The rapid proliferation of smart medical sensors and wearable healthcare devices has significantly increased the demand for low-latency computing services. The data generated by these monitoring devices must be processed efficiently by edge computing systems. However, the dynamic and unpredictable nature of incoming computational tasks requires real-time, robust, and priority-sensitive task allocation mechanisms. This study introduces an on-line task allocation algorithm for edge computing in healthcare systems, leveraging Deep Q-Learning to address these challenges. A multi-objective optimization framework is formulated, enabling the task allocator to enhance system efficiency without requiring prior knowledge of future tasks. The algorithm maximizes the number of allocated tasks while prioritizing critical ones and minimizing counterproductive operation times. A deep Q-Network is implemented to estimate Q-values for each state-action pair, empowering the task allocator to select actions with the highest Q-value consistently. The performance of the proposed algorithm is rigorously evaluated in a simulated environment. The results show that the algorithm effectively learns an optimal policy for the task allocation problem, producing an average improvement of 74.6% in the overall performance of the system.

The problem of joint shaping and precoding is studied in this paper. We introduce statistical dependencies among consecutive symbols to shape the constellation while minimizing the total transmit power when the signal goes through the precoding filter. We propose a stationary Markovian model for optimizing the transition probability of transmit symbols to avoid high-energy sequences when convolved with the precoding filter. A new algorithm based on arithmetic coding is proposed to generate a shaped sequence of symbols with the given Markov model transition probabilities.

We propose a novel digital-to-analog converter (DAC) weighting architecture that statistically minimizes the distortion caused by random current mismatches. Unlike binary, thermometer-coded, and segmented DACs, the current weights of the proposed architecture are not an integer power of 2 or any other integer number. We present a heuristic algorithm for a static mapping of DAC input codewords into corresponding DAC switches. High-level Matlab simulations are performed to illustrate the static performance improvement over the segmented structure.

In this paper, a binaural beamforming algorithm for hearing aid applications is introduced. The beamforming algorithm is designed to be robust to some error in the estimate of the target speaker direction. The algorithm has two main components: a robust target linearly constrained minimum variance (TLCMV) algorithm based on imposing two constraints around the estimated direction of the target signal, and a post-processor to help with the preservation of binaural cues. The robust TLCMV provides a good level of noise reduction and low level of target distortion under realistic conditions. The post-processor enhances the beamformer abilities to preserve the binaural cues for both diffuse-like background noise and directional interferers (competing speakers), while keeping a good level of noise reduction. The introduced algorithm does not require knowledge or estimation of the directional interferers' directions nor the second-order statistics of noise-only components. The introduced algorithm requires an estimate of the target speaker direction, but it is designed to be robust to some deviation from the estimated direction. Compared with recently proposed state-of-the-art methods, comprehensive evaluations are performed under complex realistic acoustic scenarios generated in both anechoic and mildly reverberant environments, considering a mismatch between estimated and true sources direction of arrival. Mismatch between the anechoic propagation models used for the design of the beamformers and the mildly reverberant propagation models used to generate the simulated directional signals is also considered. The results illustrate the robustness of the proposed algorithm to such mismatches.

In this paper, we propose a musical-noise-free blind speech extraction method using a microphone array for application to nonstationary noise. In our previous study, it was found that optimized iterative spectral subtraction (SS) results in speech enhancement with almost no musical noise generation, but this method is valid only for stationary noise. The proposed method consists of iterative blind dynamic noise estimation by, e.g., independent component analysis (ICA) or multichannel Wiener filtering, and musical-noise-free speech extraction by modified iterative SS, where multiple iterative SS is applied to each channel while maintaining the multichannel property reused for the dynamic noise estimators. Also, in relation to the proposed method, we discuss the justification of applying ICA to signals nonlinearly distorted by SS. From objective and subjective evaluations simulating a real-world hands-free speech communication system, we reveal that the proposed method outperforms the conventional methods.

In this letter, some results involving a form of least-squares beamformer are derived based only on directional criteria. In far-field assumptions (i.e., plane wave propagation model for the signals captured), we show that certain factors appearing in the beamformer coefficients calculation, which are usually formulated using complicated integrals, can be computed using closed-form expressions involving familiar, physically meaningful quantities. Next, by resorting to a limiting case, we demonstrate a clear theoretical link between the resulting solution and MVDR beamforming in cylindrically isotropic noise fields. We then discuss a solution to preserve spatial cues if desired which also allows to easily control and modulate the enhancement strength of the beamformer. Some experimental results are finally given in a challenging real-world environment, showing the merits of the approach.

In this correspondence, a unified approach to the autoregressive (AR) modeling of power spectral densities is described. We show that by introducing auxiliary sequences, the minimization of several customary spectral distances can be performed with the exact same convenient approach, whether a gradient-descent or a Newton/quasi-Newton descent is chosen. Moreover, we extend the usual optimization of unnormalized AR coefficients to a two-step optimization of normalized AR coefficients, and provide evidence that this alternative approach can accelerate convergence and provide robustness to erroneous initializations. Convergence and modeling results are also given.

The current generation of digital hearing aids allows the implementation of advanced noise reduction schemes. However, most current noise reduction algorithms are monaural and are therefore intended for only bilateral hearing aids. Recently, binaural in contrast to monaural noise reduction schemes have been proposed, targeting future high-end binaural hearing aids. Those new types of hearing aids would allow the sharing of information/signals received from both left and right hearing aid microphones (via a wireless link) to generate an output for the left and right ear. This paper presents a novel noise power spectral density estimator for binaural hearing aids operating in a diffuse noise field environment, by taking advantage of the left and right reference signals that will be accessible, as opposed to the single reference signal currently available in bilateral hearing aids. In contrast with some previously published noise estimation methods for hearing aids or speech enhancement, the proposed noise estimator does not assume stationary noise, it can work for colored noise in a diffuse noise field, it does not require a voice activity detection, the noise power spectrum can be estimated during speech activity or not, it does not experience noise tracking latency and most importantly, it is not essential for the target speaker to be in front of the binaural hearing aid user to estimate the noise power spectrum, i.e., the direction of arrival of the source speech signal can be arbitrary. Finally, the proposed noise estimator can be combined with any hearing aid noise reduction technique, where the accuracy of the noise estimation can be critical to achieve a satisfactory de-noising performance.

Motivated by the reportedly strong performance of particle filters (PFs) for noise reduction on essentially linear speech production models, and the mounting evidence that the introduction of nonlinearities can lead to a refined speech model, this paper presents a study of PF solutions to the problem of speech enhancement in the context of nonlinear, neural-type speech models. Several variations of a global model are presented (single/multiple neurons; bias/no bias), and corresponding PF solutions are derived. Different importance functions are given when beneficial, Rao-Blackwellization is proposed when possible, and dual/nondual versions of each algorithms are presented. The method shown can handle both white and colored noise. Using a variety of speech and noise signals and different objective quality measures, the performance of these algorithms are evaluated against other PF solutions running on linear models, as well as some traditional enhancement algorithms. A certain hierarchy in performance is established between each algorithm in the paper. Depending on the experimental conditions, the best-performing algorithms are a classical Rao-Blackwellized particle filter (RBPF) running on a linear model, and a proposed PF employing a nondual, nonlinear model with multiple neurons and no biases. With consistence, the neural-network-based PF outperforms RBPF at low signal-to-noise ratio (SNR).

This paper proposes a selective coefficient update algorithm for reducing the complexity of the proportionate normalized least- mean-square (P-NLMS) class of algorithms. It is shown that an optimal subset of coefficients to update, namely those minimizing the a posteriori error, cannot be constructed efficiently. A sub- optimal block-based coefficient selection algorithm is presented that combines proportional weighting of the input signal vector with fast ranking methods. It is compared to existing sub-optimal algorithms with respect to complexity overhead and convergence rate. Simulations show that the proposed algorithm produces performance approaching that of the optimal subset while maintaining a low coefficient selection overhead.

This paper presents a new approach for speech feature enhancement in the log-spectral domain for noisy speech recognition. A switching linear dynamic model (SLDM) is explored as a parametric model for the clean speech distribution. Each multivariate linear dynamic model (LDM) is associated with the hidden state of a hidden Markov model (HMM) as an attempt to describe the temporal correlations among adjacent frames of speech features. The state transition on the Markov chain is the process of activating a different LDM or activating some of them simultaneously by different probabilities generated by the HMM. Rather than holding a transition probability for the whole process, a connectionist model is employed to learn the time variant transition probabilities. With the resulting SLDM as the speech model and with a model for the noise, speech and noise are jointly tracked by means of switching Kalman filtering. Comprehensive experiments are carried out using the Aurora2 database to evaluate the new algorithm. The results show that the new SLDM approach can further improve the speech feature enhancement performance in terms of noise-robust recognition accuracy, since the transition probabilities among the LDMs can be described more precisely at each time point.

Among the various types of existing speech enhancement algorithms, respective advantages are usually well-known and well-studied, making it possible to choose the right algorithm given a certain type of audio quality requirement (e.g. background noise intrusiveness, naturalness of the speech, etc.). Very little has however been said so far on the quality of speech enhanced by particle filters (PFs). In this paper, we show the detailed results of the analysis of PF-enhanced speech signals. This analysis is conducted using several objective measures, and is based on the comparison with other enhancement algorithms. We find that in general, PF-based algorithms yield speech signals which are among the most natural-sounding, with a residual noise that can be seen as a white noise with its variance "modulated" by the resulting speech.

For feedforward multichannel active noise control (ANC) systems, the use of adaptive finite-impulse response (FIR) filters is a popular solution, and the multichannel filtered-x least-mean-square (FX-LMS) algorithm is the most commonly used algorithm. The drawback of the FX-LMS is the slow convergence speed, especially for broadband multichannel systems. Recently, some fast affine projection algorithms have been introduced for multichannel ANC, as an interesting alternative to the FX-LMS algorithm. They can provide a significantly improved convergence speed at a reasonable additional computational cost. Yet, the additional computational cost or the potential numerical instability in some of the recently proposed algorithms can prevent the use of those algorithms for some applications. In this paper, we propose two pseudo-affine projection algorithms for multichannel ANC: one based on the Gauss-Seidel method and one based on dichotomous coordinate descent (DCD) iterations. It is shown that the proposed algorithms typically have a lower complexity than the previously published fast affine projection algorithms for ANC, with very similar good convergence properties and good numerical stability. Thus, the proposed algorithms are an interesting alternative to the standard FX-LMS algorithm for ANC, providing an improved performance for a computational load of the same order

This paper presents an approach to enhance speech feature estimation in the log spectral domain under noisy environments. A higher-order switching linear dynamic model (SLDM) is explored as a parametric model for the clean speech distribution, which enforces a state transition in the feature space and captures the smooth time evolution of speech conditioned on the state sequence. The clean speech components are estimated by means of an Interacting Multiple Model (IMM) algorithm. Our experimental results show that increasing the order of the linear dynamic model in the SLDM and the introduction of transition probabilities among the linear dynamic models can improve the performance of SLDM systems in feature compensation for robust speech recognition.

A new algorithm for blind signal separation of speech signals that does not require pre-whitening is proposed in this paper. The algorithm is based on second order optimization using Riemannian geometry. The algorithm employs several practical approximations to the Hessian matrix of the maximum-likelihood blind separation cost function, to produce a computationally efficient algorithm that is capable of working on-line. Simulation results show the improved performance of the proposed algorithm with different mixing data

This paper examines the use of different feature sets and normalization techniques for a signature and password verifier. The verifier made use of the dynamic time warping (DTW) algorithm. Features that incorporated the pen velocity were found to be the strongest performers in tests against informed forgeries. Overall, password verification did not perform as well as signature verification. On average, the equal error rate for passwords was 2.7% higher than for signatures. Most signature and password features achieved their best performance when they were power-normalized, although normalization in both time and power was sometimes beneficial as well

Rao-Blackwellised Particle Filters (RBPFs) are a class of Particle Filters (PFs) that exploit conditional dependencies between parts of the state to estimate. By doing so, RBPFs can improve the estimation quality while also reducing the over-all computational load in comparison to original PFs. However, the computational complexity is still too high for many real-time applications. In this paper, we propose a modified RBPF that requires a single Kalman Filter (KF) iteration per input sample. Comparative experiments show that while good convergence can still be obtained, computational efficiency is always drastically increased, making this algorithm an option to consider for real-time implementations.

This letter presents a new approach to enhance speech feature estimation in the log spectral domain under noisy environments. A mixture of linear dynamic models with an architecture similar to the so-called mixture of experts (ME) is investigated to describe the clean speech feature distribution parametrically. Switching Kalman filters are adapted to the proposed model, and they estimate the clean speech components by means of a generalized pseudo-Bayesian (GPB) algorithm. Experimental results suggest that compared with previous methods, the proposed approach can be more powerful to compensate the noisy speech features for robust speech recognition

A masking threshold constrained Kalman filter for speech enhancement is derived in the paper. A key step in a traditional Kalman filter requires minimizing an estimation error variance between a clean signal and its estimation. Our new method is to minimize the estimation error variance under the constraint that the energy of the estimation error is smaller than a masking threshold, computed from both time-domain forward masking and frequency-domain simultaneous masking properties of human auditory systems. The new Kalman filter provides a theoretical base for the application of the masking properties in Kalman filtering for speech enhancement. Due to the high computation cost of the proposed perceptually constrained Kalman filter, a perceptual post-filter concatenated with a standard Kalman filter is also proposed as a heuristic alternative for real-time implementation. The post-filter is constructed to make the estimation error obtained from the Kalman filter lower than the masking threshold. A wavelet Kalman filter with post-filtering is introduced to further reduce the computational load. Experimental results with colored noise show that the new constrained Kalman filter method produces the best performance when compared with other recent methods, and that the proposed heuristics with post-filtering can also produce a significant performance gain over other recent methods.

In this paper, a switching Kalman filter (SKF) with a generalized pseudo Bayesian (GPB) algorithm of order 1 is applied to the problem of speech enhancement. It is proposed to use the masking properties of human auditory systems as a perceptual post-filter concatenated with the GPB algorithm. Experiments show that the proposed algorithm can achieve an improvement both in terms of speech quality (PESQ score, ITU-T P.862) and of word recognition rate at low SNR.

A wavelet Kalman filter combined with masking properties of the human auditory system is proposed for enhancing speech degraded by colored noise. A wavelet domain speech state-space model is first constructed, with a corresponding Kalman filter. A post-filter, considering both time and frequency masking properties, is then applied. From the calculated masking threshold, the noisy speech spectrum is further enhanced. Simulation results show that the proposed approach has the best performance compared with either classic or recently introduced methods, evaluated using ITU-T P.862 PESQ scores. The proposed approach also has a reduced complexity compared to a time-domain implementation of the Kalman filter with a perceptual post-filter.

In some communication systems, the transmitted signal is contaminated by impulsive noise with a non-Gaussian distribution. Non-Gaussian noise causes significant performance degradation to communication receivers. In this paper, we apply a recurrent neural equalizer to impulsive noise channels, for which the performance of neural network equalizers has never been evaluated. This new application is motivated from the fact that the unscented Kalman filter (UKF), which is suited for training of the recurrent neural equalizer, provides a higher accuracy than the extended Kalman filter (EKF) in capturing the statistical characteristics for non-Gaussian random variables. The performance of the recurrent neural equalizer is evaluated for impulsive noise channels through Monte Carlo simulations. The results support the superiority of the UKF to the EKF in compensating the effect of non-Gaussian impulsive noise.

This letter uses the twisted-pair common-mode signal as the input of an adaptive canceller to remove some wideband crosstalk in a digital subscriber line (xDSL) differential signal. Simulations using simple crosstalk and line balance models show the potential benefits of the canceller to improve the signal-to-noise ratio of an xDSL channel.

Real-time recurrent learning (RTRL), commonly employed for training a fully connected recurrent neural network (RNN), has a drawback of slow convergence rate. In the light of this deficiency, a decision feedback recurrent neural equalizer (DFRNE) using the RTRL requires long training sequences to achieve good performance. In this paper, extended Kalman filter (EKF) algorithms based on the RTRL for the DFRNE are presented in state-space formulation of the system, in particular for complex-valued signal processing. The main features of global EKF and decoupled EKF algorithms are fast convergence and good tracking performance. Through nonlinear channel equalization, performance of the DFRNE with the EKF algorithms is evaluated and compared with that of the DFRNE with the RTRL.

It is well known that fast affine projection algorithms can produce a good trade-off between convergence speed and computational complexity. In this paper, we propose a new pseudo fast affine projection algorithm based on the Gauss-Seidel scheme for multichannel active noise control (ANC) systems. It is shown that the proposed algorithm has a lower complexity than previously published fast affine projection algorithms, with similar convergence properties and better numerical stability.

In this work, we present five new versions of the fast Newton transversal filter algorithm (FNTF). The first algorithm is based on a simple modification of the filtering part, by introducing a scalar accelerator parameter. The second algorithm is based on the use of a block filtering technique to actualize the local filter coefficients. The third algorithm is a modification of the second algorithm by the use of the final filtering errors to actualize the filters coefficients. The fourth and the fifth algorithms are based respectively on the combination of features from the first algorithm with features of the second and third algorithms. These five new algorithms were proposed to improve the convergence speed of the original version of the FNTF algorithm for the identification of acoustic impulse responses, and to also improve their tracking ability when these systems vary in time. The different algorithms and their comparative results are presented.

Crosstalk originating from multiple high-speed data services in the same telephone bundle is a limiting factor for the maximum bit rate, the loop length and the number of data services that a bundle can support. Due to the nature of twisted-pairs, external interferences (including crosstalk) mostly couple to the twisted-pair line in common mode, and then leaks to differential mode due to line imperfect balance. The result is a degradation of the received differential signal quality. This paper uses the common mode signal as a reference to an adaptive wideband crosstalk canceller, as an attempt to remove the effect of crosstalk on the differential signal. Simulation results show the potential benefits of using this technique to reduce the crosstalk levels.

A new method for speech enhancement in colored noise is proposed. A Kalman filter concatenated with a post-filter based on the masking properties of the human auditory system is proposed for the problem. A recursive approach to compute the noise covariance matrix is used for estimating the colored noise statistics. In the post-filter, both time domain masking properties and frequency domain masking properties are taken into account. From the calculated masking level, the noisy speech spectrum is adjusted accordingly. Simulation results show that the proposed approach has the best performance compared with other recent methods, evaluated with PESQ scores.

In the field of adaptive signal processing, it is well known that affine projection algorithms or their low-computational implementations fast affine projection algorithms can produce a good tradeoff between convergence speed and computational complexity. Although these algorithms typically do not provide the same convergence speed as recursive-least-squares algorithms, they can provide a much improved convergence speed compared to stochastic gradient descent algorithms, without the high increase of the computational load or the instability often found in recursive-least-squares algorithms. In this paper, multichannel affine and fast affine projection algorithms are introduced for active noise control or acoustic equalization. Multichannel fast affine projection algorithms have been previously published for acoustic echo cancellation, but the problem of active noise control or acoustic equalization is a very different one, leading to different structures, as explained in the paper. The computational complexity of the new algorithms is evaluated, and it is shown through simulations that not only can the new algorithms provide the expected tradeoff between convergence performance and computational complexity, they can also provide the best convergence performance (even over recursive-least-squares algorithms) when nonideal noisy acoustic plant models are used in the adaptive systems.

In the field of adaptive signal processing, it is well known that fast affine projection algorithms can produce a good tradeoff between convergence speed and computational complexity. Although these algorithms typically do not provide the same convergence speed as recursive-least-squares algorithms, they can provide a much improved convergence speed compared to stochastic gradient descent algorithms, without the high increase of the computational load or the instability often found in recursive-least-squares algorithms. In this paper, a multichannel fast affine projection algorithm is introduced for active noise control. The computational complexity of the new algorithm is evaluated, and it is shown through simulations that not only can the new algorithm provide the expected tradeoff between convergence performance and computational complexity, it can also provide the best convergence performance (even over recursive-least-squares algorithms) when non-ideal noisy acoustic plant models are used in the adaptive systems.

Some recursive least-squares algorithms for multichannel active noise control have recently been introduced, including computationally efficient (i.e. "fast") versions. However, these previously published algorithms suffer from numerical instability due to finite precision computations. Numerically robust recursive least-squares algorithms for multichannel active noise control systems are introduced, using QR decompositions and lattice structures. It is shown through simulations of broadband multichannel active noise control that the recursive least-squares algorithms introduced are indeed more numerically robust than the previously published algorithms, while keeping the same convergence behavior, and therefore they are more suitable for practical implementations.

In this paper, a simple inverse structure for multichannel active noise control (ANC) is introduced, based on the use of noncausal models of the (pseudo-)inverse plant between the error sensors (typically microphones in ANC systems) and the actuators (typically loudspeakers). This structure, combined with adaptive FIR filters and filtered-x LMS-based algorithms, can produce both a reduction of the computational load and an increase of the convergence speed, compared to the multichannel filtered-x LMS algorithm or its fast exact realizations. Also, for systems combining active sound cancellation and exact sound reproduction (such as headphones combining both ANC and binaural audio with equalization), the proposed inverse structure can further reduce the computational load compared to common approaches. Simulations using exact and noisy models of a realistic acoustic plant and an inverse plant were performed to verify the performance of the proposed structure.

In recent years, a few articles describing the use of neural networks for nonlinear active control of sound and vibration were published. Using a control structure with two multilayer feedforward neural networks (one as a nonlinear controller and one as a nonlinear plant model), steepest descent algorithms based on two distinct gradient approaches were introduced for the training of the controller network. The two gradient approaches were sometimes called the filtered-x approach and the adjoint approach. Some recursive-least-squares algorithms were also introduced, using the adjoint approach. In this paper, an heuristic procedure is introduced for the development of recursive-least-squares algorithms based on the filtered-x and the adjoint gradient approaches. This leads to the development of new recursive-least-squares algorithms for the training of the controller neural network in the two networks structure. These new algorithms produce a better convergence performance than previously published algorithms. Differences in the performance of algorithms using the filtered-x and the adjoint gradient approaches are discussed in the paper. The computational load of the algorithms discussed in the paper is evaluated for multichannel systems of nonlinear active control. Simulation results are presented to compare the convergence performance of the algorithms, showing the convergence gain provided by the new algorithms.

In this paper, a simple inverse structure for multichannel active noise control (ANC) is introduced. This structure, combined with adaptive FIR filters and filtered-x LMS-based algorithms, can produce both a reduction of the computational load and an increase of the convergence speed, compared to the standard multichannel filtered-x LMS algorithm or its fast exact realizations. The proposed structure is based on the use of delayed non-causal models of the inverse plant between some error sensors (typically microphones in ANC systems) and some actuators (typically loudspeakers). Simulations using models of a realistic acoustic plant and inverse plant were performed, and the convergence gain that can be achieved with the proposed structure was demonstrated, while reducing the computational load of the control algorithm.

There has been much research on active noise control (ANC) systems and transaural sound reproduction (TSR) systems. In those fields, multichannel FIR adaptive filters are extensively used. For the learning of FIR adaptive filters, recursive-least-squares (RLS) algorithms are known to produce a faster convergence speed than stochastic gradient descent techniques, such as the basic least-mean-squares (LMS) algorithm or even the fast convergence Newton-LMS, the gradient-adaptive-lattice (GAL) LMS and the discrete-cosine-transform (DCT) LMS algorithms. In this paper, multichannel RLS algorithms and multichannel fast-transversal-filter (FTF) algorithms are introduced, with the structures of some stochastic gradient descent algorithms used in ANC: the filtered-x LMS, the modified filtered-x LMS and the adjoint-LMS. The new algorithms can be used in ANC systems or for the deconvolution of sounds in TSR systems. Simulation results comparing the convergence speed, the numerical stability and the performance using noisy plant models for the different multichannel algorithms are presented, showing the large gain of convergence speed that can be achieved by using some of the introduced algorithms.

Active control of sound and vibration has been the subject of a lot of research in recent years, and examples of applications are now numerous. However, few practical implementations of nonlinear active controllers have been realized. Nonlinear active controllers may be required in cases where the actuators used in active control systems exhibit nonlinear characteristics, or in cases when the structure to be controlled exhibits a nonlinear behavior. A multilayer perceptron neural-network based control structure was previously introduced as a nonlinear active controller, with a training algorithm based on an extended backpropagation scheme. This paper introduces new heuristical training algorithms for the same neural-network control structure. The objective is to develop new algorithms with faster convergence speed (by using nonlinear recursive-least-squares algorithms) and/or lower computational loads (by using an alternative approach to compute the instantaneous gradient of the cost function). Experimental results of active sound control using a nonlinear actuator with linear and nonlinear controllers are presented. The results show that some of the new algorithms can greatly improve the learning rate of the neural-network control structure, and that for the considered experimental setup a neural-network controller can outperform linear controllers.