You are on page 1of 4

FEATURE ARTICLE

AES67
How the net was won
Francis Rumsey
Staff Technical Writer
The AES67 standard describes a relatively straightforward
collection of networking solutions and protocols that together
enable audio networks to interoperate. A special session presented
during the 135th Convention explained the decisions that were
made and the compromises involved during the development of the
standard, led by some of the key protagonists.

T
he AES67 standard for network audio Key criteria of the AES project, originally using them. (Kevin Gross adds that when
interoperability, launched in Septem- known as X192, included low latency (less investigating the existing proprietary solu-
ber last year, stands as a beacon of than ten milliseconds) and high perform- tions, it was found that although they used
cooperation between otherwise competing ance, but the scope excluded non-IP similar approaches they were not designed
parties. During a special session that was networks, audio data compression, and low- to talk to each other, as shown in Table 1.)
part of the network audio track at the 135th performance networks such as the public These issues came down to questions
Convention, chaired by Greg Shay, some of Internet. Video was excluded, but the foun- including methods of synchronization,
the leading lights explained how the stan- dations set out in AES67 are likely to be payload formats, and the way in which the
dard was designed. Also included in this useful for video as well as audio. The group transport system was used.
article is a summary of the key features of undertaking this challenging task consisted
AES67, a standard that will change the way of members from over 100 companies, as AES67 IN A NUTSHELL
that audio networks operate together. well as broadcasters and live sound engi- The topic of synchronization was of crucial
neers. Led by Kevin Gross, it managed to importance, said Hildebrand, and the group
WHAT PROBLEMS WAS AES67 move from inception to published standard settled on IEEE1588–2008 as the way for-
DESIGNED TO SOLVE? within the impressive time scale (for stan- ward. This essentially specifies the Preci-
Andreas Hildebrand of ALC NetworX dards work) of three years. sion Time Protocol (PTP) version 2 as the
explained that the new standard aimed to Although there had been a number of means to achieve synchronization. PTP dis-
tackle the issue of interoperability for audio existing protocols for carrying audio based
networks based on Internet Protocol (IP) on IP, it had not been possible until now to
using existing solutions as much as possible. exchange data packets between networks

Technology Purveyor Date intro Synchronization Transport


Ravenna ALC NetworX 2012 IEEE 1588-2008 RTP
AVB mIEEE, AVnu 2011 IEEE 1588-2008 advanced profile Ethernet,
(IEEE 802.1AS) RTP
Q-LAN QSC Audio 2009 IEEE 1588-2002 UDP
N/ACIP EBU 2007 Data packet arrival times RTP
Dante Audinate 2006 IEEE 1588-2002 UDP
Wheatnet-IP Wheatstone 2005 Proprietary RTP
LiveWire Telos/Axia 2004 Proprietary RTP

Table 1. Existing network technologies, listing their sync and transport methods (courtesy Kevin Andreas Hildebrand explained the basic
Gross) principles of AES67.

190 J. Audio Eng. Soc., Vol. 62, No. 3, 2014 March


FEATURE ARTICLE

tributes absolute time data to devices on WHAT ARE THE MAIN


the network, whereas media clocks for indi- APPLICATIONS FOR AES67?
vidual devices are generated locally from Hildebrand also outlined some of the main
the synchronized time data. A system mas- intended application areas for AES67. These
ter clock is referenced to something like a include commercial audio applications such
GPS-referenced time signal and can tell as installed operations in stadiums, the-
slave clocks the correct time down to 25 ns aters, and theme parks, as well as fixed and
accuracy (although that level of precision is touring live sound. It is also suitable for
not used in AES 67). broadcasting houses and media production
Transport of packets uses IP version 4, operations, including interfacility connec-
although provision for IP version 6 has tions. Although wide area networks (WANs)
been made for some time in the future. such as the public Internet are not the main
The standard is based on RTP (Realtime target for AES67, special WAN services can
Transport Protocol), which itself is based be employed to enable it to operate over
on UDP (User Datagram Protocol—a rela- large corporate networks, for example, pro-
tively simple and fast method of transfer- vided that various features are implemented
ring packets that does not require to ensure the performance is adequate.
handshaking) and IP. Receivers must
Gints Linis explains the thinking behind the
support both unicast and multicast modes HOW DO EXISTING AUDIO decisions
while senders may support unicast, multi- NETWORK STANDARDS FIT IN?
cast, or both. While Layer 1 solutions It would seem that for existing proprietary
make use of the lowest layer of the OSI network technology to fit in with AES67 An AVB node is essentially a Layer 2
network model, which is the physical some form of adaptation will be required network solution, so such a thing by itself
layer (e.g., copper/fiber cable), Layer 2 in most cases. This seems likely to be cannot support AES67. However it is possi-
solutions use Ethernet protocols. Layer 3 achievable by means of additional protocol ble to run AES67 across an AVB network,
solutions (as used by existing solutions options for existing equipment (supplied and AVB devices use a variant of IEEE1588
such as Dante and LiveWire) are based on by means of firmware upgrades, for exam- synchronization, so much of the key tech-
IP, on top of which is laid UDP and RTP. ple), or by using new equipment that can nology is already present. For AVB devices
Gross adds that using the IP layer handle both AES67 and proprietary flavors to fully support AES67, though, they would
enables better scaling in terms of number of communication. have to incorporate the IP-based means of
of participants and geographical extent By offering AES67, a manufacturer of data transport. There is a chance, therefore,
than Ethernet and is more secure and audio networking equipment would essen- that AVB-biased manufacturers will also
manageable. The IEEE 802.1 AVB (Audio- tially be accepting that open standards are adopt AES67, suggested Andreas.
Video Bridging) standard, by the way, only the way forward, requiring them to distin-
enhances the Layer 2 Ethernet solution. guish themselves from the competition by MAKING THE DECISIONS
This is discussed further below. other means than the old argument of “you Gints Linis explained some of the thinking
Some recommendations on sampling have to buy all your equipment from us if behind decisions made during the forma-
rate and resolution were made, namely that you want it to work together.” It is still tion of the standard. There were three main
receivers should support 16- and 24-bit possible, of course, that operating certain conflicting factors—the technical ideal, the
linear PCM at 48 kHz, whereas senders may equipment in proprietary modes may offer costs of implementation, and solutions that
support 16 bits, 24 bits or both. Although selected advantages in terms of perform- already exist.
sampling rates of 44.1 and 96 kHz are ance or features, but this would become a Header extensions are not recommended
supported, they are not mandated in user choice. in AES67, for example, said Linis. They are
AES67. It was also decided that 1–8 channel Manufacturers of proprietary networking often poorly implemented, poorly tested, or
audio support per stream would be manda- solutions are currently at various stages in simply do not work, but receivers must be
tory, but any number of streams could be their attitudes to AES67. The RAVENNA able to tolerate whatever comes in without
carried up to the bandwidth capacity of the audio networking protocol, for example, is crashing. The 44.1 kHz sampling rate was
network. This could mean that up to already essentially AES67 compatible, said not made mandatory in the standard
around 500 channels could be carried on a Hildebrand. New nodes being manufactured because it is often considered a consumer
gigabit Ethernet link. All devices should by Axia for LiveWire will also be fully rate and seems likely to disappear eventu-
support packets lasting 1 ms (48 samples at AES67 compatible, although legacy nodes ally. 96 kHz was not made mandatory to
48 kHz), but 4, 1, 1⁄3, 1⁄4, and 1⁄8 ms packets are will not. A bridge is possible that will enable avoid placing a needless burden on devices,
also supported older nodes to talk to AES67 networks. although it is optional.
“Differentiated services” (DiffServ) has AES67 streams are compatible with EBU A packet duration of 1 ms is not compati-
been employed as a means of delivering ACIP (Audio Contribution over IP) systems, ble directly with any of the technologies
quality of service, and connection manage- although ACIP does not incorporate whose developers participated in the creation
ment is based on the Session Description absolute time in the synchronization of this standard, so it might be asked why it
Protocol. For unicast connections the stream so it would not be fully AES67 was decided to use it for AES67. Although
Session Initiation Protocol (SIP) is used. compatible. The facilitation of 192-sample not perfect, Linis explained that such a
Discovery of devices on the network has packets in AES67, for example, was specifi- packet size is sufficiently short for buffering,
been excluded from the standard for cally included to enable stream compatibil- and the packet rate is low enough not to
reasons that will be described later. ity with ACIP devices. overload the CPUs of some devices. If you

J. Audio Eng. Soc., Vol. 62, No. 3, 2014 March 191


FEATURE ARTICLE

arate connector) was regarded as undesir- AES67 AND OTHER NETWORK


CHANNEL BUNDLING able, so it was important that sync should TECHNOLOGY
The standard discusses the bundling of be a feature of the network protocol itself. Kevin Gross pointed out that it had never
related channels in the following way: Furthermore it was desired to be able to been the aim to create more confusion in the
“Although bundling multiple channels in
a stream can improve network and pro- distribute several media clocks simultane- market by creating another competing stan-
cessing efficiency, it is recommended ously on the same network, to allow for a dard. The existing approaches were all essen-
that bundling be used primarily in serv- range of audio and video systems and tially trying to achieve the same thing, so it
ice of the application. Channels of devices to be operated at different rates. had been important to attempt compatibility
related material (for example, stereo or
surround sound) are good candidates
Another requirement was to be able to run with the best parts of those, rather than
for bundling. Bundling of unrelated on standard network infrastructure, say introduce a lot of new concepts. Putting
channels destined for different receivers using existing routers and switches, as we together pieces of technology that already
in an effort to reduce network overhead can’t require IT infrastructure to be exist seemed like a much more likely
is discouraged as this complicates entirely replaced in order to handle audio. approach to succeed than “making stuff up.”
media routing configuration.”
The house clock requirement could be diffi- However AES67 was never intended to solve
cult to fulfil in this case, but basic stream- everybody’s problem, it was aimed at a spe-
ing should be possible regardless of legacy cific market that needs high-performance
want compatibility “out of the box,” he said, equipment. audio with low latency and was not going to
this is what you get here. A longer packet of The existing AES11 standard has rules be particularly useful for, say, speech codecs,
4 ms is allowed for compatibility with ACIP for synchronizing different pieces of and would not work on all networks.
devices, but this is not mandatory. It is a equipment in a larger system, and this AVB was one particular IEEE standard
common trend for devices to become more suggests that devices need to be locked to that Kevin wanted to contrast with AES67,
flexible, so it seemed likely that a variety of within 1 µs. A protocol needed to be as there can easily be misunderstandings.
packet sizes would be possible in future, and adopted for AES67 that could assist in this Both standards use IEEE1588 sync
there is also a trend towards lower latency. process, one that was likely to work across arrangements for example, but the main
Shorter buffering offers lower latency but a whole range of existing IT devices. There difference with AES67 is that it allows you
might cause dropouts, whereas longer buffer- was only really one option, and that was to use a standard IP network if you wish.
ing improves robustness but could impose PTP. However there are several variants of This gives rise to a lower sync accuracy but
more constraints on the implementation. A PTP that are mutually incompatible, so the it’s still reasonably useful, whereas AVB
compromise needed to be achieved between group settled for version 2 (2008) that had enables one to get very precise timing
options that could be implemented in a range some useful features such as profiles accuracy under all circumstances. It does,
of devices. (designed for different applications). PTP however, mean that you have to use AVB-
can run on Layer 2 Ethernet, IPv4, or compliant switches, which could be
THAT SYNCING FEELING IPv6, so the group settled on IP version 4. regarded as a barrier to interoperability.
Stefan Heinzmann delved more deeply into PTP, it must be realized, distributes AVB uses an alternative BMA (best master
matters of synchronization in AES67. absolute time data, not a clock signal, so clock algorithm) to that defined in the
Among the main requirements had been to individual nodes have to generate their basic IEEE1588 standard (the version
have phase accuracy between multiple inde- own clock signals from the time data. adopted for AES67), to choose the most
pendent nodes on a network, whether they There is explicit information in the stan- appropriate sync source on the network
were receiving the same stream as each dard about how to generate sync clocks automatically. This can mean that it takes
other or not. This was partly to ensure from the time data, so that this is not left perhaps half a second longer to set up the
things like accurate stereo imaging, which to chance. Also, even though the standard master clock source on an AES67 system,
demands close coherence between audio does not deal directly with audio/video
channels. A separate clock signal (on a sep- sync issues such as drop-frame timecode
rates, there is provision to be able to
handle such scenarios in the implementa-
tion. Liaison with SMPTE, which is devel-
oping a related networking standard,
principally for picture streams, is helping
to iron out these matters so that the two
standards are compatible.
Although network switches do not tech-
nically need to be “PTP-aware” to handle
synchronization data, some switches do not
support PTP to the required level of accu-
racy. So if one wanted to use AES67 to
distribute a house clock accurately it would
be necessary to upgrade switches to those
that could handle PTP natively. This is one
of the compromises in the standard—an
area in which legacy network equipment
Stefan Heinzmann deals with the ins and outs might not be able to handle all of the new Kevin Gross said,”We tried to avoid creating
of synchronization. requirements. more confusion by creating another standard.”

192 J. Audio Eng. Soc., Vol. 62, No. 3, 2014 March


FEATURE ARTICLE

but it is likely to work with a broader range devices. (Kevin Gross comments that
of equipment. (Annex D of the standard unicast is more secure and can be more
explains how to use the highly accurate simple and efficient if there is only one
clocking system employed in AVB as a intended receiver.) SIP (session initiation
replacement for that specified in AES67.) protocol) can be used to set this up, being a
AVB uses Ethernet as a transport proto- protocol that is also used for things such as
col and employs a VLAN tag at the start of establishing telephone calls over IP. It
each packet to set a high priority. Within involves a certain software overhead but it
these AVB packets the audio information is was considered “worth it,” said Shay.
carried as a Firewire payload. So essen- Dealing with the “discovery problem”
tially AVB is a form of “Firewire over was one of the biggest compromises that
Ethernet.” AES67 uses the 6-bit DSCP the group made, according to Shay. With
field in the IP header for setting packet many audio channels available, an auto-
priority. (VLAN tags are allowed but are matic system to find what is “out there”
not required for AES67.) Kevin explained would have many advantages. However
that the majority of IT professionals know there was no off-the-shelf solution that
Greg Shay spoke on management and
a lot more about IP than they do about would do everything for everyone, and the
discovery.
lower-level Ethernet, so this was an argu- group did not want to compromise the
ment to go with IP for AES67. Gross. If you want such a feature, you future use of the standard by saddling it
There is a bandwidth reservation always have the choice of using an AVB with a solution that was inadequate for
approach used in AVB that attempts to network, or indeed running AES67 data some use cases. An appendix to the stan-
calculate whether there is sufficient band- over an AVB network, as described in Annex dard, therefore, presents a survey of exist-
width to handle a stream on the network C of the standard. ing options, leaving it to the user to
before it is “admitted.” This is designed to implement whatever is appropriate.
ensure that if a stream is allowed onto the CONNECTION MANAGEMENT
network it will always have enough capacity AND DISCOVERY SUMMARY
to run properly. AES67, on the other hand, Connection management is about getting The AES67 standard describes a relatively
does not employ any such “admission audio communication going, said Greg straightforward collection of networking
control,” so if there is insufficient capacity Shay, whereas discovery is about finding solutions and protocols that together
you will experience unreliable transmis- out what devices are on the network, rather enable audio networks to interoperate. This
sion. It’s therefore up to the system like a phone book. In AES67 there is no may require some existing systems to be
designer to make sure that the network has connection management on top of the mul- adapted, and the approach may not work
sufficient capacity to handle the scenarios ticast mode. In essence, audio streams are under all circumstances or for all applica-
you expect. (This is also true of AVB, of broadcast on the network and other devices tions, but the building blocks are those
course, it’s just that with AVB you would pick them off as required. This is very sim- used in standard IP network systems and so
get an “access refused” response if you tried ple to implement. Internet Group Manage- should be relatively easy to implement by
to send data on a network that was already ment Protocol (IGMP) is usually imple- developers.
too full, rather than it being accepted and mented in a network switch and it simply
the result being unpredictable.) There have requires that devices advertise themselves
Editor’s note: Go to
already been some failed attempts by the as potential sources, while receivers sub- www.mobiltape.com/conference/Audio-
Internet Engineering Task Force (IETF) to scribe to a particular multicast stream. Engineering-Society-135th-Convention to
introduce forms of admission control for AES67 needed to have a unicast mode as purchase and download an mp3 (13AES-
IP-based networks, and those people have well as multicast. Unicast modes involve N05) of this workshop.
pretty much given up on the idea, said point-to-point connections between

The AES67 project group celebrates the successful completion of its work, with Richard Foss (chair, working group SC-02-12) and Kevin Gross
(leader of the project) seated center.

J. Audio Eng. Soc., Vol. 62, No. 3, 2014 March 193

You might also like