Back To Listings RSS Print

Twice as Nice?

Half truths about MPEG-2 vs. AVC-I compression

By The Sony Tech Guy | August 17, 2010

image

Good GOP, good GOP. This Group of Pictures includes 15 frames, starting with a fully-defined Intra frame (I), Predictive frames (P) and Bi-directional predictive frames (B).

Proponents of AVC-I compression have been quick to seize on a simple, and we believe simplistic, claim: "Twice the compression efficiency of MPEG-2." It's a bold promise that has a nice ring to it and has been repeated ad infinitum. The only problem: it just isn't true.

As long as there has been bit-rate reduction in video recording, there have been compression food fights. Back in 1993, Sony launched our first compressed video format. The Digital Betacam® system featured a robust data rate of 90 Megabits per second (for standard definition, mind you) and mild compression, about 2:1. Nevertheless, compression was immediately attacked as a depravity and an abomination. Of course, Digital Betacam products went on to enjoy widespread success at the high end of video production. Oddly enough, the same people who attacked compression soon embraced it across the board.


The rise of MPEG-2


Fast forward to the present day and compression is almost ubiquitous in the broadcast and video production universe. Prominent among the compression codecs is MPEG-2 Long GOP, the product of the international Moving Picture Experts Group formed by the ISO and IEC. MPEG-2 Long GOP is the technical foundation of many professional recording formats including the HDVâ„¢, XDCAM®, XDCAM HD, XDCAM EXâ„¢ and XDCAM HD MPEG422 systems.

MPEG-2 Long GOP is so widespread because it combines the efficiency of two types of compression. Its intraframe compression squeezes out the redundancy that exists within individual pictures. But wait, there's more. The system's interframe compression squeezes out the redundancy that occurs when one frame of video in a Group of Pictures (GOP) is almost identical to the others. The success of MPEG-2 Long GOP is evident in thousands of productions from the biggest reality TV hits to newsmagazines and documentaries, from sports to education to indie film production.

In the face of this success, the food fights continue unabated with extravagant claims for AVC-I.


How AVC compares


It is true that AVC is based on MPEG-4, itself an evolution of MPEG-2. It's true that AVC is a respected family of compression standards, embraced by Sony for NXCAMâ„¢ and consumer AVCHDâ„¢ products. What's also true is that by using the entire set of compression tools as stipulated for the generic form of AVC-AVC Long GOP-AVC really does have twice the compression efficiency of MPEG-2 Long GOP. The kicker is that we're talking about AVC Long GOP, but the competitive claim is being put forward for AVC-I.

AVC Long GOP takes advantage of a powerful, advanced toolkit of interframe and intraframe compression techniques. It excels at modest bitrates by taking out massive interframe redundancies. In fact, the advantages in compression efficiencies offered by AVC are primarily based on its better exploitation of these frame-to-frame redundancies. Take away the genius-level interframe processing and you're left with something less impressive: a scheme with just 15-20% more efficiency than other intra-only techniques such as DV, JPEG and MPEG-2 Intra. This slight edge of AVC-I is overwhelmed by the interframe efficiency of MPEG-2 Long GOP. In the end, MPEG-2 Long GOP can produce significantly higher picture quality than AVC-I at a similar data rate.

For authoritative information on how the codecs really compare, and what kind of thinking goes into selecting an appropriate codec, I encourage you to read "Codec Technology for XDCAM Tapeless Products and Systems," a paper by Hugo Gaggioni, Chief Technology Officer at Sony. Tell Hugo that I sent you.

Compression fun facts

The next time you're at a cocktail party and meet an attractive person, try not to use any of the following as conversational gambits.


  • A compression encoder must follow the syntax that the decoder will understand. But that still leaves considerable latitude for creativity and original engineering on the encoder side. The Sony MPEG-2 encoders of today are emphatically not the same as the encoders of yesteryear, even when the decoders are identical.
  • What happens when a scene changes in the middle of a GOP? Sony encoders immediately recognize the change, insert an I-frame and they preserve the integrity of your imagery.
  • MPEG-2 is everywhere you look, featured in hundreds of millions of devices. You'll find MPEG-2 enshrined in the ATSC high definition broadcast standard, cable TV, satellite TV, DVD as well as a compression option in the Blu-ray Discâ„¢ standard.
  • MPEG-4 encompasses quite disparate compression schemes including tools for low bit rates intended for web video, for moderate bitrates including the AVCHD format and remarkable tools for the high quality required by the MPEG-4 Simple Studio Profile (SStP) as used in the HDCAM SR system. SStP has been extended to support resolutions to 4K and data rates to 3.6 Gigabits per second.
  • While compression can be performed in software, portable products often benefit from executing compression in hardware. Sony has devoted huge resources to developing dedicated compression silicon including the TORINO chip for MPEG-2 Long GOP and the new CURIE chip for MPEG-4 Simple Studio Profile.


Under the hood

For a really close inspection of what makes video compression tick from a guy who knows whereof he speaks, take a look at Hugo Gaggioni's tutorials, on Page 2. They're the last two parts of a larger series, Basics of Broadcast Technology. The complete series is featured at sony.com/videon.

Page 1 of 2 pages 1 2 Next »

Share This

Back To Listings RSS Print

Get articles like this in your inbox: Sign Up

Disclosure, to comply with the FTC’s rules 16 CFR Part 255 This article was either written by Sony employees or for Sony by an outside contractor. It is intended for the Sony Channel on ProVideo Coalition, which Sony sponsors.

Comments

mactrix: | August, 18, 2010

This article is about compression quality, not workflow. Regarding compression the author might be not too wrong but since he does not cover workflow and impact on file structures it’s not worth to take it serious.

Long-GOP in editing workflows causes many issues, this is not mentioned here. Did you know that cut only of Long-GOP causes re-compression?!

Casey: | August, 18, 2010

Who would ever make the argument that AVC-Intra is more efficient than Mpeg-2?  That’s a ridiculous argument to make, and frankly i fear for any client that this person may work for.  For crying out loud, it’s an INTRAFRAME codec vs an INTERFRAME codec.  OF COURSE it’s less efficient!

Yes, you are right about the difference in quality and efficiency, but the argument and response put forth is absurd.  Who would ask such a question?  Are you sure they didn’t slip and mean to say AVC-HD? Was it someone trying to sound smart by using big words and mixing up terms?  Honestly, why not just make the statement, “HDCAM is less efficient than HDV at similar resolutions.”

If he’s a Sony tech, then he’s just trying to troll up controversy, and poorly at that.

Burn-E: | August, 19, 2010

“many professional recording formats including the HDV™, XDCAM

Charles Angus: | August, 19, 2010

I have never heard anyone argue that AVC-Intra compresses more than MPEG-2.

It is much higher quality, though, and an I-frame only codec. In my opinion, with today’s technology, only I-frame only codecs are suitable for serious production. I just can’t tolerate not knowing exactly how trees, grass, water, pans, etc. are going to compress - which is a problem with modern Long-GOP codecs (yes, even XDCAM).

At the end of the day, there’s no Long-GOP codec I’ve seen that compares to AVC-I at 100Mb/s in quality and reliability. Period.

The Sony Tech Guy: | August, 30, 2010

Mactrix:

You are correct; the article treats compression quality only.  Your comment on editing is also correct, as far as it goes.  However, ALL compression codecs require a cycle of decompression and recompression if you’re doing ANY kind of editing other than cuts-only.  And in those cases when you are doing cuts-only, modern NLEs only need to decompress a few frames around the editing points - the rest of the file remains intact.

The Sony Tech Guy: | August, 30, 2010

Casey:

I don’t blame you for thinking The Sony Tech Guy ginned up a bogus argument for his own amusement.  I only wish that were the case.  I assure you that the bogus argument is, in fact, a word-for-word quote from promotional material in the marketplace.

The Sony Tech Guy: | August, 30, 2010

Burn-E:

Thank you for taking the time to read so carefully and comment in such detail.  Permit me a few clarifications.

You are certainly right that Sony has many formats and many format names.

You are right again that the Europeans have found that MPEG-2 Long GOP at 50 Mbits/s is about equal to AVC-I at 100 Mbits/s.  This is true even though this version of AVC-I has 10 bit depth.  It sounds as though Auntie Beeb is confirming my main point.  While others are saying that AVC-I offers “twice the compression efficiency of MPEG-2,” it just ain’t so.

You are also correct in pointing out that at modest bit rates, AVC Long GOP is more efficient than MPEG-2 Long GOP.  This is exactly why Sony designed the NX5 around the AVCHD format. 

You’re certainly entitled to your opinion of Canon hardware codecs. 

I concede that MPEG being “everywhere you look” is overstated.  And force-fitting any codec at a given spatial resolution into the wrong data rate is never a good idea.

While AVC is newer, we do part company on the blanket statement that AVC is better, period.  In real-world implementations, there are always important questions about which flavor of AVC we’re talking about, what application, what resolution and what data rate.  As always, the devil is in the details.  I’ve tried to address just some of those questions in my original article.

Thanks for keeping me on my toes.

The Sony Tech Guy: | August, 30, 2010

Charles:

The point is not exactly that AVC-I “compresses more,” it’s that AVC-I compresses less efficiently.  And while you’re welcome to your opinion, as Burn-E commented above, independent authorities have conducted extensive tests whose results don’t quite agree with yours.  Here, for your reading pleasure, is one such authority:

http://www.ebu.ch/en/technical/trev/trev_2008-Q3_HD-Prod-Codecs.pdf

Burn-E: | August, 30, 2010

“You are also correct in pointing out that at modest bit rates, AVC Long GOP is more efficient than MPEG-2 Long GOP.  This is exactly why Sony designed the NX5 around the AVCHD format.”—The only question is: why did Sony design the NX5 for modest bit rates? The camera was released in 2010, not in 2003. After Sony had allowed usage of SD cards, the gates opened. 32GB cards are commonplace now, and 16GB cards are dirt cheap and reasonably fast. Most Class 6 cards have no problems with 35 Mbit/s, and Class 10 cards would have no problems with 50 Mbit/s. Look at JVC. They licensed XDCAM EX container from Sony and did the right thing, offering a recording format that is well supported by the industry, on cheap media. This is what I expected from Sony. Instead, Sony switched to AVCHD, the format that was suitable for consumer sector only, or so Sony claimed couple of years ago.

“While AVC is newer, we do part company on the blanket statement that AVC is better, period.”—It is a better format, period. Of course, bad implementation can ruin even the best format, but implementation aside, AVC—as a format, as a pure spec—has more features to fine-tune image and make it look better. It requires a lot of horsepower, which is why MPEG-2 is still a good enough solution, at least until everyone has an 8-core computer with 4GHz clock, and properly updated software.

And from the late news: Canon is readying the XF100/XF105 series. Think about the lens and LCD screen from the consumer HF-S series bundled with… here is the killer: two CF slots and MPEG-2 422 50 Mbit/s recording. That is right: high-end HD422 in a baby cam. With proper pricing this camera will kill the HMC40, the HM100 and the NXCAM too. It was Sony who was supposed to push MPEG-2, not JVC or Canon.

Sorry for the rant. I am going to start saving for the XF100 now. And here is the link, you can use Google to translate it into English, though numbers already say enough:

http://www.mynewsdesk.com/se/pressroom/canon/pressrelease/view/flexibilitet-maangsidighet-och-hoega-prestanda-canon-lanserar-de-nya-proffsvideokamerorna-xf105-och-xf100-460407

Please login or register to comment