Toward mobility-aware semantic communication: DeepJSCC over time-varying MIMO channels

Abstract This paper investigates the behavior of deep joint source–channel coding (DeepJSCC) for semantic image transmission over time-varying multiple-input multiple-output (MIMO) wireless channels. While DeepJSCC has demonstrated robustness to channel noise and imperfect channel knowledge, the impact of mobility-induced time variations and inter-stream semantic interference in MIMO systems requires a systematic evaluation. In this study, an autoencoder-based DeepJSCC model is trained under single-input single-output (SISO) additive white Gaussian noise (AWGN) conditions, assuming that MIMO equalization is performed independently. During the evaluation phase, image data are transmitted over time-varying MIMO channels with different precoding and postcoding configurations. Reconstruction performance is assessed using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) under a wide range of Doppler conditions and antenna configurations. Experimental results reveal that conventional separated source–channel coding (JPEG–LDPC–QPSK) exhibits a pronounced cliff effect under mobility, leading to rapid performance degradation as channel variations increase. In contrast, DeepJSCC maintains graceful degradation , demonstrating substantially higher resilience in dynamic wireless environments. Furthermore, increasing the number of receive antennas notably mitigates the effects of channel aging, highlighting the importance of spatial diversity in mobility-aware semantic communication systems.