| Author | Message | ||
     Susanne Wrang Bruun (Swb) |
Hi, I want to detect the molecular interactions in dough and hydrated gluten samples by use of NIR. I have measured in a sample cup in reflectance mode on a FT-NIR instrument. There is a lot of noise in the spectra, but I think that if I just smooth the spectra I will maybe also remove any relevant information. Can anyone suggest me what I can do to get less noise in the spectra? Is there a better way to measure these kinds of samples? Thanks, Susanne | ||
| hlmark |
Susanne - First, make sure that the instrument is performing up to it's specifications. Assuming that the instrument is OK, the best way to reduce noise is at the source, i.e., during data collection. With most FTIRs you have control over the number of scans you co-add, and the resolution setting, and using different values for already-available parameters is the simplest and most straightforward way to obtain improved results. Increasing the number of scans will improve the noise by a factor of the square root of the number of scans. More scans means scanning for longer times, and due to the square root factor you run into diminishing returns, but basically it's a matter of trading off time for noise level: scan for as long as you possibly can. Reducing the resolution setting will also reduce the noise. This has the additional advantage of reducing the time each scan takes. Therefore you can co-add more scans in a given amount of time, giving you a double bonus. A possible caveat here is the same as smoothing the spectra: if you do too much you can distort the spectral peaks. However, from your preliminary work you should have some idea by now of how broad your spectral peaks are, and so you know to what value you can reduce your instrument resolution. However, samples such as dough generally have very broad bands, and so you can reduce your resolution to fairly small values (large bandwidth). Some of this also holds true for smoothing: if your peaks are broad and spread over dozens or hundreds of individual wavelengths, smoothing can be (cautiously) done without losing information. You have to empirically try different amounts of smoothing and see if you can tell when you start to distort the peaks or change their absorbances. Beyond changing the instrumental parameters, there are changes you can make to the hardware to reduce the noise: many manufacturers, and third-party companies, can provide different components that will improve the instrument performance. The one that will probably give the most improvement is to intall a lower-noise (more sensitive) detector. However, this should be done in collaboration with either the manufacturer or someone else who can guide you through the process, because changing only the detector may not give optimum results. This can happen, for example, if the new detector's output is too low to fill the A/D converter, or too high and saturates either the A/D or the instrument's preamplifier. In either case you will not obtain the improvement you should achieve unless you compensate for the new detector's characteristics. There are many other ways to improve the noise performance, but all of them will require a greater or lesser amount of effort (and possibly other resources). Howard \o/ /_\ | ||
| Gabi Levin |
Hi Sussane, I agree with Howard, increasing the numebr of scans is one way, decreasing the resolution will buy you back some of the time lost in increasing the number of scans. Based on my knowledge of the spectral features from the product you describe, you don't need high resolution, the peaks are quite wide, and there is nothing there that shows sharp peaks. The importance of time is due to the fact that your system of composition is "alive" i.e., the composition changes due to the processes that take place in the dough, thus taking too long to scan creates another problem because you will be averaging a moving target, and possibly the correlation in time to your reference sampling will be distorted. If you pull a sample for reference measurements, "freeze" the processes so the sample will not change before the measurement is done. Wish you the best Gabi Levin |