IMPROVING DETECTION RATE WITH DCGAN IN WIRELESS END-TO-END IMAGE TRANSMISSION SYSTEM USING SEMANTIC COMMUNICATIONS

Semantic communication, which tries to go beyond Shannon's theory of communications by communicating the semantic meaning of the data instead of its bit-by-bit reconstruction at the receiver's end, is what is expected to shape mobile communication in the future. The semantic communication paradigm seeks to fill the gaps created by issues with finite bandwidth in the transmission of information for contemporary, high-volume multimedia applications. Semantic communication-based end-to-end communication systems have been made possible by the integration of AI technology with 6G communications networks. In this study, we developed an end-to-end picture transmission system based on semantic communication, and we investigate potential design factors for developing systems for semantic communication in conjunction with physical channel properties. In order to transmit the realistic image based on the semantically segmented image input at the receiver, a pre-trained DCGAN network is used. The COCO-Stuff dataset is used to train both the DCGAN network at the receiver (the decoder) and the semantic segmentation job at the transmitter (the encoder). According to the study, broadcasting the semantic segmentation map across the physical channel rather than the actual image results in significant resource gains in the form of bandwidth savings when compared to normal communication methods. The study also investigates the impact of quantization noise and physical channel aberrations on the transfer of multimedia content using semantic communication.

 Keywords: Semantic Communications, Deep Convolutional Generative Adversarial Network (DCGAN), Encoder and Decoder.