P.B. Crilly

An integrated pulse oximeter system for telemedicine applications

This paper will describe a wireless integrated pulse oximeter system that enables the measurement of blood oxygen saturation and the direct measurement of heart rate. Its implementation is relatively simple and thus easily put on a VLSI chip so it can be attached on a persons ear lobe, finger or other body part. The chip includes a radio transmitter/telemetry system, so measurement data can be linked to a data acquisition system. To conserve battery life, the sensor will be activated by external RF tagging as needed.

Adaptive noise filtering using an error-backpropagation neural network

A neural network of the feedforward-error backpropagation type proposed by D.E. Rumelhart et al. (1986) was applied to filter noise from spectral data commonly encountered in infrared absorption of molecular transitions. The purpose was to gain insight into the way a neural network can be trained to remove noise from a noise-corrupted signal with implications for signal processing in general. The neural network simulation was implemented in Fortran and run on a VAX 8800. Training of the neural network occurred on a set of spectral data with random transitions and line shape parameters. Preliminary results of the performance of the adopted neural network are reported and discussed along with observed limitations. Future improvements on noise filtering using a neural network are proposed. >

Image compression using hybrid neural networks combining the auto-associative multi-layer perceptron and the self-organizing feature map

A new image compression technique is presented using hybrid neural networks that combine two different learning networks, the auto-associative multi-layer perceptron (AMLP) and the self-organizing feature map (SOFM). The neural networks simultaneously perform dimensionality reduction with the AMLP and categorization with the SOFM to compress image data. Two hybrid neural networks forming parallel and serial architectures are examined through theoretical analysis and computer simulation. The parallel structure network reduces the dimensionality of input pattern vectors by mapping them to different hidden layers of the AMLP selected by winner-take-all units of the SOFM. The serial structure network categorizes the input pattern vectors into several classes representing prototype vectors. Both the serial and parallel structures are combinations of the AMLP and SOFM networks. These hybrid neural networks achieve clear performance improvement with respect to decoded picture quality and compression ratios, compared to existing image compression techniques. >

Improving the convergence rate of Jansson's deconvolution method

We investigate a technique to speed up the convergence of Janssons deconvolution method. The technique extends the work of Agard who applied an inverse filter to the correction term in Janssons iterative equation. Test cases include severely overlapped peaks having moderate and low signal-to-noise ratios. The results show significant reductions in the final estimates rms error.

Error analysis with deconvolution algorithms

Iterative deconvolution algorithms can be used to obtain an estimate of a signal that has been distorted by an impulse response function and corrupted by additive noise. The decision to continue iterating is usually based on some minimum error or convergence rate criterion being satisfied. However, in some classes of systems, there are pitfalls with respect to the error calculation that must be understood. One of these pitfalls is examined. >

A digital signal processing architecture for iterative deconvolution restoration algorithms

A VLSI DSP chip is presented that will significantly improve the processing throughput for a general class of iterative deconvolution algorithms. The design will be based on a systolic array concept. This will enable these algorithms to be used for real time DSP applications which formerly due to speed limitations were not possible. The increased class of applications will enable further understanding of these applications. The higher throughput will also enable the researcher to further take advantage of the features unique to iterative deconvolution.<<ETX>>

Application of neural networks on color error reductions in television receivers

The application of neural networks to color error reduction in television receivers was investigated. The best results were obtained with a backpropagation neural network using one hidden layer of seven processing elements. Approximately a 70% reduction in color errors was achieved. The color error reduction was displayed on a CIE chromaticity diagram and tested with a video graphics system. Color measurements were made with this system. Both the simulation chromaticity diagram and the video color test system verified the neural network results. >

Improved localization in GPS-denied environments using an autoregressive model and a generalized linear model

The Theater Positioning System (TPS), which can perform in GPS-denied environments and can work with, or independently of, GPS systems, was presented in [1]. The principal difficulty in optimally combining this new system and GPS is introduced by the environment, which may impart somewhat unpredictable transmission delays to the signal and thus results in less accurate performance when TPS works unaided in the environment while GPS is unavailable. In this paper, we propose two methods-an autoregressive process and a generalized linear model to model the transmission delays generated in TPS signal propagation. Using those, the unknown signal delays can be predicted and thus can be employed in the subsequent localization process. Numerical examples are provided to illustrate the performance of both methods proposed in this paper.

An introduction to neural networks based on the feed forward, backpropagation error correction network with weight space limiting based on a priori knowledge

Neural networks are introduced to instrumentation professionals. The structure of neural networks is described with particular attention paid to the backpropagation network. Both graphic and analytical descriptions are used. Examples of backpropagation networks applied to one- and two-dimensional resolution enhancement are used to exhibit characteristics of there networks. In the two-dimensional case, image recovery and enhancement of Hubble-space-telescope-like images are employed as examples. Several approaches to the effective limitation of the network weight space are reported. The conceptual basis of weight space limitation is introduced. The connection of weight space limitation to incorporation of a priori knowledge of the systems to which the networks are applied is discussed with examples.<<ETX>>

Improved results for color error reductions using neural networks and a fast training algorithm

In a previous paper by Xu and Crilly (see ibid., vol. 39, p. 630, August 1993), neural networks were able to correct color errors in television receivers. The results showed that 70% of these errors were reduced by various amounts. This paper further investigates the use of neural networks for error reduction. A simpler network is employed in conjunction with a fast training algorithm. The new results are significantly better than previously reported. It is shown that virtually all of the color errors are significantly reduced. It also presents the details of the training algorithm. >