the maximum baseband frequency achievable.
For this SDR transmitter design it was decided to maintain a minimum of four complex
samples per period. This will limit the to a useable modulation range of to /4;
which equates to an effective range of -6.25 MHz to 6.25 MHz ( = 12.5 MHz) for 16-bit
complex samples, and -12.5 MHz to 12.5 MHz ( = 25 MHz) for 8-bit complex samples.
The amplitude parameter in GNU Radio impacts both the voltage and the amplitude of
the digital signal sent to the DAC. Signal amplitude is expected to be in the range from -1.0
and +1.0. The UHD driver then normalises this into the range expected by the DAC of the
USRP. Amplitudes exceeding a magnitude of 1, will saturate the DAC causing the signal
to clip digitally. The amplitude value needs to be selected carefully since even values
lower than 1 may still cause the power output from the RF daughterboard to be
compressed. The amplitude should ideally remain at a suitably low value so the output
voltage avoids a region of non-linear output for that daughterboard [35, p7].
As the user only specifies a single gain parameter using the GRC interface (Figure 8), it is
unclear at the time of writing as to how that gain request is assigned between these two
gain sources to achieve the full 35 dB gain range available. This process became clearer
during testing.
33
2.5.6 Local Oscillator Tuning
The frequency of the daughter-board’s local oscillator (LO) can be tuned in two ways. It
may be set automatically by the USRP, where the LO is tuned as close as possible to the
target RF with the remaining difference digitally compensated for by the DUC or DDC.
Alternatively the user can manually tune the LO by specifying the frequency that the LO is
tuned to as shown in Figure 8. As this provides greater control over the frequency
selection, the distance between the LO and the RF, and the placement of the image
frequencies this method was used in all testing.
Figure 8: GNU Radio Companion GUI windows highlighting some of the key variables
The bandwidth variable adjusts the size of the baseband bandpass filter implemented by
the USRP hardware. The filter size applies to both positive and negative sides of the
complex baseband signal, thus a size of covers a range of -
to /2. The XCVR2450 offers filter sizes of 24 MHz (default), 36 MHz and 48
MHz for transmission. This variable may impact the output if it is set to a value lower than
that of the MBW (i.e. the sample rate).
34
2.6 Design Summary
This section detailed the SDR transmitter design produced using GNU Radio and the
USRP. A detailed system overview of the hardware has provided, along with an
understanding of the GNU Radio variables that control the system output. Analysis of the
system’s behaviour is the subject of the next section of this thesis, and will be addressed
there.
The total SDR transmitter design cost roughly 3000 USD5 plus shipping costs. A
successfully working system was achieved, however it is worth highlighting that
combining GNU Radio and the USRP to obtain a working system state was not a trivial
exercise. Significant trial and error was involved despite the setup guides made available
through colleagues and the GNU Radio forums. In most cases the guides were not
thorough enough to address troubleshooting, or applied to an out of date or previous
version of GNU Radio.
5Total Cost = 1700 USD (N210) + 400 USD (XCVR2450) + 750 USD (GPSDO) + 27 AUD (GPS
Antenna) + Customs Fees + Shipping Costs
35
3. Experiment Methodology
Experiments were divided up into characterisation tests and waveform verification tests.
The characterisation tests utilised three basic test waveforms to explore the response of the
USRP over a series of tests. The overall aim of this test series was to examine aspects of the
USRP’s characteristic behaviour to determine performance limitations.
The waveform verification tests involved transmitting three defined radar waveforms,
then analysing the measured results with the aim of assessing the accuracy achievable by
the SDR transmitter.
This section covers the experimental methodology used to assess the effectiveness of the
GNU Radio and USRP based SDR transmitter designed in Section 2. First, the test setup
applied throughout this thesis is discussed along with the measurement devices used to
collect data, and their configuration settings.
Secondly, the characterisation tests are discussed along with the single tone, two tone and
Gaussian noise signals used to explore the characteristic behaviour of the USRP so that
performance limits could be identified.
Finally the waveform verification tests are discussed along with the continuous waveform,
pulsed waveform, and frequency modulated continuous waveforms transmitted to assess
the accuracy of the SDR.
In each test section the GNU Radio flowgraphs used to generate the various waveforms
are provided for future use.
36
3.1 Test Setup
A block diagram of the experiment setup is shown in Figure 9, along with a photo of the
setup in Figure 10. Measurement devices consisted of a spectrum analyser, oscilloscope
and signal analyser. The spectrum analyser was used to collect data for the majority of
tests, with a few exceptions. During the Characterisation Testing, the phase noise
measurements were obtained used the signal analyser6 instead of the spectrum analyser.
During the Waveform Verification Testing the oscilloscope was as a secondary source of
data collection to support spectrum analyser measurements.
6 Due to laboratory layout, accessing the signal analyser for the phase noise measurement tests
required the use of a 10m coaxial cable running along the ceiling instead of a 1m coaxial cable.
37
3.1.1 Spectrum Analyser
3.1.2 Oscilloscope
Measurements of the waveform in the time domain were recorded using a digital
oscilloscope. The default settings were used, with adjustments summarised in Table 7,
which remained constant for all tests. The device includes an auto-measurement function
for obtaining data on the frequency and period of signals.
38
3.1.3 Signal Analyser
A digital signal analyser was used to take phase noise measurements since the available
model was equipped with a mode specifically for this function. Phase noise measurements
are referenced to the phase noise of the signal analyser. The device was set to use its
default settings for this mode, with adjustments presented in Table 8.
39
3.2 Characterisation Test Methodology
The characterisation tests were exploratory in nature and aimed to identify and investigate
limitations in the USRP’s performance. This provides an understanding of exactly ‘what’ is
being transmitted from the device and under what operating conditions the transmitted
output starts to differ from what the radar designer is expecting.
This was achieved by observing how the output and behaviour of test signals changed in
response to the adjustment of individual input variables. These tests investigated the
following characteristics and parameters by examining the power spectrum:
Three types of test waveforms were used during characterisation tests. The purpose of
these test waveforms was to define basic waveforms with well known response
characteristics, so that the impact of varying individual parameters would be easily
observable. Unlike the radar waveforms defined during waveform verification testing, the
test waveforms are tools to investigate the USRP behaviour, and are not themselves the
subject of testing.
The test waveforms are described in the following section with their standard parameters.
Variations that occur to these parameters are described in the section for that test.
Details relating to the aim, method and expected results specific to each sub-test are
discussed in the respective test section. Since the XCVR2450 encompasses both a low and
high band, tests were conducted in both bands where time was available.
40
3.2.1 Test Waveform 1: Single Tone Waveform
The single tone waveform consists of a sine wave. The parameters selected (see Table 9)
position the RF signal in the centre of the frequency band of operation. The GRC flow
graph used to generate this waveform is shown in Figure 11.
41
Figure 11: GRC flow graph for generating the single tone waveform
42
3.2.2 Test Waveform 2: Two Tone Waveform
The two tone waveform consists of two sine waves summed together and transmitted to
the USRP. This creates two tones at frequencies F1 and F2, along with a number of inter-
modulation (IM) products. The inter-modulation products at frequencies IM1 and IM2
shown in Figure 12 can be used to provide a measure of the system’s non-linearity, such as
the Third Order Output Intercept point (OIP3).
For simplicity during testing, the LO was positioned in the centre of the frequency band
tested such that the two tones were offset to the right. Although the placement of the LO
signals differs from that of the Single-Tone tests (where the RF signal is positioned in the
centre of the band) the resulting frequencies are deemed sufficiently close to the centre of
each band for examining the characteristic response. The parameters selected are shown in
Table 10. The GRC flow graph used to generate this waveform is shown in Figure 13.
Figure 12: Diagram of the two tone test waveform showing frequencies F1, F2 and inter-
modulation products IM1 and IM2 at the frequencies indicated.
43
Figure 13: GRC flow graph for generating the two tone waveform
44
3.2.3 Test Waveform 3: Wideband Gaussian Noise
The third test waveform consists of Gaussian noise spread across the frequency band. The
parameters selected (see Table 11) position the LO signal at the centre of the frequency
band tested, with the Gaussian noise distributed around the Local Oscillator as the centre.
This waveform is intended specifically for examining the baseband filter response. The
GRC flow graph used to generate this waveform is shown in Figure 14.
45
Figure 14: GRC flow graph for generating wideband Gaussian noise
46
3.3 Waveform Verification Test Methodology
To assess the accuracy of the waveforms output from the radar transmitter, a set of
waveforms was defined to be produced, transmitted and recorded for analysis. The set
consisted of three different waveform types commonly used by radars, with parameters
selected to reflect typical parameters for radar waveforms of that type, and are described
in this section.
Since these tests aimed to examine the accuracy of the waveforms generated, the USRP
parameters were selected such that the waveforms would not be subject to conditions
known to cause distorted behaviour or degraded performance as determined during
characterisation tests. This involved the following general settings:
The amplitude parameter was set to 0.25 and the gain request parameter to 20dB
to avoid regions of non-linear behaviour.
The sampling rate was lowered to 20 MSps to reduce the load on the processor and
reduce the likelihood of obtaining sample underrun errors, which were
occasionally observed during setup for signals requiring more complex flow
graphs and processing.
The baseband filter was set to a maximum width of 48 MHz, so that the signal
response could be observed with minimum filtering.
The RF frequency was set to 5.8 GHz. This frequency was selected since it was at
the high end of the available frequency range and is situated in a Defence RF band
[36, p.100], enabling free space tests to be conducted in future work.
Unlike the characterisation testing that focused on the limits of the USRP performance to
identify unexpected behaviour and regions where it occurred, the waveform testing
focused on the performance of the USRP under ideal conditions to examine how
accurately it can produce radar waveforms.
47
The following data output from the USRP was collected:
Time scope measurements using the oscilloscope
Power spectrum measurements using the spectrum analyser
Since GNU Radio Companion does not record data in the frequency domain, time scope
data was converted into the frequency domain using an FFT7 in Matlab, and used to
produce a model of the expected power spectrum within the MBW.
This enabled a frequency domain comparison between the power spectrum data obtained
from the spectrum analyser and a model of the expected power spectrum obtained from
the input data. The time domain data output to the oscilloscope was compared to the input
data as a secondary source of data to support spectrum analyser measurements. A chart
summarising the data collected is shown in Figure 15.
7The FFT algorithm used consists of a Matlab script released as part of the GNU Radio package.
The script is provided in Appendix A (Section 6.1) and incorporates a Kaiser Window, with a Beta
value of 5.
48
3.3.1 Radar Waveform 1: Continuous Waveform
The continuous waveform consisted of a cosine wave with the parameters provided in
Table 12. These were implemented in GRC using the flow graph provided in Figure 16.
Continuous Waveform
Parameter Value
BB Frequency 5 MHz
LO Frequency 5795 MHz
RF Frequency 5800 MHz
Sample Rate 20MSps
Baseband Filter 48 MHz
Amplitude 0.25
Gain 20 dB
49
Figure 16: GRC flow graph for generating the continuous waveform
50
3.3.2 Radar Waveform 2: Pulsed Waveform
The pulsed waveform was implemented in GRC by first creating a rectangular wave
vector of 1’s and 0’s representing the duty cycle on and off time periods of the desired
signal. The elements of this vector were then multiplied by samples from a cosine wave
source block to ‘chop’ the cosine wave into pulses of the desired sizes. The vector was set
to repeatedly transmit. The parameters are provided in Table 13. The flow graph used is
provided in Figure 17.
Pulsed Waveform
Parameter Value
BB Frequency 5 MHz
LO Frequency 5795 MHz
RF Frequency 5800 MHz
PRI 10 μs
PRF 100 kHz
Duty Cycle 10%
PW 1 μs
Sample Rate 20 MSps (16-bit I&Q)
Baseband Filter 48 MHz
Amplitude 0.25
Gain 20 dB
8 PRI = Pulse Repetition Interval, PRF = Pulse Repetition Frequency, PW = Pulse Width
51
Figure 17: GRC flow graph for generating the pulsed waveform
52
3.3.3 Radar Waveform 3: Frequency Modulated Continuous Waveform
Frequency modulation sweeps of 0.5, 1, 2, 5, 10, 12.5 and 25 MHz were investigated over a
20 μs cycle, where the sweep included both a 10 μs up sweep followed by a 10 μs down
sweep. To implement this in GRC, output from a triangle signal-source block (Figure 18)
was used to control the frequency shift required from a frequency modulation block. The
signal output from the frequency modulation block was then multiplied by the output
from a cosine wave signal source block.
Figure 18: Triangle signal output used to control the FMCW behaviour
The frequency modulation block in GRC uses a sensitivity parameter that defines the
rate at which the signal undergoes frequency modulation. This value is defined using the
following formula (as determined in the GRC source code), where refers to the sampling
frequency and refers to the total shift in frequency that must occur over the time period
which is the time required to sweep the frequency one way (i.e. 10 μs).
( )
Key modulation parameters are shown in Table 14. The full parameters for the Frequency
Modulated Continuous Waveforms tested are shown in Table 15 and Table 16. In these
tests the LO has under gone high side injection, with sweeps that keep the magnitude of
the during the sweep greater than or equal to ¼ of the sampling rate, thus maintaining
a number of samples per period equal to or greater than or equal to 4 samples.9 The flow
graph used is provided in Figure 19.
53
Table 14 Key parameters for applying various frequency modulation values
54
Figure 19: GRC flow graph for generating the FMCWs
55
4. Experimentation & Results
This section covers the experiments conducted and results obtained as part of the
characterisation testing and waveform verification testing. As the characterisation tests
consisted of a series of unique tests focusing on different aspects of the SDR transmitter’s
behaviour, test specific aims and methods are discussed in their respective subsections.
In both the characterisation and waveform verification test categories, predicted and
measured results are compared and summarised at the end of each subsection.
Sampling rate testing was performed to identify limitations in the achievable sampling
rates. It was assumed that based on the sample size, sampling rates from 0 up to 25MSps
or 50MSps would be achievable.
Tests were conducted using a single tone waveform for both 8-bit and 16-bit complex
samples. The requested sampling rate was set to zero then increased in steps10 of 1 MSps,
until a maximum sampling rate was determined.
Testing revealed there are restrictions on the sampling rates achievable by the USRP
hardware enforced by the GNU Radio software in most cases. If an unachievable sampling
rate is requested, the system outputs an error message and changes the requested
sampling rate to the nearest achievable value (Figure 20). The exception to this is when 16-
bit complex samples are used, and a sampling rate exceeding 25 MSps is implemented
(e.g. 33.33 or 50 MSps). In this case GNU Radio does not correct this sample request to the
maximum limit of 25 MSps for that sampling rate. The unachievable sampling rate is
applied regardless, causing the GNU Radio GUI to become dark grey indicating that the
software has stalled or frozen. Sample under-run flags (signified by a ‘U’) appear in GNU
Radio (Figure 20), whilst the signal output on the spectrum analyser may stop due to a
lack of signal samples. Table 17 shows the requested sampling rate and the actual
sampling rate set by GNU Radio for that sample size.
10Non-integer sample rates are possible but were not explored in these tests.
56
Table 17 Summary of results for sampling rates testing
Requested Sampling Rates and Actual Sampling Rates Received (MSps)
applicable to 16-bit complex and 8-bit complex samples sizes
Request Actual Request Actual Request Actual Request Actual
0.000001 0.195312 15 14.285714 30 42
1 1 16 31 43
2 2 17 16.66666 32 44
3 3.030303 18 33 45
4 4 19 34 46
5 5 20 35 47
20 33.333333*
6 5.882353 21 36 48
7 7.142857 22 37 49 50*
8 7.692308 23 38 50
9 9.090909 24 39 51
10 10 25 40 52
11 11.111111 26 25 41 53
12 27 54
12.5
13 28 55
14 14.285714 29 55+ etc.
*Causes sample under runs and GNU Radio to cease operation if using 16-bit complex samples
The modulation bandwidth testing aimed to confirm the behaviour of the RF signal as the
baseband frequency increases and approaches the limits of the MBW.
The MBW limits for the complex valued signal should theoretically be to , which
equates to limits of -12.5 MHz and 12.5 MHz away from the LO frequency for a 25 MSps
sample rate using 16-bit complex samples, and -25 MHz and 25 MHz for a 50 MSps sample
rate using 8-bit complex samples. As the baseband frequency increases in magnitude, and
crosses these boundaries the RF signal should undergo aliasing.
Tests used a single tone waveform with the baseband frequency initially set to 0, then
gradually increased past the theoretical limit to 25 MHz. The initial baseband filter width
was set to a maximum value of 48 MHz to ensure that any limits observed were due to the
25 MHz MBW, rather than the baseband filter. As the baseband frequency increased from
0 to 12.5 MHz the image frequency on the opposite side of LO frequency moved from 0 to
- 12.5 MHz from the LO. As the baseband frequency approached 12.5 MHz the RF signal
decreased in power whilst the image signal power increased until the baseband frequency
reached 12.5 MHz, at which point the signals were of the same power and symmetrical
around the LO frequency. As the baseband frequency increased further, the signal power
dropped off rapidly after passing 12.5 MHz, whilst an image signal of increasing power
was observed folding back from - 12.5 MHz approaching 0. Images showing these results
are presented in Figure 21, Figure 22 and Figure 23.
57
The reverse behaviour was observed for a RF signal with a baseband frequency decreasing
from 0 to -25 MHz. The tests were repeated for a 50 MSps sample rate using 8-bit samples
with the same behaviour observed at limits of -25 MHz and 25 MHz.
These results confirm the measured MBW limits for 25 MSps and 50 MSps sample rates
match the theoretical values and that the USRP signal responds as expected whilst the
baseband frequency approaches these limits.
Figure 21: Frequency response for a single tone waveform with a 7.5 MHz baseband frequency
58
Figure 22: Frequency response for a single tone waveform with a 12.5 MHz baseband frequency
Figure 23: Frequency response for a single tone waveform with a 15 MHz baseband frequency
59
4.1.3 Frequency Limit Testing
The frequency limit testing aimed to confirm the frequency limits of the XCVR2450
daughter-board. As discussed in the Transmitter Design section, the XCVR2450’s low band
frequency range is expected to cover 2400 to 2500 MHz, whilst the high band range is
expected to cover 4900 to either 5900 or 6000 MHz.
Low and high band testing began with the LO frequency of a single tone waveform placed
in the centre of that band (e.g. 2450 or 5400 MHz respectively), with a baseband frequency
of zero. The LO frequency was then gradually increased to identify the upper limit of that
band. The experiments were repeated with the LO frequency decreasing to identify the
lower band limits.
These tests show that the low band covers a frequency range of 2400 to 2500 MHz and that
the high band covers a frequency range of 4900 to 6000 MHz. Further testing involved
shifting the LO frequency to these limits, then increasing the magnitude of the baseband
frequencies. This clarified that the quoted frequency limits applied to the LO frequencies
only, and that through modulation, signals can be effectively modulated beyond these
boundaries, within MBW constraints as shown in Figure 24.
Figure 24: Frequency response for a single tone waveform modulated above 6000 MHz
60
4.1.4 Effects of the Amplitude Variable
The amplitude parameter impacts both the voltage and amplitude of signals sent to the
DAC. Amplitude tests aimed to examine the effects of its behaviour on the system, and
more specifically address the following:
How do changes in the amplitude variable impact the output power of the signal?
What amplitude values are suitable to avoid generating non-linear responses in the
signal output?
To verify that the signal undergoes distortion for amplitude values above 1, and
observe the extent of the distortion.
It was expected that since the amplitude parameter directly impacts the voltage of the
DAC, theoretically the output power should change by around -6 dB (i.e. a factor of four)
each time the amplitude value is halved11 [28, p.77].
Secondly, it was anticipated that as the amplitude parameter increased from 0 to 1, that the
peak power will increase linearly until crossing a threshold, at which point the power
response would become non-linear due to compression. Regions below this amplitude
value should provide linear behaviour.
Thirdly, as stated earlier it was expected that signals above with an amplitude value above
1 will clip digitally, although the extent of the distortion caused will need to be examined
in testing.
These tests involved taking samples of the peak power of a single tone waveform at
varying amplitudes to verify the predicted behavioural characteristics, and observe any
anomalous behaviour.
As anticipated, each time the amplitude parameter was halved, the peak power dropped
by approximately 6 dB in both the low band and high band. See Table 18 and Table 19
respectively. The exact power drop varied depending upon the RF frequency tested. The
results at 5.6 GHz (see Table 19) almost perfectly matched the -6.02 dB theoretical response
for each amplitude shift, whereas results at other frequencies showed greater fluctuation.
This suggests that the power amplifier for the high band path, maybe optimised at LO
frequencies around 5.6 GHz. No such optimal region was noticeable in the low band
during these tests.
For amplitudes closer to 1, the change in peak power compressed to about 4 or 5 dB.
Similarly, the RF noise skirt was significantly higher at these amplitudes as shown in
11 Gain (dB) = 20log(V2/V1), hence by halving the voltage the gain changes by approximately -
6.0206 dB
61
Figure 25 and Figure 26. For the high band the noise skirt had notably increased by around
10 dB as the amplitude shifted from 0.5 to 1.
Raising the amplitude above 1 to 1.0001 caused the signal to distort significantly as
predicted. The noise floor increased by around 5 to 20 dB, with harmonic noise spikes
occurring approximately every 70 kHz. Additionally the Spurious Free Dynamic Range
(SFDR) lowered from 49 dB to 33 dB for the low band, and from 51.5 dB to 35 dB for the
high band.
Based on these observations the amplitude parameter for the XCVR2450 should be kept at
a value well below 0.5 to optimise its impact on the signal to noise ratio, and to avoid non-
linear output from the RF daughterboard. For single tone signals, 0.25 appears to be a
suitable value.
Table 18 Amplitude reduction test results for a single tone waveform (low band)
RF Frequency (GHz)
Amplitude
2.40625 2.425 2.45 2.475 2.5
Reduction
Step Change in Output Power (dB) with Amplitude Reduction
1 to 0.5 -4.74 -4.69 -4.69 -3.68 -4.72
0.5 to 0.25 -5.67 -5.76 -5.67 -5.68 -5.75
0.25 to 0.125 -5.95 -5.94 -5.96 -6.01 -5.87
0.125 to 0.0625 -5.98 -5.96 -5.96 -6 -6.09
0.0625 to 0.03125 -6.08 -6.02 -6.07 -6.01 -6.02
Table 19 Amplitude reduction test results for a single tone waveform (high band)
RF Frequency (GHz)
Amplitude
4.90625 5 5.2 5.4 5.6 5.8 6
Reduction
Step Change in Output Power (dB) with Amplitude Reduction
1 to 0.5 -4.7 -4.8 -4.85 -4.81 -6.01 -4.77 -4.71
0.5 to 0.25 -6.02 -6.03 -5.94 -5.99 -6.01 -6.04 -6.03
0.25 to 0.125 -6.09 -6.03 -6.05 -6.03 -6.04 -6.05 -6.1
0.125 to 0.0625 -6.04 -6.14 -6.12 -6.04 -6.09 -6.01 -5.97
0.0625 to 0.03125 -5.98 -5.71 -5.94 -6.04 -6.02 -6.16 -5.92
62
Power vs Frequency for various Amplitudes
(Single Tone Signal, RF 2450MHz, LO 2493.75MHz)
10
A = 1.0001
0
A=1
-10 A = 0.5
-20 A = 0.25
A = 0.125
-30
Power (dBm)
A = 0.0625
-40 A = 0.03125
-50
-60
-70
-80
-90
-100
2449 2449.2 2449.4 2449.6 2449.8 2450 2450.2 2450.4 2450.6 2450.8 2451
Frequency (MHz)
Figure 25: Single tone waveform response to various amplitude values (low band)
63
Power vs Frequency for various Amplitudes
(Single Tone Signal, RF = 5400MHz, LO = 5393.75MHz)
10
A = 1.0001
0 A=1
A = 0.5
-10
A = 0.25
-20 A = 0.125
A = 0.0625
-30
Power (dBm)
A = 0.03125
-40
-50
-60
-70
-80
-90
-100
5399 5399.2 5399.4 5399.6 5399.8 5400 5400.2 5400.4 5400.6 5400.8 5401
Frequency (MHz)
Figure 26: Single tone waveform response to various amplitude values (high band)
64
Two Tone Tests
Additional tests were conducted in the centre of the low and high bands using the two
tone test waveform. As with the single tone test results the change in output power
remained at approximately -6 dB each time the amplitude value was halved, as shown in
Table 20 and Table 21.
The notable difference observed in the two-tone test signal is that the amplitude threshold
for distortion appeared to be 0.5 as shown in Figure 27 and Figure 28. As the maximum
amplitude of the two-tone test signal is the summation of the 0.5 amplitudes of the two
individual tones12 this threshold is still consistent with the expected maximum amplitude
of 1. For amplitudes of 0.51 or above the intermittent flashes of noise appeared across the
spectrum analyser raising the noise floor by approximately 25 dB. The presence of the
intermittent noise flashes appeared more prominently for increasing amplitude values
above 0.51. This is attributed to the fact the summation of the sinusoidal amplitudes
exceeds 1 on a more frequent basis for higher amplitude values. A spot check was
conducted using a summation of 3 test tones. Flashes of noise appeared for amplitude
values of 0.34 or higher, which is consistent with the theory that the flashes of noise are
caused by superposition of the amplitudes periodically exceeding the maximum of 1.
Ideally during a two tone test, the power of inter-modulation products IM1 and IM2 will
shift according to a linear 3:1 ratio, i.e. each 1dB shift in the input power should cause a 3
dB shift in power of IM1 and IM2 [37]. This linear response is expected when the power of
the IM1 and IM2 are around 10 dB above the noise floor. Thus, in this region halving the
amplitude should cause a -18 dB response13.
In the low band, the power response of IM1 and IM2 was roughly of a 3:1 ratio for
amplitude shifts between 0.5 and 0.125. At lower amplitude values the power drop
reduced below the linear 3:1 ratio. This is expected, given that at lower amplitude values,
these tones had lowered close to the noise floor (Figure 27).
In the high band, IM1 and IM2 were highly suppressed, being close to the noise floor for
amplitudes of 0.25 and beneath the noise floor at lower amplitudes (Figure 28). At the
maximum useable amplitude of 0.5, IM1 and IM2 were 30 dB lower in the high band then
they were in the low band.
As with single tone tests, observations show that the amplitude parameter should be kept
at a value well below 0.5 for two-tone signals, with 0.25 appearing to be a suitable value.
Additionally, the power amplifier in the high band path appears to generate lower inter-
modulation products than the one in the low band path.
12 Periodically the two peak signal amplitudes of 0.5 will combine to equal a peak of 1
13 As halving the amplitude should lower the input power by -6.0206 dB 3 = -18.0618 dB
65
Table 20 Amplitude reduction test results for a two tone waveform (low band)
RF Frequency (MHz)
Amplitude IM1 F1 F2 IM2
Reduction 2453 2454 2455 2456
Step Change in Output Power (dB) with Amplitude Reduction
1 to 0.5 -16.28 0.94 1.24 -16.98
0.5 to 0.25 -17.16 -5.09 -5.1 -17.83
0.25 to 0.125 -16.56 -5.8 -5.79 -20.83
0.125 to 0.0625 -13.9 -5.96 -5.96 -11.35
0.0625 to 0.03125 -3.14 -6.01 -6.01 -1.85
Table 21 Amplitude reduction test results for a two tone waveform (high band)
RF Frequency (MHz)
Amplitude IM1 F1 F2 IM2
Reduction 5403 5404 5405 5406
Step Change in Output Power (dB) with Amplitude Reduction
1 to 0.5 -39.05 7.64 7.89 -37.78
0.5 to 0.25 -12.33 -6.02 -5.51 -13.23
0.25 to 0.125 Noise Floor -5.77 -6.76 Noise Floor
0.125 to 0.0625 Noise Floor -6.1 -5.3 Noise Floor
0.0625 to 0.03125 Noise Floor -6.13 -6.53 Noise Floor
66
Output Power vs Frequency for various Amplitudes
(Two Tone Signal, F1 = 2454MHz, F2 = 2455MHz, LO = 2450MHz)
0
-5 A=1
-10 A = 0.51
-15 A = 0.5
-20 A = 0.25
-25 A = 0.125
-30 A = 0.0625
Power (dBm)
-35 A = 0.03125
-40
-45
-50
-55
-60
-65
-70
-75
-80
-85
2451 2452 2453 2454 2455 2456 2457 2458
Frequency (MHz)
Figure 27: Two tone waveform response to various amplitude values (low band)
67
Output Power vs Frequency for various Amplitudes
(Two Tone Signal, F1 = 5404MHz, F2 = 5405MHz, LO = 5400MHz)
0
-5 A=1
-10 A = 0.51
-15 A = 0.5
-20 A = 0.25
-25 A = 0.125
A = 0.0625
Power (dBm)
-30
-35 A = 0.03125
-40
-45
-50
-55
-60
-65
-70
-75
-80
5401 5402 5403 5404 5405 5406 5407 5408
Frequency (MHz)
Figure 28: Two tone waveform response to various amplitude values (high band)
68
4.1.5 Effects of the Gain Request Variable
These tests aimed to observe the impact of the gain variable on the signal output, to
observe any regions exhibiting non-linear behaviour and identify the maximum output
power achievable. A maximum output power of 0.1 Watts (20 dBm) is expected, with
compression (if any) occurring close to this region.
25
20
Gain (dB)
15
10
0
0 5 10 15 20 25 30
Gain Request (dB)
5
Gain (dB)
4
3
2
1 Baseband Gain Control (0 to 5 dB)
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Gain Request (dB)
Figure 29: Expected gain response for the two individual gain sources in the XCVR2450
69
Stepped Gain Testing
For this set of tests the defined single tone waveform was used with the amplitude
parameter set to 0.25 based upon the outcomes of the previous amplitude tests. Tests were
conducted in the centre of the low and high bands, with the gain parameter varied from -5
dB to 38 dB in 0.5 dB steps. The maximum output power was measured for the desired RF,
the LO frequency and the image frequency mirrored on the opposite side of the LO
frequency. Although the gain response may differ for different frequencies in the each
band, these tests will provide an understanding of what results to expect.
The low band gain test (Figure 30) showed a linear response that was consistent across the
three frequencies measured. A change in output power only occurred for gain parameter
inputs between 0 and 35 dB as expected, however the total output power only increased
by a maximum of about 28.5 dB across this range.
As the gain parameter input was increased, changes in the output power were only
observed every 1 dB, until reaching 22.5 dB at which point the output power responded to
0.5 dB increases, which is consistent with the stepping pattern observed for the VGA in
Figure 29. As the gain parameter increased from 30 to 35 dB the step response appeared
similar to those shown for the BB gain control (Figure 29). This shows that the gain
requests are implemented using the VGA first, with the baseband gain control used to
achieve values greater than 30 dB. The maximum output power achieved was 20 dBm and
is consistent with the quoted value for the RF daughterboard (e.g. 0.1 W).
The high band gain test produced a similarly linear response for the three measured
frequencies as the gain parameter increased to 23 dB (Figure 31). The power of the RF
signal continued to increase as the gain parameter increased further, but the response
became non-linear, approaching an asymptote limit at 20 dBm. During this region, the
power increase for the LO and the image signals plateau heavily, even lowering as the
gain parameter exceeded 30 dB. The total output power across the gain parameter range
increased by a maximum of about 31dB.
Comparing the response of the low band and high band tests highlights the following
observations. In operation the USRP gain output received may not match the gain request,
as shown in the low band testing. The maximum gain cannot be determined from this set
of tests as the output power reaches the maximum before the full gain is applied14.
Furthermore, non-linear behaviour may be observed as the output power approaches the
20 dBm (0.1 W) maximum.
14Furthertesting in Section 4.1.6 will show that the maximum gain was measured at 28dB for the
low band and 35dB for the high band.
70
Power vs Gain
(Single Tone Signal, Amplitude = 0.25)
25
20
15
10
5
0
Power (dBm)
-5
-10
-15
-20
-25
-30
-35 RF = 2450 MHz
-40 LO = 2443.75 MHz
-45 Image Freq = 2437.5 MHz
-50
-5 0 5 10 15 20 25 30 35 40
Gain Request (dB)
Figure 30: Stepped gain test results for a single tone waveform (low band)
Power vs Gain
(Single Tone Signal, Amplitude = 0.25)
20
15
10
5
0
Power (dBm)
-5
-10
-15
-20
-25
-30
-35 RF = 5400 MHz
-40 LO = 5393.75 MHz
-45 Image Freq = 5387.5 MHz
-50
-5 0 5 10 15 20 25 30 35 40
Gain Request (dB)
Figure 31: Stepped gain test results for a single tone waveform (high band)
71
Observations in the Frequency Spectrum
Data was recorded for the output of a single tone waveform across a 100 MHz span, for
various gain values in the low band and high band. See Figure 32 and Figure 33
respectively. As observed in the previous set of tests, the RF, LO and image signal power
values increase steadily as the gain increases. A large number of spurs are observed in
both the low and high band spectrums. The larger spurs spaced at multiples of the
baseband frequency (6.25 MHz) are harmonic frequencies15, and drop in power as they
increase in distance from LO frequency. The other spurs appear to be inter-modulation
products that can be seen increasing rapidly as the gain increases. More of these appear in
the results for the low band, along with much wider noise skirting and side-lobes as the
gain increases.
Distortion effects at higher gain values can be seen in the image frequencies (notably the
largest spur to the right of the each RF signal). The rate, at which these spurs increase in
size with gain, rises significantly at higher gain values suggesting that these are the result
of intermodulation. This is seen for the gain increase from 30 dB to 35 dB in the low band
results, and 25 dB to 30 dB in the high band results. This is consistent with the gain region
where non-linear effects are observed previously in Figure 30 and Figure 31.
15 , where n is an integer
72
Power vs Frequency for various Gains
(Single Tone Signal, RF = 2456.25MHz, LO = 2450MHz)
20
10 G = 35 dB
0 G = 30 dB
G = 25 dB
-10 G = 20 dB
Power (dBm)
-20 G = 15 dB
G = 10 dB
-30 G = 5 dB
-40 G = 0 dB
-50
-60
-70
-80
2400 2410 2420 2430 2440 2450 2460 2470 2480 2490 2500
Frequency (MHz)
Figure 32: Single tone waveform response to various gain values (low band)
73
Power vs Frequency for varying Gains
(Single Tone Signal, RF = 5406.25MHz, LO = 5400MHz)
20
G = 35 dB
10
G = 30 dB
0 G = 25 dB
G = 20 dB
-10 G = 15 dB
Power (dBm)
-20 G = 10 dB
G = 5 dB
-30 G = 0 dB
-40
-50
-60
-70
-80
5350 5360 5370 5380 5390 5400 5410 5420 5430 5440 5450
Frequency (MHz)
Figure 33: Single tone waveform response to various gain values (high band)
74
4.1.6 Power versus Gain Request and Amplitude Variables
The separate gain and amplitude tests showed the characteristic effects of these variables
on the USRP output in isolation. They further demonstrated that the USRP output may be
subject to non-linear behaviour if either of these variables is set too high.
This set of tests aims to observe the combined effects of these variables on the output
power. It is believed that these variables operate independently to each other, and that the
changing both of these variables will not result in any unexpected behaviour.
A single tone waveform was used in these tests with the RF signal placed at 2.45 GHz and
5.4 GHz as previously done in the individual gain and amplitude tests. This time the
power of the RF signal is measured at gain steps of 5 dB, with the amplitude parameter
progressively halved as before.
The results seen in Figure 34 and Figure 35 match the results obtained in isolated
amplitude and gain tests. Firstly, the gain increased at the same linear rate regardless of
the selected amplitude value. Secondly, each time the amplitude was halved, the output
power reduced by 6 dB. As before, this change in power compressed to values of 4 or 5 dB
for amplitudes closer to one. Thirdly, the output power did not exceed the stated 20dBm
maximum for the device and the response became non-linear when increases in the
amplitude and gain parameters caused the output to approach this limiting value. During
these tests a maximum gain was observed of 28 dB for the low band and 35 dB for the high
band.
75
Power vs Gain for various Amplitudes
(Single Tone Signal, RF = 2450MHz, LO = 2493.75MHz)
20
10
Power (dBm)
-10 A = 1.0001
A=1
-20 A = 0.5
A = 0.25
A = 0.125
-30
A = 0.0625
A = 0.03125
-40
0 5 10 15 20 25 30 35 40
Gain Request (dB)
Figure 34: Gain and amplitude test results for a single tone waveform (low band)
20
10
Power (dBm)
-10 A = 1.0001
A=1
A = 0.5
-20 A = 0.25
A = 0.125
-30 A = 0.0625
A = 0.03125
-40
0 5 10 15 20 25 30 35 40
Gain Request (dB)
Figure 35: Gain and amplitude test results for a single tone waveform (high band)
76
4.1.7 Power versus RF Frequency
These tests were designed to examine how the output power of the USRP varied as a
function of the waveform frequency in each band.
Examining the XCVR2450 schematic [38] shows that the two main components whose
performance is expected to impact the transmit power in both bands, are the MAX2829
Transceiver and the subsequent IRM406U power amplifier.16
Figure 36 shows plots of the transmit power versus frequency obtained from the
manufacturer datasheet for the MAX2829 Transceiver. These suggest that the power
output in the low band will remain relatively consistent across the band, whereas a
difference of up to 4dB is expected for the high band. The IRM406U data sheet does not
provide info of its output power of the frequency bands, but indicates that the output may
vary by 3dB between its minimum and typical values [39].
NOTE:
This figure has been removed due to copyright.
Alternatively, the item is available from the referenced links
Figure 36: Transmit power plots for the low band (left) and high band (right) from the MAX2829
Transceiver datasheet [39]
16A single IRM406U amplifies signals in both bands despite being presented as two separate
amplifiers in the Figure 5 block diagram.
77
Tests used the single tone waveform and involved shifting the placement of an RF signal
at regular intervals across each frequency band. The gain and amplitude remained at 0 and
0.25 respectively.17
Low band results (see Figure 37) show the output power increased almost linearly by 4 dB
across the band, whereas the response (at least from the MAX2829 component) was
expected to be reasonably constant. For the high band (see Figure 38), the output power
was typically consistent in the centre of the band (i.e. between the data points at 5350 and
5650 MHz). Beyond this region the power reducing linearly by about 8 or 8.5 dB at the
band limits, rather than the 4dB shift expected.
Both sets of results show approximately a 4 dB maximum deviation from what was is
indicated in Figure 36. Assuming the potential variation in the power amplifier output
accounts for 3dB, this leaves a 1 to 1.5 dB discrepancy, which is the ball park of a
reasonable deviation from what the datasheets indicate. Nonetheless, the results indicate
that other hardware components in the chain are causing losses and changes in
performance across each frequency band, beyond what the datasheets for the main
component indicate.
-7
Power (dBm)
-8
-9
-10
Low Band RF Signal
-11
2400 2410 2420 2430 2440 2450 2460 2470 2480 2490 2500
Frequency (MHz)
Figure 37: Peak power vs. frequency test results for a single tone waveform (low band)
17 Note: Measurements begin 6.25 MHz above the lower limit in each band since there is a +6.25
MHz offset between the LO and the RF, and the LO frequency cannot go below the lower band
limits.
78
Power vs Frequency for the High Band
(Single Tone Signal, Baseband Frequency = 6.25MHz, Amplitude = 0.25, Gain = 0 dB)
-10
High Band RF Signal
-12
Power (dBm)
-14
-16
-18
-20
4800 5000 5200 5400 5600 5800 6000
Frequency (MHz)
Figure 38: Peak power vs. frequency test results for a single tone waveform (high band)
This set of tests aims to further characterise the extent to which the output power of a
single tone waveform changes as a function of the baseband frequency offset. Additionally
it examines how the output is impacted by both the selection of the MBW and also the size
of the baseband filter.
The output power of the RF signal is expected to reduce the further it is offset from the LO
frequency, and to drop quickly as it approaches the boundaries of the MBW. It is unclear
as to what the shape of the power curve will be.
Note: Tests in this section focus on the signal at the desired RF. In many cases,
measurements are taken beyond the boundaries of the MBW or filter, to obtain an
understanding of the effects of these variables. It is important to highlight that in practice,
frequencies beyond these boundaries cannot be used due to the presence of strong image
signals, or folded RF / image signals that are not the focus of this test section.
79
Unfiltered18 Output Response
Initial tests were conducted in the low band with the local oscillator tuned to 2443.75 MHz
for consistency with previous testing. The baseband frequency was varied from 0 to the
MBW limits and beyond, with the signal power measured at the RF. The sample size
remained at 8-bit complex for most tests so that higher bandwidths of up to 50MSps could
be explored. The filter size was maximised to 48 MHz so that the output response could be
observed without interference from the baseband filter. An amplitude of 0.25 and a gain of
0 dB were used for all tests.
Tests were conducted for MBW values of 5, 15, 20, 25 and 33 MHz as shown in Figure 39.
The power curve for each MBW had a parabolic shape, including a noticeable reduction in
the RF signal power well before the baseband frequency had approached the boundaries
of the MBW. In most cases the shape of the parabola became wider with a greater plateau
response as the MBW increased. The exception to this applied to the 25 MHz MBW, which
resulted in a plateau response with a very sharp roll-off at the edges.
To examine the impact of the filter on the output signal, further tests were conducted
where the filter was set to sizes smaller than that of the MBW. These results are provided
in Figure 40. The impact of the 24, 36 and 48 MHz filter can be clearly seen creating a sharp
tent shaped roll-off to the plateau response from the 50 MHz MBW power curve. The 3 dB
cut off bandwidth frequencies were measured as 23, 33 and 42 MHz respectively.
Comparing the unfiltered 25 MHz MBW curve to the 24 MHz filtered version, showed
very similar responses since the filter boundaries are close to those of the MBW.
Comparing the unfiltered 33 MHz MBW curve to the 24 MHz filtered version, shows
distinctly different responses. The filtered version has a parabolic shape for frequencies
between 11 MHz, before undergoing a sharp tent shaped roll-off for frequencies beyond
these bounds. The unfiltered version of this signal, maintains its parabolic shape until
reaching 16.666 MHz, at which point it undergoes a sudden change in parabolic shaped
rolls off due to exceeding the boundaries of the MBW.
18Unfiltered, in this context implies that the filter was set to a 48 MHz size, considerably larger than
the modulation bandwidth of the signal so that it ideally does not interfere with the signal output.
80
Power vs BB Frequency from the LO Signal
(Single Tone Signal, LO = 2443.75MHz, Low Band)
0
-10
-20
Power (dBm)
-30
-40
BW=33MHz (48MHz Filter, 8b I&Q, Low)
-50 BW=25MHz (48MHz Filter, 8b I&Q, Low)
BW=20MHz (48MHz Filter, 8b I&Q, Low)
-60 BW=15MHz (48MHz Filter, 8b I&Q, Low)
BW=5MHz (24MHz Filter, 16b I&Q, Low)
-70
-25 -20 -15 -10 -5 0 5 10 15 20 25
Frequency (MHz)
Figure 39: Baseband frequency offset test results – Response of unfiltered single tone waveforms
(low band)
-10
-20
Power (dBm)
-30
BW=50MHz (48MHz Filter, 8b I&Q, Low)
-40 BW=50MHz (36MHz Filter, 8b I&Q, Low)
BW=50MHz (24MHz Filter, 8b I&Q, Low)
-50 BW=33MHz (48MHz Filter, 8b I&Q, Low)
BW=33MHz (24MHz Filter, 8b I&Q, Low)
-60 BW=25MHz (48MHz Filter, 8b I&Q, Low)
BW=25MHz (24MHz Filter, 16b I&Q, Low)
-70
-25 -20 -15 -10 -5 0 5 10 15 20 25
Frequency (MHz)
Figure 40: Baseband frequency offset test results – Response of filtered and unfiltered single tone
waveforms (low band)
81
Low vs. High Band Response
Some brief testing was performed with the LO tuned to 5393.75 MHz to verify that similar
behaviour was observed in the high band and the low band. The results are shown in
Figure 41 and Figure 42. The observed behaviour was the same in both bands except that
the output power for low band results was generally 3 dB higher than that in the high
band. The noise floor remained consistent throughout all tests remaining at around -80 dB
for high band tests, and -85 dB for low band tests.
Throughout the various tests, it was noticed that the some ripples were observed in the
pass-band of the filtered power curve. This was particularly apparent in Figure 41, and can
be observed at baseband frequency offsets of 0, where the signal occasionally peaked or
dropped by 1dB. These are believed to be side effects of the half band filter.
Wide band Gaussian noise (as previously defined) was generated across the frequency
spectrum as an alternate method of observing how the filter size shapes the output.
Results are shown in Figure 43, but are believed to be limited in applicability to noise only,
and not to tone signals. The noise power curve takes a parabolic shape and drop off
sharply at frequencies approaching the filter limits of , however the
reduction observed was much larger than 3 dB approaching these values. Another key
observation from this test is the presence of spurs observed at frequency values offset from
the LO frequency by 25 MHz, 33.33 MHz and 50 MHz, which appeared at up to 20 dB
above the noise floor.
It is unknown as to why these spurs appear, however they occur at frequency offsets twice
that of some of the MBW limits identified earlier19. As such, these noise spurs are believed
to be harmonics likely to be observed at multiples of the available MBW limits.
Additionally, these appear beyond the limits of the baseband filter, thus could potentially
interfere with out of band receivers.
19Test results shown in Table 17 earlier, identified that some of the achievable sampling rates for
the USRP include values of 25 MHz, 33.33 MHz and 50 MHz, which would generate modulation
bandwidth limits at frequency values of 12.5 MHz, 16.66 MHz and 25 MHz respectively.
82
Power vs BB Frequency from the LO Signal
(Single Tone Signal, LO = 2443.75MHz, Low Band)
-6
-9
-12
Power (dBm)
-15
-10
-20
Power (dBm)
-30
-40
BW=50MHz (24MHz Filter, 8b I&Q, High)
-50
BW=50MHz (24MHz Filter, 8b I&Q, Low)
BW=33MHz (24MHz Filter, 8b I&Q, High)
-60
BW=33MHz (24MHz Filter, 8b I&Q, Low)
-70
-25 -20 -15 -10 -5 0 5 10 15 20 25
Frequency (MHz)
Figure 42: Baseband frequency offset test results – Comparison of low band and high band
responses for offsets over 25 MHz
83
Power vs BB Frequency from the LO Signal
(Single Tone Signal, LO = 2443.75MHz, Low Band)
0
-10 24MHz BW Filter
-20 36MHz BW Filter
48MHz BW Filter
-30
Power (dBm)
-40
-50
-60
24 MHz
-70
36 MHz
-80
-90 48 MHz
-100
2400 2416.66 2450 2483.33 2500
Frequency (MHz)
Figure 43: Baseband frequency offset test results – Response of wideband Gaussian noise (low band)
Key observations from these tests are that the power of the signal output is visibly affected
by the choice of MBW and baseband filter size. Frequency modulated signals will provide
a more uniform power output if implemented using 25 MHz or 50 MHz sampling rates,
since these offer wider plateau regions in the power curve response with wide 3 dB cut-off
bandwidths. This is further illustrated in Table 22 below, which summarises the
approximate 3 dB bandwidth frequencies and roll-off factor20 beta from test data.
Wide band noise testing showed that spurs (harmonic or otherwise) may be generated at
frequencies beyond the baseband filter limits. This highlights a limitation in the radar
transmitter’s ability to generate band focussed or band limited noise, since significant
spurs will be created at out of band frequencies that may affect out of band receivers.
84
4.1.9 Third Order Output Intercept Point
The aim of this test was to obtain a measure of the transmitter’s non-linearity, by
determining the third order output intercept point (OIP3).
Two methods for determining the OIP3 were used, both involved transmitting the two
tone test waveform21 then obtaining power measurements of the inter-modulation tones
(IM1 and IM2) and the fundamental tones (F1 and F2) so that the OIP3 could be
determined [37].
Three criteria must be met to obtain an accurate measure of the inter-modulation tones.
No information was found on what to expect for the OIP3 of the XCVR2450 during
transmission. However, available data for the WBX daughterboard in receive, showed IIP3
values typically between 0 dBm and 15 dBm. In the absence of other information this
provides a ball park figure of what to expect from the XCVR2450 daughterboard.
21Details regarding the IM1, IM2, F1 and F2 are described under the two tone test waveform
Section 3.2.2 and are illustrated in Figure 12.
85
Method 1: Graphical Determination of the OIP3
This method of determining the OIP3, involved plotting the output power vs. input power
(in dB) of both a fundamental tone and an inter-modulation tone. Ideally the response of
both tones would be linear with a 1:1 response for the fundamental tone, and a 3:1
response for the inter-modulation tone; that is, prior to both tones experiencing
compression at higher input powers as illustrated in Figure 44.
NOTE:
This figure has been removed due to copyright.
Alternatively, the item is available from the referenced links
As discussed earlier the change in voltage and hence power is directly proportional to the
change in amplitude. The formula below expresses the normalised input power for each
amplitude, where 0 dB is the input power for an amplitude of 1.
Given that the exact input power at each amplitude value is unknown, the amplitude
input has been normalised and expressed as a power in dB relative to 0 dB (e.g. at an
amplitude of 1.) The fundamental and inter-modulation curves were then plotted as
shown in Figure 45 and Figure 46. The OIP3 was determined from these plots by
identifying the hypothetical point where straight lines extended through the ideal linear
regions of each curve would intersect22.
Microsoft Excel was used to create trend-lines based on data points that met the three
criteria for accurate measurements discussed above. Formulas for these trend-lines were
provided in Microsoft Excel, and used to determine the intersecting point at which the
OIP3 exists.
86
Output Power vs Normalised Input Power
(Two-Tone Signal, LO centred at 2450MHz, Gain = 0dB)
20
0
Output Power (dBm)
-20
-40
OIP3
7.796 dBm
-60
Fundamental Tone
Intermodulation Tone
-80
F Trendline (y=0.8603x+1.6399)
IM Trendline (y=2.8443x-12.556)
-100
-25 -20 -15 -10 -5 0 5 10
Normalised Input Power (dB)
Figure 45: OIP3 results using the graphical method at 2450 MHz
0
Output Power (dBm)
-20
-40
-60
Fundamental Tone
-80
Intermodulation Tone
F Trendline (y=0.9966x+0.7)
-100
-30 -25 -20 -15 -10 -5 0 5 10
Normalised Input Power (dB)
Figure 46: OIP3 results using the graphical method at 5400 MHz
87
As identified in earlier amplitude tests involving two tone waveforms, the high band path
highly suppresses inter-modulation products. Thus very few IM measurements obtained
in the high band were both above the noise floor, and beneath the 0.5 amplitude limit at
which DAC saturation occurs in the high band. Out of the measurements in this range,
few data points demonstrated behaviour close to a linear 3:1 response to an increase in the
input power. As such a clear OIP3 value could not be identified line from the high band
data using this method. This problem did not arise in the low band case and an adequate
line matching the data was obtained, that was close to a slope of 3. As indicated on the plot
the OIP3 values obtained for a gain of 0 dB was around 8 dBm at 2450 MHz.
This method for determining the OIP3 required only taking measurements of the
fundamental tones F1, F2 and a single inter-modulation tone; either IM1 or IM2. The
following formula [41] can then be used to obtain a value for the OIP3 that is ‘faster and
more accurate than the traditional (graphical) approach.’
This second method was applied to verify that the first set of calculations for the low band
data appeared reasonable, and to obtain an OIP3 value for the high band. Additionally,
since this method only requires a single measurement of 3 data points to determine the
OIP3, this approach was used to rapidly extend the OIP3 calculations. Calculations were
made for each gain step, covering frequencies at the start, centre and end of both
frequency bands.
Results for the low band and high band are shown in Figure 47 and Figure 48 respectively,
which show the OIP3 changing as a function of gain, for a fixed input power i.e.
corresponding to a fixed amplitude value of 0.25.
Note: this is not a traditional ‘Input Power versus Output Power’ plot of fundamental /
inter-modulation frequencies where a 3:1 ratio would be expected.
88
OIP3 vs Gain Request for various LO Frequencies in the Low Band
(Two-Tone Signal, Amplitude = 0.25)
35
30
25
OIP3 (dBm)
20
5
0 5 10 15 20 25 30 35
Gain Request (dB)
Figure 47: OIP3 results using the rapid calculation method for selected low band frequencies
35
30
OIP3 (dBm)
25
20
15 LO Frequency 4900MHz
LO Frequency 5400MHz
10 LO Frequency 6000MHz
5
0 5 10 15 20 25 30 35
Gain Request (dB)
Figure 48: OIP3 results using the rapid calculation method for selected high band frequencies
89
Comparing results from each method for a gain of 0 dB, at 2450 MHz and 5400 MHz;
method 1 produced OIP3 values of 8 dBm and an inconclusive result, whereas method 2
produced values of 10 dBm and 17 dBm. These low band values match within ~2 dB and
thus indicate that both methods have been applied correctly to the low band and provide
consistent results.
Examining the low band and high band results showed changes in the OIP3 for different
LO frequencies. In the low band the OIP3 was observed to increase as the LO frequency
changed from 2400 MHz to 2450 MHz and then to 2500 MHz. At 2500 MHz the OIP3 had
increased by 5 dB, which is consistent with the 5 dB increase in output power observed
during power versus frequency tests shown earlier in Figure 37. Examining the high band
results (see Figure 48) showed similar behaviour. The approximately 10 dB increase in the
OIP3 observed when the LO frequency was placed in the centre of the band, is consistent
with the observations observed in earlier testing (see Figure 38).
In all low band test cases the OIP3 increased linearly with gain as expected23. A similar,
somewhat linear response occurred in the high band. It is theorised that the erratic
behaviour of the OIP3 at 6000 MHz in this ‘linear region’ may be the result of operating at
the upper frequency limit of the USRP device, with the two tones offset to the right past
this limit might have impacted or suppressed the IM frequencies further, thus leading to
erratic measurements, but further testing would be required to determine this.
At higher gain values the response of the OIP3 curves began to plateau. It is believed that
at this stage the inter-modulation products and / or the fundamental tones have left their
linear regions of behaviour and are thus causing variations in the OIP3. Further
investigation would be required to verify this.
The OIP3 values observed across all tests cases were at a minimum of 5dB in each band.
23 OIP3 = IIP3 + Gain, where IIP3 refers to the third order input intercept point
90
4.1.10 Local Oscillator Suppression
The LO signal (or the image signal) may potentially be the strongest undesired signal
output from a transmitter, and may affect receivers out of the desired frequency band of
operation. The LO suppression is the difference between the power of the RF signal and
the LO signal. The image suppression is calculated in the same way, as shown in the
formulas below. These provide measures as to how much these signals may interfere with
a receiver.
=
=
During the detailed gain tests discussed in Section 4.1.5 the LO signals placed at 5393.75
MHz and 2393.75 MHz began at -31.5 dBm and -29 dBm respectively for a 0 dB gain input.
As the gain was increased from 0 dB to 23 dB (i.e. remaining in the linear region of
operation), the LO signal power increased by the same amount that the RF signal
increased. Thus the LO suppression remained near 21 dBm in both bands as summarised
in Table 23. Data applies to the linear region, defined as 0dB to 23dB.
During the ‘Power vs. Baseband Frequency’ tests discussed in Section 4.1.8, which were
conducted with a gain input of 0 dBm, the power of the LO remained at the minimum
values of -31.5 dBm and -29 dBm respectively regardless of the RF signal power, frequency
or bandwidth.
Thus without extending the tests conducted earlier, we still have an understanding of the
LO suppression. Since changes in gain increase both the LO and RF signal powers by the
same amount, the LO suppression will remain constant for gain values in the linear region
of operation for that band, and based upon the limited data collected in this study, the LO
suppression is likely to remain around 20 dB, although this is expected to shift depending
upon the LO frequency used. If the RF signal is offset sufficiently from the LO signal that a
reduction in power occurs24 the LO suppression will reduce by this amount.
24 As shown in Section 4.1.2, this will be a factor of the baseband frequency and sampling rate.
91
4.1.11 Phase Noise Measurements
Local oscillator noise can have a significant impact on the output of any transmitter, since
this noise will also be subject to following amplification stages along with the desired
signal, and thus be present when the waveform is transmitted from the device.
The aim of these was to measure the phase noise in the low and high band to obtain a
gauge of the expected performance. This involved transmitting the single tone test
waveform to a signal analyser, equipped with a mode specifically for measuring the phase
noise, referenced to the phase noise of the signal analyser. Measurements were taken of
the RF signal at various frequencies covering the low and high bands. The gain and
amplitude was also adjusted to observe any impact on the phase noise.
An indication of the expected results is provided in Figure 49, which shows phase noise
measurements available in the datasheet of the MAX2829 transceiver, which provides the
local oscillators and mixing functions of the RF daughterboard. Details regarding the test
conditions and reference signal for this data are not provided thus this data serves only as
a ball park expectation of what results will be obtained and are not suitable for a detailed
comparison.
NOTE:
This figure has been removed due to copyright.
Alternatively, the item is available from the referenced links
Figure 49: Phase noise plots from the MAX2829 Transceiver datasheet [34]
The following discussions will focus on the core of the information which is presented in
plots of the phase noise spectrum. However, tables summarising the full set of tabulated
data from these tests is available in the Appendix, Section 6.2 for the interested reader.
These tables include other measures such as jitter, residual phase modulation and residual
frequency modulation.
92
Phase Noise Response to variations in the Gain and Amplitude Parameters
The test matrix for varying the gain and amplitude values encompassed all combinations
of the following values.
Typical phase noise plots measured at the centre of each band are shown in Figure 50 and
Figure 51. The low band case (2450MHz) shows phase noise values typically around -
90dBc/Hz at 2450MHz, with negligible spurs. The high band case (5400MHz) shows phase
noise typically around -85dBc/Hz at 5400MHz, with spurs extending up to 35 dB above
the main curve of the phase noise signal. These spurs are not reflected in the manufacturer
data sheets.
Figure 50: Phase noise plot for a single tone at 2450 MHz (Gain = 0 dB, Amplitude = 0.25)
93
Figure 51: Phase noise plot for a single tone at 5400MHz (Gain = 0 dB, Amplitude = 0.25)
Adjusting the gain did not cause any noticeable change in the phase noise spectrum as
shown in Figure 52. Increases were observed in the residual noise measurements such as
jitter (see top of figure) when the gain was set to 35dB. The rising spur observed in Figure
52 at an offset of 6.5MHz on the right is the LO signal, which has risen by the gain input of
35dB as expected.
Figure 52: Phase noise plot for a single tone at 5400 MHz (Gain = 35 dB, Amplitude = 0.25)
94
Adjusting the amplitude did not have any significant impact on the phase noise in the
cases tested, except when amplitudes of 1 or greater are used. An amplitude value of 1
caused the phase noise to increase at offsets up to 30 kHz (see Figure 53). This is possibly
caused by saturation of the DACs, despite typically occurring at values greater than 1 as
observed in most tests. This noise spread across the spectrum more significant at an
amplitude value of 1.1.
Changing the amplitude (at useable values below 1) did result in notable changes to the
residual noise measurements and the RMS jitter (which varied between 0.73 and 2.33 ps.)
These values were at a minimum when the amplitude was set to 0.5.
Figure 53: Phase noise plot for a single tone at 5400 MHz (Gain = 0 dB, Amplitude = 1)
In the high band test cases, the phase noise at the 10 kHz and 100kHz offsets varied by 5 to
10dB depending upon the RF frequency. These values were at their minimum for
frequencies of 5100 and 5600 MHz. RMS jitter varied throughout testing depending upon
the RF frequency but remained beneath 1.4 ps when using an amplitude value of 0.25
which has been recommended throughout this thesis.
95
Phase Noise vs RF Frequency, measured at various Frequency Offsets
(Single Tone, Gain = 0dB, Amplitude = 0.25)
-85
-90
Phase Noise (dBc/Hz)
-95
-115
-120
2400 2425 2450 2475 2500
RF Frequency (MHz)
-80
Phase Noise (dBc/Hz)
-85
-90
-110
-115
5000 5100 5200 5300 5400 5500 5600 5700 5800 5900 6000
RF Frequency (MHz)
Figure 55: Phase noise measurements at various high band frequencies
96
4.2 Waveform Verification Testing
The aim of this section of testing was to assess the accuracy of waveforms output by the
radar transmitter. Unlike the characterisation testing that focused on the limits of the
USRP performance to identify unexpected behaviour and regions where it occurred, the
waveform testing focused on the performance of the USRP under ideal conditions to
examine how accurately it can produce radar waveforms.
This involved transmitting the set of waveforms defined earlier in Section 3.3 then:
Recording the signal output in the time domain using the oscilloscope, to compare
to the input data in the time domain from GNU Radio.
Recording the signal output in the frequency domain using the spectrum analyser,
to compare to a model of the expected frequency domain response.
The modelled results were based upon time domain data of the transmitted signal from
GNU Radio, which was converted to the frequency domain using a FFT in Matlab.
The RF waveform output from the USRP, will (ideally) share the same timing
characteristics and power spectrum around the LO frequency to that of the BB signal input
passed to the USRP from GNU Radio.
97
4.2.1 Continuous Waveform
The continuous waveform consisted of a cosine wave. This serves as simple baseline for
testing. A scope plot of the recorded CW data input to the USRP is shown in Figure 56.
(Due to the 4 samples per period minimum used in these tests, the figure appears more
like triangle wave.) The model of the expected power spectrum relative to the LO is
shown in Figure 57. This plot shows the power of the RF tone beginning to spread at -400
dB which is a highly idealised case. In reality it is expected that this will occur at a much
higher level, and close to the noise floor, which will likely be around -65 to -70 dB based
upon observations in earlier testing.
0.1
Amplitude
-0.1
-0.2
-0.3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (us)
Figure 56: Time scope plot of the baseband CW input to the USRP
-100
|S(f)| 2 (dB)
-200
-300
-400
-500
3 3.5 4 4.5 5 5.5 6 6.5 7
Frequency (MHz)
Figure 57: Modelled normalised power spectrum of the baseband CW input to the USRP
98
The output signal recorded by the spectrum analyser is shown in Figure 58. The RF, LO,
image signal were placed at the frequencies expected. The only distortion visible was a
signal at 5805MHz, which may be a harmonic of the BB frequency, or a product of
reciprocal mixing. The peak power was 2 dBm. SNR was 64 dB25 and SFDR was 16 dB.26
Figure 58: Measured power spectrum of the CW output from the USRP
99
Figure 59: Measured time scope plot of the non-interpolated CW output from the USRP (500
ps/div, 5 ns span)
Figure 60: Measured time scope plot of the interpolated CW output from the USRP (500 ps/div, 5
ns span)
100
4.2.2 Pulsed Waveform
The pulsed waveform consisted of the continuous waveform used in the previous test,
with a duty cycle of 10% applied to divide it into 1 μs pulses, repeating at 10 μs intervals.
A scope plot of the recorded CW data input to the USRP is shown in Figure 61.
0.1
Amplitude
-0.1
-0.2
-0.3
0 2 4 6 8 10 12 14 16 18 20
Time (us)
Figure 61: Time scope plot of the baseband pulsed waveform input to the USRP
The model of the expected power spectrum relative to the LO is shown in Figure 62 and
Figure 63. Several key waveform characteristics can be determined [42] from this plot,
which are expected to be visible in the USRP output.
The PRI can be identified by observing the spectral line spacing, and using the
formula below. As the spectral lines are spaced at 100 kHz (PRF) intervals, this
confirms the PRI of 10 μs.
The PW can be identified by observing the width of each side lobe. These occur at 1
MHz intervals which confirm the PW of 1 μs.
The duty cycle of 10% can then be calculated from the above values.
101
The theoretical desensitization factor can be determined using the following formula by
substituting a duty cycle value of 10% (e.g. 0.1) into the following equation. This provides
a desensitisation value of equals -20 dB. This is not reflected in the normalised power
spectrum shown in Figure 62, however when comparing the measured peak pulse power
of the output pulsed waveform and the non-pulsed CW observed previously, a power
reduction of -20 dB should be visible.
-40
2
-60
-80
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
Figure 62: Modelled normalised power spectrum of the baseband pulsed waveform input to the
USRP
-20
2
-40
-60
-80
3 3.5 4 4.5 5 5.5 6 6.5 7
Frequency (MHz)
Figure 63: Modelled normalised power spectrum (close up view) of the baseband pulsed waveform
input to the USRP
102
The output signal recorded by the spectrum analyser is shown in Figure 64. The expected
characteristics described earlier are observed. The main lobe is centred at 5800 MHz (+5
MHz relative to the LO frequency). As shown clearer in Figure 65 the width of the
sidelobes and spectral line spacing are 1 MHz and 100 kHz respectively, thus confirming
the PRI, PW, and duty cycle calculated earlier. The peak pulse power is -18.79 dBm as
shown in the top right corner of Figure 64. When compared to the 2 dBm peak power of
the CW signal from Figure 58 earlier, this produces a desensitisation factor of -20.79 dBm27
which adequately matches the theoretical value of -20 dB. This reduction lowers the
power of the main RF signal around 4 dB beneath that of the LO signal which may
interfere with receivers near the LO frequency.
Figure 64: Measured power spectrum of the pulsed waveform output from the USRP
103
Figure 65: Measured power spectrum of the pulsed waveform output from the USRP
The oscilloscope output is shown in Figure 66. Visual inspection further confirms that the
PRI is approximately 10 μs. This was verified in several tests, and matches the
observations on the spectrum analyser. However, the auto measurement function
measured a mean period of ~9.04 μs, as shown at the bottom of this figure. Given that
would imply a 10% error, it is believed that there was an error in the trigger settings for
this test. The auto measurement appears to be measuring the time difference between the
trailing edge of one pulse to the leading edge of the next, thus triggering a 9 μs period
measurement. Assuming this is the case, the measurement matches the 9 μs off time
between pulses. Converting the four auto-measured frequency values into the time
domain produces four period values differing from the four auto-measured period values
displayed. This produces eight off-time measurements. Compared to a 9 μs off time, the
maximum deviation is 0.14 μs which provides some bounds on the timing accuracy of the
pulses generated by the radar transmitter.
A close up view of a pulse is shown in Figure 67, which shows 200 ns divisions equating to
a span of 2 μs. The PW signal appears to be slightly wider than the target 1 μs PW. The
part of the signal that exceeds the start and end of the pulse is less than the 50 ns28 time per
sample, thus is interpreted as an expected artefact of the USRP voltage ramping up or
down to during the start and end of the ‘on’ period, and thus triggering the auto-
measurement function at different times, rather than due to a mistiming or error in sample
transmission. This would account for the 0.14 μs deviation in the pulse timing.
104
Figure 66: Measured time scope plot of the non-interpolated pulsed waveform output from the
USRP (5 μs/div, 50 μs span)
Figure 67: Measured time scope plot of the non-interpolated pulsed waveform output from the
USRP (200 ns/div, 2 μs span)
105
4.2.3 Frequency Modulated Continuous Waveform
The FMCW consisted of the CW used in the first test, with two-way frequency modulation
sweeps applied over a 20 μs time frame (e.g. 10 μs from the start to the maximum
frequency deviation, then 10 μs to return to the start frequency.) The RF signal was placed
at 5800 MHz, with the LO located at 5805 MHz.
Scope plots of the recorded FMCW data input to the USRP are presented for selected
frequency sweeps in Figure 68, Figure 69 and Figure 70. It should be noted as the complex
baseband frequency begins at a negative value then sweeps towards zero, although the
mixed RF signal will be of a higher frequency, the magnitude of the complex baseband
frequency reduces. Hence for the 10 MHz sweep shown in Figure 70 which is swept
through the zero region, the signal appears to reduce in frequency as it moves from -5
MHz towards zero, increases in frequency as it approaches +5 MHz, then on the return
sweep reduces in frequency as it moves from +5 MHz to zero, then increases again as it
approaches -5 MHz.
Measured output in the time domain appears in Figure 71, Figure 72 and Figure 73, which
correspond to the FM sweeps shown in Figure 68, Figure 69 and Figure 70 respectively
Due to the presence of the 5805 MHz LO frequency mixed in with the baseband signal,
increases and decreases in the frequency modulation must be determined by regions of
increased and decreased fluctuations in the amplitude. Comparing the scope plots of the
FMCW input to their corresponding measured output show that frequency modulation
over the 20 μs ramp time appeared as expected.
106
FMCW, Scope Plot
Start Frequency -5MHz, Sampling Rate 20MHz
Freq Sweep 2MHz and return, over 20us
0.3
0.2
GNU Radio
0.1
Amplitude
0
-0.1
-0.2
-0.3
0 2 4 6 8 10 12 14 16 18 20
Time (us)
Figure 68: Time scope plot of the 2 MHz sweep FMCW input to the USRP
0
-0.1
-0.2
-0.3
0 2 4 6 8 10 12 14 16 18 20
Time (us)
Figure 69: Time scope plot of the 5 MHz sweep FMCW input to the USRP
0
-0.1
-0.2
0 2 4 6 8 10 12 14 16 18 20
Time (us)
Figure 70: Time scope plot of the 10 MHz sweep FMCW input to the USRP
107
Figure 71: Measured time scope plot of the 2 MHz sweep FMCW output from the USRP (2 μs/div,
20 μs span)
Figure 72: Measured time scope plot of the 5 MHz sweep FMCW output from the USRP (2 ns/div,
20 μs span)
Figure 73: Measured time scope plot of the 10 MHz sweep FMCW output from the USRP (2
ns/div, 20 μs span)
108
Models of the expected power spectrum for these cases (relative to the LO) are shown in
Figure 74, Figure 75, and Figure 76. These provide an indication of what the signal
response is expected to look like based upon the change in frequency sweep size.
Since each frequency sweep repeats every 20 μs, we thus see spectral lines on these three
plots every 0.05 MHz. The spectral lines were not observed in the measured results, since
the measured data shows the spectrum of a single repeating pulse and not a pulse train.
The power spectrums for several FMCWs output from the USRP are shown in
Figure 77 and Figure 78. These show the response for a range of different frequency
sweeps at 20MSps and 50MSps respectively. The frequency modulation sweeps across the
range intended in each case, and the shape of the power spectrum appears as expected. An
image of the FM sweep is observed starting 30 dB below the main signal at 5810 MHz,
along with a harmonic of the FM sweep beginning at 5820 MHz.
As the frequency modulation increased, the power was reduced to spread over a wider
bandwidth. The peak power of the LO remained at the same power level in each plot
regardless of the frequency modulation (although this is difficult to see since the LO signal
overlaps for each data curve
This indicates that due to poor LO suppression (up to -20 dB for a 25MHz sweep), large
FM sweeps (i.e. greater then 10MHz) will reduce the RF power beneath the LO signal as
seen in Figure 78.
109
FMCW, Power Spectrum
Start Frequency -5MHz, Sampling Rate 20MHz
Freq Shift 2MHz and return, over 20us
10
0
-10
|S(f)| 2 (dB)
-20
-30
-40
-50
-60 500000 Sample FFT
-70
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
Frequency (MHz)
Figure 74: Modelled normalised power spectrum of the 2 MHz sweep FMCW input to the USRP
FMCW Power Spectrum
Start Frequency -5MHz, Sampling Rate 20MHz
Freq Sweep 5MHz and return, over 20us
10
0
-10
|S(f)| 2 (dB)
-20
-30
-40
-50
-60 500000 Sample FFT
-70
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
Frequency (MHz)
Figure 75: Modelled normalised power spectrum of the 5 MHz sweep FMCW input to the USRP
FMCW, Power Spectrum
Start Frequency -5MHz, Sampling Rate 20MHz
Freq Sweep 10MHz and return, over 20us
10
0
-10
|S(f)| 2 (dB)
-20
-30
-40
-50 500000 Sample FFT
-60
-70
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
Frequency (MHz)
Figure 76: Modelled normalised power spectrum of the 10 MHz sweep FMCW input to the USRP
110
FMCW, Power Spectrum
Sampling Rate 25MSps, LO Freq 5805 MHz, Start RF 5800MHz
FM Sweeps of 0MHz / 0.5MHz / 2MHz / 5MHz /10MHz
10
0 LO peak power is the
0 MHz FM Sweep
same for each FM sweep
-10 0.5 MHz FM Sweep
1 MHz FM Sweep
Power (dBm)
-20
2 MHz FM Sweep
-30 5 MHz FM Sweep
10 MHz FM Sweep
-40
-50
-60
-70
5785 5790 5795 5800 5805 5810 5815 5820 5825
Frequency (MHz)
Figure 77: Measured power spectrum of the FMCW output from the USRP, for a range of Triangular FM sweeps at 20 MSps
111
FMCW, Power Spectrum
Sampling Rate 50MSps, LO Freq 5805 MHz, Start RF 5792.5MHz
20us Triangular FM Sweeps of 12.5MHz / 25MHz
0
LO peak power is the 12.5 MHz FM Sweep
-10
same for each FM sweep 25 MHz FM Sweep
-20
Power (dBm)
-30
-40
-50
-60
-70
5785 5790 5795 5800 5805 5810 5815 5820 5825
Frequency (MHz)
Figure 78: Measured power spectrum of the FMCW output from the USRP, for a range of Triangular FM sweeps at 50 MSps.29
29 Due to an input error, the LO signal was placed at 5805 MHz for this test instead of 5812.5 MHz. Thus the test is centred at a different
frequency. Other parameters were correct. This change will not have any impact on the results.
112
4.3 Experimentation Summary
The characterisation testing and waveform verification tests were conducted successfully
with test results consistent when compared across tests.
Test Outcomes
The characterisation tests investigated and verified limitations of the USRP performance,
and identified others that were unknown at the time or did not match expected values.
The waveform verification test results showed that the SDR transmitter produced pulsed
waveforms with accurate timing and performed frequency modulation accurately over
various sweep ranges achievable by the USRP device.
As stated earlier the host computer performance (see Section 2.2.5) met the minimum
requirements recommended (see Section 1.4.3). Nonetheless, the host computer’s
processing power was not sufficient to generate / transmit signals and also display the
signals simultaneously on the GNU Radio GUI visualisers. In doing so, the packet under-
run flags began to appear in the GNU Radio output and the system occasionally stalled or
stopped if attempting to transmit pulsed waveforms or frequency modulated waveforms.
Turning off the GUI visualisers removed this problem. The system still accumulated
packet under-run flags over a long period of time; however this quantity was infrequent
enough to be considered negligible and did not impact testing.
This highlights the processor intensive requirements of running a GNU Radio and USRP
based SDR (transmitter only) even as a using a host computer with comparatively high
processing performance.
Summary of Findings
A summary of the main findings is provided in the followings tables: Table 24, Table 25,
Table 26 and Table 27. However, the full extent of the findings including detailed values
are captured in the figures, tables and discussions presented in each test section of this
thesis.
113
Table 24 Summary of characterisation test findings (Part A)
114
Table 25 Summary of characterisation test findings (Part B)
RMS jitter and other residual noise measures were notably affected by
the selected amplitude, but minimised for an amplitude value of 0.5.
115
Table 26 Summary of waveform verification test findings
FMCW Testing Unexpected spectral lines appeared in the model of the FFT based on
sample data, but identified as being caused by the FFT algorithm
used.
The host computer used in this SDR transmitter had the following
Host Computer processor specifications:
Processor Speed
Intel Core 2 Quad, @ 3GHz
The CPU was unable to maintain the desired sample rate with the
GNU Radio visualisers active. Turning the visualisers off allowed the
SDR transmitter to operate effectively.
116
5. Conclusions
GNU Radio and the USRP can be effectively combined to create a software-defined radar
transmitter, as measured by the accuracy with which the output waveforms matched the
set of target waveforms defined.
This thesis has detailed the system design of a SDR transmitter that has been developed
using these tools and used in experimentation to support this thesis.
Characterisation tests investigated and verified limitations of the USRP performance, and
identified others that were unknown at the time or did not match expected values. The
majority of test results are explainable in the context of the hardware device’s
subcomponents and datasheets. The USRP exhibited notable variations in transmit power
and phase noise across the flexible operating domain achievable. Additionally, the SDR
generated low noise spurs during low band testing that were between 5 to 18dB above the
noise floor at full scale. IM products were well suppressed in the high band compared to
the low band, by up to 30dB in some cases. The low band spurs and IM products may be
unacceptable for some radar transmission applications, and would require filtering.
The cause of some noise spurs observed in testing are not yet identified. The
characterisation test results highlight behavioural aspects of fundamental importance to
radar designers considering using these tools to produce a radar transmitter.
Waveform verification test results produced 1 μs pulsed waveforms with accurate timing
and performed frequency modulation accurately over sweep ranges from 0.5 to 25MHz.
The transmitted waveforms were not without imperfections. Poor LO suppression meant
that for FM sweeps above 10MHz the power of the RF signal was below that of the LO
signal.
Furthermore, it is recommended that future work with these tools incorporate a host
computer with higher processing power then the one used in this study. This will further
ensure the sample rate required by the USRP is maintained, preventing packet under-runs
that will degrade the integrity of the transmitted signal.
117
6. Appendix
if (nargin == 1)
sample_rate = 1.0;
end;
len = length(data);
s = fft (data.*kaiser(len, 5),len);
fft_data = abs(fftshift(s))/len; %added /len
incr = sample_rate/len;
min_x = -sample_rate/2;
max_x = sample_rate/2 - incr;
end %function
118
6.2 Appendix B - Tabulated Phase Noise Measurements
Table 28 Single tone waveform response to various amplitude values with gain values of 0 and 10dB (high band)
119
Table 29 Single tone waveform response to various amplitude values with gain values of 20 and 35dB (high band)
120
Table 30 Single tone response to stepped changes in the RF signal frequency across the low and high bands
121
7. References
1. Application Note: Selecting a USRP Device. [cited March, 2012]; Available from:
http://www.ettus.com/content/files/kb/application_note_selecting_a_usrp.pdf.
2. Patton, L.K., A GNU Radio Based Software-Defined Radar, in Department of Electrical
Engineering 2007, Wright State University: USA.
3. Fernandes, V.N., Implementation of a Radar System using Matlab and the USRP, in
Electrical Engineering 2012, California State University, Northridge: USA.
4. Williams, L., Low Cost Radar and Sonar using Open Source Hardware and Software, 2008,
University of Cape Town: South Africa.
5. Volkwin, A., Suitability of a Commercial Software Defined Radio System for Passive Coherent
Location, in Department of Electrical Engineering, 2008, University of Cape Town:
South Africa.
6. Szlachetko, B.L., A. and J. Zarzycki, Application of the Software Defined Radio in a Passive
Radar, in Military Communications and Information Systems Conference, 2009, Wroclaw
University of Technology, Poland: Czech Republic.
7. Prabaswara, A., A. Munir, and A.B. Suksmono, GNU Radio Based Software-Defined
FMCW Radar for Weather Surveillance Application, in The 6th International Conference
on Telecommunication Systems, Services and Applications, 2011, IEEE.
8. Debatty, Software Defined RADAR a State of the Art, in 2nd International Workshop on
Cognitive Information Processing, IEEE, June 2010, Royal Military Academy of
Brussels: Italy.
9. Definitions of Software Defined Radio and Cognitive Radio System, 2009, p.3: International
Telecommunications Union.
10. What are the advantages of Software Defined Radio?, July 2002: Software Defined Radio
Working Group of the ARRL. p. 1.
11. Smith, S., The Scientist and Engineer's Guide to Digital Signal Processing. 1997: California
Technical Publishing.
12. Welcome to GNU Radio. [cited March, 2012]; Available from:
http://gnuradio.org/redmine/projects/gnuradio/wiki.
13. Ettus Research Website. [cited March, 2012]; Available from: https://www.ettus.com/.
14. Corgan, J., GNU Radio in Action (Conference Overview), September 2011, Corgan
Enterprises: GNU Radio Conference 2011.
15. Capria, A.C., M. Petri, D. Martorella, M. Berizzi, F. Mese, E. D. Soleti, R. Carulli, V.,
Ship Detection with DVB-T Software Defined Passive Radar, 2010, RaSS Centre CNIT,
University of Pisa, Italian Navy CSSN ITE.
16. Godana, B., Human Movement Characterization in Indoor Environment using GNU Radio
based Radar, 2009, Delft University of Technology: Netherlands.
17. Lambert, J.C., A Radar Interrogator for Wireless Passive Temperature Sensing, 2008,
University of Central Florida: USA.
18. USRP Technical Documentation. [cited April, 2012]; Available from:
http://code.ettus.com/redmine/ettus/projects/public/documents.
19. XCVR2450 Subcomponent Specfication Sheets Archive. [cited June, 2012]; Available from:
http://code.google.com/p/microembedded/downloads/list?q=label:XCVR2450.
20. Hamza, F.A. The USRP under the 1.5X Magnifying Lens! June 2008.
122
21. Balister, P.R., J., USRP Hardware and Software Description, June 2006, Virginia
Polytechnical Institue & State University.
22. Shen, D., Tutorial 1: GNU Radio Installation Guide Step by Step (10 Tutorials), May, 2005.
23. (Forum: GNU Radio) Understanding USRP2 Flow Control. July 2010 [cited May, 2012];
Available from: http://www.ruby-forum.com/topic/213352.
24. Prasetiadi, A., A Simple Delay Compensation System in Software-Defined Frequency
Modulated Continuous (FMCW) Radar, in European Wireless 20122012, VDE VERLAG
MBH: Poland.
25. Standert, R., Software Model of a Radar Receiver, 2002, Royal Institute of Technology,
Stockholm: Sweden.
26. Frequency Modulated Continuous Wave Radar. [cited July, 2012]; Available from:
http://demonstrations.wolfram.com/FrequencyModulatedContinuousWaveFMC
WRadar/.
27. Frasier, S.J., T. Ince, and F.J. Lopez-Dekker, Performance of S-Band FMCW Radar for
Boundary Layer Observation, in 15th Conference on Boundary Layer and Turbulence2002.
28. Stimson, Introduction to Airborne Radar. 2nd ed. 1998: SciTech Publishing Inc.
29. Application Note: Selecting a RF Daughterboard. [cited March, 2012]; Available from:
http://www.ettus.com/content/files/kb/Selecting_an_RF_Daughterboard.pdf.
30. NI USRP-2921 Block Diagram, August 2011, National Instruments.
Available from: http://zone.ni.com/reference/en-XX/help/373380A-
01/usrphelp/2921_block_diagram
31. NI USRP-2920, NI USRP-2921 Universal Software Peripheral Datasheet. [June 2012];
Available from: http://sine.ni.com/ds/app/doc/p/id/ds-355/lang/en.
32. USRP N210 Schematics. [cited October, 2012]; Available from:
http://code.ettus.com/redmine/ettus/attachments/download/214/n210.pdf.
33. RF Daughterboard Notes. [cited October, 2012]; Available from:
http://files.ettus.com/uhd_docs/manual/html/dboards.html.
34. MAX2829 Datasheet. [cited November, 2012]; Available from:
http://datasheets.maximintegrated.com/en/ds/MAX2828-MAX2829.pdf.
35. Application Note: UHD Examples. [cited March, 2012]; Available from:
www.ettus.com/content/files/kb/application_note_uhd_examples.pdf.
36. Australian Radiofrequency Spectrum Plan, A.C.a.M. Authority, Editor January, 2009,
Australian Government.
37. Third Order Intercept Measurements, 2001, Agilent Technologies.
38. XCVR2450 Schematics. [cited November, 2012]; Available from:
http://code.ettus.com/redmine/ettus/attachments/download/210/xcvr2450_rev
1.1.pdf.
39. Sharp Corporation, IRM046U Target Specification Datasheet. March 2009.
http://ses.sharpmicro.com/download/IRM046U-030918pdf
40. Aniritsu Corporation, Intermodulation Distortion (IMD) Measurements. 2000;
Available from: www.aniritsu.com
41. Kundert, K., Accurate and Rapid Measurment of IP2 and IP3. Designers Guide
Community, May 2002.
42. Spectrum Analyzer Measurements. [cited October, 2012]; Available from:
http://www.microwaves101.com/encyclopedia/spectrumanalyzer.cfm.
123