Machine Learning in Visible Light Communication System: A Survey

Table 1

ML algorithms in nonlinearity mitigation.

Method	Ref.	Outcomes
GMM	[20, 26]	(1) Easy to implement; (2) Require small set of variable; (3) Higher dynamic range is achieved
-Means algorithm	[26, 53]	(1) Simple structure; (2) Extra learning model not needed; (3) Data rate up to 400 Mbit/s is achieved; (4) High Q-factor; (5) Does not work well with cluster of different sizes and densities
LM algorithm	[33]	(1) Can manage multiparameter frameworks; (2) Performance improvement is achieved in the context of training complexity; (3) It takes a long time in some scenarios
Gaussian kernel-aided DNN	[34]	(1) Can significantly decrease training iterations up to 47.6%; (2) Data rate up to 1.5 Gbps
Probabilistic Bayesian-based learning	[35]	(1) Compensate for the negative effects of VLC source nonlinearity; (2) Improved BER