Large-Scale Channel Inversion and Transmit Vector Perturbation
David Karpuk, Peter Moss · arXiv (Cornell University) · 2016
We study the behavior of channel inversion and vector perturbation schemes for large systems, wherein the transmitter has $M$ transmit antennas, and is transmitting to $K$ single-antenna non-cooperating receivers. We provide results which predict the signal-to-interference-plus-noise-ratio (SINR) for MMSE preinversion for large systems (as $K\rightarrow\infty$). We construct a vector perturbation strategy which maximizes a very sharp estimate of the SINR of the system, which we deem max-SINR vector perturbation, and similarly provide results which predict the corresponding expected SINR. We demonstrate that max-SINR vector perturbation outperforms other methods which minimize power renormalization constants. The complexity of solving integer least squares problems prohibits one from realizing the potential capacity gains of vector perturbation methods for very large systems. To that end, we use a sub-ML solver based on a sorted QR matrix decomposition to perform max-SINR vector perturbation for very large systems. The resulting SINR is shown to be comparable to the ML method for small $K$, while retaining large benefits over MMSE inversion for large $K$. We conclude by experimentally studying the resulting capacity of our scheme.