Resolving the Resolution Myth

Abstract

Myth: “High resolution is always favourable when recording NIR spectra—just to be safe...”. First of all, this myth is an outcome of modern instrumentation with adjustable spectral resolution power, whereas in earlier days one would settle for whatever “the box would offer”. The prevailing feeling nowadays often is that it is always better to collect spectral data at high resolution—just to be safe—although, paradoxically, it often turns out that one would have been better off with a lower reasonable resolution for most general NIR applications. Peak resolution has a radical effect on spectral data quality if settings are not appropriate, especially for FT-NIR systems. This column is a vade mecum for those for which this issue seems moot. A solid hint that you have applied overresolution, and therefore acquired noisy data, is when smoothing or large “windows” are needed when applying derivative pretreatment. Whereas earlier NIR equipment was dedicated to a specific task, enabling the manufacturing factory to preset factors like range, amplification, resolution and signal-to-noise ratio, the introduction of FT-NIR technology introduced a new, vast range of opportunities to collect data with less than optimal parameters, and also to acquire data of better quality than ever before—critically dependent, however, upon a certain knowledge and experience. The following should in no way be construed as harsh words directed at FT, but, rather, directed to the occasional less than optimal use of this powerful technology. With this we do not advocate going back to poor resolution, we merely state that there is no safety in blindly collecting data at high resolution as will become abundantly clear below. A crude way to put this would be whether it is better: “having less data of high quality—or more data of lower quality?”—there is clearly some kind of dilemma here. But let us first define resolution properly as some confusion exists. For example, spectral data point spacing, pixel coverage or slit size is not about resolution—and there is spectral, spatial, temporal and numerical resolution as well. For the present purpose, we speak only of “peak resolution”, which we could define as follows: resolution is a measure of the capability of an instrument to fully separate two adjacent peaks. This definition is clearly more relevant in chromatography and mid-infrared FT-IR spectroscopy than in the NIR region but it is still relevant. Resolution capabilities are defined rather differently by various OEM producers and software technologies. While a full comparison would be nice, it is too much for this column, so below we only introduce a few basic issues, enough for our didactic purpose. We suggest always running a few spectra of known samples with very narrow and well-defined peaks to find the right acquisition settings before starting off on a large dataset and at the end-of-the-day perhaps finding that noise harms the possibilities for good chemometrics more than lessthan-fully resolved peaks. For example, one could quickly check for noise with a sevenpoint second derivate SG algorithm to see if your data stills looks OK. Below follow salient illustrations of this important issue.