Coverage Hole Detection for mmWave Networks: An Unsupervised Learning Approach

The utilization of millimeter-wave (mmWave) bands in 5G networks poses new challenges to network planning. Vulnerability blockages at mmWave can cause coverage holes (CHs) the radio environment, leading link failure when a user enters these CHs. Detection CHs carries critical importance so that necessary remedies be introduced improve coverage. In this letter, we propose novel approach identify an unsupervised fashion using state-of-the-art manifold learning technique: uniform approximation and projection. key idea is preserve local-connectedness structure inherent collected unlabelled channel samples, such from service area are detectable. Our results on DeepMIMO dataset scenario demonstrate proposed method learn within data samples provide visual low-dimensional embedding while preserving CH boundaries. Once boundary determined embedding, channel-based localization techniques applied obtain geographical boundaries