The measurements were done on the night of Nov 9-10 MST. The log of the temperature differences between the heated plate and the surround is as follows:

Final Temperature offset between the two probes: 0.0C

The air temperature was about 4C throughout the test, and the humidity was quite high, about 80 percent.

Both the plate and the surround were isothermal during the test; no significant gradients were present in the IR signals from either.

The data from the images during this period is as follows:

09_1846.fit sur,sig= 2669.5 8.8 plat,sig= 2682.6 7.9 P-S= 13.1 DT= 0.8

09_1852.fit sur,sig= 2666.5 8.0 plat,sig= 2679.2 8.1 P-S= 12.7 DT= 0.6

09_1902.fit sur,sig= 2662.3 8.6 plat,sig= 2671.7 7.4 P-S= 9.4 DT= 0.1

09_1912.fit sur,sig= 2659.9 8.4 plat,sig= 2669.5 8.1 P-S= 9.6 DT= 0.0

09_1933.fit sur,sig= 2657.1 8.1 plat,sig= 2664.6 7.7 P-S= 7.6 DT= 0.2

09_1938.fit sur,sig= 2655.5 8.1 plat,sig= 2662.6 7.7 P-S= 7.1 DT= 0.3

09_1953.fit sur,sig= 2658.5 8.1 plat,sig= 2717.9 7.4 P-S= 59.4 DT= 1.7

09_2003.fit sur,sig= 2662.3 8.6 plat,sig= 2741.4 8.3 P-S= 79.1 DT= 2.9

09_2013.fit sur,sig= 2665.4 8.8 plat,sig= 2756.2 8.1 P-S= 90.7 DT= 3.5

09_2024.fit sur,sig= 2665.6 8.7 plat,sig= 2763.9 7.9 P-S= 98.3 DT= 3.8

09_2034.fit sur,sig= 2664.6 9.0 plat,sig= 2765.3 8.4 P-S= 100.8 DT= 4.1

09_2044.fit sur,sig= 2661.6 9.3 plat,sig= 2765.7 8.3 P-S= 104.1 DT= 4.2

09_2054.fit sur,sig= 2664.0 9.4 plat,sig= 2757.6 8.8 P-S= 93.6 DT= 4.2

09_2104.fit sur,sig= 2662.6 8.3 plat,sig= 2749.3 7.4 P-S= 86.7 DT= 3.9

09_2120.fit sur,sig= 2653.0 9.4 plat,sig= 2737.0 9.1 P-S= 83.9 DT= 3.3

09_2130.fit sur,sig= 2652.4 9.5 plat,sig= 2730.1 8.5 P-S= 77.7 DT= 3.5

09_2140.fit sur,sig= 2649.0 8.5 plat,sig= 2689.4 8.1 P-S= 40.4 DT= 1.9

09_2150.fit sur,sig= 2644.8 8.4 plat,sig= 2665.8 7.8 P-S= 21.1 DT= 0.9

09_2206.fit sur,sig= 2640.1 8.4 plat,sig= 2650.6 8.1 P-S= 10.5 DT= 0.0

09_2221.fit sur,sig= 2639.8 7.8 plat,sig= 2648.3 7.3 P-S= 8.6 DT= 0.1

09_2241.fit sur,sig= 2641.6 7.9 plat,sig= 2647.6 7.6 P-S= 6.0 DT= -0.3

The filenames are date_mst.fit; sur,sig give the mean level in the image of the surround and the noise. plat,sig give the mean level in the heated plate and its noise. The heater power supply was turned on about 19:40 and off about 21:30. A crude least-squares fit to these data yield the relation

DADU = 5 + 23*DT

The zero point doubless reflects some lack of perfect flatness in the camera system response, but amounts to an error of only about 0.2C. The data indicate that there is a significant lag between the IR level and the measured temperature, but this has negligible effect on the results.

Some herringbone-pattern coherent noise is evident in the frames with high sigma; the frames with sigma about 7.5 ADU appear clean. This sigma appears to be the noise floor, and represents the noise also on clean sky. It appears to be accurately gaussian.

The night was cloudy, and the sky level measured after the tester was removed was about 1700 ADU. The detector is a photoconductor and should have a signal proportional to the flux, but lacking details of the filter and the device response we will assume that the response is linear in the temperature; this is certainly satisfactory for the plate, but is approximately correct for the sky as well. Thus on this night, the sky was 2660 - 1700 DN -> 42C colder than the plate/air temperature, or about -38C. It is doubtless MUCH colder than this on a dry night. The noise represents about 7.5/23 = 0.33 C/pixel. The image of the heated plate is 70 pixels square, so the pixels are (0.5 deg)^2 as advertised. Thus the equivalent noise temperature per square degree is about 0.16 C-deg on 5 minute centers, or 0.35 C-deg-(min)^0.5

Doug Finkbeiner notes that the present system is much too slow to follow small cloud features, and suggests a cycle time at least a factor of 5 faster, yielding an image per minute. Given the extremely crude cold baffling of the current system (the effective stop and the filter are both at room temperature) this should be easy to achieve with sensitivity equal to the much slower current system. He also points out that 0.5 degree is much higher resolution than is required. I therefore suggest that we place as requirements on a new system something like the following:

FOV: 140 degrees, circular

cycle time: < 1 minute

dynamic range: 0-330K

flux resolution: >= 12 bits, so 1 bit < 0.1K near room temperature

noise: < 0.16 C-deg-(min)^0.5

histogram sampling: better than 1 ADU/sigma. Note that this may require better than 12 bit sampling. If the system has angular resolution of 1 deg, 1 minute cycle, and the minimum noise spec, the noise is 0.16C -> (0.16/330)*4096 = 2 ADU, but if the noise is more than a factor of two better, a 14-bit converter will be required.

angular resolution: < 1 deg

flat-field stability: < 0.2C p-p over one-hour timescales