The cochlea is known to exhibit a nonlinear, mechanical amplification which allows the ear to detect faint sounds, improves frequency discrimination, and broadens the range of sound pressure levels that can be detected. In this work, active artificial hair cells (AHC) are proposed and developed which mimic the nonlinear cochlear amplifier. Active AHCs can be used to transduce sound pressures, fluid flow, accelerations, or another form of dynamic input. These nonlinear sensors consist of piezoelectric cantilever beams which utilize various feedback control laws inspired by the living cochlea. A phenomenological control law is first examined which exhibits similar behavior as the living cochlea. Two sets of physiological models are also examined: one set based on outer hair cell somatic motility and the other set inspired by active hair bundle motility. Compared to passive AHCs, simulation and experimental results for active AHCs show an amplified response due to small stimuli, a sharpened resonance peak, and a compressive nonlinearity between response amplitude and input level. These bio-inspired devices could lead to new sensors with lower thresholds of sound or vibration detection, improved frequency sensitivities, and the ability to detect a wider range of input levels. These bio-inspired, active sensors lay the foundation for a new generation of sensors for acoustic, fluid flow, or vibration sensing. / Ph. D.

<p>This thesis is concerned with detection of transient signal families and detectors in nonlinear static sensor systems. The detection problems are treated within the framework of likelihood ratio based binary hypothesis testing.</p><p>An analytical solution to the noncoherent detection problem is derived, which in contrast to the classical noncoherent detector, is optimal for wideband signals. An optimal detector for multiple transient signals with unknown arrival times is also derived and shown to yield higher detection performance compared to the classical approach based on the generalized likelihood ratio test.</p><p>An application that is treated in some detail is that of ultrasonic nondestructive testing, particularly pulse-echo detection of defects in elastic solids. The defect detection problem is cast as a composite hypothesis test and a methodology, based on physical models, for designing statistically optimal detectors for cracks in elastic solids is presented. Detectors for defects with low computational complexity are also formulated based on a simple phenomenological model of the defect echoes. The performance of these detectors are compared with the physical model-based optimal detector and is shown to yield moderate performance degradation.</p><p>Various aspects of optimal detection in static nonlinear sensor systems are also treated, in particular the stochastic resonance (SR) phenomenon which, in this context, implies noise enhanced detectability. Traditionally, SR has been quantified by means of the signal-to-noise ratio (SNR) and interpreted as an increase of a system's information processing capability. Instead of the SNR, rigorous information theoretic distance measures, which truly can support the claim of noise enhanced information processing capability, are proposed as quantifiers for SR. Optimal detectors are formulated for two static nonlinear sensor systems and shown to exhibit noise enhanced detectability.</p>

This thesis is concerned with detection of transient signal families and detectors in nonlinear static sensor systems. The detection problems are treated within the framework of likelihood ratio based binary hypothesis testing. An analytical solution to the noncoherent detection problem is derived, which in contrast to the classical noncoherent detector, is optimal for wideband signals. An optimal detector for multiple transient signals with unknown arrival times is also derived and shown to yield higher detection performance compared to the classical approach based on the generalized likelihood ratio test. An application that is treated in some detail is that of ultrasonic nondestructive testing, particularly pulse-echo detection of defects in elastic solids. The defect detection problem is cast as a composite hypothesis test and a methodology, based on physical models, for designing statistically optimal detectors for cracks in elastic solids is presented. Detectors for defects with low computational complexity are also formulated based on a simple phenomenological model of the defect echoes. The performance of these detectors are compared with the physical model-based optimal detector and is shown to yield moderate performance degradation. Various aspects of optimal detection in static nonlinear sensor systems are also treated, in particular the stochastic resonance (SR) phenomenon which, in this context, implies noise enhanced detectability. Traditionally, SR has been quantified by means of the signal-to-noise ratio (SNR) and interpreted as an increase of a system's information processing capability. Instead of the SNR, rigorous information theoretic distance measures, which truly can support the claim of noise enhanced information processing capability, are proposed as quantifiers for SR. Optimal detectors are formulated for two static nonlinear sensor systems and shown to exhibit noise enhanced detectability.