Wireless Network Coding for MIMO Two-way Relaying using Latin Rectangles

TL;DR

This paper explores the design of physical layer network coding schemes for MIMO two-way relaying, using Latin Rectangles to adaptively mitigate channel fade effects and improve communication reliability.

Contribution

It introduces a novel approach to select network coding maps based on Latin Rectangles to eliminate most singular fade subspaces in MIMO two-way relaying.

Findings

Proper Latin Rectangles can remove most singular fade subspaces.

Number of non-removable fade subspaces is small for 2^λ-PSK signals.

Latin Rectangles for different antenna configurations can be derived from existing Latin Squares.

Abstract

The design of modulation schemes for the physical layer network-coded two-way MIMO relaying scenario is considered, with $n_R$ antennas at the relay R, $n_A$ and $n_B$ antennas respectively at the end nodes A and B. We consider the denoise-and-forward (DNF) protocol which employs two phases: Multiple access (MA) phase and Broadcast (BC) phase. It is known for the network-coded SISO two-way relaying that adaptively changing the networking coding map used at the relay, also known as the denoising map, according to the channel conditions greatly reduces the impact of multiple access interference which occurs at the relay during the MA phase and all these network coding maps should satisfy a requirement called the {\it exclusive law}. The network coding maps which satisfy exclusive law can be viewed equivalently as Latin Rectangles. In this paper, it is shown that for MIMO two-way relaying, deep fade occurs at the relay when the row space of the channel fade coefficient matrix is a subspace of a finite number of vector subspaces of $\mathbb{C}^{n_A+n_B}$ which are referred to as the singular fade subspaces. It is shown that proper choice of network coding map can remove most of the singular fade subspaces, referred to as the removable singular fade subspaces. For $2^{\lambda}$-PSK signal set, it is shown that the number of non-removable singular fade subspaces is a small fraction of the total number of singular fade subspaces. The Latin Rectangles for the case when the end nodes use different number of antennas are shown to be obtainable from the Latin Squares for the case when they use the same number of antennas. Also, the network coding maps which remove all the removable singular singular fade subspaces are shown to be obtainable from a small set of Latin Squares.