[jand]

This sounds like fun tech.

But isn't this dead before it really lifts off? Intel 1588 is not supported by every customer-NIC, and far from supported by most switches and such.

So if i have to build a "special" hardware environment anyway, why use AES67 and not some other proprietary solution without IP which may (? or may not) deliver better latency and jitter?

Last time i checked (which is a while back), the same support-argument would apply to multicast.

Am i missing the point here? Please enlighten me. Thank you!

[jcrawfordor]

While this can't be dropped into a lot of existing networks, that doesn't really eliminate the advantages of IP.

This might be best explained by analogy: VoIP. Various issues including security model, DHCP-based provisioning, and QoS mean that corporate VoIP phones are typically deployed on a specially prepared network using a dedicated VLAN. While various VoIP vendors claim that you can just drop their solution on your existing network and have it work, this rarely pans out, and it's very common for introduction of VoIP to involve upgrades to network appliances and possibly architectural changes.

It might also be important context that prior to the adoption of VoIP most corporate environments were already using a proprietary digital telephone system such as Nortel Meridian, so it's not even a matter of upgrade from analog to digital.

Rather, the IP-based network, even with special requirements, is more flexible (due to the large set of routing, switching, etc protocols available for IP) and less costly to maintain (due to common skillset and equipment with computer networks). Even better, while there may be a new network investment required to switch to VoIP, that investment is "dual purpose" and the new network equipment will also serve your computers.

I'm not an expert in this field, my experience being limited to some work with Dante equipment years ago---VoIP is more my wheelhouse. But I think the situation is very similar here. Adopters of live audio over IP will almost certainly need to invest in new network equipment and possibly rearchitect their existing network. VLAN segregation and special QoS policy will presumably be the norm.

But choice of vendors and common skillset, if nothing else, will make the IP network less costly to maintain. There's a wide variety of technology available for moving IP around in interesting situations, fiber is popular in large theater contexts because of a perceived improvement in reliability over long runs (probably less significant in the days of GbE but I haven't been involved in this kind of thing in a while). Further, with sufficient precautions in place the network can be dual-use and can also serve purposes like administrative networks and even front-of-house wifi.

Protocols like Dante are already being widely adopted, and compatibility can be a big headache, so I think a uniform standard for this kind of thing will be very popular.

[jand]

I get your point, and you may be right and it might take off.

But is a latency of 2ms up to 50ms really attractive? I am in no way near audio engineering, but i can remember the MIDI-folks swearing about their 2ms latency.

[bubblesorting]

Not very good keyboard player here: higher than ~~8ms is annoying for playing/recording. But 50ms probably isn't bad if say, you're blasting a message through a factory intercom, or piping music through a dental office, or stuff like that.

[detaro]

And for large venues, remember that sound only goes ~3.4 m in 10 ms. loudspeaker-to-ear latency can be larger than all of your processing. Doesn't mean it doesn't matter, but makes the last few ms less important.

[lukeh]

AES67-2018 specifies packet times (latencies) of 125us, 250us, 333us, 1ms and 4ms. Senders and receivers may support additional packet times.

[qppo]

It's not DoA, the tech has been in production for close to 15 years. If you've been to Disneyland you've heard audio broadcast over IP, or watched a match of League of Legends (or virtually anything that comes out of a stadium). Hell if conventions were still around, go to NAB or AES and look for the "we speak Dante" signs everywhere.

The problem AES67 solves is that there are multiple competing standards and it's difficult or impossible to connect them over the same network. The standard is there to unify under one protocol. The tech problem was delivering audio in sync over large networks, and rather than use custom hardware they used commodity networking gear and network protocols (not doing that might seem crazy, but that's generally what happens in high end audio). But even then, you probably have some special hardware in your environment to connect microphones and speakers to the network in the first place.

[detaro]

It's somewhat widely already in use for large events etc, so clearly it works.

Not for consumer products, but it's not optimized for those. If you buy any kind of professional NICs and switches, 1588 support isn't that hard to find. Wide support means less trouble if one manufacturer changes things, more choice in hardware than a proprietary system would likely have. Can live with existing networks.

E.g. lets say you're building out an event at a large convention center. They probably can give you Ethernet or fiber from one end of the building to the other, but can they give it to you for your proprietary thing of choice?

Or you are making a permanent installation: do you want to have to put one vendors stuff everywhere, as a parallel network of proprietary stuff, or do you prefer something that works on the network infrastructure you already have, with maybe some upgrades to some components, but keeping it with the stuff your network people already know? Especially if it means you don't need to pull additional cabling or fiber everywhere?

[lukeh]

There are economy of scale advantages to using standard protocols and hardware: the R&D is paid for by the computer industry, which is much larger than the audio industry. Many NICs have support for PTP hardware clocks and hardware time stamping, and (unlike AVB/TSN) AES67 will work with most network switches. It's also possible to implement software endpoints by pointing an RTP client at the AES67 peer (albeit, this won't be a true AES67 endpoint as it won't play out synchronously without a PTP synchronised media clock).

[bsder]

> But isn't this dead before it really lifts off?

No, because automotive is the use case driving this and consumer/professional audio will come along for the ride. Think about how you synchronize the LCDs and speakers in an SUV--Ethernet is way easier than just about any other solution.

The biggest obstacle to adoption in the consumer/prosumer space is the fact that everybody killed Ethernet ports on their laptops.

[qppo]

What? I've been going to AES67 meetings and talks for several years and worked on AoIP products, I don't think I've ever met someone from the automotive industry or heard they used the protocols that AES67 is going to supplant/integrate with (Dante, Ravenna, etc). The standards committees members were professional and commercial audio manufacturers.

This isn't how you connect something trivial like an LCD to your speakers, it's how you rig up the audio system at a theme park or network the audio feeds for all the broadcasters at a World Cup match. Those are two applications that AES67 committee members had worked on and where it will probably be deployed.

[bsder]

I was being kind and lumping AES67 in with TSN/AVB (time-sensitive networking/audio-video bridging).

If I'm being snarky, AES67 is a last ditch effort by the proprietary vendors to remain relevant before AVB/TSN wipes them out.

Before Covid, it seemed like Presonus was wiring up every new church I knew of with AVB/TSN.

> This isn't how you connect something trivial like an LCD to your speakers

Automotive companies do not regard that as trivial. Copper is heavy and expensive and difficult to route. LCDs are in the ceiling and speakers are in the floor. The only common point is at the (literal, in this case) firewall.

Collapsing everything to a few pairs of Ethernet is a big deal for them.

[lukeh]

AVB/TSN has a lot of great things about it but it's a market failure outside of proprietary audio systems (e.g. Avid, Presonus) and the automative world. The fact that it requires not just the endpoints (which themselves require an Ethernet chipset with multiple send/receive queues and hardware time stamping) but all network switches to support 802.1AS, MSRP, MVRP, etc creates a stalemate between vendors (waiting for a deployed base) and customers (waiting for market availability).

My personal opinion is that AES67 will eventually win. Dante has the installed base for now, and it can operate in an AES67 interoperability mode (albeit with quite a few limitations). SMPTE 2110-30 is essentially AES67. AES/TSN will limp along within homogeneous networks with Dante/AES67 at the edges where interoperability is required.

Don't get me wrong, I like AVB (see other post about AES67/AVB bridge), even though it has some quirks, e.g. the relationship between PTP and media clocks is a lot more straightforward in AES67.

[detaro]

Re automotive audio and Ethernet, AVB is kind of a thing in automotive, or at least some vendors try to make it one.

[aidenn0]

What's the advantage to this versus a more traditional PA system? Just fewer wires to run?

[qppo]

The big win is distributed audio networks. For example an esports match might have play-by-play and commentary in 20 languages, in traditional sportscasting they fly those folks out on location. What they do for League of Legends (and a lot of international sportscasting) is use AoIP to send the stream back to their studio, multicast to commentator crews around the world, multicast back to the studio, multiplex into the simulcast, and broadcast out on their streaming platforms. (Also I'm recalling this info from a talk at an AES meeting from about 18 months ago so if anyone from Riot has better intel, please correct me!)

Onsite, it's not feasible to use miles of cable for analog signals (and it is miles, in large venues). Even with balanced connections you start getting noise problems after a few hundred feet. Digital conversion closer to the sources solves this issue, which means those DACs and ADCs need to be networked somehow. Building infrastructure for corporate networks onsite is a mostly solved problem with lots of cheap hardware available, so they just use that and place custom gear with the converters at the input/output locations. Not to mention if lighting is involved, noise can get really quite bad.

It's just all around more modular, cheaper at scale, effective, and foolproof than full analog.

There's also a whole bunch of less sexy use cases in corporate environments where a PA doesn't work at all.

[tonyarkles]

Even just a digital snake! A pair of Ethernet cables is way way way easier to set up than a 24 channel analog snake. Plus... need a couple more inputs? Just throw another stage box up there and connect it to the switch.

I’m not in the industry but my wife is, and AES67 seems like a complete game changer.

[sethhochberg]

Yep. I was heavily involved in the live sound world right in the chapter where copper snakes were being replaced with early digital snakes, and even that was a game-changer enough on its own. Less weight, less unpredictability (even in good copper snakes, the wires are _tiny_ and prone to breakage in entertainment use), more channels, less need to differentiate between your sends and returns and keep corresponding adapters around if you had a return-heavy mix, far better options for performers to make their own unique monitor mixes on stage instead of relying on whatever submixes could be sent back from the console on the returns that happened to be available.

And that was just the early days, when even the protocols that sometimes used Ethernet were only L1 and L2 and not IP-compatible/routable. You might have had several lines of ethernet running back and forth to the stage, but couldn't comingle snakes with IP-based control systems.

AES67 and friends pushed the industry to a point where, basically, everything speaks IP. The stage remains an analog realm up to the DIs and mic premaps, and everything can be routed and distributed in almost infinite combinations from there with commodity hardware.

Its seriously as impactful as the shift from tape to digital recording in terms of the new workflow options and paradigms it opened up.

[rasz]

I know Americans love big SUVs, but I dont think even they managed to build something so large as to require compensating for the sound propagation delay.

[Zenst]

I had a look at 1588 and was a rabbit hole of paywalls - latest draft seems to be https://standards.ieee.org/standard/1588-2019.html

Why is a standards group paywalled!

[__d]

Many standards groups require payment for their published standards: the ISO and most national standards bodies (ANSI, etc) are maybe the most prominent examples.

This is their business model. You pay for access to the standards documents, and that money funds the development and maintenance process. It's supported by governments requiring adherence to these standards, so implementors are obliged to purchase them.

Gratis access to standards is a newer model which relies on a different funding stream. The IETF is one such example.

[jand]

I once made the shocking discovery, that most ISO and IEEE documents can be found on chinese PDF dump sites. Since this is obviously against all copyright laws, i would recommend to explicitly exclude such domains from your google search. Just to be safe.

[jimmySixDOF]

Yes almost anything to do with Industrial Ethernet type standards are all over the place. 1588 PTP mixes with IETF TICTOC workgroup output. Synchronous data also gets used in Psuedowire backhaul for low level circuit emulation signaling like from SCADA or GSM towers etc. And that's without going into IECxxxxx's for powerplant comms.

Certifiable Precision Timeing over anything is not trivial.

[lukeh]

You can also read the linuxptp code, it's beautifully written.