在無線網路傳輸上基於馬可夫決策利用自動重傳機制及錯誤更正碼以達到彈性的傳送方式

Abstract

在即時無線網路環境之下，我們結合FEC和ARQ的方式以達到有彈性的通路。因為網路環境通常沒有足夠的頻寬來正確的傳送所有的封包，所以傳輸端就必須根據封包的重要性以及時間限制來決定是否要傳送此封包或提早丟掉不傳。我們延續廖教授的研究。廖教授考慮相同的問題但是只使用了ARQ的機制。在我們的論文中進一步考慮使用RS code和傳送不同長度封包的問題。同樣的，我們也是將系統模擬成馬可夫決策的問題，我們更利用了policy-iteration 的方法找出一個好的決策。在我們的論文中，傳送端做的動作不只是事先丟掉多少封包，還包括了要用多少的code rate傳輸。此外，每次可能傳送不同大小的封包，這會使得定義worse case變的更加困難。在我們的模擬中，我們介紹另一種方法用來比較MDP做出來的方法，因為此方法只考慮通道的影響，所以利用MDP會得到比較好的結果。

We consider using an adaptive hybrid of FEC and ARQ via Reed–Solomon (RS) code for real-time flexible communication over fading wireless channel. Because quite often there is not enough bandwidth to correctly transmit the packet in time, so it is worthwhile to decide to send or early drop the packet according to the importance of the packet and its timing constraints. To handle such a problem, we extend Liao and Yeh’s approach which deals with the same problem but with only ARQ in use. However, our problem in hand is made more difficult by the fact that the bandwidth demands for packets are different because the size of packet is lengthened if the RS code is used. As their approach, we also formulate the problem as a Markov-Decision problem and we also use the policy-iteration method to find out a good policy. The main difference from theirs is that the action in our policy is in the form of not only dropping packets but also the code rate for packet to be sent. Besides, the worse case in our design is more complicated than that in theirs by involving the consideration of different packet size. In our simulation, as a comparison we introduce another method which adapts the RS code only according to the channel state, and we show that our policy from MDP is better.