In most application programs, loops usually contain most of the computation in a program and are the most improtant source of parallelism. When loops are executed on multiprocessors, the cross iteration data dependences need to be enforced by synchronization between processors. In this paper, we propose a new synchronization scheme(Free/Hold) for reducing overgeads occured by synchronization variables in data oriented scheme and delay of time occured by synchronization instruction in statement oriented scheme. The Free/Hold mechanism enforces the correct execution order by inserting synchronization instruction between each instance with data dependence relationship using the RD(Real dependence Distance). We also present an algorithm for removing unnecessary dependences in one-to-many dependences.

We propose two architectures using two dimensional systolic arrays for the DCT/DST/DHT. One decomposes the N-point DCT/DST/DHT into even-and odd-numbered frequency samples, and then computes them independently at the same time. In addition, the proposed architecture can be used for the IDCT/IDST/IDHT. Anogher is the modified version for the DHT/IDHT. Two proposed architectures generate outputs sequentially using real multiplications and additions. As compared to the conventional methods the proposed systolic arrays exhibit many advantages in terms of simplicity of the processing element (PE), latency, and throughput. Teh simulation results using VHDL, international standard language for hardware description, show the effectiveness of the proposed architecture.

This paper describes the multichannel combined R2MFC/DTMF/CCT reciver which provides a signaling service functions for call processing control in digital switching system. Using the TMS320C25 DSP chip, we have implemented multi-function receriver shich processes 8 channels of R2MFC, DTMF, and CCT signals simultaneously. In order to increase the channel multiplicity of the combined receiver. R2MFC and CCT receiver were employed by discrete Fourier transform(DFT) method using Goertzel algorithm, and DTMFreceiver was employ by infinite impulse reponse(IIR) filtering method using 4KHz subsampling technique. The combined receiver has 4 function modes for each channel such as R2MFC, DTMF, CCT, and Idle modes. The function mode of each channel may be selected at any time by single-chip micro-controller(.mu.C). Hence, the number of channels assigned for each function mode can be adjusted dynamically according to the signaling traffic variations. From the experimental test results using the test-bed, it has been proved that the combined receiver statisfies all receiver satisfies all receiver specifications, and provides good channel multiplicity and performance, Therefore, it may give a great improvement than existing receiver in cost, reliability, availability, and serviceability.

This paper presents new QRD-based recursive least squares algorithms for bilinear lattice filter. Bilinear recursive least square lattice algorithms are derived by using the QR decomposition for minimization covariance matrix of predication error by applying Givens rotation to the bilinear recursive least squares lattics algorithms. The proposed algorithms are applied to the bilinear system identification to evaluate the performance of algoithms. Computer simulations show that the convergence properties of the proposed algorithms are superior to that of the algorithms proposed by Baik when signal includes the measurement noise.

In this paper, we assume that the dynamics of DC motor and nonlinear load are unknown. We propose an inverse dynamic model of DC motor and nonlinear load using the artificial neural network and construck speed control system based on the proposed dynamic model. We also propose another dynamic model with speed prediction scheme using the artificial neural network that removes the undesirable time delay effect caused by the computation time during the real-time control. We suggest a dynamic model which has arbitrary number of speed arguments and is especially effective when the motor and load has large moment of inertia. Next, we suggest a controller that combine the neurocontrol and PID control with constant gain. We show that the proposed neurocontrol systems have capabilities of noise rejection and generalization to have good velocity tracking through computer simulations and experiments.

In this paper, a fuzzy expert system is developed for fault diagnoisis of a drum-type boiler system in fossil power plants. The develped fuzzy espert system is composed of knowledge base, fuzzification module, knowledge base process module, knowledge base management module, inference module, and linguistic approximation module. The main objective of the fuzzy expert system is to check the states of the system including the drum level and detect faults such as the feedwater flow sensor fault, feedwater flow control valve fault, and water wall bube rupture. The fuzzy expert system diagnoses faults using process values, manipulated values, and knowledge base which is built via interviews and questionaries with the experts on the plant operations. Finally, the validity of the developed fuzzy expert system is shown via experiments using the digital simulator for boiler system is Seoul Power Plant Unit 4.

In this paper we suggest a 32 bit high speed parallel multiplier which plays an important role in digital signal processing. We employ a bit-pair recoding Booth algoritham that gurantees n/2 partial product terms, which uniformly handles the signed-operand case. While partial product terms are generated, a special method is suggested to reduce time delay by employing 1's complement instead of 2's complement. Later when partial products are added, the additional 1 bit's are packed in a single partial product term and added to in the parallel counter. Then 16 partial product terms are reduced to two summands by using successive parallel counters. Final multiplication value is obtained by a BLC adder. When this multiplier is simulated under 0.8$\mu$CMOS standard cell we obtain 30ns multiplier speed.

It is AGC(Automatic Gain Control) amplifier to decide characteristics of IF(Intermediate Frequency) processing IC. When demodulated IF signal by PLL type demodulator, the amplitude of input singla should be maintained at a certain amplitude. The AGC amplifier is an important factor to achieve this condition. The AGC amplifier needs the wide dynamic range, the wide AGC range and better noise characteristics. We designed the AGC amplifier to satisfy these characteristics.

In this thesis, we perform the experiment of speaker recognition by identifying vowels in the pronounciation of each speaker. In detail, we extract the vowels from the pronounciation of each speaker first. From it, we check the frequency energgy of 29 channels. After changing these into fuzzy values, we employ the fuzzy inference to recognize the speaker by text-dependent and text-independent methods. For this experiment, an algorithm of extracting vowels is developed, and newly introduced parameter is the frequency energy of the 29 channels computed from the extracted vowels. It shows the features of each speakers better than existing parameters. The advanced point of this paramter is to use the reference pattern only without the help of any codebook. As a rewult, test-dependent method showed about 95.5% rate of recognition, and text-independent method showed about 94.2% rate of recognition.

In this paper, each step of microprogramming was discussed using a microprogrammable machine that would grasp the characteristics of the microprogram in close relation to the hardwares and microinstruction definable language which make an effective microcode. This was proposed for an environment which minimizes costs.

A motion picture compression scheme using subband coding with motion compensation is presneted in this paper. A hierarchical subband decomposition is used to split the image signal into 10 subbands with a 3-layer pyramid structure and motion compensation is used in each band. However, in this case, motion vector information is drastically increased; therefore, initial motion vectors are estimated in the highest pyramid and motion vectors are refined using the reconsructed subband signal in each layer. Simulation results show that the proposed method compares favorably in terms of prediction error energy and side informatio with methods requiring additional information. Images recostructed from the proposed method show good quality compared to those reconstructed using blockwise DCT.

Number plates of vehicles operating in real world are sometimes difficult to recognize due to the number plate degradation (bent or dirty plates). To recognize the vehicle number from a number plate with severe degradation, good segmentation is necessary, which in turn requires good thresholding. This paper proposes a binarization method that combines the fast processing speed of global thresholding methods with the local thresholding methods' ability to adapt to lacal gray level characteristics. The proposed method overcomes the degradation of number plates quickly and maintains the widths of digit strokes uniform. The paper presents results of comparison with existing global and local thresholding methods.

The segmentation of range image is essential to recognize the three dimensional object. Generally, surface curvature is well-known feature for segmentation and classification of the fange image, but it is sensitive to noies. In this paper, we propose the structure of hierarchical neural network using surface curvature for segmentation of range images. The hierarchical structure of neural networks is robust to noise and the result of segmentaion is better than conventional optimization method of single level.

Video compression algorithms are based on removing spatial and/or temproal redundancy inherent in image sequences by predictive(DPCM) encoding, transform encoding, or a combination of predictive and transform encoding. In this paper, each 8$\times$8 DCT coefficient of DFD(displaced frame difference) is adaptively quantized by one of the four quantizers depending on total distortion level, which is determined by characteristics of HVS(human visual system) and buffer status. Therefore, the number of possible quantizer selection vectors(patterns) is 4$^{64}$. If this vectors are coded, toomany bits are required. Thus, the quantizer selection vectors are limited to 2048 for Y and 512 for each U, V by the proposed method using SWAD(sum of weighted absolute difference) for discriminating vectors. The computer simulation results, using the codebook vectors which are made by the proposed method, show that the subjective and objective image quality (PSNR) are goor with the limited bit allocation. (17Mbps)

본 논문에서는 입력영상에 존재하는 복수 패턴을 한번의 패스(pass)로 인식할 수 있는 변형된 네오코그니트론을 제안한다. 변형된 네오코그니트론은 특징을 추출하는 S세포층 및 V세포층과 패턴의 크기, 이동, 회전 등 변형에 덜 민감하게 하는 C세포층으로 구성된 계층적 신경망이다. S세포와 V세포들은 DCC(don't care condition)을 적용하여 영상에 존재하는 모든 패턴에 대한 특징을 추출하는 역할을 하며, S세포와 C세포는 추출된 패턴에 대한 위치정보도 함께 가진다. 위치정보는 추출된 특징들이 서로 간섭을 받지 않고 올바른 특징만이 추출되도록 하는 데 이용된다. 제안한 방법은 피드백이 존재하는 선택적 주의 방법과는 달리 단일패스로 인식이 가능하므로 인식시간이 짧으며, DCC와 위치정보를 이용하므로 서로 접촉되어 있는 복수 패턴의 인식도 가능하다.

A method of applying a recurrent backpropagation network to identifying or modelling a dynamic system is proposed. After the recurrent backpropagation network having both the characteristicsof interpolative network and associative network is applied to XOR problem, a new model of recurrent backpropagation network is proposed and compared with the original recurrent backpropagation network by applying them to XOR problem. based on the observation thata function can be approximated with polynomials to arbitrary accuracy, the new model is developed so that it may generate higher-order terms in the internal states Moreover, it is shown that the new network is succesfully applied to recognizing noisy patterns of numbers.