Time-frequency BSS of biosignals

Abstract

Time–frequency (TF) representations are very important tools to understand and explain circumstances, where the frequency content of non-stationary signals varies in time. A variety of biosignals such as speech, electrocardiogram (ECG), electroencephalogram (EEG), and electromyogram (EMG) show some form of non-stationarity. Considering Priestley's evolutionary (time-dependent) spectral theory for analysis of non-stationary signals, the authors defined a TF representation called evolutionary Slepian transform (EST). The evolutionary spectral theory generalises the definition of spectra while avoiding some of the shortcomings of bilinear TF methods. The performance of the EST in the representation of biosignals for the blind source separation (BSS) problem to extract information from a mixture of sources is studied. For example, in the case of EEG recordings, as electrodes are placed along the scalp, what is actually observed from EEG data at each electrode is a mixture of all the active sources. Separation of these sources from a mixture of observations is crucial for the analysis of recordings. In this study, they show that the EST can be used efficiently in the TF-based BSS problem of biosignals.

Keywords

• medical signal processing
• spectral analysis
• electroencephalography
• blind source separation
• Gaussian processes
• iterative methods
• time-frequency analysis
• signal reconstruction
• biosignals
• blind source separation problem
• active sources
• EST
• TF-based BSS problem
• TF BSS
• time–frequency representations
• important tools
• frequency content
• nonstationary signals varies
• nonstationarity
• Priestley's evolutionary
• time-dependent
• TF representation
• evolutionary spectral theory
• bilinear TF methods