The Real CODEC Torture Test: 35Mbps vs 100Mbps
 

Last week I posted some of the CODEC torture test material. The CODEC torture consisted of a fast horizontal pan, which would cause a considerable change in each frame and, in theory, break the Long-GOP CODEC. While you could definitely see a difference in the 35Mbps 4:2:0 (EX1 native) and the 100 Mbps 4:2:2 (Flash XDR) footage, the Long-GOP held you surprisingly well at either bit-rate.

Then Barry Green suggested that the real Long-GOP torture test was to rotate the camera while shooting a complex scene. This is a much more difficult video for MPEG2 Long-GOP as the temporal compression (P, B frames) only looks for motion vectors along the X and Y axis. If an object rotates, then the algorithm falls apart, in theory.

Well, Jim Arthurs was too tempted. So, he took his EX1 and the Flash XDR and shot some outdoor footage while rotating his camera. The results (left image) were astounding (IMO), the 35Mbps 4:2:0 image show heavy artifacts, while the 100 Mbps 4:2:2 looks very good.

Important note: this image is not indicative of the normal quality from a Sony EX1, remember this is an extreme test designed to break the CODEC.

Jim also shot some footage from his grinder, as further evidence that the Long-GOP format holds up very well, even when the video changes from frame to frame (another way the I-Frame experts claim that Long-GOP fails). The 35 Mbps 4:2:0 and 100 Mbps 4:2:2 comparison image is shown below.

Zoom in on each image and you'll see the clear advantage of the 100Mbps 4:2:2 CODEC, which improves the quality in three ways:
1) Improved Color Space: 4:2:2 vs 4:2:0
2) 3X Higher bit rate: 100 vs 35
3) Two CODEC chips are used to compress the video (EX1 uses only one CODEC).

In conclusion, I don't see any advantage using an I-Frame only CODEC when you can get superb quality video under very extreme conditions using the Long-GOP 100Mbps 4:2:2 CODEC in Flash XDR / nanoFlash. Furthermore, you can playback the video directly off the CF cards w/o rewrap or transcode (in FCP). Finally, the low-cost Transcend 32GB CF cards will support the 100Mbps rate. (Transcend 32GB CF cards are currently 1/20th the price of equivalent P2 or SxS cards).

Special thanks to Jim Arthurs for these great shots!

All I can say is wow, that is quite a difference.

On the house photos, not only is the image broken up in the 35Mbps example, the almost blur of the green siding detail compared to the 100Mbps sample is quite noticeable.

Thanks for posting.

Any talk of trying the HD-SDI out on the HPX170 and testing it with your codec, Mike?

The stills are amazing. You guys have a sure winner on your hands!

mc

Hi Michael-
I have a HPX170 user ready to test as soon as I can break free a demo XDR to send him. I too am quite interested in the results.

However, I will have resolution chart image comparison shortly using an uncompressed capture from an EX1 camera. We will compress the image using our 100 Mbps CODEC and compare to DVCProHD and HDV. It should be quite telling.

I agree. Breaking the 4:2:0 barrier is what is really going to make the EX shine IMO - There are still issues with CMOS but it's a tradeoff against CCD. Can't wait to see further tests.

Jim Arthurs test ...
 

I downloaded Jim's test and it proved the Flash XDR 100 mg speed to be clear of artifacts. Good example of the strength of the codec.

Would the BBC eventually declare this acceptable for broadcast?

Hi Jack-
We hope to find out in the very near future. I strongly suspect it will be more than acceptable, as the 100 Mbps CODEC is above HDCAM in quality.

Funny you should ask, as I've just read an article in TVB Europe about this very issue, you can read it here: TVB Europe : October 2008

The article in question starts on page 34, the relevant paragraph ("Making The Cut") is on page 37. Phil Myers is the product manager for XDCAM in Europe, so that's a pretty big endorsement in my book. (Also, on page 13 of this issue, the XDR is given a Best of IBC award)

I have previously looked into the BBC's guidelines for HD programming, and for footage to be considered HD it needs to meet the following criteria:

-be acquired at a resolution of 1920x1080@25 frames/sec, interlaced or progressive (though 1440x1080 prefiltering is *currently* acceptable)
-720p footage is only accepted from the Panasonic AJ-HDC27 or AJ-HDX900 (others may be accepted *by request*)
-the CCD chip on the camera needs to be 1/2" or bigger
-the bitrate for intra-frame compression codecs (e.g. MPEG2 I-Frame only, AVC-Intra, DVCProHD, JPEG2K) needs to be 100Mbps or higher
-the bitrate for inter-frame compression codecs (e.g. MPEG2 long-GOP, AVC HD, H.264) needs to be 50Mbps or higher

Adhering to these rules by the letter excludes a lot of cameras: Canon's XL and XH range all use 1/3" chips, as do many of Panasonic's models.

Sony's EX1 and EX3 both have 1/2" chips but fall at the bitrate hurdle, only supporting the XDCAM codec up to 35Mbps.
However, use an XDR to record from the HD-SDI out and suddenly you've got full-raster 100Mbps intra-frame codec video from a camera with a 1/2" CCD - well within the BBC's guidelines!

Dear Paul,

Thank you for posting that information.

Unreal.
I'm sold on using 100 Mbps 4:2:2 Long-GOP.

Thanks Paul great find. Wow TVB is loaded with up to date info.

[QUOTE=... you've got full-raster 100Mbps intra-frame codec video from a camera with a 1/2" CCD - well within the BBC's guidelines![/QUOTE]

The EX1 has 1/2" CMOS, not CCD, I wonder if that changes things or not? I wouldn't think so.

Hi Buck-
I seriously doubt that would be objectionable to the BBC. CMOS and CCD sensors both have their limitations and "issues". You can get great results from either sensor if you stay within the limits.

Very true.
I know some who have invested in over $40K on the RED ONE (another CMOS sensor based camera) with accessories.
This certainly has not stopped them from producing great results.