Investigation of an iterative groupwise soft input/soft output multiuser detection algorithm

This thesis investigates the performance of a novel iterative soft input/soft output groupwise multiuser detector (MUD). In general, multiuser detectors improve performance by explicitly accounting for the interference when detecting all users together. In this thesis, the interference is assumed to be a product of users whose channel gains are known. The novel technique extends the group detector MUD with an iterative process, and is aptly named iterative MUD (IMUD). The success of IMUD is measured by how close it can get to the performance of the computationally prohibitive optimum technique (joint maximum likelihood MUD) with as little complexity as possible. To do this, IMUD was compared to existing MUDs in terms of complexity and performance. Since MMSE V-BLAST was seen as the best low complexity competitor, it is used as a performance reference.; The results from IMUD with equal numbers of users and receivers were excellent. In a 12 user/12 receiver system at an error rate of 10-3 , IMUD had a 2dB gain on MMSE V-BLAST and roughly equal complexity. In the case when there are more users than receivers, IMUD's gains are even more pronounced. In a 12 user/8 receiver flat fading system at an error rate of 10-3, IMUD had a 7dB gain on MMSE V-BLAST with roughly equal complexity. IMUD was also tested in a serially concatenated system in which all users employed convolutional codes. The results are promising. In a 6 user/4 receiver system at an error rate of 10-3, the IMUD/SISO detector performs within 2dB of a single user coded system with the same number of antennas. With a rate 2/3 code, IMUD/SISO performs within less than 2dB of a single user coded system.