Aggressive Look-Up Table Compression by Principal Component Analysis for Digital Pre-Distortion in Optical Communication

Abstract

In this paper, we propose to reduce the size of look-up table (LUT) for digital pre-distortion in optical communication using principal component analysis (PCA). In this scheme, the full-size LUT array is rearranged into a matrix and then compressed into several matrices using the PCA algorithm. This procedure can be iterated to achieve an aggressive reduction of LUT size. Extensive experiments are conducted to validate the performance of the proposed method. For LUT with a memory length of 3 (LUT-3), the size is reduced to 15.3% with 0.25 dB and 0.1 dB signal-to-noise ratio (SNR) penalties compared to the full-size LUT-3 for uniform 64-ary quadrature amplitude modulation (64QAM) and probabilistically shaped 64QAM (PS-64QAM), respectively. For LUT with a memory length of 5 (LUT-5), the size is reduced to 2.2% and 1.5% with SNR penalties less than 0.1 dB compared to the full-size LUT-5 for uniform 64QAM and PS-64QAM, respectively. Finally, the computational complexity of obtaining a pattern error in the compressed LUT is discussed.