JSCC

Technical Explanation

Classical communication follows Shannon's separation theorem: source coding compresses data (e.g., JPEG, H.265) into bits, and channel coding (e.g., LDPC, Polar) adds redundancy to protect those bits. In real-world noisy, fading wireless channels, this separation causes catastrophic cliff-edge threshold drops. Deep JSCC trains encoder and decoder neural networks jointly across a simulated noisy wireless channel, achieving graceful signal degradation, eliminating threshold cliffs, and vastly outperforming classical codecs in low SNR.

Key Functions

• End-to-end autoencoder training directly outputting analog I/Q transmission symbols
• Elimination of catastrophic cliff-edge signal collapse under severe channel degradation
• Direct adaptation to instantaneous channel state information (CSI) and fading
• Substantial latency reduction bypassing multi-stage digital video compression pipelines
• Tailored feature preservation prioritizing salient regions of images and lidar scans