The scheme is to measure radiation loss to the dark sky. If we take a thermocouple and paste one side to something at the local temperature, and past the other side to something looking just at outer space, then if heat is lost (or gained) to outer space a voltage will be generated by the thermocouple.
For this implementation, I use a Thermo Electric Cooler (TEC). These devices have a lot of PN junctions connected in series. The particular one used, the Melcor CP1.0-127-05L has 127 in series. They thus generate a big voltage for a small temperature difference. (In relative terms - a millivolt might be typical for this application.) The neat thing is that this voltage goes to very near zero if the temperature difference between the two sides is zero. I once put such a device in a hole bored in a big block of aluminum. This to insure that both sides were at the same temperature. The whole mess was wrapped in insulation. After a couple of days the signal went to within a microvolt of zero. This was just the best that I could measure.
The device is glued to something at local temperature and surrounded with insulation arranged so that a hole in the insulation points in the same direction as the telescope. At the suggestion of Paul Bartholdi, this version puts the detector behind a "storm window" to limit local convection loss, and servos the voltage across the device to zero.
Also built into this implementation is a light sensor to tell the difference between daylight and darkness. A nice feature is the ability to detect clear sky during the day. (This assumes you turn off the dark sky part of the circuit.) But expect a bit signal if the sun shines directly on the detector. The sun IS hot. So it drives the detector as implemented in the negative direction. If it does not see the sun, the day time signal for a clear sky is similar to the night time signal.
This creates a single output which is positive for clear sky and negative for daylight. It's output is somewhere in between when there are clouds.
It is an old trick to measure the radiation to the dark sky. I first heard about it in college physics in 1947. I heard about Peltier devices even earlier from a series of books put out for kids by Hugo Gernsback on learning about electricity. I apologize to anyone who thinks this idea is theirs for use with astronomy. I would reference you if I knew who to properly credit. I used TECs as differential temperature sensors in my "Cold Fusion" experiments starting in 1988 (Gosh! Is it that long ago?).
I used a 2" x 1/2' aluminum bar long enough to clamp one end to the frame of the telescope. Given a reasonable length of time, it will come to the same temperature as the telescope frame. A reasonable time can be an hour or so the first time the device is mounted. The TEC was glued to one end of the bar. A balsa wood (A good insulator) well was built around the TEC. In front of the TEC, three balsa framed saran wrap windows were constructed so that it was behind a "triple pane" storm window. I think there is nothing special about the material used for the window. Put a piece of the material you use between your face and a source of radiant heat such as a gas flame as I did. If it does not block the heat, the material is OK. Try the same thing with a piece of aluminum foil to calibrate the experiment. You will find the foil really blocks the heat.
The TEC was wrapped with some pipe insulation. This material (bought at the local hardware store) consists of double bubble wrap with a layer of aluminum foil on each side. It thus reflects heat, and is also a poor conductor of heat. If you can't find that material, use thin foam and glue foil on one side. Then make a tube out of several layers of the material so formed.
My design was 6" long with an inner diameter of about 1 1/2". Two layers of the bubble wrap were used. The construction is such that the TEC looks out through a tube to the sky. In every other direction, it sees either the back plate from the back side, or the reflective foil on the inside of the tube from the front side of the TEC. Take care that the aluminum foil/insulator tube completely covers the area where the TEC is glued to the plate.
Click on the diagram below to see a full-size version.
The electronics uses a low drift operational (OP07) amplifier to bring the signal level up to a reasonable level. Any low offset amplifier will do. Look for measures in microvolts on the data sheet. The OP07 is a common type and the most likely one to find. Gain is set at about 5000. I measured the impedance of the TEC at 2 ohms. Your device may vary and require selection of R5 for the best range. (A schematic will follow) Per the suggestion of Paul Bartholdi, the TEC is operated so that the voltage across it is held to zero by the Op Amp. Any difference from zero volts across the TEC causes an input to the Op Amp which then drives current back through the feedback resistor R5 to hold the TEC potential to zero. This is more a nicety to people that appreciate what Op Amps can do than a necessity. The Mark I version of this detector worked with the conventional input and feedback resistor.
To get the same polarity as my device, it is necessary to mount the TEC so that it gives the expected signal. For this, the TEC is glued with the side where the two leads are attached against the metal bar. This is no big deal, if you get the wrong polarity, just reverse the black and red leads.
For the light detector, I just went to the internet and searched. The PNA4603H device was found. I have no idea if this is the best possible device, but it does what I want and cost a couple of bucks from Digi-Key. One disadvantage of this device is that it puts out 0.3 volt or so in darkness. To match with the very low voltage generated by the TEC, it was necessary to offset this voltage. I just used 3 conventional silicon diodes in series. So it takes of order 2 volts out of the light sensor before any appreciable signal is sent to the op amp. Since it's full scale output voltage is of order 4 volts there is still 2 volts to do the work.
I found it convenient to put a couple of switches in the circuit to simplify set up. One does not get either clear sky or daylight on order for testing. You can solder in jumpers and save the cost of the switches once you have things set up.
First open both switches. Now R6 can be adjusted to bring the Op Amp output to zero. Use a good meter and get it to a millivolt or so. Well, if you don't have a good meter, do the best you can. It should not be critical.
Now close switch S2. This connects the TEC. Point the detector at the ceiling. The output should go to near zero. Let things settle until they quit changing. A drift of around of a couple of hundred millivolts is expected. This is a sensitive device with a lot of gain so don't expect the "Rock of Gibraltar". Now place your hand over the opening. (I hope you were not hovering over the device with your face and expecting it to settle.) You should see the signal go negative. Depending on how far away your hand is, there might be a negative change of a volt. Now place a piece of aluminum foil over the top of the insulating tube. I.e. just cover the opening with a cap. Now it should no longer see your hand. If it does not see your hand, then take a soldering iron and move it about the device. If you have built a good shield, then the device should not see the soldering iron. At the side, in front, etc.. But don't get the iron too close or you will melt the bubble wrap. Just for fun, remove the foil cap and hold the soldering iron in front of the device. It should peg at -12 volts or so.
I have run many hour long tests to see how much the device drifts. About +/-0.1 volt is typical. Mostly this is because the room temperature is not constant. But try it, running for a day with aluminum foil covering the tube end.
Now close switch 1. With lights in the room measure the voltage across R3. It should be a couple of volts with light shining on the daylight detector, and near zero in darkness. Unless you work at night and can turn out all the lights there may be a problem seeing "near zero". If this all works, then you are set up.
Figure 1, above, is a data run taken with this device on an evening that started with clear sky which was followed by intermittent clouds and then solid clouds. A sequence of images was taken through the night and continued until well after daylight. The mean sky background was measured for each image. The simultaneous "Clear and Dark Sky Detector" (CDSD) output was also recorded. The time scale is the fractional Julian day x 1000.
The I detector was used for the mean sky background. This is normally around -23000 or so for a very clear night. At around -18000 one can usually see gradients or other evidence of clouds in the image. At -10000 or so the images are clearly bad and fuzzy from the cloud cover. As I measured each image (Yes, I did this by hand) I made a judgement as to whether the CDSD matched what I was seeing.
This match is good, but not great. It is worst when the clouds are patchy. This is because the CDSD is looking at a much larger area than the telescope. If a cloud is sitting right in front of the telescope, the CDSD may be looking at clear sky around it. Thus it might not indicate as low a value as if it were looking at the same field as the telescope.
As you can see from Figure 1. the data starts out "pretty clear" but not great. The CSD has about 3 volts output. When patchy clouds come by at time 720 or so, the CDSD dips as the sky background increases. These are heavy clouds at first, then at about 800 they thin out. At 980 or so the sun rises and the dark sky circuit pushes the output to a big negative value.
I will probably set my program level at 2.5 volts. If the CDSD is below 2.5 volts, I will take darks.
One might mount the TEC at the focal point of a small mirror. One would not need a great mirror. An automobile headlight or some such might do just fine. It would then be important to figure some way to connect the back side to the local "ground" temperature. A big aluminum bar would probably be OK. Just bring it out through the side of the mirror and clamp it to some big thermal mass.