Abstract
Many wireless transceivers, such as massive-MIMO and Rake, are designed to provide high diversity order. In order to reduce cost, low-complexity switched transceivers, e.g., hybrid antenna preprocessing with selection, or S-Rake, which employ a limited number of RF chains or correlators, are popular. This paper shows that the presence of limited number of RF chains or correlators in such transceivers introduces an inherent trade-off between system performance and the channel estimation overhead. We prove that to maximize achievable rate, it is optimal to perform channel estimation for only a subset of diversity branches with the highest second moments, if the diversity branches have the same fading parameters but different mean powers. We also prove that the achievable data-rate is a unimodal function of the size of this subset L which ensures that any locally optimum L is also globally optimum. A computationally efficient approximation for the ergodic capacity is introduced, which reduces the cost of finding the optimal L significantly. Simulation results for some practically important settings suggest that the optimal choice of L can improve the data rate by a factor of 20 - 30%.