SoftCast achieves this by encoding videos using fixed-precision real numbers and maintaining that representation throughout the video codec to the physical layer. This departs from traditional encoding methods in which transmitted video undergo repeated modulation schemes to convert pixel values to code-words lacking numerical properties of the original values. SoftCast pushes the real representation all the way to the channel by making the magnitude of the transmitted signal proportional to the pixel value such that channel noise directly translates to a perturbation in pixel values. This has a small impact on video quality and allows for smooth degradation as channel noise increases.
While traditional compression and error-correcting cannot be applied to real values, SoftCast achieves the analogous result of redistributing the bits in a data stream, removing redundant bits and allowing for correction, by directly redistributing power among the real values. In this way, a source can compute the optimal power allocation to produce a compressed and accurate representation of the video signal, subject to a hardware power budget. The source can perform this optimization purely based on the video data, without any need for receiver feedback. This allows SoftCast to achieve a smooth degradation property of real numbers, while maintaining compactness and resilience.