Speed and accuracy comparison of techniques to solve a binary quadratic programming problem with applications to synchronous CDMA

Abstract

We compare methods for solving the NP-hard binary quadratic programming (BQP) problem. Various methods are discussed, including box-constrained quadratic programming, branch and bound, coordinate descent, group decision making and semi-definite relaxation. An algorithm from target-tracking, the probabilistic data association filter, is modified to the BQP application. Simulation results show that this and several other methods can significantly outperform the decision feedback detector (DFD) or its group counterpart, GDFD.