| Author | Message | ||
     Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 91 Registered: 9-2001 |
Bruce - I guess whoever wrote the article didn't know the difference between a Latin Square and a full-factorial, then! \o/ /_\ | ||
| Bruce H. Campbell (campclan) Moderator Username: campclan Post Number: 99 Registered: 4-2001 |
Howard, I don't think I can send a copy of the article without getting permission from the publisher. I will ask them. Bruce PS My representation of a Latin Square, excluding typos, was taken from the article. | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 90 Registered: 9-2001 |
Bruce - the design you showed isn't a Latin Square, it's a five-level, 5x5 full-factorial design (BTW - you have a typo: B3 is present twice). A Latin Square design manages to include a third factor into the same number of experiments, while making sure that the third factor is present in the design with all combinations of the first two. For example, using your 5x5 as a basis and indicating levels of the third factor with small letters, a Latin Square would look like this: A1a A2b A3c A4d A5e B1e B2a B3b B4c B5d C1d C2e C3a C4b C5c D1c D2d D3e D4a D5b E1b E2c E3d E4e E5a Check it out: "a" exists with all of "A", "B", etc., also with all of 1, 2, 3, etc. So does "b" and each of the other levels of the third variable. There's also a design called a Graeco-Latin square, that manages to squeeze in a fourth variable, also in all combinations with the first three. I'd have to see the article from the New Scientist to see how they use the Latin Square for noise reduction; it may or may not be appropriate for NIR. Could you send a copy? The requirements for sending digital signals over a power line are considerably different than for the low-level analog signals we deal with in an NIR instrument, after all. \o/ /_\ | ||
| Bruce H. Campbell (campclan) Moderator Username: campclan Post Number: 98 Registered: 4-2001 |
In a recent issue of NEW SCIENTIST (March 24-30, 2007,p. 48) an article went into how digital signals could be sent over power lines with all the attendant noise that may interfer. The gist of the article is the use of a Latin Square to "filter" out the noise, leaving a usable signal. (See below for an example of a Latin Square.) After reading the article I thought this may be a way to decrease the noise in NIR (and other detection systems) thereby improving the precison as well as the detection limit. Basically the illustration shows a five by five Latin Square generated with a basic signal in box A1, twice the fequency in Box A2, three times the frequency in Box A3, etc. Box B1 has the same frequency as Box A2, box B2 has the sampe frequency as box A3, etc. Boxes C1, D1, and E1 have the same frequencies as boxes A3, A4 and A5. The rest of the Latin Square box values should be evident from the above pattern description. Anyway, if noise is present in, say, box A3, the logic of the Latin Square can remove that noise. Different noise in more than one box can also be removed. If n boxes exist, the system can filter out n-1 noise outbreaks. Here I get to the crux of the discussion. When NIR spectrometers do multiple scans, could not the first scan be "put into" box A1, the second scan into box A2, etc? And could the Latin Square be then used to remove most if not all the noise? I don't have enough of a mathematical background to answer this question. (Note: Euler was the discoverer of Latin Squares.) Bruce The Latin Square is an n by n array with a 5X5 array shown below. A1 A2 A3 A4 A5 B1 B3 B3 B4 B5 C1 C2 C3 C4 C5 D1 D2 D3 D4 D5 E1 E2 E3 E4 E5 | ||
| yiwu |
I want to detect proteins in some pharmaceuticals (both in liquid and solid samples), but the concentrations of the proteins may be less than 0.1% level, 0.01% level or just a few ppm. So my purpose is to enhance the sensitivity of NIR, or maybe decrease the detection limit. any idea is welcome | ||
| Michel |
NIR is probably not your best choice for low concentrations. Much will depend on the (stability of the) background material (solvent). If you work in very pure solvents and your analyte concentration is the biggest change between samples then these concentrations could be feasible. If your background is organic and has few OH groups, you can seriously increase your pathlength and in this way improve your chances. If you have a fluctuating background, I would consider a different technology (Fluorescence?). | ||
| hlmark |
I have to agree with Michel: using NIR to measure concentrations below 0.1% is risky, at best. But if you're bound and determined to do it, then to maximize your chances of success, in addition to Michel's recommendations you want to start by taking the best spectral data you possibly can. This would include things like stabilizing the instrument, improving it's noise performance, controlling the sample temperature, etc. I would also recommend that you start by trying to measure liquids (clear, non-scattering liquids, if possible), to avoid all the messy physical effects that you encounter with solids (particle size variation, etc.). Watch out for effects such as variation in stray light, reflections from the surfaces of the sample cell, and any other optical effects that might arise. You might have to do some customization of your instrument to improve its stability and noise characteristics. Of course, you can also reduce the noise by co-adding spectra; you should do as much of this as you can spend measurement time on. Longer measurement times will, in turn, require more stringent control of any and all potential causes of drift. Your measurement is likely to be marginal in any case, so this may be a situation where you will want to take advantage of known noise theory: the best signal-to-noise is obtained when the pathlength is such that the absorbance at the measurement wavelength (the peak of the protein band) is 0.3 - 0.35 (ideally 0.33, but the curve is fairly broad there, so some leeway is available). Howard \o/ /_\ | ||
| yiwu |
Thanks for your suggestions! I have to agree with both of you. NIR is not good method for trace component, but as Howard said, I am bound to detect trace protein in both liquid and solid smaples using NIR. That is my project. I currently use FT-NIR spectromether (Spectrum one NTS, PerkinElmer), but I am not sure if it exactly fit into my experiments. Does anyone know which kind of instrument is more suitable for low concentration measurment? Thank you, Howard. But how can we customize the instrument to improve its stablility? Would you mind giving some examples? Regards, yiwu | ||
| Michel |
+ aircondition the lab/instrument + reduce ambient light (paint the table/wall black) + sample holder which is thermostatically controlled (thermo-electric or circulate to waterbath) + the colder your detector is, the better: check if it is possible to cool the detector + buy one of these "smart UPS" which you put between the 220Volt and your instument. It cleans your incomming electricity. (www.apc.com) + keep some distance between instrument and PC (EMI) | ||
| hlmark |
Yiwu - hard to say that any one instrument is "best". All are designed for general-purpose use, and your application is definitely "pushing the envelope" as they say. Michel's pieces of advice are all good ones, but except for cooling the detector, all of them involve stabilizing the instrument's environment, which is a good thing to do, but may not suffice. Once you start making changes to the instrument itself, even just cooling the detector, you get involved in re-engineering it. This is because normally, each part (or at least each section) is designed to work best with the other parts or sections as they are also designed, so when you change one, you may find that you have to change others in order for the original change to have the desired effect. For instance, if you cool the detector or, more drastically, change to a different type of detector, then you can reduce the noise of the detector. But you will also change the detector's other operating characteristics so that the reduced noise may not come through to the final spectral values unless you change other parts of the signal-processing channel as well. For example, you may need to use a lower-noise preamplifier, or one that has a different input impedance, or provides a different voltage or current to the detector, or has some other characteristics that are different than the one in the off-the-shelf instrument. Even then, to realize the benefit of lower noise you may also need to change the A/D converter, to one with more bits than is currently installed. Some of these changes will require fairly extensive modifications to the instrument, and should not be made lightly. There is a domino effect here: if you make these changes then you may find that you need to make still other changes (including the software, perhaps) so that you wind up re-engineering the entire instrument. At the very least, you should get advice from the manufacturer, if they are willing to work with you on this. | ||
| hlmark |
Yiwu - I just thought of another part of the environment you may want to stabilize. Taking a lead from mid-IR spectroscopy, eliminating water vapor in the optical path will eliminate variations due to humidity changes. Due to their history, the design of most FTIR instruments include provision for purging the air inside the instrument, and air dryers are commerically available for this purpose. We don't usually need to do this in the NIR, but when you push the envelope you need to consider all possibilities. There are also devices that remove both moisture and CO2 from the air (alternatively, if liquid nitrogen is readily available, the headspace nitrogen over the liquid is eminently suitable for this purpose, as well). Howard \o/ /_\ | ||
| yiwu |
Howard - As your suggestion, my current attempt is measuring liquids in a wavelength spectral window (not whole NIR region, for example from 950 to 1050nm), and the results seemed not bad. But here I have two questions: 1) Is dispersive spectrometer better than FT one in a narrow spectral window (100nm or less)? 2) Does the buffer effect the detection limit? Same concentration protein in water or in tris-, is there any difference? thanks yiwu | ||
| hlmark |
Yiwu - As I tried to indicate before, there is no general answer to "FTIR versus dispersive?". When you're pushing the limits of the technology you have to consider the performance of a particular FTIR versus a particular dispersive instrument, for the particular measurement. You also have to consider the question of flexibility of the instrument in regard to modifying the measurement parameters without making major changes to the instrument itself, in order to optimize the measurement you want to make. Ultimately, however, the limiting concentration you can measure will depend on: 1) The energy available from the source 2) The noise performance of the instrument (detector, etc) 3) The width of the absorbance bands of the analyte 4) Stability of the instrument 5) Time available for the measurement In this analysis I'm ignoring the errors due to other (interfering) constituents in the sample, since we're concentrating our attention on the instrument for now. But the listed characteristics are independent of whether the instrument is FTIR or dispersive, but very much dependent on the design and construction of a particular instrument, of either type; these are the fundamental limits to a spectroscopic measurement. I'm also wondering about the choice of wavelength region you are using for your measurements. You may want to consider trying to use a different wavelength range, particularly the range 1100-2500 nm, at least so that you can compare the results to see which range is better. Are you being limited by the type of detector in the instrument you are using? If so then you may want to consider a different instrument for that reason. I do have some more detailed comments I could make, but I think that discussion is not appropriate for this forum, so I will send you an e-mail message and we can hold further disucssions. Howard \o/ /_\ |