Fig. 1. SLD device and electro-optical characteristics.(a) Light-output-power – current – voltage (L–I–V) and external quantum efficiency (EQE) characteristics of the SLD at CW injection current. (i) Scanning electron microscope image of the SLD showing the metal contact (on p-side) and the 12° tilted front facet. (b) Electroluminescence (EL) spectra of the SLD under various injection currents. (ii) SLD under operation coupled with a collimating lens. (c) Full-width at half maximum (FWHM) and peak position of the SLD as measured from (b).

Fig. 2. (a) The schematic of the MPANN equalization pre-processing for the received symbols. (b) The structure of the applied MPANN. (c) The structure of the DNN for comparison.

Fig. 3. VLC experimental setup.(a) Schematic of the VLC system and the DNN-aided CAP post-equalizer. (b) Image of the optical channel and components. (c) and (d) are detail images of the transmitter (TX) and receiver (RX) units. CAP: Carrierless amplitude and phase. AWG: Arbitrary waveform generator. DC: Direct current. TEC: Thermoelectric cooler. APD: Avalanche photodiode. OSC: Oscilloscope.

Fig. 4. (a) The relation of the injection current–Vpp–BER in the SLD-based VLC system. (b) The frequency response of the received signal (RX) and the transmitted signal (TX). (c) The constellation diagram of the recovered signal without the first-stage equalizer showing nonlinear effects at the data rate of 2.4 Gbit/s using the setup of Fig. 3.

Fig. 5. BER performance when using(a) LMS linear equalizer under different numbers of linear memory depth; (b) VOLT2 nonlinear equalizer under different numbers of nonlinear memory depth and fixed number of linear memory depth (71); (c) DPD under different numbers of linear memory depth and fixed number of nonlinear memory depth (4); (d) DPD when using different numbers of nonlinear memory depth and fixed number of linear memory depth (31). All the measurements are taken when the bandwidth of CAP-16-QAM is 600 MHz and 700 MHz. D: the optimal and utilized memory depth

Fig. 6. BER performance when varying(a) the number of nodes of the first layer of DNN, (b) the number of nodes of the first hidden layer of DNN, (c) the number of nodes of the second layer of DNN, (d) the linear memory depth of MPANN, (e) the nonlinear memory depth of MPANN, and (f) the number of nodes of the first hidden layer of MPANN when the bandwidth of CAP-16-QAM is 600 MHz and 700 MHz. N: The optimal and utilized number of nodes.

Fig. 7. (a) Q factor performance comparison versus data rate with the equalization schemes of LMS, VOLT2, DPD, DNN and MPANN. (b) Spatial complexity comparison among the different equalization schemes.

Fig. 8. The constellation diagram of CAP-16-QAM at 2.4 Gbit/s using the setup in Fig. 3 when the first-stage equalization scheme is LMS, DPD, VOLT2, DNN and MPANN, respectively.