Research On Implementation And Application Of Wavelet Transform Using Switched-current Technique

Wavelet transform (WT) is a widely used mathematical tool for signal processing,due to its good estimation of time and frequency localization. It is also one of themost effective tools for local analysis of nonstationary and fast transient localizations.In order to meet the application requirement, hardware implementations of WT havebeen be researched in the passed few years. However, nowadays, systems employingthe WT are implemented using digital devices. Since these digital systems requireanalog-digital (A/D) converter to transfer analog sensor information to the digitaldomain for processing analog signal, the WT implementations by means of digitalcircuits have the disadvantages of large chip area, high power consumption and hightransmission delay, which is in contradiction with the trend of low voltage, low powerconsumption and low cost for implementing WT, and has hindered the application ofthe WT. Because the real-time operation and power consumption of analog system aresuperior to those of ditigal systems, the analog implementation of WT has attractedmuch attention. In the design of analog integrated circuit, current mode circuit hasbecome the main technology, which the circuit has the advantages of low voltage, lowpower consumption, wide frequency band and large dynamic range. Switched-current(SI) technology as the representative of the analog sampled-data signal processingcircuit has received much attation. The time constant of SI circuit is only relative tothe ratio of component parameters and clock frequency, so the accuracy of the SIcircuit can reach very high. Furthermore, the expansion coefficient of the SI circuitcan be accurately obtained by adjusting the clock frequency and they are fullycompatible with current trends in digital CMOS process technology, because they donot require linear floating capacitors. Therefore, SI technology is very suitable forimplementing analog VLSI WT with low voltage, low power and multi-scalecharacteristics.At present, the WT implementation using analog filter is the main method. Thework is concentrated on three issues:(1) Wavelet approximation.(2) Wavelet filtertopology design.(3) Wavelet filter circuit design. Over the years, the design methodsfor the above issures have been researched, but still have some shortcomings asdescribed below: Frist, Padé and L2approximations is non-ideal such as the accuracy,the stability and the convergence. Second, most filters employ cascade and ladder topologies, which has highly sensitivity and complex circuit architecture. Finally, thein-phase integrators are usually used to implement the desired wavelet filter. However,the literatures about inverse integrators, inverse differentiators and multi-outputcurrent mirrors (MOCM) to implement the wavelet filter are rarely reported.To address the above issues, the implementation theory and method of WT usingSI filter are researched by combining the WT theory with SI technology in this paper.The design methods, steps and improvements of SI WT circuit are systematicallypresented. The practical applications demonstrate the proposed method is feasible.Themain research work in this dissertation focuses on the following several aspects:1. The implementation principle of WT using analog filter are firstly analysed.Then the implementation scheme and procedure are summed up, which provides theclear design ideas for ananlog WT synthesis. Moreover, the WT implementationemploying SI filter is specifically analyzed.2. The time-domain and frequency-domain approximations of the wavelet havebeen investigated. According to the theory of linear systems, the universalapproximation model has been firstly deduced and structured, which is suitable forapproximating the arbitrary wavelet function and has advantages of strongcommonality, high accuracy and good stability. Secondly, the frequency-domainapproximation of the wavelet function is presented based on function-chain neuralnetwork (FLNN) and frequency domain function fitting (FDFF) approach. Amongthen, the FLNN method can obtain the simple transfer function and may implementthe wavelet filter with compact structure. The FDFF has advantages of high accuracy,simple solution procedure. The experimental results show the feasibility of ourmethods, which enriches and developes the time and frequency domain approximationmethod of wavelet function.3. The approximation theories and schemes of complex wavelet have beenanalyzed based on wavelet approximation approaches. Subsequently, theapproximation method of complex wavelet has been presented based on the modifieddifferential evolution algorithm (MDE) and multi-objective optimization (MOO)strategy. In order to simplify the approximation network structure, we have focusedon the same-poles approximation algorithm (SPAA) of complex wavelet. Thesimulation results verify the effectiveness of the proposed complex waveletapproximation which has high precision, simple approximation networkcharacteristics.4. The multiple-loop feedback (MLF) structures of discrete-time filter (DTF) have been investigated in the analog filter topologies. The follow-the-leader feedback (FLF)and inverse follow-the-leader feedback (IFLF) structures are proposed based on SIinverse integrators, inverse differentiators and MOCM as the main building blocks.And then, the calculation method of each parameter in the structures has been given.The flexible design methodology and simple circuitry of MLF SI filter, combined withtheir lower magnitude sensitivity to component various, offers for analog WTimplementation. These structures are particularly suitable for high order SI waveletfilter design.5. The implementation of SI wavelet filter has been researched. To begin with,Simple SI wavelet filter based on FLF structure with inverse integrators and MOCMas the main building blocks is presented. The simple transfer function is obtained byFLNN method and then the FLF SI filter is used to implement the approximationnetwork. Next, the SI wavelet filter based on IFLF structure with inversedifferentiators and current mirror circuit as the main building blocks is presented. Thetransfer function of the wavelet filter is gained using MDE algorithm, then the weightcoefficients of the filter network is realizated employing current mirror circuit.Finally, the SI complex wavelet filter (CWF) with shared structure is designed basedon SPAA. The results demonstrate the advantages of the proposed SI CWF.6. The application about ECG detecting has been investigated based on SI WTcircuit. A novel method for detecting QRS of ECG based on SI WT circuit ispresented. According to the principle of QRS detecting based on WT, the QRSdetecting scheme is proposed based on SI WT circuit. Subsequently, the SI WT andWT modulus maxima (WTMM) circuits are designed. The experimental results showthat the detecting results of QRS using SI WT agrees with the ones using software.Furthermore, this approach is particularly suitable for the ECG detecting applications,which are low voltage, low power consumption and high speed.