標題: 高速渦輪解碼器晶片設計及其在CCSDS系統上的應用
High Throughput Turbo Decoder Chip Implementation for CCSDS System Applications
作者: 莊翔琮
Chuang, Hsiang-Tsung
方偉騏
Fang, Wai-Chi
電子研究所
關鍵字: 渦輪碼;提早停止;迭代;基數-4;Turbo Codes;Early Stopping;Iteration;Radix-4
公開日期: 2008
摘要: 由於渦輪碼有著優異的錯誤更正能力,所以在近十年來已經被廣泛的運用在通訊系統上。然而由於渦輪碼複雜的結構使得其速度無法有效提升,本論文將改善解碼器的架構使渦輪解碼器速度有效提升。 由於渦輪碼的時脈是被遞迴結構所限制的,我們利用偏移加法-比較-選擇器和一級CSA的架構,來減少主要路徑延遲;除此之外,我們更進一步提出了hybrid 4-inputs addition/subtraction基數-4的遞迴結構使得此架構的吞吐量和傳統的遞迴結構相比有近80%的提升。另一方面,傳統渦輪解碼必須跑到固定次數的迭代以確保事前資訊已經收斂,但如此一來造成速度慢,高延遲和功率浪費。事實上,當通道狀況好的時候,渦輪解碼會提早收斂,因此,藉由分析,我們選用HDA2提早停止方法來降低迭代次數來達到高吞吐量的目的。 根據實驗分析,此渦輪解碼器在UMC90 nm製程下最高能達到的時脈頻率為357.14MHz,以及在單塊MAP解碼器之下,渦輪解碼器能達到77.62MS/s的傳輸速度,晶片面積為1.59mm2。另外,由於平行化的渦輪解碼會發生記憶體碰撞的問體,我們可以利用修正過的退火演算法將這問題解決,並且在十四塊MAP解碼器之下,渦輪解碼器能達到884.91MS/s的傳輸速度,晶片面積為17.64mm2。
Turbo codes have been applied widely in communication systems over the last decade due to its excellent error correction ability. However, because of complex structure, the data rate of turbo decoder could not improve more efficiently. Therefore, the thesis presents improved architectures to increase its data rate. The operating frequency of turbo decoder is greatly limited by the recursion unit. In order to decrease the critical path delay, the OACS and one stage CSA structure is employed. Furthermore, the hybrid 4-inputs addition/subtraction radix-4 recursion architecture is presented for CCSDS turbo decoder and finally the relative throughput of proposed recursion unit is faster than traditional one around 80%. On the other hand, the decoding process has to run a certain number of iterations to ensure the extrinsic have converged. In fact, turbo decoder may converge earlier when the channel condition is good. Hence, an early stopping criterion could be employed to reduce the number of iterations. After chip implementation in 90nm process, the maximum clock rate 357.14MHz can be achieved, and the 1.59mm2 core area can support the maximum data rate 77.62MS/s of turbo decoder with single MAP decoder. Besides, if the parallel MAP decoders are considered, the memory collision could be happened. We can introduce the modified annealing algorithm to solve the collision problems. The 17.64mm2 core area can support the maximum data rate 884.91MS/s of turbo decoder with fourteen MAP decoders.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009411647
http://hdl.handle.net/11536/80559
顯示於類別:畢業論文


文件中的檔案:

  1. 164701.pdf

若為 zip 檔案,請下載檔案解壓縮後,用瀏覽器開啟資料夾中的 index.html 瀏覽全文。