Deep learning approach for automatic modulation format identification in vehicular visible light communications

Automatic Modulation Format Identification(AMFI) is an important issue in Visible Light Communication (VLC). There are several factors that affect the performance of vehicular VLC (V-VLC) systems. To accomplish the required communication quality for different traffic scenarios, It needs adaptive modulation techniques. Adaptive modulation requires modulation format identification at the receiver to avoid the overhead required to determine the proper modulation format. In this paper, we propose an AMFI scheme based on Deep Learning (DL) for Infrastructure-to-Vehicle (I2V) VLC. Our scheme utilizes the Hough transform to project the constellation diagram of the received modulated signals onto another space to extract features efficiently. Hough transform is estimated for eight different VLC modulation formats, including Q/8/16-Phase Shift Keying (PSK) and 4/16/32/64-Quadrature Amplitude Modulation (QAM) at signal to noise ratio (SNR) ranging from 5 to 25 dB. Conventional classifiers such as Alexnet, Sequeeznet and Googlenet are used to perform the task of AMFI. A study of the impact of the weather conditions and the communication distance on the accuracy of the used classifiers is introduced. The obtained results prove that the proposed scheme can identify the modulation format with high classification accuracy up to 100% at lower SNR levels and different weather conditions.