Friso G. De Boer

R wave detection using Coiflets wavelets

Accurate detection of QRS complexes is important for ECG signal analysis. In this paper, a generic algorithm using Coiflet wavelets is introduced to improve the detection of QRS complexes in Arrhythmia ECG Signals that suffer from: 1) non-stationary effects, 2) low Signal-to-Noise Ratio, 3) negative QRS polarities, 4) low QRS amplitudes, and 5) ventricular ectopics. The algorithm achieves high detection rates by using a signal-to-noise ratio threshold instead of predetermined static thresholds. The performance of the algorithm was tested on 48 records of the MIT/BIH Arrhythmia Database. It was shown that this adaptive approach results in accurate detection of the QRS complex and that Coiflet1 achieves better detection rate than the other Coiflet wavelets.

Applying the APG to measure Heart Rate Variability

The Acceleration Plethysmogram (APG) is used to calculate the heart rate (HR) and HRV. The APG is an optical technique that has been developed for experimental use in vascular diseases. It is considered a promising tool that may replace some of the current traditional cardiovascular diagnostic tools. The performance of the proposed algorithm when tested on 26 records measured at rest showed very promising results. Our results demonstrate that the HRV indices, SDNN and rMSSD can be calculated using APG signals collected at rest.

Adaptive feedforward control to compensate cogging torque and current measurement errors for PMSMs

Torque ripple minimisation is achieved by an adaptive feedforward control method to reduce cogging torque and current measurement errors. Experiments have been conducted to show that torque ripples are highly sensitive to changes in current offset and cogging torque. An easy way to adapt to cogging torque changes (due to temperature) is proposed and implemented after an initial estimation of the cogging torque is done. This initial estimation is necessary for manufacturing inaccuracies and ongoing adaptation is to cope with operating conditions changes. Higher harmonics of the cogging torque can also be compensated through this way. Current offset which contributes to a large proportion of the torque ripple is also compensated through an adaptive method. The advantages of the proposed scheme combine the benefits of preprogrammed waveform methods which are easy to implement and a simple adaptive system to effectively reduce RMS torque ripple factor from 11.30% to 1.7%.

Recognition of T waves in ECG signals

The method described in this paper deals with the problems of T-wave detection in ECG signals. Determining the position of a T-wave can be complicated, due to the ambiguous and changing form of the complex and the presence of noise. We developed a method to detect T-waves in noisy signals. The performance of the proposed method was tested on 33 records of the MIT/BIH Arrhythmia Database resulting in 0.48% incorrectly detected T waves.

Indirect adaptive feedforward control in compensating cogging torque and current measurement inaccuracies for Permanent Magnet motors

Theoretically, if the inverse transfer function of the plant is known, feedforward control can achieve zero error between the reference and the output. The main aim of this paper is to design a simple feedforward model and to compensate for the errors brought about by a change in the cogging torque and current measurement inaccuracies. The reference position, speed and acceleration have been designed so that they can be used as inputs to the feedforward model. The feedforward model is a stable form of the inverse transfer function of the motor dynamics of a PM motor. MRAS technique has been used to compensate for changes to the cogging torque and current measurement inaccuracies. A successful indirect adaptive feedforward control has been designed for three different shapes of the back EMF (sinusoidal, trapezoidal and a mix of the first two shape using real motor back EMF) and the control scheme is validated through hardware experimentation.

P wave demarcation in electrocardiogram

Efficient and effective feature extraction algorithms are required in the analysis of long records electrocardiographic (ECG) signals. In this paper a computationally efficient method is proposed as a feature extractor for P waves in ECG signals. The performance of the proposed algorithm was tested on 29 records of the MIT/BIH Arrhythmia Database resulting in 0.72% incorrectly detected P waves.

Torque ripple estimation and minimisation independent of sensor type

Torque ripple exists at the output of permanent magnet synchronous motors (PMSMs) as a result of current error, non-sinusoidal back-EMF and cogging torque. Since part of the cogging torque component is based on manufacturing error, motor design may not eliminate all torque ripple. One possible method for reducing torque ripple is by injecting the inverse of a torque ripple estimate to the current controller. Torque estimators that include cogging torque rely on feedback from different sensors such as vibration sensors and accelerometers, microphones, speed and torque sensors. This paper outlines a control method that can utilize any sensor with a linear transfer function between the torque and the sensor input. The method determines an estimate of the torque ripple for a number of harmonics simultaneously, as well as the transfer function between the torque and the sensor. Experimental results show different sensor types provide similar estimates of the torque with less than 2% variation in magnitude for the sensor types used. Experimental results show that regardless of which sensor type is used, critical torque harmonics are reduced by at least a factor of 3.

An acquisition method for the MLR of auditory evoked potentials

The study is focused on the recording of the auditory evoked potential to stimuli that result in binaural (two-ear) interaction. The auditory evoked potential is derived from small bioelectric potentials recorded from the scalp. The AEP is categorized on the basis of the latency of the response following the auditory stimulus. For example, the auditory brainstem response (ABR) occurs in the first 20 ms after the stimulus, the middle latency response (MLR) from 20 to 70 ms, and the slow vertex response (SVR) up to 500 ms after stimulation. The study of auditory evoked potentials may provide insight in the mechanism of auditory processing in the brain. The study presents a methodology to measure AEP related to binaural hearing.

A New Approach Correlating Binaural Hearing and the Brain’s Response

Normal binaural hearing allows the auditory system to determine the direction and distance of sound sources and to detect certain sounds at much lower intensity levels. Different stimuli may have different impact on binaural processing and may generate different brain responses. The mechanism by which this occurs is poorly understood. Time averaged EEG responses of normal hearing subjects to repeated stimuli were analyzed. The stimuli, 500 Hz Blackman windowed pure tones, were presented as homophasic or anti-phasic and were also mixed with various noise conditions. Auditory evoked potentials (AEP) were obtained by averaging 500 trials of in-phase and 500 trials of outphase of each EEG epoch. The results show that the amplitude of the dominant frequency component in the 20 50 Hz range of the middle latency response of the AEP was larger for the anti-phasic condition than for the homo-phasic condition. The normalised amplitude differences were larger when the stimuli were embedded in noise resulting in a higher mean value of the normalized amplitude difference than for noise free stimuli. These results are likely to relate to binaural masking level difference which finds that the detection of a signal in a background noise is easier when the signal has a different inter-aural phase difference than the noise.

Position based iterative learning control to minimise torque ripple for PMSMs

Torque ripple minimisation is achieved by an adaptive feedforward control for mass produced permanent magnet synchronous motors. Torque ripple, a major problem for such motors can be reduced using appropriate estimation scheme from speed information. The proposed method employs online zero phase filtering which can effectively minimise all torque harmonics at low speed operation. Experimental results validate the proposed scheme reducing the torque ripple factor from 6.87% to 1.28% using the proposed method. The adaptation scheme make use of PD type iterative learning control scheme which has the advantage of fast convergence compared to the more commonly used P type iterative learning control. The PD-ILC scheme is also quite robust to an initial error in the estimation of the parameter J.