利用預棄技術和狀態再利用的技巧實現低功率渦輪解碼器

Abstract

考慮傳輸通道品質的情況下，為了使晶片能量能有效率的被使用，而提出了新的想法“提早放棄”以利用在渦輪解碼的流程上。渦輪解碼器是利用反覆的方式，完成解碼的動作，解碼的次數和通道的品質高度相關。如果通道狀況良好，可以利用提早停止的方式，停止解碼程序。然而，當通道品質惡劣，傳統的做法是解到最大解碼次數之後，若發現解碼失敗，再要求傳送端重傳該封包。因為解碼所消耗的能量會正比於解碼迴圈的次數，所以就浪費能量在無意義的迴圈上。我們提出的作法是放棄可能解碼失敗的封包其解碼程序，以節省不必要能量的浪費，並能及早重傳。利用模擬觀察渦輪解碼資料，我們增加一個簡單的硬體，檢查解碼中所產生的外部資訊，當作判斷提早放棄的機制。除此之外，我們也提出在假設重傳的封包資料一致的情況下，如何再利用之前提早放棄解碼的計算，使得整體解碼所需的迴圈次數下降。我們實驗結果顯示，在通道狀況惡劣的情況下，綜合我們所提的想法，對達成一定數量成功解碼的封包，可以減少平均10%到37%的解碼所需迴圈次數。

A novel early give-up algorithm for turbo decoding process undergoing poor channel SNR is proposed for energy-efficient consideration. Turbo decoding involves an iterative process and the number of iterations required to correctly decode the information packet is highly dependent of the channel condition. If the channel SNR is good enough, the iterative process could likely be reduced, i.e. early stop or termination. When the channel is contaminated at the time of transmission, the process will keep going until a maximal number of iterations is reached, and request a packet to be re-submitted. Because energy consumption of the decoding algorithm is proportional to the number of iterations, this would cost extra energy resource. The proposed approach is to give up the decoding process earlier during bad channel SNR and request data to be re-sent immediately. Based on observations from the simulations of turbo decoding process, a simple hardware checking the average absolute value of the extrinsic information on-the-fly is involved into the original turbo decoder architecture. Besides, we apply another technique to reuse the prior MAP information of the given-up process based on the assumption of correlation between same packets transmitted at different times. Our results shows that the average iterations required to decode fixed amount of valid packets can be reduced from 10% to 37% under bad channel conditions.