Timing synchronization of LACO-OFDM under non-linear distortions using ELM

Layered asymmetrically-clipped optical orthogonal frequency division multiplexing (LACO-OFDM) is a modulation technique for visible light communication (VLC) that provides power and spectral efficiencies. The LACO-OFDM technique utilizes all the available subcarriers and will have a relatively higher peak-to-average power ratio (PAPR). Hence, the modulated samples are prone to distortions caused by the non-linear devices and diffused channels, resulting in reduced accuracy of delay offset estimation for timing synchronization. In this paper, we propose a method to improve the timing synchronization accuracy of LACO-OFDM modulated systems using extreme learning machines (ELM). The ELM is trained offline to identify the delay offset using the timing metric (TM) as a feature identification problem. Among all types of neural networks that are used as classifiers and non-linear estimators, the ELMs are the least complex and require significantly less computational capacity. The simulations are performed to evaluate the accuracy of delay offset estimation in the presence of unknown non-linear effects of high-powered light-emitting diodes (LEDs) and diffused channel characteristics. The results show that the probability of error (PoE) of delay offset estimation and bit error rate (BER) is significantly reduced for scenarios like different frame lengths, channel coefficients, and unknown LED parameters in contrast to conventional methods. The proposed approach attains a PoE of 10−1 at signal-to-noise ratios (SNR)s of 4 dB and 6 dB with various TM, surpassing conventional methods which only achieve this beyond 15 dB. Moreover, the PoE remains below 10−2 for SNRs exceeding 7 dB and 10 dB in the proposed method, while conventional methods require SNRs above 25 dB to achieve comparable results.