As a student and a hobbyist, getting access to nice proper equipment can be quite difficult. Using the wrong sort of equipment can lead to poor results and big hassles. In the recent review of the OSRAM LIGHTIFY system, I was using a fairly old spectrometer on loan to try and characterize the spectral performance of the lamps with inconsistent results which implied problems with the calibration and response of the spectrometer in use. I marked those results as “preliminary” and with a disclaimer to ignore the absolute figures due to doubts over their validity.
Of course, I personally wasn’t too pleased that it had to go out like that, but I wanted to make sure the review was delivered in a timely manner. Finding a solution to the issue wasn’t necessarily going to be quick and easy.
Knowing the spectrometers cost about AU$3,000, to calibrate one would require a reference light source of about AU$2,500 which is quite expensive. A hand-held lighting meter which could perform CCT, CRI, Lux measurements costs about AU$2,200. The reviews I have done to date are often blessed by the test equipment I have hard-won through review processes and have been put to good use in assessing other products for the internet community as a whole. My budget obviously doesn’t extend that far.
Calibrating the Spectrometer
As a result, I turned to a method which I had used in my research to calibrate the spectrometer – namely using the sun as a reference source. The spectrometer had a small aperture made of a pin-prick hole in a piece of foil attached to the input to attenuate the incoming sunlight, which was measured through a break in the clouds today (27th November 2015) at solar noon (12:42:49 Sydney time). Should it have been fully overcast, I would have had to wait for another day.
The recorded value was compared with the spectral shape as simulated by SMARTS v 2.9.5 solar modelling software using similar inputs to IEC 60904-3:2008 adjusted to compute the spectrum at the given date, time and location. This allows us to develop a set of calibration/compensation values to improve the result. Again, the resulting corrected spectral data was analyzed using a MATLAB script from the Rensselaer Polytechnic Institute’s Lighting Research Center.
The resulting correction multiplier values show that there is indeed a very low sensitivity in the blue wavelengths as I had surmised earlier. At 450nm, it is about half as sensitive as it should be. In the green region, it is a little more sensitive than it should be, but on the whole, this is the sort of thing you expect. Each spectrometer has a range of working wavelengths and a small region of best-efficiency. As a result, it is important to calibrate them to get the most accurate performance – my reliance on the supplied typical grating performance was inadequate.
The bumps in the curve represent atmospheric absorbance bands which may not have been clearly resolved by the spectrometer, or may represent differences in the real-life atmospheric conditions and those of the model. The differences themselves are not too great, but do result in the corrected spectra growing many “bumps”.
However, the correction factors now do provide a more accurate picture of the relative intensities of the LED wavelength components than before. Below about 430nm, the rising values that appear are the result of the correction factors amplifying measurement noise and the LEDs do not really emit significant energy in the UV-A portion of the spectrum.
The resulting new CCT and CRI figures are above. The script ran into an error calculating the CRI for the 6500k setting, the reason is not known at this time. It is likely that these values better represent the real performance of the globe, but may still not be entirely accurate due to low signal-to-noise ratio of the blue measurements between 370-430nm which are used in calculations. Comparing with a test done by Cnet which tested the Tunable White globe, they reported 2680-5273K and CRIs ranging from 78-83, which is vaguely similar to the results obtained.
A low-cost, rapid method of calibrating a spectrometer was trialled and served to improve the accuracy of the recorded data. Recalculated values for the light output are now more representative than those based on the typical grating efficiency curves as calculated in the original review, and are comparable to reported figures elsewhere.