As referenced in a previous Note: http://publiclab.org/notes/stoft/05-14-2013/in-search-of-spectrometer-measurement-and-calibration , here are some observations on adding external attenuation for the spectrometer's incoming light. As most have probably observed, it is difficult to adjust the light level to provide a strong signal while avoiding RGB saturation.
These observations all show the relative signal change when adding attenuation. A "reference level" is first established and then attenuation is added while looking for decreasing signal strength.
Neutral Density DSLR Filter An adjustable neutral density filter was placed a few inches in front of the spectrometer's input slit and the relative spectral attenuation curve was captured for each of a series of 5 arbitrary attenuation filter settings.
DSLR ND Filter Relative Attenuation:
Notice that simply adding the ND filter produces a transmission response which is clearly quite variable with respect to wavelength. This is not one of the really expensive ND camera filters so non-uniformity is expected. However, note that the curves all have similar shapes. It seems that ND filters have a flatter response from 400-600nm but then peaking at ~700nm. The response plotted here is similar. The obvious question is how the filter response would be calibrated.
Transparency Film Attenuators Purchasing an expensive ND filter is probably unwarranted for a relatively inexpensive device which is still in development. As a cheap alternative, grey-shaded rectangular areas of InkJet transparency film were printed. The image pattern was created at 600dpi and printed in random-dithering mode for uniformity and the "pattern area colors" were all shades of grey. 8-10 patterns can easily be printed per sheet.
The printed transparency attenuators have the advantages that they are cheap to make and can be "stacked" in combinations and the attenuation values repeated "by the numbers". The downside is again, calibration. At a single wavelength, like ~600nm, a preliminary set of transparency filters appears to show a linear attenuation. In this plot, grey values were referenced to 128 and attenuators of 8, 16, 32, 64 and 96 were added to get variable attenuation settings.
Printed-Transparency Relative Attenuation:
Other Observations: 1) Neither inexpensive (~$40) DSLR ND filters nor cheap transparency filters have a flat transmission response vs wavelength. 2) For use with this webcam spectrometer, the DSLR ND filter has no clear measurement advantage. 3) Attenuation filters do make setting light levels (to maximize signal and avoid RGB saturation) much easier. 4) No single attenuation is sufficient -- a range of attenuation, to mix-n-match, helps. 5) Amplitude calibration remains a difficulty; even with attenuators.
Dynamic Range and SNR remain a significant issue for the spectrometer but Printed Transparency attenuators might provide cheap means for making measurements easier. However, these are only preliminary results and only represent a starting point.
Cheers,
Dave
Added: 5/15/13 A set of attenuation curves using "known" values could provide some view to system linearity with respect to wavelength. So, a set of Transparency Attenuators were printed with a range of value (16,32,64,128,192) representing specific grey-scale density. If the system is linear, then plots of transparency grey-scale vs measured signal amplitude change should show straight lines even if the sensitivity is wavelength dependent.
So, the signal level change was obtained and the data re-plotted. Notice that this plot looks very similar to the ones above and shows the change in sensitivity over wavelength.
Now, for every 50nm over the 300 to 800nm range, the measured signal attenuation is plotted against the transparency attenuator value.
Note that there is reasonably consistency in an approximately linear response of the system -- enough so, that I suspect that any non-linearity is not the imager chip but results from a variety of other factors. Transparency film is certainly not equally transparent at all wavelengths, but being extremely thin, those effects are minimized.
Suggestions for printing transparency attenuators: 1 - Use 600 dpi for the image 2 - Use the finest print resolution and use the dithering print setting 3 - Grey levels below 8-16 are of little use 4 - Use increments of 16 or above 5 - Grey levels above 192 may not print uniformly - it would be better to double-up
Dave
14 Comments
Hi, Dave - very interesting... is this fairly related then to the suggestion of an "HDR" neutral density filter where we could get 0/25/50/75/100% light and achieve both ideal exposure for any wavelength as well as a plot of linearity of the sensor response?
Also -- an optically interested friend suggested using polarizer film at different angles to achieve varying amounts of neutral density filtering. That could be a relatively expensive technique if it is even across wavelengths, but I don't know that much about polarizers.
Is this a question? Click here to post it to the Questions page.
Reply to this comment...
Log in to comment
Jeff, Thanks. I've just posted an addendum at the end of this research note to add some plots specifically looking for linearity. I believe the image sensor is, in fact, linear (assuming the chip exposure is stable at whatever level -- max gain I believe). I also believe that the anomalies I see are artifacts of a combination of other issues -- namely the system "un-flattness" for lack of a better term -- plus the overall "un-flattness" of the system response to be understood.
Besides the very low cost, the print-transparency attenuators have the advantage of fixed attenuation values. So, if we assume linearity, then fixed attenuation provides more value for measurement ... in addition to just preventing signal overload. (There is another, simple method for gross-level attenuation of a source like an incandescent lamp -- just place a cardboard baffle with a slit (like 1/8-1/4-in wide) in front of the lamp. This also cuts down on stray light and reflections.)
On using fixed attenuators. For example: You were attempting to observe fluorescence in the presence of a vary large signal where the wavelengths were well separated. If you use a fixed attenuation to keep the source signal from saturation (then record it's peak) and then remove that known attenuation and record the peak of the target signal, you have (theoretically) a measure of the difference -- whose value is greater than the dynamic range of the image sensor itself. The non-uniformity of the sensor vs wavelength might still be an issue -- I just don't yet understand why the system exhibits so much variance. A fixed attenuator can be roughly "calibrated" for any single wavelength of interest by providing a strong, non-saturating signal at that frequency (from any source) and then inserting a fixed attenuation (but while keeping the signal above the noise). The measured difference, at that target wavelength, is now the calibration for that attenuator for that wavelength.
ND filters are just a pair of polarizing filters -- one mounted the revers of the other so that 180-deg rotation goes from a min to max attenuation that is continuously variable. Good for photography but w/o fixed "known" settings of attenuation. Unless there is an additional way to calibrate them, they become less useful for amplitude measurements. I think there are fixed-value filters, but that would really multiply the cost.
Dave
Reply to this comment...
Log in to comment