Performance Improvement of Random Access by Prioritizing Collisions

Abstract

The importance of unlicensed spectrum is highlighted in terms of the flexibility of network deployment for various services envisioned in 5G and beyond. Since listen before talk is mandatory for channel access in unlicensed spectrum and it causes an unavoidable waste of resources due to collisions, an efficient random backoff mechanism is required. In the existing backoff schemes that impose waiting penalties on collided packets, a degraded fairness performance is observed. In this work, we analyze how prioritizing collided packets can improve performance compared to existing schemes. To this end, we devise a random backoff scheme called the Collision Priority Backoff (CPB) under the concept of giving priority to collided packets. We apply Bayesian optimization to carefully determine channel access parameters of the CPB to maximize network throughput. Since the optimized access parameters require the number of stations in the network, we also devise an adaptive version of the CPB called the Adaptive CPB (ACPB). We deal with an environment where the number of stations changes as a switching bandit problem, and employ a variant of upper confidence bound policy in the ACPB. Various simulation results validate that the proposed backoff scheme shows high throughput and fairness performance.

Keywords

• Random access
• backoff
• Bayesian optimization
• switching bandit
• fairness
• throughput