Design and Optimization of Swin Transformers-Based Image Semantic Communication System Under SIMO Channels

Semantic communication is a new type of communication that improves bandwidth efficiency by transmitting semantic information of data, which can transmit a variety of modalities such as image, text, video, and audio. Semantic models based on convolutional neural networks (CNNs) make it difficult to capture global features resulting in degradation of transmission performance, and the traditional independent source channel coded image transmission method is susceptible to changes in the channel environment. To resolve these problems, this paper proposes a semantic model called SwinJSCC-SIMO, a joint source-channel coding (JSCC) technique using Swin Transformer as the model backbone for image transmission over SIMO channels. Specifically, the SIMO channel sets multiple antennas at the receiving end to receive signals, which improves the reliability of the received signals. The reconstructed image output from the decoder is input to the multiscale discriminator network (DSN) to identify the authenticity of the reconstructed image, guide the semantic model to reconstruct the image. The experimental data show that SwinJSCC-SIMO has better image reconstruction capability, stronger robustness, and signal-to-noise ratio adaptation.