Convergence and errors in turbo-decoding

Turbo-codes are decoded using iterative decoding algorithms with somewhat elusive convergence properties. In this paper, the modes of convergence observed in an extensive series of turbo-code simulations are categorized by examining the bit convergence behavior for each frame. Based on these results, a new method for decoder termination based on average log-likelihood ratios is presented and compared with other methods. As an application, a selective repeat automatic repeat request (ARQ) system is considered.