This technology describes distilled sensing (DS), a novel, multi-step, and selective (adaptive) sampling procedure for recovering sparse signals from noisy observations. DS results in dramatic quantifiable improvements over the best non-adaptive sensing methods for the estimation and detection of sparse signals in noise. This technology is an adaptive measurement procedure using selective sampling based on a feedback input from the previous measurements, allowing successive iterations of the algorithm to refine and focus measurement resources towards relevant signal components and away from locations where no signals are present. There are several procedures for extracting image, spatial, or temporal waveform signals and these methods have proven to be successful in most cases. However current methods can have trouble in resolving sparse signals in noise.
This technology allows signals to be transferred efficiently using fewer computational resources and producing a cleaner output signal. Additionally, this technology improves upon previous methods by using a specially designed feedback process that prescribes how and where measurements should be acquired based on the outcomes of previous measurements. Doing so systematically improves the effectiveness of subsequent measurements, resulting in fewer total measurements. In this way, gains are seen in increased signal-to-noise ratio, reduced acquisition time, reduced power/energy expenditure, reduced use of acquisition resources, improved reliability of signal detection, and increased fidelity for signal estimation.
The benefits of this technology were validated in a simulation of astronomical imaging applications. The technology outperformed the best possible passive imaging techniques in terms of visual clarity and reconstruction fidelity error.
Patent Issued (US 8,521,473)
Tech Ventures Reference: IR M09-079