Slashdot Mirror
100GbE To Slash the Cost of Producing Live Television
New submitter danversj writes "I'm a Television Outside Broadcast Engineer who wants to use more IT and Computer Science-based approaches to make my job easier. Today, live-produced TV is still largely a circuit-switched system. But technologies such as 100 Gigabit Ethernet and Audio Video Bridging hold the promise of removing kilometres of cable and thousands of connectors from a typical broadcast TV installation. 100GbE is still horrendously expensive today — but broadcast TV gear has always been horrendously expensive. 100GbE only needs to come down in price just a bit — i.e. by following the same price curve as for 10GbE or 1GbE — before it becomes the cheaper way to distribute multiple uncompressed 1080p signals around a television facility. This paper was written for and presented at the SMPTE Australia conference in 2011. It was subsequently published in Content and Technology magazine in February 2012. C&T uses issuu.com to publish online so the paper has been re-published on my company's website to make it more technically accessible (not Flash-based)."
100GbE is huge demand for core infrastructure people due to backbones being strained everywhere by the explosion of online video usage. Tier 1 providers are simply at a demand level that current foundries can't even come close to providing. Thus no one has an incentive to slash prices.
As well, not everybody viewing HD footage has a shitty provider, and giving providers the excuse "it comes that way" won't help anybody.
The intent here is to replace so much of the specialized cabling for lighting controls, audio, video, camera control systems, etc. with a single, multi-purpose system that can handle uncompressed data, thereby supporting existing models of data acquisition. Each level of re-compression and transcoding results in a loss of quality.
Thirty four characters live here.
Well that's a failure of imagination. I'll admit technically speaking it often is *somewhat* compressed, - eg. 422 Subsampled chroma at least. But there is a massive difference between a delivery codec and a signal you're still working with. To start with H264 and their ilk are computationally expensive to do anything with. A single frame of 1080p is a pretty big dataset, and it's painful enough doing basic matrix transforms, but adding a bunch of higher level computations on top of that?... For example just cutting between two feeds of an inter frame compressed codec requires that the processor decompress the GOP and recreate the missing frames. Several of orders of magnitude more complicated than stopping one feed and starting another. And generally speaking the uncompressed feed you have in broadcast situation you're doing *something* oo. Switching, mixing, adding graphics, etc. But the biggest question is one of generation loss. Even one round trip through one of those codecs results in a massive drop in quality (as you rightly point out). You don't want to be compressing footage out of the cameras any more than you can, because you KNOW that you're going to be rescaling, retiming, wiping, fading, keying etc etc etc...
-Steve http://www.stevennicholson.com
You don't see that all the time on slashdot.
Great article.
I think many are getting confused here and think that this article is about reducing the cost of producing live TV on a shoestring. The figures in this article are very high, but for professional video production, existing figures are also very high.
If you take into account that this could allow production trucks to shrink in size a bit (RG6 takes up a lot of space), the price of this new way could be even lower.
http://lkml.org/lkml/2005/8/20/95
Replacement tech rarely catches up. 1080p signal? Please, that is so last year. 4k is the new norm. No TV's for it yet? Actually, they are already on sale which means that if you are not recording your repeatable content right now in 4k, you will have a hard time selling it again in the future. That is why some smart people recorded TV shows they hoped to sell again and again on film and not video-tape. Because film has a "wasted" resolution in the days of VHS video tapes but when DVD and now Blu-ray came out, these shows can simply be re-scanned from the original footage and voila, something new to flog to the punters.
I don't know how much data a 100GbE link can truly handle but the fact is that trying to catch up to currect tech means by the time you are finished, you are obsolete. the 4k standard created by the Japanese (and gosh doesn't that say a lot about the state of the west) isn't just about putting more pixels on a screen it is about all the infrastructure needed to create such content. And you better be ready for it now because if you are not, you will be left behind by everyone else.
The future may not be now, but it sure needs to have been planned for yesterday.
MMO Quests are like orgasms:
You may solo them, I prefer them in a group.
You're just jealous because Australia is a significant source of crappy stories, and some of them are extremely low quality.
Our crappy stories per capita ratio is truly astounding.
hmmm. i should write an article about this. I'm sure I can get it published.
The latency problem i can understand, but that will be a problem regardless of compression or not.
The trouble is that the more effective codecs tend to require an entire frame before they can do any compression (so that they can compress more effectively by taking the whole frame into consideration). So if you have a series of pieces of equipment processing the video (camera, distribution, control desk(s), effects etc), then each one has to wait until it's received the last lines of a frame before it can even start sending out the first lines of that frame - so each element in the chain adds a whole frame's worth of latency. Whereas if you do it uncompressed, most equipment can start sending out the first line of a frame before it's even received the second line.
Encoding and decoding will not add that much cost compared to the network.
That's dependent on a lot of factors. 100Gbps Ethernet has the potential to reach much bigger economies of scale than broadcast-quality codec hardware (though it has a long way to go before reaching that far as yet).
Compressing/uncompressing only destroys the pic if its lossy. There are numerous lossless codecs that should do the trick and save tons of money in the process.
The trouble with lossless codecs is that they can never guarantee to make a frame smaller - mathematically there must be some frames that are uncompressible. Over the course of a long video, the codec will win on average, but when working with live streams, if you get just one frame that doesn't compress nicely (or worse, a few in succession) then your network has to be able to handle that bandwidth - so you might as well not use the compression in the first place.
Need to type accents and special characters in Windows? Use FrKeys
this is about live TV. Live TV is a different. The infrastructure relies on point-to-point circuit switching. One video signal is sent down one coax cable. 8 cameras is 8 coax cables, now have 1km of cable that's 8km just for the live camera feeds to the OB truck. 100GbE means one cable. 8km of coax or fibreoptic isn't cheap, and usually requires a truck and a team of sparks to transport all these cables.
Back to caferace's conversation. It is a bottleneck indeed for content that is not live. Digitising rushes to intermediate codecs takes time, tape is usually played back at normal speed or double speed, output via HD-SDI from the deck and the workstation that transcodes on the fly in realtime. Tapeless workflows speeds the process up as you can import faster that 1-2x but still takes time to transcode. However, having this slow down is not a problem, the rushes have to be logged, while they are converting this logging process can be done manually.
Cinematic filming have the workflow sorted to some extent. High end cameras shoot direct to an intermediate codec, a DIT works on set and logs as the footage is shot, and sound and continuity departments can log electronically to the same system now too. The problem at the moment is it is not one system but many have to come down to one. I work in the sound department in film as an assistant. One of my responsibilities is keeping time-code correct on set, I have to go round each department times a day* and "jam" each system, recorder, slate, camera etc so they are correctly in time so when all the data is put together by the DIT. One day they will get unified
Logging while shooting cannot be done for news or reality TV as everything happens too quick.
* Three times a day because Sony can't make a $100,000 camera that doesn't have an internal clock that doesn't drift by +/- 2-3 frames a day.
Insightful write up. Getting rare here on ./
For those not RTFA, they are referring to using Ethernet in professional live broadcast situations. Aka, newsroom or outdoor sporting broadcasts where cable bundles are still common. I believe they are imagining a world where a broadcast truck rolls up to a stadium and runs a few pair of 100Gbe fiber vs a large coax bundle. This could save considerable time and money. Some interesting bw numbers:
SD 270 Mbit/s
Interlaced HD 1485 Mbit/s
Progressive HD 2970 Mbit/s
> The intent here is to replace so much of the specialized cabling
Yup. I'm glad I work in radio, where we've been ferrying oversampled, high-quality audio over IP for some years now.
The digital switching and input assignments are a dream as well. Not that many years ago, if someone came into Engineering and said, "sorry, forgot! We have a paid ballgame going on at 4PM!" ... my assistants and I would literally grab a punch tool and some Belden wire and start frantically running cables. Many was the time we'd put something on air by literally throwing a pair across the floor with gaffer's tape. "Watch Yer Step!" :)
Nowadays, any source in our facility can be assigned to any input on any mixer in any control room. Run once, use many times. Ah, it's a beautiful thing. I can move an entire radio station from one control to another literally in a matter of minutes. It takes longer for the staff to physically grab their coffee cups and lucky charms than it does for my staff to move the signals.
My poor brethren in TV just have entirely too much data. If we'd all go back to RADIO drama, see, this wouldn't be a problem, now woodit? :D
Cogito, igitur comedam pizza.
In the last studio upgrade we did, we retrofitted everything with Ethernet -- 10G switches. Cameras are all ASI -> GigE (MPEG-2 Multicast), switchers, and final outs.
Uncompressed, at full rate, an ASI feed uses 380 MB/s. An uncompressed 1080p melted feed is 38 MB/s.
You need to do careful network planning, but remember these are switches -- you shouldn't see traffic you didn't request. Right now we usually have about 8 cameras, plus the mixer, plus the groomer, plus the ad-insert. It then goes right out via the internet (Internet2 -- FSN is also a partner so we can send right to them), and a satellite truck as a backup. Our plan next year is not to have the satellite tuck on site anymore.
This is for a live-sports studio that feeds about 300 cable / satellite providers, reaching about 73M homes.
Network Architect here, who's worked on many varied systems. I predict what the consumer will see is a drop in reliability.
Real time communication is just that, real time. Gear of old (5ESS switches, TDM networks, Coax analog video switchers) were actually built around this notation from the ground up, and many design decisions were made to keep things operating at all costs. Of course, this added cost and complexity.
Packet based networks were built on the assumption that losing data was a-ok. Packet drops are how problems are signaled. Protocols are just barely in some cases starting to figure out how to properly deal with this for real time situations, and largely the approach is to still throw bandwidth at the problem.
So yes, running one 100Gbe cable will be cheaper in the future, but it's going to introduce a host of new failure modes that, no offense, you probably don't understand. Heck, most "Network Architects" sadly don't understand, not knowing enough about the outgoing or incoming technology. However I've seen the studies, and it's not pretty. VoIP is not as reliable as circuit switched voice, but it's pretty darn close as it's got more mature codecs and low bandwidth. iSCSI is laughably unreliable compared to even fiber channel connections, much less some kind of direct connection methodology. The failure mode is also horrible, a minor network blip can corrupt file systems and lock up systems so they need a reboot. Of course, it's also a straight up redundancy thing; when you're covering the Super Bowl having every camera feed leave the building on a single cable sounds like a great cost and time reducer, until it fails, or someone cuts it, or whatever, and you lose 100% of the feeds, not just one or two.
With the old tech the engineering happened in a lab, with qualified people studying the solution in detail, and with reliability as a prime concern for most real time applications. With the new tech, folks are taking an IP switch and IP protocol, both of which were designed to lose data as a signally mechanism and who's #1, #2, and #3 design goals were cheap, cheap, and cheap and then multiplexing on many streams to further reduce costs. The engineering, if any, is in the hands of the person assembling the end system which is often some moderately qualified vendor engineer who's going to walk away from it at the end. It's no wonder when they fail it's in spectacular fashion.
I'm not saying you can't move live TV over 100Gbe (and why not over 10Gbe, even 10x10Gbe is cheaper than 100Gbe right now), but if I owned a TV station and my revenue depended on it, I don't think that's the direction I would be going...
HDSDI uncompressed video is 1.5Gb/s. That is the standard for moving uncompressed video around inside a TV truck, whether 720p or 1080i. It rises to 3Gb/s if you're doing multiple phases of video (3D video, super slo-mo, etc). Within that 1.5Gb/s is still more than enough headroom to embed multiple datastreams and channels of audio (8 stereo pairs is the norm, some streams do up to 16). So I fail to see why 100Gb/s is necessary to transmit uncompressed video.
It's also a chicken-and-egg scenario. I'm a broadcast engineer and audio specialist. I had Ma Bell contact me about 7 years ago asking about how important uncompressed video transmission was, as they were trying to gauge a timeframe for a network rebuild to allow for uncompressed video transmission. My answer hasn't changed much in 7 years, because although moving uncompressed video from site to (in the case of Fox) Houston and then back to your local affiliate would be nice, it's completely unnecessary because by the time it reaches your house your local cable or satellite operator has compressed your 1.5Gb/s signal down to between 4Mb/s and 10Mb/s typically, making the quality gains negligible.
It will solve one problem, which is image degradation due to multiple passes of compression. Think about it... the 1.5Gb/s leaves our TV truck and gets ASI compressed into 270Mb/s (best case scenario, satellite transmission is significantly lower bandwidth, and most networks don't use an entire 270M circuit, they use less). It then arrives at the network hub, where it gets decompressed. If it's live it then goes through several switchers and graphics boxes, then gets re-compressed to ASI and sent either to another hub or to your local affiliate. (If not live, it gets put into a server which re-compresses the video even harder before playout.) Your local affiliate then decompresses it, it passes through more switchers and graphics boxes, then it gets either broadcast using 8VSB, or it gets re-compressed and passed on to your cable or satellite provider, who then un-compresses it, processes it into MPEG or some other flavor, and re-compresses it into its final 3-12Mb/s data stream for your receiver to decompress one final time.
This would eliminate several compression steps, and mean a better final image quality because you're not recompressing compression artifacts over and over and over again. A real 1.5Gb/s video frame looks like staring out a window compared to the nastiness you see when you hit pause on your DVR during a football game (also a best-case scenario, most cable/broadcast/sat providers ramp up the bitrate to the max for live sports and then set it back down shortly thereafter).
But the 100Gb/s makes no sense to me. Are you (crazy) overcompensating for latency? Are you sending 100% redundant data for error correction? Why in the world would you need that much overhead? I can't imagine it's to send multiple video feeds, the telco companies don't want you to do that because then you order less circuits from them. Plus you'd want at least two circuits anyways in case your primary circuit goes down for some reason.
(Side note: The one benefit to a TV truck using Ethernet as a transmission medium is the fact that these circuits are bi-directional. Transmission circuits nowadays are all unidirectional, meaning you need to order more circuits if you need a return video feed, meaning higher transmission costs. The ability to send return video or even confidence return signals back down the same line would be huge for us and a big money saver.)