Skip to main content

Todd Moon

Followers

23

Following

2

Co-author

1

Public Views

Ernest Jouriles

University of California, San Diego

Md.Touhidur Rahman

Alyson Fletcher

University of California, Santa Cruz

Boston University

Indian Institute of Science

pratyush agrawal

University of Massachusetts Amherst

Interests

Uploads

Papers by Todd Moon

Sparse Bayesian learning boosted by partial erroneous support knowledge

Recovery of sparse signals with unknown clustering pattern in the case of having partial erroneou... more Recovery of sparse signals with unknown clustering pattern in the case of having partial erroneous prior knowledge on the supports of the signal is considered. In this case, we provide a modified sparse Bayesian learning model to incorporate prior knowledge and simultaneously learn the unknown clustering pattern. For this purpose, we add one more layer to support-aided sparse Bayesian learning algorithm (SA-SBL). This layer adds a prior on the shape parameters of Gamma distributions, those modeled to account for the precision of the solution elements. We make the shape parameters depend on the total variations on the estimated supports of the solution. Based on the simulation results, we show that the proposed algorithm is able to modify its erroneous prior knowledge on the supports of the solution and learn the clustering pattern of the true signal by filtering out the incorrect supports from the estimated support set.

Quasicontinuous state hidden Markov models incorporating state histories

The Markovity assumed in conventional hidden Markov models (HMMs) does not necessarily match the ... more The Markovity assumed in conventional hidden Markov models (HMMs) does not necessarily match the statistical structure of many real signals, since many signals have long-term dependencies. In this paper, we generalize the concept of the HMM state to include the history of states or previous models leading to a state, while still limiting the number of basic states to a finite number. This expanded state is efficiently represented using real-numbered states, the fractional part representing the history. State sequence estimation is accomplished using an extension of the Viterbi algorithm. Parameters estimation for state transition probabilities and output distributions is presented.

GPU Acceleration of DSP for Communication Receivers

PubMed, Sep 1, 2017

Graphics processing unit (GPU) implementations of signal processing algorithms can outperform CPU... more Graphics processing unit (GPU) implementations of signal processing algorithms can outperform CPU-based implementations. This paper describes the GPU implementation of several algorithms encountered in a wide range of high-data rate communication receivers including filters, multirate filters, numerically controlled oscillators, and multi-stage digital down converters. These structures are tested by processing the 20 MHz wide FM radio band (88-108 MHz). Two receiver structures are explored: a single channel receiver and a filter bank channelizer. Both run in real time on NVIDIA GeForce GTX 1080 graphics card.

Phase Effects on Speech and Its Influence on Warped Speech

2020 Intermountain Engineering, Technology and Computing (IETC), Oct 2, 2020

A method of transforming speech from one speaker's voice to another is discussed, which opera... more A method of transforming speech from one speaker's voice to another is discussed, which operates by moving speech magnitude information from a source speaker to a target speaker using a process involving dynamic warping in both the time domain and the frequency domain. This process involves only spectral magnitudes, and has been found to introduce significant deleterious signal processing artifacts. It has been found that by reconstruction of phase information significantly improves the quality of the transformed speech.

Time and frequency corrections in a distributed radio network

When using low-cost analog to digital converters (ADCs), synchronization between multiple ADCs is... more When using low-cost analog to digital converters (ADCs), synchronization between multiple ADCs is often difficult to achieve but is desirable for applications such as direction-of-arrival or time-of-arrival processing. A lack of synchronization between multiple ADCs can result in three offsets that are introduced when the sampling begins: a sample timing offset, a frequency offset, and a phase offset. In this paper, an adaptive method for correcting these offsets using software feedback loops is presented. The system is implemented in GNURadio with low-cost RTL-SDR receivers as a proof-of-concept.

Performance of maximum likelihood temperature/emissivity separation of hyperspectral images with correlated Gaussian downwelling radiance

Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processin... more Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processing to extract the emissivity spectra and temperature of the materials in the image. We review a model for the hyperspectral image pixels that we have been developing, for which a maximum likelihood can be used for tem-perature/emissivity separation. Optimization of the ridge-like log likelihood function is accomplished using an “eigenstep” method, in which an eigenvector of the Hessian determines motion along the ride. We report on new testing performance, including bias, showing that bias can be reduced by increasing the number of image samples used.

Detection of Amorphously Shaped Objects Using Spatial Information Detection Enhancement (SIDE)

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Apr 1, 2012

Pattern recognition of amorphously shaped objects such as gas plumes, oil spills, or epidemiologi... more Pattern recognition of amorphously shaped objects such as gas plumes, oil spills, or epidemiological spread is difficult because there is no definite shape to match. We consider detection of such amorphously shaped objects using a neighborhood model which operates on a concept of loose spatial contiguity: there is a significant probability that a pixel surrounded by the object of interest itself contains that object of interest, and boundaries tend to be smooth. These assumptions are distilled into a single-parameter prior probability model to use in a maximum a posteriori hypothesis test. The method is evaluated against synthetic data generated from hyperspectral imagery and DIRSIG simulation results. These tests indicate significant improvement on the ROC curves.

Kurtosis-Based Symbol Timing and Carrier Phase/Frequency Tracking

Entropy, Jun 27, 2021

This article is an open access article distributed under the terms and conditions of the Creative... more

Details on CSA-SBL : An Algorithm for Sparse Bayesian Learning Boosted by Partial Erroneous Support Knowledge

This report provides details on CSA-SBL(VB) algorithm for the recovery of sparse signals with unk... more This report provides details on CSA-SBL(VB) algorithm for the recovery of sparse signals with unknown clustering pattern. More specifically, we deal with the recovery of sparse signals with unknown clustering pattern in the case of having partial erroneous prior knowledge on the supports of the signal. In [1], we provided a modified sparse Bayesian learning model to incorporate prior knowledge and simultaneously learn the unknown clustering pattern. For this purpose, we added one more layer to support-aided sparse Bayesian learning algorithm (SA-SBL) that was proposed in [2]. This layer adds a prior on the shape parameters of Gamma distributions, those modeled to account for the precision of the solution elements. We made the shape parameters depend on the total variations on the estimated supports of the solution. The inference technique for implementing this algorithm is variational Bayes (VB). Part of this work has been published in [1, 3].

Details on Gaussian Process Regression (GPR) and Semi-GPR Modeling

This report tends to provide details on how to perform predictions using Gaussian process regress... more This report tends to provide details on how to perform predictions using Gaussian process regression (GPR) modeling. In this case, we represent proofs for prediction using non-parametric GPR modeling for noise-free predictions as well as prediction using semi-parametric GPR for noisy observations. 1. GAUSSIAN PROCESS REGRESSION Gaussian processes (GPs) are widely used for modeling a phenomenon based on the observed spatiotemporal data [12]. A GP can be used as a tool for either classification or regression [1, 2, 12]. GPs have been used for decades as a supervised learning tool for regression problems known as Gaussian process regression (GPR) models [1, 2], and are also referred to as kriging, named after the mining engineer D.G. Krige in the geostatistics literature [3-5]. GPR models and kriging methods are applicable to a variety of problems such as the prediction and estimation of temperature, precipitation, missing pixel and un-mixing of pixels in hyperspectral imaging (HSI), human head pose estimation, concentration of carbon dioxide in the atmosphere, etc. [2, 6-11]. In GP modeling, it is assumed that the phenomenon of interest (PoI) can be evaluated via an unknown and probably nonlinear function, which we denote by f (•). The arguments of the function comprise a variable set u referred to as the input data. For example, u can be defined as u = [u x , u y , u z , t] T , where (u x , u y , u z) and t denote the spatial and temporal information about the measurements, respectively. Unlike parametric models such as linear regression, GP is non-parametric. In GP one defines a probability distribution function as a prior over the unknown function f (•), directly. In other words, GP defines a distribution over functions in the function space and the inference is performed directly in this space [2]. This is more general than a parametric model such as Bayesian linear regression, where the prior distribution is defined over the space of parameters. The GP model treats any observation as an outcome of a Gaussian random variable, and all of these random variables are jointly Gaussian. With this setting, any well-defined GP model only needs a mean accompanied with a positive definite covariance function. Under this assumption, GP provides a posterior distribution over the unknown function f once data are observed. Therefore, for any set of N observations with the input data set {u 1 ,. .. , u N }, GP assumes that the distribution p(f (u 1),. .. , f (u N)) is jointly Gaussian with some mean µ(U) and a covariance matrix K(U), where U := [u 1 , ..., u N ]. The entry in row i and column j of K(U) is denoted by [K(U)] ij = κ(u i , u j), where κ(., .) is a positive definite kernel function. The kernel function specifies

Details on Amp-B-SBL: An Algorithm for Recovery of Clustered Sparse Signals Using Approximate Message Passing

HAL (Le Centre pour la Communication Scientifique Directe), Jan 9, 2019

Solving the inverse problem of compressive sensing in the context of single measurement vector (S... more Solving the inverse problem of compressive sensing in the context of single measurement vector (SMV) problem with an unknown block-sparsity structure is considered. For this purpose, we propose a sparse Bayesian learning (SBL) algorithm simplified via the approximate message passing (AMP) framework. In order to encourage the block-sparsity structure, we incorporate the concept of total variation, called Sigma-Delta, as a measure of blocksparsity on the support set of the solution. The AMP framework reduces the computational load of the proposed SBL algorithm and as a result makes it faster compared to the message passing framework. Furthermore, in terms of the mean-squared error between the true and the reconstructed solution, the algorithm demonstrates an encouraging improvement compared to the other algorithms.

Sparse Recovery with Unknown Sparsity Pattern via Multiple Measurement Vectors

Source Localization and Room Mapping Using Information Derived from Independent Component Analysis

Convolutive independent component analysis (ICA) algorithms, which have proven capable at separat... more Convolutive independent component analysis (ICA) algorithms, which have proven capable at separating convolutively mixed signals, can also provide information about the geometry of the setting. This geometry includes source location information and wall locations (room shape). From the multi-input/multioutput impulse responses learned from convolutive ICA, peaks indicating direct path or reflection delays are extracted. The location of sensors (or sources) is obtained using hyperbolic geometry based on direct path time delays. Delays from reflected paths learned from the impulses responses correspond geometrically to ellipses, which are tangent at the reflecting point. Ellipse tangent directions are clustered to determine wall locations. Following a summary of the method, experiments are presented on actual room measurements.

Using Dual Approximation for Best Linear Unbiased Estimators in Continuous Time, with Application to Continuous-Time Phase Estimation

2022 Intermountain Engineering, Technology and Computing (IETC)

CAMP: An Algorithm to Recover Sparse Signals with Unknown Clustering Pattern Using Approximate Message Passing

I n f o r m a t i o n D y na m ic s L a b o r a t o r y U t a h S t a te U n i v e r s i t y

A Tool for Training Speech Imitation Accuracy

2018 52nd Asilomar Conference on Signals, Systems, and Computers, 2018

Dysarthria is a neurological motor speech disorder that commonly results in reduced intelligibili... more Dysarthria is a neurological motor speech disorder that commonly results in reduced intelligibility. Communication partners can learn to better understand the speech of someone with dysarthria through perceptual training. Vocal imitation of the degraded speech during perceptual training has been shown to elevate this learning. A tool that provides the learner with real-time feedback regarding the accuracy of their imitation attempts during training may further enhance this learning. We describe a training tool that compares dysarthric speech productions with the imitation attempts of healthy subjects, using a two-level dynamic warp that accounts for both spectral and temporal degradation. Feature vectors derived from both the spectrogram and LPC are examined.

Contravariant Adaptation on the Manifold of Invertible Matrix Transfer Functions

IEEE Signal Processing Letters, 2017

Adaptation using contravariant, sometimes called the natural gradient, provides true steepest des... more Adaptation using contravariant, sometimes called the natural gradient, provides true steepest descent learning on a manifold such as the manifold of invertible matrices or the manifold of invertible matrix transfer functions. In this paper, the contravariant gradient on the manifold of invertible matrix transfer functions is presented using the language of differential geometry. The covariant gradient (the usual gradient) is converted to the contravariant or natural gradient using the inverse of a Riemannian metric tensor which is derived for the manifold of invertible matrix transfer functions. This has not been previously done using the language of differential geometry.

Target codes

International Journal of Mobile Network Design and Innovation

A Tree-Search Algorithm for Signal Multi-User Communications Detection and Parameter Estimation in

This paper considers the problem of simultaneously detecting the information bits and estimating ... more This paper considers the problem of simultaneously detecting the information bits and estimating signal amplitudes and phases in a K-user asynchronous direct-sequence spread-spectrum multiple-access communication system. The joint maximum-likelihood (ML) estimator has a computational complexity that is exponential in the total number of bits transmitted, and thus does not represent a practical solution to the problem. We consider an estimator that combines a suboptimum tree-search algorithm with a recursive least-squares estimator of com lex signal amplitude. The complexity of this estimator is close to that of the joint ML receiver. This receiver has the advantage that the transmitted signal powers and phases are extracted from the received signal in an adaptive fashion without using a test sequence.

Performance of maximum likelihood temperature/emissivity separation of hyperspectral images with correlated Gaussian downwelling radiance

2016 50th Asilomar Conference on Signals, Systems and Computers, 2016

Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processin... more Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processing to extract the emissivity spectra and temperature of the materials in the image. We review a model for the hyperspectral image pixels that we have been developing, for which a maximum likelihood can be used for tem-perature/emissivity separation. Optimization of the ridge-like log likelihood function is accomplished using an “eigenstep” method, in which an eigenvector of the Hessian determines motion along the ride. We report on new testing performance, including bias, showing that bias can be reduced by increasing the number of image samples used.

Sparse Bayesian learning boosted by partial erroneous support knowledge

Recovery of sparse signals with unknown clustering pattern in the case of having partial erroneou... more Recovery of sparse signals with unknown clustering pattern in the case of having partial erroneous prior knowledge on the supports of the signal is considered. In this case, we provide a modified sparse Bayesian learning model to incorporate prior knowledge and simultaneously learn the unknown clustering pattern. For this purpose, we add one more layer to support-aided sparse Bayesian learning algorithm (SA-SBL). This layer adds a prior on the shape parameters of Gamma distributions, those modeled to account for the precision of the solution elements. We make the shape parameters depend on the total variations on the estimated supports of the solution. Based on the simulation results, we show that the proposed algorithm is able to modify its erroneous prior knowledge on the supports of the solution and learn the clustering pattern of the true signal by filtering out the incorrect supports from the estimated support set.

Quasicontinuous state hidden Markov models incorporating state histories

The Markovity assumed in conventional hidden Markov models (HMMs) does not necessarily match the ... more The Markovity assumed in conventional hidden Markov models (HMMs) does not necessarily match the statistical structure of many real signals, since many signals have long-term dependencies. In this paper, we generalize the concept of the HMM state to include the history of states or previous models leading to a state, while still limiting the number of basic states to a finite number. This expanded state is efficiently represented using real-numbered states, the fractional part representing the history. State sequence estimation is accomplished using an extension of the Viterbi algorithm. Parameters estimation for state transition probabilities and output distributions is presented.

GPU Acceleration of DSP for Communication Receivers

PubMed, Sep 1, 2017

Graphics processing unit (GPU) implementations of signal processing algorithms can outperform CPU... more Graphics processing unit (GPU) implementations of signal processing algorithms can outperform CPU-based implementations. This paper describes the GPU implementation of several algorithms encountered in a wide range of high-data rate communication receivers including filters, multirate filters, numerically controlled oscillators, and multi-stage digital down converters. These structures are tested by processing the 20 MHz wide FM radio band (88-108 MHz). Two receiver structures are explored: a single channel receiver and a filter bank channelizer. Both run in real time on NVIDIA GeForce GTX 1080 graphics card.

Phase Effects on Speech and Its Influence on Warped Speech

2020 Intermountain Engineering, Technology and Computing (IETC), Oct 2, 2020

A method of transforming speech from one speaker's voice to another is discussed, which opera... more A method of transforming speech from one speaker's voice to another is discussed, which operates by moving speech magnitude information from a source speaker to a target speaker using a process involving dynamic warping in both the time domain and the frequency domain. This process involves only spectral magnitudes, and has been found to introduce significant deleterious signal processing artifacts. It has been found that by reconstruction of phase information significantly improves the quality of the transformed speech.

Time and frequency corrections in a distributed radio network

When using low-cost analog to digital converters (ADCs), synchronization between multiple ADCs is... more When using low-cost analog to digital converters (ADCs), synchronization between multiple ADCs is often difficult to achieve but is desirable for applications such as direction-of-arrival or time-of-arrival processing. A lack of synchronization between multiple ADCs can result in three offsets that are introduced when the sampling begins: a sample timing offset, a frequency offset, and a phase offset. In this paper, an adaptive method for correcting these offsets using software feedback loops is presented. The system is implemented in GNURadio with low-cost RTL-SDR receivers as a proof-of-concept.

Performance of maximum likelihood temperature/emissivity separation of hyperspectral images with correlated Gaussian downwelling radiance

Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processin... more Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processing to extract the emissivity spectra and temperature of the materials in the image. We review a model for the hyperspectral image pixels that we have been developing, for which a maximum likelihood can be used for tem-perature/emissivity separation. Optimization of the ridge-like log likelihood function is accomplished using an “eigenstep” method, in which an eigenvector of the Hessian determines motion along the ride. We report on new testing performance, including bias, showing that bias can be reduced by increasing the number of image samples used.

Detection of Amorphously Shaped Objects Using Spatial Information Detection Enhancement (SIDE)

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Apr 1, 2012

Pattern recognition of amorphously shaped objects such as gas plumes, oil spills, or epidemiologi... more Pattern recognition of amorphously shaped objects such as gas plumes, oil spills, or epidemiological spread is difficult because there is no definite shape to match. We consider detection of such amorphously shaped objects using a neighborhood model which operates on a concept of loose spatial contiguity: there is a significant probability that a pixel surrounded by the object of interest itself contains that object of interest, and boundaries tend to be smooth. These assumptions are distilled into a single-parameter prior probability model to use in a maximum a posteriori hypothesis test. The method is evaluated against synthetic data generated from hyperspectral imagery and DIRSIG simulation results. These tests indicate significant improvement on the ROC curves.

Kurtosis-Based Symbol Timing and Carrier Phase/Frequency Tracking

Entropy, Jun 27, 2021

This article is an open access article distributed under the terms and conditions of the Creative... more

Details on CSA-SBL : An Algorithm for Sparse Bayesian Learning Boosted by Partial Erroneous Support Knowledge

This report provides details on CSA-SBL(VB) algorithm for the recovery of sparse signals with unk... more This report provides details on CSA-SBL(VB) algorithm for the recovery of sparse signals with unknown clustering pattern. More specifically, we deal with the recovery of sparse signals with unknown clustering pattern in the case of having partial erroneous prior knowledge on the supports of the signal. In [1], we provided a modified sparse Bayesian learning model to incorporate prior knowledge and simultaneously learn the unknown clustering pattern. For this purpose, we added one more layer to support-aided sparse Bayesian learning algorithm (SA-SBL) that was proposed in [2]. This layer adds a prior on the shape parameters of Gamma distributions, those modeled to account for the precision of the solution elements. We made the shape parameters depend on the total variations on the estimated supports of the solution. The inference technique for implementing this algorithm is variational Bayes (VB). Part of this work has been published in [1, 3].

Details on Gaussian Process Regression (GPR) and Semi-GPR Modeling

This report tends to provide details on how to perform predictions using Gaussian process regress... more This report tends to provide details on how to perform predictions using Gaussian process regression (GPR) modeling. In this case, we represent proofs for prediction using non-parametric GPR modeling for noise-free predictions as well as prediction using semi-parametric GPR for noisy observations. 1. GAUSSIAN PROCESS REGRESSION Gaussian processes (GPs) are widely used for modeling a phenomenon based on the observed spatiotemporal data [12]. A GP can be used as a tool for either classification or regression [1, 2, 12]. GPs have been used for decades as a supervised learning tool for regression problems known as Gaussian process regression (GPR) models [1, 2], and are also referred to as kriging, named after the mining engineer D.G. Krige in the geostatistics literature [3-5]. GPR models and kriging methods are applicable to a variety of problems such as the prediction and estimation of temperature, precipitation, missing pixel and un-mixing of pixels in hyperspectral imaging (HSI), human head pose estimation, concentration of carbon dioxide in the atmosphere, etc. [2, 6-11]. In GP modeling, it is assumed that the phenomenon of interest (PoI) can be evaluated via an unknown and probably nonlinear function, which we denote by f (•). The arguments of the function comprise a variable set u referred to as the input data. For example, u can be defined as u = [u x , u y , u z , t] T , where (u x , u y , u z) and t denote the spatial and temporal information about the measurements, respectively. Unlike parametric models such as linear regression, GP is non-parametric. In GP one defines a probability distribution function as a prior over the unknown function f (•), directly. In other words, GP defines a distribution over functions in the function space and the inference is performed directly in this space [2]. This is more general than a parametric model such as Bayesian linear regression, where the prior distribution is defined over the space of parameters. The GP model treats any observation as an outcome of a Gaussian random variable, and all of these random variables are jointly Gaussian. With this setting, any well-defined GP model only needs a mean accompanied with a positive definite covariance function. Under this assumption, GP provides a posterior distribution over the unknown function f once data are observed. Therefore, for any set of N observations with the input data set {u 1 ,. .. , u N }, GP assumes that the distribution p(f (u 1),. .. , f (u N)) is jointly Gaussian with some mean µ(U) and a covariance matrix K(U), where U := [u 1 , ..., u N ]. The entry in row i and column j of K(U) is denoted by [K(U)] ij = κ(u i , u j), where κ(., .) is a positive definite kernel function. The kernel function specifies

Details on Amp-B-SBL: An Algorithm for Recovery of Clustered Sparse Signals Using Approximate Message Passing

HAL (Le Centre pour la Communication Scientifique Directe), Jan 9, 2019

Solving the inverse problem of compressive sensing in the context of single measurement vector (S... more Solving the inverse problem of compressive sensing in the context of single measurement vector (SMV) problem with an unknown block-sparsity structure is considered. For this purpose, we propose a sparse Bayesian learning (SBL) algorithm simplified via the approximate message passing (AMP) framework. In order to encourage the block-sparsity structure, we incorporate the concept of total variation, called Sigma-Delta, as a measure of blocksparsity on the support set of the solution. The AMP framework reduces the computational load of the proposed SBL algorithm and as a result makes it faster compared to the message passing framework. Furthermore, in terms of the mean-squared error between the true and the reconstructed solution, the algorithm demonstrates an encouraging improvement compared to the other algorithms.

Sparse Recovery with Unknown Sparsity Pattern via Multiple Measurement Vectors

Source Localization and Room Mapping Using Information Derived from Independent Component Analysis

Convolutive independent component analysis (ICA) algorithms, which have proven capable at separat... more Convolutive independent component analysis (ICA) algorithms, which have proven capable at separating convolutively mixed signals, can also provide information about the geometry of the setting. This geometry includes source location information and wall locations (room shape). From the multi-input/multioutput impulse responses learned from convolutive ICA, peaks indicating direct path or reflection delays are extracted. The location of sensors (or sources) is obtained using hyperbolic geometry based on direct path time delays. Delays from reflected paths learned from the impulses responses correspond geometrically to ellipses, which are tangent at the reflecting point. Ellipse tangent directions are clustered to determine wall locations. Following a summary of the method, experiments are presented on actual room measurements.

Using Dual Approximation for Best Linear Unbiased Estimators in Continuous Time, with Application to Continuous-Time Phase Estimation

2022 Intermountain Engineering, Technology and Computing (IETC)

CAMP: An Algorithm to Recover Sparse Signals with Unknown Clustering Pattern Using Approximate Message Passing

I n f o r m a t i o n D y na m ic s L a b o r a t o r y U t a h S t a te U n i v e r s i t y

A Tool for Training Speech Imitation Accuracy

2018 52nd Asilomar Conference on Signals, Systems, and Computers, 2018

Dysarthria is a neurological motor speech disorder that commonly results in reduced intelligibili... more Dysarthria is a neurological motor speech disorder that commonly results in reduced intelligibility. Communication partners can learn to better understand the speech of someone with dysarthria through perceptual training. Vocal imitation of the degraded speech during perceptual training has been shown to elevate this learning. A tool that provides the learner with real-time feedback regarding the accuracy of their imitation attempts during training may further enhance this learning. We describe a training tool that compares dysarthric speech productions with the imitation attempts of healthy subjects, using a two-level dynamic warp that accounts for both spectral and temporal degradation. Feature vectors derived from both the spectrogram and LPC are examined.

Contravariant Adaptation on the Manifold of Invertible Matrix Transfer Functions

IEEE Signal Processing Letters, 2017

Adaptation using contravariant, sometimes called the natural gradient, provides true steepest des... more Adaptation using contravariant, sometimes called the natural gradient, provides true steepest descent learning on a manifold such as the manifold of invertible matrices or the manifold of invertible matrix transfer functions. In this paper, the contravariant gradient on the manifold of invertible matrix transfer functions is presented using the language of differential geometry. The covariant gradient (the usual gradient) is converted to the contravariant or natural gradient using the inverse of a Riemannian metric tensor which is derived for the manifold of invertible matrix transfer functions. This has not been previously done using the language of differential geometry.

Target codes

International Journal of Mobile Network Design and Innovation

A Tree-Search Algorithm for Signal Multi-User Communications Detection and Parameter Estimation in

This paper considers the problem of simultaneously detecting the information bits and estimating ... more This paper considers the problem of simultaneously detecting the information bits and estimating signal amplitudes and phases in a K-user asynchronous direct-sequence spread-spectrum multiple-access communication system. The joint maximum-likelihood (ML) estimator has a computational complexity that is exponential in the total number of bits transmitted, and thus does not represent a practical solution to the problem. We consider an estimator that combines a suboptimum tree-search algorithm with a recursive least-squares estimator of com lex signal amplitude. The complexity of this estimator is close to that of the joint ML receiver. This receiver has the advantage that the transmitted signal powers and phases are extracted from the received signal in an adaptive fashion without using a test sequence.

Performance of maximum likelihood temperature/emissivity separation of hyperspectral images with correlated Gaussian downwelling radiance

2016 50th Asilomar Conference on Signals, Systems and Computers, 2016

Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processin... more Temperature/Emissivity separation (TES) techniques are desirable in hyperspectral image processing to extract the emissivity spectra and temperature of the materials in the image. We review a model for the hyperspectral image pixels that we have been developing, for which a maximum likelihood can be used for tem-perature/emissivity separation. Optimization of the ridge-like log likelihood function is accomplished using an “eigenstep” method, in which an eigenvector of the Hessian determines motion along the ride. We report on new testing performance, including bias, showing that bias can be reduced by increasing the number of image samples used.