Memory-controlled deep LSTM neural network post-equalizer used in high-speed PAM VLC system

Area of Science:

• Optical Communications
• Artificial Intelligence
• Signal Processing

Background:

• High-speed visible light communication (VLC) systems face performance limitations due to linear and nonlinear impairments.
• Neural network-based equalizers offer a promising solution for enhancing optical communication system performance.

Purpose of the Study:

• To propose and demonstrate a memory-controlled deep long short-term memory (LSTM) neural network post-equalizer.
• To mitigate both linear and nonlinear impairments in pulse amplitude modulation (PAM)-based VLC systems.

Main Methods:

• Implementation of a deep LSTM neural network as a post-equalizer.
• Experimental demonstration using 1.15-Gbps PAM4 and 0.9-Gbps PAM8 VLC systems with a single red LED.
• Comparison with traditional finite impulse response (FIR) and nonlinear hybrid Volterra equalizers.

Main Results:

• Successful demonstration of PAM4 and PAM8 VLC systems with bit error ratio (BER) below the hard decision forward error correction (HD-FEC) limit.
• Achieved a 1.2 dB improvement in Q factor and a one-third increase in transmission distance compared to FIR equalizers.
• Demonstrated significant complexity and performance advantages over traditional Volterra equalizers.

Conclusions:

• The proposed deep LSTM equalizer effectively mitigates impairments in high-speed PAM-based VLC systems.
• This work represents the first demonstration of deep LSTM application in VLC systems.
• The LSTM-based equalizer offers superior performance and efficiency compared to conventional methods.