AV-over-IP Architecture: What Is Actually Moving?

Why an AV engineer should care

An AVoIP drawing often shows encoders, decoders and a cloud labeled NETWORK. That is approximately as useful as labeling an electrical riser ELECTRICITY HAPPENS HERE.

To design or troubleshoot the system, you need to know:

• What traffic each endpoint generates
• Whether it is unicast or multicast
• Which links carry each stream
• What is media versus control or management
• Where bandwidth can become oversubscribed
• What happens after a link or switch fails

Learning objectives

By the end of this lesson, you should be able to:

• Describe how an AV signal becomes network traffic.
• Distinguish unicast from multicast distribution.
• Separate media, control, management and timing functions.
• Calculate bandwidth at endpoint and uplink levels.
• Read an AVoIP topology as a set of traffic paths rather than a collection of boxes.

---

1. From HDMI to Ethernet—and back

A basic AVoIP path looks like this:

1. A source produces HDMI, SDI, audio or another baseband signal.
2. An encoder captures the signal.
3. It may compress the audio or video.
4. It divides the resulting data into small payloads.
5. Protocol headers are added.
6. The endpoint transmits Ethernet frames through the network.
7. A decoder receives the packets.
8. It buffers, reorders and possibly decompresses them.
9. It reconstructs an HDMI, SDI, audio or USB output.

The switch does not understand that the packets contain a football game, camera image or Dante channel. It principally acts on Ethernet and IP information:

• Source and destination MAC addresses
• VLAN tag
• Source and destination IP addresses
• Multicast group
• QoS markings
• Transport protocol and port, when relevant to features such as ACLs

The encoder and decoder understand the media. The network moves packets according to addressing and policy.

“Stream” and “flow”

Manufacturers use these words inconsistently.

For our purposes:

• A stream is one logical media transmission, such as an encoded video source.
• A flow is a particular packet transmission from a sender, potentially containing one or more channels.

Dante, for example, can place multiple audio channels into one network flow. Other platforms may treat every video source as its own stream and multicast group.

Always check the manufacturer’s definition before calculating a system from the word alone.

---

2. Compression determines the network

An encoder does not automatically make video small. Its codec and operating mode determine the data rate.

Uncompressed

Uncompressed video preserves the original samples without a compression process. Quality and latency can be excellent, but the bandwidth is substantial.

A rough active-pixel calculation for 1080p60, 10-bit, 4:2:2 video is:

1920 × 1080 × 60 × 20 ≈ 2.49 Gb/s

Packet headers and other transport requirements increase the carried rate. Uncompressed 4K60 is roughly four times that pixel payload.

This is why uncompressed professional-video systems commonly require 10, 25 or faster Ethernet.

SMPTE ST 2110 can transport video, audio and ancillary data as separate, accurately timed streams. ST 2110-20 defines uncompressed video, while ST 2110-30 covers PCM audio by reference to AES67. (smpte.org)

Light or visually lossless compression

JPEG XS, JPEG 2000 and proprietary low-latency codecs can reduce video sufficiently for 1 or 10 Gb/s networks while preserving high image quality.

Depending on the product and format, a 1 Gb/s encoder might generate several hundred megabits per second. Never use the Ethernet port speed as the assumed stream rate; obtain the manufacturer’s maximum media bitrate for the actual resolution and mode.

Heavier compression

H.264, H.265 and similar codecs can reduce video to tens of megabits per second—or less—but normally introduce more encoding delay and can visibly degrade difficult material.

These are useful for:

• Signage
• Recording
• Remote contribution
• Streaming
• Large-scale distribution where ultra-low latency is unnecessary

Codec selection is therefore an architectural choice involving image quality, latency, bandwidth and interoperability.

---

3. Unicast versus multicast

Unicast: one sender, one receiver

A unicast transmission has a specific destination.

If an 800 Mb/s encoder sends independently to six decoders, it may need to transmit six copies:

800 Mb/s × 6 = 4.8 Gb/s

That is impossible through a 1 Gb/s encoder port.

Unicast is simple and often sensible when only one or a few receivers need a source.

Multicast: one sender, a receiver group

A multicast sender transmits one copy addressed to a multicast group. The network replicates it where paths branch toward interested receivers.

With properly operating IGMP snooping:

• The encoder sends one 800 Mb/s stream.
• Each link that needs the group carries one copy.
• A switch replicates the stream onto receiver-facing ports.
• Adding receivers on an existing branch does not require additional copies from the encoder.

This is the fundamental scaling advantage of multicast.

But multicast is not magically free:

• Every receiver port still carries the complete stream.
• Each uplink carrying the group consumes its full bitrate.
• Separate downstream branches each need a copy.
• Without IGMP snooping, switches may flood the traffic.
• Without a working querier, membership state can age out or behave unpredictably.

Audinate similarly distinguishes Dante unicast and multicast flows: multicast does not require additional sender bandwidth as receivers are added, but it can consume bandwidth across links where it propagates. (dev.audinate.com, dev.audinate.com)

Important distinction: switching versus routing

An AV operator may say, “Route Encoder 4 to Decoder 12.”

That does not necessarily mean an IP router is involved. The control system may merely tell Decoder 12 to leave one multicast group and join another. Ethernet switches then change where the multicast frames are forwarded.

AV routing and Layer 3 IP routing are related words describing different operations.

---

4. The functional planes

These are useful design categories, even when a product combines them on one port or VLAN.

The planes do not automatically require four VLANs. They may share one network, or a product may allow some to be separated.

The documentation should nevertheless identify each function because they have different requirements:

• Media needs bandwidth and consistent delivery.
• Control must reach the correct devices and servers.
• Management needs secure administrator access.
• Timing needs predictable behavior and the correct PTP domain.

ST 2110 intentionally separates elementary video, audio and data streams, then uses PTP to keep them aligned. (smpte.org) A system can therefore have perfectly healthy media links but fail because the control server is unreachable—or have functional control pages while packet loss destroys the video.

---

5. Bandwidth exists at several places

Do not calculate only the total number of encoders. Calculate every potential bottleneck.

Endpoint port

Can the encoder’s port carry its maximum output? Can a decoder receive all required simultaneous streams?

Access-switch capacity

How many sources enter the switch? How many receiver copies leave it? Switches normally advertise large switching capacities, but confirm the actual model and installed modules.

Uplinks

Suppose ten encoders each generate 850 Mb/s:

10 × 850 Mb/s = 8.5 Gb/s

A 10 Gb/s uplink may technically pass that traffic, but it leaves limited room for packet overhead, bursts, other services or future additions.

If that link fails over to a 1 Gb/s backup, the topology is redundant only in the most decorative sense.

Core, stack and ring paths

Determine the worst credible traffic direction after:

• An uplink failure
• A stack-member failure
• A spanning-tree topology change
• A ring operating through its remaining side
• Multiple receivers selecting different sources

The failure path must carry the required load, not merely establish link lights.

Link aggregation

Two 10 Gb/s links in a LAG can provide 20 Gb/s of aggregate capacity, but a single stream is normally assigned to one physical member according to a hash.

A LAG does not generally turn one 12 Gb/s stream into two neat 6 Gb/s halves. It distributes different conversations across members, sometimes unevenly.

---

6. Commercial-AV examples

Example 1: Visionary Solutions Duet with VLAN tagging

The Duet encoder scenario we discussed is a good example of multiple logical functions sharing one physical switch connection.

With its VLAN features configured, the endpoint can place different traffic onto separate tagged VLANs—for example:

• PacketAV video
• Dante/AES67 audio
• The expansion Ethernet port

The connected switch port must be configured as a tagged member of the corresponding VLANs using matching VLAN IDs.

With tagging disabled, those functions occupy one untagged network. That can prevent Dante from living on BMT’s preferred audio VLAN and can place a device connected to the expansion port directly into the video network.

This is not merely organization. An exposed expansion port could give a rogue or accidental device access to the AVoIP VLAN.

The drawing should show the endpoint as a tag-aware multi-service endpoint, not as an ordinary access-port device.

Example 2: Compressed-video access switches and a Ruckus core

Imagine eight encoders producing 750 Mb/s each on one ICX access switch:

8 × 750 Mb/s = 6 Gb/s

A 10 Gb/s uplink to the core is reasonable for the present load.

Twelve decoders on another access switch can all watch one source. With multicast, its uplink carries approximately 750 Mb/s—not 9 Gb/s—because the receiving switch replicates the stream locally.

But if the twelve decoders select eight different sources, that receiver-switch uplink may carry all eight streams: approximately 6 Gb/s.

Receiver count alone does not determine uplink usage. The important value is the number of distinct streams required across that link.

---

BMT standards takeaway

Every AVoIP riser should identify:

• Endpoint link speed
• Maximum stream bitrate and codec mode
• Unicast or multicast behavior
• Multicast group strategy
• VLAN for media, control, management and timing
• Normal and failure-path uplink loads
• IGMP snooping and querier location
• PTP domain and clock path
• LAG membership and limitations
• USB traffic, when transported
• Required bandwidth headroom
• Who owns each network configuration

The cloud can remain on the overview drawing—but the engineering sheets must explain what happens inside it.

Practical exercise

A system has six 800 Mb/s encoders on Switch A and twelve decoders on Switch B. The switches have a 10 Gb/s connection.

Calculate the A-to-B traffic when:

1. All twelve decoders watch Encoder 1 using multicast.
2. The decoders collectively watch all six encoders using multicast.
3. Encoder 1 sends twelve independent unicast copies.
4. The 10 Gb/s link fails over to 1 Gb/s.

Then identify which condition cannot operate as designed.

Compact glossary

• Codec: Method used to encode and possibly compress media.
• Stream: Logical media transmission.
• Unicast: One sender to one destination.
• Multicast: One sender to a receiver group.
• Plane: Conceptual category of system traffic.
• RTP: Real-time Transport Protocol.
• Oversubscription: Potential traffic exceeds available link capacity.
• LAG: Link Aggregation Group.
• Headroom: Unused capacity reserved for overhead, bursts and growth.

Review questions

1. Why does multicast reduce sender bandwidth but not receiver-port bandwidth?
2. What determines the load on an uplink serving multicast receivers?
3. Why can a two-link LAG still fail to carry one exceptionally large stream?

Answers

1. The sender transmits one group-addressed copy, but every receiver must still receive the complete stream.
2. The combined bitrate of the distinct multicast groups needed across that uplink—not simply the number of receivers.
3. A single flow is normally hashed onto one member rather than divided packet-by-packet across both links.

Where we are headed

We now have the overall AVoIP architecture. Lesson 20 will dissect Dante and AES67: channels versus flows, unicast and multicast subscriptions, PTP clock election, latency settings, primary/secondary networks and what AES67 interoperability actually does—and does not—guarantee.