Study On Signal Filtering And Parameter Extraction Of Overlapped Peaks

With the deepening of the study in the fields of medicine, chemistry, biological and etc., the analyte usually tend to be complex. Due to the experiment conditions, the material properties, the instrument resolution and other factors, components with similar wavelength will influence each other. Peaks mixing and overlapping, which directly affect the qualitative and quantitative analysis of signal. Besides, in the process of signal acquisition and transmission, the interference of noise is inevitable. Therefore, it is a key issue to filter and enhance the noisy signal, and extract accurate parameters.After overview of merits and drawbacks of some traditional overlapped peak resolution methods, taking the Gaussian smoothing function as tool to overcome the influence of noise on peak resolution and enhancement, two signal enhancement methods with noise immunity were proposed. Then curve fitting was adopted to extract parameters of the overlapped peaks. Three main results were shown following:1. A peak-sharpening method with noise immunity was proposed based on traditional peak sharpening algorithm. Taking the 2nd order derivative of the Gaussian function as sharpened template, then, convolution operation was performed between sharpened template and the noisy signal, thus, a sharpened signal will be obtained. Finally, some simulated signals and a mass spectrum were conduct to verify the algorithm. Results indicated that the proposed peak-sharpening method is an efficient method. As a result, it can be directly used to distinguish peak numbers, extract the peak positions of the characteristic peaks, which can provide effective information for the follow-up extraction of characteristic parameters.2. In order to decrease the influence of noise on the differential of signal. A fractional order differentiation method with noise immunity was proposed. The fractional order differential of Gaussian function was utilized as smoothing operator. And then convolution operation was performed between smoothing operator and noisy signal to obtain noisy signal’s fractional order differential. By this method, noisy signal’s fractional order differential can be obtained directly, which simplified the progress of processing. Not only integer order differentiation, but also arbitrary order differentiation of a signal can be acquired. The effectiveness of proposed method was verified by differential signals between a signal with or without noise.3. In general, peak numbers and peak positions can be obtained by peak enhancement or peak sharpening, while peak widths and peak heights cannot be obtained directly. Our research team has proposed and verified the method that combined continuous wavelet and curve fitting to extract peak parameters. Based on this idea, sub-peaks’initial parameters were extracted from sharpened signal. And then these parameters were used as initial guess of curve fitting to obtain final peak parameters. Specific progress is implemented as follows:First, utilizing the proposed signal enhancement method to sharpen peak signal. The purpose is to distinguish the overlapped peaks. Then, taking 5 or 7 points around-each peak as peak fitting data, initial parameters of each peak can be obtained. Final parameters of each peak were obtained by taking the initial parameters as initial guess of curve fitting. An advantage of the proposed method is that it can avoid the nonuniqueness of curve fitting and make the result more accurate. At last, verifications were performed by fitting multi-peak signals and effects of peak height, peak width, separation resolution and other factors on fitting results were analyzed in detail.Research results showed that, unknown overlapped components signal can be resolved and parameters can be extracted, which laid a foundation for overlapped peaks signal resolution and separation...