No description available.

A major break-through in the real-time digital simulation of dynamic models has occurred with the introduction of the Intel 2920 digital signal processing chip. The problems and potentials for this new device are demonstrated by implementing an elliptic function digital low pass filter via the bilinear z-transform approach. The software implementation is presented. Debugging and software verification are accomplished via manufacturer's simulation software tools. The hardware performance is verified in the laboratory. The results of these efforts point to much promise for wide scale applications, however, problems associated with performance indicate early version chip problems.

This research report is devoted to an investigation of finite impulse response (FIR) filters and their implementation via windowing techniques. Analysis is restricted to those FIR filters that possess both a nonrecursive structure and a linear phase response. Initially an informative yet concise overview of FIR filters is presented. Some knowledge of "discrete – time" systems is assumed. Next, the concepts associated with FIR filter implementation through the utilization of window functions is developed. Analysis is performed in both the time and frequency domains. Results obtained define optimal window function criteria. The last design technique is devoted to computer aided FIR filter design via windows. The purpose of this section is to demonstrate the advantages of using the computer to solve the FIR filter – window design problem. An example problem supplements this section. Conclusions, along with a brief summary are documented.

Cross correlator systems, analog and clipped input channels, with correlated random narrow band Gaussian noise as inputs are computer simulated. The performance of each system is evaluated on the basis of the output signal-to-noise ratio. The output SNR of each system is compared with theoretical asymptotic approximations computed as a function of the SNR of the input channels. The output of the simulation compares within 3 dB of the asymptotic approximations for the analog correlator for all values of the correlation coefficient tested and for the systems with clipped input channels with uncorrelated inputs. For the systems with clipped input channels, certain combinations of the input SNR with non-zero correlation cause the output SNR to be zero. This causes discontinuities in the dB plot. For the systems with clippers and non-zero correlation of the input channels, the output of the simulation compares within 3 dB of the results of the asymptotic approximations when no discontinuities occur in the output plot.

The Fujitsu MB8764 digital signal processing chip is designed to operate with a machine cycle of up to 10 MHz. The chip’s ability to perform a 16-by-16 bit multiply and add operation in one machine cycle makes it a good candidate for real time digital signal processing. Unlike the Intel 2920 the Fujitsu MB8764 does not have an onboard analog-to-digital, digital-to-analog converter. Therefore, this paper will be restricted to the operation of this device with digital data input and output. The use of the MB8764 as a digital filter is analyzed. Performance limitations due to finite word length, memory size and configuration, and clock rate are considered. The MB8764 capabilities in computing fast Fourier transforms are discussed. Development of a working digital filter with the MB8764 work station is presented.

When a digital process is realized on a general-purpose computer or a special-purpose hardware, errors due to finite register length are introduced. These errors are due primarily to arithmetic roundoff, coefficient quantization, and scaling rules. This paper addresses the effects of finite word length on a direct-form implementation of a high order H (z) transfer function. The development and analysis of a modified direct-form realization suggested by Dr. Fred O. Simons, are carried out via FORTRAN emulation of a fourth-order low-pass Butterworth filter. The results are presented as a parametric tradeoff of signal-to-noise ratio at the filter output versus word length. Conclusions are drawn by comparing the modified direct-form with the canonic direct-form. The analysis presented here is intended to illustrate how a high order transfer function can be realized directly without decomposing into a group of low-order subfilters.

As the use of computer in engineering design as well as other areas increase, it becomes more imperative that the application software used be as simple, convenient, and powerful as possible. The engineer is not interested in the internal workings of the computer or its operating system. It is the design itself that takes precedence. The filter design package developed for this project, known as FILTER, is such an application. With FILTER, coupled with the Digital Research Graphics Environment Manager, the engineer is led through the analog and digital filter design phase on a personal computer with carefully designed interactive computer graphics requiring little or no computer knowledge.

Thesis (MSc (Mathematics))--Stellenbosch University, 2008. / LULU smoothers is a class of nonlinear smoothers and they are compositions of the maximum and minimum operators. By analogy to the discrete Fourier transform and the discrete wavelet transform, one can use LULU smoothers to create a nonlinear multiresolution analysis of a sequence with pulses. This tool is known as the Discrete Pulse Transform (DPT). Some research have been done into the distributional properties of the LULU smoothers. There exist results on the distribution transfers of the basic LULU smoothers, which are the building blocks of the discrete pulse transform. The output distributions of further smoothers used in the DPT, in terms of input distributions, has been a challenging problem. We motivate the use of these smoothers by first considering linear filters as well as the median smoother, which has been very popular in signal and image processing. We give an overview of the attractive properties of the LULU smoothers after which we tackle their output distributions. The main result is the proof of a recursive formula for the output distribution of compositions of LULU smoothers in terms of a given input distribution.

published_or_final_version / Speech and Hearing Sciences / Doctoral / Doctor of Philosophy

Collaborative filtering (CF), a very successful recommender system, is one of the applications of data mining for incomplete data. The main objective of CF is to make accurate recommendations from highly sparse user rating data. My contributions to this research topic include proposing the frameworks of imputation-boosted collaborative filtering (IBCF) and imputed neighborhood based collaborative filtering (INCF). We also proposed a model-based CF technique, TAN-ELR CF, and two hybrid CF algorithms, sequential mixture CF and joint mixture CF. Empirical results show that our proposed CF algorithms have very good predictive performances. In the investigation of applying imputation techniques in mining incomplete data, we proposed imputation-helped classifiers, and VCI predictors (voting on classifications from imputed learning sets), both of which resulted in significant improvement in classification performance for incomplete data over conventional machine learned classifiers, including kNN, neural network, one rule, decision table, SVM, logistic regression, decision tree (C4.5), random forest, and decision list (PART), and the well known Bagging predictors. The main imputation techniques involved in these algorithms include EM (expectation maximization) and BMI (Bayesian multiple imputation). / by Xiaoyuan Su. / Vita. / Thesis (Ph.D.)--Florida Atlantic University, 2008. / Includes bibliography. / Electronic reproduction. Boca Raton, Fla., 2008. Mode of access: World Wide Web.