Review: Panasonic AG-HPX370 1/3” 3-MOS P2 HD Camcorder
The HPX370 offers 50/60Hz AVC-Intra and DV-format recording, and improves noise, sensitivity, and skew over the HPX300.
By Adam Wilt | July 29, 2010
The 4.5-77mm 17x Fujinon lens shows no ramping at all in the center of the image: with the aperture wide open, there's no loss of level from full wide to full telephoto. There is some portholing / vignetting at the extremes of focal length with the iris wide open; zoom in from 20mm and the edges of the image darken slightly, with the darkening spreading to form an even vignette across most of the image at 77mm. Maximum light loss at the edge of the picture is about a stop, and it's not really visible other than on the WFM until you're past 40mm. Stop down to f/3.5, though, and the vignetting is banished. There is also a bit of vignetting when zooming wider than 10mm, with just a tiny and visually imperceptible darkening of the corners; again, stopping down to f/2.8 makes it go away.
At 4.5mm there's a little complex mustache distortion, with a bit of barrel bulging in the middle of the image while the extreme edges show pincushioning (straight lines bow inwards at their centers), morphing to slight pincushioning around 10mm, cleaning up to a rectilinear rectitude at full telephoto. Even at the wide end, the distortion isn't really noticeable unless you do something like frame the edge of a doorway in the outermost 10%-15% of the picture.
Minimum focus is a rather lengthy 3 feet, 1 inch (measured from where the shade meets the lens proper; the front of the lens shade is nearly an inch farther forwards). Using the separate macro ring brings M.O.D. down to 2 feet 5 inches at 77mm, and all the way down to inside the front of the lens at 4.5mm. Because the macro ring is a separate control, as is typical on high-end broadcast lenses, shooters transitioning from smaller cameras with integrated zooms may be caught out by the near-focus limits.
As with all prism cameras, there's a bit of vertical green/magenta color fringing on out-of-focus details, but it's not excessive.
The lens shows a slight bit of blue/yellow lateral chromatic aberration at full wide angle, with CA diminishing to an insignificant amount about halfway between 4.5mm and 10mm. Red/green CA starts becoming noticeable around 30mm and is quite prominent by 77mm. Panasonic's CAC (Chromatic Aberration Correction) eliminates any visible trace of lateral CA (if by "lateral" you mean "horizontal"; correction is only done horizontally, so vertical color smear is still visible at the extremes of the focal range).
Chromatic Aberration Correction at full telephoto.
Chromatic Aberration Correction at full wide angle.
The lens is quite sharp even wide open, but the 1/3" true-1080P sensor clearly shows the diffraction-based limits of smaller apertures for small formats. At f/5.6, resolution drops to about 800 TVl/ph and the image is visibly getting softer; go smaller still and the image degrades further. The sweet spot is in a two- to three-stop range from wide open to around f/2.8 or even f/4, but close the lens down further and sharpness suffers.
With focal length starting at 4.5mm, the lens isn't particularly wide by modern standards; the built-in lens on the 1/3" HVX200 goes to 4.2mm, and the HPX170's starts at 3.9mm. HPX370 owner/operators may want to consider a wide-angle adapter, or one of the wider 1/3" zooms available—though when you see what the wider zooms cost, you'll appreciate the stock lens all the more.
If this lens has a weak point, it's breathing: racking focus from near to far causes a substantial change in image magnification, perhaps as much as 20%.
Heavy breathing: the 17x Fujinon changes magnification with focus.
Resolution and Detail
The HPX370 uses 1/3" MOS sensors with 1920x1080 active photosites: a full-res HD sensor. From the full-size image, the camera derives a number of other rasters through downsampling.
I did my usual test of shooting a DSC Labs multiburst-square wave sweep chart and pulling out pixel-for-pixel extractions for your viewing pleasure. Have a look not only at limiting resolutions and over-limit aliases, but also at compression quality, especially around the text labels. These charts were all shot around f/2.8 or f/4, with detail at -6 (on a range from -7 to +7).
1920x1080p AVC-Intra 100 (pixel-for-pixel).
1440x1080p AVC-Intra 50 (expanded back to 1920x1080 in FCP).
1280x720p AVC-Intra 100 (pixel-for-pixel).
960x720p AVC-Intra 50 (expanded back to 1280x720 in FCP).
Pixel-for-pixel 576i and 576p DVCPRO50 (720x576).
Pixel-for-pixel 480i and 480p DVCPRO50 (720x480).
Eyes glazed over yet? My take on these results is that the camera properly resolves (and records) 1000+ TVl/ph in full 1920x1080 AVC-Intra 100 mode, with corresponding reductions in limiting resolution as the sampling raster and/or the frame size is reduced. There's minimal aliasing at full res; downsampled formats show some green/magenta aliasing in the vertical direction, but it's not something that normally makes itself visible in real-world shooting.
In practical shooting, I found that the camera made very smooth, naturalistic images with plenty of detail and no noticeable aliasing of any sort.
Sensitivity and Noise
The AG-HPX370 Series brochure says:
...the same levels of sensitivity and image quality that are found in Interlace mode are now possible in Progressive mode thanks to P.A.P. (Progressive Advanced Processing), a 3D adaptive processing technology.
That means that we're not going to see a raw, unprocessed output from these sensors; it's going to be fiddled with in unusual ways. Indeed, I measured the HPX370's basic sensitivity at 0dB gain as ISO 800 in 1080p, 720p, and all 576- and 480-line modes. In 1080i, the camera was about a third of a stop faster: ISO 1000.
By comparison, I rate the 1/2" PMW-EX1 and EX3 at ISO 500 in 1080p. RED suggests rating their new, S35-sized Mysterium-X sensor at ISO 800. Arri similarly suggests shooting the 35mm-sized Alexa at ISO 800. Here's a 1/3" sensor making a full-HD 1080p image at ISO 800 when at 0dB gain. Not too shabby.
The camera lets you set gain from -3dB to +12dB, and also offers +24dB on a user-definable pushbutton. As part of an exposure test, I shot the same scene at 0dB, +6db (1 stop faster), +12dB (2 stops faster), and +24dB (four stops faster). Here's a pixel-for-pixel extract from those tests, with the gain setting, the effective ISO rating, and the f-stop at which it was shot:
Pixel-for-pixel details at four gains. Inset: source image with detail area outlined.
I point out the f-stop to show the diffraction-limited resolution loss that occurs with small apertures on 1/3" sensors. Note the loss of resolution as the aperture closes down; the +24dB image is sharp despite its noise because I dialed in some ND and opened the iris back up. When you have a small sensor as sharp as those in the HPX370, even relatively large apertures like f/5.6 cause noticeable losses. A separate test I ran, varying light levels instead of apertures, showed no change in resolution with changes in gain.
I compared the HPX370 with the Sony PMW-EX1, a 1/2" 1920x1080 3-CMOS camera. The HPX370 is quieter at 0dB gain than the EX1, despite being nearly a stop more sensitive. As gains rise, the HPX370's edge diminishes; with both cameras at ISO 3200 (EX1 at +18dB and the HPX370 at +12dB), the two cameras are about equally noisy, and with both cameras at +12dB, the EX1 is quieter (if almost a stop less sensitive).
The HPX370's noise pattern is very isomorphic, like film grain; some older Panasonics tended towards horizontally streaky chroma noise signatures, almost like the noise seen on noisy analog tapes.
How can the HPX370 get such clean, sensitive images from a 1/3" sensor? It's that "3D adaptive processing technology" at work. Presumably, the third "D" is time: correlating the images in multiple frames to find and eliminate noise. Previous small-sensor Panasonics have used 3D, temporally-recursive noise reduction techniques to make better-than-they-should-be images.
The problem with recursive noise reduction of this sort is that it can lead to visible trailing (almost like old tube-camera artifacts) behind fast-moving subjects. This danger means that you can't use it as aggressively as you might like as noise increases; trail visibility climbs too quickly. This tradeoff may explain why the noise climbs faster on the HPX370 than it does on the EX1.
I have two clips out of all that I shot that may show evidence of this. They include rapid motion of contrasty, dense foliage in front of smooth and featureless backgrounds; in one case the sky, in the other an out-of-focus shadowed area. If I look very carefully, frame-by-frame—especially with some contrast boosting applied—I can see what looks like faint eddies of noise in the picture for a frame or two after the foliage has moved on, echoing where the leaves were in the previous frame or two. It's very faint, and it's not something that anyone would normally see (or "that anyone normal would see", grin). I only saw it because I was playing around with the images, pushing them around in post to see how well the AVC-Intra codec held up under extreme manipulation.
Now, AVC-Intra 100 (which is what I shot the clips with) is an intraframe-only codec, and Panasonic assures me that the entire thing, including prediction, is done on an intra-frame basis—so there should be no way the codec could cause any trailing compression artifacts in a scene.
I'm left to suspect that, unless Panasonic has managed to sneak three low-power, perfectly-registered, full-HD-capable 1/3" Newvicon tubes into the HPX370 [note: not too bloody likely!], 3D recursive noise reduction is being used, and that accounts for the slight trailing I saw, as well as the impressive low-noise performance of the 370's fast, yet tiny, 1080p sensors.
Panasonic indirectly confirmed this speculation when tech-reviewing the article; they said trailing of this sort had been mentioned on DVXuser, too. Panasonic is coming up with a firmware update offering two different P.A.P. modes, the current one and a new mode designed to reduce this sort of artifact. They may send me another 370 with the new firmware for testing; stay tuned...
Tonal Scale Reproduction
The HPX370 offers three manual knee settings as well as an auto knee. The auto knee is a bit laggy in its reaction and can sometimes lead to slight "pumping" in highlight values as overall levels change, such as when the aperture is suddenly opened or closed a couple of stops. The manual knees are fixed, so they don't exhibit any adaptive behavior; good for consistency, if less ideal for coping with unpredictably changing scenes.
Here's the really exciting bit: the knees appear to be gradual, "soft shoulder" limiters, more like Sony's cine gammas and hypergammas than a traditional, sharp-cornered, hard-onset knee. Bright, saturated gradients, like skies or overexposed skin tones, desaturate smoothly as they approach the clipping level, and crush smoothly into the 109% exposure ceiling.
Knees on lesser cameras cause such gradients to exhibit sudden hue shifts as they cross into the knee's levels of operation; skies typically go cyan while skins turn yellow, before both hard-clip into white. No so with the HPX370's knees. Previously, I had to use the Cine-like V gamma on affordable Panasonics to get clean skintone rolloff; Cine-like V applies a film-like S-curve to the tonal scale. With the HPX370's knees, I get the same, smooth highlight handling in all the gamma settings.
Here are three pix, shot off the camera's LCD screen, showing the HPX370's waveform display for a rather overexposed and unevenly-lit chart (the hue shift with changing exposure level is an artifact of the LCD and of my slapdash white-balancing of these stills; the camera was white-balanced on the chart and its pix were uncolored):
High knee, HD Norm gamma, +1.5 stops.
Mid knee, HD Norm gamma, +1.5 stops.
Low knee, HD Norm gamma, +1.5 stops.
The camera has several gamma curves predefined, allowing several looks. I've got two sequences of images following; the first is at a nominally correct exposure level for the HD Norm gamma, so you can see what happens to the midtones as the gamma setting is changed with the exposure held constant:
HD Norm gamma, normal exposure.
Low gamma, normal exposure.
SD Norm gamma, normal exposure.
High gamma, normal exposure.
Black Press gamma, normal exposure.
Cinelike D (data) gamma, normal exposure.
Cinelike V (video) gamma, normal exposure.
I then opened the aperture a stop, to see more information in the highlights and how it was affected by gamma changes. Again, pay more attention to the waveform itself than to the picture on the LCD beneath it, as the LCD clips its highlights:
HD Norm gamma, open 1 stop.
Low gamma, open 1 stop.
SD Norm gamma, open 1 stop.
High gamma, open 1 stop.
Black Press gamma, open 1 stop.
Cinelike D (data) gamma, open 1 stop.
Cinelike V (video) gamma, open 1 stop.
Observe how smoothly the highlights are handled in all the gammas. When you have a DSP with a 14-bit input and 20-bit internal processing doing your in-camera processing, you have the dynamic range and discrimination in your math to do this sort of smooth, clean rolloff.
At 0dB gain, with the knee set to "low", I was able to see ten discernable stops of dynamic range between the onset of clipping and lost-in-the-shadows blacks, and I never saw a highlight I didn't like. Sweet.
If you want to pull more out of your shots than the gammas and knees alone can provide, the camera offers three strengths of Dynamic Range Stretch. DRS works like "Auto Lighting Optimizer" in Canon still cameras, "Active D-Lighting" in Nikons, or the Shadows and Highlights controls in Photoshop, Aperture, and other photo-processing software; it performs localized level adjustments, lifting shadows and lowering highlights without flattening overall contrast.
Real-world examples of DRS at work: more detail held in clouds, shadowed areas.
Like the Highlights and Shadows controls, though, DRS can sometimes lead to localized haloing effects, especially when a sharp contrast is seen at the edge of an otherwise flat area.
DRS on and off. Notice slight "haloing" at contrasty edges.
I was only able to get noticeable haloing on my test chart; the subtlety of DRS's adjustments seems to be improved in the HPX370 over that used in the HPX170 and HMC40, without a loss in effectiveness.
The HPX370 captures the same, famously naturalistic colors that other Panasonics do. There are four different color matrices available in the camera's scene file settings; they look a bit more saturated in my LCD pix than they did did on the HD-SDI outputs, but the pix still offer a useful comparison:
Norm1 color matrix (Litepanel Micro 5600K LED lights).
Norm2 color matrix (Litepanel Micro 5600K LED lights).
Fluorescent color matrix (Litepanel Micro 5600K LED lights).
Cine-like color matrix (Litepanel Micro 5600K LED lights).
Flash Band Compensation
The 370 includes Flash Band Compensation in its firmware (it's in a firmware update available for the HPX300, too). FBC detects and compensates for "split frames" caused when a photo flash is shot with a "rolling shutter" CMOS sensor.
Four sequential frames without Flash Band Compensation.
In this sequence with FBC off, I panned the HPX-370 shooting 1080i, and fired off a Nikon SB-600 flash. The flash fired as the image scan was about 80% completed on the second field of frame #2, so it lit up the remainder of that field as well as the first 80% of frame #3's field 1.
Details of frames #2 and #3 without Flash Band Compensation.
There's a very slight overlap, about four scanlines tall, where both the outgoing and incoming fields are flash-illuminated; camera flashes are not instantaneous, and the overlap shows the duration of the flash's output.
FBC works by taking a frame sequence like this before it's recorded, and juggling portions of adjacent fields (or frames, in the progressive case) back and forth as needed to marry the flash-exposed regions together; a full-field (or frame) "flash frame" is much less distracting than a flash split across two adjacent fields or frames.
Four sequential frames with Flash Band Compensation on.
This time, I turned FBC on, panned the camera, and popped off a flash. Note the motion blur in frames 1, 3, and 4 that's absent in frame 2; something funny happened there... and frame #3 also has differing motion blur above and below the "splice point".
Details of frame #3 with Flash Band Compensation on.
What appears to be happening is that the HPX370 takes the bright halves of the two adjacent fields and puts them together, shifting one of the slices up or down a scanline to match the field order properly.
Close examination of frame #2, the preceding unflashed frame, shows it to be without interlacing artifacts; it looks like one field was interpolated from the other. About half the FBC-fixed sequences I shot show a de-interlaced frame #2, it only seems to happen when the first field in the FBC-fixed frame #3 is flashed. When the second field in frame 3 winds up flashed, frame 2 is unaffected.
I surmise that FBC is working entirely within a single frame when the flash is contained within the two fields of that frame, swapping and line-shifting the tops and bottoms of the two fields as needed.
When the flash spans two frames (the outgoing field 2 of frame #2 and the incoming field 1 of frame #3), I'm guessing the flashed field 2 of frame 2 is stolen to fill in field 1 of frame 3, and replaced with an interpolated copy of frame 2's field 1.
In their tech review, Panasonic neither confirmed nor denied this speculation, but they didn't complain about it.
It's a clever trick, if one that can indeed lead to minor motion and resolution irregularities, as Panasonic warns. In practice, though, the shock of a flashed frame is disconcerting enough to hide any minor stutters caused by field-stitching; I never saw hiccups in real-time playback.
FBC is a clever fix for one of the most distracting rolling-shutter artifacts. I only wished it worked in more modes; it's currently limited to interlaced recording modes and non-variable-frame-rate 720p, both with the shutter off.
I measured skew (the "jellocam effect") by panning the camera at a constant rate past a vertical line in each of its frame-size/frame-rate combinations, and comparing the distance traveled by the line between adjacent frames or fields (which happens at the frame or field rate) with the intra-frame or intra-field lateral displacement of the bottom of the line compared to the top (which gives a measure of the readout speed for the frame or field).
In 60i / 60p formats, the two figures were comparable; the line skewed within the frame or field over the same amount of distance that the line moves between frames or fields. In 30p the line skewed over only half the inter-frame distance; in 24p it took only about 40% of the inter-frame distance.
From these test I infer that the 370 reads out its sensors in about 1/60 second, regardless of frame rate or frame size. (I did not test the camera in 50Hz modes, alas, so I don't know if the camera maintains the 1/60 second read time or if it drops to 1/50 second with the change in the camera's overall timebase.)
This skew figure is the same as that of the Sony PMW-EX1. When I taped an EX1 on top of the 370 and panned them together, I got comparable skew on both cameras.
One thing I noticed while doing this test was that short shutter times really make skew noticeable: when everything is sharp, you see skew a lot more clearly. Even though the read time was the same in both cases, a slow pan at 1/120 sec shutter looks far worse than a similar slow pan at 1/24 second. This may help explain why so much footage from tiny consumer CMOS cams is so wobbly—consumer cams often shorten the shutter drastically as part of their autoexposure methods. It also suggest that if CMOS jello is a big concern, you'll want to use the slowest shutter speed you can get away with to reduce its visibility.
I didn't do a formal test of the audio on the camera, but the ambient sound recorded from the supplied shotgun mike was clean and noise-free. Panasonic's audio front ends have been very clean ever since the days of the DVX100, and the 16-bit, 48 kHz uncompressed recording available in all formats the camera records should yield perfectly good sound.
100 Mbit/sec AVC-Intra 100 captures full-raster images with 10-bit, 4:2:2 sampling. The format shows minimal compression artifacts while holding excellent detail; short of HDCAM-SR it's one of the cleanest, most transparent HD field codecs around. AVC-Intra 50 uses DVCPROHD-style spatial subsampling—1440x1080 and 960x720 pixels—but retains 4:2:2 chroma and 10-bit depth. At half the data rate of AVC-Intra 100 it showed more artifacts (see the mosquito noise around the text in the resolution chart pix above, for example); it seems to be roughly equal to DVCPROHD in overall fidelity.
DV, DVCRO50, and DVCPROHD recordings held no surprises; Panasonic has these nailed. They may be a bit old-school, but DV is instantly and efficiently editable on decade-old computers; DVCPRO50 is a superb 8-bit 4:2:2 SD codec; and the DVCPROHD formats continue to be widely supported, especially on computers overstressed by the more computationally complex AVC-Intra formats.
The Anton-Bauer Tandem is a clever bit of work; it can stand alone as a battery charger or mount up on the camera as an AC adapter using a battery for backup.
Anton-Bauer "Tandem" AC adapter/charger and Dionic90 battery on HPX370.
In my testing it worked very well, though on some occasions the camera didn't want to power up consistently with a battery on the Tandem and the Tandem plugged into AC. Panasonic tells me that Anton-Bauer is aware of this and has "an upgrade fix" in the works, although they had no more information about it. When it did work (which was most of the time), I found I could yank out the AC cable and the camera would run, undisturbed, on the battery.
The HPX370 will genlock to an analog HD Y signal or an SD composite signal in HD modes; in SD, an SD composite signal must be used. Panasonic warns that the subcarrier of the composite output doesn't lock to the genlock input, so you can't use this camera as a feed to an analog composite production switcher without a TBC on the input.
Next: Conclusions; Pros, Cons, and Cautions; more info...
Get articles like this in your inbox: Sign Up