Complementary Correlation: New Features for Audio Analysis and Enhancement
Abstract
What information is available in the spectral asymmetry of noise-like signals Unknown signal characterization is often done at low signal-to-noise ratios where to make the problem more challenging, the unknown signal is noise-like but does not have the statistics of random noise but the underlying structure is not characterized. One problem example would be detection of noise-like signals with long-tailed autocorrelations, such as a chaotic signal. With the electromagnetic signal being frequency limited the signal can only be accurately described in time-frequency and ambiguity space. (The ambiguity space is the 2-dimensional Fourier transform of the time-frequency description.) More generally the problem is how to characterize timefrequency signals, signals that are not quasi-periodic or at least having characteristics that are not known ahead of time. Long tailed autocorrelations are essentially impossible to estimate since the signal levels are lower than the quantization limit of A/D converters. Thus analog processing of quantities related to the generation of chaos could offer a sensitive (i.e. effectively high dynamic range) and low power alternative. We can start with a standard analog square law detector, which squares the incoming signal and low passes it to find its energetic envelope. This low frequency, essentially DC term, the spectral auto-correlation, formed by simply squaring the signal and keeping its low pass information retains none of the spectral asymmetry of the signal source. Yet if the same squared output is high passed, with the same corner as the low pass filter, the a) spectral asymmetry of the signal is enhanced and b) the high passed squared output is located exactly at twice the frequency of the input signal. The, for example, a ratio test (division) of the high pass squared output to the usual energetic low pass squared output could satisfy the mathematical properties of a maximum likelihood detector of spectral asymmetry. The degree of this asymmetry could be used to add usually missing information to diagnose the source, natural or manmade, of the signal.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 25, 2021
- Source ID
- W911NF1210277
Entities
People
- Les Atlas
Organizations
- Army Contracting Command
- United States Army
- University of Washington