search

melexis / mlx90640-library

MLX90640 library functions
Apache License 2.0
241 stars 192 forks source link

Strange temperature distribution when creating a histogram #81

Open spamekkel opened 3 years ago

spamekkel commented 3 years ago

Hi,

I opened this issue a week ago, then closed it as I wanted to be sure.

While examining the pixel values I noticed there an unexpected distribution of the temperaures. With the latest mlx90640 driver I created a histogram of pixel #400. Below is the outcome after ~40.000 frames @4Hz. I'd expect a sort of normal distribution of the pixel values.

mlx90640_histogram_px400

Did anyone investigate this?

slavysis commented 3 years ago

Did you check if your object temperature changed during that long measurement? Can you share the linear plot of the data?

Best regards

spamekkel commented 3 years ago

Hi,

Thanks for your reply, your help is much appreciated.

There probably was a temperature change during that measurement of 3 hours. The histogram was recorded on the embedded system without storing the while data set. I will create another histogram of 30 minutes and record the data from the whole image. I'm currently too busy to pick it up now, but will do this next week.

Kind regards, Spam.

spamekkel commented 3 years ago

I should have said why the distribution seems strange. It is not the second bump to the right.

Looking at the device's targeted temperature range (-40 to +300'C) and a resolution of 18-bits I expected much smaller discrete temperature steps. E.g. for this range 340'C / 2^18 = 0.0013'C. But the frame data is only 16-bits, so perhaps steps of 340'C / 2^15 = 0.0052'C are more reasonable. But the temperature steps in the recorded histogram (2650 equals 26.50'C) are approx. 0.08'C, much higher then I'd expected. I admit I haven't gone through the device manual's complex explanation of the temperature calculation. But I'm aware of the noise figures, I'm not expecting a 16-bit accuracy of the final To. The datasheet does not explain the ADC resolution vs the end result. Perhaps this and all the compensations for Vdd, Ta etc. lead to the resultion of 0.08'C. ?

slavysis commented 3 years ago

Well, the thing is that the step is not solely defined by the resolution. The sensitivity of the sensor also plays major role as this parameter would define more or less what change of the calculated temperature would be induced by changing the raw data. So for example, if your sensor has such a sensitivity that when the raw data changes by 1LSB, the calculated temperature would change by 0.08C then whatever the resolution is, you just would not be able to see smaller step.

Best regards