Eliminating-highest-error and fastest-metric-descent criteria and iterative algorithms for bit-synchronous CDMA multiuser detection

In this paper, the approach of iterative, metric descent algorithms to bit-synchronous CDMA multiuser detection is proposed. An iterative algorithm in the sequential flip mode is presented. Then the fastest-metric-descent (FMD) and eliminating-highest-error (EHE) criteria are proposed for the determination of the bit to be flipped at each time. The FMD criterion achieves the largest amount of metric descent at each bit flip, while the EHE criterion achieves the highest probability that the bit flip is correct. In perfect power control, the FMD and EHE criteria are equivalent. Two algorithms under the FMD and EHE criteria are proposed for matched filter outputs. The EHE based algorithm for decorrelating noise whitening filter outputs is also proposed. It is shown that in a finite number of bit flips the proposed four algorithms converge to fixed points. It is also shown that these fixed points are local minimum points of the metric, and vice versa. When a user is strong enough, it approaches the single user bound. If the signature waveforms are orthogonal, all users achieve the single user bound. Simulation results show that the FMD and EHE based algorithms have improved performance. The proposed algorithms are attractive in the near-far situation. They have linear computational complexity per demodulated bit in the number of users and are suitable for implementation in Hopfield neural networks for fast parallel computation.