This proposal is for a knowledge exploration and building effort focused on optimizing separation of signals of interest from the accompanying background. In surveillance research, a primary goal of measuring sensor signals is to provide the basis for establishing a clear linkage between a measurement element and the source of the signal. 

The measurement of sensor streams produces a collection of the desired signals embedded in a complex unknown background. For example, in seismic data the separation of signals is from accompanying noise is crucial for seismic imaging needed for oil & gas exploration. The information bearing signal may be smeared in the ambient random noise and disregarded. Enhancing the useful signal while preserving edge properties of the seismic profiles by attenuating random noise can reduce interpretation difficulties. This issue is common across all systems depending upon detection of signals in noise.

Typically, signal processing techniques focus on the isolation of peaks from background; as such the focus is on location of the position of peaks and the strength of the peak. A common approach is to apply  a constant threshold level is used to separate signals from lower level noise and background. In a strong SNR environment this may be sufficient as is obvious from the plot. In low SNR situations and in cases where higher fidelity signal characterization is required more sophisticated advanced techniques are required. The subtraction of the background in a spectrum has been the subject of many investigations and different techniques, varying from filtering to polynomial function fits, have been developed. 

What is needed is a means to separate the background from the signal in a rigorous and self-consistent manner that enables optimal separation of signal elements and noise elements. On that basis we strive to infer a unique signature for the source from the measurement. This is especially important when our goal is to isolate features of signals that enable establishing signatures of sources.

key words

Signal plus Noise Separation; Background estimation; Information Extraction; Information Peeling; Fitting polynomial functions to sensor data; Residue processing of noise; Savitzky–Golay filters; recursive Bayesian estimation

Citizenship:  Open to U.S. citizens

Level:  Open to Postdoctoral and Senior applicants