Failure Prediction for Proactive Beam Recovery in Millimeter-Wave Communication

This paper proposes beam failure prediction to recover from inevitable link failures in beam-based mmWave communication proactively. The proposed system consists of two components. First, a prediction engine to foresee future beam failures and their severity. For this purpose, machine learning and deep learning are proposed to perform prediction. The second component is a proactive recovery mechanism, that matches the prediction failure results with a suitable recovery action, with the goal to maintain seamless connectivity and prevent service interruptions. The performance of the proposed system is compared against conventional beam failure detection and recovery. Simulations were carried out using real beamforming data. The results indicate a substantial improvement in the network performance. The improvement is measured in terms of prediction accuracy, beam failure probability and successful beam failure probability. This paper also assesses a drawback of the proposed system, particularly the increase in handover rate, and shows that the achieved gain outweighs this weakness.