Convergence properties of iterative turbo decoding based on the number of searched states

Abstract

We investigate the evolution of the trellis path through the state diagram of a turbo code versus the number of iterations n/sub i/. Simulations show that the BER and the number of states with probability above a given threshold have the same convergence behavior versus n/sub i/. As a consequence we propose a new criterion to stop the iterative decoding process. A new algorithm to avoid the channel parameters estimation is proposed too. The joint combination of the proposed algorithms perform the same as the optimum solution with reduced computational complexity.