Deep Learning-Based Automatic Modulation Format Identification For I2V Visible Light Communication

Abstract

The performance of Infrastructure-to-vehicle (I2V) visible light communication (VLC) links is influenced by multiple elements, rendering fixed modulation order inadequate for data transmission. Consequently, adaptive modulation techniques become necessary for optimal communication quality in different traffic scenarios. However, to enable proper modulation type selection and avoid unnecessary overhead, modulation format identification at the receiver end becomes crucial. Therefore, this paper proposes an automatic modulation format identification (AMFI) scheme based on deep learning (DL) specifically for the I2V-VLC application scenario. Two different features are considered, the first one utilizes the constellation diagram of received modulated signals, while the second is a novel scheme, it employs the Voronoi diagram. Various VLC modulation formats are further taken into account, including Q/8/16-phase shift keying (PSK) and 4/8/16/32/64-quadrature amplitude modulation (QAM). Benchmark pre-trained classifiers such as AlexNet, SqueezeNet, and GoogleNet are employed and their accuracies are compared. Additionally, the impact of different weather conditions on the model’s accuracies is investigated. The findings illustrate the efficiency of the proposed approach in accurately identifying the VLC modulation format, even in diverse weather conditions and at low signal-to-noise ratio values.