We consider a wireless relay network where two pairs of nodes exchange information with their partners through a single amplify-and-forward two-way relay with each node equipped with multiple antennas. We propose a new relaying scheme employing OFDMA for the multiple access transmission in the first time slot and OFDM/SDMA for the broadcast transmission in the second time slot to improve the sum rate of the network. To fully utilize spatial diversity, we design the relay beamforming matrices according to two methods respectively, i.e., signal to leakage and noise ratio (SLNR) and block diagonalization based zero-forcing (BDZF), on per subcarrier basis. We also derive the upper bound on the capacity region of this two-way relay network by using cut-set theory. Simulation results show that the proposed scheme outperforms three other relaying schemes in terms of sum rate and can approach the upper bound of capacity region.

Automatic facial expression recognition is the kernel part of emotional information processing. This paper dedicates to develop an automatic facial expression recognition approach based on confusion-crossed support vector machine tree (CSVMT) to improve recognition accuracy and robustness. After the pseudo-Zernike moment features were extracted, they were used to train a CSVMT for automatic recognition. The structure of CSVMT enables the model to divide the facial recognition problem into sub-problems according to the teacher signals, so that it can solve the sub-problems in decreased complexity in different tree levels. In the training phase, those sub-samples assigned to two internal sibling nodes perform decreasing confusion cross, thus, the generalization ability of CSVMT for recognition of facial expression is enhanced. The compared results on Cohn-Kanade facial expression database also show that the proposed approach appeared higher recognition accuracy and robustness than other approaches.

We propose a subspace-based blind channel estimation method for space-time coded OFDM system. Using only the redundancy induced by OFDM modulation and space-time block coding (STBC), channel state information (CSI) can be blindly estimated up to two scalar ambiguities, even with single receiving antenna. Compared with other blind channel estimation method for space time OFDM systems, this method needs neither pre-coding nor over-sampling, and thus has higher system data rate and lower complexity.

Neural network tree (NNTree) is a modular neural network with the overall structure being a decision tree (DT), and each non-terminal node being an expert neural network (ENN). One advantage of using NNTrees is that they are actually gray-boxes because they can be interpreted easily if the number of inputs for each ENN is limited. To design interpretable NNTrees, we have proposed a multiple objective optimization based genetic algorithm. This algorithm, however, is good only for solving problems with binary inputs. In this paper, we propose a method to solve problems with continuous inputs. The basic idea is to find a small number of critical points for each continuous input using self-organized learning, and quantize the input using the critical points. Experimental results with several public databases show that the NNTrees built from the quantized data are much more interpretable, and in most cases they are as good as those obtained from the original data.

Synchronization of neural networks by mutual learning has been demonstrated to be possible for constructing key exchange protocol over public channel. However, the neural cryptography schemes presented so far are not the securest under regular flipping attack (RFA) and are completely insecure under majority flipping attack (MFA). We propose a scheme by splitting the mutual information and the training process to improve the security of neural cryptosystem against flipping attacks. Both analytical and simulation results show that the success probability of RFA on the proposed scheme can be decreased to the level of brute force attack (BFA) and the success probability of MFA still decays exponentially with the weights' level L. The synchronization time of the parties also remains polynomial with L. Moreover, we analyze the security under an advanced flipping attack.

Quasi-entropy (QE), a class of new independence measures, is proposed. By optimizing QE, blind separation of signals with arbitrary continuous distributions is achieved. Simulations verify these results.

Evolutionary programming has been applied to many optimization problems. However, on some function optimization problems its convergence rate is slow. In this paper, swarm directions are embedded in fast evolutionary programming. The swarm direction for an individual supplies its place to be mutated. The experimental results show its effectiveness and efficiency.

A novel method of blind image restoration based on sparse code shrinkage and ICA algorithms is presented in this paper. In the Sparse Code Shrinkage method by Hyvarinen, ICA-like algorithm is used to remove gaussian noise in natural images. And we can also assume that the pixels in blocks of images are independent nongaussian random variables. Thus an ICA algorithm can be applied to do deblurring. The method is tested to be useful in both synthesized images and natural images.

The population-based incremental learning (PBIL) algorithm is extended to a form where multiple traits for each gene reflect the pleiotropic and polygenic characteristics in natural evolved systems. This method is used to solve the traveling salesman problem. Some results are better than the best existing algorithms for evolutionary computation of the problem. The results show that the method proposed is comparable to the advanced level of solvers for the traveling salesman problem.

Statistical complexity measures are proposed as general indicators of structure or correlation. Lopez-Ruiz (1995) introduced another measure of statistical complexity C/sub LMC/ that, like others, satisfies the boundary conditions of vanishing in the extreme ordered and disordered limits. Feldman examined some properties of C/sub LMC/ and found that it is neither an intensive nor extensive thermodynamic variable, and proposed a simple alteration of C/sub LMC/ that renders it extensive. However the remedy results in a quantity that is a trivial function of the entropy density and hence of no use as a measure of structure or memory. We alter the disequilibrium term by the information of time irreversibility (information of nonlinear dynamics) and present a novel statistical complexity measure which is used to quantify the complexity caused by nonlinear dynamics. This statistical complexity measure allows reliable detection of periodic, quasi-periodic, linear stochastic and chaotic dynamics. When applied to the analysis of heart rate data, correlational structure in heart rate series is found, and the estimated statistical complexity appears to be correlated with different cardiac dynamics.

In this paper binary population-based incremental learning is extended to an integer form, and a new approach to the traveling salesman problem (TSP) is proposed based on linkage relations between cities. The properties of this method are distributed stochastic tour construction, probability distribution initialization, accelerated search based on some power of probability, decision of entropy of probability distribution for terminate condition on process of evolution, and improvement of population solutions using 2-opt/3-opt. Thirteen TSP problems are solved, including ten international contest problems on symmetric and asymmetric TSP problems. The results show that the method proposed in this paper is comparable to the international advanced level on TSP problems and is capable of finding high quality solutions to TSP problems in short times, particularly for large TSP instances.

Many blind beamforming algorithms, such as C-CAB, use the signal characteristics to estimate the steering vector. The conventional LCMV algorithm is then adapted to obtain the optimum solution. However, the LCMV-like methods are sensitive to the mismatch. In this paper, the cause of this mismatch is discussed in detail. A robust blind beamforming algorithm is presented. Using a neural network structure the algorithm can decrease the computational complexity and make it possible to realize the method in real time. Results of computer simulations are included to support our analysis.

Chaotic behavior can be found in nonautonomous cellular neural networks (CNNs). The impulsive control method to control this kind of chaos is used and some satisfactory result are achieved. The condition which ensure existence of periodic solution in the impulse controlled system is provided and proved, and numerical simulation shows the chaos can be eliminated by adding a very small external force. Through observing the time waveform diagram of some controlled periodic orbits, we find the time waveform is very similar to encephalic electric activity or cardiac electric activity in biomedical field study. This illustrate that the nonautonomous CNNs model can successfully model physiological electric response activity signals. Moreover, the physiological explanation is given for the nonautonomous CNNs model.

It is by now widely appreciated that normal heart rhythms show features of deterministic chaos, and it is also corrupted by a great deal of physiological noises from surroundings as the inputs to the chaotic system. We study the effects of measurement and dynamical noises on low-dimensional chaotic systems, and present a numerical algorithm for estimating to what extent heart rhythm is corrupted by dynamical noises. Further analysis shows that adding increasing amounts of dynamical noise increasingly develops the complexity of chaotic dynamics. The method need not fully reconstruct an attractor, thus it is robust to measurement noises. Finally, the ECG of 53 subjects falling into three groups were used to test the algorithm. From the measured R-R intervals, a time series of 1000 first order difference points was used to estimate dynamical noise levels. We found that the estimated dynamical noise levels were correlated to the different classes of ECG.

The complex dynamics of the chaotic neural networks makes it possible for them to escape from local minimum of the simple gradient descent neurodynamics. In this letter we use a transiently chaotic neural network to detect the CDMA multiuser signals and hence obtain an implementation scheme of the CDMA multiuser detector (TCNN-MD). Computer simulation results show that the proposed detector is clearly superior to Hopfield neural-network-based detector.

In this paper, the fast implementation of cosine-modulated filter bank (CMFB) is revisited. A class of paraunitary CMFBs with arbitrary length is considered. By further reorganizing of the polyphase component matrix and using the linear-phase property of the prototype filter, we obtain a more efficient implementation structure for the CMFB, in which we use 2/spl times/2 lossless matrices instead of 2/spl times/1 ones. In the new implementation, the number of two-channel lossless lattices is reduced by a factor of two.

An adaptive quantization algorithm for MPEG-2 video coding using a neural network is presented in this paper. The proposed algorithm uses a backpropagation neural network to divide the macroblock activity into one of four categories: flat, edge, texture, fine-texture, and thus the macroblock can be quantized adaptively according to the human vision system (HVS) sensitivity. Experimental results show that this method can reduce block artifacts of flat area and distortion at edge effectively. Meanwhile, the picture subjective quality and objective quality of each frame are improved.

This paper investigates the application of a recurrent wavelet neural network (RWNN) to the blind equalization of nonlinear communication channels. We propose a RWNN based structure and a novel training approach for blind equalization, and we evaluate its performance via computer simulations for a nonlinear communication channel model. It is shown that the RWNN blind equalizer performs much better than the linear CMA and the RRBF blind equalizers in the nonlinear channel case. The small size and high performance of the RWNN equalizer makes it suitable for high speed channel blind equalization.

The bit-number allocation strategy of Test Model 5 (TM5) may cause the coding quality to decrease and the buffer to overflow at scene changes, therefore a new algorithm on MPEG-2 target bit-number allocation is proposed in this letter. Here the influence of scene changes on coding quality is analyzed and an approach to improve the coding performance at scene changes is presented to avoid a sharp quality decrease of scene-change frames. Simulation results indicate that this algorithm provides a promising coding quality at scene changes.

This paper presents a decision feedback neural network (NN) coherent receiver scheme for continuous phase modulation based on frequency domain. Through decision feedback pre-processing, the effect of the previous transmitted symbols can be removed for the present symbol decision. By employing a Karhunen-Loeve transform (KLT) or discrete cosine transform (DCT), the input data number of neural networks can be reduced significantly. To obtain more sufficient convergence of neural networks a modified "delta-bar-delta" BP learning algorithm is proposed. Despite the low complexity in NN training and implementation, computer simulation results show that our NN receivers can achieve near optimal demodulation performance.

Estimation of the trajectory of a target from a passive sonar's bearings and frequency measurements in the presence of multivariate normally distributed noise, with unknown inhomogeneous general covariance, is modelled as a nonlinear multiresponse parameter estimation problem. It is shown that maximum likelihood estimation in this case is identical to optimizing a determinant criterion which has a concise form and contains no elements of unknown covariance matrix. An effective Gauss-Newton type algorithm, using only the first-order derivatives of the model function, is presented to implement such estimation. The simulation shows that the proposed approach is superior to the traditional estimation methods especially under the condition of strong inhomogeneity of noise covariance and high correlation between different types of measurement noises.

A modified fractal transform (MFT) is presented. In the function part of the MFT, the conventional greyscale function of an image block is replaced by the greyscale function of an error image block whose mean is removed. This fractal transform is used to approximate an image which is to be encoded. The simulation results show that with the MFT the image decoding process is very fast, typically only 1 to 3 iterations are required to reconstruct the image while the quality of reconstructed image remains high.

In this paper, we addressed a theoretical analysis for the capacity of parallel intra-connected bidirectional associative memory (MIBAM) and proved the conclusions: two MIBAM with the equal total number of neurons have the equal recalling probability for m pairs of stored pattern pairs if m is not too large. The results of computer simulation support the conclusions well.

This paper presents a subband filter bank model of N subbands for orthogonally multiplexed QAM system with 2N subchannels. Then, a fast implementation algorithm using polyphase decomposition and discrete cosine transform (DCT) is derived from this model. Compared with the implementation method through discrete Fourier transform, only half the computation cost is needed. Finally, we give the adaptive equalization scheme based on subband adaptive filter theory, and its efficiency is verified by experiments.

By computer simulation, we detect a new type of strange attractor in a three-cell cellular neural network which defers from that found by other researchers. Bifurcation phenomena are analyzed.

In this paper, two codebook search methods and a coding scheme are proposed for fast and low rate vector quantization using the self-organizing feature maps (SOFM). Based on the topology preservation property of the SOFM, the search methods use the distance between adjacent input vectors to guide the codebook search process and to determine searching sequence of codevectors. The novel coding scheme, which can be considered as a vector version of delta modulation, eliminates the correlation buried in the source sequence and hence reduces the rate. Simulation results demonstrate the effectiveness of proposed methods and better performances than those obtained previously.

In this paper, a new competitive learning algorithm based on the partial distortion theorem is proposed for the on-line vector quantizer design. The novel algorithm is called partial-distortion-equivalent competitive learning (PDECL) algorithm, which aims at making the partial distortions for each neuron (code-vector) be uniform to overcome the neuron underuse problem as well as to minimize the average distortion for the designed vector quantizer. Compared with the Kohonen learning algorithm (KLA), the frequency-sensitive competitive learning (FSCL) algorithm and the soft competition scheme (SCS) algorithm, the PDECL consistently shows the better performance than all of them and the LBG algorithm for the design of vector quantizers with different codebook sizes especially when the codebook size is large enough.>

Neural-network-based algorithms for multipath time delay estimation are presented. Two cases were considered: (1) the echo is a pulselike signal; and (2) the echo is a nonGaussian broadband random process. The algorithms estimate the time delays based on a neural network by relating the Lyapunov energy function to the cost function of time delay estimation. These methods have several advantages over the conventional algebraic methods. They are less time consuming, and the echo number detection and time delay estimation are performed simultaneously.>

A multilayer neural network used for nonlinear predictive image coding is described. Two coding schemes, nonadaptive and adaptive, are shown. Owing to matching of the local properties of the image, nonlinear predictive coding gives a better performance than linear predictive coding. A series of computer experiments shows the method has not only the ability to generalize but also noise reduction capabilities. Compared with differential pulse code modulation (DPCM), it greatly reduces the number of bits to be transmitted.>

A method is presented for the design of 3-D recursive digital filters with approximately spherically symmetric magnitude frequency response. The desired 3-D recursive filter is obtained by cascading a number of 3-D transformed filters which can be designed by rotating a 1-D analog filter and using the triple bilinear z-transformation. The stability of the 3-D filters is discussed in detail, and a theorem which gives the sufficient conditions for stable 3-D filters is derived.>

A method is presented for integrating the problems of matching and motion estimation. A general principle of consistent motion for such a fusion is formulated. Using this principle, the rigidity constraint is incorporated into the correspondence process. A fusion algorithm for point matching is presented. The overdetermination of essential motion equations is utilized directly to give a matching error measurement of any hypothesized correspondence, so that the correspondence and motion estimation are constructed as a single combinatorial optimization problem. A coarse-to-fine strategy for point-matching is also put forward. The method is especially applicable to cases where the set of possible correspondence has been found. Examples show the effectiveness of the proposed method.>

The authors present a novel subband coding method, directional subband coding of images. In order to match human visual sensitivity, an image is decomposed into a low-frequency component and many high-frequency directional components using a separable quadrature mirror filter and a directional filter. Effective coding methods are then used to encode the component images. The method is computationally simple and avoids potentially difficult design problems. Computer simulations show that the method performs well.>

The authors present a unified camera calibration method in which radial distortions can be considered or ignored. The usually tangled process of calibrating intrinsic and extrinsic camera parameters in the zero-distortion case is separated by using vanishing points as calibration points. By ignoring distortions, rough estimates for intrinsic camera parameters are solved linearly. Refinements are made on the basis of the invariance properties of the cross ratio and the collinearity of perspectivity, so that radial distortion can be considered. With the corrected image plane, the usual linear methods can be used to solve extrinsic camera parameters with low computational cost.>

An important relationship is established between the Gram-Schmidt orthogonal procedure and least-squares solutions of the overdetermined modified Yule-Walker (OMYW) equations of order from 0 to p, where p is the largest possible value of the model order. Based on this relationship, the authors present an order-recursive approach to harmonic retrieval in the presence of white noise interference which is shown to offer significantly improved numerical features compared to the commonly used generalized-inverse method. The performance of the present approach is close to the SVD (singular-value decomposition) method, but the computational complexity savings are rather considerable. The detailed simulation results are comparable to other well-known methods.>

An efficient blockwise algorithm, named the block sequential least-squares (BSLS) algorithm, is presented for sequentially solving exact LS problems in real time. The information is carried from block to block by iterating some correlation vectors. The average operations of the BSLS are fewer than sample-by-sample fast RLS algorithms. The BSLS algorithm allows efficient use of the FFT technique to make gains in computational complexity savings. It is shown that the new algorithm can provide improved numerical features over the existing fast-RLS algorithms due to the new adaptation version.>

Three fast, recursive, in-order, least-squares (LS) algorithms are presented to estimate efficiently the two-dimensional simultaneous autoregressive (SAR) model parameters and to determine the model orders at the same time. The algorithms are for a causal model with a nonsymmetric half plane neighborset, a semicausal model with a symmetric half plane neighborset, and a noncausal model with the corresponding neighborset being a rectangle plane centered at

An edge-based, fast and effective stereo matching method is proposed, which is characterized by two matching steps: initial and directed matching. In the initial matching a standard epipolar constraint and local geometrical and area constraints are applied among edge segments to reduce the search space. Then the directed matching proceeds under the guidance of an edge-connectivity relational graph which represents the structural relationship among edge segments. By using a heuristic labelling technique the combinational search is greatly reduced in the matching process. The implementation and results are presented to show the efficacy of the proposed stereo matching method.>

A method of acquiring 3-D data of an object via binocular technique is presented in this paper. It attempts to improve the accuracy of 3-D measurement of the passive technique and the main feature of this method are: 1) accurate (subpixel) location of the noisy image edge segments by parameterization; 2) using generalized cones for 3-D measurement with an accuracy beyond resolution limit of the digital sensor.

Adaptive maximum entropy spectrum estimator (AMESE) can track variation of frequency. Because of the variation of frequency, it's necessary to discuss average detection performance of AMESE. In this paper, we discuss average detection performance The discussion gives a theoretical basis for designing AMESE that can achieve given false probability and detection probability.

In this paper, we propose a new algorithm for maximum entropy reconstruction of the original image from a part of its noise corrupted linear measurement. The algorithm we propose here is a iterative way which uses the fast Fourier transform (FFT) so that it consumes less computer time than others do.

The techniques for harmonic analysis in colored noise have attracted much more attentions in the last two years. This paper presents a novel approach, namely modified Yule -Walker equations method, to harmonic analysis in unknown colored noise, based on the fact that an ARMA model can be fitted to the time sequence consisting of harmonics plus a MA noise process. It shows that the present method can do without any prior knowledge about the noise model. New fixed-order fast algorithms, including both nonadaptive and adaptive situations, are presented for efficient solution of MYW equations. The performance of ours, FFT and maximum entropy are illustrated by simulation results.