Fairness in Multiuser Systems with Polymatroid Capacity Region

TL;DR

This paper develops algorithms to find fair rate allocations on the sum-capacity facet of polymatroid-structured multi-user systems, balancing complexity and fairness for different system scenarios.

Contribution

It introduces efficient algorithms for achieving fair rate points on the sum-capacity facet, addressing both high and low complexity cases in polymatroid-based systems.

Findings

Greedy algorithm for max-min corner point identified.

Recursive method for maximizing minimum rates among users.

Fast algorithm for computing time-sharing coefficients.

Abstract

For a wide class of multi-user systems, a subset of capacity region which includes the corner points and the sum-capacity facet has a special structure known as polymatroid. Multiaccess channels with fixed input distributions and multiple-antenna broadcast channels are examples of such systems. Any interior point of the sum-capacity facet can be achieved by time-sharing among corner points or by an alternative method known as rate-splitting. The main purpose of this paper is to find a point on the sum-capacity facet which satisfies a notion of fairness among active users. This problem is addressed in two cases: (i) where the complexity of achieving interior points is not feasible, and (ii) where the complexity of achieving interior points is feasible. For the first case, the corner point for which the minimum rate of the active users is maximized (max-min corner point) is desired for signaling. A simple greedy algorithm is introduced to find the optimum max-min corner point. For the second case, the polymatroid properties are exploited to locate a rate-vector on the sum-capacity facet which is optimally fair in the sense that the minimum rate among all users is maximized (max-min rate). In the case that the rate of some users can not increase further (attain the max-min value), the algorithm recursively maximizes the minimum rate among the rest of the users. It is shown that the problems of deriving the time-sharing coefficients or rate-spitting scheme can be solved by decomposing the problem to some lower-dimensional subproblems. In addition, a fast algorithm to compute the time-sharing coefficients to attain a general point on the sum-capacity facet is proposed.