Method: I made a light box with internal diffusing material and a translucent plastic front, and applied layers of tinted window film to form several areas of fixed, constant brightness. I was using the film as a stepped-zone ND filter. The tinted film is actually quite transparent in the near-IR, but I sidestepped this problem by using a compact fluorescent lightsource which has almost no IR output.
I calibrated the lightbox using my Sekonic L-508 lightmeter as a 1 degree spot meter. It reads out in EV (exposure value) which can be converted to cd/m^2 (candelas per square meter) based on a conversion formula I (cd/m2) = 2 ^ (EV-3). The brightest area on the target is EV 13.4 = 1351 cd/m^2. This EV conversion formula is specific to this meter; apparently the response of light meters in EV units is not very accurately standardized. The meter resolution is 0.1 EV. These meter readings of luminance are shown on the horizontal axis on the chart.
I took video (normal interlaced scan, 1/60 sec) of the target, using several different f/stops, and imported via firewire to my PC. I took a still frame from each f/stop setting, imported into Photoshop, converted to greyscale, and used the eyedropper (5x5 average mode) to read out the pixel value from 0-100% of the selected areas of my luminance target. I used the digital value to IRE level transfer function I measured to convert these to IRE levels, and the vertical axis on the plots below shows these readings.
Results: I have two plots for the TRV900 from two sets of measurements, one with the internal ND filter off, and one with it on. You can see that the internal ND filter makes the image very nearly 3 f/stops darker. I also have a plot from the one-chip TR7000 (steadyshot off, 1/60 sec exposure). Comparing the f/stops needed to achieve the same IRE video output for a given EV input, the TR7000 is between one and two stops less sensitive than the TRV900. Here is the raw data used to make these plots.
The minimum value I ever recorded, for the darkest area of the target (a patch of opaque black felt) is 10% white pixel output = 3 IRE units, this is the "dark level" output apparently (on both TRV900 and TR7000, 0% White (RGB 0,0,0) is -7.5 IRE units). Ideally the curves should all go down to this point, but it seems stray light/scattering (lens flare) causes the dark level to rise as the overall image becomes brighter.
Ideally the different f/stop curves should all look the same, just translated horizontally on the chart. This is not quite true here, but recall that there are roughly equal areas of each intensity value up to EV13.5 in the frame during every measurement. Taking video of a scene with significant image area at a luminance corresponding to 4 or 5 stops above full saturation is asking a lot of your camera. For any given iris setting, I'd say the useful dynamic range of both cameras is around 6 stops, the TRV900 perhaps slightly more.
The maximum white level from both cameras is about 109 IRE units, which exceeds the NTSC standard of 100 IRE. This normally doesn't cause a problem with standard VCRs, but it might be a problem for some analog video capture cards. It is not, of course, an issue with firewire transfers. I note the upper end of the TRV900 response rolls off slightly as it approaches the maximum, perhaps an anti-saturation analog circuit, in effect extending dynamic range. The TR7000 shows very little such rolloff.
Here is a chart comparing the light response (transfer function) of the TRV900 with 35mm film. I used Kodak Royal Gold 100 in my Canon Elan at f/5.6 and 1 sec. exposure. In this case, the brightest part of the exposure chart at EV 13.5 is 8.5 stops above the 18% grey point (EV 5). The negatives were developed (Walgreens 1-hr service) and scanned (Photosmart negative scanner, Vusmart software at gamma=2.0). The EV point "0" is an approximation; my L-508 spotmeter could not read that level which should just be considered "very dark". The TRV900 was set at f/5.6 and 1/60 sec. The TRV900 image has a few units of noise so there is no real difference below EV 6, for at most 7.5 stops of range, whereas the film really does have a dynamic range exceeding 12 stops. Note the scanned pixel value for film at the 18% grey point for this exposure (EV 5) is actually quite close to 18%, but how this appears on the display or hard copy depends on the gamma of your output device. I think the difference between the two curves below is a big part of the "film" look as opposed to video.
