Impulsive-Noise-Aware EP Detector for Massive MIMO under a Bernoulli-Gaussian Model

Expectation propagation (EP) provides an attractive complexity-performance tradeoff for uplink massive multiple-input multiple-output (MIMO) detection. However, its standard formulation relies on the additive white Gaussian noise assumption and is therefore vulnerable to impulsive, non-Gaussian disturbances. This work proposes an impulsive-noise-aware EP (INA-EP) detector for massive MIMO systems under Bernoulli-Gaussian (BG) impulsive noise. At each iteration, INA-EP performs scalar BG inference on the receive-antenna residuals to obtain posterior statistics of the impulsive component. These statistics are then incorporated into the EP update through an impulse-mitigated effective observation and the corresponding effective noise variance, thereby reducing the impact of sparse large-amplitude outliers. Simulation results demonstrate that INA-EP consistently outperforms conventional EP in both impulsive-noise estimation mean squared error (MSE) and uncoded bit error rate (BER), with larger gains at higher impulsive activity, while remaining robust under imperfect channel state information.