In this paper, an adaptive multiresolution speech enhancement algorithm based on wavelet transform is put forward. It can make adaptive filtering to noise speech both at scales and among scales. So that the noise parts during the frequency intervals which decrease hearing quality mostly are reduced efficiently. Both the SNR and subject hearing quality of denoised speech are high and good.

In this paper, A decoupled parameter estimation(DPE) algorithm for estimating parameters of the harmonic-group signal in colored noise is researched. In this algorithm, a "iterated harmonic-group clean algorithm" is presented to obtain the estimates of the parameters of the harmonic signal based on FFT. Then, the harmogram is used to decide the number of the harmonic signal. This algorithm not only is computationally simple and can provide supperresolution frequency estimation, but also is robust to mismodeling errors.

An algorithm for simultaneous estimation of two dimentional angle and polarization of multiple signals is proposed in this paper. Using an L-shaped array consisting of crossed dipoles, the algorithm exploits rotational invariance properties involved in array structure, so that both angle and polarization estimates may be computed via matrix pencil method, and these estimates are automatically paired. Simulation results are presented to verify the efficacy of the proposed algorithm.

A multiscale analysis method for precisely estimating the time of arrival (TOA) of radar return is proposed in this paper. Through synthetically analyzing the characteristics of echoes in different scales, our method not only can detect the radar return and measure its TOA precisely but also can cancel the multipath interferences correctly and reject noise strongly. Only a three-scale wavelet transform (WT) is needed for every receiving signal and a two-scale WT is performed on each candidate radar returns in this method. Our algorithm is more computationally efficient by using the famous Mallat algorithm. The experimental results in this paper have verified the effectiveness of our method.

A novel scheme of motion estimation for video sequences, spatial-temporal correlated multiresolution algorithm, is proposed in this paper. This algorithm employs the correlation presented in motion vector field, including spatial-temporal correlation and father-children correlation from hierarchical blocks, to reduce calculating complexity. Motion compensated gain and entropy of motion vector can be improved evidently, but the number of search is less then that of three step method.

Based on analysis of the "stop and go" decision-directed (DD) algorithm, a method of selecting (3 is proposed, which guarantees the quick convergent characteristic of the "stop and go" DD algorithm. In order to improve the convergent speed of "stop and go" DD algorithm, the momentum "stop and go" DD algorithm is proposed and then based on this, an adaptive momentum "stop and go" DD algorithm is also proposed. Computer simulations show that the proposed techniques are very effective for the "stop and go" DD algorithm.

Estimating the arrival time of overlapping echoes by high resolution array processinh algorithms can efficiently expand the ability of target resolution in radar.Lowering the SNR trhreshold at which the algorithms take effect has particularly been an area of focus for practical applications.In this paper a novel method based on pulse comprression for range super-resolution is presented.The proposed method,especially suitable to large time-bandwidth product (TB) signals,can be applied at very low SNR and greatly reduce the computational load in comparison with the known approaches.

A family of binary QF sequences are constructed and their periods, size of the family, correlation functions and linear span are given, which shows that C_max= ^qn/2+1+1,L_max= m(3n/2)^m-1 when n is even, C_max= q^n/2+3/2+1,L_max= 2^m-1mn^m-1 when n is odd.

For both the periodic case and aperiodic case, the relations between the run numbers of a binary sequence and autocorrelation, cross-correlation functions of sequences are investigated.And the correlation functions expressed by the runs of a binary sequence, which are refered to as RCF,are derived in this paper.

This paper proposes a new media-access protocol for multichannel ring networks. The protocol adopts reservation schemes and provides each node with two channels. The mathematical analysis and computer simulation show that the protocol has good properties.

It is important to design survivability for a broadband communication network. Network survivability is affected by many factors including routing. The effect of splitting factors of diverse routing is analyzed and the particular advantages of the even splitting factor is discovered. This research result provides a reasonable consideration to network survivability design.

In mmW radar system, SNR affect heavily on radar range. To lower the SNR demand of target recognition, a new approach of feature extraction based on the bispectrum technique is presented. The feature vectors defined from bispectrum have high stability even when SNR is very low. Thus, the sensitivity and range of mmW radar are improved.

A method finding the global optimum value of misalignment is presented.It is used to increase the accuracy of range alignment.There exist the drift and jump erros when use the classical amplitude correlation method to aling the envelope of the echoes.Sometimes,these errors are so great that they are as large as the sidth of several range bins.So the quality of image is badly affected.By usign global optimum method,the range alignment is more accurate,teh drift and jump phenomena of echo envelope are decreased and the quality of the image is improved.Form the image result of real measured data,it can be seen that this method is much better than the classical method.

This paper,at first ,expounds the relation between the singular singal and the modular maximum for wavelet transform,and the different features of singal and noise in wavelet transform.Then,in the light of high-resolution radar target one dimensional range profiles,it is decomposed in multi-resolution and the multi-resolution feature description are obtaiend.According to the multi-resolution feature description of radar target,the conception of difference degree for the range profiles is defined.Using the difference degree,the fast matching for target recognition can be obtained.

This paper studies on the speckle reduction in multi-look polarimetric SAR image.A novel method,named the multi-look polarimetric whitening filtering (WPMF) method,is presented,and is further extended to construct a fully polarimetric filter with multi-channel output.The paper also quantifies the amount of speckle reduction achievable by using the MPWF,and compares this method with the span,weighting and power equalization (PE) methods in theory.Experimental results with the NASA/JPL L-band 4-look polarimetric SAR data demonstrate the effectiveness and superority of the MPWF,and show that the MPWF is of great advantage for enhancing SAR image classification.

In this paper,the temporary high resolution radar target range profile sequences were used to identify aircraft targets, and a real time recurrent neural network classification method was proposed based on range profile sequences.The experiments of classification of three kinds of aircraft targets were performed, and demostrated that there were high classification performance using the proposed method.

Some probles about SAR moving targets imaging by joint time-frequency analysis are discussed in this paper.Three kinds of time-frequency transforms--Short Time Fourior Transform(STFT),Wavelet Transform(WT) and Wigner-Ville Distribution (WVD) are compared.It is presented that only WVD could meet the requirement.Moveover,Hough Transform(HT) is employed in restraining the distribing corss terms,rpoduced by the WVD due to its bilinear property.Experiments test that the combined WVD and HT could provide satisfactory detection performance even if the signal-to-noise rato is very low.

A novel complex FIR filter design approach based on Neural Network Optimiza-tion(NNO) technique is proposed in this paper. To demonstrate the feasibility of the NNO design approach, the weight least mean square criterion between the disired frequency response and the designed filter response is defined as the Lyapunov energy function of a continuous Hopfield network, and the network state equations are drived. The implementation of the NNO approach is described together with some design guidelines. A few design examples are given and the advantages of NNO approach over conventional methods are illustrated.

It is very hard to solve the packing problem especially for the three dimensional packing problem. A new evolution strategy is presented in this paper which combines heuristic experience with genetic algerithms, its evalutionary thought is common to whatever dimensional packing problems. Finally, an experiment is given proving the strategy is better.

When performing RCS measurements for outdoor targets, usually the whole space range under measurement is greatly longer than the actual target zone, which causes very large data storage and makes further data processing difficulty. To solve this problem, an effective method is presented in this paper, which can largely reduce measurement data. In this method, a related local oscillation signal is generated to be mixed with the received signal, which can greatly decrease the carrier frequency of the return signal from targets. With this method, a much lower sampling rate is guaranteed, which can greatly reduce the data storage and signal processing time.

A fast direction-of-arrival(DOA) estimation algorithm is proposed that applies to coherent wide-band emitter signals impinging on an uniform linear array.This algorithm,where spatial smoothing technique is employed,is based on obtaining the array mainfold of a desired narrow-band covering the center frequency of the whole wide-band.The fast algorithm,which has super-resolution,leads to the realization of real-time bearing estimation of multiple coherent wide-band soucres.Simulations are presented with comparison to coherent signal-subspace method (CSM).

This paper provides array signal processing super-resolution fast algorithm. This algorithm based on the eigenvalue value shift of covariance matrix and its power iteration. It does not needs the inverse matrix. It converges quickly and only needs several iteration to converge to solution. Its architecture is very simple and easily implemented.

In this paper, the performances of ATM switch architectures based on different buffer strategies and some switch design technologies which can improve switching performance under uniform traffic mode are studied and compared with that of nonuniform traffic mode. On the basis of above analyses, a high-performance switch design is proposed.

The Perfectly Matched Layer(PML)is very effective in absorbing propagating wave, so that it has been successfully applied in FDTD research. Katz(1994) derived 12 differential equations in three-dimensional PML, but there is no conclusion for the propagation characters of plane wave in PML such as wave expression ,wave number and wave impedance. These characters are investigated in this paper.

The device structure and technical processings of quasi-planar self-aligned silicon avalanche electron emission array are introduced. The processing step at the edge of electron emission region is about 100 nm only and the width of self-aligned current channel of shallow As implantation is about 3 μm. Its I-V characteristics show a larger linear region and lower series resistance than that of the previous silicon avalanche electron emission devices. Some of the electron emission characteristics are also discussed in the paper.