Faculty Bibliography

Synchrosqueezing transform (SST) can effectively improve the readability of the time frequency (TF) representation (TFR) of nonstationary signals composed of multiple components with slow varying instantaneous frequency (IF). However, for signals composed of multiple components with fast varying IF, SST still suffers from TF blurs. In this paper, we introduce a time-frequency analysis (TFA) method called matching synchrosqueezing transform (MSST) that achieves a highly concentrated TF representation comparable to the standard TF reassignment methods (STFRM), even for signals with fast varying IF, and furthermore, MSST retains the reconstruction benefit of SST. MSST captures the philosophy of STFRM to simultaneously consider time and frequency variables, and incorporates three estimators (i.e., the IF estimator, the group delay estimator, and a chirp-rate estimator) into a comprehensive and accurate IF estimator. In this paper, we first introduce the motivation of MSST with three heuristic examples. Then we introduce a precise mathematical definition of a class of chirp-like intrinsic mode-type functions that locally can be viewed as a sum of a reasonably small number of approximate chirp signals, and we prove that MSST does indeed succeed in estimating chirp-rate and IF of arbitrary functions in this class and succeed in decomposing these functions. Furthermore, we describe an efficient numerical algorithm for the practical implementation of the MSST, and we provide an adaptive IF extraction method for MSST reconstruction. Finally, we verify the effectiveness of the MSST in practical applications for machine fault diagnosis, including gearbox fault diagnosis for a wind turbine in variable speed conditions and rotor rub-impact fault diagnosis for a dual-rotor turbofan engine. (C) 2017 Elsevier Ltd. All rights reserved.

We introduce a new wavelet-based algorithm to enhance the quality of speech corrupted by multi-talker babble noise. The algorithm comprises three stages: The first stage classifies short frames of the noisy speech as speech-dominated or noise-dominated. We design this classifier specifically for multi-talker babble noise. The second stage performs preliminary de-nosing of noisy speech frames using oversampled wavelet transforms and parallel group thresholding. The final stage performs further denoising by attenuating residual high frequency components in the signal produced by the second stage. A significant improvement in intelligibility and quality was observed in evaluation tests of the algorithm with cochlear implant users.

OBJECTIVE:To evaluate the performance of the EyeTribe compared to the EyeLink for eye movement recordings during a rapid number naming test in healthy control participants. BACKGROUND:With the increasing accessibility of portable, economical, video-based eye trackers such as the EyeTribe, there is growing interest in these devices for eye movement recordings, particularly in the domain of sports-related concussion. However, prior to implementation there is a primary need to establish the validity of these devices. One current limitation of portable eye trackers is their sampling rate (30-60 samples per second, or Hz), which is typically well below the benchmarks achieved by their research-grade counterparts (e.g., the EyeLink, which samples at 500-2000 Hz). METHODS:We compared video-oculographic measurements made using the EyeTribe with those of the EyeLink during a digitized rapid number naming task (the King-Devick test) in a convenience sample of 30 controls. RESULTS:EyeTribe had loss of signal during recording, and failed to reproduce the typical shape of saccadic main sequence relationships. In addition, EyeTribe data yielded significantly fewer detectable saccades and displayed greater variance of inter-saccadic intervals than the EyeLink system. CONCLUSION/CONCLUSIONS:Caution is advised prior to implementation of low-resolution eye trackers for objective saccade assessment and sideline concussion screening.

OBJECTIVE:To determine if native English speakers (NES) perform differently compared to non-native English speakers (NNES) on a sideline-focused rapid number naming task. A secondary aim was to characterize objective differences in eye movement behaviour between cohorts. BACKGROUND:The King-Devick (KD) test is a rapid number-naming task in which numbers are read from left-to-right. This performance measure adds vision-based assessment to sideline concussion testing. Reading strategies differ by language. Concussion may also impact language and attention. Both factors may affect test performance. METHODS:Twenty-seven healthy  NNES and healthy NES performed a computerized KD test under high-resolution video-oculography.  NNES also performed a Bilingual Dominance Scale (BDS) questionnaire to weight linguistic preferences (i.e., reliance on non-English language(s)). RESULTS:Inter-saccadic intervals were significantly longer in  NNES (346.3 ± 78.3 ms vs. 286.1 ± 49.7 ms, p = 0.001), as were KD test times (54.4 ± 15.1 s vs. 43.8 ± 8.6 s, p = 0.002). Higher BDS scores, reflecting higher native language dominance, were associated with longer inter-saccadic intervals in  NNES. CONCLUSION/CONCLUSIONS:These findings have direct implications for the assessment of athlete performance on vision-based and other verbal sideline concussion tests; these results are particularly important given the international scope of sport. Pre-season baseline scores are essential to evaluation in the event of concussion, and performance of sideline tests in the athlete's native language should be considered to optimize both baseline and post-injury test accuracy.

We address the problem of estimating a sparse low-rank matrix from its noisy observation. We propose an objective function consisting of a data-fidelity term and two parameterized non-convex penalty functions. Further, we show how to set the parameters of the non-convex penalty functions, in order to ensure that the objective function is strictly convex. The proposed objective function better estimates sparse low-rank matrices than a convex method which utilizes the sum of the nuclear norm and the El norm. We derive an algorithm (as an instance of ADMM) to solve the proposed problem, and guarantee its convergence provided the scalar augmented Lagrangian parameter is set appropriately. We demonstrate the proposed method for denoising an audio signal and an adjacency matrix representing protein interactions in the 'Escherichia coli' bacteria. (C) 2017 Elsevier B.V. All rights reserved.

Sparse approximate solutions to linear equations are classically obtained via L1 norm regularized least squares, but this method often underestimates the true solution. As an alternative to the L1 norm, this paper proposes a class of nonconvex penalty functions that maintain the convexity of the least squares cost function to be minimized, and avoids the systematic underestimation characteristic of L1 norm regularization. The proposed penalty function is a multivariate generalization of the minimax-concave penalty. It is defined in terms of a new multivariate generalization of the Huber function, which in turn is defined via infimal convolution. The proposed sparse-regularized least squares cost function can be minimized by proximal algorithms comprising simple computations.