Calculated estimate of HVX200 CCD resolution at DVinfo.net

Steve Mullen · January 27th, 2006, 08:03 PM

A post in another thread asked:

"I was a bit confused about the resolution test. What do the many question marks in the HVX200 column mean?

1080i AND 720p camera resolutions at 24F:

TVL/ph (H rez.) TVL (V rez.)

550? 540?

My initial response: The HVX200 resolution seems very low. It's a shame if these numbers are accurate."

--------

For past work, I've built a math model that estimates measured resolution given CCD resolution. It incorporates: an Interlace Factor, a Kell Factor, a Green-shift Factor, and a De-interlacer Factor.

For existing HD cameras -- including those in the test except the HVX200 -- the total (horizontal and vertical) AVERAGE error of my model's estimates is zero lines. (The vertical average error is zero. The horizontal average error is zero.)

So I plugged in the measured HX200 resolution numbers -- assuming they were valid -- and solved for the minimum error.

The solution indicated the three CCDs were 1080 by 812. Hmmm! 1080x810 is a perfect 4:3 ratio.

Putting in 1080x810 yielded a total (horizontal and vertical) average error, for ALL HD cameras, of -9-lines. This estimate did not use Green-shift which it is rumored the HVX200 chips use.

So next I plugged in 960x720 (which I have always believed is the CCDs resolution) and turned ON Vertical and Horizontal Green-shift, and the total (horizontal and vertical) average error for ALL HD cameras was +6-lines.

This model assumed 720p and 1080 24F is obtained by de-interlacing the chips' output. If this is correct -- there should be little difference in sensitivity between 720p/1080f24 and 1080i modes because the chips are always running in interlace mode.

Two other CCD resolutions have been proposed: 960x576 and 960x480. These is are widescreen PAL or NTSC CCD's resolutions. Running the model with these resolutions, and assuming the CCDs could run in both interlace and progressive modes, indicated the 960x480 option was a possible option. It yielded a total (horizontal and vertical) average error of 2-lines. If this option is valid, there should be a sensitivity difference between progressive and interlace modes -- with 720p24/1080f24 being 1-stop less sensitive.

The Vertical resolutions estimated by the models:

INTERLACED 960x720 CCDs (720p and 1080 24F mode) = 567 TVL

SWITCHABLE 960x480 CCDs (720p and 1080 24F mode) = 540 TVL

MEASURED (720p) = 540 TVL

Given that my model predicts Vertical resolution across all HD cameras (without the HVX200) with ZERO error, my model's estimate should be very close. The obvious best match, to the measured vertical resolution, is the SWITCHABLE 960x480 CCDs model.

HOWEVER:

Using 960x480, the estimate for 1080i60 vertical resolution is only 378-lines which is way too low. (This mode was not tested!) Using 960x576 increases estimated vertical resolution for 1080i60 to 454-lines which is better, but still low. But, with 960x576 the 1080 24F and 720p vertical resolution estimates are too high at 648-lines. For this reason, neither of these options seem valid to me.

--------------------

Now to your question about the question marks.

I think the reason why the measures were so hard to get is that Panasonic is not using a typical design. I believe that using INTERLACED 960x720 CCDs running with both vertical and horizontal Green-shift AND that significantly under-sample 1080 AND that de-interlace for 720p and 1080 24F -- output resolution patterns that are quite unusual looking. Hence the question marks.

--------

Five more points:

1) Although not tested, the INTERLACED 960x720 CCDs option estimates that in 1080i60 mode (not tested) the resolution measures will be the same as when running at 1080 24F. This result support the INTERLACED 960x720 CCDs option.

2) Panasonic does not have SSE so we can speculate they avoided this by using a much lower rez. CCD than did JVC. Both options use a lower rez. than the JVC pixels.

3) The NTSC 960x480 CCDs would make the HVX200 more sensitive than any of the other low-cost HD camcorders. And, that's exactly what the test found. However, because the 960x720 is also lower resolution than the other low-cost HD camcorders, one would expect it to also be more sensitive. Both options are supported by the data.

4) Panasonic refuses to give us the CCD spec. and were they to be using PAL or NTSC CCDs one can see why they would really not want that fact known. This favors the SWITCHABLE 960x576/480 CCDs options. Although, because of the flack Sony got for not using HD resolution CCDs, Panasonic may have wanted to hide even 960x720 since it is not the 1280x720 resolution offered by JVC.

5) The sensitivity in 720p verses 1080i will tell us which option is correct -- not tested, unfortunately!

Jeff Kilgroe · January 28th, 2006, 11:22 AM

Quote:

Originally Posted by Steve Mullen

I think the reason why the measures were so hard to get is that Panasonic is not using a typical design. I believe that using 960x720 CCDs running with both vertical and horizontal Green-shift AND that significantly under-sample 1080i AND that use de-interlacing for 720p -- output resolution patterns that are quite unusual looking. Hence the question marks.

PLEASE REMEMBER -- THIS IS SPECULATION ONLY!

While we're speculating, I have a few other possibilities and comments to offer. :)

First, it's pretty well accepted that the EIA resolution charts don't hold up all that well with resolutions in excess of about 800 to 900 lines. Even with cameras like the Viper or F950U CineAlta with their full 1920x1080 censors. So, resolution chart tests with these new 1/3" HD cameras are a little sketchy... Looking at the limited res chart tests done with the HVX and XLH1, we seem to be seeing 650 to 700 lines out of the HVX and about 750 to 800 lines out of the XLH1. But looking at footage from both cameras, especially the HVX footage, I don't think those res chart tests are telling the whole truth - nowhere close. It's difficult to judge CCD resolution from this when there are obviously limiting factors with the glass and with the chart and test methods.

We have other ways to determine CCD resolution and how it may be sampled. One approach may be simply looking at the specs we do know and trying to match that up with available CCD components.

We also know that the HVX CCD block does in fact scan progressively all the way up to 60Hz. It is also a native 16:9 CCD.

I suppose we could start cross referencing that and other little tidbits along with known component suppliers to Panasonic and see what we come up with.

This is also speculation on my part, but I am of the opinion that the CCD block is a 16:9 1024x768 imager. The pixel shift would effectively allow for a scanning area of up to 1536x1152. Which is a very nice fit for interpolating a 960x720 (DVCPRO 720p) or 1280x1080 (DVCPRO 1080) image.

It is possible that the censor is 960x720. That would in fact give them an equivalent scanning area of 1440x1080. But I would doubt that the censor is any lower than 960x720 and I think the 1024x768 is far more likely. Higher resolutions, like 1280x720, are also highly unlikely.

Steve Mullen · January 28th, 2006, 12:15 PM

Quote:

Originally Posted by Jeff Kilgroe

While we're speculating, I have a few other possibilities and comments to offer. :)

First, it's pretty well accepted that the EIA resolution charts don't hold up all that well with resolutions in excess of about 800 to 900 lines.

THEY did NOT use the EIA charts. See Adam Wilt's dv.com story.

Edwin Huang · January 28th, 2006, 12:50 PM

I'm pretty positive the ccds based on the reported behavior of the camera's in various tests and fact that DVCPRO HD media format captures at 960x720. It only makes sense that the CCDs are that size.

Barry Green · January 28th, 2006, 01:58 PM

On an EIA chart I could make a case for 730 lines vertical, although I'd probably be more comfortable calling it 670.
http://www.fiftv.com/HVX200/1956-Chart-Extraction-1.jpg

Steve, the HVX doesn't work like the DVX. There is no interlace scanning ever. It's always progressive.

On the DVX, going into interlace gains you a full stop of sensitivity. On the HVX it's identical for both 720/60p and 1080/60i -- interlace and progressive sensitivity are identical. And, in fact, it's a little FASTER at 24P and 30P than it is at 60P or 60i -- most likely due to the impossibility of executing a 60fps refresh rate at a full 1/60 shutter speed. In the FILM CAM mode it actually limits the shutter to 350 degrees so there's a little bit of free time to dump the sensor's charge. I believe that is the reason the full 60hz modes are a little bit slower than the others.

Interlace is created from the internal 1080/60p scan. We don't know what the pixel count is, but we have been told that it scans the CCD at 1080/60p (or whatever your frame rate is), and then downconverts that scan to 720p, slices and interleaves it for 60i, etc.

Stephen L. Noe · January 28th, 2006, 02:39 PM

Quote:

Originally Posted by Edwin Huang

I'm pretty positive the ccds based on the reported behavior of the camera's in various tests and fact that DVCPRO HD media format captures at 960x720. It only makes sense that the CCDs are that size.

I'm leaning that way as well, however, it's almost impossible to reverse engineer without solid constants to begin with. The rez results they've come up with and the fact that they are using pixel shift and from experience with other Panasonic camera's I'm 51% sure the CCD's are 960x720. But maybe not!?!

Shannon Rawls · January 28th, 2006, 03:53 PM

Steve,

For us non-rocket scientists.....are you saying the HVX might be scanning interlaced?

Give us your hypothesis using a little street talk with a touch of slang. *smile*

- ShannonRawls.com

Steve Mullen · January 28th, 2006, 03:56 PM

Quote:

Originally Posted by Barry Green

On an EIA chart I could make a case for 730 lines vertical, although I'd probably be more comfortable calling it 670.
http://www.fiftv.com/HVX200/1956-Chart-Extraction-1.jpg

Steve, the HVX doesn't work like the DVX. There is no interlace scanning ever. It's always progressive.

Since all the cameras were not measured using an EIA chart, I can't use that data in my model.

Barry, I've seen no information from Panasonic that describes anything about the way the HVX works -- in particular any notion it has 1080p chips.

All we have is the resolution data from the tests.

Thank you for the information that 720/60p and 1080/60i have equal sensitivity. That is the data I needed to reject the 960x576/960x480 options. So that supports the 960x720 resolution option.

Curious if you have a resolution number for 1080i60.

Steve Mullen · January 28th, 2006, 04:04 PM

Quote:

Originally Posted by Stephen L. Noe

I'm leaning that way as well, however, it's almost impossible to reverse engineer without solid constants to begin with.

That's why I presented the fact that my model has ZERO average error with ALL the HD cameras including the Varicam and CineAlta.

That step was necessary to get ALL the constants correct before running the model in reverse with the HVX200 measurements.

It even makes it possible for me to test the idea the HVX has 1080p chips.

Which I'll do -- but given JVC's experience with 720p chips running too hot that seems very unlikely. Of course, Panasonic could be running it in dual 540-line mode and doing a better job of matching both halves. And, you have to ask why Panasonic would hide a "feature" like 1080p.

ESTIMATE: Using 960x1080p CCDs (with H. Green shift) or 1080x1080 (without H. Green-shift), the model's error is a whopping 74-lines! If the CCDs were 1080p, the model predicts V. rez. in 720p and 108024F to be 756 TVL. Since they measure at 540-lines it seems very unlikely 1080p CCDs are being used.

I'm now wondering if the question marks indicate a behind the scenes disagreement because the data measured clearly do not match the "concept" of 1080p CCDs.

Moreover, the EIA 1956 chart is not the correct chart for HD -- see Adam's dv.com comments. So I'll stick with the data.

Bob Grant · January 28th, 2006, 04:33 PM

This might be a totally nuts idea but I think there's another way to determine the CCD size, use the dark current noise. With no light hitting the CCDs each pixel of noise is one element firing by itself, stack enough frames and look at each channel should reveal the number of elements generating the noise and hence the true CCD resolution, probably best done by capturing 4:4:4.

Here's a question that I don't quite get. For any camera using pixel shifting just how much advantage is there in 4:2:2 sampling over 4:2:0, if the pixels aren't coplanar isn't the result of recording 4:2:2 going to be about the same as interpolating 4:2:0 to 4:2:2?

Barry Werger · January 28th, 2006, 05:49 PM

Steve...

Can you share your model with us, or is it proprietary? At least, tell us what you can about it...?

Ash Greyson · January 28th, 2006, 06:18 PM

I believe it was confirmed by Panny that the HVX would record 1080p... I never saw anything that said the chips were NATIVE 1080P... Barry could confirm I am sure. 960X720 seems likely to me, I edit my Varicam footage in that and it looks great...

ash =o)

Shannon Rawls · January 28th, 2006, 06:19 PM

Quote:

Originally Posted by Ash Greyson

I believe it was confirmed by Panny that the HVX would record 1080p... I never saw anything that said the chips were NATIVE 1080P...

Ash....the Panasonic website says:
1/3" 16:9 native high-sensitivity progressive 3-CCD with 1080/60p scanning
Pretty clear to me. So obviously it's a Progressive chip like Barry says. We just don't know if its 1080 horizontal or verticle. If anything else was discovered....that would be catastrophicly bad for Panasonic.

Quote:

Originally Posted by Bob Grant

For any camera using pixel shifting just how much advantage is there in 4:2:2 sampling over 4:2:0, if the pixels aren't coplanar isn't the result of recording 4:2:2 going to be about the same as interpolating 4:2:0 to 4:2:2?

Good question Bob. But I think it may still be better because it's not losing any color information during the shift. Whereas 4:2:0 has to "guess" what color is next.

But again, I'm not sure. Barry? DSE?? Chris??

- ShannonRawls.com

Chris Hurd · January 28th, 2006, 06:42 PM

Quote:

Originally Posted by Steve Mullen

you have to ask why Panasonic would hide a "feature" like 1080p.

Perhaps we haven't been reading the same sources, but I've noticed numerous occasions where Panasonic went out of their way to state that this camera did have 1080p. In fact to me it seemed like a broken record at times.

Matthew Wauhkonen · January 28th, 2006, 07:54 PM

Jan went on the record as saying that the camera did not have 960X720 CCDs or 960X540 CCDs.

It could be like 960X719, though....

Personally, I'm betting it's something way lower, like 720X540 with lots of pixel shifting. We're seeing similar amounts of noise to the dvx at a similar sensetivity (in progressive mode) so I think it's got a similar pixel density, just with intense pixel shifting and true 16X9 chips.

January 27th, 2006, 08:03 PM	#1
Steve Mullen HDV Cinema Join Date: Feb 2003 Location: Las Vegas Posts: 4,007	REVISED: Calculated Estimate of HVX200 CCD Resolution A post in another thread asked: "I was a bit confused about the resolution test. What do the many question marks in the HVX200 column mean? 1080i AND 720p camera resolutions at 24F: TVL/ph (H rez.) TVL (V rez.) 550? 540? My initial response: The HVX200 resolution seems very low. It's a shame if these numbers are accurate." -------- For past work, I've built a math model that estimates measured resolution given CCD resolution. It incorporates: an Interlace Factor, a Kell Factor, a Green-shift Factor, and a De-interlacer Factor. For existing HD cameras -- including those in the test except the HVX200 -- the total (horizontal and vertical) AVERAGE error of my model's estimates is zero lines. (The vertical average error is zero. The horizontal average error is zero.) So I plugged in the measured HX200 resolution numbers -- assuming they were valid -- and solved for the minimum error. The solution indicated the three CCDs were 1080 by 812. Hmmm! 1080x810 is a perfect 4:3 ratio. Putting in 1080x810 yielded a total (horizontal and vertical) average error, for ALL HD cameras, of -9-lines. This estimate did not use Green-shift which it is rumored the HVX200 chips use. So next I plugged in 960x720 (which I have always believed is the CCDs resolution) and turned ON Vertical and Horizontal Green-shift, and the total (horizontal and vertical) average error for ALL HD cameras was +6-lines. This model assumed 720p and 1080 24F is obtained by de-interlacing the chips' output. If this is correct -- there should be little difference in sensitivity between 720p/1080f24 and 1080i modes because the chips are always running in interlace mode. Two other CCD resolutions have been proposed: 960x576 and 960x480. These is are widescreen PAL or NTSC CCD's resolutions. Running the model with these resolutions, and assuming the CCDs could run in both interlace and progressive modes, indicated the 960x480 option was a possible option. It yielded a total (horizontal and vertical) average error of 2-lines. If this option is valid, there should be a sensitivity difference between progressive and interlace modes -- with 720p24/1080f24 being 1-stop less sensitive. The Vertical resolutions estimated by the models: INTERLACED 960x720 CCDs (720p and 1080 24F mode) = 567 TVL SWITCHABLE 960x480 CCDs (720p and 1080 24F mode) = 540 TVL MEASURED (720p) = 540 TVL Given that my model predicts Vertical resolution across all HD cameras (without the HVX200) with ZERO error, my model's estimate should be very close. The obvious best match, to the measured vertical resolution, is the SWITCHABLE 960x480 CCDs model. HOWEVER: Using 960x480, the estimate for 1080i60 vertical resolution is only 378-lines which is way too low. (This mode was not tested!) Using 960x576 increases estimated vertical resolution for 1080i60 to 454-lines which is better, but still low. But, with 960x576 the 1080 24F and 720p vertical resolution estimates are too high at 648-lines. For this reason, neither of these options seem valid to me. -------------------- Now to your question about the question marks. I think the reason why the measures were so hard to get is that Panasonic is not using a typical design. I believe that using INTERLACED 960x720 CCDs running with both vertical and horizontal Green-shift AND that significantly under-sample 1080 AND that de-interlace for 720p and 1080 24F -- output resolution patterns that are quite unusual looking. Hence the question marks. -------- Five more points: 1) Although not tested, the INTERLACED 960x720 CCDs option estimates that in 1080i60 mode (not tested) the resolution measures will be the same as when running at 1080 24F. This result support the INTERLACED 960x720 CCDs option. 2) Panasonic does not have SSE so we can speculate they avoided this by using a much lower rez. CCD than did JVC. Both options use a lower rez. than the JVC pixels. 3) The NTSC 960x480 CCDs would make the HVX200 more sensitive than any of the other low-cost HD camcorders. And, that's exactly what the test found. However, because the 960x720 is also lower resolution than the other low-cost HD camcorders, one would expect it to also be more sensitive. Both options are supported by the data. 4) Panasonic refuses to give us the CCD spec. and were they to be using PAL or NTSC CCDs one can see why they would really not want that fact known. This favors the SWITCHABLE 960x576/480 CCDs options. Although, because of the flack Sony got for not using HD resolution CCDs, Panasonic may have wanted to hide even 960x720 since it is not the 1280x720 resolution offered by JVC. 5) The sensitivity in 720p verses 1080i will tell us which option is correct -- not tested, unfortunately! __________________ Switcher's Quick Guide to the Avid Media Composer >>> http://home.mindspring.com/~d-v-c Last edited by Steve Mullen; January 28th, 2006 at 03:44 PM. Reason: Better information

January 28th, 2006, 03:53 PM	#7
Shannon Rawls Major Player Join Date: Nov 2003 Location: Los Angeles, California Posts: 853	Steve, For us non-rocket scientists.....are you saying the HVX might be scanning interlaced? Give us your hypothesis using a little street talk with a touch of slang. smile - ShannonRawls.com __________________ Shannon W. Rawls ~ Motion Picture Producer & huge advocate of Digital Acquisition.

January 28th, 2006, 12:50 PM	#4
Edwin Huang Regular Crew Join Date: Jul 2005 Posts: 43	I'm pretty positive the ccds based on the reported behavior of the camera's in various tests and fact that DVCPRO HD media format captures at 960x720. It only makes sense that the CCDs are that size.

January 28th, 2006, 01:58 PM	#5
Barry Green Barry Wan Kenobi Join Date: Jul 2003 Location: North Carolina Posts: 3,863	On an EIA chart I could make a case for 730 lines vertical, although I'd probably be more comfortable calling it 670. http://www.fiftv.com/HVX200/1956-Chart-Extraction-1.jpg Steve, the HVX doesn't work like the DVX. There is no interlace scanning ever. It's always progressive. On the DVX, going into interlace gains you a full stop of sensitivity. On the HVX it's identical for both 720/60p and 1080/60i -- interlace and progressive sensitivity are identical. And, in fact, it's a little FASTER at 24P and 30P than it is at 60P or 60i -- most likely due to the impossibility of executing a 60fps refresh rate at a full 1/60 shutter speed. In the FILM CAM mode it actually limits the shutter to 350 degrees so there's a little bit of free time to dump the sensor's charge. I believe that is the reason the full 60hz modes are a little bit slower than the others. Interlace is created from the internal 1080/60p scan. We don't know what the pixel count is, but we have been told that it scans the CCD at 1080/60p (or whatever your frame rate is), and then downconverts that scan to 720p, slices and interleaves it for 60i, etc.

January 28th, 2006, 04:33 PM	#10
Bob Grant Trustee Join Date: Nov 2005 Location: Sydney Australia Posts: 1,570	This might be a totally nuts idea but I think there's another way to determine the CCD size, use the dark current noise. With no light hitting the CCDs each pixel of noise is one element firing by itself, stack enough frames and look at each channel should reveal the number of elements generating the noise and hence the true CCD resolution, probably best done by capturing 4:4:4. Here's a question that I don't quite get. For any camera using pixel shifting just how much advantage is there in 4:2:2 sampling over 4:2:0, if the pixels aren't coplanar isn't the result of recording 4:2:2 going to be about the same as interpolating 4:2:0 to 4:2:2?

January 28th, 2006, 05:49 PM	#11
Barry Werger Regular Crew Join Date: Jan 2006 Posts: 104	Steve... Can you share your model with us, or is it proprietary? At least, tell us what you can about it...?

January 28th, 2006, 06:18 PM	#12
Ash Greyson Trustee Join Date: May 2003 Location: Tulsa, OK Posts: 1,689	I believe it was confirmed by Panny that the HVX would record 1080p... I never saw anything that said the chips were NATIVE 1080P... Barry could confirm I am sure. 960X720 seems likely to me, I edit my Varicam footage in that and it looks great... ash =o)

January 28th, 2006, 07:54 PM	#15
Matthew Wauhkonen Major Player Join Date: Apr 2005 Posts: 285	Jan went on the record as saying that the camera did not have 960X720 CCDs or 960X540 CCDs. It could be like 960X719, though.... Personally, I'm betting it's something way lower, like 720X540 with lots of pixel shifting. We're seeing similar amounts of noise to the dvx at a similar sensetivity (in progressive mode) so I think it's got a similar pixel density, just with intense pixel shifting and true 16X9 chips.