Improving mmWave backhaul reliability: A machine-learning based approach

WiGig technologies, such as IEEE 802.11ad and later IEEE 802.11ay, provide multi-gigabit short-range communication at 60 GHz for bandwidth-intensive applications. However, this band suffers from high propagation losses that can only be compensated using highly directional antennas, making millimeter-wave (mmWave) links susceptible to blockage and errors. This high sensitivity to blockage leads to unstable and unreliable connections, since proprietary IEEE 802.11ad mechanisms, such as beamforming training, have high overhead, and can only be triggered when performance degradation is already detected, which compromises QoS and QoE even more. This article proposes a proactive machine learning framework that uses real-life data acquired in an outdoor setting to improve the reliability and resilience of a blockage-prone WiGig-based network. In particular, we propose a link quality classifier, which can differentiate between normal, long-term blockage and short-term operation with a test F1-score of 97%. Moreover, we introduce a novel deep learning forecasting model that can accurately capture the interactions between past multi-layer observations under different environments to produce accurate forecasts for 16 KPIs.