JVC adopts AVCHD, CMOS for new HD Everios

JVC Everio GZ-HD40

(Credit: JVC America)

Updated 6/23/08 with information about the sensors.

In an interesting turn, JVC's newest hard-disk-based HD camcorders--the Everio GZ-HD40, GZ-HD30, and GZ-HD10--will support AVCHD in addition to its traditional MPEG-2 formats. This is a great idea, at least for the interim, since it theoretically provides you with more flexibility in your video-editing options--especially if its AVCHD files aren't supported right out of the gate--and allows it to deliver high bit rate video beyond the 24Mbps maximum for AVCHD. Plus, the camcorders include FireWire (i.Link) interfaces, which many of today's models have dropped, for dealing with MPEG-2's HDV-compatible datastream.

The three models also integrate new CMOS chips; I've been waiting to hear back from the company with some details about the chips both of the CMOS chips, a 1/4.5-inch, 1.75-megapixel version for the HD10 and a 1/3-inch, 2.68-megapixel version for the HD30/40 represent a significant switch for JVC, which usually uses low-resolution CCDs. The sensors for these models should be high enough resolution to deliver a decent HD image. However, the press information provided is odd. The press release states:

This CMOS uses proprietary interpolation technology based on what JVC developed for its 3CCD cameras. This algorithm makes it possible to generate virtual pixel data from the red, green, and blue physical pixels, thereby providing image information that actually surpasses what's required for 1,920x1,080 Full HD imaging, or for 1,440x1,080 imaging in the GZ-HD10.

But the sensors are "high resolution" enough that interpolation shouldn't be necessary. And not only is interpolation undesirable, generating the "virtual pixel data" from a single chip is far less useful than generating it from three chips that each capture those specific primaries.

The key details on the new models:

	GZ-HD40	GZ-HD30	GZ-HD10
Hard-disk size	120GB	80GB	40GB
Format(s)	AVCHD and MPEG-2	AVCHD and MPEG-2	AVCHD
Sensors	1/4.5-inch, 1.75 megapixels	1/3-inch, 2.6 megapixels	1/3-inch, 2.6 megapixels
Max video dimensions	1920x1080	1920x1080	1440x1080
Maximum bit rate at highest resolution	17Mbps AVCHD 26.6Mbps MPEG-2	17Mbps AVCHD 26.6Mbps MPEG-2	17Mbps AVCHD
Hours recorded (highest/lowest quality)	10/15	6/33	1/3.3
Price	$1,299.95	$999.95	$799.95

Also, the GZ-HD40 comes bundled with a docking station. The three models are slated to ship in August.

[viskoundis]

when a CMOS chip is listed as 2.68-mega pixels 1/3 of those pixels are red 1/3 are green and 1/3 are blue.

True HD has 1080x1920 pixels each of which include the {red,green,blue} information. Just like your HD monitor has 1080 lines and each pixel has three colors (1080x1920 32bit pixels with 10bit per color).

Clearly in order to capture full HD information the CMOS must be at least 3x1080x1920 pixels, about 6 megapixels.
Another way to do the same thing is to use three CMOS or three CCDs with two mega pixels each.

Now JVC or any other consumer "HD" camera does not have 6 mega pixel CMOS or 3x2 mega pixel CCDs (or CMOSs), so in order to record the full HD format they have to interpolate.

A simple way to do the interpolation is, if for instance a CMOS has:

red -green-blue-red -green-blue-red =true

pixel sequence, one could fit a curve between the 1st second and 3rd reds and use the interpolated result in place of the green and blue.

red -green-blue-red -green-blue-red =true
-red -red -red -red =interpolated

Similar process happens for the green and blue pixels.

red -green-blue -red -green-blue -red =true
-red -red -red -red =interpolated
green -green-green -green-green=interpolated
blue -blue - -blue -blue -blue =interpolated

so this way you complete the missing bits.
The same is done for still photography as well. A 9 mega pixel camera is not really 9. It interpolated 9. The sensor chip has 9 mega pixels, but in terms or RGB you only have 3 mega pixels.

The results are fairly ok, but its cheating. That s why the professional (cinema) HD cameras cost $40k now. They have 3 full HD CCDs. That requires supper fast electronics for image processing.

I suspect that as electronics get faster and sensors get cheaper we ll get true HD cameras without interpolation for the consumer market.

[hazydave]

While some of this is technically correct, you're missing a big point.

Back in the days of SD, it was fairly critical to sample one RGB pair per pixel. This wasn't strictly due to any need to see one RGB sample per pixel, but largely due to the small number of pixels on the sensor. Early 3-CCD camcorders did this, early single CCD camcorders didn't. Later on, as now, single CCD camcorders grew much larger sensors, and also provided a full RGB sample per SD pixel, at the cost of light sensitivity over the 3-CCD solution.

Of course, that's not the end of the story. Once the image was sampled, software then downsampled the color... either 4:1:1 subsampling for an NTSC camcorder, or 4:2:0 subsampling for a PAL camera. Either way, you're tossing out 3/4 of the color information in the process of going from CCD output to DV tape input.

The same is true of HD camcorders, both MPEG-2 and AVC models... 3/4 of the color information is tossed out. This is true for professional cameras, too, up to the point where you find much higher bitrates, and only 2:1 subsampling (4:2:2)... you can find this in the XDCAM HD422 format (regular XDCAM and XDCAM EX are 4:2:0 subsampled)

The good news is that, there's so much color in a full HD image (even using a Bayer pattern sensor, which is, by the way, 1/2 green, 1/4 red, and 1/4 blue... not 1/3 of each), your eye doesn't notice the subsampling in most cases. That's part of human anatomy.. you have about 6 million color sensors in your eye (cones... only 1/10th of those are blue sensitive), and about 120 million luma sensors (rods and cones). So enough color is important, more than enough can generally be tossed out without notice.

Light gathering, on the other hand, is pretty critical... thus, many of the 3-sensor pro/prosumer cameras don't have full HD sensors, anyway (this is what JVC was talking about here, their interpolations from the 3CCD world). If you build a 1 megapixel sensor at 1/3" versus a 2 megapixel sensor at 1/3", each pixel has twice the light gathering capability... minimum (perhaps more, since there's usually some gap between sensing elements). So many 3CCD cameras are built with overlapping smaller sensors. They do color interpolation, just as in the single CCD case, then the decimation to 4:2:0 (all MPEG-2 and AVC camcorders use 4:2:0 until you get to some of those higher end XDCAMs and such). Some newer models do offer full HD sensors x 3, but it's still a trade-off.

Even low-end cinema-style camcorders, like Sony's current CineAlta, use 3 chips, 2K / HD format, and compression (though there's usually an uncompressed output) on tape, even if less of it. But these are mostly used in television these days... cinema is evolving to 4K. And most 4K cameras use a single sensor.

And it's usually a CMOS sensor... something much more like what you'll find in a digital SLR these days. Real film replacement cameras, like The Red Camera, record 4K (nominally 4000x2000 pixels). and outputs an uncompressed stream of RAW Bayer Pattern data (eg, the same thing you get when you record RAW with a digital SLR). A number of Arriflex models also use a single CMOS sensor. The Dalsa Origin also uses a single 4K sensor, and puts out RAW Bayer pattern data.

So, no need to be such a snob about a single sensor. For HD, it's basically trading off light for simplicity and cost, not color, as was the case in the SD world. And even HD camcorders with three chips make this decision.