Applications in Multi-band Isolation of Spectra with Data-Adaptive Sub-banding (MIDAS): Using Multi-criteria Decision Analysis to Optimize MIDAS-based De-noising Methods When Processing Infrasound and Other Signals of Interest

Abstract

The ever-present challenge faced by the signal processing analyst is to get more from the available data, whether it be exploiting the same data in new ways to garner new information, or simply to increase the confidence in existing qualitative metrics. Traditional techniques include filtering (to improve the signal to noise ratio of detected signals or images or to isolate and possibly remove interfering signals), feature detection/extraction (identifying key characteristics within the signal) and signal decomposition (identification of dominant signals of interest relative to noise terms). Current research by our team began with an emphasis on the filtering of signals of interest within the infrasound band but has been shown to also be effective in other applications including image processing. The Multi-stage Isolation of Spectra with Data-Adaptive Sub-banding (M.I.D.A.S.) filter begins with a wavelet pre-processing stage and follows with a spectral sub-banding stage for isolation of key signal content. The MIDAS filter is a coherent filter, so the filtering of a complex input produces a phase-preserved complex output. With many other infrasound and seismic data filtering tools such as the Pure State Filter [Olsen, 2009], a real-valued time-domain input is required and thus no phase information can be extracted from the filtered output data set. A typical signal processing scenario where phase preservation is critical is image processing. If the applied filter distorts or even destroys the phase data it will be impossible to correctly recover the image. Both qualitative and quantitative image quality metrics are used within this research and demonstrate that the MIDAS filter is effective at removing channel noise artifacts from images while preserving the phase information. Such noise artifacts might be caused by bit errors during the transmission of an image, or by quantization errors during the digitization or capture of the image.