Clementine NIR and LWIR Camera - Comments on Raw Data Sets

January 25, 1995 - I. Lewis (camera system engineer; camera operator during Clementine operations)

The LWIR electronics are functionally equivalent to the NIR, and the comments on the NIR performance apply generally (except as noted) to the LWIR.

Before attempting "science" with the Clementine raw data sets, the user should be aware of the basic features seen in the images, and assess how these types of features affect the conclusions that are drawn. As a casual observer scans the data sets from the NIR camera (and to some degree the LWIR camera), he would notice that the camera states (integration time, gain ID, and offset ID) vary widely throughout an orbit. Without a final in-flight calibration of the meaning of signal digital counts (DNs), one cannot even compare global intensity data. The following (brief) discussion describes the features in the data that must be accounted for in the interpretation of signal received at the camera as a function of DN. It is recommended that the user delay scientific use of the NIR and LWIR data until after the calibration effort is released (estimated to be end of calendar year 1995), if possible, if relative accuracy of 5% or better is needed between pixels and frames.

Control Settings

The camera state variables were set throughout the mission in an attempt to maximize the dynamic range of the data within the 8-bit A/D converter. Since the scene was much brighter at the equator than near the poles, the integration times and gain were longer for high latitude images.

Basic Camera Functioning

As with all infrared (IR) cameras, the signal received from the focal plane comes not only from signal generated from the scene, but also from dark current generated internally within each pixel and from thermal emission generated from the camera optical path itself. The magnitudes of these contributions was very large, varying both through the orbit and through the mission.

The signal from the focal plane is clocked out, and every pixel signal has a global offset (voltage subtraction) applied, followed by a global gain (voltage multiplier), followed by an A/D digitization. Global gain and offset should be the same for all pixels in the stream. Functionality of these global parameters with temperature will be checked in calibration. The pixel responsivity to external scene, the pixel responsivity to internal thermal signal, the pixel dark current, and the individual pixel offset must be considered separate variables for each row/column.

Camera integration time is selected from 4 possible settings. Global gain is selected from a 5-bit table that results from a resistor switching network. Global offset is selected from 255 offset states, with a full offset subtracting (roughly) 1 detector full well.

Camera Performance

The first aspect is easier to quantify: the standard deviation of adjacent dark frames at equivalent camera states yields the pixel readout noise. Half way through the mission, it was discovered that the temporal noise a) was much larger than digitization and shot noise could justify, and 2) varied as a function of gain ID setting, disproportional to the gain value. Preferred gain settings were established after that point. The explanation of these differences is the open resistors in the switching selection act as noise pickups. Temporal noise will depend on which, how many, and where the location of the open resistors are on the board.

Dark frames at the end of the various orbits can easily be used to determine the temporal noise of each gain setting. Note that dark frames were not compressed. Upon performing this subtraction, the user may notice intermittent repetitive tracks through the difference frame. These have been tentatively tracked to the FPA readout clock being tied to the cryocooler clock, which cycles intermittently as the FPA temperature controller commands. While unsightly, this has insignificant impact on the temporal noise of the data. Typical temporal noises are 1 counts rms to 5 counts rms, pending gain setting.

Note that both the NIR and LWIR had large variance of dark current generated per pixel and of responsivity to external signal. This gives a large “salt and pepper” contribution to each scene which will be correctable and should not be considered “noise”.

The second camera performance issue is calibration uncertainty. The dark signal offset level in operation is at present the largest contributor to data misinterpretation. In the NIR camera, the focal plane cool down time to a stable level at high gain/offset settings is more than 45 minutes. The FPA cools down in less than 10 minutes. The rest of the time is attributed to the residual heat capacity in the cold filter, and the fact that the cryocooler starts to cycle off based only of the FPA reading. Throughout every orbit, the continuous heat load of operating all sensors increased the cryocooler rejection temperature, making cooling at the nominal FPA temperature less efficient (increasing cryocooler current draw), and heating the local optical (including dewar window) temperature. It is thought that this increase in the radiative heat load from the window during the orbit results in increased cold filter temperature and increase of dark signal in the focal plane.

The LWIR cryocooler, on the other hand, never reached set point, so the cryocooler ran full on at all times. The FPA temperature rises throughout the orbit as the cryocooler rejection temperature increases. It should be noted that the value read out from the FPA temperature is not the actual temperature of the unit, but it is uniquely related to FPA temperature. (I.e. “95” from the LWIR temperature signal means 95 units of temperature, which may correspond to 75 kelvins; 95 will always mean 75.)

There is a marked change in camera performance near orbit 155 (the mid-lunar mapping mission orbital correction burn). At this time, a large spacecraft burn occurred, which had the net effect of reducing overall thermal mass on the spacecraft. The beryllium block heat sink started increasing faster and more during a mapping orbit than prior to that point. This increase in operational temperature had a marked effect on the IR cameras, which (functionally) increased the dark counts (i.e. the background counts generated from pixel dark current and the thermal emission from within the optical path.) One should note that the 10W generated by each of the IR cameras is a very major source of the heat load going into the beryllium block, which was originally sized to accommodate only partial use of the LWIR and less total operating time for the NIR camera. After launch, the operational plan changed to accommodate as much LWIR data as possible. After orbit 155, the amount of LWIR operation during an orbit pass had to be reduced to avoid complete saturation of the LWIR signal with background (FPA temperature rise created excessive dark current)

At the high sensitivity settings (used near the poles), the background counts in the NIR camera increased roughly 20 counts beginning to end of orbit before orbit 155, and increased roughly 120 counts beginning to end of orbit after orbit 155. A quick examination of two cool down sequences found that the background contribution of (presumably) the cold filter might differ by 20 counts if tracked only as a function of cryocooler case temperature. I expect that the eventual calibration effort will be able to predict dark level at the highest polar integration time/gain settings to about 5 counts. Equatorial offset should be determined to better than 1/2 count. Each of these values will mean that the temporal noise (as a function of gain setting) is expected to be the ultimate limitation in data interpretation.

One last word to the brave who venture forward before calibration rules are in effect: the max/min of the pixel is not necessarily 255/0. In some (not too many), the settings selected use an offset high enough that the following gain circuitry cannot achieve 255. For examples (the numbers here are changed to protect the innocent): the maximum output of a pixel is 4.2 V. The offset setting of “66” subtracts 1.3 V, leaving a saturated pixel at 2.9 V. The gain state of 61 (?) is a 1.5X gain, leaving 4.35 V. The A/D converter reads 255 at 4.5 V. Thus, no 255’s will result. In the NIR and LWIR images, there are hot pixels, and should always be a 255. If not, the camera state artificially prevents this. Zero is a different matter: since the background from dark current and thermal emission from the camera is always significant, the user may never see the residual circuitry voltage that comes from a “0” output form a pixel. This will follow in the calibration rules.

Miscellaneous observations