EECS CSPL SEMINAR SERIES
FALL TERM 1999

Lei Wei

Dr. Lei Wei

Faculty of Engineering and Information Technology

Australian National University

Canberra, ACT, Australia

Lei.Wei@anu.edu.au

Monday, October 4
4:00 - 5:00 P.M.
Room 1200 EECS

Iterative Viterbi Decoding Algorithm
-A way to achieve near Shannon limit decoding for current standard systems

Abstract-
Over the last 30 years, the Viterbi algorithm and the Forney concatenated systems have been widely applied in digital communications and commonly adopted in many current standards. Recently, excitement about the turbo codes was sparked by their close approach to the Shannon limit (within 0.7 dB of the limit) with a very low decoding complexity. Can we achieve near Shannon limit decoding without (or with very little) modification of the current standard systems? Yes, in this seminar we will present our recent research results on the iterative Viterbi algorithm.

The seminar is divided into three parts. In the first part, we will study the algebraic properties of the Forney concatenated codes (i.e., using a convolutional code as the inner code, a block code as the outer code and a block interleaver). These properties are important to construct good iterative decoding algorithms. In the second part an iterative decoding algorithm based on the famous Viterbi algorithm (thus, called iterative Viterbi algorithm) is proposed to decode the codes. We also analyze the performance of the IVA and conclude better performance could be obtained if we replace the powerful block codes by some simple parity codes.

At the third part, we will present some numerical results for two cases. Case (a) is the CRC concatenated code used in the Qualcomm CDMA system (CC + 12-bit CRC for a block length of 192, which consists of 172 information bits, 12 CRC bits and 8 tail bits). We will show that without any modification of the system except applying the IVA, the performance (Eb/N0) can be improved by 0.6 dB at a packet error rate of 1%. Simply replacing the 12-bit CRC code by a simple parity code, the performance can be improved by 1 - 1.2dB, which is only 0.9 dB from Shannon sphere packing bound. Case (b) is the Forney concatenated code (CC+ (255,231) BCH). We show its performance using the IVA is better than current NASA system. Finally, we show that for block lengths ranging from 56 information bits to 11970 information bits, the Forney codes can achieve a performance about 1 dB away from the Shannon sphere packing bound for block error rate of 10^-4 and discuss the advantages and disadvantages of the IVA, compared with turbo codes.


Biosketch-
Dr. Lei Wei received ME and Ph.D degrees in electrical engineering from University of New South Wales and University of South Australia, Australia, in 1993 and 1995, respectively. Since 1995, he has been with the Australian National University (ANU). Now he is a senior lecturer (= associated professor in US system) at Department of Engineering, FEIT, ANU. Over the last few years, he has been working in areas of error control coding, multiuser detection, fast simulations for markov systems. He proposed the hierarchical decoder for decoding long convolutional codes and the iterative Viterbi algorithm for near Shannon limit decoding.

return to CSPL Seminars homepage