Neural circuit architectures for real-time signal processing in video rate communication systems

The authors describe the algorithms and hardware used to vector quantize predicted pixel intensity differences for real-time video compression. In this approach, both the algorithms and hardware are derived from aspects of neural network research, which can be thought of as providing new types of heuristics. The hardware is designed for rapid vector quantization performance, which entails the development of application specific associative memory circuits. The real-time associative memory is a key component of the signal processing hardware. Analog hardware is used to perform transform calculations on the source signal intensities, based on a Herault-Jutten network.<
>