A branching process analysis of the average number of computations of the stack algorithm

A new analysis for bounding the average computational effort of sequential decoding is presented. It is based on a branching process model of the subset of incorrect paths explored by the decoder. Closed-form expressions formulated in terms of the set of branch metrics and their probability assignments as actually used by the decoder are derived. Code characteristics in the form of Hamming weights of the 'incorrect paths or column distance function are incorporated, making the analysis directly applicable to specific cases. Evaluation of the bounds is not difficult. Results are in good agreement with those obtained by lengthy computer simulations.