Recreating Spectral Response from 6-Channel Color Sensor

[AxisCat]

Hi All,

This is a project I started a couple of years ago then got stuck or bored and stopped working on it. I am thinking about picking back up where I left off. I am using a AMS AS7262 color sensor with the following overlapping channels: 450, 500, 550, 570, 600 and 650 nm, each with 40nm FWHM.

I used a Digikröm CM110 Monochromator and stepped through the spectrum in 1nm increments saving the output from the sensor to a file. My light source was a standard 100 watt halogen household bulb. This is what I measured:

The continuous dark blue curve is from a photodiode I am trying to use as a reference. I understand this is really hard stuff to do with the limited equipment I have available. Is it even possible to recreate a fairly accurate spectral response using just these 6 discrete channels? The manufacturer provides some information that eludes to it being possible.

Anyone have any thoughts?

Thanks,
Axis

 

[Andy Resnick]

Hi All,

This is a project I started a couple of years ago then got stuck or bored and stopped working on it. I am thinking about picking back up where I left off. I am using a AMS AS7262 color sensor with the following overlapping channels: 450, 500, 550, 570, 600 and 650 nm, each with 40nm FWHM.

I used a Digikröm CM110 Monochromator and stepped through the spectrum in 1nm increments saving the output from the sensor to a file. My light source was a standard 100 watt halogen household bulb. This is what I measured:
View attachment 336147
The continuous dark blue curve is from a photodiode I am trying to use as a reference. I understand this is really hard stuff to do with the limited equipment I have available. Is it even possible to recreate a fairly accurate spectral response using just these 6 discrete channels? The manufacturer provides some information that eludes to it being possible.

Anyone have any thoughts?

Thanks,
Axis

Off the top of my head, I suggest that you first normalize each channel and then assign/compute wavelengths according to ratios of the different channels- that's generally how color vision works.

 

[AxisCat]

I appreciate the reply. The "mapping function" is the part I am stuck on. Someone suggested Fourier transforms because the curves looked like filters and the math can be reversed. That looked really complicated so I haven't researched it any further. Another person's thought was using a matrix to calibrate. The input matrix would be 6x1 (sensor output x 6-channels), the correction matrix would be 6x300 (700nm - 400nm), and the output matrix would be a 1x300 representing the spectral response from 400-700nm. I really don't know which direction to take with this.