Deep learning-enhanced carrier less amplitude and phase modulation for high-speed visible light communication: mitigating LED nonlinearity and bandwidth limitations

Visible Light Communication (VLC) has emerged as a promising wireless technology, yet faces critical challenges in LED nonlinearity and bandwidth limitations that constrain performance. This work presents three key innovations: (1) A hybrid Convolutional Neural Network (CNN) – Deep Neural Network (DNN) architecture combining a 1D-CNN demodulator for joint modulation classification (7 modulation types) and distance estimation with a DNN equalizer that achieves 4–15 dB BER improvement over conventional methods; (2) A bandwidth enhancement technique enabling 6 × expansion (25 MHz signaling over 4.5 MHz LED bandwidth) while maintaining 4.67b/s/Hz spectral efficiency; and (3) An optimized LED operating point (I_DC = 0.5A, ζ = 0.8) that avoids saturation while maximizing SNR. Our system demonstrates robust 100 Mb/s transmission (64-QAM) at the 7 % FEC threshold (BER = 3.8 × 10 3 ) with Eb/N0 = 20 dB under both LOS/NLOS conditions, representing a 2 × improvement over prior CAP-VLC systems. The integrated learning-based framework addresses nonlinear distortion, bandwidth constraints, and channel variability simultaneously, offering a practical solution for high-speed optical wireless networks.