ESPN Primes Its World Cup Pipe

Fiber network will deliver 2D, 3D coverage to U.S.

With its multi-platform coverage of the 2010
FIFA World Cup starting June 11, and the coincident launch of its new 3D
network, sports giant ESPN says its robust global fiber network is ready to
handle the reams of traffic from South Africa.

ESPN's private network will
link temporary studios in South Africa with its regular facilities in Bristol,
Conn.; Los Angeles; New York; London; Argentina and Brazil, via an OC-12 (622
megabit-per-second) pipe. The fiber links, fed by Net Insight terminal
equipment, will support coverage on its linear networks, its Web and mobile
offerings, and ESPN 3D, which also launches June 11 on DirecTV and Comcast with
a live broadcast of the match between South Africa and Mexico.

ESPN had initially planned
to use Ericsson (formerly Tandberg) MPEG-2 encoders to backhaul signals from
South Africa. But after testing Ericsson's new CExH42 MPEG-4 encoder a couple
of months ago, it decided to make the move to MPEG-4.

"We want to add more
services and improve the production between the U.S. and South Africa," says
Emory Strilkauskas, ESPN lead engineer for transport and special projects. "We
decided to make the investment after trialing with Ericsson."

While MPEG-4 is often touted by encoder manufacturers as yielding a 50%
effective reduction in bandwidth over MPEG-2, "that's not what we typically
see," adds Bill Lamb, VP of transmission and transport for ESPN. "A 25% bitrate
reduction is a nice, conservative number," Lamb says. "We're able to do quality
feeds at 30 megabits per second, and we'll increase it to 40 once we get out
there and fine-tune the pipe. When we get back to the States, we'll definitely
use higher rates."

Format-conversion technology will be crucial for handling the feeds. The
World Cup matches are being produced by host broadcaster HBS in the European
1080-line-interlace/50 hertz format, which needs to be converted to ESPN's
720-line progressive/60 frames-per-second format to be shown in the U.S. on

ESPN is using about 20 Snell
Alchemist Ph.C-HD systems for both 2D and 3D standards conversion for its Cup
coverage, with some on-site in South Africa and others in Bristol. The network
has also purchased seven FOR-A FRC-8000 frame-rate converters for its
transmission facilities in Bristol and Brazil.

What is unique about the 3D feeds is that ESPN won't be converting them
from interlace to progressive for U.S. distribution. ESPN, like ABC and other
Disney networks, has long been a believer in the quality and bandwidth
efficiency of progressive over interlace, which is why it produces and
transmits its programming in 720p/60. But it will deliver the 3D World Cup
coverage to DirecTV and Comcast in the 1080i/60 HD format, which will then be
transmitted to consumers in the side-by-side "frame-compatible" 3D format
supported by both new 3D TVs and existing set-top boxes.

Converting the 3D signals from 50Hz to 60 Hz was complicated enough,
Lamb explains, and the network didn't want to risk a loss of picture quality.
"Technically, we can change it quite fluidly from the 1080i/50 original," Lamb
adds. "But we didn't want to harm the signal any more than we have to, and we
want to preserve very high quality."

While ESPN plans
to produce its 3D events going forward in 720p/60, it may continue to deliver
ESPN 3D in 1080i/60 to certain operators, depending on their requirements. ESPN
is delivering ESPN 3D directly over fiber to DirecTV and Comcast, so there
isn't a standard satellite feed of the network being broadcast today.

"To call this a work in progress is to underestimate the amount of
movement here," Lamb says. "We're trying to stay flexible."

HBS will produce the 3D coverage and deliver separate left and right
eye signals to ESPN at the International Broadcast Center in Johannesburg using
JPEG compression at a rate of 300 Mbps. According to Lamb and Strilkauskas,
it's one of the first events to do so, as most 3D productions to date have
created a frame-compatible 3D signal right in the production truck.

ESPN will decode the left-and-right signals to baseband (uncompressed
HD at 1.5 Gbps), then re-encode them in MPEG-4, at an average bit-rate of 20
Mbps each. The network will also be creating a "side-by-side", frame-compatible
3D feed in Johannesburg, as it has done with its previous 3D productions. It
will multiplex that side-by-side feed, which will be encoded at 40 Mbps, with
the left-and-right eye signals, then send the three streams back to the U.S. at
a total payload of 80 Mbps.

"Throughout the event, we're going to test them, and we'll go full left
and right once we're comfortable with it," says Strilkauskas.

In the long run ESPN believes it can deliver better quality by keeping
the left- and right-eye signals separate for as long as possible in the
contribution and distribution chain.

"For the most part, everyone's been doing frame-compatible mode right
from remote," says Strilkauskas. "When we decided to launch the network, we
felt we needed to come back full left and right [eye]. We knew there would be
different needs from affiliates, and if you bring back one format it might not
suit the other."

"We're going to let affiliates decide what they need," says
Strilkauskas. "Either way, that's an interim solution for us. We really want to
progress to be full left/right to consumers. But right now, that's not

Double-Duty for Standards Converters

Snell has demonstrated two modes of standards-conversion operation for
stereoscopic 3D; one with two Alchemists, where one is handling the left-eye
signal and the other is handling the right-eye; and a single Alchemist
processing a frame-compatible signal where the left-right images have been
squeezed into a side-by-side or top/bottom format. Both worked well, says Gerard
Phillips, GM of conversion for Snell.

"Our view would be that the ultimate quality would be two channels with
two independent Alchemists, as there's no possibility of any crossover between
the left and right images," says Phillips. "But the experiments we've done with
left/right squeezed into a single frame [and one Alchemist processing both] are
pretty much indistinguishable."

Performing standards conversion for sports, particularly in fast-moving
panning shots, is already challenging in the 2D world, notes Phillips. That's
because measuring motion with a high degree of accuracy from frame to frame is
essential for a high-quality conversion.

As Phillips explains, compression systems can use motion-compensation
techniques such as block matching to predict motion from one frame to another
without having to be 100% accurate---as long as the parts of the picture they
identify between two frames are similar, the end result will still look good.
But converting from one frame rate to another in standards conversion requires
more accuracy.

That is why Snell's Alchemist Ph.C-HD product uses a technique called
phase correlation (Ph.C) that is designed to actually measure the speed and
direction of moving objects in a scene, what Phillips calls "true motion," and match
pixels in each frame precisely with their counterparts in adjacent frames. That
gives a portrayal of smooth motion essential for sports coverage.

"With Alchemist, we're looking for true motion, rather than imagining
motion," says Phillips. "So we're looking for stuff that is the same stuff;
it's not predictive like compression. If you're panning across a white picket
fence, if you're compressing it, it's doesn't matter what part of the fence
you're looking at. But in standards conversion, if the motion estimation is not
done well, bits of the picket fence jump backward."

3D ups the ante, says Phillips, because it requires a
standards-converter like Alchemist to apply identical processing to separate,
synchronized left-and-right eye images that have small but very precise
differences that create the effect of depth when they are displayed together on
a 3D set with accompanying glasses. The precise alignment of the left-and-right
eye images [technically, the "horizontal disparity"] must be maintained through
any downstream processing, including standards conversion, or the depth
perspective will be change. Theoretically, the processor could handle them
differently enough to damage the 3D effect.

"You do have separate viewpoints," notes Phillips. "You have two
similar and correlating pictures, but two different pictures. So each standards
converter, each processor, has a different input, and it has the opportunity of
making different decisions. Conceptually, the decisions could be different
enough to break the 3D spell, and push you out of the suspension of reality.
The suspension of reality can also be broken by bad production techniques, such
as having things too close to the viewpoint."

But in testing the Alchemist over the past few months with different sports,
including rugby, soccer and golf, Snell hasn't experienced any major image

"The decisions made independently between the two units are actually
very coherent," says Phillips. "Where I thought it might cause trouble was
where there a big difference from the left-hand side of one image to the
left-hand side of the other. Those would be places where I would suspect a good
standards converter to make different decisions, but we haven't experienced

Lamb concurs, and says that performing standards conversion in 3D
turned out to be less challenging than ESPN originally thought.

"We thought that was going to be the dealbreaker, to be honest with
you," he says. But significant testing by ESPN and Snell showed the Alchemist
could handle it.

"It turns out the algorithms they use are very predictable, and do the
same job [with both feeds]," says Lamb.

No Rest for the Weary

Lamb and Strilkauskas are confident in their preparation in both South
Africa and domestically and expect ESPN 3D to launch without major problems. As
Lamb jokes, "I have a high level of confidence in ESPN's ability to do the

They will have little time for major changes in ESPN 3D's transmission
scheme. The day after the World Cup ends, on July 12, ESPN 3D will be airing
the first ESPN-produced telecast for the new channel with the State Farm Home
Run Derby from Angel Stadium in Anaheim, Calif., site of the 2010 Major League
Baseball All-Star Game. ESPN 3D's schedule going forward will include the X
Games, college basketball, college football (including the 2011 BCS National
Championship game in Glendale, Ariz. next January) and NBA games.

There will be a combination of MPEG-4 and JPEG-2000 encoders on the
NEP/PACE 3D truck that ESPN will use to produce domestic events for ESPN 3D. As
ESPN 3D moves into the fall for college football coverage, it will likely send
left and right eye discreet signals over satellite back to Bristol, using
MPEG-4 compression, and format them there. When it has significant bandwidth over
a fiber link, it will also try JPEG-2000 compression, which it already uses
extensively on its private network between its U.S. locations.

JPEG-2000 is billed by vendors as providing "lossless" encoding for
high-quality HD transmission, even after multiple encodes and decodes, and has
lower latency than MPEG-4. Maintaining synch between 3D's left-and-right-eye
signals with JPEG-2000 compression is a little easier, says Strilkauskas,
because it uses I-frames only. With MPEG-4 it's more challenging, as GOP [group
of pictures] alignment is an issue.

JPEG-2000 hardware itself is also about 25% cheaper than MPEG-4 gear,
adds Lamb. But JPEG-2000 requires a bit-rate of 100 megabits per second or
more, which is cost-prohibitive when backhauling signals internationally over
fiber such as for the World Cup.

In the U.S., where links are cheaper and ESPN already had extensive
full-time fiber capacity, it's a different story.

"In the U.S. running 100 Mbps for an event doesn't cost you much more,"
says Strilkauskas.

E-mail comments to