Polyphonic sound event localization and detection (SELD) has many practical applications in acoustic sensing and monitoring. However, the development of real-time SELD has been limited by the demanding computational requirement of most recent SELD systems. In this work, we introduce SALSA-Lite, a fast and effective feature for polyphonic SELD using microphone array inputs. SALSA-Lite is a lightweight variation of a previously proposed SALSA feature for polyphonic SELD. SALSA, which stands for Spatial Cue-Augmented Log-Spectrogram, consists of multichannel log-spectrograms stacked channel-wise with the normalized principal eigenvectors of the spectrotemporally corresponding spatial covariance matrices. In contrast to SALSA, which uses eigenvector-based spatial features, SALSA-Lite uses normalized inter-channel phase differences as spatial features, allowing a 30-fold speedup compared to the original SALSA feature. Experimental results on the TAU-NIGENS Spatial Sound Events 2021 dataset showed that the SALSA-Lite feature achieved competitive performance compared to the full SALSA feature, and significantly outperformed the traditional feature set of multichannel log-mel spectrograms with generalized cross-correlation spectra. Specifically, using SALSA-Lite features increased localization-dependent F1 score and class-dependent localization recall by 15% and 5%, respectively, compared to using multichannel log-mel spectrograms with generalized cross-correlation spectra.

Sound event localization and detection (SELD) consists of two subtasks, which are sound event detection and direction-of-arrival estimation. While sound event detection mainly relies on time-frequency patterns to distinguish different sound classes, direction-of-arrival estimation uses amplitude and/or phase differences between microphones to estimate source directions. As a result, it is often difficult to jointly optimize these two subtasks. We propose a novel feature calledSpatial cue-Augmented Log-SpectrogrAm(SALSA) with exact time-frequency mapping between the signal power and the source directional cues, which is crucial for resolving overlapping sound sources. The SALSA feature consists of multichannel log-spectrograms stacked along with the normalized principal eigenvector of the spatial covariance matrix at each corresponding time-frequency bin. Depending on the microphone array format, the principal eigenvector can be normalized differently to extract amplitude and/or phase differences between the microphones. As a result, SALSA features are applicable for different microphone array formats such as first-order ambisonics (FOA) and multichannel microphone array (MIC). Experimental results on the TAU-NIGENS Spatial Sound Events 2021 dataset with directional interferences showed that SALSA features outperformed other state-of-the-art features. Specifically, the use of SALSA features in the FOA format increased the F1 score and localization recall by$6 \,\%$each, compared to the multichannel log-mel spectrograms with intensity vectors. For the MIC format, using SALSA features increased F1 score and localization recall by$16 \,\%$and$7 \,\%$, respectively, compared to using multichannel log-mel spectrograms with generalized cross-correlation spectra.

Polyphonic sound event detection and localization (SELD) task is challenging because it is difficult to jointly optimize sound event detection (SED) and direction-of-arrival (DOA) estimation in the same network. We propose a general network architecture for SELD in which the SELD network comprises sub-networks that are pre-trained to solve SED and DOA estimation independently, and a recurrent layer that combines the SED and DOA estimation outputs into SELD outputs. The recurrent layer does the alignment between the sound classes and DOAs of sound events while being unaware of how these outputs are produced by the upstream SED and DOA estimation algorithms. This simple network architecture is compatible with different existing SED and DOA estimation algorithms. It is highly practical since the sub-networks can be improved independently. The experimental results using the DCASE 2020 SELD dataset show that the performances of our proposed network architecture using different SED and DOA estimation algorithms and different audio formats are competitive with other state-of-the-art SELD algorithms. The source code for the proposed SELD network architecture is available at Github1.

Polyphonic sound event detection and direction-of-arrival estimation require different input features from audio signals. While sound event detection mainly relies on time-frequency patterns, direction-of-arrival estimation relies on magnitude or phase differences between microphones. Previous approaches use the same input features for sound event detection and direction-of-arrival estimation, and train the two tasks jointly or in a two-stage transfer-learning manner. We propose a two-step approach that decouples the learning of the sound event detection and directional-of-arrival estimation systems. In the first step, we detect the sound events and estimate the directions-of-arrival separately to optimize the performance of each system. In the second step, we train a deep neural network to match the two output sequences of the event detector and the direction-of-arrival estimator. This modular and hierarchical approach allows the flexibility in the system design, and increase the performance of the whole sound event localization and detection system. The experimental results using the DCASE 2019 sound event localization and detection dataset show an improved performance compared to the previous state-of-the-art solutions.

Many signal processing-based methods for sound source direction-of-arrival estimation produce a spatial pseudo-spectrum of which the local maxima strongly indicate the source directions. Due to different levels of noise, reverberation and different number of overlapping sources, the spatial pseudo-spectra are noisy even after smoothing. In addition, the number of sources is often unknown. As a result, selecting the peaks from these spectra is susceptible to error. Convolutional neural network has been successfully applied to many image processing problems in general and direction-of-arrival estimation in particular. In addition, deep learning-based methods for direction-of-arrival estimation show good generalization to different environments. We propose to use a 2D convolutional neural network with multi-task learning to robustly estimate the number of sources and the directions-of-arrival from short-time spatial pseudo-spectra, which have useful directional information from audio input signals. This approach reduces the tendency of the neural network to learn unwanted association between sound classes and directional information, and helps the network generalize to unseen sound classes. The simulation and experimental results show that the proposed methods outperform other directional-of-arrival estimation methods in different levels of noise and reverberation, and different number of sources.

When distances between microphone pairs are larger than the half-wavelength of signals, source localization methods using cross-correlation such as time-difference-of-arrival (TDOA), steered response power (SRP) are commonly used in practice. We present here a novel model that expresses microphone pairwise cross-correlations as a sum of autocorrelations of source signals shifted by the relative delays of the signals arriving at the microphone pairs, and weighted by the source power and the distances between the sources and the microphone pairs. The model is formulated as a linear inverse problem and is sparse with respect to the source power map. The source power map, which directly shows the locations of all the sound sources, can be reconstructed using ℓ1-norm minimization algorithms. We demonstrate the effectiveness of our model in a wildlife monitoring application, where the goal is to locate multiple frogs in a dense chorus.

Acoustic beamforming has played a key role in the robust automatic speech recognition (ASR) applications. Accurate estimates of the speech and noise spatial covariance matrices (SCM) are crucial for successfully applying the minimum variance distortionless response (MVDR) beamforming. Reliable estimation of time-frequency (TF) masks can improve the estimation of the SCMs and significantly improve the performance of the MVDR beamforming in ASR tasks. In this paper, we focus on the TF mask estimation using recurrent neural networks (RNN). Specifically, our methods include training the RNN to estimate the speech and noise masks independently, training the RNN to minimize the ASR cost function directly, and performing multiple passes to iteratively improve the mask estimation. The proposed methods are evaluated individually and overally on the CHiME-4 challenge. The results show that the proposed methods improve the ASR performance individually and also work complementarily. The overall performance achieves a word error rate of 8.9% with 6-microphone configuration, which is much better than 12.0% achieved with the state-of-the-art MVDR implementation.

This paper presents an eigenvector clustering approach for estimating the direction of arrival (DOA) of multiple speech signals using a microphone array. Existing clustering approaches usually only use low frequencies to avoid spatial aliasing. In this study, we propose a probabilistic eigenvector clustering approach to use all frequencies. In our work, time-frequency (TF) bins dominated by only one source are first detected using a combination of noise-floor tracking, onset detection and coherence test. For each selected TF bin, the largest eigenvector of its spatial covariance matrix is extracted for clustering. A mixture density model is introduced to model the distribution of the eigenvectors, where each component distribution corresponds to one source and is parameterized by the source DOA. To use eigenvectors of all frequencies, the steering vectors of all frequencies of the sources are used in the distribution function. The DOAs of the sources can be estimated by maximizing the likelihood of the eigenvectors using an expectation-maximization (EM) algorithm. Simulation and experimental results show that the proposed approach significantly improves the root-mean-square error (RMSE) for DOA estimation of multiple speech sources compared to the MUSIC algorithm implemented on the single-source dominated TF bins and our previous clustering approach.

This paper presents a generalized multiple-point sparse constrained deconvolution approach for mapping acoustic noise sources in large regions using a movable array. Extended from our previous MPSC-DAMAS approach, we first derive a generalized inverse problem relating to the source powers and the array manifold using a generic beamformer and an explicit measurement noise model. We then propose a generalized MPSC-DAMAS (GMPSC-DAMAS) approach for resolving the inverse problem. A new parameter setting method based on a multiple-point minimum-variance-distortionless-response (MVDR) beamformer is also presented. The realizations of the GMPSC-DAMAS approach using the delay- and-sum (DAS) beamformer and the MVDR beamformer are evaluated. Simulation results show the proposed GMPSC-DAMAS approach achieves much lower absolute power estimation errors and processing time than the MPSC-DAMAS approach in terms of number of sources and robustness to measurement noise.

This paper presents a robust speech recognition system using a microphone array for the 3rd CHiME Challenge. A minimum variance distortionless response (MVDR) beamformer with adaptive microphone gains is proposed for robust beamforming. Two microphone gain estimation methods are studied using the speech-dominant time-frequency bins. A multichannel noise reduction (MCNR) postprocessing is also proposed to further reduce the interference in the MVDR processed signal. Experimental results for the ChiME-3 challenge show that both the proposed MVDR beamformer with microphone gains and the MCNR postprocessing improve the speech recognition performance significantly. With the state-of-the-art deep neural network (DNN) based acoustic model, our system achieves a word error rate (WER) of 11.67% on the real test data of the evaluation set.

This paper presents a learning-based approach to the task of direction of arrival estimation (DOA) from microphone array input. Traditional signal processing methods such as the classic least square (LS) method rely on strong assumptions on signal models and accurate estimations of time delay of arrival (TDOA) . They only work well in relatively clean conditions, but suffer from noise and reverberation distortions. In this paper, we propose a learning-based approach that can learn from a large amount of simulated noisy and reverberant microphone array inputs for robust DOA estimation. Specifically, we extract features from the generalised cross correlation (GCC) vectors and use a multilayer perceptron neural network to learn the nonlinear mapping from such features to the DOA. One advantage of the learning based method is that as more and more training data becomes available, the DOA estimation will become more and more accurate. Experimental results on simulated data show that the proposed learning based method produces much better results than the state-of-the-art LS method. The testing results on real data recorded in meeting rooms show improved root-mean-square error (RMSE) compared to the LS method.

Mapping environmental noise with high resolution on a large scale (such as a city) is prohibitively expensive with current approaches, which use a large, dense array spanning the entire region of interest, or sequential noise measurements at thousands of locations on a dense grid. We propose instead a new acoustic measurement scheme using a small movable array (for example, mounted on a vehicle driving along the streets of a city) to rapidly acquire measurements at many different locations. A multiple-point sparse constrained deconvolution approach for the mapping of acoustic sources (MPSC-DAMAS) and a multiple-point covariance matrix fitting (MP-CMF) approach are developed to accurately estimate the locations and powers of stationary noise sources across the region of interest. Computer simulations of large region acoustic mapping demonstrate that superior resolution and much lower power estimation errors are achieved by the proposed approaches compared to the state-of-the-art SC-DAMAS approach and CMF approach.

Modern-day train operations rely on wireless communications. Unlike other mobile systems, the train vehicle operations are tightly interwound with and remain physically close to the railway and the trackside infrastructure, providing a suitable platform to deploy leaky-waveguide-based communication. Due to the train system's safety-critical application and its exposure to the public, it is critical to address security in train communications. To investigate the availability of leaky waveguide communications, we first study prior leaky waveguide implementations in train systems and, based on those studies, construct a model to characterize the path loss of inside-waveguide propagation and the repeater implementations. Using our model, we analyze the jamming impact and contrast with jamming in free space without a waveguide. As a result, we establish that jamming the waveguide takes advantage of the waveguide infrastructure to extend its impact beyond the traditional jamming range and breaks the spatial dependence on the jamming source.

In the nervous system, sensory neurons encode signals as a sequence of action potentials (spikes). However, spike generation is metabolically expensive. Achieving high coding fidelity may require a high spike rate. Here we propose that neurons achieve a trade-off by optimally timing spikes so that maximum fidelity is achieved for a given spike rate. The proposed neural encoder generates spikes which are reconstructed by a linear filter, with energy modeled as a constraint proportional to the average spike-rate. We develop expressions for the encoding error and derive the optimal parameters in the limit of high spike-firing rates. The energy-constrained neural encoder is compared with experimental spike-times from two sensory neurons, one cortical and one peripheral. The proposed energy-constrained neural encoder closely approximates the experimentally recorded spike-times, and the decoded experimental inputs are within 2dB of the predicted distortion-energy curve for both neurons.

In this brief, we present a novel architecture for pseudorandom (PN) code acquisition based on statistical error compensation (SEC), which achieves significant power savings. SEC treats errors in hardware as noise in communication networks, and employs robust estimation theory to compensate for errors. We apply SEC to a 256-tap PN code acquisition filter in a 180-nm CMOS process. Multiple (five) dies were tested under voltage overscaling to achieve a near constant detection probability (Pdet) above 90%. The minimum energy consumption ranged from 72.89 to 210.59 pJ (ave 122.52 pJ) for supply voltages between 0.69 and 0.70 V. These operating conditions result in raw error rates of 85.83%-91.23% (ave 88.99%). Energy savings over a conventional errorfree design ranges from 2.4× to 5.8× (ave 3.86×). Energy savings over past work ranges from 1.55× to 3.79× (ave 2.52×). Improvement in error-tolerance over existing error-tolerant designs range from 2146× to 2281× (ave 2225×). The large energy savings were found to be due to a combination of voltage scaling and activity factor reduction. The proposed design achieves a 2.5× improvement in the figure of merit [normalized power/(#taps * precision * sample rate)] compared with conventional PN code acquisition filters.

It is challenging to determine the directions of arrival of speech signals when there are fewer sensors than sources, particularly in noisy and reverberant environments. The coherence test by Mohan et al. exploits the time-frequency sparseness of non-stationary speech signals to select more relevant time-frequency bins to estimate directions of arrival. With no prior knowledge about the incoming sources, this work proposes a combination of noise-floor tracking, onset detection and a coherence test to robustly identify time-frequency bins where only one source is dominant. After that, the largest eigenvectors of covariance matrices corresponding to these bins are clustered and the directions of arrival of the sources are estimated based on the cluster centroids. Simulation and experimental results show that this method is able to localize 8 sources with small errors using only 3 omnidirectional microphones. The proposed method is robust to background noise and reverberation.

Current sensory array systems do not fully exploit tactile sensing strategies widely used by vibrissal sensing animals to explore their surroundings. We develop a new tactile fluid-flow imaging technique, which relates rat's whisker movements to tomographic imaging to extract fluid-flow characteristics with a robotic whisker array. At high Reynolds numbers, the drag force on a whisker segment is proportional to the relative velocity squared, and acts as a distributed load along the whisker length. Therefore, we propose that it is possible to map out the 2-D cross-sectional mean fluid-flow velocity field with a robotic whisker array by measuring the moment sensed at each whisker base from different directions for tomographic reconstruction. The associated inverse problem for the tomographic image formation is formulated as a MAP estimation problem and solved computationally. The experimental results demonstrate that this new approach offers a fundamentally novel sensor technology for flow-field measurements.

In this paper, we propose a new auxiliary-vector (AV) algorithm using the conjugate orthogonality for speech enhancement. When only a limited data record is available, the AV algorithm is the state-of-the-art for obtaining the minimumvariance-distortionless (MVDR) filter. However, the current AV algorithms suffer from convergence problems when applied to the speech enhancement. Based on the conjugate GramSchmidt process, we develop new auxiliary vectors that are conjugate orthogonal and apply them to the AV algorithm. The proposed conjugate AV algorithm converges to the optimal MVDR solution within finite steps no greater than the filter dimension. Theoretical analysis establishes formal convergence of the proposed conjugate AV algorithm. Our experiments using the synthetic and real speech data show favorites of the new proposal over the state-of-the-art approaches. Index Terms: speech enhancement, microphone arrays, correlation, convergence, adaptive signal processing

Simultaneous detection and estimation is important in many engineering applications. In particular, there are many applications where it is important to perform signal detection and Signal-to-Noise-Ratio (SNR) estimation jointly. Application of existing frameworks in the literature that handle simultaneous detection and estimation is not straightforward for this class of application. This paper therefore aims at bridging the gap between an existing framework, specifically the work by Middleton et al., and the mentioned application class by presenting a jointly optimal detector and signal and noise power estimators. The detector and estimators are given for the Gaussian observation model with appropriate conjugate priors on the signal and noise power. Simulation results affirm the superior performance of the optimal solution compared to the separate detection and estimation approaches.

Many applications involve sparse signals with unknown, continuous parameters; a common example is a signal consisting of several sinusoids of unknown frequency. Applying compressed sensing techniques to these signals requires a highly oversampled dictionary for good approximation, but these dictionaries violate the RIP conditions and produce inconsistent results. We consider recovering both a sparse vector and parameter perturbations from an initial set of parameters. Joint recovery allows for accurate reconstructions without highly oversampled dictionaries. Our algorithm for sparse recovery solves a series of linearized subproblems. Recovery error for noiseless simulated measurements is near zero for coarse dictionaries, but increases with the oversampling. This technique is also used to reconstruct Radio Frequency data. The algorithm estimates sharp peaks and transmitter frequencies, demonstrating the potential practical use of the algorithm on real data.

Sensing systems with multiple sensors and operating modes warrant active management techniques to balance estimation quality and measurement costs. Existing literature shows that in the joint sensor-scheduling and state-estimation problem for HMMs, estimator optimization can be done independently of the scheduler at each time step. We investigate the special case when a MAP estimator is used, and show how the joint problem can be converted to a standard Partially Observable MarkovDecision Process (POMDP), which in turn enables us to use POMDP solvers. As this approach is highly redundant, we derive a direct solution, which exploits the separability property while still utilizing standard solvers. When compared to standard techniques, the direct algorithm provides savings by a factor of the state-space dimension. Numerical results are given for an example motivated by wildlife monitoring.

In many detection applications with battery powered or energy-harvesting sensors, energy constraints preclude the use of the optimal detector all the time. Optimal energy-performance trade-off is therefore needed in such situations. In many applications, the signal and noise power may vary greatly over time, and this can be exploited to constrain energy consumption while maintaining the best possible performance. A detector scheduling algorithm based on the signal and noise power information is developed in this work. The resulting algorithm is simple due to its threshold-test structure and can be easily implemented with almost no overhead. A detection system with two detectors using the proposed scheduling scheme is estimated to greatly reduce the energy consumption for a wildlife monitoring application.

This paper presents a Spatialized Audio Multiparty Teleconferencing (SAMT) system with a radically new communication experience for group teleconferencing. The system includes our recently developed 3D audio technologies: 3D sound source localization (SSL) and 3D audio capture and reproduction using a low-cost and compact design microphone array. In essence, the SAMT system offers 3D audio capture capability and spatial audio perception with multiple participants at a site, which still falls short in teleconferencing solutions. In addition to being able to identify and automatically track the active speaker, the system allows more compelling visual presentation for effective communication. Requiring only a low-cost microphone array and a consumer depth camera, the proposed system runs reliably and comfortably in real time on a commodity laptop or desktop PC. With such a minimal deployment requirement, we present a variety of user experiences created by SAMT.

Real-time 3D sound localization is an important technology for various applications such as camera steering systems, robotics audition, and gunshot direction. 3D sound localization adds a new dimension, but also significantly increases the computational requirements. Real-time 3D sound localization continuously processes large volumes of data for each possible 3D direction and acoustic frequency range. Such highly demanding compute requirements outpace current CPU compute abilities. This paper develops a real-time implementation of 3D sound localization on Graphical Processing Units (GPUs). Massively parallel GPU architectures are shown to be well suited for 3D sound localization. We optimize various aspects of GPU implementation, such as number of threads per thread block, register allocation per thread, and memory data layout for performance improvement. Experiments indicate that our GPU implementation achieves 501X and 130X speedup compared to a single-thread and a multi-thread CPU implementation respectively, thus enabling real-time operation of 3D sound localization.

Applications such as long-term environmental monitoring and large-scale surveillance demand reliable performance from sensor nodes while operating within strict energy constraints. There is often not enough power for sensors to make measurements all of the time. In these cases, one must decide when to run each sensor. To this end, we develop a one-step optimal sensor-scheduling algorithm based on expected-utility maximization. “Utility” is an application-specific measure of the benefit from a given sensor measurement. In sensing environments that can be modeled using a hidden Markov model, selecting the appropriate combination of sensors at each time instant enables maximization of the expected utility while operating within an energy budget. For some budgets, the utility-based algorithm shows more than 300% utility gains over a constant duty-cycle scheme designed to consume the same amount of energy. These benefits are dependent on the energy budget.

Real-time detection of intermittent events requires continual monitoring and processing of sensor data. A battery-powered device that supports multiple sensing modalities and processing algorithms has the potential to save energy by using expensive sensors and algorithms only when the event of interest is most likely to occur. To develop a policy for sensing and processing management, we adopt maximum sequential information gain as an objective criterion for such energy-limited systems, which can be solved via dynamic programming. For binary hypothesis testing with two sensing options, the optimal management policy is a simple two-threshold test on the posterior belief. Detection of bird presence/absence in a wildlife monitoring application shows up to a 37% reduction in error rate over standard constant-duty-cycle sensing.

Existing techniques utilizing bio-inspired robotic whisker sensory systems generally address object feature extraction with artificial whiskers as a mechanical problem. We present an alternative signal-processing approach that formulates the object shape recognition as a 2-D tactile imaging problem. Observing that the whisker position at the very initial contact is similar to a `ray path' in X-ray computed tomography; the 2-D cross-sectional image of the object can be tomographically reconstructed by measuring only the angle at the whisker base and the whisker-base location at the initial contact for each viewing angle. This approach has the important practical advantage of eliminating the need for the calibration of the force/moment measurements. Experimental results demonstrate the promising potential for bio-inspired systems using arrays of vibrissal sensors for object shape recognition.

In this paper, we present a fast-converging adaptive frequency-domain minimum-variance-distortionlessresponse (MVDR) beamformer (FMV) for speech enhancement. The well-known FMV solution is optimum in the microphone array processing. However, the direct computation of the optimum FMV solution is often undesirable due to the the inversion of the spatio-spectral correlation matrix which is often unstable and is expensive for large arrays. To avoid the matrix inversion, we develop a fast-converging conjugate gradient (CG) algorithm for iteratively computing the FMV solution. Compared to the existing steepest descent (SD) algorithm, the CG algorithm can dramatically improve the convergence speed for the case of multiple interfering signals in speech enhancement. Therefore, the computational load and processing time can be significantly reduced. The speech enhancement experiments using a four-channel acousticvector-sensor (AVS) microphone array are demonstrated for the target speech signal corrupted by two and five interfering speech signals and superior performance are achieved.

Stochastic computation, as presented in this paper, exploits the statistical nature of application-level performance metrics, and matches it to the statistical attributes of the underlying device and circuit fabrics. Nanoscale circuit fabrics are viewed as noisy communication channels/networks. Communications-inspired design techniques based on estimation and detection theory are proposed. Stochastic computation advocates an explicit characterization and exploitation of error statistics at the architectural and system levels. This paper traces the roots of stochastic computing from the Von Neumann era into its current form. Design and CAD challenges are described.

Future microprocessors increasingly rely on an unreliable CMOS fabric due to aggressive scaling of voltage and frequency, and shrinking design margins. Fortunately, many emerging applications can tolerate computational errors caused by hardware unreliabilities, at least during certain execution intervals. In this paper, we propose scalable stochastic processors, a computing platform for error-tolerant applications that is able to scale gracefully according to performance demands and power constraints while producing outputs that are, in the worst case, stochastically correct. Scalability is achieved by exposing to the application layer multiple functional units that differ in their architecture but share functionality. A mobile video encoding application here is able to achieve the lowest power consumption at any bitrate demand by dynamically switching between functional-unit architectures.

The main drawback of Orthogonal Frequency Division Multiplexing (OFDM) systems is the high peak-to-average power ratio (PAPR), which leads to a significant reduction in performance and power efficiency. Tone injection (TI) is a promising PAPR reduction technique that cyclically extends QAM constellations to allow an alternative encoding with lower PAPR at the transmitter. We present a new efficient complex-baseband algorithm which performs TI by using aggressive clipping to combat large peaks. An approximated-analog PAPR reduction of up to 5.1 dB at a 10-5 symbol-clip probability is obtained for 64-channel 16-QAM OFDM. This is a very fast and practical peak-power reduction method for OFDM systems that essentially achieves the same PAPR as single-carrier modulation for large constellations.

In this paper, the stochastic networked computation (SNC) paradigm for designing robust and energy-efficient systems-on-a-chip in nanoscale process technologies, where robust computation is treated as a statistical estimation problem is presented. The benefits of SNC are demonstrated by employing it to design an energy-efficient and robust pseudonoise-code acquisition system for the wireless CDMA2000 standard (http://www.3gpp2.org). Simulations in IBM's 130-nm CMOS process show that the SNC-based architecture enhances the average probability of detection (PDet) in the presence of process variations by two to three orders of magnitude, reduces power by 31%-39%, and reduces the variation in PDetby one to two orders of magnitude at a typical false-alarm rate of 5% over a conventional architecture. SNC performance in the presence of voltage overscaling and across technology nodes (90, 65, 45, and 32 nm) is also studied.

Weakly electric fish have the ability to navigate and locate prey in the dark using a unique weak electrosense system. Imitating that ability of electric fish, we develop an electric-field sensing system capable of high-resolution imaging of the surrounding environment, by use of a novel beamforming technique that exploits the sparsity of sources in a scanned space. Both simulation and experimental results show that the sparse beamforming technique accurately images not only a single object but also multiple objects with highly correlated signals.

Two-dimensional ultrasound (US) is widely used in minimally invasive cardiac procedures due to its convenience of use and noninvasive nature. However, the low quality of US images often limits their utility as a means for guiding procedures, since it is often difficult to relate the images to their anatomical context. To improve the interpretability of the US images while maintaining US as a flexible anatomical and functional real-time imaging modality, we describe a multimodality image navigation system that integrates 2D US images with their 3D context by registering them to high quality preoperative models based on magnetic resonance imaging (MRI) or computed tomography (CT) images. The mapping from such a model to the patient is completed using spatial and temporal registrations. Spatial registration is performed by a two-step rapid registration method that first approximately aligns the two images as a starting point to an automatic registration procedure. Temporal alignment is performed with the aid of electrocardiograph (ECG) signals and a latency compensation method. Registration accuracy is measured by calculating the TRE. Results show that the error between the US and preoperative images of a beating heart phantom is 1.7 +/-0.4 mm, with a similar performance being observed in in vivo animal experiments.

Electrophysiological cardiac data mapping is an essential tool for the study of cardiac rhythm disorders, such as atrial fibrillation. Over the past decade, various advanced cardiac mapping systems have been developed to create detailed cardiac maps and assist physicians in diagnosis and therapy guidance. While these systems have increased the ability to study and treat cardiac arrhythmias, inherent limitations exist. The objective of this paper is to describe and evaluate a system that extends current approaches to cardiac mapping, to create a dynamic cardiac map, using patient-specific cardiac models. This paper details novel approaches to collecting a stream of electrophysiological cardiac data, registering the data with patient-specific dynamic cardiac models, and displaying the data directly on the dynamic model surface, giving a more accurate and comprehensive visualization environment when compared to current systems. To validate the system, a series of laboratory and in vivo experiments were conducted. In the laboratory studies, the system was used to test the user's ability to accurately locate a landmark in physical space, as well as their ability to accurately navigate to a virtual location. In the in vivo studies the overall system performance was compared to an existing electrophysiological recording system, where right atrial cardiac maps were created during sinus and paced cardiac rhythms. The results showed that the new dynamic cardiac mapping system was able to maintain high accuracy in locating physical and virtual landmarks, while being able to create a dynamic cardiac map displayed on a dynamic cardiac surface model.

The stochastic sensor network-on-chip (SSNOC) was recently proposed as an effective computational paradigm for jointly achieving energy-efficiency and robustness in nanoscale processes. In this paper, we study the trends in energy-efficiency and robustness exhibited by an SSNOC architecture as the feature size scales from 130nm to 32nm for a PN-code acquisition application. The conventional architecture exhibits a 3 orders-of-magnitude loss in detection probability P_{det} due to process variations in the 130nm and smaller technology nodes. At the 130nm and 90nm nodes, the proposed SSNOC architecture recovers from this performance loss, and exhibits a 2 orders-of-magnitude smaller variation in P_det compared to the conventional architecture. However, for the 65nm and 45nm technology nodes, the SSNOC architecture with assistance from circuit level techniques such as adaptive body bias (ABB) and adaptive supply voltage (ASV) shows a 2-3 order-of-magnitude better detection performance. In addition, the SSNOC architecture with ABB/ASV achieves 22% to 31% energy savings. For the 32nm node, the current version of SSNOC with ABB/ASV is not robust enough and thus motivates the need to explore even more powerful versions of SSNOC.

Modern integrated circuits (ICs) are designed as massively parallel systems as a consequence of diminishing silicon feature sizes. This has adversely impacted reliability because of increased errors due to process and environmental variations, and particle hits. Viewing hardware errors as analogous to measurement or system noise allows us to borrow results from estimation theory and extend Moore's law. The estimation-theoretic framework provides a design optimization formalization that enables power/reliability trade-off in broad classes of applications. Two applications described here show that specific instantiations of the framework yield significant power savings and system reliability.

Presented in this paper is an energy-efficient and variation-tolerant PN-code acquisition architecture for the wireless CDMA2000 standard. The architectures is based on the recently proposed stochastic sensor network-on-chip (SSNOC) computational paradigm. The latter employs the principles of statistically similar decomposition and robust estimation theory to compensate for timing errors due to process variations. Performance of the SSNOC-based PN-code acquisition architecture at the slow process corner indicates that the average probability of detection PDetimproves by up to 3 orders-of-magnitude over that of the conventional architecture, while the variation in PDet(sigma / mu) is reduced by up to 2 orders-of-magnitude over that of the conventional architecture while simultaneously achieving a power reduction of 39%.

We address the issue of improving information availability in a class of delay-tolerant sensor network applications, where the sensor nodes are deployed in disconnected environments. In such environments, since there is no continuous access to a remote base station, there is a need to leverage the collaborative resources of the sensor network to support in-network storage. The stored data can then be retrieved opportunistically by mobile collectors in the proximity. We have developed a data-centric, in-network storage architecture that partitions the network into storage zones. In such a scheme, some of the storage nodes may be unavailable at the time of storage and retrieval, because they are either sleeping to conserve energy or because they have failed.We present two schemes based on random linear network coding for improving information availability within such a storage architecture. In the centralized scheme, the encoding is performed by the managers in each storage zone, whereas in the decentralized scheme, the encoding is done locally by the zone members. We have implemented the network coding schemes in TinyOS and we present results that show the impact of the zone size, duty cycle, and the degree of encoding on the decoding probability, based on our experiments on a testbed of Micaz motes.

Biologists have reported that fish form sensory maps to assist in the imaging of flow disturbances underwater, including those created by predator or prey, using the lateral-line flow-sensing system. Our group has invented artificial lateral-line systems based on both micromachined and commercial hot-wire anemometer sensors. In this paper, we report on the successful application of adaptive beamforming (Capon's method), in conjunction with an artificial lateral line, to visualize flow disturbances, including acoustic dipoles, underwater. Our preliminary results showcase, for the first time, an ability to image an entire region underwater using micro and millimeter scale arrays, thereby mimicking the flow-imaging abilities of fish.

A modern DSP laboratory course should teach students how to quickly develop efficient applications using a mixture of C and assembly instructions. Since typical irregular DSP microprocessor architectures present some challenges to traditional optimizing compilers, we introduced a novel new interactive optimizing compiler, VISTA, in a DSP laboratory course to teach students about optimizing compilers and the trade-offs between code development in C and assembly. Students generally found VISTA educational and gained improved insight into effective DSP software development.

A high peak-to-average-power ratio (PAPR) is a primary drawback of orthogonal frequency-division multiplexing (OFDM). Many methods have been proposed to reduce the OFDM PAPR for single-antenna and for MIMO systems. Some systems, however, use multiple antennas for frequency-selective transmit beamforming either to a single or multiple simultaneous receivers. This paper presents a new method for PAPR reduction that minimizes the peak power across all antennas while preserving the maximal ratio transmission (MRT) gain and zero-crosstalk criteria at the receivers. Because the gain and zero crosstalk is preserved at the receivers, no performance loss is seen by the receivers when using the PAPR-reduced transmit vector. With eight transmit antennas, the algorithm demonstrates an average of 5.5 dB PAPR reduction for single-user beamforming and about 4.5 dB PAPR reduction for four-user transmit spatial division multiple access (Tx-SDMA)

Optimal estimation of a two-dimensional (2-D) multichannel signal ideally decorrelates the data in both channel and space and weights the resulting coefficients according to their SNR. Many scenarios exist where the required second-order signal and noise statistics are not known in which the decorrelation is difficult or expensive to calculate. An asymptotically optimal estimation scheme proposed here uses a 2-D discrete wavelet transform to approximately decorrelate the signal in space and the discrete Fourier transform to decorrelate between channels. The coefficient weighting is replaced with a wavelet-domain thresholding operation to result in an efficient estimation scheme for both stationary and nonstationary signals. In contrast to optimal estimation, this new scheme does not require second-order signal statistics, making it well suited to many applications. In addition to providing vastly improved visual quality, the new estimator typically yields signal-to-noise ratio gains 12 dB or higher for hyperspectral imagery and functional magnetic resonance images.

Mobile devices primarily processing multimedia data need to support multimedia quality with limited battery energy. To address this challenging problem, researchers have introduced adaptation into multiple system layers, ranging from hardware to applications. Given these adaptive layers, a new challenge is how to coordinate them to fully exploit the adaptation benefits. This paper presents a novel cross-layer adaptation framework, called GRACE-1, that coordinates the adaptation of the CPU hardware, OS scheduling, and multimedia quality based on users' preferences. To balance the benefits and overhead of cross-layer adaptation, GRACE-1 takes a hierarchical approach: It globally adapts all three layers to large system changes, such as application entry or exit, and internally adapts individual layers to small changes in the processed multimedia data. We have implemented GRACE-1 on an HIP laptop with the adaptive Athlon CPU, Linux-based OS, and video codecs. Our experimental results show that, compared to schemes that adapt only some layers or adapt only to large changes, GRACE-1 reduces the laptop's energy consumption up to 31.4 percent while providing better or the same video quality.

High peak to average power ratio (PAR), a major drawback in single-input-single-output (SISO) systems using orthogonal frequency division multiplexing (OFDM), has already found many methods of reduction. The active channel extension (ACE) method reduces analog PAR by as much as 3.5 dB with no reduction in data rate by altering constellation points. We present an extension of this original method for PAR reduction for multi-input-multi-output (MIMO) systems in general. This new scheme, compatible with variable power allocation, minimizes PAR on each antenna in a similar manner as in the SISO system. It also supports any number of unique constellations and number of antennas by moving outer constellation points appropriately depending on the constellation for the particular channel and antenna. Simulations with 4 antennas at both transmitter and receiver show analog peak reductions of more than 3 dB overall for all 4 antennas.

There is significant interest in battery-powered sensor networks to be used for detection in a wide variety of applications, from surveillance and security to health and environmental monitoring. Severe energy and bandwidth constraints at each sensor node demand system-level approaches to design that consider detection performance jointly with system-resource constraints. Our approach is to formulate detection problems with constraints on the expected cost arising from transmission (sensor nodes to a fusion node) and measurement (at each sensor node) to address some of the system-level costs in a sensor network. For a given resource constraint, we find that randomization over the choice of measurement and over the choice of when to transmit achieves the best performance (in a Bayesian, Neyman-Pearson, and Ali-Silvey sense). To facilitate design, we describe performance criteria in the send/no-send transmission scenario, where the joint optimization over the sensor nodes decouples into optimization at each sensor node.

It has long been known that a fixed ordering of optimization phases will not produce the best code for every application. One approach for addressing this phase-ordering problem is to use an evolutionary algorithm to search for a specific sequence of phases for each module or function. While such searches have been shown to produce more efficient code, the approach can be extremely slow because the application is compiled and possibly executed to evaluate each sequence's effectiveness. Consequently, evolutionary or iterative compilation schemes have been promoted for compilation systems targeting embedded applications where meeting strict constraints on execution time, code size, and power consumption is paramount and longer compilation times may be tolerated in the final stage of development, when an application is compiled one last time and embedded in a product. Unfortunately, even for small embedded applications, the search process can take many hours or even days making the approach less attractive to developers. In this paper, we describe two complementary general approaches for achieving faster searches for effective optimization sequences when using a genetic algorithm. The first approach reduces the search time by avoiding unnecessary executions of the application when possible. Results indicate search time reductions of 62%, on average, often reducing searches from hours to minutes. The second approach modifies the search so fewer generations are required to achieve the same results. Measurements show this approach decreases the average number of required generations by 59%. These improvements have the potential for making evolutionary compilation a viable choice for tuning embedded applications.

It has long been known that a fixed ordering of optimization phases will not produce the best code for every application. One approach for addressing this phase ordering problem is to use an evolutionary algorithm to search for a specific sequence of phases for each module or function. While such searches have been shown to produce more efficient code, the approach can be extremely slow because the application is compiled and executed to evaluate each sequence's effectiveness. Consequently, evolutionary or iterative compilation schemes have been promoted for compilation systems targeting embedded applications where longer compilation times may be tolerated in the final stage of development. In this paper we describe two complementary general approaches for achieving faster searches for effective optimization sequences when using a genetic algorithm. The first approach reduces the search time by avoiding unnecessary executions of the application when possible. Results indicate search time reductions of 65% on average, often reducing searches from hours to minutes. The second approach modifies the search so fewer generations are required to achieve the same results. Measurements show that the average number of required generations decreased by 68%. These improvements have the potential for making evolutionary compilation a viable choice for tuning embedded applications.

A stochastic wireless sensor network consisting of low-duty-cycle, unsynchronized nodes offers a simple, robust, and ultra-low-power solution for many sensor network applications. We demonstrate the feasibility of this approach by proving that such a network exhibits quasi-stable behavior as long as the mean number of active nodes in a neighborhood exceeds a bound between one and two, even in the presence of collisions. As this number increases, the mean lifetime of messages in the network grows rapidly and the average response time decreases. Simulation results correspond closely to the analytical bounds for quasistability, lifetime, and response time. A stochastic sensor network appears to he a promising approach for many wireless sensor network applications with modest data throughput requirements.

The reverberation time (RT) measures the persistence of a sound in enclosed acoustic spaces. In a previous work, a method for the blind maximum-likelihood estimation (MLE) of RT using passively received microphone signals was presented. The procedure overcomes the drawbacks of current methods that use a controlled sound source for RT determination. Here, fast algorithms for online implementation of the method are developed. One algorithm, suitable for a one-time determination of the RT, requires O(N) computations for a data frame of length N. A second IIR algorithm, based on Q-levels of quantization, requires O(Q) computations. Results for speech data and choice of algorithms are discussed.

As medical ultrasound imaging moves to larger apertures and higher frequencies, tissue sound-speed variations continue to limit resolution. In geophysical imaging, a standard approach for estimating near-surface aberrating delays is to analyze the time shifts between common-midpoint signals. This requires complete data-echoes from every source/receiver pair in the array. Unfocused common-midpoint signals remain highly correlated in the presence of delay aberrations; there is also tremendous redundancy in the data. In medical ultrasound, this technique has been impaired by the wide-angle, random-scattering nature of tissue. This has made it difficult to estimate azimuth-dependent aberration profiles or to harness the full redundancy in the complete data. Prefiltering the data with two-dimensional fan filters mitigates these problems, permitting highly overdetermined, least-squares solutions for the aberration profiles at many steering angles. In experiments with a tissue-mimicking phantom target and silicone rubber aberrators at nonzero stand-off distances from a one-dimensional phased array, this overdetermined, fan-filtering algorithm significantly outperformed other phase-screen algorithms based on nearest-neighbor cross-correlation, speckle brightness maximization, and common-midpoint signal analysis. Our results imply that there is still progress to be made in imaging with single-valued focusing operators. It also appears that the signal-to-noise penalty for using complete data sets is partially compensated by the overdetermined nature of the problem.

Due to inherent factors such as a small and fragmented market and rapid hardware obsolescence, the conventional textbook is inadequate for DSP laboratory education. Freely available open-content materials that enable and promote both local customization and further development by a community of educators offers a fresh approach to lab text development that can surmount these barriers. We overview a joint effort under the aegis of the Connexions Project to develop a large pool of DSP lab modules sufficient to serve as the complete, stand-alone text for several types of DSP lab courses.

There has been a lot of interest recently in applying channel coding with side information (CCSI) concepts to data hiding. Part of the interest stems from the fact that information theoretic bounds can be derived for such systems. In this paper, we model the audio data embedding problem as a parallel CCSI problem. This is done by dividing the audio spectrum into frequency bins and then treating each bin as a separate CCSI channel. A perceptual mask is derived from the audio signal to determine the amount of power to use in each channel. It acts as a "water-filling" formula by determining the amount of distortion that can be introduced into each frequency bin without introducing audible distortion. As a result, our data embedding scheme will be imperceptible to the human ear. An exciting application for our audio data embedding solution is to embed data within the audio signal that will enable surround sound at the receiver. The resulting surround sound system will be better than existing surround sound systems that are based on manipulating two stereo channels to derive the other surround channels.

The high peak-to-average power ratio (PAR) in orthogonal frequency division multiplexing (OFDM) modulation systems can significantly limit performance. Methods exist which alter or introduce new signal constellations to combat large signal peaks. We present an improved PAR-reduction method that dynamically extends outer constellation points in active (data-carrying) channels, within margin-preserving constraints, in order to minimize the peak magnitude. This scheme simultaneously decreases the bit error rate slightly while substantially reducing the peak magnitude of an OFDM transmit block and requiring no side information. PAR reduction for an approximated analog signal is considered, and a 4.5 dB reduction is obtained for 256-channel QPSK OFDM. The results show great promise for use in commercial systems.

The paper develops nonnegative, high-resolution time-frequency representations (TFRs) that correspond with intuitive notions of energy distribution. These so-called consistent TFRs require the desired representation to be consistent with a set of spectrogram-based energy measurements. By formulating the desired representation as the solution to a constrained optimization problem, it can be solved using a gradient-projection technique. The consistent TFR demonstrates superior performance compared to existing techniques on a variety of test signals and biological data. The result is a high-resolution, nonnegative, intuitively-satisfying TFR that should prove to be an excellent tool for exploratory data analysis.

In this paper, we present a new wavelet-based estimator for 2-D multichannel signals. Estimation of a 2-D multichannel signal relies on the ability to decorrelate the signal in both space and channel. The new estimator we present uses a 2-D discrete wavelet transform to approximately decorrelate the signal in space, allowing for efficient estimation of both stationary and nonstationary signals. When channel correlation is unknown, but depends only on channel separation, the DFT may be utilized to decorrelate the channel efficiently. In contrast to the optimal estimation scheme, our new estimator does not require second-order signal statistics, making it well suited to many applications. In addition to providing vastly improved visual quality, the new estimator typically yields signal-to-noise ratio gains of over 12 dB.

Traditionally, video encoders have been designed assuming that the more redundancy is removed, the better the encoder. However, on current laptops, reducing the compression efficiency of the video encoder by reducing the number of instructions used to perform compression can actually reduce the total energy used to encode and transmit a sequence. The correct balance between computation and compression efficiency may change dynamically, motivating adaptive encoders. At the same time, recent general-purpose processors also employ energy-driven adaptations. For best gains, the adaptations in the hardware and application layers must be coordinated. From a system design viewpoint, this coordination must happen through minimal, well-defined interfaces. This paper develops (1) an adaptive video encoder for general-purpose processors that trades computational complexity for compression efficiency to minimize total system energy, and (2) a method for determining the best configuration for such an encoder when running on a processor that is also adaptive. Our adaptive processor employs recent energy saving techniques of dynamic voltage and frequency scaling and architectural adaptation. Using a detailed simulator, we show that our cross-layer adaptive application algorithm reduces energy significantly, when employed on a fixed or adaptive processor.

Current acoustical recording systems provide limited flexibility in sound recovery; single microphone systems reconstruct sound over time at a single spatial location, or "one-dimensional" (temporal) recording, whereas arrays may provide isolation in an azimuthal direction of along a pencil beam (two or three dimensional, respectively). Full four-dimensional recovery of a source at any spatial location, particularly using an arbitrary, modest array of acoustical sensors, enables many new applications. A modified frequency-domain MVDR adaptive beamforming algorithm especially tailored for high-performance recovery of nonstationary targets in similar background demonstrates in simulation SIR gains of over 12 dB for speech-in-speech acoustic target recovery, even in situations with more sources than sensors.

In this paper, we present a novel DFT- and wavelet-based estimation scheme for hyperspectral imagery. Optimal hyperspectral image estimation relies on the ability to decorrelate the signal in both space and channel at the cost of requiring second-order signal statistics. This statistical requirement is removed by the proposed estimator, which approximately decorrelates the signal in space using a 2D discrete wavelet transform and in channel using a discrete Fourier transform. In addition to allowing extremely efficient estimation, the proposed estimator vastly improves visual quality and yields typical signal-to-noise ratio gains of over 14 dB.

We report the development of a multi-modal, flexible tactile sensing skin based on polymer substrates and integrated micromachining technology with increased mechanical robustness relative to silicon tactile devices. This polymer-based tactile skin is very unique because it includes the following sensing modalities beyond surface roughness and contact force measurement: thermal conductivity, hardness, temperature, and its own curvature. These new modalities allow the demonstrated tactile sensors to characterize an object or a contact event in a more comprehensive fashion. Sensing is accomplished via thin film metal gold heaters, nickel RTD's (Resistance Temperature Device), and NiCr (nichrome) strain gauges. Experimental characterization of the sensors' performance and potential for application to robotics are presented.

Connexions is a new approach to authoring, teaching, and learning that aims to fully exploit modern information technology. Available free of charge to anyone under open-content and open-source licenses, Connexions offers custom-tailored, current course material, is adaptable to a wide range of learning styles, and encourages students to explore the links among courses and disciplines. In contrast to the traditional process of textbook writing and publishing, Connexions fosters world-wide, cross-institution communities of authors, instructors, and students, who collaboratively and dynamically fashion “modules” from which courses are constructed. We believe the ideas and philosophy embodied by Connexions have the potential to change the very nature of textbook writing and publishing, producing a dynamic, interconnected educational environment that is pedagogically sound, both time and cost efficient, and fun. This paper overviews the philosophy and technology behind Connexions and describes a nascent community developing material for DSP education.

The high peak-to-average power ratio (PAR) in OFDM systems can significantly limit performance. Tone reservation techniques use unavailable or reserved tones to design a peak-cancelling signal that lowers the PAR of a transmit data block. These techniques have been shown to reduce the PAR for real-baseband OFDM signals, such as for ADSL. Our fast-converging real-baseband tone-reservation approach based upon active-set methods is extended to reduce PAR in complex-baseband signals. This new technique converges very quickly toward a minimum-PAR solution at a low cost.

The spatial channel in applications such as radar, sonar, and wireless communications is typically characterized by complex signal scattering leading to multiple signal components arriving at the array from a spread of angles. This multipath angle spread is well known to lead to loss of spatial signal coherence across a receiving sensor array, requiring complicated combining schemes to achieve optimal performance, particularly when the signal is only partially coherent across the array. We show that the discrete Fourier transform serves as an efficient, robust, and asymptotically optimal spatial combiner for uniform linear arrays in multipath channels. In addition, the proposed spatial processing allows for convenient integration of conventional frequency-domain methods for angle-of-arrival searches. Simulation results show that the proposed combining scheme provides near-optimal performance at significantly less computation, even for arrays of modest size.

The multi-disciplinary nature of a DSP laboratory course makes it depend more heavily on the surrounding curriculum than many other courses. We argue that the most effective DSP laboratory courses are tailored to their specific curricula. Based on almost 15 years of experience in designing DSP laboratory courses for three different electrical and computer engineering (ECE) curricula at two different institutions, we present a general philosophy for designing DSP laboratory courses and evaluate our previous course designs as case studies.

Quadratic distributions such as time-frequency distributions and ambiguity functions have many useful applications. In some cases it is desirable to have a quadratic distribution of more than two variables. Using the technique of applying operators to variables, general quadratic distributions of more than two variables can be developed. We use this technique to develop a four-parameter quadratic distribution that includes variables of time, frequency, lag, and Doppler effect. A general distribution is first developed and some of the mathematical properties are discussed. The distribution is then applied to the improvement of an adaptive time-frequency distribution. An example signal is shown to evaluate the performance of the technique.

Users of media broadcasting services generally experience different grades of overall performance because of the non-uniform quality of their channel conditions. In order to provide all users equally satisfactory performance, we propose a new performance criterion named "minimax disappointment", based on layered service levels and the basic principle of joint source-channel coding. This criterion minimizes for all users the maximum value of performance degradation between the received performance and the expected optimal performance, given each user's individual channel situation. In support of this criterion, we develop two broadcasting systems and a gradient-based optimization scheme. Our systems achieve universally near-maximal performance for multiple user classes by providing layered service.

We present an optimal, computationally efficient, integer-bit power allocation algorithm for discrete multitone modulation. Using efficient lookup table searches and a Lagrange-multiplier bisection search, our algorithm converges faster to the optimal solution than existing techniques and can replace the use of suboptimal methods because of its low computational complexity. Fast algorithms are developed for the data rate and performance margin maximization problems.

With the rapid growth of multimedia content in wireless communication, there is an increasing demand for efficient image and video transmission systems. We present a joint source-channel matching scheme for wireless video transmission which jointly optimizes the source and channel coder to yield the optimal transmission quality while satisfying real-time delay and buffer constraints. We utilize a parametric model approach which avoids the necessity of having detailed a priori knowledge of the coders, thus making the scheme applicable to a wide variety of source and channel coder pairs. Simulations show that the scheme yields excellent results and works for several different types of source and channel coders.

The vector constant modulus algorithm (VCMA) was previously introduced as an extension of the constant modulus algorithm (CMA) which can equalize data from shaped sources having nearly Gaussian marginal distributions. Some simple changes in the structure of VCMA allow it to be used in fractionally spaced equalizers with their attendant benefits. Although developed with shell mapping in mind, VCMA can also equalize data from other shaping methods such as trellis shaping. Furthermore, the vector modulus concept from VCMA can be successfully applied to other algorithms based on constant modulus criteria, including reduced constellation algorithm (RCA) and multimodulus algorithm (MMA). Simulations have verified all of these results.

Many algorithms for blind source separation have been introduced in the past few years, most of which assume statistically stationary sources. In many applications, such as separation of speech or fading communications signals, the sources are nonstationary. We present a new adaptive algorithm for blind source separation of nonstationary signals which relies only on the nonstationary nature of the sources to achieve separation. The algorithm is an efficient, online, stochastic gradient update based on minimizing the average squared cross-output-channel-correlations along with deviation from unity average energy in each output channel. Advantages of this algorithm over existing methods include increased computational efficiency, a simple on-line, adaptive implementation requiring only multiplications and additions, and the ability to blindly separate nonstationary sources regardless of their detailed statistical structure.

The Department of Electrical and Computer Engineering at the University of Illinois recently adopted new undergraduate curricula. The most radical change was the introduction of ECE 210, Analog Signal Processing, in place of both the sophomore-level circuit analysis course and the junior-level signals and systems course. The new course combines core material from these traditional courses, along with applications such as AM radio and a modest laboratory component, in a way that improves both the students' understanding and their motivation. The new course still serves well as the base of the required curriculum and as a prerequisite for subsequent courses, while realizing savings in the early curriculum and allowing more time for advanced signal processing and systems courses in future semesters.

Multichannel sensor array processing has received considerable attention in many important areas of signal processing. Almost all data recorded by multisensor instruments contain various amounts of noise, and much work has been done in developing optimal processing structures for estimating the signal source from the noisy multichannel observations. The techniques developed so far assume the signal and noise processes are at least wide-sense-stationary so that optimal linear estimation can be achieved with a set of linear, time-invariant filters. Unfortunately, nonstationary signals arise in many important applications and there is no efficient structure with which to optimally deal with them. While wavelets have proven to be useful tools in dealing with certain nonstationary signals, the way in which wavelets are to be used in the multichannel setting is still an open question. Based on the structure for optimal linear estimation of nonstationary multichannel data and statistical models of spatial signal coherence, we propose a method to obtain an efficient multichannel estimator based on the wavelet transform.

This paper describes an approach to boundary detection in ultrasound speckle based on an image enhancement technique. The enhancement algorithm works by filtering the image with "sticks," short line segments which are varied in orientation to achieve the maximum projected value at each point. The statistical properties of this approach have been described in an earlier paper; in this work we present three significant extensions to improve the performance of the basic method. First, we investigate the effect of varying the size and shape of the sticks. We show that these variations affect the performance of the algorithm in very fundamental ways, for example by making it more or less sensitive to thinner or more tightly curving boundaries. Second, we present a means of improving the performance of this technique by estimating the distribution function of the orientation of the line passing through each point. Finally, we show that images can be "stained" for easier visual interpretation by applying to each pixel a false color whose hue is related to the orientation of the most prominent line segment at that point. Examples are given to illustrate the performance of the different settings on a single image.

Summary form only given. With the rapid growth of wireless communications systems there is an increasing demand for efficient image and video transmission. Significant performance gains can be obtained from joint source channel matching where system resources are assigned based on the tradeoff between data and redundancy. We develop a more general approach for joint source channel matching based on a parametric distortion model that incorporates the flexibility and constraints of both the source and the channel. We use parametric models describing source and channel. We use parametric models describing source and channel characteristics that can be accurately applied to most classes of source and channel coders. To show the generality of our approach, we applied it to the familiar Said-Pearlman progressive image coder with two types of channel coders, a coder with orthogonal symbols of different power, and fixed-rate BPSK modulation overlaid with Reed Solomon codes.

Quadratic time-frequency representations (TFRs) and time-scale representations (TSRs) have been shown to be very useful for detecting nonstationary signals in the presence of nonstationary noise. The theory developed thus far is only for the single observation case; however, in many situations involving signal detection, there are advantages in using an array of receiving sensors. Sensor arrays allow for target or source localization and can provide a large gain in the SNR. We show that time-frequency and time-scale representations provide a natural structure for the detection of a large class of nonstationary signals in the presence of nonstationary noise using an array of sensors. That is, time-frequency and time-scale provide a detection structure that is both optimal and allows for efficient implementation. In developing the TFR/TSR-based optimal quadratic array processor, we consider several types of array environments including those with full, partial, and no coherence.

Application of joint source-channel matching in heterogeneous multimedia environments will demand general source-channel optimization schemes suitable for a wide variety of source coding standards, channel coders, and variable channel conditions. We develop a general approach for joint source-channel matching based on a parametric distortion model that can be accurately applied to most classes of source and channel coders. Our simulations indicate that it may be possible to obtain nearly all of the benefits of joint source-channel optimization by matching existing source and channel coding standards using the simple and general approach we propose.

We present an optimal, efficient power allocation algorithm for discrete multitone modulation (DMT). Using efficient lookup table searches and a Lagrange multiplier bisection search, our algorithm converges much faster to the optimal solution than existing techniques and can replace the use of suboptimal methods because of its low computational complexity. A fast algorithm is developed and a pseudocode is provided.

Application of joint source-channel matching in heterogeneous multi-media environments will demand general source-channel optimization schemes suitable for a wide variety of source coding standards, channel coders, and variable channel conditions. We develop a general approach for joint source-channel matching based on a parametric distortion model that can be accurately applied to most classes of source and channel coders. Our simulations indicate that it may be possible to obtain nearly all of the benefits of joint source-channel optimization by matching existing source and channel coding standards using the simple and general approach we propose.

Due to high data rates and limited bandwidth as well as limited battery power, wireless multimedia communications systems must be optimized in every possible way. We develop a generic matching scheme for wireless image and video communication in which the three most significant components: the source coder, the channel coder, and hardware power consumption, are jointly optimized. That is, we maximize the end-to-end image quality subject to a total power constraint on both the RF transmission power and the power consumption of the digital implementation of the channel coder, which represents a major portion of the total hardware power in short-range applications.

The constant modulus algorithm (CMA), a popular method for performing blind equalization, has the drawback of failing when the input data has a Gaussian distribution or a kurtosis larger than three. This means that source shaping (which is used to increase the shaping gain) can produce data unequalizable by CMA and other important blind equalization techniques. This paper proposes a simple extension of CMA that blindly equalizes many types of shaped constellations, including those shaped by shell mapping.

This paper considers the problem of detecting lines in speckle imagery, such as that produced by synthetic aperture radar (SAR) or ultrasound techniques. Using the physical principles that account for the speckle phenomenon, we derive the optimal detector for lines in fully developed speckle, and we compare the optimal detector to several suboptimal detection rules that are more computationally efficient. We show that when the noise is uncorrelated, a very simple suboptimal detection rule is nearly optimal, and that even in colored speckle, a related class of detectors can approach optimal performance. Finally, we also discuss the application of this technique to medical ultrasonic images, where the detection of tissue boundaries is considered as a problem of line detection.

Interference suppression in spread spectrum communication systems is often essential for achieving maximum system performance. Existing interference suppression methods do not perform well for most types of nonstationary interference. We first consider interference suppression schemes based on adaptive orthogonal time-frequency decompositions, such as wavelet packet and arbitrary dyadic time-frequency tilings. These methods often reduce interference substantially, but their performance can vary dramatically with minor changes in interference characteristics such as the center frequency. To circumvent these drawbacks, we propose a multiple overdetermined tiling (MODT) with an accompanying blind interference excision scheme which appears very promising for mitigating time-frequency-concentrated interference. Simulations with narrowband, impulsive, and simultaneous impulsive and narrowband interference compare the performance of the various methods and illustrate the promise of approaches based on multiple overdetermined tilings.

It is commonly known that wavelet expansions are very sensitive to translation of an input signal due to their dyadic structure. Without proper time alignment of the signal to the decomposition basis vectors, a compact expansion may be unattainable. Recently, work has been done to combat translation sensitivity by determining the best translation-invariant expansion of a signal with respect to a set of bases. However, the dyadic structure of wavelet expansions also results in frequency alignment sensitivity. In this work, we introduce a frequency-shifted wavelet packet library as well as an efficient best basis algorithm for determining the best signal representation among all frequency shifts of a signal.

This article presents a method of adaptively adjusting the window length used in short-time Fourier analysis, related to our earlier work in which we developed a means of adaptively optimizing the performance of the cone kernel distribution (CKD). The optimal CKD cone length is, by definition, a measure of the interval over which the signal has constant or slowly changing frequency structure. The article shows that this length can also be used to compute a time-varying short-time Fourier transform (STFT). The resulting adaptive STFT shares many desirable properties with the adaptive CKD, such as the ability to adapt to transient as well as long-term signal components. The optimization requires O(N) operations per step, less than the fast Fourier transform (FFT) used in computing each time slice, making it competitive in complexity with nonadaptive time-frequency algorithms.

Although several kernel design methods have been proposed for improved time-frequency responses (TFRs) for various signal classes, the family of simple periodic signals has not been specifically considered. Many signals, such as speech and music, which are analyzed using time-frequency techniques exhibit a strong periodic structure. Here we develop an improved TFR designed for periodic functions, under the assumption that the fundamental period of the signal is known or can at least be accurately estimated.

Reliable detection of engine knock is an important issue in the design and maintenance of high performance internal combustion engines. Cost considerations dictate the use of vibration signals, measured at the engine block, for knock detection. Conventional techniques use the energy in a bandpass filtered version of the vibration signal as a measure. However, the low signal-to-noise ratio (SNR) in the vibration measurements significantly degrades the performance of such bandpass energy detectors. We explore the design and application of more general quadratic detection procedures, including time-frequency methods, to this challenging problem. We use statistics estimated from labeled training data to design the detectors. Application of our techniques to real data shows that such detectors, by virtue of their flexible structure, improve the effective SNR, thereby substantially improving the detection performance relative to conventional methods.

Generalized joint signal representations (JSRs) extend the scope of joint time-frequency representations (TFRs) to a richer class of nonstationary signals, but their use, just as in the case of TFRs, has been primarily limited to qualitative, exploratory data analysis. To exploit their potential more fully, JSR-based statistical signal processing techniques need to be developed that can be successfully applied in real-world problems. We present an optimal detection framework based on arbitrary generalized quadratic JSRs, thereby making it applicable in a wide variety of detection scenarios involving nonstationary stochastic signals, noise and interference. For any given class of generalized JSRs, we characterize the corresponding class of detection scenarios for which such JSRs constitute canonical detectors, and derive the corresponding JSR-based detectors. Our formulation also yields a very useful subspace-based interpretation in terms of corresponding linear JSRs that we exploit to design optimal detectors based on only partial signal information.

Generalized joint signal representations extend the scope of joint time-frequency representations to a richer class of nonstationary signals. Cohen's marginal-based generalized approach is canonical from a distributional viewpoint, whereas, in some other applications, for example, in a signal detection framework, a covariance-based formulation is needed and/or more attractive. We present a canonical covariance-based recipe for generating generalized joint signal representations. The method is highlighted by its simple characterization and interpretation, and naturally extends the concept of the corresponding linear representations.

Time-frequency based methods, particularly quadratic (Cohen's-class) representations, are often considered for detection in applications ranging from sonar to machine monitoring. We propose a method of obtaining near-optimal quadratic detectors directly from training data using Fisher's optimal linear discriminant to design a quadratic detector. This detector is optimal in terms of Fisher's scatter criterion as applied to the quadratic outer product of the data vector, and in early simulations appears to closely approximate the true optimal quadratic detector. By relating this quadratic detector to an equivalent operation on the Wigner distribution of a signal, we derive near-optimal time-frequency detectors. A simple example demonstrates the excellent performance of the method.

Joint time-frequency representations have proven very useful for analyzing signals in terms of time-frequency content. In an attempt to tailor joint representations to a richer class of signals, two approaches (Cohen's (1991) and Baraniuk's (see Proc. ICASSP 1993, p.III-320, 1993)) to obtaining joint representations of arbitrary variables have been proposed. Baraniuk's generalization appears broader than Cohen's, since the latter can be recovered from the former as a special case. One of the main results of this paper is that, despite being apparently quite different, the two approaches to generalized joint representations are exactly equivalent. We explicitly characterize the mapping which relates the representations of the two methods, and also determine the corresponding relationship between the operators of the two methods. A practical implication of the results is that one can avoid the group transforms in Baraniuk's approach, which may not be computationally efficient, by replacing them with Fourier transforms in Cohen's method.

This paper presents a novel adaptation of the median filter to the problem of boundary-preserving speckle reduction in ultrasonic imaging. The technique involves applying a bank of oriented one-dimensional median filters to the image, and retaining at each point the largest value among all the filter bank outputs. The result is an operator which suppresses speckle noise while retaining the structure of the image, particularly the thin bright streaks, which tend to occur along boundaries between tissue layers. The technique is compared to a block median filter and an algorithm discussed by Loupas et al. (1989), and is shown to be far superior to the median filter, and noticeably better than the Loupas filter at enhancing thin lines.

Current theories of a time-varying spectrum of a nonstationary process all involve, either by definition or by implementation, an assumption that the signal statistics vary slowly over time. This restrictive quasi-stationarity assumption limits the use of existing estimation techniques to a small class of nonstationary processes. We overcome this limitation by deriving a statistically optimal kernel, within Cohen's (1989) class of time-frequency representations, for estimating the Wigner-Ville spectrum of a nonstationary process. Both time-frequency invariant and time-frequency varying kernels are derived. It is shown that, in the presence of any noise, optimal performance requires a nontrivial kernel, and that optimal estimation may require smoothing filters very different from those based on a quasi-stationarity assumption. An example illustrates that the optimal estimators can potentially yield tremendous improvements in performance over existing methods.>

Presents a statistical approach to edge detection in ultrasound speckle, and uses actual noise statistics to derive an expression for an optimal detection rule. The authors compute the optimal detector for the special case of uncorrelated speckle, and an approximation to the optimal detector for the case when signal-to-noise ratio (SNR) is high. They also discuss a simple sub-optimal detection scheme, and show that its performance is close to that of the optimal uncorrelated speckle detection rule, and surpasses that of the high-SNR approximation except in the case of extremely high SNR, in which case they both perform equally well.>

A new class of signal analysis tools that generalizes the popular wavelet and short-time Fourier transforms is introduced. The class allows skews and rotations of the analyzing wavelet in the time-frequency plane, in addition to the time and frequency translations and scalings used by conventional transforms. In addition to providing a unifying framework for studying existing time-frequency representations, the general class provides a systematic method for designing new representations with properties useful for certain types of signals.>

The design of finite/wordlength finite impulse response, (FIR) digital filters is of much interest and importance in practical digital systems. In fixed-point implementations of digital filters, the designer is allowed only a finite number of bits for each coefficient. In general, the best choice for the finite wordlength coefficients is neither the truncated nor rounded infinite-precision values. A new method for quantizing cascaded subfilters to produce an overall filter that improves upon or compares favorably with the optimal quantized direct-form implementation is introduced. In particular, this method enhances performance in designs requiring deep stopband suppression, which is very common in applications. By cascading subfilters, it is possible to improve the stopband suppression. A difficulty with designing cascaded subfilters is the nonlinearity of the problem. A Taylor series approximation is used to develop a linear integer program for the optimal cascaded coefficients, improving the stopband suppression with no additional loss in the passband deviation.>

An optimization formulation for designing signal-dependent kernels that are based on radially Gaussian functions is presented. The method is based on optimality criteria and is not ad hoc. The procedure is automatic. The optimization criteria are formulated so that the resulting time-frequency distribution (TFD) is insensitive to the time scale and orientation of the signal in time-frequency. Examples demonstrate that the optimal-kernel TFD offers excellent performance for a larger class of signals than any current fixed-kernel representation. The technique performs well in the presence of substantial additive noise, which suggests that it may prove useful for automatic detection of unknown signals in noise. The cost of this technique is only a few times greater than that of the fixed-kernel methods and the 1/0 optimal kernel method.>

The application of a multiplicative modification function, or mask, to the Wigner distribution (WD) of a signal has been proposed as an implementation of time-varying filtering in the time-frequency plane. Because the least squares synthesis technique involves an eigenvalue/eigenvector decomposition, the effects of mask size, shape, and placement on the reconstructed system output are not well understood. However, for the case of a Gaussian input signal and a Gaussian shaped mask, the result of the least squares synthesis procedure on the modified WD can be expressed analytically. This expression for the system output is a function of the input signal and mask Gaussian parameters and can be used to analyze the effects of the mask on the output. For masks that are smaller than the WD of the input, the reconstructed signal has a WD related to the mask and not the system input.>

Two signal components are considered resolved in a time-frequency representation when two distinct peaks can be observed. The time-frequency resolution limit of two Gaussian components, alike except for their time and frequency centers, are determined for the Wigner distribution and smoothed variants of the Wigner distribution (the pseudo-Wigner distribution, the smoothed Wigner distribution, and the squared magnitude of the short-time Fourier transform). The matched-window short-time Fourier transform has the highest resolution in time-frequency; it is about 24% greater than that of the Wigner distribution. An explanation of this rather surprising result is presented. The pseudo-Wigner distribution provides more resolution in one direction than the other representations, if substantial resolution in the other direction can be sacrificed.>

We present a data-adaptive time-frequency representation that obtains high resolution of signal components in time-frequency. This representation overcomes the often poor resolution of the traditional short-time Fourier transform, while avoiding the nonlinearities that make the Wigner distribution and other bilinear representations difficult to interpret and use. The new method uses adaptive Gaussian windows, with the window parameters varying at different time-frequency locations to maximize the local signal concentration in time-frequency. Two methods for selecting the Gaussian parameters are presented: a parameter estimation approach, and a method that maximizes a measure of local signal concentration.

Since 1965, when Cooley and Tukey published their famous paper on the radix-2 fast Fourier transform, much effort has gone into developing even more efficient algorithms. Most algorithms, however, do not directly handle real-valued data very well, and them exist several ways to solve that problem. This paper derives a new algorithm; the decimation-in-time real-valued split-radix FFT, which can transform any length N = 2Msequence but uses less operations than any other known real-valued FFF, which is the fastest Cooley-Tukey real-valued transform in use. Instead of breaking the transform down equally as in traditional algorithms, the even and odd indexed parts are broken down differently in the split-radix algorithm. This gives a significant savings in both additions and multiplications over any fixed radix Cooley-Tukey FFT. The paper compares the split-radix transform with several of the already existing methods such as the Hartley transform, the prime factor, Winograd, Cooley-Tukey etc, and shows in which cases a specific algorithm is faster than the rest.