Search

Delirium is an under-diagnosed yet frequently occurring clinical complication with potentially serious consequences for intensive care unit (ICU) patients. Diagnosis is currently reactive and based upon qualitative assessment of the patient's cognitive status by ICU staff. Here, we conducted a preliminary investigation into whether emerging quantitative electroencephalography (QEEG) analysis techniques can accurately discriminate between delirious and non-delirious patients in an ICU setting. Resting EEG recordings from 5 ICU patients in a state of delirium and 5 age matched control patients were analyzed using autoregressive spectral estimation for quantification of EEG power and renormalized partial directed coherence for analysis of directed functional connectivity. Delirious subjects exhibited pronounced EEG slowing as well as severe general loss of directed functional connectivity between recording sites. Distinction between groups based on these parameters was surprisingly clear given the low sample size employed. Furthermore, by targeting the electrode positions where effects were most apparent it was possible to clearly segregate patients using only 3 scalp electrodes. These findings indicate that quantitative diagnosis and monitoring of delirium is not only possible using emerging QEEG methods but is also accomplishable using very low-density electrode systems.

Objective: the purpose of this paper was to investigate the effect of spectral computation methods on the estimation of the four ultrasonic backscatter parameters, namely, the apparent integrated backscatter (AIB), the zero frequency intercept of apparent backscatter (FIAB), the frequency slope of apparent backscatter (FSAB), and the backscatter spectral centroid shift (SCS) from the backscattered signal of interest (SOI), and their subsequent correlations with cancellous bone parameter [bone volume/total volume (BV/TV)]. Methods: ultrasonic backscatter measurements were performed on 26 bovine cancellous bone specimens using a 1.0-MHz focused transducer. Four spectral estimation algorithms, including the classical periodogram, the autoregressive (AR) Burg algorithm, the AR covariance algorithm, and the AR modified covariance algorithm, were used to calculate the ultrasonic parameters. Influence of the signal's delay time (T1) and its length (T2) on the strength of correlation between BV/TV and backscatter parameters was also studied. Results: the results have demonstrated that the AR-based estimators provide much more reliable and stronger correlations between the BV/TV and backscatter parameters than the classical periodogram. Recommendations for choosing SOI were also suggested. Conclusion: these results indicate that the AR-based method has a promising potential to enhance the performance of evaluation and diagnosis of the cancellous bone using the ultrasonic backscatter method. Significance: the enhancement of the correlation may provide a positive impact on the ultrasonic backscatter method to diagnose and monitor the bone quantity.

The manuscript reports on the implementation of autoregressive spectral estimation aimed at improving the accuracy of detecting short-lasting events in signals acquired by a retinal birefringence scanning device that was reported earlier. A signal consisting of two frequency components is generated, where each frequency is a multiple of the scanning frequency. One frequency is produced during central fixation, while another one prevails during off-central fixation. These components may be of a very short duration, presenting a challenge for the FFT to identify them with sufficient time- and frequency resolution. Autoregressive spectral estimation using the Burg algorithm provided a satisfactory solution, capable of reliably differentiating between the two frequency components (96 and 192 Hz) on signal segments of duration as short as 5 ms. The device and the signal analysis methods were developed originally with the purpose of checking for eye alignment and strabismus – a major risk factor for amblyopia. The method enables the technology to work with less-cooperative patients, such as young children. Other medical and non-medical applications are possible.

Versions of implementing initial synchronization in systems with pseudorandom operating frequency tuning are studied. These versions are based on applying a recurrent acquisition with an autoregressive spectral estimation. Known algorithms are analyzed, and versions of their modification are proposed. Simulation models of considered solutions are constructed, and quantitative data that characterize their noise immunity are obtained.

The electrical activity of the human stomach, which normally shows a frequency of about 0.05 Hz, may be studied non-invasively by either cutaneous electrogastrography (EGG) or surface magnetogastrography (MGG). Detection of changes in frequency with time may be useful to characterize gastric disorders. The fast Fourier transform (FFT) has been the most commonly used method for the automated spectral analysis of the signals obtained from the EGG or the MGG. We have used an autoregressive (AR) parametric spectrum estimator to analyse simulated signals of gastric electrical activity, and to evaluate the results of human studies using EGG and MGG. In comparison with the FFT, our results showed that the AR spectrum estimator provided more detailed qualitative information about frequency variations of short duration simulated signals than the FFT. In the human studies, the AR estimator was as good as the conventional FFT methods in detecting physiological changes in frequency and in identifying abnormal recordings. We conclude that the AR spectral estimator may provide a better qualitative analysis of frequency variations in small portions of the signal, and is as useful as the FFT to analyse human EGG or MGG studies.

A theoretical derivation of autoregressive spectral estimation is briefly presented, at variance with a recent paper by Kauppinen and Saario.

An improved “frequency-weighted” method of spectral estimation is described. Discussion of the method is directed toward the problem of estimating autoregressive process parameters from discrete-time observations corrupted by additive independent noise with known power spectral density. This problem often occurs in speech analysis and other applications. The estimation procedure is cast as the problem of minimizing a frequency-weighted information measure and has theoretical foundations relating it to asymptotic maximum likelihood and minimum information divergence procedures as well as the linear predictive procedures popular for speech analysis. Topological properties of the frequency-weighted measure are developed and used to prove that the solution to the minimization problem is global and unique. Computational procedures are developed for both the estimation problem and the related problem of testing multiple simple hypotheses. Simulation results are used to support the theoretical development and demonstrate the advantages of the method as compared to “noise filtering” methods that have been popular for speech analysis.

This paper introduces recent advancements in speaker recognition system. It incorporates the background concepts and emphasizes the signal model design technique most commonly used in the state-of-the-art of speaker recognition system. An analytical description is provided of procedures applied in speaker recognition system, covering several methods that are utilized to investigate the autoregressive spectral estimation of the Automatic Speaker Verification (ASV) system. Recent additions in the signal processing area are illustrated that tremendously improve the efficiency of the spectral properties of a speech frame. This paper is meant to explain the concepts of recognition system pragmatically. Also, it envelops some portions of speech recognition system, which is the further enhancement of speaker recognition system. The problems that have been recently invoked when the system is implemented in a large scale through networks have been forwarded. Realization of the ASR system is established in TMS320C30 EVM floating-point processor. Conclusion is presented highlighting the signal modelling techniques in speaker recognition and discussing some current problems which need to be resolved to implement the system ubiquitously. Keywords: speaker recognition system, signal modeling, autoregressive spectral estimation, speech frame Botswana Journal of Technology Vo 14(1) 2005: 63-70

In this paper, we address the problem of blindly separating convolutive mixtures of spatially and temporally independent sources. Source densities are modelled as mixtures of Gaussians. We present an EM algorithm to compute Maximum Likelihood estimates of both the separating filters and the source density parameters, whereas in the state-of-the-art separating filters are usually estimated with gradient descent techniques. The use of the EM algorithm, as an alternative to the usual gradient descent techniques, is advantageous as it provides a faster and more stable convergence and as it does not require the empirical tuning of a learning rate. Besides, we show how multichannel autoregressive spectral estimation techniques can be used in order to properly initialize the EM algorithm. We demonstrate the efficiency of our EM algorithm together with the proposed initialization scheme by reporting on simulations with artificial mixtures. Finally, we discuss the theoretical and practical limitations of our EM approach and point out possible ways of addressing these issues in future works.

Summary form only given. Autoregressive (AR) spectral estimation models the signal whose spectrum is to be evaluated as the output of an all-pole linear filter excited with white noise. The advantages of this approach versus classical spectral estimation techniques are well known. Known algorithms introduced for 2-D AR spectral estimation have been extended to the case of spatio-temporal sequences in order to develop a technique suitable to obtain AR spectral estimates of television sequences. The problems encountered in the evaluation of AR spectral estimates of PAL signal sequences and an experimental tuning procedure have been investigated. >

Time-invariant algorithms of autoregressive spectral estimation were employed in the study of the heart rate variability signal to quantify, on a beat-by-beat basis, the effect of the sympatho-vagal balance in the assessment of the heart rate. Thus, it was possible to obtain the relevant spectral parameters during transient phenomena, in which the hypotheses of stationarity, required for a batch analysis, are not respected. In particular, the focus is on the study of the cardiac transient ischemic pathology, aiming at analyzing the dynamical changes corresponding to the beginning of the attack and during the attack itself. For this purpose, the European ST-T database provides a significant number of cases of ischemic attacks that have been studied through time-invariant spectral analysis for a better comprehension of the origin mechanisms and of the involvement of the autonomic nervous system in the ischemic episodes. >

This thesis treats the topic of multi-dimensional autoregressive (AR) spectral estimation. An iterative algorithm for the solution of toeplitz block- toeplitz matrix equations is presented. This leads to a fast solution of the two dimensional normal equation compared with direct inversion of the autocorrelation matrix. The covariance method is used to estimate the autocorrelation function. Because the resulting matrix is not toeplitz block- toeplitz, a modified iterative algorithm is presented. Quarter-plane and nonsymmetric half-plane support are used, as well as combined quadrant averaging. Results of computer simulation show that in some cases a single iteration is sufficient to produce an acceptable spectral estimate. Because the AR parameters are estimated from previous values, this suggests the possibility to estimate spectral densities of slowly varying random processes. http://archive.org/details/multidimensional1094530666 Lieutenant, United States Navy Approved for public release; distribution is unlimited.

On montre par une analyse asymptotique comprenant les termes du second-ordre, que les estimateurs des moindres carres doivent etre preferes aux estimateurs de Yule-Walker

It is proposed that the autoregressive coefficients (as computed from the estimated autocorrelation function) in an autoregressive spectral estimation scheme may be used for feature extraction purposes. An example is presented where such extractors are used to describe seismic wave traces originating from shallow earthquakes and underground nuclear explosions.

This report is concerned with the development and application of improved techniques for digital signal processing, based on use of residue number system (RNS) to implement the processing functions associated with isolated-word speech recognition. Specifically, the use of RNS in combination and systolic architectures for implementation of speech recognition algorithms was explored. The implementation of time-warping speech algorithm in RNS is described. Keyword include: Signal processing, Speech recognition, systolic architecture, Residue number system (RNS), Autoregressive spectral estimation, and Dynamic time-warping.

The lattice structure offers a convenient visual realization of the Levinson recursion used in solving the Yule-Walker equations. Based on the lattice, a number of methods have been developed recently for all-pole or autoregressive spectral estimation. Of particular importance are methods that do not require windowing of the data, especially for short data lengths. This paper focuses on the importance of the lattice as a tool in spectral estimation.

Fetal breathing movement (FBM) in utero may be an indicator of fetal health. This paper provides a second-by-second estimate of FBM rate. In the absence of a statistical model for the fetal breathing movement, block data structured autoregressive spectral estimation is used. The optimum tapered Burg algorithm provides a minimum variance breathing rate estimate from a short block of data. The data were recorded using a PVDF (PolyVinyliDeneFluoride) transducer which picks up maternal abdominal wall movements. A peak tracking algorithm is used to extract the fetal breathing rate. Results from these signals are presented in graphical form. Further analysis of the fetal breathing rate has revealed periodicities, similar to that observed in the fetal heart rate.

The aim of the present study is to identify the slow wave (SW) of the myoelectric activity of the small bowel (electroenterogram) recorded at the abdominal surface. For this purpose autoregressive spectral estimation (AR) was used. Internal and external recordings were acquired simultaneously from three Beagle dogs (in nine recording sessions). Three methods for obtaining the optimal order of AR model were applied to one-minute-length: abdominal myoelectric signal: Final prediction error (FPE), Akaike information criterion (AIC) and minimum description length (MDL). The optimal order was of 10, with a frequency resolution /spl Delta/f= 0,025 Hz. The frequency of the maximum peaks of the signal spectrum can be found around 0,3 Hz. AR spectral analysis shows that the SW can be detected on the abdominal recordings of the small bowel myoelectrical activity without breathing interference.