Learning Based CoMP Clustering for URLLC in Millimeter wave 5G networks with Blockages

URLLC will be a use case of 5G which requires high reliability, low latency and high availability to be satisfied simultaneously. 5G will be using millimeter wave (mmw) communication which suffers from frequent and dynamic blockages impacting reliability. In addition to high SNR line-of-sight (LOS) links and low SNR non-line-of-sight (NLOS) links, complete outage (blockage) links are also anticipated. Link status will be changing dynamically between these three states. Coordinated multipoint joint transmission (CoMP-JT) is an ideal candidate to ensure high reliability, where a group of base stations (BSs) transmits the same data to a user equipment (UE). Due to highly dynamic blockages and backhaul constraints, BSs selected to be part of CoMP cluster based on the reference signal received power (RSRP) alone will be outdated by the time of data transmission. In this paper, a CoMP clustering scheme is proposed in which a neural network algorithm running in each BS learns the spatiotemporal pattern of blockages and predicts the BS-UE link status based on the clock time and location of UE. The BSs with predicted blockage shall be removed and LOS links shall be given higher priority over NLOS links during CoMP clustering, thereby increasing the reliability and availability. Analytical channel model is combined with stochastic geometry based model to characterize the real world spatio-temporal blockages. A modified control flow of events for CoMP-JT in URLLC is proposed to address the issue of backhaul constraints. Simulation results show that the proposed CoMP clustering scheme outperforms the RSRP based CoMP clustering in terms of BLER and SNR.