Secondary Access to Spectrum with SINR Requirements Through Constraint Transformation

TL;DR

This paper proposes a constraint transformation and heuristic algorithm to efficiently allocate spectrum with SINR requirements in dynamic networks, achieving near-optimal solutions within 10% of genetic algorithm results.

Contribution

It introduces a novel constraint transformation for the spectrum allocation problem with SINR constraints, enabling heuristic solutions for arbitrary pool sizes and single-channel transmission.

Findings

Heuristic solutions are within 10% of genetic algorithm solutions.

The transformation converts a non-convex problem into a binary quadratic constraint problem.

The approach is effective for dynamic spectrum access scenarios.

Abstract

In this paper we investigate the problem of allocating spectrum among radio nodes under SINR requirements. This problem is of special interest in dynamic spectrum access networks where topology and spectral resources differ with time and location. The problem is to determine the number of radio nodes that can transmit simultaneously while still achieving their SINR requirements and then decide which channels these nodes should transmit on. Previous work have shown how this can be done for a large spectrum pool where nodes allocate multiple channels from that pool which renders a linear programming approach feasible when the pool is large enough. In this paper we extend their work by considering arbitrary individual pool sizes and allow nodes to only transmit on one channel. Due to the accumulative nature of interference this problem is a non-convex integer problem which is NP-hard. However, we introduce a constraint transformation that transforms the problem to a binary quadratic constraint problem. Although this problem is still NP-hard, well known heuristic algorithms for solving this problem are known in the literature. We implement a heuristic algorithm based on Lagrange relaxation which bounds the solution value of the heuristic to the optimal value of the constraint transformed problem. Simulation results show that this approach provides solutions within an average gap of 10% of solutions obtained by a genetic algorithm for the original non-convex integer problem.