Learning Personalized Codebook for TDD Non-Antenna Switching System

Abstract

In this paper we investigate the codebook design for a time division duplex (TDD) non-antenna switching system, where the user device is equipped with two receive radio frequency (RF) chains but only one transmit RF chain. To acquire the channel state information at transmitter (CSIT), we resort to a mixture of channel sounding and limited feedback, where the former is employed to obtain the CSIT of the antenna connected to the transmit RF chain and the latter is employed for the other antenna. We propose a deep learning method to design the codebook for limited feedback. The learned codebook is distinguished from the traditional ones in two ways. First, the learned codebook is so-called personalized, which is not fixed but adapt to the partially known CSIT. Second, the codebook exhibits different beam patterns from the traditional codebook that is designed for quantization error minimization. Simulation results demonstrate that the learned codebook can achieve higher data rate with lower complexity than traditional codebook.