Stability and Delay of Zero-Forcing SDMA with Limited Feedback

TL;DR

This paper analyzes the stability and delay in SDMA systems with zero-forcing beamforming under limited feedback, deriving the stability region, feedback scaling laws, and delay impact due to quantization errors.

Contribution

It characterizes the stability region under limited feedback, derives feedback scaling laws, and quantifies delay effects in SDMA systems with zero-forcing beamforming.

Findings

Stability region is a convex polytope with perfect CSI.

Feedback bits scale logarithmically with inverse of stability margin.

CSI quantization errors increase queueing delay and delay tail probability.

Abstract

This paper addresses the stability and queueing delay of Space Division Multiple Access (SDMA) systems with bursty traffic, where zero-forcing beamforming enables simultaneous transmission to multiple mobiles. Computing beamforming vectors relies on quantized channel state information (CSI) feedback (limited feedback) from mobiles. Define the stability region for SDMA as the set of multiuser packet-arrival rates for which the steady-state queue lengths are finite. Given perfect CSI feedback and equal power allocation over scheduled queues, the stability region is proved to be a convex polytope having the derived vertices. For any set of arrival rates in the stability region, multiuser queues are shown to be stabilized by a joint queue-and-beamforming control policy that maximizes the departure-rate-weighted sum of queue lengths. The stability region for limited feedback is found to be the perfect-CSI region multiplied by one minus a small factor. The required number of feedback bits per mobile is proved to scale logarithmically with the inverse of the above factor as well as linearly with the number of transmit antennas minus one. The effects of limited feedback on queueing delay are also quantified. For Poisson arrival processes, CSI quantization errors are shown to multiply average queueing delay by a factor larger than one. This factor can be controlled by adjusting the number of feedback bits per mobile following the derived relationship. For general arrival processes, CSI errors are found to increase Kingman's bound on the tail probability of the instantaneous delay by one plus a small factor. The required number of feedback bits per mobile is shown to scale logarithmically with this factor.