AI/ML for CSI Compression

In wireless communications, particularly for multi-antenna systems, the information about the wireless channel is leveraged to minimize interference and enhance received signal quality (Signal to Interference Plus Noise Ratio). The wireless channel (known as Channel State Information (CSI)) is often estimated at the mobile device (also known as User Equipment (UE)) and fed back to the Base Station (also known as Network (NW)).

In multi-antenna systems, the feedback CSI usually large and thus needs to be compressed. Traditionally, CSI feedback compression has been done using codebook-based methods. Recently, AI/ML-based CSI Compression has gained significant traction. AI/ML-based CSI Compression involves a two-sided AI/ML model consisting of an encoder model at the UE and a decoder model at the NW. The encoder compresses the CSI and the decoder reconstructs the CSI. In its simplest form, both the encoder and decoder are jointly trained and then split up during deployment.