在無線影像傳輸上使用自動重傳對於TMN8的Frame-Layer 速度控制機制的影響

Abstract

我們都知道無線通道傳輸易受所在位置、時間和連續錯誤等特性影響傳輸品質。所以在本篇論文中主要研究在緩慢Rayleigh衰減通道中對於影像傳輸所造成的影響並且用混合式自動重傳機制來改善影像的品質。因為在傳送端接收到錯誤重傳的訊號時需要時間來重傳錯誤的封包，這樣的話就會降低通道的有效輸出。如此一來相對的就會造成影像編碼器中暫存器的快速累積，並且有可能導致TMN8速度控制法則降低每張影像frame的位元個數及skipped frame的個數的增加。為了改善這個問題，我們利用了將Gilbert模型和回饋通道所傳回的ACK/NAK考慮進TMN8的速度控制法則的改良型速度控制法則來解決所面臨的問題。 主要概念就是利用過去發生傳輸錯誤的數量來決定現在目前通道狀況的好壞並且用簡單化的Gilbert通道來預測未來一段時間內平均的輸出，有了這些消息後我們就可以用來預先在所編碼的每張frame中扣除所可能會造成重傳量，如此一來即可改善因為錯誤重傳所導致影像編碼器暫存區中的快速累積。我們是利用GE 模型來模擬實際通道傳輸的狀況，假使在較低的訊號雜音比下我們使用通道編碼來保護資料，那麼在GE模型中所求出的參數也會反映出無線通道在有編碼和沒有編碼的差異。而當接收的訊號雜音比變差時，位元錯誤率會提高導致接收端更容易收到錯誤的資料。這時我們會考慮利用BCH碼的解碼能力來提高有效輸出。雖然在通道狀況變差時需要降低資料速率當作保護位元來保護資料，但是我們可以從模擬中所得到的訊號雜音比對有效輸出的圖看出這樣的犧牲是值得的。我們也可以從訊號雜音比對有效輸出的圖形得知在哪個訊號雜音比下使用多強的通道編碼來保護資料是有效率的。所以當訊號雜音比變差時，我們就可以再結合預測模型和通道編碼來比較出在有編碼和沒有編碼的條件下skipped frame個數、每一張frame所可壓縮的位元數和影像編碼器暫存器的影響。

Wireless channel transmission is known to suffer location-dependent, time-varying and burst errors. We study the problem of rate control for video transmission over wireless flat and slow Rayleigh fading channels, and use the hybrid ARQ retransmission scheme to improve the video quality. After receiving NAK information, sender needs time to retransmit the relative error packet. It will reduce effective channel throughput and the video encoder buffer may fill up quickly and lead the TMN8 rate-control algorithm to reduce the bit allocation of each frame and increase the number of skipped frame. We use a modified rate-control scheme that takes into consideration an a priori channel model, i.e. Gilbert channel model and channel feedback information, i.e., ACK/NAK to diminish the problem of filling up the encoder buffer and many skipped frames. After the acknowledgments received through the backward channel, we can estimate of the current channel states and use simplified Gilbert channel to predict the future channel throughput. We combine these two kinds of information from a priori channel condition and statistically predict the future channel throughput to reach our goal. The method we used to simulate channel transmission is GE model. GE model is the extension of Gilbert model. If we use channel code to protect data in the lower SNR, the parameter of GE model would also reflect the difference of with coding and without coding. When the received SNR becomes lower, the bit error probability would increase and the receiver would receive more erroneous data. We will consider using correcting ability of BCH code to increase effective throughput. Although it needs to decrease source data rate to form redundant bit to protect source data when channel condition become worse, it is worthy from comparison of the effective throughput between different code rate of BCH code in different SNR. From the comparison of the effective throughput, we could know that choosing which code rate of BCH code to encode source date in which SNR would be more effective. When the received signal is in lower SNR, we can use prediction model and coding method to see the difference of number of skipped frames, target number of bit and buffer fullness between with coding and without coding.