Bit Allocation Law for Multi-Antenna Channel Feedback Quantization: Single-User Case

TL;DR

This paper develops an asymptotically optimal channel quantization scheme for multi-antenna single-user systems, minimizing power with limited feedback by optimally allocating bits between channel magnitude and direction.

Contribution

It introduces a product codebook structure with optimal bit allocation between magnitude and direction for large feedback bit regimes in multi-antenna systems.

Findings

Uniform quantization in dB is asymptotically optimal.

Optimal bit ratio for real channels is (M-1)/2 between direction and magnitude.

System performance approaches perfect CSI as feedback bits increase.

Abstract

This paper studies the design and optimization of a limited feedback single-user system with multiple-antenna transmitter and single-antenna receiver. The design problem is cast in form of the minimizing the average transmission power at the base station subject to the user's outage probability constraint. The optimization is over the user's channel quantization codebook and the transmission power control function at the base station. Our approach is based on fixing the outage scenarios in advance and transforming the design problem into a robust system design problem. We start by showing that uniformly quantizing the channel magnitude in dB scale is asymptotically optimal, regardless of the magnitude distribution function. We derive the optimal uniform (in dB) channel magnitude codebook and combine it with a spatially uniform channel direction codebook to arrive at a product channel quantization codebook. We then optimize such a product structure in the asymptotic regime of $B\rightarrow \infty$, where $B$ is the total number of quantization feedback bits. The paper shows that for channels in the real space, the asymptotically optimal number of direction quantization bits should be ${(M{-}1)}/{2}$ times the number of magnitude quantization bits, where $M$ is the number of base station antennas. We also show that the performance of the designed system approaches the performance of the perfect channel state information system as $2^{-\frac{2B}{M+1}}$. For complex channels, the number of magnitude and direction quantization bits are related by a factor of $(M{-}1)$ and the system performance scales as $2^{-\frac{B}{M}}$ as $B\rightarrow\infty$.