Biyolojik işaretler için adaptif gürültü azaltma sistemi

Abstract

Biyolojik işaretler genellikle toplamsal gürültü ile bozulmuş olarak algılanırlar. Bu gürültü, diğer bir biyolojik işaret veya ortam gürültüsü olabilir. Gürültü ile bo2ulmuş bir işareti, yeniden elde etmenin klasik yöntemi, işareti bir filtreden geçirmektir. Bu filtre, sabit katsayılı veya adaptif bir filtre olabilir. impuls cevabını, her giriş örneği için otomatik olarak ayarlayan filtrelere, ADAPTÎF FİLTRE denir. Adaptif filtreler, sabit katsayılı filtrelerden farklı olarak işaret ve gürültü istatistikleri hakkında, fazlaca ön bilgi olmadan tasarlanabilirler. Biyolojik işaretler ve bunlara eklenen gürültüler çoğu kez, rasgele olaylardan etkilenirler. Ayrıca, bu işaretlerin vücutta ki iletim ortamlarının özellikleri, zamanla bilinmeyen bir şekilde değişebilmektedir. Dolayısı ile, gürültü azaltma probleminin çözümünde adaptif filtre kullanılması uygun olur. Gürültü Azaltma Sisteminde kullanılan adaptif filtre, bir adaptif algoritma ile kontrol edilerek, katsayılarını her giriş örneği için yeniden hesaplar. Bu tezde, biyolojik işaretlerdeki gürültülerin Adap tif Gürültü Azaltma Sistemi kullanılarak temizlenmesi araştırılmış ve bu amaçla hızlı bir sistem gerçekleşti rilmiştir, işlemci olarak, Yüksek Performanslı Mikro- denetleyici (HPC) kullanılmıştır. Vücuttan algılanan, gürültülü işaretler, bilgisayar ile kontrol edilen, Analog /Sayı s al Çevirici ünitesinde sayısal forma sokul muştur. HPC için yazılan bir adaptif filtre programı ile, Elektrokardiyografi (EKG) ölçümünde karşılaşılan şebeke gürültüsü azaltılmıştır. Çıkış işareti, Sayısal/ Analog Çevirici ünitesi ile analog işarete dönüştürülmüş ve izlenmiştir. Daha sonra bu sistem, çeşitli biyolojik işaretler üzerindeki gürültüleri azaltacak şekilde, genel amaçlı hale getirilmiş ve Fetal EKG üzerindeki anne EKG gürültü sünün azaltılmasında denenmiştir.

Alive body, generates certain electrical signals, while performing its different functions. For instance, Electrocardiogram (ECG) for heart, Electroensef alogram (EEG) for brain and Electromiyogram (EMG) for muscle activity. Perceiving and comment on these biological signals, necessary to diagnose and treat of many disease and during the investigations on Biomedical Engineering. But, any biological signal can not be purely percieved from the body. The other biological signals generated in the body and the environmental noises add and corrupt to the desired signal. For example; power line interference on ECG, the mother's ECG noise on fetal ECG, EMG noise on EEG signal and the heart sounds on the lung sounds are generally prevent to comment on the desired signal. The classical method to estimate a signal, corrupted by the additive noise, is to pass it through a filter. This filter, used for the above purpose, can be fix or adaptive. A prior knowledge about the signal and the noise characteristics is necessary to design a fix filter. Adaptive filters on the other hand, can be design without requirement to this prior knowledge and can adjust the own parameters automatically. An adaptive filter consists of, tap weights adjusted always, an estimation response generated with the lineer combination of input set and the tap weights, and an adaptive algorithm controlled by the difference between the desired response and the estimated signal. Noise Cancelling is an application area of adaptive filters. A reference input, perceived from the noise source by the sensors, and a primary input, contained the corrupted signal, are applied to the Noise Canceller. Noise component on the primary input is tried to estimate by filtering the reference input. This noise estimation is subtructed from the primary input. Thus, the noise on the primary input, has been cancelled or the signal to noise ratio of the primary input has been increased. Subtructing the noise estimation from a receiving signal, is an important operation. The output noise power can be increased, if done faulty. But, if the filtering and subtructing operations controlled by an adaptive algorithm, risk of increasing of output noise power has been eliminated. This system called ADAPTIVE NOISE CANCELLER. The recursive algorithm starts from an initial state and tries to estimate signal characteristics. For stationary inputs, after a few iterations, which is called rate of convergence, the filter converges to the optimum Wiener solution. In a nonstationary environment, the algorithm have the capability of tracking the time variations in the statistics of the signal. There is no unique solution to the adaptive filtering problem. Wiener and Kalman Filter Theories can be used for adaptive filter approximation. Transversal filters are used to implement adaptive filter, based on the Wiener Filter Theory. The finite impulse response of such a filter, is a set of tap weights. For stationary inputs, mean square error (mean square of the difference between the desired response and the output of the transversal filter) is a second order function of these tap weights. The relation between the mean square error and these tap weights, is determine a mult i -dimensional surface with a unique minimum point. This surface is named Error Performence Surface. The optimum tap weight vector at the minimum point of this surface, can be calculated by using Wiener-Hopf equation. This equation requires the inverse of the input correlation matrix and the cross-correlation wector between the input and the desired response. VI However, matrix inversion operation takes a long time. Therefore, different algorithms have been developed instead of the Wiener-Hopf equation, for updating the tap weights of the transversal filter. Two of them are Steepest Descent and Least Mean Square (LMS) Algorithms. In the steepest descent method, which is a well-known technique in optimization theory, gradient vector is used to calculate the tap weights. Gradient vector is the first derivative of the mean square error and depends on the input correlation matrix and the cross-correlation vector between the input data and the desired response. Least Mean Square Algorithm uses the estimation of the gradient vector, instead of the real gradient vector. LMS algorithm is simple and does not require to calculate the correlation functions and matrix inversion. But, its relative rate of convergence is low and sensitivity on eigenvalue spread of the input correlation matrix is high. For nonstationary signals, the error performence surface varies with time. The LMS algorithm will track the minimum point of the error performence surface in this case. The desired features of adaptive filters are effectively work in unknown characteristics and tracking the time variations of the input signal statistics. Adaptive filters with the above features, are successfully applied in many fields, like biomedical signal processing, radar, sonar, sismology, image processing, pattern recognition etc. In these application, adaptive transversal filter is used to modelling a dinamic system. In biological signal processing, a biological medium is modelled with adaptive filter. For example, filter in the noise cancelling system designed for a biological signal, is a model of transmission line of the signal in the body and try to estimate the noise. Adaptive transversal filter produces the lineer combination of the of input data set, which are generated from an input signal with the delay line and the tap weights. The estimation error is obtained by comparing the filter output the desired response of the dinamic system. The algorithm, which corrects the tap weights for the next input data, is controlled by this error signal. Vll The first study, on Adaptive Noise Cancelling System has been done by Howells and Applebaum in 1957-1960. Widrow and Hopf have being worked on Least Mean Sguare Algorithm at Stanford University in 1959. The first Noise Cancelling System has been implemented as a term work by two students at Stanford University. In this thesis, an Adaptive Noise Cancelling System, which can be operated in real-time, has been implemented. A High Performence Microcontroller (HPC) has been used, as a processor in the system. Adaptive Noise Cancelling System has been designed to cancel, power line interference on ECG signal firstly. This system, consists of, an Analog/Digital Converter (ADC) unit, a High Performence Microcontroller (HPC) Development Board, a Digital /Analog Converter (DAC) unit, and a computer. The corrupted ECG signal, by the 50 Hz. noise, has been perceived from the body with an ECG amplifier and used as the primary input for the system. This signal has been sampled with 400 Hz. and digitized by the ADC unit. As the reference input, one period of the 50 Hz. cosine and sine waves have been sampled with 400 Hz. Two sets of reference input data have been obtained, and stored to the memory, to use for all primary inputs. An LMS Adaptive Filter program has been developed in machine language of HPC. The output signal of the noise canceller has been applied to the DAC unit, and analog ECG signal has been obtained without 50 Hz. noise. After that, ANC system was modified to take also the reference input from the outside. The corrupted ECG signal has been applied to the first channel of the ADC unit. 50 Hz. reference signal, has been taken from the wall -outlet with a transformer, and applied to the second channel of the ADC unit. The adaptive filtering program has been modified for this application. Thus, the desired ECG signal has been obtained. As a second and developed application, cancelling of the mother ECG noise on the fetal ECG signal has been examined. ANC system has been enhanced to solve this problem. The corrupted fetal ECG signal, perceived from vxxi the abdominal leads, will be applied to the system as the primary input, and the mother's ECG signal, precieved from the chest leads, as the reference input. A fifth-order adaptive filter program has been developed for this system.