Quantitative analysis of the stacked autoencoder method in MIMO-ACO-OFDM VLC systems

In this paper, a stacked autoencoder network is utilised to realise the signal constellation and transceivers adapted to the dimmable indoor visible light communication system in order to acquire lower symbol error probability. Its decoder parts function as denoising and the equaliser for the proposed multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) visible light communication system which can compensate the non-linear transfer function and the crosstalk between multiple LED data streams. The bit error rate performance as well as the influence of LED spatial intervals on root mean square delay spread, impulse response and bit rate have been analysed considering multipath reflections of the indoor MIMO-VLC system. The numerical results show that the a stacked autoencoder technique performs better in bit error rate reduction compared with state-of-art the zero forcing and minimum mean squared error algorithm. The experiment also shows, when the semi-angle at half power of LEDs and the field of view of Photodetectors become small, better performance can be achieved at the centre of the room, which can be explained by strong beam converge and the decreased multipath interference. Moreover, enlarging the separation between LEDs leads to improved bit error rate performance and reduced channel correlation of channel matrix, which need to be optimally chosen in practice.