Blind MIMO Semantic Communication

In this paper, we propose a novel blind multi-input multi-output (MIMO) semantic communication (SC) framework named Blind-MIMOSC that consists of a deep joint source-channel coding (DJSCC) transmitter and a diffusion-based blind receiver. The DJSCC transmitter encodes source data into the transmitted signal, while the blind receiver employs a parallel variational diffusion (PVD) model to simultaneously recover the channel and the source data from the received signal without using any pilots. This work presents the first fully pilot-free solution for joint channel estimation and source recovery in block-fading MIMO systems. Extensive experiments show that Blind-MIMOSC with PVD significantly improves channel and source recovery accuracy compared to state-of-the-art approaches, with drastically reduced channel bandwidth ratio (defined as the ratio of the channel input symbol count to source data dimensions).