Duncan Campbell

A Fuzzy Logic‐Controlled Classifier for Use in Implantable Cardioverter Defibrillators

Purpose: Implantable cardioverters defibrillators (ICDs) are increasingly used in the management of life‐threatening arrhythmias. Correct recognition of a treatable arrhythmia is crucial to this application. However, the computational power of microprocessors currently used in ICDs limits the range of traditional algorithms available for this application. Methods: Classification based on fuzzy inference systems (FIS) were trained to recognize different cardiac rhythms (AF, VF, SVT, VT) from the Ann Arbor Electrogram Library. The FIS used were designed using adaptive‐network‐based fuzzy inference methods to optimize the classification procedure. Only computational techniques suitable for ICD design were used. Results: After pretraining with the ANFIS correct rhythm classification was observed for the rhythms studied. Conclusion: In this preliminary study, successful rhythm classification was demonstrated using fuzzy logic techniques. In view of the computational efficiency this may have application in ICD design.

A reconfigurable parallel inference processor for high speed fuzzy systems

An application specific parallel rule inference architecture is presented which is capable of performing an entire rule inference within one clock cycle. The architecture is composed of asynchronous self routing min and max blocks that are interlinked into a network as determined by the rules for the intended application. The inference processor does not have to fetch rules from memory because the rules are configured in the firmware structure; hence the inherent speed advantage. The design is targeted for high capacity Complex Programmable Logic Devices (CPLDs), whose ability to be reconfigured allows the application specific rule structure to be practical for real world systems.

A high-speed reconfigurable defuzzification architecture for true centre-of-gravity computations

An application specific parallel defuzzification architecture is presented which is capable of performing high-speed true centre-of-gravity (CoG) computations. This architecture uses values determined by the integrals of the output membership functions (MFs), as opposed to using singleton values. This is done by storing pre-calculated values of areas and moments within embedded memory blocks. During run-time the defuzzification processor uses these individual pre-calculated values to compute the final CoG result. The design is especially well suited for families of high capacity complex programmable logic devices (CPLDs) that have a large number of I/O pins and whose ability to be reconfigured allows the application specific pre-calculated structure to be practical for real world systems.

Dynamic weight leakage for LMS adaptive linear predictors

Adaptive line enhancers provide a means of adaptively estimating the spectral characteristics of low SNR stationary signals. The LMS algorithm is a computationally efficient method for updating filter weights and is commonly used. Insufficient modal stimulation can lead to false spectral peaks. These peaks may be eliminated by applying fixed weight leakage however this produces a biased spectral estimate. Dynamic weight leakage and hybrid combined weight leakage schemes have been developed and are applied within the LMS structure. They provide mechanisms for suppressing state spectral modes with minimal or no bias in the spectral estimates. These schemes can be implemented with only a minor impact on the computational effort.

Adaptive EEG transient event discrimination using dynamic LMS filter weight leakage

The EEG is a highly complex and dynamic signal comprising a large ensemble of time-varying, statistical properties. Such diverse signal properties pose significant challenges in processing the EEG. A dynamic weight leakage based LMS adaptive linear predictor has been developed to discriminate for transient events within the EEG, and in particular, epileptiform discharges. The resulting procedure improves the SNR of these events by at least two-fold, leading to greater selectivity in subsequent epileptiform event detection stages.

Transient discrimination in nonlinear time series using linear-phase time-delay neural networks

A linear-phase adaptive nonlinear predictor is presented which allows the adaptive discrimination of short-term transient signals from longer-term periodic signals whilst maintaining the phase integrity of both. A linear-phase time-delay neural network is used in an adaptive predictor configuration and is adaptively trained as a predictor for signals arising from nonlinear processes. The targeted application of this technique is the decomposition of electroencephalograms to classify waveform transient and background rhythmic characteristics which is useful in managing neurophysiological conditions such as epilepsy.

Fully parallel fuzzy logic processor architecture: exceeding one billion rules per second

A novel, very high-performance fuzzy logic processor architecture has been developed and conceptually proven. Processing of over 1.2 billion fuzzy logic instructions per second is possible. It is an 8-bit, fully parallel, synchronous, pipelined employing max-min based rule inferencing. The concept has been proven using complex programmable logic devices (CPLDs), exploiting both the high gate count and I/O pin count, as well as the reconfigurable structure. True non-singleton center-of-gravity defuzzification has also been developed incorporating an optimized dividing speeds significantly greater than the currently available commercial deices. Implementation in CPLDs allows reconfigurability in the fuzzy logic design, while custom devices allow a much greater degree of integration and potential for even greater processing speeds. High speed fuzzy logic processing is particularly suited to high bandwidth data processing applications such as virtual reality.

Feature enhancement from nonlinear time series using linear-phase and nonlinear-phase time-delay fuzzy combiners

Time-delay fuzzy combiners are used to extract short-term transients embedded in longer-term, nonlinear, periodic signals. Linear-phase structures maintain the phase integrity of these signals and provide transient extraction with reduced phase distortion and lower prediction errors. Inference rule reduction within the fuzzy combiner can lead to greater processing efficiencies but at the expense of higher prediction errors and malformation of the extracted transients. This research is aimed at the decomposition of electroencephalograms to classify waveform transient and background rhythmic characteristics which is useful in managing neurophysiological conditions such as epilepsy.

Frequency estimation of travelling ionospheric disturbances

The use of adaptive filters in estimating the instantaneous frequency of Doppler shifted signals reflected from the ionosphere is addressed in this paper. With the aid of a software simulator, it has been shown that an adaptive linear predictor is effective in extracting the required Doppler shift. Analysis of such signals provides information on velocity fluctuations that arise from travelling ionospheric disturbances and is important when trying to link a disturbance with its source. Spectral analysis of the Doppler itself can provide additional information about a disturbance and limitations in the performance of Fourier transforms in this application are highlighted.