Neural quantizer for fronthaul compression: Design and FPGA implementation

Abstract

Signal compression for fronthaul links has been actively studied, and the most common fixed compression method is the application of nonlinear quantizers. However, optimization such as clipping ratio (CR) is necessary for proper use of nonlinear quantizers. This paper proposes a nonlinear quantizer based on deep learning, called a neural quantizer. In this paper, we present the configuration of the neural quantizer and show that its performance is comparable to that of a CR-optimized nonlinear quantizer. We implement the proposed neural quantizer in FPGA, measure the processing delay, and show that it works within the desired processing time.