The TSL2591 sensor has two channels, CH0 and CH1. The CH0 spectrum is vaguely similar to the "visual" density spectrum we are trying to measure, and the CH1 spectrum has lower sensitivity and leans towards the infrared.
The current approach subtracts the reading of CH1 from CH0, in basic counts, to get a combined reading that then has a slope correction applied and density calculations performed. So far this has worked out fine, despite some comments that combining sensor channels like this is not advisable.
However, recent analysis efforts have shown that CH1 has behavior that may not be desirable:
Its log error characteristic curve, the thing slope correction exists to deal with, is completely different from CH0.
Its thermal response curve (temperature-vs-reading) looks different from CH0, with a steeper slope
Above 30C it introduces substantial "dark" reading errors, which cause issues on high density targets
For the most part, its been easy to ignore these issues because:
The contribution of CH1 to the final reading is usually very small
Most testing has occurred under room temperature conditions
Measurement errors related to spectrum mismatch, as can happen with color targets, are more likely to be on the blue side of CH0. Meanwhile, CH1 only sees on the red side.
Making this change would only have a minor effect on the device's spectral sensitivity:
As such, making this change is likely a "safe" course of action for standardizing how the device processes sensor data. For the most part, it would involve removing one math operation from a few routines and recalibrating the device. It would also be possible to remove CH1 from the gain calibration data, however it does no harm to leave that in there.
The TSL2591 sensor has two channels, CH0 and CH1. The CH0 spectrum is vaguely similar to the "visual" density spectrum we are trying to measure, and the CH1 spectrum has lower sensitivity and leans towards the infrared.
The current approach subtracts the reading of CH1 from CH0, in basic counts, to get a combined reading that then has a slope correction applied and density calculations performed. So far this has worked out fine, despite some comments that combining sensor channels like this is not advisable.
However, recent analysis efforts have shown that CH1 has behavior that may not be desirable:
For the most part, its been easy to ignore these issues because:
Making this change would only have a minor effect on the device's spectral sensitivity:
As such, making this change is likely a "safe" course of action for standardizing how the device processes sensor data. For the most part, it would involve removing one math operation from a few routines and recalibrating the device. It would also be possible to remove CH1 from the gain calibration data, however it does no harm to leave that in there.