Non-optimal is Good! Resource Allocation in Presence of Dynamic Obstacles in D2D Networks

To cope with high bandwidth demands of modern applications, device-to-device (D2D) communications using millimeter-wave (mmWave) signals are being harnessed. The major challenge of mmWave signals is that they require a strict, obstacle-free line-of-sight communication. Static obstacles are easier to avoid; dynamic obstacles pose the main hurdle, their movement not being known. In this work, we propose a way to learn link blockages due to dynamic obstacles, using the link activation history. For this, one might have to explore non-optimal link activations. This ensures that all links are tried a sufficient number of times, ensuring adequate knowledge about link failures, thus creating an exploration-exploitation dilemma. To this end, we propose a systematic way of exploring such non-optimal channel allocations, so that the number of link failures is minimized. Given the hardness of this problem, we devise a greedy solution, and show its effectiveness over existing strategies through simulations.