Tag: Audio over IP

In legacy recording studios, concert arenas, and broadcast centers, routing 64 or 128 channels of balanced analog audio between live stage boxes, mixing consoles, outboard racks, and DAW recorders required massive, multi-ton copper multicore “snakes” terminated with hundreds of solder joints and patchbay normals—each vulnerable to ground loops, capacitive high-frequency roll-off, and crosstalk. Today, modern facility engineering has transitioned entirely to High-Speed Digital Audio Interconnects and Audio-over-IP (AoIP) Layer-3 Switched Ethernet Networks capable of routing over 512 bidirectional channels of uncompressed 24-bit/32-bit PCM audio down a single Category-6 copper cable or single-mode optical fiber with sub-millisecond latency.

How can a standard non-deterministic corporate Gigabit Ethernet switch transport hundreds of real-time studio audio streams simultaneously without dropping a single sample or introducing clock jitter? The answer lies in the precision hardware timestamping of the IEEE 1588 Precision Time Protocol (PTPv2), Differentiated Services Code Point (DSCP) Quality of Service prioritization, and Real-time Transport Protocol (RTP) packetization. This comprehensive engineering reference examines legacy point-to-point digital buses (AES3, S/PDIF, ADAT Lightpipe, MADI) alongside modern AoIP standards: Audinate Dante, AES67, SMPTE ST 2110-30, Ravenna, and AVB/Milan.

Before packetized Ethernet networking conquered audio routing, digital studio interconnects relied on dedicated, point-to-point synchronous serial transmission lines where the sampling clock was embedded directly inside the data transitions via biphase-mark coding:

Designed by Alesis in 1991, the original ADAT Optical Interface runs at a fixed frame clock of $48\text{ kHz}$ ($12.288\text{ Mbps}$), carrying 8 channels of 24-bit audio. When studios transitioned to $96\text{ kHz}$ and $192\text{ kHz}$, engineers devised a clever backward-compatible trick called S/MUX (Sample Multiplexing): instead of doubling the optical LED clock speed, an $88.2/96\text{ kHz}$ converter puts odd samples (Sample 1, 3, 5...) onto ADAT Channel 1 and even samples (Sample 2, 4, 6...) onto ADAT Channel 2! The receiving interface stitches those two $48\text{ kHz}$ channels back together into a single $96\text{ kHz}$ channel—trading channel count for sample rate ($4\text{ channels}$ at $96\text{ kHz}$ via S/MUX2, or $2\text{ channels}$ at $192\text{ kHz}$ via S/MUX4).

While AES10 MADI remains rock-solid for point-to-point trunking between a live FOH console and a stage rack, MADI is still a one-way daisy-chain or point-to-point pipe: if you want to route Microphone #14 from Stagebox A simultaneously to the Front-of-House Console, the Monitor Console, the Broadcast Truck, and the Pro Tools Recorder, you need dedicated hardware MADI splitters and routers.

Audio-over-IP (AoIP) replaces dedicated audio cables with standard Layer-3 IP packets flowing across commercial off-the-shelf (COTS) Gigabit ($1\text{ Gbps}$) and 10-Gigabit ($10\text{ Gbps}$) managed Ethernet switches. Developed by Audinate in 2006, Dante (Digital Audio Network Through Ethernet) has become the de facto global industry standard, alongside the open interoperability umbrella standard AES67:

1. Raw Uncompressed Audio Bitrate per Channel (24-bit / 48 kHz):
   Bitrate_1ch = 48,000 samples/sec * 24 bits = 1.152 Mbps (or 1.536 Mbps w/ 32-bit slot)

2. Standard Dante / AES67 Packet Framing:
   - Transport Layer: UDP/IP Unicast or IGMPv2/v3 Multicast (Port 5004 RTP / 4321 Dante)
   - Packet Time (Latency Profile):
     * Dante Ultra-Low Latency: 150 microseconds (0.15 ms) -> 8 samples/packet @ 48 kHz
     * Dante Default / AES67:   1,000 microseconds (1.0 ms) -> 48 samples/packet @ 48 kHz
     * Broadcast High-Density:  4,000 microseconds (4.0 ms) -> 192 samples/packet @ 48 kHz

3. Maximum Bidirectional Channel Capacity on a Single 1 Gbps Cat6 Cable:
   512 Channels In + 512 Channels Out (1,024 total channels @ 48 kHz / 24-bit)!

In a Dante or AES67 network, audio is routed in “Flows” (bundles of typically 1 to 8 channels per IP packet stream):

• Unicast Flows (Point-to-Point): When a stagebox sends a private 4-channel flow solely to the FOH console, it addresses the UDP packets directly to the console’s specific IP address. Only that switch port receives the traffic.
• Multicast Flows (One-to-Many): If 10 different devices on the network all need to listen to the same 8 stage microphones, sending 10 separate Unicast copies would multiply network bandwidth by $10\times$. Switching the flow to Multicast transmits a single stream of packets to an IANA Multicast Group IP address (239.x.x.x). However, if left unmanaged, an Ethernet switch treats multicast packets like broadcast traffic and floods them out every single port on the switch—overwhelming lower-speed devices! Therefore, managed AoIP switches require IGMP Snooping (Internet Group Management Protocol) enabled so the switch hardware inspects subscription requests and forwards multicast audio streams exclusively to ports that explicitly requested them.

Here lies the deepest engineering question of Audio-over-IP: Ethernet switches store and forward packets in variable-length queues. Depending on network congestion, Packet #1 might take $12\text{ microseconds}$ to cross a switch, while Packet #2 takes $85\text{ microseconds}$ (Packet Delay Variation / Network Jitter). If a DAC clocked its converters directly off the arrival time of Ethernet packets, the resulting audio would suffer catastrophic jitter and glitches!

AoIP decouples packet arrival time from audio sample clock generation using the IEEE 1588 Precision Time Protocol (PTPv1 in legacy Dante; PTPv2 in AES67, ST 2110, and modern Dante):

1. Grandmaster Clock sends 'Sync' message at exact time t_1; Follower records arrival time t_2.
2. Grandmaster sends 'Follow_Up' packet containing the exact hardware departure timestamp t_1.
3. Follower sends 'Delay_Req' message at time t_3; Grandmaster records arrival time t_4.
4. Grandmaster sends 'Delay_Resp' packet containing timestamp t_4 back to Follower.

Follower Computes Exact One-Way Network Propagation Delay (d) and Clock Offset (theta):
   Mean_Path_Delay (d) = ( (t_2 - t_1) + (t_4 - t_3) ) / 2
   Clock_Offset (theta) = ( (t_2 - t_1) - (t_4 - t_3) ) / 2

By exchanging these four timestamps multiple times per second (and utilizing Boundary Clock or Transparent Clock switches that measure and subtract the exact nanosecond residence time each PTP packet spent sitting inside the switch fabric), every single Dante and AES67 receiver synchronizes its local hardware Voltage-Controlled Crystal Oscillator (VCXO) to the Grandmaster’s absolute wall-clock time within less than $1\text{ microsecond}$ ($< 100\text{ nanoseconds}$ with PTP-aware switches) across an entire stadium!

Each outgoing audio RTP packet is stamped with its exact absolute presentation timestamp. When the packet arrives at the receiving DAC, it sits inside a small FIFO playout buffer until the receiver’s local PTP-synchronized VCXO ticks to that exact scheduled nanosecond—resulting in sub-picosecond local converter jitter that is completely immune to Ethernet switch packet delay variation!

To guarantee glitch-free audio transport over standard enterprise switches (such as Cisco Catalyst, Netgear M4250/M4300, or Luminex GigaCore) when control data or file transfers share the same switch fabric, engineers configure two mandatory Layer-2/Layer-3 switch rules:

Rule #2 — Differentiated Services (DiffServ) DSCP QoS Queues: Managed switches inspect the 6-bit DSCP field inside every IPv4 packet header and sort incoming packets into 4 or 8 strict-priority hardware egress queues:

• Queue 4 (Highest Strict Priority — DSCP EF / Value 46 or CS7 / Value 56): Assigned exclusively to IEEE 1588 PTP Clock Sync Packets. Even if a port is $100\%$ saturated with file downloads, a PTP clock packet jumps straight to the front of the line with zero queuing delay.
• Queue 3 (High Priority — DSCP AF41 / Value 34 or EF / Value 46 in AES67): Assigned to Real-Time Audio RTP Packets, guaranteeing deterministic sub-millisecond delivery before any general network traffic.
• Queue 1 (Best Effort — DSCP 0): Assigned to Dante Controller UI discovery, web browsers, and bulk file transfers.

In mission-critical live broadcast (such as the Super Bowl, Olympics, or Broadway theater), a single severed Cat6 cable or failed switch power supply cannot be allowed to interrupt the audio show. AoIP interfaces achieve zero-downtime hardware fault tolerance by offering two physical Ethernet ports (Primary and Secondary):

1. Switched Mode (Daisy-Chain): Connects the two ports to an internal 2-port Ethernet switch inside the audio unit, allowing small studio setups to daisy-chain interfaces without an external switch (offers zero redundancy).
2. Redundant Mode (Dual Air-Gapped Networks & SMPTE ST 2022-7): Connects the Primary port to an entirely separate physical Ethernet switch fabric (Red VLAN/Switch) and the Secondary port to an independent parallel switch fabric (Blue VLAN/Switch). The transmitter sends identical, sample-locked audio streams simultaneously down both networks. The receiving interface buffers both streams in parallel; if a forklift slices the Primary fiber optic trunk in half, SMPTE ST 2022-7 Seamless Protection Switching continues reading uninterrupted samples from the Secondary port buffer with zero dropped samples and zero audible click!