Wang Ying, Mo
Jinyuan Received Jul.13, 2002; Supported by the National Natural Science Foundation of China (No. 29975033) and the Natural Science Foundation of Guangdong Province (No. 980340 and 01237) Abstract A new method named Threshold
Fitting Technique is developed, which is applied to estimate drifting baselines of CE
signals. It uses threshold to remove peaks from the signals and least square fitting with
Mexican Hat wavelet to obtain smooth baselines, then the baselines can be subtracted.
Simulated and experimental signals are proceeded. All the results are satisfactory. This
method solves the baseline drift problem in CE signals successfully. And this technique
can process signals with high noise directly, too. Capillary electrophoresis ( CE ) is routinely used for analysis in a wide area[1]. It has been proved to be an excellent technique for the separation of mixtures with a high separation efficiency, fast analysis time and low consumption of reagents and samples. However, the signals obtained from CE often include drifting baselines, resulting in the obscuration of useful information, changes of peak's shape and large errors as the calculation of peak areas. Thus, accurate analysis on the qualitative or quantitative test would be limited with drifting baselines. In order to improve the baseline data, a baseline subtraction technique is employed by the separation of the peaks and baselines. However, such baseline processing method has not been found in the previous literatures. Also, no techniques about baseline processing for CE signals have been reported. Wavelet has been put forward only in a short time, but it has become a hot topic in different areas quickly owing to its excellent characters[2]. Wavelet as a high performance signal processing technique leads to new methods for signal processing. It has begun to be applied in analytical chemistry for signal treatment in recent years[3]. On the baseline extracting, a valuable approach using wavelet[4], which has been reported to be used in HPLC, divides the signal into high frequency and low frequency regions. The low frequency region is regarded as the baseline. But when it was applied to CE signal treatment, the baseline would be found to be distorted because there is no distinct boundary between the baseline's frequency and the peak's frequency. Thus, to develop a technique for baseline subtraction is urgent and useful for CE signal treatment. In this paper, we will describe a newly developed method using wavelet technique, named threshold fitting technique ( TFT ). TFT adopts threshold to remove peaks in signals and uses least square fitting with Mexican Hat wavelet to obtain smooth baselines. TFT can subtract baseline accurately from CE signal even when it has high noise. 2 THEORY The base of TFT is curve fitting. The principle of curve fitting is introducing a new function f(x) - fitting function, to approximate the original signal points which are a series of discrete data points . During the course of fitting, a criterion called least square must be followed. The least square makes the sum square of errors between the new function and the original data points to be minimum. That is minimize ( 1 ) where and are the ith point on the original signal and the fitting function respectively, k is the number of signal points, is the weight coefficient which always equals to 1 in this paper. The function f(x) is called as the useful signal extracted from the original one. Fitting function is a key of curve fitting. Here, Mexican Hat wavelet[5] ( 2 ) is applied in the fitting function. It has a simple explicit expression and a smooth figure. The value of Mexican Hat wavelet function reduces rapidly with the coefficient. It is like the same characteristic with human watch in space. With this reason, Mexican Hat wavelet is suitable as a fitting function. When the original CE signal, which has a typical form shown as Fig. 1 at curve a, is fitted by Mexican Hat wavelet followed least square criterion, it is found that the fitted result (see Fig. 1 at curve b ) does not agree with the original signal well on the whole curve. They show an agreement during the segments where the curve varies slowly, those are the regions between signal peaks. But, at the sharp peak regions, the fitting result departs strongly from the original curve and the peaks become much lower and wider as shown in Fig. 1. According to the above theory, the fitting technique leads to a smooth and slowly changing curve as the result of considering all the points on the whole curve. At the peak regions, a few points are away from the majority of points, so the fitting curve will not pass through these disparate points. But those points do take part in the fitting, so the fitting curve will protuberate to the direction of peaks in virtue of the peaks' influence. Therefore, the fitting curve forms lumps at the peak regions ( those are the segments where peaks add to the baseline), but it can accord with the original signal at other regions ( those are the segments contain the baseline only ). In order to obtain an even and exact baseline, the primary task is to eliminate the pumps at the peak regions. In estimating the baseline, the points on the signal curve that lie on top of peaks can be considered as outliers, and thus one can imagine using a technique to subtract a baseline by ignoring those points that lie on peaks. Therefore threshold is introduced. Fixed threshold requires judgement by the operator ( which may introduce bias ) and sometimes it is impossible to find a right threshold in advance. Threshold in TFT is decided by the arithmetic automatically and achieved gradually. Fig. 1 Signal and its fitted curve a. simulated CE signal b. fitted curve of a Fig. 1(B) is a part of Fig.1( A) after magnified. M and N are the cross points of curve a and b. The process of TFT is going through the
following steps: 3.1 Reagents Roxithromycin dispersible tablets, MeOH/formamide ( 50/50 v/v ), supporting electrolyte: 10mmol/LNH4AC-2mmol/LHAC. 3.2 Apparatus High performance capillary electrophoresis with amperometric detection system, Spellman High Voltage Electronics Corporation ( CZESOPN 10MCNZ2 ) were employed. The electrodes were platinum electrodes. A micro computer was used to process the data. 3.3 Data processing All the data processing can be performed with our self-written program. During the course, the threshold changes and reaches the ideal values step by step. 4 RESULTS AND DISUSSION Fig. 2-D shows a peak of simulated CE signal, Fig. 3-B shows a peak of experimental CE signal. From the figures we can see that the signals have drifting baselines, so that the calculation of peak areas will meet much difficulty, such as how to find the starting point and the end-point of each peak, how to deduct the blank value at the peak regions. But if we can estimate the baselines and extract them from the original signals, all the difficulty will be solved. 4.1 The processing of simulated signals As a new technique, simulated signals are processed at first to test TFT's performance. We have simulated the signals with different kinds of baselines. Each result of subtracted baselines was satisfactory. Fig. 2-A is an example of simulated signal which is composed of two peaks and a beeline as baseline. Its estimated baseline is shown in the same figure. It is shown that the baseline is estimated accurately. Fig. 2-B is another simulated signal whose baseline is a sine wave. The result of TFT ( Fig. 2-B at curve b ) tallies completely with the theoretical baseline. When signals synthesized by peaks with different numbers, different positions, different heights, different widths and the same baselines are processed , the estimate baselines are still shown to be the same. It indicates that the results are not affected by the outliers ( the peaks ) .
Fig. 2 Simulated signals Table 1 The relative errors of peak height and peak area
4.2 The processing of experimental signals
Table 2 The quantitative results of experiment signal
TFT does provide a powerful technique for estimating baseline of CE signals. It is very simple and easy to apply in the treatment of CE signal, it avoids the intervention from operators and the results are accurate. In addition, TFT is not tailed for specific data sets, it can be applied in many kinds of signals of analytical chemistry to subtract the baselines. That will make the analyses more precise. REFERENCES
¡¡ |
|