A new gradient type adaptive algorithm is proposed in this paper. It is formulated based on the least squares criteria while the conventional gradient algorithms are based on the least mean square criteria. The proposed algorithm has two variable parameters and by changing them we can adjust the characteristic of the algorithm from the RLS to the LMS depending on the environment. This capability of adjustment achieves the possibility of providing better solutions. However, not only it provides better solutions than the conventional algorithms under some conditions but also it provides a very interesting theoretical view point. It provides a unified view point of the adaptive algorithms including the conventional ones, i.e., the LMS or the RLS, as limited cases and it enables us to analyze the bounds for those algorithms.

An alias free parallel structure for adaptive digital filters (ADF's) is considered. The method utilizes the properties of the Frequency-Sampling Filter (FSF) banks to obtain alias free points in the frequency domain. We propose a new cost function for parallel ADF's. The limiting value analysis of system identification using proposed cost function is given in stochastic sense. It is also shown by simulation examples that we can carry out precise system identification. The cost function is defined in each bin; accordingly, it enables the parallel processing of ADF's.

This paper proposes a new subband adaptive filtering algorithm for adaptive FIR filters. The number of taps for each subband filter is adaptively controlled based on a sum of the absolute coefficients or the coefficient power in conjunction with the subband signal power. Keeping the total number of taps constant, redundant taps are redistributed to subbands where the number of taps is insufficient. Simulation results with a white signal show that the number of taps in each subband approaches an optimum as each subband filter converges. For a colored signal, tap assignment by the new algorithm is as stable as for a white signal.

For system identification problems, such as noise and echo cancellation, FIR adaptive filters are mainly used for their simple adaptation and numerical stability. When the unknown system is a high-Q resonant system, having a very long impulse response, IIR adaptive filters are more efficient for reduction in the order of a transfer function. One way to realize the IIR adaptive filter is a separate form, in which the numerator and the denominator are separately realized and adjusted. In the actual applications, the order of the unknown system is not known. In this case, it is very important to estimate the total order and the order assignment on the numerator and the denominator. In this paper, effects of the order estimation error on the residual error are investigated. In this form, indirect error evaluation called "equation error" is used. Through theoretical and numerical investigation, the following results are obtained. First, under estimation of the order of the denominator causes large degradation. Second, over estimation can improve the performance. However, this improvement is saturated to some extent due to cancellation of the redundant poles and zeros. Third, the system identification error is proportional to the equation error as the adaptive filter approaching the optimum. Finally, there is possibility of recovering from the unstable state as the order assignment approaches to the optimum in an adaptive process using the equation error. Computer solutions are provided to aid in gaining insight of the order assignment and stability problem.

Contextual classification of multispectral image data in remote sensing is discussed and concretely two improved contextual classifiers are proposed. The first is the extended adaptive classifier which partitions an image successively into homogeneously distributed square regions and applies a collective classification decision to each region. The second is the accelerated probabilistic relaxation which updates a classification result fast by adopting a pixelwise stopping rule. The evaluation experiment with a pseudo LANDSAT multispectral image shows that the proposed methods give higher classification accuracies than the compound decision method known as a standard contextual classifier.

An unsupervised segmentation technique is presented that is based on a layered statistical model for both region shapes and the region internal texture signals. While the image partition is modelled as a sample of a Gibbs/Markov random field, the texture inside each image segment is described using functional approximation. The segmentation and the unknown parameters are estimated through iterative optimization of an MAP objective function. The obtained tesults are subjectively agreeable and well suited for the requirements of region-oriented transform image coding.

We present a method of extracting the digital inphase (I) and quadrature (Q) components from oversampled bandpass signals using narrow-band bandpass Hilbert transformers. Down-conversion of the digitized IF signals to baseband and reduction of the quantization noise are accomplished by the multistage decimator with the complex coefficient bandpass digital filters (BPFs), which construct the bandpass Hilbert transformers. Most of the complex coefficient BPFs in the multistage decimator can be replaced with the lowpass filters (LPFs) under some conditions, which reduces computational burden. We evaluate the signal to quantization noise ratio of the I and Q components for the sinusoidal input by computer simulation. Simulation results show that the equivalent amplitude resolution of the I and Q components can be increased by 3 bits in comparison with non-oversampling case.

The relation between computing part of the FFT spectrum and the so-called generalized FFT (GFFT) is clarified, leading to a new algorithm for performing partial FFTs. The method can be applied when only part of the output is required or when the input data sequence contains many zeros. Such cases arize for example in decimation and interpolation and also in computing linear convolutions. The technique consists of decomposing the DFT into several generalized DFTs. Efficient algorithms for these generalized DFTs exist. The computational complexity of the new approach is roughly equal to the complexity of previous techniques, but the structure is superior, because only one type of butterfly is used and a few lines of code are sufficient. The theoretical properties of the GDFT are given. The case of multidimensional signals, defined on arbitrary sampling lattices is also considered.

Kalman filter is an essential tool in signal processing, modern control and communications. The filter estimates the states of a given system from noisy measurements, using a mean-square error criterion. Although Kalman filter has been shown to be very versatile, it has always been computationally intensive since a great number of matrix computations must be performed at each iteration. Thus the exploitation of this technique in broadband real time applications is restricted. The solution to these limitations appears to be in VLSI (very large scale integration) architectures for the parallel processing of data, in the form of systolic architectures. Systolic arrays are networks of simple processing cells connected only to their nearest neighbors. Each cell consists of some simple logic and has a small amount of local memory. Overall data flows through the array are synchronously controlled by a single main clock pulse. In parallel with the development of Kalman filter, the square root covariance and the square root information methods have been studied in the past. These square root methods are reported to be more accurate, stable and efficient than the original algorithm presented by Kalman. However it is known that standard SRIF is less efficient than the other algorithms, simply because standard SRIF has additional matrix inversion computation and matrix multiplication which are difficult to implement in terms of speed and accuracy. To solve this problem, we use the modified Faddeeva algorithm in computing matrix inversion and matrix multiplication. The proposed algorithm avoids the direct matrix inversion computation and matrix multiplication, and performs these matrix manipulations by Gauss elimination. To evaluate the proposed method, we constructed an efficient systolic architecture for standard SRIF using the COMPASS design tools. Actual VLSI design and its simulation are done on the circuits of four type processors that perform Gauss elimination and the modified Givens rotation.

The classical supervised learning algorithms for optimizing multi-layered feedforward neural networks, such at the original back-propagation algorithm, suffer from several weaknesses. First, they have the possibility of being trapped at local minima during learning, which may lead to failure in finding the global optimal solution. Second, the convergence rate is typically too slow even if the learning can be achieved. This paper introduces a new learning algorithm which employs a genetic-type search during the learning phase of back-propagation algorithm so that the above problems can be overcome. The basic idea is to evolve the network weights in a controlled manner so as to jump to the regions of smaller mean squared error whenever the back-propagation stops at a local minimum. By this, the local minima can always be escaped and a much faster learning with global optimal solution can be achieved. A mathematical framework on the weight evolution of the new algorithm in also presented in this paper, which gives a careful analysis on the requirements of weight evolution (or perturbation) during learning in order to achieve a better error performance in the weights between different hidden layers. Simulation results on three typical problems including XOR, 3-bit parity and the counting problem are described to illustrate the fast learning behaviour and the global search capability of the new algorithm in improving the performance of back-propagated network.

In this paper we introduce the Piecewise Linear Radial Basis Function Model (PWL-RBFM), a new nonlinear model that uses the well known RBF framework to build a PWL functional approximation by combining an l₁ norm with a linear RBF function. A smooth generalization of the PWL-RBF is proposed: it is obtained by substituting the modulus function with the logistic function. These models are applied to several time series prediction tasks.

A method for evaluating the degradation of subband adaptive digital filters (ADF) is presented. The performance of a simple ADF that uses critical sampling is mainly influenced by the subband filter bank's characteristics and the finite precision arithmetic operations used. This paper considers a two-channel mirror filter bank and a normalized least mean square algorithm with floating point arithmetic. The theoretical ERLE (Echo Return Loss Enhancement) and the theoretical relationships between the output error of the ADF and the circuit parameters considering finite precision A/D conversion and finite word length effects in floating point arithmetic operation are obtained using an equivalent noise model. Simulation results are found to be in good agreement to analytical values; the difference is only 3 to 5 dB.

A new steepest descent linear adaptive algorithm, called the proportion-sign algorithm (PSA), is introduced and its performance analysis is presented when the signals are from zero-mean jointly stationary Gaussian processes. The PSA improves the convergence speed over the least mean square (LMS) algorithm without overly degrading the steady-state error performance and has the robustness to impulsive interference occurring in the desired response by adding a minimal amount of computational complexity. Computer simulations are presented that show these advantages of the PSA over the LMS algorithm and demonstrate a close match between theoretical and empirical results to verify our analysis.

For a complex object model, a form of range restriction called specialization constraint (SC), has been proposed, which is associated not only with the properties themselves but also with property value paths. The domain and range of an SC, however, were limited to single classes. In this paper, SCs are generalized to have sets of classes as their domains and ranges. Let Σ be a set of SCs, where each SC in Σ has a set of classes as its domain and a non-empty set of classes as its range. It is proved that an SC is a logical consequence of Σ if and only if it is a finite logical consequence of Σ. Then a sound and complete axiomatization for SCs is presented. Finally, a polynomial-time algorithm is given, which decides whether or not an SC is a logical consequence of Σ.

We developed a simple, practical, adaptive data compression algorithm of the LZ78 class. According to the Lempel-Ziv greedy parsing, a string boundary is not related to the statistical history modeled by finite-state sources. We have already reported an algorithm classifying data into subdictionaries (CSD), which uses multiple subdictionaries and conditions the current string by using the previous one to obtain a higher compression ratio. In this paper, we present a practical implementation of this method suitable for any kinds of data, and show that CSD is more efficient than the LZC which is the method used by the program compress available on UNIX systems. The CSD compression performance was about 10% better than that of LZC with the practical dictionary size, an 8k-entry dictionary when the test data was from the Calgary Compression Corpus. With hashing, the CSD processing speed became as fast as that of LZC, although the CSD algorithm was more complicated than LZC.

In this paper, a fault model for multiple-valued programmable logic arrays (MV-PLAs) is proposed and the equivalences of faults of MV-PLA's are discussed. In a supposed multiple-valued NOR/TSUM PLA model, it is shown that multiple-valued stuck-at faults, multiple-valued bridging faults, multiple-valued threshold shift faults and other some faults in a literal generator circuit are equivalent or subequivalent to a multiple crosspoint fault in the NOR plane or a multiple fault of weights in the TSUM plane. These results lead the fact that multiple-valued test vector set which indicates all multiple crosspoint fault and all multiple fault of weights also detects above equivalent or subequivalent faults in a MV-PLA.

In this paper, we present Branching Oriented System Equation based on-line error correction scheme for recursive digital signal processing. The target digital signal processing is linear and time-invariant, and the algorithm includes multiplications with constant coefficient, additions and delays. The difficulties of the algorithm-level fault tolerance for such algorithm without structural regularity include error distribution problem and right timing of error correction. To escape the error distribution problem, multiple fan-out nodes in an algorithm are specified as the nodes at which error corrections are performed. The Branching Oriented Graph and Branching Oriented System Equation are so introduced to formulate on-line correction schemes based on this strategy. The Branching Oriented Graph is treated as the collection of computation sub-blocks. Applying checksum code independently to each sub-block is our most trivial on-line error correction scheme, and it results in, with appropriate selection of error identification process, TMR in sub-block level. One of the advantages of our method is in the reduction of redundant operations performed by merging some computation sub-blocks. On the other hand, the schedulability of the system is an important issue for our method since our on-line error correction mechanism induces additional data dependencies. In this paper, the schedulability condition and some modifications on the scheme are also discussed.

This paper presents an approach for assembly plan generation from an assembly illustration. Previously, we have already proposed an approach for the assembly plan related information acquisition from an assembly illustration, in which auxiliary lines were taken as clues. However, some ambiguity remains in dynamic information such as assembly operations and their execution order. We have verified through experiments that the ambiguity could be made clear by referring to the feedback information from the completed assemblage after the assembly operations shown in the current illustration. But in fact, in an assembly illustration there are not only the figures of mechanical parts and the auxiliary lines for visualizing their assembly relations, but explanatory words and explanatory lines as well. Explanatory words can basically be classified into two categories: instructions on assembly operations and mechanical part names. The former explicitly describes dynamic information such as the details of assembly operations. The latter also implies dynamic information such as the function of a mechanical part. Explanatory lines are usually drawn for making clear the explanatory relations. Naturally we consider that to integrate the information from explanatory words with that already obtained through the extraction of auxiliary lines will probably enable us to generate an unambiguous assembly plan from the currently observing illustration.

Kumar and Billinton have presented a new technique for obtaining the steady-state probabilities from a flow graph based on Markov model. By examining the graph and choosing suitable input and output nodes, the steady-state probabilities can be obtained directly by using the flow graph. In this paper this graphical technique is applied for a k-out-of-n: G repairable system. Consequently a new derivation way of the formulae for the steady-state availability and MTBF is obtained.

A new current-mode dual-input configuration for the generation of a ratio (Y₂/Y₁) type network function using the second generation current conveyor (CC ) elements has been proposed. With nonideal CC s, a precise compensation of output current can be obtained simply by insertion of a grounded admittance (Y_c). The design of precise insensitive current-mode integrator/differentiator type functional circuits follows directly if the Y_1,2 admittances are suitably chosen as RC components.

This paper discusses a new design method for 2-D variable FIR digital filters, which is an extension of our previous work for 1-D case. The method uses a 3-D prototype FIR filter whose cross-sections correspond to the desired characteristics of 2-D variable FIR filters. A 2-D variable-angle FIR fan filter is given as a design example.