Partitioning for SA(B,C) for CPM signals

The SA(B,C) algorithm is considered for continuous phase modulation (CPM). The SA(B,C) detector performs a breadth first maximum likelihood sequence detection in the trellis under structural and complexity constraints. Given a trellis, the SA(B,C) detector partitions the states into C classes and for every time interval selects B paths into each class. To find a partition that minimizes C for a given B and CPM system, such that the performance (error event probability) approaches unconstrained MLSD when the signal to noise ratio goes to infinity, is an NP-hard problem. Here a suboptimum method for finding the partitioning is considered. Substantial savings in complexity can be made by using the SA(B,C), while achieving almost optimal (unconstrained) error performance