Wireless Bidirectional Relaying, Latin Squares and Graph Vertex Coloring

TL;DR

This paper links the problem of designing network coding maps for two-way relay channels to graph coloring of singularity removal graphs, providing a novel graph-theoretic approach to optimize Latin Square-based solutions.

Contribution

It introduces the concept of singularity removal graphs and shows how Latin Squares satisfying certain constraints can be obtained via vertex coloring, advancing the design of network coding maps.

Findings

Latin Square solutions correspond to proper vertex colorings of singularity removal graphs.

Minimum symbols in Latin Squares equal the chromatic number of the associated graph.

For M-QAM, some graphs require more than M colors; for certain PSK sets, M colors suffice.

Abstract

The problem of obtaining network coding maps for the physical layer network coded two-way relay channel is considered, using the denoise-and-forward forward protocol. It is known that network coding maps used at the relay node which ensure unique decodability at the end nodes form a Latin Square. Also, it is known that minimum distance of the effective constellation at the relay node becomes zero, when the ratio of the fade coefficients from the end node to the relay node, belongs to a finite set of complex numbers determined by the signal set used, called the singular fade states. Furthermore, it has been shown recently that the problem of obtaining network coding maps which remove the harmful effects of singular fade states, reduces to the one of obtaining Latin Squares, which satisfy certain constraints called \textit{singularity removal constraints}. In this paper, it is shown that the singularity removal constraints along with the row and column exclusion conditions of a Latin Square, can be compactly represented by a graph called the \textit{singularity removal graph} determined by the singular fade state and the signal set used. It is shown that a Latin Square which removes a singular fade state can be obtained from a proper vertex coloring of the corresponding singularity removal graph. The minimum number of symbols used to fill in a Latin Square which removes a singular fade state is equal to the chromatic number of the singularity removal graph. It is shown that for any square $M$-QAM signal set, there exists singularity removal graphs whose chromatic numbers exceed $M$ and hence require more than $M$ colors for vertex coloring. Also, it is shown that for any $2^{\lambda}$-PSK signal set, $\lambda \geq 3,$ all the singularity removal graphs can be colored using $2^{\lambda}$ colors.