Rank Matching for Multihop Multiflow

TL;DR

This paper investigates the degrees of freedom in a layered 2x2x2 MIMO interference channel with arbitrary antenna counts and rank constraints, revealing a fundamental rank-matching phenomenon akin to impedance matching in circuits.

Contribution

It introduces the concept of rank-matching for maximizing DoF, providing tight outer bounds and achievability results, and extends insights to more complex network configurations.

Findings

Maximum DoF achieved when source and destination ranks match.

Rank mismatch causes a DoF loss proportional to the mismatch.

Results are tight for almost all channels satisfying the rank constraints.

Abstract

We study the degrees of freedom (DoF) of the layered 2 X 2 X 2 MIMO interference channel where each node is equipped with arbitrary number of antennas, the channels between the nodes have arbitrary rank constraints, and subject to the rank-constraints the channel coefficients can take arbitrary values. The DoF outer bounds reveal a fundamental rank-matching phenomenon, reminiscent of impedance matching in circuit theory. It is well known that the maximum power transfer in a circuit is achieved not for the maximum or minimum load impedance but for the load impedance that matches the source impedance. Similarly, the maximum DoF in the rank- constrained 2 X 2 X 2 MIMO interference network is achieved not for the maximum or minimum ranks of the destination hop, but when the ranks of the destination hop match the ranks of the source hop. In fact, for mismatched settings of interest, the outer bounds identify a DoF loss penalty that is precisely equal to the rank-mismatch between the two hops. For symmetric settings, we also provide achievability results to show that along with the min-cut max-flow bounds, the rank-mismatch bounds are the best possible, i.e., they hold for all channels that satisfy the rank-constraints and are tight for almost all channels that satisfy the rank-constraints. Limited extensions - from sum-DoF to DoF region, from 2 unicasts to X message sets, from 2 hops to more than 2 hops and from 2 nodes per layer to more than 2 nodes per layer - are considered to illustrate how the insights generalize beyond the elemental 2 X 2 X 2 channel model.