Impulse Response Identification Applied to Estimation of Event-Related Potentials and Non-Intrusive Echo Path Analysis

Abstract: This thesis consists of five parts in which problems in two fields, medical and telecommunication, have been studied. In the first part, a prediction algorithm, the Prony method, for estimation of frequencies and damping factors of sinusiods is developed for a coloured noise disturbance. The method is applied to estimating a single Event-Related Potential, sERP, and the estimation principle is evaluated on both simulated ERPs and measured data. In part two, a new system for automatic tracking of the depth of anaesthesia is proposed. The amplitude and latency of the Nb-trough are used as tracking parameters. The system detects changes in the state of consciousness with a maximum delay of 52 s. Part three presents algorithms in a system which performs in-service non-intrusive measurements of the telephone channel. The new principle used in this work is to divide the problem into delay estimation and echo path identification. A block-adaptive frequency-domain technique is used, leading to low computational complexity and high quality of transfer function estimate. The double-talk detector (DTD) introduced in part three is thoroughly described and analysed in part four. An analysis of the detector is carried out where its approximative receiver operation characteristic is derived. The simulations show that the proposed DTD handles wider ranges of signal levels and hybrid attenuations than a standard detector. Part five develops and analyses robust transfer function estimation based on the non-parametric spectrum estimation technique described in part three. The transfer function estimation algorithm is reformulated to be block recursive in data rather than expressed in spectral estimates. This development paves the way for the robust recursive algorithm where techninques from robust statistics theory are adapted to the complex parameter case.