A television stage often mixes several audio-over-IP families. A console on Dante, a mic park on Ravenna, a video infrastructure on SMPTE ST 2110: three dialects, one network. AES67 exists so they understand each other. The standard, published by the Audio Engineering Society in 2013 and since revised, replaces none of these systems. It describes a minimal common ground, a subset of settings that all of them accept.
An interoperability standard, not another system
The distinction matters. Dante, Ravenna, Livewire+, Q-LAN and WheatNet-IP remain proprietary solutions, each with its own configuration tools and strengths. AES67 does not try to unify them, only to guarantee that a stream leaving one system arrives readable in another. When a maker claims AES67 compatibility, it promises that bridge, not the end of its in-house format. It is the same shared-standard logic you find one layer up, in the audio layer of SMPTE ST 2110.
The rules the standard enforces
For the common ground to hold, AES67 sets a few non-negotiable points. Transport runs over RTP on UDP and IP, most often multicast, so one stream can serve several receivers. Audio travels as linear PCM, 48 kHz mandatory, with 24-bit required at the receiver, while 44.1 and 96 kHz stay optional. Each stream is described with SDP, the same format IP telephony uses to announce a session.
Synchronisation is the heart of it. Every device locks to a master clock over PTP version 2, the IEEE 1588 standard. Without that shared reference, two streams sampled at the same rate would slowly drift apart, and the network would produce clicks. PTP distributes a single time at the microsecond level, from which each device derives its audio clock. The kinship with studio clocks is no accident: the move to PTP also reaches video sync generation.
Latency, a matter of packets and buffer
AES67 thinks in packet time. The default, one packet per millisecond, offers a good balance between network load and latency. Devices can drop to 125, 250 or 333 microseconds to cut delay, at the cost of denser traffic and more sensitivity to jitter. At the receiver, a link offset sizes a buffer that absorbs network variation: too short and it lets faults through; too long and it adds delay. That setting stays one of the field moves when you tie two domains together, much like managing a large network in our Dante Domain Manager guide.
What the standard leaves open
AES67 deliberately stops at transport and clocking. Device discovery and connection management are not mandated: a stream announces itself through its SDP, but nothing dictates how a receiver learns it exists. In practice, each system keeps its own mechanism, and ST 2110 infrastructures rely on the NMOS specifications to catalogue and connect streams. Quality of service also falls to the installer, who prioritises PTP and audio on the network to protect them from congestion. Redundancy is not part of the core either: it is handled on top, following the two-path method of ST 2022-7.
Where AES67 fits in an installation
In a modern control room, AES67 is rarely the main system: it is the handshake between systems. You enable it to route a few streams from a Dante console into a Ravenna infrastructure, or to inject audio into an ST 2110 core whose sound layer, ST 2110-30, rests precisely on AES67. For links that leave the building, you stay on other tools, such as the contribution codec, and MADI keeps its place in point-to-point interconnection. AES67 replaces nothing: it holds the whole together.
My take
AES67 is one of those standards you only notice when it is missing. The day you need two parks from different brands to talk, it turns a headache into a checkbox. Its merit is its modesty: it fixes the bare minimum, clock, transport, description, and leaves the rest to the systems and their management tools. The trap is thinking that enabling AES67 is enough to make everything communicate without setup. PTP and link offset still need care, but once they are in place, the network becomes surprisingly easy to live with.