在多輸入多輸出高速下行封包擷取系統中採用乏晰Q-Learning技術之混合自動重傳機制

Abstract

為了提供更高速的資料傳輸與更有效的資源利用,第三代合作夥伴計畫（3rd generation partnership project, 3GPP）提出了多輸出多輸入高速下行封包擷取技術（Multiple input multiple output high speed downlink packet access, MIMO HSDPA）來提供更高速且安全的下鏈路資料封包傳送。在3GPP的規格裡面,有一個重要的服務品質要求：根據通道狀況決定原始傳送的MCS時，必須使封包的錯誤率小於0.1。因此，我們提出在多輸出多輸入高速下行封包擷取技術之混和自動重傳機制下,採用乏晰Q 學習演算法來解決這個問題（FQLM-HARQ）。乏晰Q學習法同時結合了乏晰邏輯運算與Q學習法的優點。在此,我們將混合自動重傳機制（hybrid automatic retransmission request, HARQ）程序模擬為離散時間馬可夫決策過程（Markov decision process, MDP）。根據BLER的表現,乏晰系統規則會設計成不同的部分，來實現BLER的服務品質需求。Q學習演算法可以在不同的環境下,經由不斷的學習,選出最適當的MCS並且修正乏晰系統規則。在學習之後，我們期望多輸出多輸入高速下行封包擷取系統可以達到最高的資料輸出而又不違反QoS需求。 從模擬結果可知，所提出的FQLM-HARQ機制在通道訊息延遲的情況下，可以達到最大的輸出並且盡力維持BLER的要求。相較於其他系統，FQLM-HARQ可以更適應通道的變化。

Multiple input multiple output high speed downlink packet access (MIMO HSDPA) system is proposed by 3rd generation partnership project (3GPP) to provide higher transmission data rate and more resource utilization. An important QoS requirement defined in spec is to choose a suitable MCS based on the channel quality indicator while maintaining the initial block error rate (BLER) smaller than 0.1. Therefore, we proposed a fuzzy Q-learning based MIMO HARQ (FQLM-HARQ) scheme for MIMO HSDPA system to solve this problem. The FQLM-HARQ scheme can take the advantage from both fuzzy logic and Q-learning. Here, the HARQ scheme is modeled as a Markov decision process (MDP). The fuzzy rule is designed to separate different parts according to the BLER performance and the Q-learning algorithm can learn the optimal MCS under different environment. After learning, we can expect the MIMO HSDPA system with higher throughput while not violating the BLER requirement. From simulation results, the proposed FQLM-HARQ scheme can achieve higher system throughput and endeavor to maintain the BLER requirement with channel quality indicator delay consideration. Comparing to other traditional schemes, the FQLM-HARQ scheme can accommodate well in channel variation.