Channel learning and enhanced adaptive reconstruction for robust image transmission over complex time varying wireless channels

Wireless transmission of sensor-generated images over complex time-varying channels suffering from multipath fading, Doppler shifts, and phase noise remains challenging. Traditional bit-oriented systems degrade sharply under such conditions and fail to preserve semantic integrity. We propose CLEAR, a channel-aware semantic communication framework that integrates deep joint source–channel coding (DeepJSCC) with an adaptive diffusion denoising model (ADDM). Unlike prior DeepJSCC variants that adapt only to SNR, CLEAR conditions the iterative denoising process on Channel State Information (CSI)—including impulse response, Doppler shift, and phase noise—enabling the receiver to dynamically adjust denoising behavior to the specific distortion profile. Experiments on CIFAR-10, CIFAR-100, DIV2K, and Kodak24 show that CLEAR outperforms DeepJSCC, ADJSCC, DeepJSCC-A, DeepJSCC-V, and SCAN across AWGN, Rayleigh fading, and time-varying channels. Under complex channel conditions (Doppler shift $$D_s = 0.5$$ , phase noise $$P_n = 0.5$$ ), CLEAR achieves up to 2.3 dB PSNR gain and 0.041 SSIM improvement over DeepJSCC-V, while the ADDM module contributes an additional 0.81−1.56 dB over a CNN-based denoiser at comparable parameter count.