This is a story about making a highly stabile light source that was used, and occasionally still is, to test professional astronomical cameras. It was concieved out of a necessity to test one of those very expensive gadgets that seemed to have behaved erratically. The light source served its purpose and faulty camera was eventually sent back to the producer and replaced by a correctly working unit.
When we think about light sources, we don't often think about how stabile their output is over time. That is because people don't need hihgly stabile light sources at all. Not only is our vision rather poor in dynamic range i.e. it scores merely 40 dB (compare that to hearing dymnamic range of 120dB), but we also have several automatic mechanism that adjust the eye response to continually changinl lighting conditions: our retinas get senzitized and desenzitized, pupils expand and contract etc. Because of all those negative feedback mechanisms regulating our less than perfect receptors, we are simply totally unable to estimate the absolute brightness of what we see. But that should not make any reader sad - if there were such need in life, nature would have figured out how to do it long ago. There isn't so nobody does it. Except perhaps mantis shrimps
The situation is different in the world of artificial light sensors. They actually do need to report absolute brightness of objects and scenes. For some purposes the tollerances can be quite large e.g. a typical camera taking a snapshot would do equally good job if it captured brightness of a family gathering scene twice as dark or twice as bright as it actually was because everyone would be able to recognize their favourite uncle in it without a problem. Sounds too loose? Just compare photos of the same scene taken by several different cameras and see for your self.
But for cameras used in industrial and scientific setups requirements for consistent absolute brightness measurements come into play. Astronomy is probably the most demanding of all possible applications one can imagine for a camera for several rasons. Firstly, astronomical objects are universally very far away and consequently very faint and dark. If you think a little about it, very small number of photons eventually ends on Earth and further on some specific camera sensor who knows where after traveling thousands and millions of light years, no matter how bright the original object was. Secondly, astronomers need to rely heavily on mutually comparing pictures of various celestial objects to draw conclusions and figure out laws of nature. Thus if cameras weren't keeping absolute sensitivity constant over long periods of time, you would only get bad astronomy
So how accurate and precise astronomical cameras actually are? Modern camera sensors, being they CCD or CMOS or whatever, are quite efficient so that they typically capture 90% of infalling photons. Not only that, but they are so precise that they are able to discern the actual number of photons that each single pixel managed to gather! Dynamic range of high quality sensors is 16-bit i.e. each pixel typically reports between 0 and 66.000 captured photons. It stands to reason to assume that absolute stability of sensor sensitivity should be of the same order otherwise investing in 16-bit equipment would be a waste of time and money.
There we are. In order to test a camera with dynamic range of the order ~ 100.000, we need to have a light source of comparable absolute stability, and possibly a bit better. In other words, our light source should not vary its light output by more than 1 in 100 thousand during the complicated testing procedures if we wish to get meaningfull conclusions about the validity and quality of camera operation.
Let's see how that compares to the stability of everyday electronics. A typical volume control of an audio power amplifier has no more than 100 steps. Either it is digital or you simply cannot physically position the volume control potentiometer to more than ~ a hundred steps. The same goes with brightness and contrast adjustments of TV screens and PC monitors. What about controling speed of a vehicle? Are you able to discern more than a hundred steps in pressing a pedal by your foot? The conclusion is that common stuff that surorunds us is rarely required to be adjusted and monitored by precision higher than 1%. The world is quite rough if you think about it.
But remember, what we want is 10E-5. Now we have to ask ourselves, is it at all possible to build what we aim of building without some exotic components made of unobtainium and how much would the test equipment of our dreams cost?