Joint Power and Antenna Selection Optimization in Large Cloud Radio Access Networks

Large multiple-input multiple-output (MIMO) networks promise high energy efficiency, i.e., much less power is required to achieve the same capacity compared to the conventional MIMO networks if perfect channel state information (CSI) is available at the transmitter. However, in such networks, huge overhead is required to obtain full CSI especially for Frequency-Division Duplex (FDD) systems. To reduce overhead, we propose a downlink antenna selection scheme, which selects $S$ antennas from $M > S$ transmit antennas based on the large scale fading to serve $K\leq S$ users in large distributed MIMO networks employing regularized zero-forcing (RZF) precoding. In particular, we study the joint optimization of antenna selection, regularization factor, and power allocation to maximize the average weighted sum-rate. This is a mixed combinatorial and non-convex problem whose objective and constraints have no closed-form expressions. We apply random matrix theory to derive asymptotically accurate expressions for the objective and constraints. As such, the joint optimization problem is decomposed into subproblems, each of which is solved by an efficient algorithm. In addition, we derive structural solutions for some special cases and show that the capacity of very large distributed MIMO networks scales as $O(K\log M)$ when $M\rightarrow\infty$ with $K, S$ fixed. Simulations show that the proposed scheme achieves significant performance gain over various baselines.