Black Box Linearization for Greater Linear Dynamic Range: The Effect of Power Transforms on the Representation of Data

Abstract

Power transformations are commonly used in image processing techniques to manipulate image contrast. Many analytical results, including chromatograms, are essentially presented as images, often to convey qualitative information. Power transformations have remarkable effects on the appearance of the image, in chromatography, for example, increasing apparent resolution between peaks by the factor √n and apparent column efficiency (plate counts) by a factor of n for an nth-power transform. The profile of a Gaussian peak is not qualitatively changed, but the peak becomes narrower, whereas for an exponentially tailing peak, asymmetry at the 10% peak height level changes markedly. Using several examples we show that power transforms also increase signal-to-noise ratio and make it easier to discern an event of detection. However, they may not improve the limit of detection. Power responses are intrinsic to some detection schemes, and in others they are imbedded in instrument firmware to increase apparent linear range that the casual user may not be aware of. The consequences are demonstrated and discussed.

Sampling frequency, response times and embedded signal filtration in fast, high efficiency liquid chromatography: A tutorial

“…The firmware commonly applies a power function to the signal S and the output is S y where 1 < y < 2. This results effectively in a greater LDR and has the (un)intended consequence of making the chromatographic peaks to appear more efficient than they really are [22][23][24]. For the more commonly used optical absorbance detectors, the available choices in detector electronics typically allow a user to choose a particular data sampling rate (a.k.a.…”

Recent applications of chemometrics in one‐ and two‐dimensional chromatography

“…However, these types of analytical systems are not always available and even when accessible, they sometimes still do not provide resolution of all components. Techniques for peak-resolution enhancement, such as even-derivative sharpening (see Section 2.4.1) [82], derivative symmetrization [83], or power-law methods may be used [84,85]. These techniques also have the potential for peak detection, as they highlight any small difference in peak shapes.…”

Wavelet transforms in separation science for denoising and peak overlap detection

“…CWT provides a huge flexibility in the choice of wavelets. Names of continuous wavelets that can be used for detecting peak shoulders or overlaps: Daubechies family (1-9), Symlet family (2-8), Coiflet family (1-5), and Gauss family (1)(2)(3)(4)(5)(6)(7)(8). Gaussian wavelet transform gives the most accurate estimation of frequency components localized in time.…”

“…These problems are currently tackled by exploiting highly selective detection technologies, surface chemistries, and obtaining high peak capacities by using small particles or narrow open tubular columns [1,2]. Another less trodden path for solving separation problems is signal processing for enhancing peak to peak resolution and noise reduction [3][4][5][6][7].…”