SPLIT SPECTRUM ALGORITHMS RELY ON INSTANTANEOUS PHASE INFORMATION - A GEOMETRICAL APPROACH

Split spectrum algorithms for suppression of interference noise in ultrasonic nondestructive evaluation are known to work well when properly tuned. However, the algorithms are sensitive to certain parameter values and it is not clear how the algorithms use the phase and amplitude information available. This is partly because most split spectrum algorithms have been suggested heuristically without any detailed signal and noise model. Here, a first step towards a model based approach to split spectrum is initiated. Based on a fairly detailed signal model, geometrical interpretations of the split spectrum concept are presented in an attempt to trace the phase and amplitude information utilized by the algorithms. In the light of gained theoretical knowledge, the conventional algorithms, polarity thresholding, minimization and geometric mean are evaluated. Finally, the geometrical approach is related to statistical pattern recognition and neural network approaches and the signal model is verified experimentally on real ultrasonic signals.