Pilot Allocation for Cell-Free Massive MIMO based on Channel Charting

We consider an uplink pilot allocation scheme for cell-free massive multiple-input multiple output (mMIMO) networks based on channel charting (CC). The channel chart, intended to act as a proxy of the user equipment (UE) physical locations, is created in an offline phase, with prior knowledge of the channel covariances among all access points (APs), for a certain number of offline UEs. In an online phase, the active UEs are placed on the chart based only on partial covariance information, restricted to their serving AP sets. Pilot allocation is based on the CC locations of the active UEs. For that, distance-based clustering is considered, where each cluster exhausts all pilots. Additionally, we consider a weighted graph coloring (WGC) approach, where the weighted edge between two UEs reflects the similarity of their channels. We evaluate the channel estimation quality of the CC based framework, and compare it to that when using the UE physical locations. The results are contrasted with a benchmark method, considering knowledge of the channel covariances among all APs for the active UEs. Furthermore, we evaluate the computational complexity of the studied approaches. Numerical results show that the CC framework performs similarly to using the UE physical locations.