IBC: Sony announces HVR-V1e

[Steve Mullen]

Originally Posted by Marvin Emms

The mistake Sony made was in releasing the HC1 in the first place, it was too good.

I too wonder about this. The HC1/A1 had a very sophisticated DSP processing system for wide-lattitude. Was that one R&D group's solution while another group was working on ClearVid?

Or, has the DSP from the HC1/A1 been extended into the new cameras?

Somehow, I suspect the ClearVid concept gives Sony the same recorded rez with half the number of pixels so the lattitude issues may take care of themselves.

It certainly helps to have your own LSI R&D facility. :)

By the way, Fuji digital cameras have used diagonal placement for years. It would be nice if you explained these "banding issues" you say Sony doesn't care about.

-------------------------
Steve Mullen
My "Sony HDV Handbook" is available at:
www.mindspring.com/~d-v-c

[Marvin Emms]

"is simply a restatement of the assertion."

Thomas suggested Sony should be calling them 2Mpixel chips, which suggested to me he hadn't followed my point, I felt if I restated in different terms it might be clearer.

please do not link to forums/sites that are non-sponsors

The important thing for resolution on one axis is the projection of the center of each pixel on the edge. The pixel pitch in this case is 2.05 with the twisted pixels, and 2.9 without. Sony are claiming a 40% increase in vertical and horizontal resolution, the actual math on these numbers yeilds a 41% improvement. I'll come back to this.

The simplest way I can think of to explain it is the same way I double checked my numbers, on graph paper.

If you draw a square on graph paper 40 units on a side, the total area is 40*40 units, it contains that many subsquares - 1600. In this analogy each subsquare is a pixel.

From the midpoint of each side connect to adjacent midpoints to make a diamond. (Its another square, but rotated 45 degrees relative to the first).

If you project the center points of each pixel inside the diamond onto its edge you find you get 40 pixels, the same it true for each side by symmetry.

The edges when you do this properly end up jagged, as one side must lose its half a pixel to be attached to the other side to make complete pixels, but you get the idea.

The number of center pixel projections on a side is 40, the same as for the big square, but the diamond contains exactly half the number of pixels - even though they are exactly the same size (The diamond cuts each quarter square in half so the area is exactly half). We have a 40x40 sensor containing half the number of pixels as the real thing.

A hypothetical 1920x1080 must follow the same result as it can be broken up into square 40x40 blocks recursivly, and since each of those follows this result, they all must. This can be proved explicitly I'm sure, but I don't feel the need and I don't have the concentration left.

So a 1920 by 1080 resolution contains half of the number of pixels it would were it a grid.

Next part.

What is the improvement in resolution on a side?

From simple pythagoras the hypotenuse is SQR( 20^2 + 20^2 ) which boils down to 20 * SQR(2), the old number of pixels per length is 40, the new is 20SQR(2), so the proportionate increase in resolution is therefore 40 / 20SQR(2) = 2/SQR(2) = SQR(2) = 1.41 ish.

A 41% increase in resolution, almost exactly the 40% Sony are claiming and exactly the 41% in the numbers on dpreview.

1920x1080 / 2 = 1'036'800 predicted pixels, sony are claiming "1.03 million" for the kin product FX7 in video mode (1.2Mpixel gross in still mode), I am assuming this has the same sensors, they do not give a specific value for the V1e that I can see. This matches.

Native 1920 horizontal and 1080 vertical resolution, which is what Sony claim the camera works from internally. This matches.

I would have included a picture of my construction on graph paper but its a lot of effort to redraw on the computer and with the math being so simple I figure people will either get it or not at all.

I'm done with all the 'evidence' I have to hand that the system works the way I say it does but since I'm in simple math mode, its worth considering what the tradeoff is. If this were interpolation a single sensor would have a lousey measured resolution, it isn't and the trick does genuinly deliver 1980 h and 1080 v resolution with only 1.03Mpixels. Since these are what camcorder reviewers measure Sony are very happy bunnies. The kicker is the diagonal resolution. While the diagonal resolution should by the same maths be SQR(2) higher than the v or h rez (rez here meaning number of pixels per given length), on the clearvid it is SQR(2) lower. 1/SQR(2) * 1/SQR(2) = 1/2. The diagonal resolution is half what it would be on a genuine 1920x1080 camera, but noone measures this on a review. This is why the HC3 aliases badly with diagonal bands when the HC1 does not, and yet the reported total resolution of HC3 is higher, even given its lower number of video pixels.

Pixel shift done properly should even the score on the diagonals somewhat for these two newer cams, assuming they use it. Hopefully neerer release someone can do some proper tests including diagonal resolution on the real thing.

[Steve Mullen]

Originally Posted by Marvin Emms

If you draw a square on graph paper 40 units on a side, the total area is 40*40 units, it contains that many subsquares -- 1600. In this analogy, each subsquare is a pixel.

>>> Agreed. Looking at the 2x2 diagram you linked to -- thank you -- there are 4 cells. <<<


If you project the center points of each pixel inside the diamond onto its edge you find you get 40 pixels, the same it true for each side by symmetry. The edges when you do this properly end up jagged, as one side must lose its half a pixel to be attached to the other side to make complete pixels, but you get the idea.

>>> OK -- let's take FOUR 2x2 squares, which is 16 cells.

In a SQUARE arrangement we have 4x4 = a total "resolution" of 16. I can project 4 verticals and 4 horizontals through these 16 which matches this resolution.

In a DIAMOND arrangement, we have 7x7 = 49 "resolution" as I can project 7 verticals and 7 horizontals through these 16.

However, if I leave out the ragged edge cells, I drop 4 cells, leaving 16-4=12. Now I can can project 5 verticals and 5 horizontals through these 12. So that would be 5x5 "resolution" or 25/12 = 2.08333333333X relationship. You are quite correct! <<<


So a 1920 by 1080 resolution contains half of the number of pixels it would were it a grid.

>>> 995,328 cells -- which if a symetric arrangement is used would be a 998x998 array. If our diamonds are equivilent to square pixels, which I think they are. <<<



What is the improvement in resolution on a side?

From simple pythagoras the hypotenuse is SQR( 20^2 + 20^2 ) which boils down to 20 * SQR(2), the old number of pixels per length is 40, the new is 20SQR(2), so the proportionate increase in resolution is therefore 40 / 20SQR(2) = 2/SQR(2) = SQR(2) = 1.41 ish.

A 41% increase in resolution, almost exactly the 40% Sony are claiming and exactly the 41% in the numbers on dpreview.

>>> Here we must very clear that the 1.4X increase is over the resolution of a "different" chip. But, what is the "different" chip's physical resolution? If the "different" chip is a sensor that has Square elements -- that is 998x998 -- then it's much like a 960x1080 CCD.

With pixel-shift, Sony CLAIMED such chips offered a 1.5X increase in Horizontal resolution -- enough to support 1440x1080. But, the actual increase under dynamic (real-world) testing was only about 15% -- not 50%.

Sony's FX7/V1 should be able to deliver 1.4X TOTAL resolution. Assuming this is even on both X and Y axes -- then the DYNAMIC measured Horizontal resolution of the FX7/V1 should be slightly more (20% verses 15%) than the FX1/Z1 while the Vertical measured resolution should be much more.

The STATIC measured Horizontal resolution of the FX7/V1 would, however, be less or equal to the FX1/Z1 while the Vertical measured resolution would be slightly more. (I don't remember the Z1's static resolution.)

However, were Horizontal pixel-shift used by the FX7/V1 -- the Horizontal resolution would increase under DYNAMIC testing by about 15% and under STATIC testing far more. So, it's possible the FX7 and V1 could static measure as well as the new Canons. So one of the key questions is "is pixel-shift used?" The other question, are square pixels used? And, what are the X and Y pixel counts? <<<


If this were interpolation a single sensor would have a lousey measured resolution, it isn't and the trick does genuinly deliver 1980 h and 1080 v resolution with only 1.03Mpixels.

>>> Carefull here. You do not mean a resolution of 1920x1080 -- you mean the expected effective resolution FROM a 1920x1080 chip. <<<


The diagonal resolution is half what it would be on a genuine 1920x1080 camera, but no one measures this on a review. This is why the HC3 aliases badly with diagonal bands when the HC1 does not, and yet the reported total resolution of HC3 is higher, even given its lower number of video pixels.

>>> Agreed, but other than the A1/HC1 and HDCAM camcorders -- there aren't any 1920x1080 Sony camcorders. So the real question is how does diagonal resolution (aliasing) compare to Sony camcorders with 960x1080 CCDs? <<<


Great job, Marvin!

-------------------------
Steve Mullen
My "Sony HDV Handbook" is available at:
www.mindspring.com/~d-v-c

[Barry Green]

Originally Posted by Steve Mullen

I doubt 960 is scaled by 2X to 1920 only to be down-scaled to 1440.

I'm only going off of what Sony themselves is saying. Sony representatives at IBC said that the V1E and the FX7 have the same sensors. 1/4" CMOS with ClearVid, at a resolution of 960x1080. See Mikko Wilson's report here:
http://www.dvxuser.com/V6/showthread.php?t=70634

Furthermore, Sony's marketing material claims that the internal processing is done at 1920 x 1080.
http://www.sonybiz.net/cgi-bin/bvisa...-V1E(brch).pdf

And we know that the recording format is 1440 x 1080.

So somehow they are sampling or interpolating the 960x1080 chip to deliver a 1920x1080 matrix to the DSP, which then gets scaled to 1440 x 1080 when recording. According to Mikko's report, the Sony rep chose the words "a new interpolation method to improve quality," and I believe I read similar wording in some other report (whether it was their brochure or camcorderinfo's interview or someone else's firsthand report I don't remember).

[Steve Mullen]

Originally Posted by Barry Green

So somehow they are sampling or interpolating the 960x1080 chip to deliver a 1920x1080 matrix to the DSP, which then gets scaled to 1440 x 1080 when recording.

You need to read Marvin's post and mine following his as you are about 2 posts behind us.

There are both a "new processing technology" (Marvin) and the processing is at 1080p (Me) -- so unless we are both wrong -- what Sony says matches what we say.

There is no scaling from 960 to 1920 despite the chips being 960x1080. The 1920x1080 buffer really does get filled with 1920x1080 elements that are directly read-out from the CMOS chips.

And, I think it gets filled 50 or 60 times each second. Which means Sony can get progressive with full V resolution just like JVC.

So now Sony, Canon, and JVC get the pixels to match their HD formats using neither "scaling" nor "pixel-shift."

AND YES -- I am assuming pixel-shift is not used with ClearVid. If I'm correct on this, the DYNAMIC measured Horizontal resolution of the FX7/V1 should be slightly more than the FX1/Z1, while the Vertical measured resolution should be much more.

The STATIC measured Horizontal resolution of the FX7/V1 should, however, equal to or be less the FX1/Z1 while the Vertical measured resolution should be slightly greater.

-------------------------
Steve Mullen
www.mindspring.com/~d-v-c

[Thomas Smet]

Ah yes but how does it look?

...

[Steve Mullen]

Originally Posted by Thomas Smet

Ah yes but how does it look?...

Who cares? It's a Sony. And, if it has true 24p or 25p -- plus 1080i50 or 1080i60.

It is small, light, and likely will have low street price. My feeling to really really replace DV, a camcorder has to have a street price of well under $3000.

Can it look worse than the VX1000? The VX1000 sparked a revolution, although it was far from perfect. :)

Seriously -- the vast majority of videographers are used to the Sony look and so will buy another Sony. The fact that many think other camcorders look better, will not prevent the V1 from being a hit.

But even more seriously, Sony is leveraging its LSI R&D to create an engine that shows real "pro" promise if it is implemented as 1/3-inch chips.

But, I'm surprised that there are no reports from IBC.

[Lawrence Bansbach]

Originally Posted by Steve Mullen

AND YES -- I am assuming pixel-shift is not used with ClearVid. If I'm correct on this, the DYNAMIC measured Horizontal resolution of the FX7/V1 should be slightly more than the FX1/Z1, while the Vertical measured resolution should be much more.

The STATIC measured Horizontal resolution of the FX7/V1 should, however, equal to or be less the FX1/Z1 while the Vertical measured resolution should be slightly greater!

Another forum says that Sony claims 800 lines for the V1 vs 650 for the Z1. I'm assuming that the figure is for static resolution using a chart, but in any case I consider a 23% improvement more than slight. Of course, the figure is Sony's and may be inflated (due to testing optimization) to promoted the V1, but I would be hard-pressed to explain why Sony downplay the Z1 to do this.

And, if it has true 24p or 25p . . .

That would be nice, but Sony has been spitefully resistant to 24p on a prosumer camcorder. I don't know why Sony can't do 50i/60i/24p/25p/30p on the same camera, particularly since CMOS chips are more flexible than CCDs.

[Bob Zimmerman]

maybe they will. I would just like to know soon when something here will be announced.

[Steve Mullen]

Originally Posted by Lawrence Bansbach

I'm assuming that the figure is for static resolution using a chart, but in any case I consider a 23% improvement more than slight. Of course, the figure is Sony's and may be inflated (due to testing optimization) to promoted the V1, but I would be hard-pressed to explain why Sony downplay the Z1 to do this.

Since or moderator has turned-off my ability to EDIT, I have no choice to restate and fix my error:

Assuming no pixel-shift, Sony's FX7/V1 should be able to deliver 1.4X TOTAL resolution. Assuming this is even on both X and Y axes -- measured Horizontal resolution of the FX7/V1 should be slightly more (20% verses 15%) than the FX1/Z1 while the Vertical measured resolution should be 20% greater.

What's different is that FX7/V1 rez should not go down with motion. This is significant.

I'm sure Sony knows that Z1 buyers will not buy the V1. But, this ClearVid in 1/3-inch chips could power the Z2.

[Lawrence Bansbach]

Originally Posted by Steve Mullen

Since or moderator has turned-off my ability to EDIT. . .

My previous post was edited (not by me) to remove the link to the other forum, but the thread is the same one that Barry cites above for Mikko Wilson's report .

[Marvin Emms]

Lawrence,

"Sony claims 800 lines for the V1 vs 650 for the Z1. I'm assuming that the figure is for static resolution using a chart, but in any case I consider a 23% improvement more than slight."

But the Z1 does not use clearvid, so it isn't boosting horizontal and vertical resolution at the expense of diagonal resolution.

Steve,
"Assuming no pixel-shift, Sony's FX7/V1 should be able to deliver 1.4X TOTAL resolution"

I don't follow this, without pixelshift what we have is a 1Mpixel camera.

Its also worth remebering that to make video that looks remotely watchable at 25fps you need something like a 1/50th shutter. So there will be detail lost to motion blur with movement.

[Steve Mullen]

Originally Posted by Marvin Emms

Steve,
"Assuming no pixel-shift, Sony's FX7/V1 should be able to deliver 1.4X TOTAL resolution"

I don't follow this, without pixelshift what we have is a 1Mpixel camera.

Its also worth remebering that to make video that looks remotely watchable at 25fps you need something like a 1/50th shutter. So there will be detail lost to motion blur with movement.

I'm quoting you:

"From simple pythagoras the hypotenuse is SQR( 20^2 + 20^2 ) which boils down to 20 * SQR(2), the old number of pixels per length is 40, the new is 20SQR(2), so the proportionate increase in resolution is therefore 40 / 20SQR(2) = 2/SQR(2) = SQR(2) = 1.41 ish.

A 41% increase in resolution, almost exactly the 40% Sony are claiming and exactly the 41% in the numbers on dpreview."


A 40% increase is 1.4X. Which given the almost 1000x1000 arrangment would be 20% on each axis -- which matches the 21% V. Rez. improvement quoted over the 1000x1000 CCDs in the Z1.

Also, given the pitch spec and drawing -- the ClearVid pixels are square. So to image a 16:9 pix the chips must be 16:9 and not 4:3. (There were IBC photos of the block.)

Lastly, of course, the shutter-speed drops at 25p -- but the motion blur is what film folks want. Same story at 24p and 30p. Drives me nuts to freeze a frame and see that anything with even a tiny motion is blurred. Effective resolution is dropped which kind of reverses the point of HD.

Which is why I love 60p. But, everyone wants to avoid the look of "video" these days -- not realizing that HD has it's own "through the window" look that isn't like the "video look" they don't want. The problem may be that buyers of HDV camcorders typically have zero experience watching HD at home. Worse, many watch their work on a small computer monitor.

HD, as someone wrote, should stand for "High Impact TV." At 8-feet the screen must be at least 50-inches, and 60-inches is even better. Watch 60p like this and it is better than "film" and sure doesn't look like "video." Think IMAX at home.

-------------------------
Steve Mullen
"Sony HDV Handbook"
www.mindspring.com/~d-v-c

[Marvin Emms]

"A 40% increase is 1.4X. Which given the almost 1000x1000 arrangment would be 20% on each axis "

The 41% increase in resolution is on the vertical and horizontal axes. There is no improvement in the total resolution of the system - there cannot be, this is determined by the total number of pixels.

To give an extreme example of what could be done, they could make a sensor that had half a million pixels in a line vertically, and half a million pixels in a line horizontally and try to make an image out of that information. It would look rubbish of course, but hold a series of vertical or horizontal striped lines up to the sensor and it would aprear to have amazing resolution. A quarter of a million lines on each axis. But the total information you have to make an image is the same. 1 Mpixel.

41% is ideal and mathematical, it also assumes that the processing algs are intent on making the most of this resolution, which is why in the real world the answer is less.

I've gone over websites and reports and some sites say clearvid and talk about diagonal pixel structures and some give 960x1080 as the sensor resolution (a claim I cannot find on the Sony site). The problem is they can't both be true. There simply isn't a self consistant numbering system for diagonal pixels where that number works and provides a pixel count of 1.03Mpixels. Something somewhere is being simplified. Beyond this point I don't think guessing with maths will help so I'm just going to have to wait for real reviews.

[Steve Mullen]

Originally Posted by Marvin Emms

The 41% increase in resolution is on the vertical and horizontal axes. There is no improvement in the total resolution of the system -- there cannot be, this is determined by the total number of pixels.

41% is ideal and mathematical, it also assumes that the processing algs are intent on making the most of this resolution, which is why in the real world the answer is less.

I've gone over websites and reports and some sites say clearvid and talk about diagonal pixel structures and some give 960x1080 as the sensor resolution (a claim I cannot find on the Sony site). The problem is they can't both be true. There simply isn't a self consistant numbering system for diagonal pixels where that number works and provides a pixel count of 1.03Mpixels

1) I'm confused totally. You demonstrated how 1M pixels can fill a 2Mpixel buffer -- for each chip. Right? The 1920x1080 buffer rightly will have greater measured resolution. Right? You compute a 40% increase -- which given it reflects the TOTAL increase would logically imply that it would be 20% greater on the Vertical axis over the FX1/Z1. And, that's the increase reported. So, I'm not sure why you seem to be backing-up on your claims of a 1.4X resolution increase.


2) If the chips have 995,328 cells -- which were a symetric arrangement used -- they would have a 998x998 active pixel array. Since the edges must be rejected, that means about 4,000 cells present, but not used. So, now we are up to 1Mpixels.

Now, with 960x1080 pixels the active count is 1,036,800 pixels. The difference between 999,328 and 1,036,800 is so small I would assume that either a 999x999 or 960x1080 array is used. Either way, both are about 1000x1000. So, I'm betting on the standard 960x1080 array.

Since I'll be under NDA in a few hours -- reading the Sony material carefully, the new DSP is a continuation of the DSP used in the HC1/A1. This, means ten things to me:

1) The foreground/background processing system is still being used to allow much greater latitude.

2) The DSP chip has 6 input ports that input 6 pixels at the same time. Given 3 chips, that means 2 pixels are read from each CMOS chip at the same time.

3) By reading 2 pixels at once, the data-output "clock-rate" from a CMOS chip can be cut in half. This lowers power and, therefore, heat. (JVC also reads out 2 pixels at once from each CCD for the same reasons.)

4) Which leads to the question of which 2 pixels get read-out. JVC reads two from the same row. The read-out starts at the edges and works toward the middle. Sony, I expect uses a different system because CMOS can be read randomly. I suspect the read-out begins at the top and bottom rows and works toward the middle. No possibility of SSE, but on very high-speed action the top and bottom pixels will be obtained at an "earlier" time than are the middle pixels. This leads to the bowed-shape (to the right) of a golf club being swung, for example.

Of course, given the diagonal pixel arrangment, it may be better to think of this as the top and bottom rows of the 1920x1080 buffer being filled slightly before the middle rows.

5) It is the ability of the DSP to read pixels randomly, as though it were addressing RAM, that makes it possible to input into the DSP buffer a linear series of 1920-pixels that do not come from a linear row of CMOS elements. This is something unique to CMOS.

6) Pixel-shift is NOT used because the DSP buffer already has 1920-pixels. Which means the FX7/V1 dynamic and static V. & H. Rez measures will be the same. Wonderful! Just like JVC.

7) There will greater V. rez (by 20%) from the FX7/V1 than the FX1/Z1 -- when the FX1/Z1 is measured either statically or dynamically.

8) There will slightly greater H. rez from the FX1/Z1 than the FX7/V1 -- when the FX1/Z1 is measured statically.

9) There will be very slightly greater H. rez from the FX7/V1 than the FX1/Z1 -- when the FX1/Z1 is measured dynamically.

10) The reality is that the image is being formed in and by the DSP. The imaging chips are now only "collectors" of information the DSP needs to create an image. Once created, I believe at 50hz or 60Hz, the DSP is able to scale -- IF necessary for today's limited recording systems -- to 1440x1080 and also order the frames and fields to generate interlace or progressive video AND THEN, if necessary for tape recording, to place progessive video into interlace fields.

The choice of 24p is, therefore, a political one. I expect it not to be present in the FX7.

In short, I belive the CMOS/DSP engine is capable of outputting 1080p50 or 1080p60 to an encoder and to a recording system capable of handling such video. In fact, I suspect the DSP can not only downscale 1920 to 1440 it can also downscale to 1280. And, with 50p or 60p -- 1080 can be downscaled to 720. Thereby, offering super-sampled, 720p50/720p60.

-------------------------
Steve Mullen
"Sony HDV Handbook"
www.mindspring.com/~d-v-c