Speech enhancement is fundamental for various real time speech applications and it is a challenging task in the case of a single channel because practically only one data channel is available. We have proposed a supervised single channel speech enhancement algorithm in this paper based on a deep neural network (DNN) and less aggressive Wiener filtering as additional DNN layer. During the training stage the network learns and predicts the magnitude spectrums of the clean and noise signals from input noisy speech acoustic features. Relative spectral transform-perceptual linear prediction (RASTA-PLP) is used in the proposed method to extract the acoustic features at the frame level. Autoregressive moving average (ARMA) filter is applied to smooth the temporal curves of extracted features. The trained network predicts the coefficients to construct a ratio mask based on mean square error (MSE) objective cost function. The less aggressive Wiener filter is placed as an additional layer on the top of a DNN to produce an enhanced magnitude spectrum. Finally, the noisy speech phase is used to reconstruct the enhanced speech. The experimental results demonstrate that the proposed DNN framework with less aggressive Wiener filtering outperforms the competing speech enhancement methods in terms of the speech quality and intelligibility.
A navigation complex of an unmanned flight vehicle of small class is considered. Increasing the accuracy of navigation definitions is done with the help of a nonlinear Kalman filter in the implementation of the algorithm on board an aircraft in the face of severe limitations on the performance of the special calculator. The accuracy of the assessment depends on the available reliable information on the model of the process under study, which has a high degree of uncertainty. To carry out high-precision correction of the navigation complex, an adaptive non-linear Kalman filter with parametric identification was developed. The model of errors of the inertial navigation system is considered in the navigation complex, which is used in the algorithmic support. The procedure for identifying the parameters of a non-linear model represented by the SDC method in a scalar form is used. The developed adaptive non-linear Kalman filter is compact and easy to implement on board an aircraft.
A detailed study about the suitable perturbation element shape and location for tunable BW dual mode microstrip filter which has circular ring resonator is presented. BW tuning is achieved by resonator geometry modification. The study explains the effect of a perturbation element on the stability of the center frequency during BW tuning. Different cases have been studied for two shapes of perturbation element; which one is a rectangular and the other is a radial. The treated cases discuss whether the perturbation element is located in the inner or in the outer circumference of the ring, and whether it is a patch or a notch. BW tuning simulation treated the case of FBW3dB increase for two and three times. The best case of perturbation element which has the best center frequency stability has been modeled, simulated, and fabricated at 2.4 GHz. Geometry modification of the filter took into account the RF MEMS modeling. The filter has an elliptic frequency response, and its FBW has been increased in five steps from 1.7% to 5%. The designed filters were evaluated experimentally and by simulation with very good agreement.
In this work, the design of current mode Fractional order filter using VDTAs (Voltage differencing trans-conductance amplifier) as an active element with grounded capacitors has been proposed. The approximate transfer functions of low and high pass filters of fractional order on the basis of the integer order transfer has been shown and the form of those functions of filters is also implemented using VDTA as an active building block. In this work, filters of the different sequence have been realized. The frequency domain simulation results of the proposed filters are obtained on Matlab and PSPICE with TSMC CMOS 180 nm technology parameters. Stability and sensitivity is also verified.
Various components of surface texture are identified, namely form, waviness and roughness. Separation of these components is done by digital filtering. Several problems exist during analysis of two-process surfaces. Therefore the Gaussian robust profile filtering technique was established and has been studied here. The computer generated 2D profiles and 3D surface topographies having triangular scratches as well as measured stratified surfaces were subjected to filtration. However even robust filter applications cause distortion of profiles having valleys wider than 100 μm. In order to minimize the distortion associated with wide and deep valleys, the robust filter should be modified. A special procedure was elaborated for minimizing distortion of roughness profiles caused by filtration. Application of this method to analyses of several profiles was presented. The difference between 1-D and 2-D filtering of surface topography using the same kind of filter was discussed. As a result we found that modification of a 2-D surface topography filter was not necessary.
The paper deals with the basic set-up of single-frequency microchip laser - so called Lyot filter configuration. Description of its operation and practical realization is given. Some results obtained for Nd:YAG/KTP microchip laser are presented. The evidences of single-frequency operation and its limits are emphasized. Described construction constitutes the base for building the frequency stabilization of green 532 nm microchip laser.
Hybrid filter material was obtained via modification of polypropylene (PP) nonwoven with nanosize zinc oxide particles of a high aspect ratio. Modification was conducted as a three-step process, a variant of hydrothermal method used for synthesis of nano-ZnO, adopted for coating three dimensional polymeric nonwoven filters. The process consisted of plasma treatment of nonwoven to increase its wettability, deposition of ZnO nanoparticles and low temperature hydrothermal growth of ZnO rods. The modified nonwovens were investigated by a high resolution scanning electron microscopy (HR-SEM). It has been found that the obtained hybrid filters offer a higher filtration efficiency, in particular for so called most penetrating particle sizes.
Trials of cast steel filtration using two types of newly-developed foam filters in which carbon was the phase binding ceramic particles have been conducted. In one of the filters the source of carbon was flake graphite and coal-tar pitch, while in the other one graphite was replaced by a cheaper carbon precursor. The newly-developed filters are fired at 1000o C, i.e. at a much lower temperature than the currently applied ZrO2-based filters. During filtration trials the filters were subjected to the attack of a flowing metal stream having a temperature of 1650°C for 30 seconds. Characteristic of the filters’ properties before and after the filtration trial were done. It was found, that the surface reaction of the filter walls with molten metal, which resulted in local changes of the microstructure and phase composition, did not affect on expected filter lifetime and filtration did not cause secondary contamination of cast steel.
The influence of wrong information about transition and measurement models on estimation quality has been presented in the paper. Two methods of a particle filter, with and without the Population Monte Carlo modification, and also the extended and unscented Kalman filters methods have been compared. A small 5-bus power system has been used in simulations, which have been performed based on one data set, and this data set has been chosen from among 100 different – to draw the most general conclusions. Based on the obtained results it has been found that for the particle filter methods the implementation of the slightly higher standard deviation than the true value, usually increases the estimation quality. For the Kalman filters methods it has been concluded that optimal values of variances are equal to the true values.
Single-branch filters are still popular and are commonly used for power quality improvement purposes. Analysis of a single-branch filter is a relatively simple task. Although individual filters tuned to specific harmonics can be easily designed, after connecting them into a group it turns out that the capacitance and inductance mutually influence each other, distorting the resulting frequency characteristics. This article presents a matrix method for design a group of single-branch filters, so that the resultant frequency characteristic satisfies the design requirements including the requirements for location of the frequency characteristic maxima. Designer indicates the frequencies of the parallel resonances.
Many studies have been developed aiming to improve digital filters realizations, recurring to intricate structures and analyzing probabilistically the error's behavior. The work presented in this paper analyzes the feasibility of fixed-point implementation of classical infinite impulse response notch filters: Butterworth, Chebyshev I and II, and elliptic. To scrutinize the deformations suffered for distinct design specifications, it is assessed: the effect of the quality factor and normalized cut-off frequency, in the number of significant bits necessary to represent the filter's coefficients. The implications brought to FPGA implementation are also verified. The work focuses especially on the implementation of power line notch filters used to improve the signal-to-noise ratio in biomedical signals. The results obtained, when quantizing the digital notch filters, show that by applying second-order sections decomposition, low-order digital filters may be designed using only part of double precision capabilities. High-order notch filters with harsh design constraints are implementable using double precision, but only in second-order sections. Thus, it is revealed that to optimize computation time in real-time applications, an optimal digital notch filter implementation platform should have variable arithmetic precision. Considering these implementation constraints, utmost operation performance is finally estimated when implementing digital notch filters in Xilinx Virtex-5 field-programmable gate arrays. The influence of several design specifications, e.g. type, and order, in the filter's behavior was evaluated, namely in regard to order, type, input and coefficient number of bits, quality factor and cut-off frequency. Finally the implications and potential applications of such results are discussed.
The paper presents the equalization problem of non-linear phase response of digital IIR type filters. An improved analytical method of designing a low-order equalizer is presented. The proposed approach is compared with the original method. The genetic algorithm is presented as an iterative method of optimization. The vector and matrix representation of the all-pass equalizer are shown and introduced to the algorithm. The results are compared with the analytical method. In this paper we have also proposed the use of an aging factor and setting the initial population of the genetic algorithm around the solution provided by the analytical methodology
In this paper, the sensitivity analysis of the elliptic filters realized by using biquadratic structures was carried out. The influence of spread the structure parameter values on the shape of the frequency characteristic of the filter transmittance modulus was analyzed. The analysis was limited to the case of even order low-pass filter. Defining the proper class of the sensitivity coefficients, the changes influence of individual structure parameters on the deviation of basic parameter values of the characteristic was considered. The considerations were illustrated by the numerical example.
An LLCL-filter is becoming more attractive than an LCL-filter as the interface between the grid-tied inverter and the grid due to possibility of reducing the copper and the magnetic materials. The efficiency of the LLCL-filter based single-phase grid-tied inverter also excites interests for many applications. The operation of the switches of the VSI is various with different modulation methods, which lead to different efficiencies for such a single-phase grid-tied inverter system, and therefore important research has been carried out on the effect of the choice of PWM schemes. Then power losses and efficiencies of the LLCL-filter and the LCL-filter based single-phase grid-tied inverters are analyzed and compared under the discontinuous unipolar, the dual-buck and the bipolar modulations. Results show that the efficiency of LLCL-filter based inverter system is higher than the LCL- filter based independent on the modulation method adopted. Experiments on a 2 kW prototype are in good agreement with results of the theoretical analysis.
Bogusław Wolniewicz, inspired by his formal ontology of situations, has put forward a question on semilattices with a unit (A question about joinsemilattices, Bulletin of the Section of Logic 19/3, 1990). The present paper is entirely devoted to this problem in the formulation given by Wolniewicz. First, the meaning of the question is analyzed and its lattice-theoretical and Boolean algebraic contents are exhibited. Second, set-theoretical and topological counterparts of the question are formulated and commented upon.
The paper presents a concept of a control system for a high-frequency three-phase PWM grid-tied converter (3x400 V / 50 Hz) that performs functions of a 10-kW DC power supply with voltage range of 600÷800 V and of a reactive power compensator. Simulation tests (in PLECS) allowed proper selection of semiconductor switches between fast IGBTs and silicon carbide MOSFETs. As the main criterion minimum amount of power losses in semiconductor devices was adopted. Switching frequency of at least 40 kHz was used with the aim of minimizing size of passive filters (chokes, capacitors) both on the AC side and on the DC side. Simulation results have been confirmed in experimental studies of the PWM converter, the power factor of which (inductive and capacitive) could be regulated in range from 0.7 to 1.0 with THDi of line currents below 5% and energy efficiency of approximately 98.5%. The control system was implemented in Texas Instruments TMS320F28377S microcontroller.
Reliable estimation of longitudinal force and sideslip angle is essential for vehicle stability and active safety control. This paper presents a novel longitudinal force and sideslip angle estimation method for four-wheel independent-drive electric vehicles in which the cascaded multi-Kalman filters are applied. Also, a modified tire model is proposed to improve the accuracy and reliability of sideslip angle estimation. In the design of longitudinal force observer, considering that the longitudinal force is the unknown input of the electric driving wheel model, an expanded electric driving wheel model is presented and the longitudinal force is obtained by a strong tracking filter. Based on the longitudinal force observer, taking into consideration uncertain interferences of the vehicle dynamic model, a sideslip angle estimation method is designed using the robust Kalman filter and a novel modified tire model is proposed to correct the original tire model using the estimation results of longitudinal tire forces. Simulations and experiments were carried out, and effectiveness of the proposed estimation method was verified.
In this paper, a filtering stage based on employing a Savitzky-Golay (SG) filter is proposed to be used in the spectrum sensing phase of a Cognitive Radio (CR) communication paradigm for Vehicular Dynamic Spectrum Access (VDSA). It is used to smooth the acquired spectra, which constitute the input for a spectrum sensing algorithm. The sensing phase is necessary, since VDSA is based on an opportunistic approach to the spectral resource, and the opportunities are represented by the user-free spectrum zones, to be detected through the sensing phase. Each filter typology presents peculiarities in terms of its computational cost, de-noising ability and signal shape reconstruction. The SG filtering properties are compared with those of the linear Moving Average (MA) filter, widely used in the CR framework. Important improvements are proposed.
In this paper we introduce a self-tuning Kalman filter for fast time-domain amplitude estimation of noisy harmonic signals with non-stationary amplitude and harmonic distortion, which is the problem of a contactvoltage measurement to which we apply the proposed method. The research method is based on the self-tuning of the Kalman filter's dropping-off behavior. The optimal performance (in terms of accuracy and fast response) is achieved by detecting the jump of the amplitude based on statistical tests of the innovation vector of the Kalman filter and reacting to this jump by adjusting the values of the covariance matrix of the state vector. The method's optimal configuration of the parameters was chosen using a statistical power analysis. Experimental results show that the proposed method outperforms competing methods in terms of speed and accuracy of the jump detection and amplitude estimation.
The accuracy and reliability of Kalman filter are easily affected by the gross errors in observations. Although robust Kalman filter based on equivalent weight function models can reduce the impact of gross errors on filtering results, the conventional equivalent weight function models are more suitable for the observations with the same noise level. For Precise Point Positioning (PPP) with multiple types of observations that have different measuring accuracy and noise levels, the filtering results obtained with conventional robust equivalent weight function models are not the best ones. For this problem, a classification robust equivalent weight function model based on the t-inspection statistics is proposed, which has better performance than the conventional equivalent weight function models in the case of no more than one gross error in a certain type of observations. However, in the case of multiple gross errors in a certain type of observations, the performance of the conventional robust Kalman filter based on the two kinds of equivalent weight function models are barely satisfactory due to the interaction between gross errors. To address this problem, an improved classification robust Kalman filtering method is further proposed in this paper. To verify and evaluate the performance of the proposed method, simulation tests were carried out based on the GPS/BDS data and their results were compared with those obtained with the conventional robust Kalman filtering method. The results show that the improved classification robust Kalman filtering method can effectively reduce the impact of multiple gross errors on the positioning results and significantly improve the positioning accuracy and reliability of PPP.
The paper presents a concept of an active filter with energy storage. This solution can be used for the compensation of momentary one phase high power loads with discontinued power consumption (e.g. spot welding machines). Apart from the typical filtering capabilities, the system’s task is also the continuity of the input power from the feeder line and limiting its fluctuation. The proposed by the author’s solution can produce measurable economic benefits by reducing the rated power necessary to energize periodically operating loads and improving the indicators of electrical energy quality. The developed method of active power surges compensation enables a flexible approach to requirements concerning the rated power of the point to which the periodically operating loads with high peak current value are connected. The tests were conducted on a simulation model specially developed in Matlab & Simulink environment, proving high effectiveness of the presented solution.
The considered shunt active power filter can be controlled not only to compensate non-active current in the supply source, but additionally to optimize energy flow between the source and the load. In such a case the filter shapes the source current to be active and simultaneously regulates its magnitude. The presented filter/buffer can operate properly even when the load contains AC or DC variable energy source of any characteristic. The device can optimize energy flow for a single load, but also for a group of loads as well. The distinctive feature of the employed control method of the filter/buffer is that certain changes of energy stored in the device are utilized as the source of information concerning the active current of the load. This control method is very flexible and can be implemented to nearly all structures of active filters, for DC, single- and multiphase circuits.
This paper presents a Kalman filter based method for diagnosing both parametric and catastrophic faults in analog circuits. Two major innovations are presented, i.e., the Kalman filter based technique, which can significantly improve the efficiency of diagnosing a fault through an iterative structure, and the Shannon entropy to mitigate the influence of component tolerance. Both these concepts help to achieve higher performance and lower testing cost while maintaining the circuit.s functionality. Our simulations demonstrate that using the Kalman filter based technique leads to good results of fault detection and fault location of analog circuits. Meanwhile, the parasitics, as a result of enhancing accessibility by adding test points, are reduced to minimum, that is, the data used for diagnosis is directly obtained from the system primary output pins in our method. The simulations also show that decision boundaries among faulty circuits have small variations over a wide range of noise-immunity requirements. In addition, experimental results show that the proposed method is superior to the test method based on the subband decomposition combined with coherence function, arisen recently.
A hybrid method is presented for the integration of low-, mid-, and high-frequency driver filters in loud-speaker crossovers. The Pascal matrix is exploited to calculate denominators; the locations of minimum values in frequency magnitude responses are associated with the forms of numerators; the maximum values are used to compute gain factors. The forms of the resulting filters are based on the physical meanings of low-pass, band-pass, and high-pass filters, an intuitive idea which is easy to be understood. Moreover, each coefficient is believed to be simply calculated, an advantage which keeps the software-implemented crossover running smoothly even if crossover frequencies are being changed in real time. This characteristic allows users to efficiently adjust the bandwidths of the driver filters by subjective listening tests if objective measurements of loudspeaker parameters are unavailable. Instead of designing separate structures for a low-, mid-, and high-frequency driver filter, by using the proposed techniques we can implement one structure which merges three types of digital filters. Not only does the integration architecture operate with low computational cost, but its size is also compact. Design examples are included to illustrate the effectiveness of the presented methodology