Department Publications

Objective The Temporal Response Function (TRF) is a linear model of neural activity time-locked to continuous stimuli, including continuous speech. TRFs based on speech envelopes typically have distinct components that have provided remarkable insights into the cortical processing of speech. However, current methods may lead to less than reliable estimates of single-subject TRF components. Here, we compare two established methods, in TRF component estimation, and also propose novel algorithms that utilize prior knowledge of these components, bypassing the full TRF estimation. Methods We compared two established algorithms, ridge and boosting, and two novel algorithms based on Subspace Pursuit (SP) and Expectation Maximization (EM), which directly estimate TRF components given plausible assumptions regarding component characteristics. Single-channel, multi-channel, and source-localized …

This paper proposes a single-channel speech enhancement method to reduce the noise and enhance speech at low signal-to-noise ratio (SNR) levels and non-stationary noise conditions. Specifically, we focus on modeling the noise using a Gaussian mixture model (GMM) based on a multi-stage process with a parametric Wiener filter. The proposed noise model estimates a more accurate noise power spectral density (PSD), and allows for better generalization under different noise conditions such as harmonic and babble noise environments compared to traditional Wiener filtering methods. Simulations show that the proposed approach can achieve better performance in terms of speech quality (PESQ) and intelligibility (STOI) at low SNR levels.

We address the detection of material defects, which are inside a layered material structure using compressive sensing-based multiple-input and multiple-output (MIMO) wireless radar. Here, strong clutter due to the reflection of the layered structure’s surface often makes the detection of the defects challenging. Thus, sophisticated signal separation methods are required for improved defect detection. In many scenarios, the number of defects that we are interested in is limited, and the signaling response of the layered structure can be modeled as a low-rank structure. Therefore, we propose joint rank and sparsity minimization for defect detection. In particular, we propose a non-convex approach based on the iteratively reweighted nuclear and ℓ1-norm (a double-reweighted approach) to obtain a higher accuracy compared to the conventional nuclear norm and ℓ1-norm minimization. To this end, an iterative algorithm is designed to estimate the low-rank and sparse contributions. Further, we propose deep learning-based parameter tuning of the algorithm (i.e., algorithm unfolding) to improve the accuracy and the speed of convergence of the algorithm. Our numerical results show that the proposed approach outperforms the conventional approaches in terms of mean squared errors of the recovered low-rank and sparse components and the speed of convergence.

We address the compensation of distortions in terahertz time-domain spectroscopy (THz-TDS) systems. In this context, the main sources of possible distortions are misalignment of the hardware and the sample during the scan, vibration of the measurement system and time delay variations due to the unevenness of the material surface. This heavily affects the achievable accuracy of material characterization by the THz-TDS system. Therefore, it is important to develop approaches to mitigate these effects. In this work, we profit from the low-rankness of the distortion profile. To this end, we formulate an iterative algorithm based on rank minimization to estimate the material parameters in the presence of the aforementioned impairments. In contrast to the standard convex relaxation of the rank, we propose an iteratively reweighted nuclear norm for improved estimation performance. Further, we compare our approach with …

The Kalman filter is often used with plants without zeros. However, when the plant has zeros, the Kalman filter tends to neglect the effects of zeros, leading to errors in output estimation. Without the effects of zeros, the Kalman filter estimation would lag behind the true output, or it would lead to true output if the zeros are on the RHP. The parallel Kalman filter bank, instead of a single filter is proposed to remedy this problem. Each Kalman filter in the bank operates on a plant submodel that is obtained by decomposing the plant model. The output estimations of all of the Kalman filters in the bank are added together to synthesize the plant output estimation. The method was simulated in a PID-controlled system where better results have been obtained.

The quadrotor is an under-actuated nonlinear system that poses challenges in modeling and controller design. This work uses the translational (Newtonian) and rotational (Eulerian) dynamics of the quadrotor to build a decoupled control system. A linear quadratic Gaussian (LQG) regulator is used as the controller, and two extended Kalman filters are used for state estimation in the two dynamic models. The entire design process from modeling to flight simulation is presented. Model linearization while preserving controllability, state-space modeling under gravity disturbance, feedforward compensation, LQR gain adjustment, actuator dynamics, sensor system for observability, adaptation of extended Kalman filter, and its limitations are presented.

Wireless signals-based non-destructive material characterization and object identification is a vital research area that has many potential day-to-day and industrial applications. Here, one of the main challenges is suppression of the unwanted reflections coming from the surroundings that are not in our interest, ie, clutter. The clutter echoes are much stronger than the reflection that comes from the desired object. Also, in many scenarios, the clutter source and the desired objects are in close proximity. Therefore, clutter may mask the reflections coming from the desired objects. This dissertation intensively discusses clutter suppression for the identification of objects/defects. In this regard, we utilize advances in supervised learning and nonconvex optimization for clutter suppression. In this context, we profit from the low-rankness and the sparseness of the responses of the clutter and the objects. Instead of the convex relaxation of rank and sparsity commonly applied in state-ofthe-art, we propose a non-convex iteratively reweighted nuclear and ℓ1− norm-based iterative algorithm to obtain higher accuracy. Moreover, in contrast to the standard parameter tuning of the iterative algorithm, we propose supervised deep learning to tune the parameters, which is referred to as algorithm unrolling. Our numerical results show that deep learning-based parameter tuning significantly improves the accuracies of the recovered low-rank and sparse components and the speed of the convergence of the iterative algorithm. Additionally, we investigated the robustness of our approach by introducing artificial errors into the measurement matrices. Our findings show that …