A new stopping rule for turbo decoders

The iterative decoding process to decode turbo codes is rather time consuming and stopping rules are applied to prematurely quit the process, for the decoder Might have already reached a decision in an earlier iteration. In this paper we set up a new soft stopping ride based on probability density evolution of the extrinsic information. We adopted a model in which the evolution is expressed through a figure of merit defined by the evolving mean of the extrinsic's pdf. A value of this figure is then used as a threshold to stop the decoding process. The performances of some threshold values are compared with the performance of a "magic" rule and of the standard decoder using fixed numbers of iterations. The criteria used for performance comparison are decoding speed, error performance and computational complexity. The results show a significant increase in the average decoding speed at almost no cost concerning the performance. The negligible computational requirements give this ride an advantage.