The restrictions on movement which are such a necessary part of the social distancing required to fight the spread of COVID-19 has brought the advances made in online collaborative working in recent years into sharp focus. In Podcast 411, we discussed the things we now take for granted in terms of online collaboration and file sharing. However, one area which remains beyond our current capabilities is real-time, online performance, with musicians playing together from remote locations across the internet.

AoIP Across A LAN

The development and proliferation of Dante, RAVENNA and AVB/TSN have shown how useable computer networks are for the distribution of low latency, multichannel, uncompressed audio but this has always been across a LAN or local area network. The kind of simultaneous live performance being discussed here has always been impractical across the internet as the route any particular packet of data will take passes beyond the control of the user. Ultimately the choice has been between studio-quality audio or operating over a long distance. You can’t have both.

Something which struck me straight away was that Source-Connect offers high-quality audio at a distance. Isn’t that a solution? Unfortunately not as although using Source-Connect is it possible to record remotely, the workflow is recording to DAW playback rather than live input at both ends and also Source-Connect uses compressed audio for the live monitoring and “backfills” the placeholder, compressed audio with uncompressed audio post record. Clever and effective but not quite the same thing.

A recent showcase event at the University of West London and Edinburgh Napier University entitled Video Killed the Radio Star: How the Future Began illustrated how high quality simultaneous live performance by musicians at opposite ends of the UK is possible but it took some extremely clever exploitation of currently available software and hardware products to achieve it. The crucial inspiration came from Professor Justin Paterson and Dr Paul Ferguson. Using some recent additions to the Dante Domain Manager software a possible way to provide stable, reliable clock to UWL and Edinburgh simultaneously was envisaged.

Clocking - The Fundamental Problem

Talking specifically about Dante, the reason a Dante network doesn’t work effectively across the internet is that there is no way to provide an accurate clock signal to all sites on such a system. In the same way, as we wouldn’t digitally connect audio equipment without a suitable word clock, network audio relies on a network clock to enable solid timing and eliminate jitter.

Ferguson explains…

“in a standard digital audio scenario, we wouldn't dream of sending audio between two devices without rock-solid and consistent clock. With Dante, that fundamental principle is no different. Certainly with Dante Domain Manager, we can now send a Dante stream containing clock packets from anywhere to anywhere. However, where the network between two devices spans hundreds of miles and isn't ours, that network is beyond our jurisdiction. We don't know the hardware hops or the traffic levels, so we'll always run the risk of not having that all-important, low jitter, consistent clock. It might be OK, or it might not. If it's not, the main symptom for the audio user at either end is generally silence!"

How Did They Do It?

There are four elements which enabled this project:

JANET - A wide bandwidth internet connection. Being universities both UWL and Edinburgh Napier have access the Joint Academic NETwork, a dedicated network shared by UK universities.

Dante Domain Manager - Dante Domain Manager allows for logical groupings of Dante devices into Dante Domains, within which are user-defined zones. A single Domain can contain member devices from different zones but clocking remained a serious issue.

Support for SMPTE 2110 / AES67 compliant devices in Dante workflows, and support for GPS Synchronisation. Using SMPTE 2110 / AES67 for audio transport opened up the possibility for clocking between sites using GPS

The Sonifex AVN-GMCS, an AES67 compliant grandmaster clock designed for use with AoIP applications. The AVN-GMCS uses PTPv2 (Precision Time Protocol v2) to synchronise devices on a network, having derived its own absolute time from the atomic clocks built into GPS satellites.

So there we have it. The way to establish a common clock between remote sites is to use separate master clocks which are both tightly synchronised to a very accurate common clock source - GPS.

What About Latency?

Having sufficient bandwidth available courtesy of the JANET and having solved the clocking issue via GPS it would be seen that all is well, except that the issue of latency remains. Dante latency isn’t deterministic even across a LAN, though it is so short as to be negligible in all but the most extreme of cases. Once audio is taking an unknown route across the internet to make a round trip through the system, latency is always going to be a hurdle.

In the case of Video Killed the Radio Star: How the Future Began, the best-effort nature of Dante’s latency could present issues. Professor Ferguson explains…

“much like when setting buffer size on a DAW, it's a trade-off between acceptable latency for the performing musicians versus system performance. Even with a firewall to contend with, early experiments revealed a 9.5ms round-trip time for data packets sent from London to Edinburgh and back. Whilst initially encouraging, the nature of the JANET network having 18 million users means that at peak times, the increased traffic would force this round-trip time to 15ms or more, with occasional peaks of up to 30ms. Given this was a live, keynote performance to open the conference, I erred on the side of caution. Even though evenings are generally quiet times for traffic, we simply set the buffer on our RedNet interfaces to 20ms at each side: 40ms overall. A little longer than you'd ideally want for drums, but safe, and absolutely useable for the synthetic textures we were getting from Edinburgh."

So the limiting factor on this experiment is the speed of the network infrastructure, and as network capacity and speed inevitably rise, it’s hard to see any reason why live tracking across the internet, with performers sharing cue mixes in real-time across potentially unlimited distances shouldn’t become a reality in the near future.

If you want to know more about the showcase event at UWL for which this performance was devised then read the detailed report provided by Focusrite.