While scaling of electronics has enabled lower power consumption and increasingly dense integration of digital circuits, the associated reduction in supply voltage has become a fundamental limitation in design, as most traditional analog-to-digital converters represent values as an increment of voltage. Therefore, the resolution of these converters decreases with each generation of technology. The technology described here uses novel time encoding and decoding techniques to perform real-time asynchronous encoding and error free decoding of visual signals. This new encoding and decoding method therefore promises improved video quality compatible with lower voltage requirements.
This technology uses a neuron architecture similar to human retinal cells to decode and encode visual signals. A time encoding machine receives an input signal and generates an asynchronous set of binary transitions that are related to the amplitude of the input. A time decoding machine then uses optimization with recurrent neural networks to accurately recover the input signal with no error introduced. Unlike other decoding machines, this technology does not require a numerical algorithm to recover the visual signal, saving bandwidth during the process. Together, this creates a decoding process that is more affordable and feasible while simultaneously improving on the quality of the resulting output signals.
In addition, this technology can be built on silicon chips, is highly parallelized, and can be implemented with very large scale integration (VLSI) technology. Computer simulations of the proposed analog circuitry of this technology demonstrated high-quality reconstruction of videos with fast recovery speed.
Patent Issued (US 7,573,956)
Tech Ventures Reference: IR M11-061, M02-057, M08-071