Free Trial

Safari Books Online is a digital library providing on-demand subscription access to thousands of learning resources.

Share this Page URL
Help

Chapter 8. Multiuser MIMO Receiver Proce... > 8.8 Conclusions - Pg. 368

368 CHAPTER 8 Multiuser MIMO Receiver Processing for Time-Varying Channels 10000 1000 Megaflops 100 10 1 User User Chip Chip LMMSE Krylov LMMSE Krylov Hard SD Soft SD FIGURE 8.11 Computational complexity of the multiuser detector. We compare the complexity using (1) exact LMMSE filtering or Krylov subspace approximation with dimension S = 5, (2) parallel interference cancelation in chip space or user space, and (3) sphere decoding ("SD") with hard or soft outputs. 8.8 CONCLUSIONS Multiuser detection in time-varying channels is a demanding task from the computational complexity viewpoint. In this chapter, we traded accuracy for efficiency by introducing several complexity reduc- tion methods enabling a real-world low-complexity implementation. In particular, we used an iterative approximation of the MAP detector in combination with a reduced-rank model for the time-varying channel. This reduced-rank channel model projects the time-varying channel onto a subspace spanned by band-limited and time-concentrated prolate spheroidal sequences. We investigated two different multiuser detection methods: First, we used the Krylov subspace method to reduce the complexity of multiuser detection using LMMSE filtering. We showed that for chip-space linear multiuser detection, no complexity reduction due to the Krylov subspace method can be achieved. However, for a user-space LMMSE filter, the Krylov subspace method yields a significant complexity reduction. Second, we investigated sphere decoding and developed a sphere decoder that exploits the reduced-rank channel model for complexity reduction. Furthermore, we showed that a soft-output sphere decoder can be devised using the same approach. . . We provided simulation results and complexity comparisons for all presented low-complexity architectures. The complexity of channel estimation and multiuser detection could be reduced by more than one magnitude while retaining the good performance of exact LMMSE filters.