Channel Charting Based Pilot Allocation in MIMO Systems

We consider uplink pilot allocation based on multipoint
channel charting (CC) to mitigate pilot contamination in a
multi-cell network with spatially correlated MIMO channels. The
channel chart is created in an offline phase with full information,
i.e. user channel covariance matrices are estimated at multiple
base stations (BSs). In the online phase, we assume that only
partial information about a user’s channel covariance is known,
i.e., it is available only at the serving BS. A machine learning
framework is developed to predict the CC locations in the
online phase. Pilots are allocated to active users in the online
phase based on weighted graph colouring. CC locations are
used as proxies of user locations; similarity weights between
users are constructed from CC distances. Simulation results
show that the CC based approach with partial information in
the online phase outperforms a solution based on full angle-ofarrival
information, and performs closely to an algorithm with full
covariance information. We also consider a partial information
machine learning framework to predict the channel covariance
matrices at other BSs, which slightly outperforms CC based
approach, with the price of a