Stopping Rules for Turbo Decoders

Abstract

Decoders for turbo codes are iterative in nature, i.e., they have to perform a certain number of iterations before reaching a satisfactory degree of confidence regarding a frame to be decoded. Until now standard turbo decoders have used a fixed number of iterations. In this article, we propose some simple stopping rules that can be used to reduce the average number of iterations. This technique ofiers a trade-ofi between speed and performance and can provide a significant increase in the average decoding speed while not sacrificing decoder performance. We tested several types of stopping rules for turbo decoders. One type is based on comparing decoded bits (hard bit decisions) with previous decoded bits; a second type is based on comparing reliabilities (soft bit decisions) with a threshold; and a third type uses a cyclic redundancy check (CRC) code applied to hard decoded bits. We simulated turbo decoder performance using these rules (including several variations of the first two types) and further required that the decoder cease after 20 iterations if the stopping rule is not yet satisfied. Specifically, we analyzed the decoder-error rates and the average number of iterations for each rule. We found that the average number of iterations was roughly between 4 and 7 for a bit signal-to-noise ratio, Eb/N0, near the waterfall threshold, as compared with b 0 the 10 fixed iterations used by the current turbo decoder. In addition, the resulting error rates were noticeably lower than those for 10 fixed iterations, and in fact were very nearly equal to the error rates achieved by a decoder using 20 fixed iterations.

Keywords

Details

Volume

42-142

Published

August 15, 2000

Pages

1–22

File Size

497.2 KB