Ι3Ό8012 九、發明說明： 【發明所屬之技術領域】 本發明係關於一種適應性RTO,特指一種在無線通訊 系統下適應性RTO的預測方法。 【先前技術】 . 傳輸控制協定（TCP )爲資料網路中最爲普遍信賴的 傳輸協定，其係爲一種連接導向（connection oriented) • 的協定，並且提供流量控制與壅塞控制。而使用三向交 握方法，TCP能夠實行虛擬的連接，並確保連接已被建 立，且TCP能夠藉由控制發送緩衝器與接收緩衝器來預 防傳輸溢位。藉由緩啓動與傳輸逾時控制，TCP可避免 網路壅塞。然而，設計於有線網路的TCP協定並不適用 在無線系統中，此乃肇因於無線鏈接與有線鏈接係有數 種不同的特性。有線鏈接具有低封包錯誤率、低頻寬變 動與較小的延遲變動，但在無線鏈接中，則有較大的封 包錯誤率，較大的頻寬變動，與較高的延遲變動。肇因 於頻寬變動的在無線連接中的封包錯誤與假逾時，會致 使擁擠控制的效能變差。 另外，關於已知的先前技術，如美國專利第 US006791945 號名爲『Timeout threshold shaping for wireless TCP communications』，該專利案係有關於系統 中RTO的計算，其採用多重通道，如SCH與DCCH通 道。對於各個通道都有不同的延遲特性。該系統處理器 6 1308012 係僅維持運行整體系統延遲特性之平均値’當一個時間 週期中大部分封包被在DCCH上以較短的往返（r〇und ' trip )時間延遲被傳輸後，該習知的處理器接著使用計 算逾時的値將高於實際運行平均數，當被傳輸於具有較 長往返時間的封包的SCH上時實際延遲將高於該逾時 之値，此時即使目標節點已正確的收到該封包，可是系 統將被認定爲逾時。因此，該網路並未良好使用。該發 I 明藉由自動增加運行時間來最低化這種不必要逾時的 發生。該處理器亦被程式化爲具有一種演算法能夠自動 插入較長、較慢的封包。該演算法被特製爲不顯著影饗 系統效能。該演算法對於每個X封包於SCH頻道上插 入較長的封包，該X封包係由來源所傳送，且由特定的 系統參數來決定。結果，該較長的插入封包增加了運行 平均値與逾時値，該發明解決了不同頻道變動的問題， 他使用適當的配置使其計算適用於所有的頻道。但是， • 由於頻寬的單一頻道之延遲特性的變動，則未被加以考 量。在本發明中，該演算法能夠被在具有單一頻道的系 統上執行。對於多重頻道，該演算法亦能夠被應用到各 個頻道上，使得對於各者之適當的RT0可被計算出來。 在專利US006493316,《用於控制基於封包之連接頻 寬的裝置及方法》中，該發明著重於調整視窗大小以增 加總通量。藉由修改TCP以將其傳送速率限制於某最大 値與最小値之間，而不是限制在數値一與該接收器視窗 7 1308012 大小之間。該最小値將保證該連接運行於一個最小速率 上’而該最大値防止將所有可用頻寬指派給某一對節點 上。此演算法可有選擇地防止在該視窗大小太小而不允 - 許快速重發及恢復時發生逾時。該方法包括若干步驟。 (1)計算擁擠視窗（C-WND)。（2)設定最小頻寬視窗 (MIN-WND)及最大頻寬視窗（MAX-WND)。（3)根據擁 擠視窗決定調用最小頻寬視窗還是最大頻寬視窗。（4) | 如果獲得許可，允許一或多個資料封包之傳輸進入該連 接。在最小頻寬視窗及最大頻寬視窗上存在兩種操作。 第一種操作係擴充視窗。當接收到一重複的ACK時， 最大頻寬視窗及最小頻寬視窗會以一的數値擴充。當接 收到一非重複ACK時，最大頻寬視窗及最小頻寬視窗 重設爲其原始大小（對最大頻寬視窗，係最大允許頻寬X 往返時間（RTT)，對於最小頻寬視窗，係所保證之最小 頻寬X往返時間）並且移至右邊，如此在視窗中之第一 • 位元組係最先的未確認位元組。即使當一 ACK丟失時’ 擴充後的視窗亦確保該連接可發送新的封包。此種操作 防止該連接暫停但不增加網路中之封包數。當發送速率 在設定的限制値之間時，執行正常的TCP演算法。如果 發送速率低於最小値時，則調用最小頻寬窗口。如果發 送速率低於最大値，則調用最大頻寬視窗。在該發明中 利用了 一重發計時器週期，其具有一大於超出視窗發送 計時器範圍（在此時間之後，即使TCP擁擠視窗之狀態 1308012 通常不允許發送封包，亦可發送該封包）及往返時間之 値。重發計時器週期確保即使當沒有接收到ACK時， ' 仍然發送最小速率之封包以激勵ACK，最終重新發送而 - 不被TCP逾時暫停。因此，增加了總通量。該發明藉由 在不同的情況下利用合適之視窗大小增加通量，並且防 止一些不必要之逾時。在本發明中，該演算法在不同通 道情況下，計算合適的RTO以防止不必要之逾時。該 φ 發明沒有指定用於無線網路。當無線通道變化顯著且快 速時，該發明之演算法將不能很快適應該種情況，並且 將產生不必要之逾時。在本發明中，基於每一傳輸之通 道情況修改往返時間。此可更快地適應當前情況並且顯 著地減少不必要之逾時。 另查美國專利『Enhancement of explicit congestion notification (ECN) for wireless network applications』係 揭露使用增進的明顯式壅塞通知（ECN )技術，來從低 • 頻寬延遲感度TCP連接避免封包的不必要延遲。此錯誤 依舊導致系統進入壅塞狀態。而這類的壅塞係由高BER 而不是缺乏緩衝而發生的。當這些錯誤發生時，TCP錯 誤的假設網路是壅塞的，且大幅度的降低了其之新舊封 包的傳輸率。而增進的ECN之主要工作，就是從個別 的封包丟失分辨由於BER之壅塞封包丟失，來當BER 導致丟失封包時，拒絕進入低起始態，並且當偵測到早 期壅塞通知時減低其傳送速度，來增加連接的輸貫量。 9 1308012 該發明著重於分辨肇因於高BER之壅塞的實際壅塞。 然而，由於頻寬的單一頻道之延遲特性之變動，則未被 列入考慮，頻寬變動導致其他類型的壅塞，而在本發明 中，我們致力於分辨從頻寬變動導致而來的壅塞中的真 實壅塞。 綜合上述，當頻寬變動發生在TCP傳輸時，原來的 RTO測量法無法有效工作，該RTI估算係被新的RTT 0 測量法更新，且在一個長的穩定傳輸之後，將趨近於 RTT値。該RTT於傳輸率上改變基底（base)。當傳輸 率從高變爲低的時候，該RTT可能超出RTO逾時臨界 並因而激發再傳輸。此再傳輸爲一假性的傳輸，因爲其 並非由未經傳遞的封包所激發，而是由RT0的變動所 激發。該假性的傳輸導致資料的不必要傳輸，而該資料 並非丟失。該相關的擁擠視窗降至一最大的節區大小 (MSS )，且該慢起始臨界將被目前視窗尺寸的一半所 # 切割。結果，該頻道未被有效使用。 【發明內容】 本案發明人等有鑑於習知技術上述之缺失，特別提出 一種在無線通訊系統下適應性RTO的預測方法，其中 一適應性的以計時器爲基礎的控制演算法能增進TCP 執行效能，該演算法動態性的改變往返時間（RTT )評 估値與適應目前頻寬狀態。由於更正的重傳逾時（RTO ) 値會被重新的適應性RTT預估値計算出來，使得頻寬變 10 1308012 動造成的不正確RTO値之假性逾時問題能夠藉此 決，並且也提升了在無線傳輸上的TCP執行效能° - 【最佳實施例之實施方式】 - ' 在原本的壅塞控制流中，TCP係藉由下列步驟來 壅塞： 第一，基地台傳輸一封包，並計算該RT0。 第二，假使該封包之往返時間（RTT )大於RT0，該 | 會認爲產生壅塞，並進入（緩）慢啓動狀態 在無線網路中，時強時弱的頻道導致頻寬變動。 頻寬突然變得比先前還要小，則具有較小頻寬的頻 會需要更多的時間來傳輸這些封包，也就是說，這 包的RTT將會變得很長。當介於TCP封包之傳輸與 封包之認知間的時間大於RT0時，將會發生壅塞 種類型的壅塞係不同於有線網路的。其係導致源頻 震盪。假使我們能避免這種類型的壅塞，我們即能 • 壅塞視窗進入緩啓動狀態（SLOW START PHASE )。 適應性RTO演算法系被提出來通知TCP有關將導 必要的TCP封包逾時並進入緩起動（狀）態的頻 動。此演算法的主要目地爲偵測頻寬變動並計算 RTO。其將發送一個新的RT0回TCP，來避免RT0 誤估算。 原來的RTT與RT0估算値係被提出用於有線環 而有線環境並不被認爲會具有不穩定的傳輸速率（ 被解 偵測 TCP 〇 假使 道將 些封 TCP 。這 寬之 避免 致不 寬變 新的 之錯 境， 尤其 11 1308012 是源於頻寬變動的不穩定）。該計算係如下所示： R=-xR + ^xRTT 8 8 Z) = ^xZ) + ^x abs(R -RTT)Ι3Ό8012 IX. Description of the Invention: [Technical Field] The present invention relates to an adaptive RTO, and particularly to a method for predicting adaptive RTO in a wireless communication system. [Prior Art] Transmission Control Protocol (TCP) is the most commonly trusted transport protocol in data networks. It is a connection oriented protocol and provides flow control and congestion control. Using the three-way handshake method, TCP can perform virtual connections and ensure that connections are established, and TCP can prevent transmission overflow by controlling the transmit and receive buffers. With slow start and transmission timeout control, TCP avoids network congestion. However, the TCP protocol designed for wired networks is not suitable for use in wireless systems because of the many different features of wireless links and wired links. Wired links have low packet error rates, low frequency wide variations, and small delay variations, but in wireless links, there are larger packet error rates, larger bandwidth variations, and higher delay variations.封 Due to bandwidth variations in packet errors and false timeouts in wireless connections, the performance of congestion control is degraded. In addition, with respect to known prior art, such as U.S. Patent No. US006791945 entitled "Timeout threshold shaping for wireless TCP communications", the patent relates to the calculation of RTO in the system, which employs multiple channels, such as SCH and DCCH channels. There are different delay characteristics for each channel. The system processor 6 1308012 maintains only the average of the overall system delay characteristics of the operation 値 'When most of the packets are transmitted on the DCCH with a short round trip (r〇und ' trip ) time delay in a time period, the The known processor will then use the calculated timeout 値 to be higher than the actual running average, and the actual delay will be higher than the timeout when transmitted on the SCH with the longer round-trip time, even if the target node The packet has been received correctly, but the system will be considered timeout. Therefore, the network is not well used. This is to minimize the occurrence of this unnecessary timeout by automatically increasing the runtime. The processor is also programmed to have an algorithm that automatically inserts longer, slower packets. The algorithm was tailored to not significantly affect system performance. The algorithm inserts a longer packet for each X packet on the SCH channel, which is transmitted by the source and is determined by the particular system parameters. As a result, the longer insertion packet increases the running average and timeout. The invention solves the problem of different channel variations, and uses appropriate configuration to make its calculations applicable to all channels. However, • The variation in the delay characteristics of a single channel of bandwidth is not taken into account. In the present invention, the algorithm can be executed on a system having a single channel. For multiple channels, the algorithm can also be applied to each channel so that the appropriate RT0 for each can be calculated. In the US 006 493 316, "Attachment of Connection-Based Connection Bandwidth and Apparatus", the invention focuses on adjusting the window size to increase the total throughput. By modifying TCP to limit its transfer rate to between a maximum and a minimum, rather than limiting it between the number and the size of the receiver window 7 1308012. This minimum 値 will ensure that the connection runs at a minimum rate' and the maximum 値 prevents all available bandwidth from being assigned to a certain pair of nodes. This algorithm can selectively prevent the timeout when the window size is too small to allow rapid retransmission and recovery. The method includes several steps. (1) Calculate the crowded window (C-WND). (2) Set the minimum bandwidth window (MIN-WND) and the maximum bandwidth window (MAX-WND). (3) Decide whether to call the minimum bandwidth window or the maximum bandwidth window according to the congestion window. (4) | If licensed, one or more data packets are allowed to enter the connection. There are two operations on the Minimum Bandwidth Window and the Maximum Bandwidth Window. The first type of operation is an augmented window. When a repeated ACK is received, the maximum bandwidth window and the minimum bandwidth window are expanded by one. When a non-repetitive ACK is received, the maximum bandwidth window and the minimum bandwidth window are reset to their original size (for the maximum bandwidth window, the maximum allowed bandwidth X round trip time (RTT), for the minimum bandwidth window, The guaranteed minimum bandwidth X round trip time) and move to the right, so the first • byte in the window is the first unacknowledged byte. Even when an ACK is lost, the expanded window ensures that the connection can send a new packet. This action prevents the connection from being suspended but does not increase the number of packets in the network. A normal TCP algorithm is executed when the transmission rate is between the set limits. If the send rate is below the minimum ,, the minimum bandwidth window is called. If the send rate is below the maximum 値, the Maximum Bandwidth window is called. In the invention, a retransmission timer period is utilized, which has a range greater than the window transmission timer (after this time, even if the status of the TCP congestion window 1308012 is generally not allowed to transmit the packet, the packet can be sent) and The time is right. The retransmission timer period ensures that even when no ACK is received, 'the packet with the lowest rate is still sent to stimulate the ACK, and finally retransmitted - not suspended by the TCP timeout. Therefore, the total flux is increased. The invention increases throughput by utilizing the appropriate window size in different situations and prevents unnecessary timeouts. In the present invention, the algorithm calculates an appropriate RTO in the case of different channels to prevent unnecessary timeouts. The φ invention was not specified for wireless networks. When the wireless channel changes significantly and quickly, the algorithm of the invention will not adapt quickly to this situation and will create unnecessary timeouts. In the present invention, the round trip time is modified based on the channel conditions of each transmission. This adapts to the current situation faster and significantly reduces unnecessary timeouts. Also, the U.S. patent "Enhancement of explicit congestion notification (ECN) for wireless network applications" discloses the use of enhanced explicit congestion notification (ECN) techniques to avoid unnecessary delays in packet transmission from low-bandwidth delay sensitivity TCP connections. This error still causes the system to enter a blocked state. This type of congestion occurs with high BER rather than lack of buffering. When these errors occur, the TCP error assumes that the network is congested and greatly reduces the transmission rate of its old and new packets. The main work of the enhanced ECN is to distinguish the loss of the packet from the BER due to the loss of the BER packet, and refuse to enter the low start state when the BER causes the packet to be lost, and reduce the transmission speed when the early congestion notification is detected. To increase the amount of connected connections. 9 1308012 This invention focuses on the resolution of the actual congestion caused by high BER congestion. However, due to the variation of the delay characteristics of the single channel of the bandwidth, it is not taken into consideration, and the bandwidth variation causes other types of congestion. In the present invention, we are working to distinguish the congestion caused by the bandwidth variation. The real congestion. In summary, when the bandwidth variation occurs in TCP transmission, the original RTO measurement method cannot work effectively. The RTI estimation is updated by the new RTT 0 measurement method and will approach the RTT after a long stable transmission. . The RTT changes the base at the transmission rate. When the transmission rate changes from high to low, the RTT may exceed the RTO timeout threshold and thus trigger retransmission. This retransmission is a spurious transmission because it is not triggered by an untransferred packet but by a change in RT0. This spurious transmission results in unnecessary transmission of the data, which is not lost. The associated crowded window is reduced to a maximum node size (MSS) and the slow start threshold will be cut by half of the current window size. As a result, the channel is not being used effectively. SUMMARY OF THE INVENTION The inventors of the present invention have proposed a method for predicting adaptive RTO in a wireless communication system in view of the above-mentioned shortcomings of the prior art, wherein an adaptive timer-based control algorithm can enhance TCP execution. Performance, the dynamic change of the algorithm's round-trip time (RTT) is evaluated and adapted to the current bandwidth state. Since the corrected retransmission timeout (RTO) will be calculated by the re-adaptive RTT estimate, the false RTO of the incorrect RTO caused by the bandwidth change of 10 1308012 can be used, and also Improved TCP execution performance over wireless transmissions - [Embodiment of the preferred embodiment] - 'In the original congestion control flow, TCP is blocked by the following steps: First, the base station transmits a packet, and Calculate the RT0. Second, if the round-trip time (RTT) of the packet is greater than RT0, the | is considered to be choking and entering the (slow) slow start state. In the wireless network, the strong and weak channel causes the bandwidth to change. The bandwidth suddenly becomes smaller than before, and the frequency with a smaller bandwidth requires more time to transmit these packets, which means that the RTT of this packet will become very long. When the time between the transmission of the TCP packet and the recognition of the packet is greater than RT0, the type of congestion that occurs in the congestion type is different from that of the wired network. It causes the source frequency to oscillate. If we can avoid this type of congestion, we can • slam the window into the slow start state (SLOW START PHASE ). The adaptive RTO algorithm is proposed to inform the TCP of the frequency at which the necessary TCP packets are timed out and enters the slow start state. The main purpose of this algorithm is to detect bandwidth variations and calculate RTO. It will send a new RT0 back to TCP to avoid RT0 mis-estimation. The original RTT and RT0 estimation schemes were proposed for wired loops and the wired environment is not considered to have an unstable transmission rate (the TCP is de-detected. If the channel is to be sealed, TCP will be blocked. The newer error, especially 11 1308012 is due to the instability of bandwidth variation). The calculation is as follows: R=-xR + ^xRTT 8 8 Z) = ^xZ) + ^x abs(R -RTT)
RTO=R + 4xD 其中：RTT爲封包往返時間的測量 P爲用於封包傳輸率的評估 D爲RTT之偏差評估器（estimator) RTO爲往返時間RTO=R + 4xD where: RTT is the measurement of packet round trip time P is used for evaluation of packet transmission rate D is RTT bias estimator (estimator) RTO is round trip time
首先，這三個公式的常數係被微調以用於有線網路。 然而’其可能不適用於無線網路，再者，由於頻寬震盪 可能導致錯誤評估，我們需要增加一個或更多個因子與 頻寬進入公式中，去修該這些公式。 頻寬變動可以是正的也可以是負的，且RT0將會基 於頻寬的變動而改變，對於各個基地台控制器之RT0 之計算公式如下： RTT_esti = (RTT esti-pradelqy)*{^reVl〇US~^)+pradelay (1) _ present _bw v ’ RTO = RTT _ esti + Deviation _ index x Deviation (2) 其中 pro-delay = propgation_也1町=RTT -(^^-) (3) 並且 ^ 爲有線阔路上之資料傳輸延遲 在方程式（3 )中，RTT減去傳輸延遲係等於全部的 傳輸延遲（包含無線延遲與有線延遲），由於有線之傳 輸延遲很小，我們可以忽略他。將兩個因子Previous_bw 與 present_bw加進方程式（1 )中，則可得到RTT之 較佳的評估，來計算新的RT0。假使頻寬（Present_bw) 12 1308012 是小的’我們就可得到較大的RTT_esti，反之亦然 使僅增加RTO値，則可避免壅塞逾時。然而，一 ' 生網路壅塞，TCP必須花費更多的時間來等待重傳 此，適應性TCP演算法在此議題中被考慮。此外， 計算使用了原來的方程式，來確保新的RTO對於 的有線網路係恰當的。 方程式（2 )之型態係近似於原始的TCP，但是 0 以常數4來取代Deviation_index。該常數4爲具有 傳輸率之有線網路的微調値。在無線網路中，我們 臨因爲頻寬變動導致的時強時弱的頻道特色，所以 能爲一個常數，而是一個頻寬的函數。對於該偏差 的方程式，將使用和有線伺服器相同的値。 爲了更理解本發明，請參閱第1圖，如圖所示， 應性RTO演算法具有比原始的TCP協定更佳的輸 並且在速率改變的短週期中更具增效。原始的TCP 0 具有壅塞逾時問題，另一方面，適應性的RTO演 具有幾乎相同的執行效能，其乃該演算法消除了來 頻寬變動的壅塞逾時。 接著再比較兩個不同的頻寬模型，該有限的突發 (burst)模型具有比均勻（uniform)分配模型更 改良。這是因爲有限的突發傳輸模型之逾時事件比 勻分配模型中多的緣故。在有限的突發傳輸模型中 塞逾時在當頻寬從高變至低時發生（約50% )。但 。假 旦發 。因 RT0 原來 我們 穩定 會面 其不 方面 該適 出， 協定 算法 自於 傳輸 好的 在均 ，壅 在均 13 1308012 勻分配模型中壅塞逾時僅在頻寬下降至9.6Kbps或 19.6Kbps (少於20%)時發生。該適應性的RTO演算 / 法能夠消除大部分的在突發傳輸模型中的逾時’且輸出 丨 效能更被提升。 接著請參閱第2圖，如圖所示，該適應性的RTO演 算法能夠消除壅塞逾時且維持通常逾時時間於所有情 況中。原始的TCP協定在不同的情況中具有不同的逾時 時間。在不同的頻寬模型裏，有限的突發傳輸模型比均 勻分配模型具有更多的逾時，此乃因爲在有限的突發傳 輸模型中存有壅塞逾時之較高的可能性，在不同速率改 變的週期中，當週期較大時，就有比較少的逾時的次數。 本發明提出的演算方法能有效的解決非壅塞產生的 逾時問題，其於無線系統中具有顯著的影響。在無線資 料傳輸中，該系統具有較高的頻寬變動，當頻寬從高掉 至低時，則將發生壅塞逾時。但採用了本發明之適應性 # RTO演算法之後，系統可以動態性的改變RTT評估値， 來適應目前的頻寬狀態，對應的RTO値將因而被計算。 因此，本發明所提出之適應性的RTO演算法能夠消除 因爲頻寬變動所導致的壅塞逾時，並增進輸出效能。 本發明所提出之方法確實解決了存在於先前技術當 中的瓶頸，且具新穎性與進步性及產業利用性，敬請核 准專利，實感德便。 14 1308012First, the constants of these three formulas are fine-tuned for use in wired networks. However, it may not be suitable for wireless networks. Furthermore, since bandwidth fluctuations may lead to erroneous evaluation, we need to add one or more factors and bandwidths to the formula to fix these formulas. The bandwidth variation can be positive or negative, and RT0 will change based on the bandwidth variation. The formula for RT0 for each base station controller is as follows: RTT_esti = (RTT esti-pradelqy)*{^reVl〇 US~^)+pradelay (1) _ present _bw v ' RTO = RTT _ esti + Deviation _ index x Deviation (2) where pro-delay = propgation_ also 1 town = RTT - (^^-) (3) and ^ Data transmission delay for wired wide road In equation (3), RTT minus transmission delay is equal to the total transmission delay (including wireless delay and wired delay). Since the transmission delay of the cable is small, we can ignore it. Adding the two factors Previous_bw and present_bw to equation (1) gives a better estimate of the RTT to calculate the new RT0. If the bandwidth (Present_bw) 12 1308012 is small 'we can get a larger RTT_esti, and vice versa, if only RTO値 is added, the congestion time can be avoided. However, for a network congestion, TCP must spend more time waiting for retransmissions. Adaptive TCP algorithms are considered in this topic. In addition, the calculations use the original equations to ensure that the new RTO is appropriate for the wired network. The equation (2) is similar to the original TCP, but 0 replaces Deviation_index with a constant of 4. This constant 4 is a fine tuning of the wired network with a transmission rate. In the wireless network, we are due to the strong and weak channel characteristics caused by the bandwidth variation, so it can be a constant, but a function of bandwidth. For the equation for this deviation, the same 値 will be used as the wired server. To better understand the present invention, please refer to Figure 1, which shows that the RTO algorithm has a better transmission than the original TCP protocol and is more efficient in the short period of rate change. The original TCP 0 has a congestion timeout problem. On the other hand, the adaptive RTO performance has almost the same performance, which eliminates the congestion timeout of the bandwidth variation. Then compare two different bandwidth models, the finite burst model is more improved than the uniform distribution model. This is because the time-out event of the limited burst transfer model is more than the evenly distributed model. In the limited burst transfer model, the plug-over time occurs when the bandwidth changes from high to low (about 50%). but . False hair. Because RT0 turns out that we have a stable meeting, the agreement algorithm is good. Since the transmission algorithm is good, the congestion time in the 13 1308012 evenly distributed model only drops to 9.6Kbps or 19.6Kbps (less than 20). %) occurs. This adaptive RTO calculus/method eliminates most of the timeouts in the burst transfer model and the output 效能 performance is improved. Next, please refer to Figure 2, which shows that the adaptive RTO algorithm can eliminate the congestion timeout and maintain the usual timeout in all cases. The original TCP protocol has different timeouts in different situations. In different bandwidth models, the limited burst transfer model has more time-over than the even-distribution model because of the high probability of congestion over time in a limited burst transfer model. In the period of rate change, when the period is large, there are fewer times of timeout. The calculation method proposed by the invention can effectively solve the problem of overtime caused by non-blocking, which has a significant influence in the wireless system. In wireless data transmission, the system has a high bandwidth variation, and when the bandwidth is from high to low, a congestion timeout occurs. However, after adopting the adaptive # RTO algorithm of the present invention, the system can dynamically change the RTT evaluation 値 to adapt to the current bandwidth state, and the corresponding RTO 値 will be calculated accordingly. Therefore, the adaptive RTO algorithm proposed by the present invention can eliminate congestion due to bandwidth variation and improve output performance. The method proposed by the present invention does solve the bottleneck existing in the prior art, and has novelty and advancement and industrial utilization. Please verify the patent and feel it. 14 1308012
【圖式簡單說明】 第1圖爲本發明之傳輸率與頻寬改變週期之比較圖。 第2圖爲本發明之逾時次數與速率改變週期之比較圖 【主要元件符號說明】 J \ 15BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a comparison diagram of a transmission rate and a bandwidth change period of the present invention. Figure 2 is a comparison diagram of the timeout period and rate change period of the present invention. [Main component symbol description] J \ 15