Parameter Estimation in Spectral Resolution Enhancement Based on Forward-Backward Linear Prediction Total Least Square Method.

This paper presents a novel approach for estimating autoregressive (AR) model parameters using deep neural network (DNN) in the AR-Wiener filtering speech enhancement. Unlike conventional DNN that predicts one kind of target, the DNN used in this paper is trained to predict the AR model parameters of speech and noise simultaneously at offline stage. We train this network by minimizing the Euclidean distance between the output of DNN and the AR model parameters of clean speech and noise. At online stage, the acoustic features are first extracted from noisy speech as the input of the DNN. Then, AR model parameters of speech and noise are estimated by the DNN simultaneously. Finally, the Wiener filter is constructed by the AR model parameters of speech and noise. However, the AR model parameters only models the spectral shape not the spectral details, there are still some residual noise between the harmonics. In order to solve this problem, we introduce the speech-presence probability (SPP), that is, in the test stage, the SPP is estimated and is used to update the Wiener filter. The experimental results show that our approach has higher performance compared with some existing approaches.

Multi-sensor remote sensing image super-resolution aims to provide better characteristics for different types of resolution and compensate the limitations of the particular imaging systems. However, existing super-resolution techniques consider spectral and spatial resolution enhancement separately, i.e. only spatial or only spectral resolution can be enhanced. Among spatial super-resolution methods maximum a posteriori estimation approach with B-TV regularization stands out as one of the best method for spatial resolution enhancement. But existing implementations were designed only for RGB and grayscale photographic imagery. Unlike photographic RGB imagery, multispectral remote sensing images captured by optical sensors often contain more than three spectral channels (red, green and blue) and, moreover, different remote sensing systems produce a different spectral response for the similar spectral components. Therefore, a more complex image acquisition model should be regarded to take into account the variations in bandwidth and number of spectral channels in the case of remote sensing images. In this article, we propose an algorithm aiming to provide spectral-spatial multi-sensor remote sensing image super-resolution. We apply a joint spectral-spatial image acquisition model, that is typical for remote sensing systems, and investigate the super-resolution algorithm streaming from this model and the maximum a posteriori estimation approach with B-TV regularization. We propose a simple way to adapt B-TV regularization in the case of multiple spectral channels. Our experimental results confirm the enhancement in the spectral and spatial resolution of the output image in comparison with the input images. The results of our research demonstrate that the proposed method achieves both spectral and spatial super-resolution.

Remote sensing image pan-sharpening is an appropriate way to fuse panchromatic (PAN) image with high spatial/low spectral resolution and multispectral (MS) image with low spatial/high spectral resolution to obtain a high spatial/high spectral resolution MS image (HRMS). Spectral information preservation and spatial resolution enhancement are the key issues of pan-sharpening problem. We propose a novel pan-sharpening method to decrease the spectral distortion of traditional methods inspired by sparse representation theory. Unlike previous works, details which must be injected to MS image should be extracted directly from a so-called ‘detail dictionary’, trained by high frequency information of PAN image. With two proposed assumptions, we prove that the sparse coefficients of details on the detail dictionary can be obtained from a low pass version of PAN image and the MS image. We use QuickBird and WorldView-2 data sets to evaluate the proposed method. The simulation results show that the proposed method can effectively preserve spectral information and also improve the spatial resolution. The results show that the performance of proposed method is better than the state-of-the-art and TSSC methods quantitatively and perceptually.

Maximum likelihood autoregressive model parameter estimation with noise corrupted independent snapshots

<p/> A parametric signal processing approach for DNA sequence analysis based on autoregressive (AR) modeling is presented. AR model residual errors and AR model parameters are used as features. The AR residual error analysis indicate a high specificity of coding DNA sequences, while AR feature-based analysis helps distinguish between coding and noncoding DNA sequences. An AR model-based string searching algorithm is also proposed. The effect of several types of numerical mapping rules in th proposed method is demonstrated.

Accurately identifying faults in rolling bearing systems by analyzing vibration signals, which are often nonstationary, is challenging. To address this issue, a new approach based on complementary ensemble empirical mode decomposition (CEEMD) and time series modeling is proposed in this paper. This approach seeks to identify faults appearing in a rolling bearing system using proper autoregressive (AR) model established from the nonstationary vibration signal. First, vibration signals measured from a rolling bearing test system with different defect conditions are decomposed into a set of intrinsic mode functions (IMFs) by means of the CEEMD method. Second, vibration signals are filtered with calculated filtering parameters. Third, the IMF which is closely correlated to the filtered signal is selected according to the correlation coefficient between the filtered signal and each IMF, and then the AR model of the selected IMF is established. Subsequently, the AR model parameters are considered as the input feature vectors, and the hidden Markov model (HMM) is used to identify the fault pattern of a rolling bearing. Experimental study performed on a bearing test system has shown that the presented approach can accurately identify faults in rolling bearings.

Speaker and language recognition in noisy and degraded channel conditions continue to be a challenging problem mainly due to the mismatch between clean training and noisy test conditions. In the presence of noise, the most reliable portions of the signal are the high energy regions which can be used for robust feature extraction. In this paper, we propose a front end processing scheme based on autoregressive (AR) models that represent the high energy regions with good accuracy followed by a modulation filtering process. The AR model of the spectrogram is derived using two separable time and frequency AR transforms. The first AR model (temporal AR model) of the sub-band Hilbert envelopes is derived using frequency domain linear prediction (FDLP). This is followed by a spectral AR model applied on the FDLP envelopes. The output 2-D AR model represents a low-pass modulation filtered spectrogram of the speech signal. The band-pass modulation filtered spectrograms can further be derived by dividing two AR models with different model orders (cut-off frequencies). The modulation filtered spectrograms are converted to cepstral coefficients and are used for a speaker recognition task in noisy and reverberant conditions. Various speaker recognition experiments are performed with clean and noisy versions of the NIST-2010 speaker recognition evaluation (SRE) database using the state-of-the-art speaker recognition system. In these experiments, the proposed front-end analysis provides substantial improvements (relative improvements of up to 25%) compared to baseline techniques. Furthermore, we also illustrate the generalizability of the proposed methods using language identification (LID) experiments on highly degraded high-frequency (HF) radio channels and speech recognition experiments on noisy data.

The present study aims to establish normal limits of short–term Heart Rate Variability (HRV) indices with the two popular spectral domain methods of Fast Fourier Transform (FFT) and Autoregressive (AR) model in three psychosomatically important age groups of healthy male subjects. The three considered age groups are 18–30 years, 30–45 years and 45–60 years and HRV decreases in the higher age groups when evaluated with both the FFT and AR models. Vagal control becomes weak with the ageing process. However, the corresponding HRV indices computed with both the FFT and AR models differ quantitatively, which may be due to methodological differences. The FFT–based technique evaluates HRV indices on the actual RR interval series length and hence, the HRV indices using this technique are considered to be highly accurate, where as the AR model works on the basis of linear prediction. This study may help towards optimisation of AR models such that the HRV indices calculated with both FFT and AR models are nearly the same.

Autoregressive (AR) modeling is a very popular method for time series analysis. Being linear by nature, it obviously fails to adequately describe nonlinear systems. In this paper, we propose a kernel-based AR modeling, by combining two main concepts in kernel machines. One the one hand, we map samples to some nonlinear feature space, where an AR model is considered. We show that the model parameters can be determined without the need to exhibit the nonlinear map, by computing inner products thanks to the kernel trick. On the other hand, we propose a prediction scheme, where the prediction in the feature space is mapped back into the input space, the original samples space. For this purpose, a pre-image technique is derived to predict the future back in the input space. The efficiency of the proposed method is illustrated on real-life time-series, by comparing it to other linear and nonlinear time series prediction techniques.

An innovative Bayesian system identification method using autoregressive model

The aim of this paper is to present a non‐linear statistical model to fit and forecast the signal‐to‐interference plus noise ratio (SINR) in two‐tier heterogeneous cellular networks which consist of macrocells and femtocells. Since in these networks the number and locations of femtocell base stations (FBS) are variable, SINR forecasting can be useful in some areas such as power control and handover management. So far, linear autoregressive (AR) models have commonly been used in forecasting the received signal strength (rss) in macrocellular networks. However, AR modelling results in high mean square error (MSE) when data are non‐linear. This paper focuses on SINR which takes into account signal strength, interference and noise effects. Moreover, macro‐femto cellular network is considered. The F ‐test results show that the SINR data are non‐linear, leading to use non‐linear models instead of AR model. A non‐linear logistic smooth threshold AR (LSTAR) model is utilised to model and forecast the SINR data. Kolmogorov–Smirnov (K‐S) test demonstrates that LSTAR provides good fitness to the SINR samples. The results indicate that LSTAR model achieves much better performance in modelling and forecasting of SINR data than the AR model.

Speaker recognition in noisy environments is challenging when there is a mis-match in the data used for enrollment and verification. In this paper, we propose a robust feature extraction scheme based on spectro-temporal modulation filtering using two-dimensional (2-D) autoregressive (AR) models. The first step is the AR modeling of the sub-band temporal envelopes by the application of the linear prediction on the sub-band discrete cosine transform (DCT) components. These sub-band envelopes are stacked together and used for a second AR modeling step. The spectral envelope across the sub-bands is approximated in this AR model and cepstral features are derived which are used for speaker recognition. The use of AR models emphasizes the focus on the high energy regions which are relatively well preserved in the presence of noise. The degree of modulation filtering is controlled using AR model order parameter. Experiments are performed using noisy versions of NIST 2010 speaker recognition evaluation (SRE) data with a stateof-art speaker recognition system. In these experiments, the proposed features provide significant improvements compared to baseline features (relative improvements of 20% in terms of equal error rate (EER) and 35 % in terms of miss rate at 10 % false alarm).

The best accuracy for estimated spectra is obtained with parsimonious time series models, which have the smallest number of parameters to guarantee unbiased models. Durbin 's method for the estimation of moving average (MA) parameters uses the estimated parameters of a long autoregressive (AR) model to calculate MA parameters. Probably all pejorative remarks on the quality of Durbin's method in the literature are based on suboptimal or wrong choices for the method of AR estimation or for the order of the intermediate AR model. That AR order should be considerably higher than the order of the best predicting AR model and it should grow with the sample size. Furthermore, the Burg estimates for the AR parameters give the best results because they have the smallest variance of all AR methods with a small bias. The triangular autocorrelation bias of the popular Yule-Walker method of AR estimation can cause large bias errors in finite samples, which makes it unsuited. Durbin's method applied to the proper number of AR parameters estimated with Burg's method outperforms all other known MA estimation methods, asymptotically as well as in finite samples. The accuracy is generally close to the Cramer-Rao bound.

In this paper, we systematically established a chirp modulation theoretical model for spectral focusing coherent anti‐Stokes Raman scattering technology (SF‐CARS). Through in‐depth analysis of the dispersion characteristics of fiber optic dispersion and grating pairs, the control mechanism of dispersion effect on the time frequency characteristics of laser pulses is elucidated, and the influence of laser pulse parameters on spectral resolution and signal‐to‐noise ratio (SNR) is revealed. Through theoretical optimization of precise chirp modulation, we improved the spectral resolution of the SF‐CARS system to 6.7 cm −1 and simultaneously enhanced the SNR to 340, achieving a concurrent improvement in both resolution and detection sensitivity. With the improvement of spectral resolution and SNR of SF‐CARS, its application boundaries for studying dynamic processes will be further expanded.

This paper proposes an adaptive noise suppression method for non-stationary noise based on the Bayesian estimation method. The following conditions are assumed: (1) speech and noise samples are statistically independent, and they follow auto-regressive (AR) processes. (2) The prior distribution of the parameters of the noise AR model of a current frame is identical to the posterior distribution of those parameters calculated in the previous frame. Under these conditions, the proposed method approximates the joint posterior distribution of the AR model parameters and the speech samples by using the variational Bayesian method. Furthermore, we describe an efficient implementation by assuming that all involved covariance matrices have the Toeplitz structure. The proposed method was tested on real speech and noise signals and compared with other noise suppression methods.