Randomized user-centric clustering for cloud radio access network with PHY caching

We propose a randomized user-centric remote radio head (RRH) clustering (UCRC) scheme for cloud radio access network (C-RAN) with physical layer (PHY) caching. The randomized UCRC design is modeled as a multi-timescale non-convex problem to optimize the average weighted sum rate (WSR). We first decompose the problem into short-term precoding and medium-term UCRC subproblems. Then we propose a low complexity algorithm which finds the optimal solution for week interference regime and achieves good performance for general cases. Finally, we exploit PHY caching at RRH node of the C-RAN to reduce the fronthaul loading. We show that there is a phase transition of the PHY caching gain according to the skewness of the popularity distribution. Specifically, when the skewness parameter is larger than 1, a large saving in C-RAN fronthaul loading is possible even when the RRH cache capacity is small compared to the total content size.