• Blogs
• Forums
• Tradeshows
• Partners
• White Papers
• Events

 

Shoot Review: Sony HVR-V1U, Part 2

May 1, 2007 12:00 PM, Reviewer: D.W. Leitner

Uncompressed HD capture with Blackmagic Design Intensity.

In my last review of Sony's 3-CMOS HVR-V1U (see digitalcontent
producer.com/
cameras/revfeat/
sony_hvrvu), I sketched the V1's pedigree, layout, operation, and unique features. In this article, I'll tackle the V1's image quality and new opportunities in high-quality image capture made possible by the V1's HDMI output and Blackmagic Design's new Intensity HDMI capture card.

As noted in my last article, the V1 is a complex imaging system with myriad setups and results. So I won't attempt the forensics of every picture profile, a topic better left to the intersection of technique and taste. Rather, I want to convey a broader sense of how this unique 3-CMOS camcorder is different from the 3CCD camcorders we've grown accustomed to since the late 1980s.

Every image from the lens of a digital camcorder is spatially sampled by a pixel-based imager; then converted from analog to digital; then encoded into video components for signal processing; then compressed by a codec for recording to tape, disk, or flash memory. Sony's V1 introduces key innovations at each of the first three steps.

As noted in a previous look at the V1 (see digitalcontentproducer.com/hdhdv/depth/sony_hvrv1u_
09192006), Sony's 16:9 1/4in. ClearVid CMOS chips, 1 megapixel each, use a novel pixel grid that's tilted 45 degrees. Pixels, in other words, are diamond-shaped instead of square. Why on earth would Sony do this?

Briefly, in a CCD, charges accumulate at each pixel (photodiode) during exposure. To create an analog voltage for output, charges are then transferred out one column (or row) at a time in strict ordinal sequence with precise timing. As passive collectors of photoelectric charges, CCDs are, therefore, dumb devices. They're basically an area array of incident light meters. (Remember those?)

Unlike CCDs, which emerged from advanced computer memory research (Bell Labs, late 1960s), CMOS evolved from common integrated circuit (IC) design. The idea behind CMOS was to create a simpler, cheaper (theoretically), and smarter imager, with built-in A/D and signal processing located at every pixel. This is why CMOS is sometimes called “retina on a chip.” (The eye's retina is actively part of the central nervous system.) Unlike CCD pixels, whose accumulated charges are marched out of a CCD's pixel architecture like soldiers in single file, CMOS pixels are individually and randomly addressable, and they're output from a CMOS chip in a torrent of parallel streams — often six or more.

So why did Sony tilt the V1's pixel grid 45 degrees? Unlike 3CCD cameras that shift red and blue CCDs a half-pixel horizontally and/or vertically relative to the green CCD (“pixel shifting”) to simulate a greater pixel count and, therefore, more resolution (e.g. Panasonic AG-HVX200), the V1 introduces a new interpolation technique in which virtual pixels are created at each spot where four adjacent diamond-shaped photodiode sites meet. (I'm tempted to say, “X marks the spot.”)

Take a piece of paper and draw a checkerboard grid with at least nine squares. Make a dot in the center of each square. These dots represent square pixels. Next, at each point where four squares meet, make another set of dots. These would be interpolated pixels, created from each of the four surrounding squares. Notice that the new set of dots doesn't fit pre-existing columns and rows of pixels. Now turn your drawing 45 degrees. Draw a new set of horizontal and vertical lines connecting all the dots, real and interpolated alike, to form a new pattern of columns and rows. Notice that each of the new pixels (in reality, one each for R, G, and B) is interpolated from not two but four surrounding pixels. Pretty clever, huh?

CMOS enables this because it lacks physical registers (pipelines) situated between each pixel column (or row) along which a CCD's photodiode charges must flow. It simply isn't possible to angle a CCD's pixel grid. The problem with classic pixel shifting is that while it may possibly improve stairstep aliasing in the luma signal, it does little for aliasing in chroma. The V1's three 1080×960 CMOS imagers, which are not shifted relative to one another, employ this angled-pixel technique to double their horizontal count of 960 photo sites to 1,920 effective sites, delivering better aliasing than pixel shifting can.

Now, what if each CMOS pixel could have its own ASA? (Another antiquated term, but you know what I mean.) In other words, what if CMOS pixels capturing image highlights could have a lower ASA, and pixels capturing shadows a higher ASA? Not remotely possible with CCDs.

Although we're not quite there yet, some incipient version of this seems to be at play in the V1. Consider: Each CMOS pixel has its own transistor amplifier; gain is applied at the CMOS pixel level; and the downstream Enhanced Imaging Processor, a large Sony ASIC for CMOS signal processing (introduced in the single-CMOS HVR-A1U), contains all the CMOS driver circuitry.

Among other things, the V1's 14-bit EIP teases apart what Sony calls “textures” — or delicate high-frequency image data that conveys the visual feel of fine detail — and “brightnesses” — or coarser tonal scale detail that can accommodate broader manipulation — and processes them as separate components. This discourages the user from bludgeoning the image with over-enhancement while affording new opportunities in tonal reproduction, including separate processing of the extremes of highlight and shadow that CMOS can register. (EIP also enables the V1's active viewfinder histogram.)

Prior to encoding, EIP converts the V1's diamond-shaped pixels into conventional square pixels, then proceeds to process them as 1920×1080p, 4:2:2. This is what the HDMI jack makes available in interlaced form (contrary to reports elsewhere that it is the lesser 1440×1080i). By comparison, when recorded to HDV tape, the resulting signal is resolution-reduced to 1440×1080i, chroma-reduced to 4:2:0, and compressed using MPEG-2.

When snow was still on the ground in Manhattan, I shot a series of V1 tests under blue skies. In other words, extreme contrast. I tested various functions at 24p, 30p, and 60i. (24p exhibited the expected judder.) I tested for the appearance of the CMOS's rolling shutter, but I couldn't detect it at any frame rate. At 24p, there was the occasional moiré — in one case, an unusual diagonal moiré on a brick building in the distance. When I switched to 60i, it went away.

Cinematone, as expected, darkened the appearance of the image, which made a significant difference in capturing highlight details of snow. (Picture Profile 2, Cinematone Gamma Type 1.) For instance, other things being equal, the visual comparison between 30p with Cinematone and 60i without Cinematone was the difference between a full tonal scale with smooth highlights and a garish, video-ish image with blown-out highlights.

Out of curiosity, I tested the Contrast Enhancement feature (CNTRST ENHCR in CAMERA SET menu) meant to soften high-contrast scenes. Not a single JND (just noticeable difference) as far as I could tell. But I did notice an odd phenomenon. I would lock exposure (1080i, f/5.2, 0dB, 1/60 sec, Contrast Enhancement off) and slowly pan past a large patch of snow. As the snow entered the frame, 100-percent zebras would be present, but after a beat, the zebras would go away. Upon playback, the effect was definitely there. These scenes were shot with Picture Profile off, so it seems there's some sort of default Auto Knee or Dynamic Contrast Control circuit in operation when the V1 is not in the Picture Profile mode where Knee Point adjustment can be selected.

Outdoors, I captured clips to HDV tape, and indoors, I captured uncompressed HD via HDMI to Blackmagic Design's Intensity card using a RAID 0-striped set of internal drives in a dual-3GHz dual-core Intel Xeon-based Mac Pro with Apple Final Cut Pro 5.1.4.

It's worth noting that in the same spirit that Sony's DCR-VX1000 launched the FireWire era in 1995, the V1 is the first professional/prosumer camcorder to output uncompressed HDMI. (Sony's tiny consumer HDR-HC3 from February 2006 may have been the very first HDMI-enabled camcorder.) And Blackmagic Design's timely $249 Intensity card is the first HDMI capture solution. I must say I never, ever thought I'd see the day when I'd be capturing and editing uncompressed HD in my bathrobe.

I had intended to write at length about my adventures in the strange, new land of HDMI, but because (apologies to Apple) Intensity “just works,” there's not a lot to report. As of this writing, Intensity supports only Windows systems and Intel-based Mac Pros, but you're going to want nothing less than the fastest, brawniest box to handle uncompressed HD at upwards of 1Gbps. The quad-core Mac Pro I have been using, which does the trick nicely, has 4GB of RAM and four internal 500GB 7200rpm Hitachi SATA drives. I set aside one internal drive for OS X and applications, including Final Cut Pro 5.1.4, and striped the remaining three internal drives using Apple's Disk Utility into a 1.4TB RAID 0 volume. They proved fast enough for uncompressed HDMI capture, playback, and editing.

After Intensity's software was installed, I selected “Blackmagic HDTV 1080i 59.94 Uncompressed” from FCP's Capture Preset list and clicked FCP's Capture Now button, because there's no machine control of HDMI. (HDMI doesn't carry timecode, either.) Under “Item properties” the captured clip's data rate was listed as 118.8MBps, or a blazing 950Mbps. I made a series of cuts and added simple dissolves. It all played back in realtime and at pixel-per-pixel, 100-percent scale on the 23in. Apple Cinema Display. The longest single clip I captured was 20 minutes.

To preview clips before capturing, I simply strung an HDMI cable between the V1 — whose robust HDMI jack is full-sized, unlike mini-HDMI jacks on other HDV and AVCHD camcorders — and the Infinity card mounted in the Mac Pro. Then I selected “Log and Capture” in Final Cut. When I pressed play on the V1, uncompressed HD appeared in the Log and Capture preview window as expected. I routinely previewed footage with the waveform monitor and vectorscope that are available in the Clips Settings tab.

I found that a minute of uncompressed 1080i HDMI gobbled up 6.9GB of the internal 1.4TB RAID. At that rate, 1.4TB would capture 2 hours and 20 minutes. But, of course, it's wise to keep 25 percent to 30 percent of disk capacity free as headroom for disk data management, variations in disk speeds, and fragmentation, so we're really talking well less than 2 hours. Impractical for long-form unless you intend to invest in a big, fast Fibre Channel or SCSI RAID system.

The affordable alternative is to capture uncompressed HDMI using compression, which still yields results far superior to HDV compression. For capture, I chose Maxtor's 1.5TB OneTouch III Turbo drive, which proved to be a perfect fit. Like the MacPro's internal drives, I striped the Turbo to RAID 0, which yielded a single 1.4TB volume. (The 1.5TB Turbo contains twin 750MB drives that use new perpendicular magnetic recording for greater recording density and higher capacity.) While the Turbo's FireWire 800 connection isn't nearly fast enough to reliably capture uncompressed HD — I tried it for giggles, and got stuttering playback and dropped frames — it's perfect for direct capture of HDMI to DVCPRO HD at 100Mbps.

I found that a minute of uncompressed 1080i HDMI captured to DVCPRO HD (1280×1080) via Final Cut consumed about 840MB — a huge savings in disk space over uncompressed capture. The same clip captured to higher-resolution JPEG (1920×1080) used a little more than 1GB. (It's a “Blackmagic HDTV 1080i 59.94 — JPEG” capture preset in FCP.) I have since downloaded the beta of CineForm's wavelet-based Intermediate QuickTime Mac OS X codec and expect to try this soon. Then there's Apple's own 220Mbps ProRes 422, just announced at NAB, waiting in the wings. You can see where this is going.

Last month at NAB, Gefen showed preproduction models of its $700 Wireless for HDMI Extender, which uses radio signals to transmit 1080i HDMI up to 30ft. Meant for the wireless living room, it consists of a compact transmitter and receiver that guarantee up to 480Mbps. By my reckoning, 1440×1080i, 4:2:2, 8-bit video should require 747Mbps, but Gefen's system applies “realtime, visually lossless” wavelet compression, according to the company. I cite this as another example of HDMI potentially meeting low-cost field production needs. Especially as powerful software waveform/vectorscope tools such as Adobe DV Rack HD (for Windows, reincarnated for its July release as Adobe OnLocation CS3) and ScopeBox (Mac, version 1.1 shown at NAB) gain fluency in uncompressed HDMI.

Hey, don't scoff. While much of this is beyond the capabilities of current laptops, Necessity, Invention's mother, is a busy gal these days.

---

bottomline

Company: Sony
www.sony.com/professional

Product: HVR-V1U

Assets: Superior image quality, new image-capture options afforded by Blackmagic Design Intensity HDMI card.

Caveats: Uncompressed HD capture requires large, fast RAID arrays.

Demographic: Anyone wanting high-quality images with uncompressed HD capture.

PRICE: $4,890

© 2008 Penton Media, Inc.