Machine-Learning-Enhanced Bayesian Detection for α-Stable Noise Channels in 5G/6G DS-CDMA Networks

Impulsive, non-Gaussian interference in urban 5G/6G wireless scenarios invalidates Gaussian-noise-based assumptions of conventional detectors. This work proposes the use of the Machine Learning-Enhanced Bayesian Detector for DS-CDMA systems over α-stable noise channels. The framework merges probabilistic Bayesian inference with a recurrent neural-network estimator that continuously learns α-stable parameters α, β, γ, and δ from the received data. Closed-form derivations of the detection and false alarm probabilities are obtained using characteristic-function-based likelihood ratios, whereas the proposed approach is corroborated via MATLAB simulations. The results demonstrate that ML-BD provides a 3-dB SINR gain, ≈ 45% BER reduction, and ≈ 15% increase in the detection probability compared to classical Bayesian and energy detectors. This work demonstrates that the marriage of adaptive learning with Bayesian reasoning results in a robust, interpretable, and computationally efficient detector for interference-limited 5G/6G metropolitan networks.