Performance analysis of (M,L) algorithms

The performance of the (M,L) algorithm applied to the decoding of convolutional codes is investigated. Asymptotic analytical results and simulated points for low values of the signal-to-noise ratio are presented. Different versions of the algorithm are compared; in particular, the influence of the decoded path continuity on the performance is explored. The results obtained by simulation at low signal-to-noise ratios show that optimal detection can be achieved with a saving in the number of retained states by a factor of two. Asymptotically, a memory saving of one order of magnitude is foreseen from the analysis.<>