Clementine HIRES "Hysteresis" Measurements

1/25/95 - Travis White and Bruce Wilson, Larwence Livermore National Lab

Observations.

Sequences of flight images of Vega and of the Moon indicate that the response of the HiRes sensor tends to have a slight lag in responsivity after a change of image intensifier gain. The effect has been dubbed "gain hysteresis." General Atomics, manufacturer of the image intensifier tube used in the HiRes, had previously reported that the high voltage power supply on the multichannel plate gain stage required around 15 ms to charge or discharge (10-90%). Therefore, we at LLNL postulated that the in-flight delay between setting the gain and taking a picture may have been too short to allow stabilization of gain. There is general interest among the Clementine Science Team to use a spare HiRes camera to duplicate the effects observed in flight. Not having the exact in-flight timing of command sequences at our disposal, we requested help from Josh Resnick of RSI to determine these. At the same time, we tested Clementine HiRes sensor 312 for gain hysteresis and observed three important effects.

Figure 1. Mean counts per pixel following an increase of image intensifier gain from a setting of 160 to a setting of 170. (HiRes SN 312; Filter B, 560 nm integration time, 25 ms; offset, 0; video gain, "350 e- per count"; illumination, fluorescent white light scattered from off-white walls.) The dashed line is a simple exponential approximation to the measured values (squares).

First, the camera did exhibit lag after a change of the image intensifier's gain, with an exponential time constant on the order of 2–5 ms in our tests and a possible variation of 10–20% in net counts. (Figs. 1, 2)

Second, the mean count per pixel depended upon the time between successive images; the shorter the time, the greater the count — all else being constant. (Fig. 3) To a good approximation, the mean counts per pixel as a function of time between images decline exponentially towards a scene- and camera setting-dependent value with a characteristic time of several seconds.

Figure 2. Mean counts per pixel following a decrease of image intensifier gain from a setting of 170 to a setting of 160. (HiRes SN 312; Filter B, 560 nm; integration time, 25 ms; offset, 0; video gain, "350 e- per count"; illumination, fluorescent white light scattered from off-white walls.) The dashed line is a simple exponential approximation to the measured values (squares) .

Figure 3. Camera response as a function of interval between exposures. Mean dark count at this exposure is about 29 counts/pixel. Camera settings are the same as in Fig. 1, with intensifier gain setting = 170.

Third, we observed a scene-dependent lag in response, so that following sudden changes from dark to light scenes, the camera required several exposures (at 2 second intervals) to reach steady-state values, camera settings and illumination being held constant. (Table 1) The magnitude of the shortfall in net counts between the initial image and the steady-state value was about 3–5% in our tests. The reverse was not true; following sudden changes from bright to dark the camera seemed to exhibit no lag in response.

Sequence #                         1         2         3         4
Dark count prior to unblocking   27.9      27.9      28.0      28.0
1st Exposure after unblocking   128.8     132.2     134.4     134.3
2nd Exposure after unblocking   133.1     134.6     137.4     137.2
Steady-state after unblocking   135.3±0.5 135.6±0.5 138.6±0.5 137.9±0.3
1st dark image after re-blocking 27.9      27.9      28.0      28.0

Experimental description.

The readout time of the HiRes camera (27 ms) is greater than the delay that we are trying to characterize (1 – 20 ms). Therefore, Figs. 1 and 2 represent superimposed results obtained over many seconds rather than in a single sweep. In our experiment, we took sequences of images, changing intensifier gain every eighth image. We used only the first image of each sequence of 8 images to characterize the camera response as a function of gain- delay. Starting with the shortest delay possible between setting the gain and taking the next picture (150µs on our system), we performed a cycle of 48 images: 8 at a gain setting of 170, 8 more at a setting of 160, then back to a setting of 170 for 8 more images, and so forth. On each subsequent set of 48 images, we increased the delay by 1 ms up to a maximum of 20 ms, which made a total of 21 sets of 48 images. System jitter allowed the delay to reach almost 30 ms in one case (Fig. 1). Through-out this procedure, the interval between successive images was about 0.23 s ±0.07 s (peak-peak).

Our experimental setup (Fig. 4) consisted of a computer-controlled HiRes sensor (SN312), a special "breakout" cable connected to a custom-made SASI-to-RS232 converter box, and a Phillips 3585 Logic Analyzer (DOE property id #5263581, calibrated 5/93). The computerized controller was a stand-alone Sun workstation ("vagabond") operating under unix and equipped with a SPARC 2CE/16 cpu, a VME interface bus and an LLNL-made frame grabber interface board with SASI bus output. A standard 24V Clementine power supply activated the camera. Immediately behind its 51-pin input jack, the SASI-to-RS232 converter box provided a set of convenient probe points to which we connected the Logic Analyzer without electrically loading the SASI bus. We ran our tests in an ordinary office at room temperature. The camera pointed at a nearby (30–45 cm distant) off-white wall illuminated by conventional fluorescent ceiling lamps. Although we expected that 60Hz flicker might cause major variations in the output of the sensor, the output was quite stable. Our integration time was 25ms; offset, 0; video gain, "350 electrons/count" (gain setting 1 from {0, 1, 2}), filter position, 5 (Filter B, 560 nm); and intensifier gain, variable (typically, 160 – 170). As a measure of camera performance, we used the mean count per pixel in a box whose corners lay at (127, 95) and (207, 175). In this notation, the first value of each ordered pair represents a column number between 0 and 383, and the second represents a row number between 0 and 287. The image exhibited noticeable spatial roll-off, consistent with the way intensifier sensitivity of Clementine HiRes cameras decreases towards the edge of the FOV. The domain of sampled pixels was roughly centered on the area of greatest response. We relied on computer-generated time hacks to get all of our timing data.

Figure 4. Experimental layout.

Discussion of results.

Figure 5 on the next page shows what appear to be 3 cycles of exposures. In reality, there are 21 different sets overlaid on top of each other, 48 images per set as described previously. If there were no lag in the response of the camera, then the data would have formed 6 straight line segments in a perfect square-wave pattern with 3 line segments corresponding to gain 170 and 3 to gain 160. The fact that the 6 line segments are hooked as they are shows clearly that there is a lag in camera response. To be sure that the results shown in Fig. 5 are not misinterpreted due to their dependence on the interval between exposures, we have replotted them in Fig. 6 (next page) as a function of the interval between successive images. What we see are two nearly horizontal rows of points at about 165 and 95 counts, respectively. (Actually, the rows have a slightly negative slope to them, as we now expect with increasing intervals between images.) In addition there are a few stray points and, most importantly, there are two series of intermediate values which asymptotically approach the bounding values near 165 and 95 counts. All points that do not lie within one count of the top or bottom bounding lines in Fig. 6 (squares) correspond to the first images taken after changing the intensifier gain. The rest of the points (diamonds) correspond to images taken about 200ms or longer after the gain change. The stray points in the figure result from the fact that the strongest variations in the output of the camera depend more upon time since gain change than upon time since previous image. As shown in Figs. 1 and 2, when the same data (squares) are plotted against time-since-gain-change, there are no stray points of which to speak.

Figure 5. Lag of camera response when intensifier gain is changed.

Figure 6. Effects of gain change separated from effects of changing rate of picture taking. Fig. 2 replotted.

Careful scrutiny of the curved sequences in Fig. 6 shows that they do not quite reach their neighboring limiting values of 165 or 95. This discrepancy suggests that our two main observations (dependence of mean signal upon time elapsed between pictures and lag in camera response following a gain change) may have different origins. Since the bounding rows of points come from randomly spaced instants through-out the experiment, the short- fall is probably a characteristic of the camera rather than an artifact of external influences such as changing light conditions.

Our experiments do not identify the source of the gain hysteresis. Possible explanations include either persistent phosphorescence or electronic lag — due, for example, to recharging the high-voltage stage of the intensifier tube after an exposure. The 10–90% time constants associated with the power supply on the intensifier tube are 4 and 19 ms for charging and discharging the gain of the prototype unit. ("Clementine Receiver Project Critical Design Review," General Atomics, 23 October 1992. Courtesy, Joe Kordas.) We have no measured values on other intensifier stages. Note that the design of the power supply causes an asymmetry in the time constants, which we do not observe in our experiments. Within experimental uncertainty, the characteristic decay times shown in Figs. 1 and 2 are identical. Having equal decay and rise times is inconsistent with the electronic behavior we expect. As for the hypothesis that phosphorescence could account for the effect, the decay time of the P20 phosphor depends upon current density. For P20, decay to 10% occurs in 1.8 ms at 2 µA/cm2 and in 180 ms at 0.5µA/cm2. ("Optical characteristics of cathode ray tube screens," EIA Tube Engineering Advisory Council (TEPAC), TEPAC Publication No. 116, Dec. 1980, p. 56.) This is a broad range of possibilities, spanning the range of current densities in our experiment. Note, also, multiple decay processes are evident in P20 phosphor decays.

What about the origin of the effect shown in Fig. 3? Why does the output from the camera depend upon the amount of time between consecutive exposures? (Note that the effect shown in Fig. 3 can be large, as much as 25 counts out of a net of 110 at the settings we used.) One possible explanation might be increased dark current as a function of camera repetition rate. We eliminated that possibility experimentally by placing the lens cover over the camera and showing that dark counts are unaffected by the camera rep rate in our experiments. Dark counts do not depend on the intensifier gain, either. In addition, not even the first image after installing the lens cover has a residual output. That is, within shot-to-shot variation, as soon as the camera is blocked the output is the same as the steady-state value. Therefore, whatever the cause of the dependence of output upon the interval between consecutive pictures, the effect (a) depends upon light, (b) does not depend upon the FPA dark current, and (c) does not depend upon the shot-to-shot persistence of the phosphor. These observations suggest that the origin of the dependence on time between images lies in the front end of the image intensifier tube.

Scene-dependent lag.

If, in our experiment, there are lags in response that are associated with the front end of the intensifier tube, the possibility arises that an unrecognized circuit — specifically, the bright-source protection circuit — may be contributing to the bias on the front end, altering the gain of the camera. There is also the possibility that the output depends upon the length of time that the camera has had photons coming into it.

We tested this latter idea by placing the lens cover over the camera and then suddenly removing it just before an exposure. Although the mean count per pixel in the first image after unblocking the camera was lower than subsequent images, the timing of the test was too uncertain when the interval between images was less than 2 seconds. It is possible we got lower counts at fast rep rates simply because we photographed ourselves while removing the lens cap. When the interval was 2 seconds, though, we had plenty of time to install or remove the lens cover. In this case, our results definitely showed that the output of the first image or two after removing the lens cap was lower than the steady-state value, which was reached within about 6 seconds (3 exposures). Table 1 above shows typical results obtained by blocking and unblocking the camera. A more accurate retest of this experiment would be useful, using electronically timed shutters and varying the delay between the instant the shutter begins and the instant the exposure starts.

Note that the dark count reported in Table 1 was about 28 counts, which is 1 or 2 counts less than the dark count obtained previously at the same integration time. Since the data of Figs. 1–6 were obtained on a separate day from the data of Table 1, this change of dark current shows that there is a daily drift in dark count on the order of a count or more.

Conclusions.

Our results show that there is indeed a few-ms lag in the camera response that is associated with changes of gain, that the output of the camera depends upon the interval between exposures, that the dark count varies from day to day, and that there is a lag in the response of the camera following a change in scene brightness. All of these effects could affect the radiometric interpretation of Clementine flight data.

Our results so far imply that the two effects we observed (lag associated with gain change and rep-rate- dependent output) probably have different causes, but we do not know those causes. From our data, we cannot tell to what extent the effects are empirically predictable — which is necessary if one is to back out extremely precise (±2%) relative radiometric values from the Clementine data. Results may depend upon the particular sequence of exposures taken.

If it can next be established that the in-flight Vega results are duplicable on the ground, then one should try to determine empirical relations for the lag in camera response. The goal of additional tests is to improve the accuracy (or, alternatively, to establish the limits with which we know the accuracy) of Clementine's lunar spectral radiometric imagery. For example, answers to the following questions would be useful: does the lag that accompanies changes of gain always have a quasi-exponential fall-off with a fixed time constant, independent of filter, integration time and intensifier gain? Is the lag spatially homogeneous? Is there a well defined relationship between changes in gain (or changes in mean output) and the amplitude of the exponential curves like those of Figs. 1 and 2? Designed and built by Darron Nielson, LLNL.