This paper investigates the effect of noises added to hidden units of AutoEncoders linked to multilayer perceptrons. It is shown that internal representation of learned features emerges and sparsity of hidden units increases when independent Gaussian noises are added to inputs of hidden units during the deep network training. It is also shown that the weights that connect the contaminated hidden units with the next layer have smaller values and outputs of hidden units tend to be more definite (0 or 1). This is expected to improve the generalization ability of the network through this automatic structuration by adding the noises. This network structuration was confirmed by experiments for MNIST digits classification via a deep neural network model.

In this paper, a grey filtering approach based on GM(1,1) model is proposed. Then the grey filtering is applied to speech enhancement. The fundamental idea in the proposed grey filtering is to relate estimation error of GM(1,1) model to additive noise. The simulation results indicate that the additive noise can be estimated accurately by the proposed grey filtering approach with an appropriate scaling factor. Note that the spectral subtraction approach to speech enhancement is heavily dependent on the accuracy of statistics of additive noise and that the grey filtering is able to estimate additive noise appropriately. A magnitude spectral subtraction (MSS) approach for speech enhancement is proposed where the mechanism to determine the non-speech and speech portions is not required. Two examples are provided to justify the proposed MSS approach based on grey filtering. The simulation results show that the objective of speech enhancement has been achieved by the proposed MSS approach. Besides, the proposed MSS approach is compared with HFR-based approach in [4] and ZP approach in [5]. Simulation results indicate that in most of cases HFR-based and ZP approaches outperform the proposed MSS approach in SNRimp. However, the proposed MSS approach has better subjective listening quality than HFR-based and ZP approaches.

The two-sample locally optimum rank detector test statistics for composite signals in additive, multiplicative, and signal-dependent noise are obtained in this letter. Compared with the structure of the one-sample locally optimum rank detector, that of the two-sample locally optimum rank detector is shown to be simpler, although it needs more computations. It is known that there is a trade-off of computational complexity and structural simplicity between the one- and two-sample detectors.

The one-sample locally optimum rank detector test statistics for composite signals in multiplicative and signal-dependent noise are obtained. Since the one-sample locally optimum rank detector makes use of the sign statistics of observations as well as the rank statistics, both 'even' and 'odd' score functions have to be considered. Although the one-sample locally optimum rank detector requires two score functions while the two-sample detector requires only one score function, the one-sample detector requires fewer calculations since it has to rank fewer observations.

This paper deals with the problem of autoregressive (AR) spectral estimation from a finite set of noisy observations without a priori knowledge of additive noise power. A joint technique is proposed based on the high-order and true-order AR model fitting to the observed noisy process. The first approach utilizes the uncompensated lattice filter algorithm to estimate the parameters of the over-fitted AR model and is one-pass. The latter uses the noise compensated low-order Yule-Walker (LOYW) equations to estimate the true-order AR model parameters and is iterative. The desired AR parameters, equivalently the roots, are extracted from the over-fitted model roots using a root matching technique that utilizes the results obtained from the second approach. This method is highly accurate and is particularly suitable for cases where the system of unknown equations are strongly nonlinear at low SNR and uniqueness of solution from the LOYW equations cannot be guaranteed. In addition, fuzzy logic is adopted for calculating the step size adaptively with the cost function to reduce the computational time of the iterative total search technique. Several numerical examples are presented to evaluate the performance of the proposed scheme in this paper.

In the present paper we shall examine the real-time restoration of biomedical signals under additive noises. Biomedical signals measured by instruments such as catheter manometers, ambulatory electrocardiographs and thermo-dilution sensors are susceptible to distortion and noise. Therefore, such signals must be restored to their original states. In the present study, nonstationary biomedical signals are observed and described using a mathematical model, and several restoration filters that are composed of a series of applications of this model are proposed. These filters restored band-limited approximations of the original signals in real-time. In addition, redundancy is introduced into these restoration filters in order to suppress additive noise. Finally, an optimum filter that accounts for restoration error and additive noise is proposed.

The additive noise response of a charge pump phase-locked loop in the synchronous mode of operation has been studied. In order to determine the tracking and noise performances of the loop, mean square values of tracking error and local oscillator phase jitter have been analytically obtained. Analytical results agree well with the simulation results obtained here and elsewhere. The analysis performed can be used in choosing different system parameters for optimum system operation.

We present a new method for the identification of time-invariant multichannel autoregressive (AR) processes corrupted by additive white observation noise. The method is based on the Yule-Walker equations and identifies the autoregressive parameters from a finite set of measured data. The input signals to the underlying process are assumed to be unknown. An inverse filtering technique is used to estimate the AR parameters and the observation noise variance, simultaneously. The procedure is iterative. Computer simulation results that demonstrate the performance of the identification method are presented.

In this report, we propose a robust block adaptive digital filter (BADF) which can improve the accuracy of the estimated weights by averaging the adaptive weight vectors. We show that the improvement of the estimated weights is independent of the input signal correlation.