Btw, if you want to assess the noise, you could measure the light source say 100 times, and compute the standard deviation of the 100 readings at each spectral band, and/or plot the 100 curves on top of each other in order to visualize the same (Make sure though, that the light source is stable during the time of the measurements, otherwise you might mis-interpret a drift of the light source as noise, unless you'd estimate and compensate the drift). And you can average the 100 readings. Note that averaging 100 independent readings is supposed reduce random noise by a factor of sqrt(100) = 10. So plotting the average of the 100 readings should result in a smoother curve, with less noise. Any deviation/anomaly you observe in a single reading, which does not decrease by a factor of at least 5 in the average of the 100 readings, is with ~95% probability NOT caused by random noise, but likely a systematic component (i.e. either it is actually present in the measured spectrum, or it's a systematic measurement error). Best Regards Gerhard Am 03.10.2011 13:01, schrieb Gerhard Fürnkranz:
Well, the sensor's sensitivity is typically bell-shaped, so the S/N ratio is certainly better at green wavelengths, while it can be significantly worse at the blue and red end. "János, Tóth F."<janos666@xxxxxxxxxx> schrieb:I noticed it when I made my previous charts that the wavelength interval is extended in HiRes mode and the curves seem "noisy" in that extra range. [...]