Thermal camera noise and Noise Equivalent Temperature Difference (NETD)

[Fraser]

I am often asked to provide a recommendation or comment on a thermal camera based upon its published specification sheet and NETD. I cannot provide such recommendations based on these data sheets.

I have been asked to explain why a very expensive Prosumer or Industrial thermal camera costs so much more than a budget offering like the FLIR One G2/G3, Seek Thermal Pro or some other Budget camera when the paper specifications appear so similar. There will, of course, be the build quality differences but some potential purchasers of budget cameras focus on resolution and NETD. This can be a mistake.

NETD and pixel count are prime marketing tools for a manufacturer but those specifications in isolation can be very misleading. How ? Well look at this video by Santa Barbara Infrared and learn about noise in the thermal camera system. Note the presenters comments at time index 11:18 regarding TVH  Manufacturers of products want to highlight the good specifications and down play those that are less impressive. The buyer rarely gets to see the whole picture and that may be intentional.

https://youtu.be/9LGwjsulf2M

So when people ask me why the image quality of a budget camera is not as good as that of a prosumer or Professional camera with similar specifications ..... it could be for a number of reasons..... mostly the cost reduction required to build a budget camera ! There are many types of noise that effect the imaging performance of a camera and then there is also the Radiometric performance to consider but that is a topic for another day.

I have looked at the power supply design of the Seek Thermal imaging cores and it is most definitely built down to a price and, if my tests were accurate, there appears to be an awful lot of power rail noise on the microbolometers power input to the ROIC. If such noise reaches the bias and ADC  power supplies inside the ROIC you have injected power rail noise issues to deal with. As a best case scenario, the ROIC is creating the noise on its own power rail and protects the biases and and ADC from such internally. It is still poor noise management on the power rail though. The NETD of the raw Microbolomter and ROIC die may not have suffered from this noise in a well designed NETD test jig  All my personal opinion and not fully proven yet though so YMMV.

Fraser

[Fraser]

For anyone interested in the testing of thermal Imaging system performance, you could do worse than watch the various videos from Santa Barbara Infrared  To replicate the tests you will need expensive test equipment though !

https://youtube.com/c/SantaBarbaraInfrared

I have also purchased text books on the performance testing of both the sensor array and complete systems. They are expensive books and the testing is very involved !

Fraser

[AkiTaiyo]

Some of the ULIS Lynred bolometers have a specification that the bias voltages must be under 20nV/√Hz from 1Khz to 10Mhz..
Even a 'low noise' LDO such as a LT1763 can't meet this requirement (its about 500nV√Hz at 1KHz) so some good circuit design is needed to be able to get the best low-noise performance from a detector.  Typically buffering and filtering power supplies with low noise op-amps.  Need to make sure all your voltage references are nice and clean too to get them to meet the specifications if your using a DAC to supply the bias voltages.  There’s a limit to what you can do when you’re trying to build to a budget, but low quantity industrial / pro gear can afford the extra expense of getting the noise as low as possible.
I also agree that noise specs on thermal cameras can be very misleading and often don’t specify the circumstances and/or settings under which the headline spec was achieved!

[Max Planck]

For anyone interested in the testing of thermal Imaging system performance, you could do worse than watch the various videos from Santa Barbara Infrared  To replicate the tests you will need expensive test equipment though !

https://youtube.com/c/SantaBarbaraInfrared

I have also purchased text books on the performance testing of both the sensor array and complete systems. They are expensive books and the testing is very involved !

Fraser

I will try to add my 0.02$.

1. The standards for NETD measurement (not only them) are still dedicated for thermal scanners, where measurement is made at the voltage output of a single detector preamplifier. There is a huge terra incognita between those standards and cameras using FPAs. Basically, NETD is a measure of temporal noise at the preamplifier output. The procedure should be done for a blackbody temperature 7-8 degC above ambient.

2. Given above, there is quite some room for manfacturers "resourcefulness". For FPAs, NETD describes a spatio-temporal noise: temporal and X-Y. No one said, that NETD measurement has to be done for the entire matrix. Why not some windowing and measurement in the central area of a detector matrix.

3. For  microbolometer cameras, manufacturers are giving the NETD value for a fast reference lens - f/1. The slower the lens, the higher NETD. For a f/1.6 lens it will almost triple.

Those are the most important elements, I think.

Max 

[IR_Geek]

Excellent points!   

NETD is almost useless and unfortunately is a leftover from much older analog systems.   3D noise is a much better indicator of spatial uniformity and ultimately the 'sensitivity' of the system for modern 2D focal plane arrays.    As mentioned above, proper test equipment is crucial and really expensive.    A true understanding of the sensors waveband, integration settings, NUC, optical system, etc... are key to making a true apples to apples comparison.