Meta-Heuristic Optimization of Ensemble Learning for Cooperative Spectrum Sensing in 5G Cognitive Radio Networks

Abstract

The scarcity of spectral resources has positioned Cognitive Radio (CR) as a critical technology for next-generation wireless communications, relying heavily on accurate spectrum sensing to identify unutilized frequency bands. While machine learning techniques like Random Forest (RF) combined with Empirical Mode Decomposition (EMD) have proven effective, their performance at ultra-low Signal-to-Noise Ratios (SNR) is often constrained by sub-optimal hyperparameter configurations. This paper proposes a novel framework that integrates Particle Swarm Optimization (PSO) with an RF classifier to dynamically optimize hyperparameter selection—specifically the number of learning cycles and minimum leaf size. By utilizing a custom fitness function based on Out-of-Bag (OOB) error validation, the PSO algorithm minimizes the probability of false alarms while maximizing detection probability. The primary user is modelled using a 5G-compliant Orthogonal Frequency-Division Multiplexing (OFDM) waveform. Simulation results demonstrate a significant enhancement in cooperative sensing performance. The proposed PSO-RF-EMD architecture achieves a perfect 100% probability of detection (Pd) at an SNR of -11 dB, effectively pushing the SNR wall further than conventional unoptimized ensemble methods, thus ensuring highly robust spectrum sensing in severely noisy environments.