A New Algorithm for Multiuser DetectionBased on a Gradient Guided

| This paper introduces a new multiuser detection algorithm based on a gradient guided search that can achieve near-optimum performance while its implementation complexity is linear in the number of users. The new algorithm attempts to perform jointly optimum multiuser detection by updating one user's bit decision each iteration in the best possible way. When a decorrelating multiuser detector was employed to initialize the proposed algorithm, in all cases tested the gradient guided search converged to a solution with performance very close to optimum. Further, tests indicate that the proposed algorithm provides excellent performance for cases where other suboptimum algorithms perform poorly. Although we focus only on synchronous multiuser detection, the algorithm is equally applicable to asynchronous multiuser detection problems. I. Introduction Communication systems employing code division multiple access (CDMA) continue to attract attention 1, 2]. CDMA systems exhibit a limit on the maximum number of users that can simultaneously communicate over the channel for a spec-iied level of individual user performance. This limitation is brought about by the ultimate domination of the multiple access interference (MAI) over the noise 7]. Over the years researchers have sought ways to extend the user capacity of CDMA systems by employing optimum multiuser detectors, linear multiuser detectors and interference cancellation (IC) detectors 3]. With regard to the optimum detectors, the work of Verdu 4] is the most cited in the literature and the one upon which much of the other work is based. While indeed such optimum detectors ooer signiicantly improved performance, they unfortunately suuer from the fact that the combinatorial optimization problem for multiuser detection is NP-hard and its complexity grows exponentially with the number of users unless the crosscorrelation matrix has a special structure 1]. Therefore, the computational complexity of optimum detectors prohibits their implementation in many practical applications. However, their performance can serve as a benchmark against which to compare other suboptimum schemes. Recognizing the complexity of optimum detectors, a class of linear detectors 5], for instance the decorrelating detector and the MMSE detector, have been proposed which achieve