A deep learning framework for blockage mitigation and duration prediction in mmWave wireless networks

Abstract

Millimeter-Wave (mmWave) communication can be highly affected by blockages, which can drastically decrease the signal strength at the receiver side. To overcome the impact of blockages, predicting the optimal mitigation technique and accurately estimating the duration of the blockage events are crucial for maintaining reliable and high-performance mmWave communication systems. Prior works on mitigating blockages have proposed a variety of model and protocol-based blockage mitigation solutions that concentrate on a singular technique at a time, like switching the current beam to an alternative beam at the current base station or client. In this paper, we tackle the overarching question: what blockage mitigation technique should be employed? and what is the optimal sub-selection within that technique? We also address the blockage duration estimation problem. We solve these problems by developing a Gated Recurrent Unit (GRU) model, trained on data from periodic message exchanges in mmWave systems. We tested our neural network models by utilizing a mmWave simulator that is commercially available and widely used in wireless communication to compile a large amount of dataset for this purpose. Our findings reveal that our proposed method introduces no extra communication overhead, while achieving remarkable accuracy, exceeding 91%, in predicting the optimal blockage mitigation technique. Moreover, the blockage duration estimation model achieves a very high accuracy with a residual mean error of less than 0.04 s. Finally, we demonstrate that our proposed blockage mitigation method substantially boosts the volume of data transferred in comparison to various other blockage mitigation strategies.