- 14-bit per Column Distributed A/D- D/AD
- Correlated Multi-Sampling- CMS
- 4812 x 3624 Pixel Array
- 5.0um pixel - Global/Rolling Shutter
- >90FPS @ 4800 x 3600 (10-bit), >120FPS w/ROI
- 2 Gigapixels/sec throughput
- On-chip 125dB HDR in rolling shutter mode
- As low as 5e- noise
- 3 or 6 LVDS lanes
- 14 bit Single Data Rate (SDR) per port can be folded to 7 bit Dual Data Date (DDR) to save pins
"Panavision Imaging is pleased to offer another high performance CMOS image sensor for industrial imaging applications", said Derrick Boston, President and CEO of Panavision Imaging.
This is the third area sensor to be launched in PVI’s family of Dynamax sensors, joining the 3.2MP sensor (DYN0032) and HDTV 2.1MP (DYN0021) which are already in the market.
Dynamax 17.4MP test samples have been released to various customers for their different market applications. The engineering grade devices will be available in a CPGA package by the end of 2012, in either color or monochrome versions. High volume production is planned in the middle of 2013.
are there any camera's using the dyn0032 and dyn0021. Is it available for others or just for panavision?
ReplyDeleteOn the datasheet I got on the VISION show in Stuttgart yesterday,dynamic range is specified as 62dB and read noise as 8e. So I guess the 125dB can be achieved with multiple exposures.
ReplyDeleteDual exposure on alternate rows, actually. Might lead to some movement artifacts after interpolation, but still better than multiple frames.
DeleteWell, from 62dB to 125dB with a dual exposure... that's one hell of an exposure time differential! Ten stops of light, no less. If normal rows are at 1/48s (as they usually are on motion picture cameras), the "highlight recovery rows" must be at 1/50000s.
DeleteAlso, of the 21 stops of DR that you get, the first one (the darkest one) and the 12th one (where your skin tones may end up) have the same amount of noise.
Oh, and due to that fast exposure time, anything above that 12th stop looks terribly strobey and generates anxiety in the viewer.
Am I doing something wrong here, or does this make no sense at all?
If I understand this mode of operation correctly, one has to trade-off vertical resolution for high dynamic range capturing. Am I correct?
DeleteIt may be vertical or horizontal resolution, but yes
DeleteWait: 125 dB is only in HDR mode AND with rolling shutter. Was that 62 dB spec in global shutter mode? Then all this could make some sense...
DeleteWhat dynamic range can a camera lens put on the focal plane? Can a lens deliver an image where one region is 1.7 million times (125db) brighter than another region? I have no idea what this spec is for lenses.
ReplyDeleteGreat question. I think high end lenses (e.g., reflective telescopes) can achieve this. But, I have no idea what the limit is in a refractive lens barrel with full baffling etc. Probably 125 dB is doable but may be near the commercial limit. Probably also depends where in the FOV the bright (or dark) region is, and probably also depends on the average scene lighting as well.
DeleteYes, 125dB is definitely pushing the optics. More elements means more reflections and scattering. So fewer elements is better, and as Eric says, one or two mirrors is about the best you can do.
DeleteMost practical applications will have a hard time keeping the optics clean enough to maintain 125dB. Then again, this isn't your typical camera.
The definition of DR is sometimes misleading. We need a minimum number of nuances to appreciate une scene, for example 100. Suppose that the nuance difference is defined as the noise floor in dark. In this case, 100 nuances will consume 40dB from the dynamic range. This is to say that a 60dB DR sensor will give only 20dB exposure tolerance. This value matches well the limitation observed on CCD or CMOS sensors.
DeleteA lot of studies have been conducted by Kodak people in order to feature out the maximum dark to white range inside camera optical chamber. Their conclusion was that the range is limited to 80-90dB by taking into account only the parasite reflexion in the optical chamber.
So 125dB scene is almost inexisting. But if you want to be able to distinguish 100 nuances in the darkest zone and alos 100 nuances in the brighest zone, we need more than 80-90dB DR.
Annoncing 120dB DR is very often misleading on a dual-exposure system because you can always cover the dark and the bright extremes by using very different exposure. But the key issue is that you should be able to see all the nuances in the 120dB DR claimed. This is not always possible.
If one claims 120dB, the simplest test to check this value is to image a scene from 0.1lux to 100klux by fixing the sensor's setting. If it cannot do this, then it doesn't have 120dB DR.
-yang ni
I still like the diffraction ring technique. Just one image and count the number rings you can see. There is a straight forward intensity ratio between the rings so you can just look up the intrascene DR.
DeleteThe maximum intensity of the diffraction ring is not intense enough. The DR measurement can be limited by the noise floor. If the peak intensity on the diffraction pattern is 1000lux, then the sensor should be able to pop out 1mLux which is not that easy. Just fix your sensor setting, then from moon light to direct Sun shine. If you can image the scene without change the imaging parameters, then you can really 120dB, otherwise it's not true.
Delete-yang ni
Sorry Yang Ni, but I believe you are incorrect here. What counts is lux-sec (photoelectrons, actually) so one can just increase the integration time if the diffraction pattern is not intense enough (although creating the pattern directly on the sensor should result in plenty of signal). I stand by this technique. And while failing your proposed test means, as you say, the sensor does not have 120 dB DR, passing your proposed test does not prove the sensor has 120dB intrascene dynamic range.
DeleteEric, I would like to say that your diffraction ring is a good way to measure the useful overall DR of a sensor including the parasite light influence. My method can reveal the real DR tolerated by a sensor by excluding the parasite light issues.
ReplyDelete-yang ni
I don't understand what you mean by "parasite light". Do you mean stray light? If so, where does such stray light come from?
DeleteYes I mean stray light. When you shine light on the sensor, it will not be fully absorbed. Some will be reflected back onto the sensor surface. For me, it's important to check at first the pixel's DR and then check the intra-scene DR of the pixel array.
Delete-yang ni
The lenses in the Google StreetView R5 and R7 cameras look straight into the sun, and can simultaneously make out details on buildings on the shady side of the street. The sun is approximately EV-32, and R7 can see into shady building walls around EV-11 or so. That's an intensity difference of 2 million, or 126 dB, between different parts of the scene.
ReplyDeleteThis works in the real world. It does require cleaning each lens cover every morning and operators checking the cleanliness of the lenses in real time, and having the drivers stop the car to clean the lenses if they collect bugs or excessive dirt. We also assemble the cameras and lenses under a hood, but I think that's pretty common.
I did a bunch of work to get that lens baffled as well as I could. I talked about some of it in this talk: http://www.youtube.com/watch?v=tNvFsOvVkOg The good part starts at 36:19, but after reviewing it I'm not thrilled. I thought I did a better talk at some point, but I can't find it so it must have been an internal talk.
After baffling the lens (don't use internal threads -- they give you rainbows), and blackening the edges of the elements, the worst problems were double-bounce ghosts off first the sensor and then other air/glass surfaces. The basic techniques I used were:
- Get Aptina to do an AR coating on the sensor dust cover
- Reduce the number of air/glass surfaces behind the aperture stop
- Make those rear surfaces with negative radii, so that the double-bounce ghosts off the sensor and them are spread a lot and not much energy ends up on the sensor.
- Make the exit pupil distance short enough that if the sun is much off center, most of the reflected light ends up in the baffles.
Then we started playing whack-a-mole with the double-bounce ghosts:
- Arrange the lens design so that the ghosts were spread very large, so that the intensity was knocked down a lot. For instance, if the sensor is 10% reflective, and the air/glass is 0.5% reflective, and the ghost is 50x the diameter of the sun, then the intensity drops by 22 stops, which is enough.
- Arrange the lens design so that the ghosts end up on top of the sun. It's already blown out, so blowing it out more doesn't matter.
- Arrange the lens design so that the ghosts are much farther from the image center than the primary image. This isn't great, but if the sun is far enough off-axis it works.
During the design I forgot to check for ghosts of the sun when the sun is just outside the field of view. As a result, there is a top-hat-shaped flare when the sun is just outside the field. That's why I designed that damn bulky spherical baffle outside the camera. It doesn't work in the corners, so you can see the top hat in a few images. I checked around just now and couldn't find any, so it's satisfyingly uncommon.
Bottom line: you can make a camera that delivers 125+ dB dynamic range to the sensor, so long as only a small amount of the scene is really bright.
Very informative, thanks!
Delete