How we got here

Nothing in this module was designed. It accumulated, one problem at a time, and every generation solved the previous one's limit and introduced a new one of its own.

This page is not examinable as history. It is here because knowing why something exists tells you what it refuses to do, and that is the knowledge that keeps working when the product names change.

One pattern runs through all of it:

Each step put more on one cable, and charged for it in something you now have to configure.

Every date, port number and multicast range on this page was checked against showstack, the open index of live entertainment technology, which carries a citation on each entry. Where this page summarises, showstack has the source.

Block 0: The chain with no numbers in it#

Before any of the rest of this, there was a complete working signal path that counted nothing. It is worth walking once, because everything that came after is a replacement for one link in it, and because most of the words on a modern desk were named here.

From the air to the wire, and back to the air#

A microphone turns pressure into voltage. A preamplifier makes that voltage bigger. A fader takes a fraction of it, a bus adds the fractions together, and a power amplifier makes the sum big enough to move a coil in a magnet, which moves a cone, which pushes air. Five stages, and the voltage at every one of them is a scale model of the pressure in the room.

The loudspeaker at the end is the microphone at the start, running backwards. Same three parts, same physics, arrows reversed. Connect a loudspeaker to a microphone input and shout into it and you get a signal out, which is exactly what a talkback speaker is doing.

A record, which is the only chain you will meet with no converter in it#

Students ask whether vinyl is really analogue all the way, and whether it works without a computer anywhere. Yes, and yes.

A shape cut into plastic shakes a diamond, which shakes a magnet next to a coil, which makes about five millivolts, which is equalised, amplified and used to move a cone. No clock, no sample rate, no number at any point. It is the same transducer as the microphone at both ends of the chain.

One honest footnote, because it is the thing people get wrong in both directions: the playback path being analogue tells you nothing about the recording. Most records pressed since the 1980s passed through a digital master on the way to the lathe. The chain in your living room is genuinely analogue; the chain that made the disc usually was not.

Both eras, side by side#

Lighting, video and sound each made the same move at different times, and in each case the change was not to the job but to what the wire was carrying. An analogue wire carries a quantity, so it degrades a little at a time and nothing reports it. A digital wire carries a count, so it is exact until it is not.

This is the single most useful thing to take from the whole page: the two eras fail differently, and knowing which failure you are looking at is most of fixing it. Softness, hum and drift are analogue symptoms. Sparkles, dropouts and a device that ignores you are digital ones.

Block 1: Audio, from a voltage to a network#

Audio started as the thing it is: a voltage that looks like the sound. Everything since has been an argument about how to get more of those down fewer cables without the noise.

The three moments that matter.

AES3 made it digital but not networked. Two channels on one pair, self clocking, point to point. Worth noticing that it carries its own timing in the transitions, exactly like the LTC timecode in Class 4: the same trick, twice, for the same reason.

CobraNet proved audio over Ethernet, and showed the bill. A dedicated layer 2 network that could not be routed, a latency you took rather than chose. It worked, and everything since has been an argument about how much of that restriction to give back.

Dante won by being ordinary. Routable IP, standard managed switches, latency you set. What it charges you is the clock: your system now depends on a PTP election you did not make and cannot see. That is why Class 5 spends so long on grandmasters and PTP aware switches, and it is the direct cost of the convenience.

AVB took the opposite bet. Rather than asking the network to behave, it makes the switches promise bandwidth in advance. Better engineering, and it requires every switch in the path to support it, which is why it kept the systems designed around it and lost the general market. Milan is its professional audio profile, so devices from different manufacturers interoperate.

Ravenna took the open route. Built on published standards (RTP, PTPv2) rather than one company's stack, which is why it feels less automatic than Dante and why it slots into a broadcast plant more comfortably. Dante discovers; Ravenna is configured. AES67 later formalised almost exactly what Ravenna was already doing, which is why the two get along so easily.

AES67 is not a competitor. It is the peace treaty: a narrow set of rules (RTP, PTPv2, 48 kHz, 1 ms packets) that lets Dante, Ravenna, Q-LAN and Livewire exchange audio. You use it at a boundary, because switching a device into AES67 mode usually means giving up whatever its own ecosystem provides. ST 2110-30, the audio part of the broadcast video standard, is AES67.

Block 2: Lighting, from one wire per dimmer to the network#

DMX512 is nearly forty years old and is still the last metre of almost every rig. It survives because it is dumb: no addressing, no negotiation, no state, no return path. A fixture reads the slots at its address and obeys. Everything after it is a way of carrying DMX, not of replacing it, which is why the universe is still the unit even on a network that has no idea what a universe is.

Then it ran out of room twice.

First, the number of universes. Art-Net (1998) put universes in UDP packets years before there was a standard, it was free to implement, and it is still the default on a great deal of kit. Two weaknesses matter on a large rig: it broadcast by default, which floods a network, and it has no priority field, so two senders on one universe is settled by whichever packet arrived last.

Then, the way they were carried. sACN (E1.31, 2009) is the ANSI standard: multicast by default, so a switch with IGMP snooping delivers each universe only to the ports that asked; a priority field from 0 to 200, so a backup console is a configuration rather than a gamble; and synchronisation packets so universes change together. Both of its advantages depend on the network being set up properly, which is exactly why Class 3 exists.

The legacy trap worth repeating. Art-Net traditionally used 2.x.x.x, which is real public address space belonging to somebody else. Isolated, it works and nobody notices. Connected to a venue's internet, traffic for real hosts disappears into your rig. Use 10.x.x.x for new work.

Block 3: Video connectors, one number getting bigger#

The whole connector story is a single number growing, because somebody kept wanting more pixels, more frames or more bits than the previous cable could carry.

Three things worth carrying away.

Digital did not make it better, it made it different. VGA degrades gradually: a long run gets soft and you can still work. Digital links are fine until they are not, and then there is no picture at all. That cliff is why an HDMI run that "works on the bench" is not a result.

DisplayPort sends packets, HDMI sends a stream. That is why DisplayPort can carry several independent displays down one cable and one port on a card can drive a wall, and it is why the GPU comparison in Class 2 is really a DisplayPort conversation.

SDI persists because it refuses to negotiate. No handshake, no EDID, no HDCP, no content protection to fail at the worst moment. It is a locking BNC that either carries a picture or does not, over long runs, and that predictability is why broadcast never left.

The connector table changes every few years. The arithmetic does not: pixels × lines × frames × bits, then compare it to what the cable carries. Learn the arithmetic.

Block 3b: Video on a network, and why there are three answers#

This one is not a progression, and treating it as one is the mistake. NDI did not replace ST 2022-6 and ST 2110 did not replace NDI. They are three different trades between bandwidth, latency, and how much network you had to build first.

Read the latency column, not the bandwidth column. Bandwidth is a purchasing decision and latency is an artistic one. A frame at 25 fps is 40 ms, which is where lip sync starts to show, so a feed that goes to a screen beside a live performer is a different problem from a feed that goes to a director's monitor.

NDI won the middle of the market by asking for nothing. Plug it into the network you already have and devices find each other. That is genuinely valuable, and it is also why it turns up on show critical paths where nobody designed the network, which is the one place it does not belong. Know which of those two situations you are in.

ST 2110's separated essences is the idea to steal. Video, audio and ancillary data travel as separate multicast flows, so a device subscribes to what it needs: a timecode reader takes 40 kbit/s instead of 2.5 Gbit/s. The cost is that the parts can now arrive separately, so PTP stops being infrastructure and becomes the thing the picture is made of. And note where the two families finally meet: ST 2110-30, the audio part, is AES67. The audio lineage and the video lineage end in the same place.

Block 4: Parallel and serial, and why serial won everything#

Sending eight bits at once down eight wires is obviously faster than sending them one after another. It was obviously faster for about twenty years, and then it stopped.

Skew is the reason. Every conductor is a slightly different length with slightly different characteristics, so its bit arrives slightly early or late. That difference is fixed by the cable. As speed rises the bit period shrinks and the skew does not, so eventually the eight bits no longer fit inside one sampling window and the byte cannot be reassembled. Making the cable faster makes the problem worse.

A single line has nothing to agree with, so it can be driven far faster than any of the eight individually, and eight times a very high rate beats eight lines at a low one.

Every one of those is the same story. This is also why DMX is serial, at 250,000 bits per second down one pair, and why AES3, SDI, HDMI and Ethernet all are. There is no fast parallel interface left on any machine you will use.

Block 5: Storage, and how the same space kept holding more#

What is physically on the thing#

Every medium in that list solves the same problem: make a room full of sound survive being put down and picked up again. Two of them keep a shape, and three of them keep a count, and that single difference decides what damage does to each one.

A scratch on a record is not a lost bit; it is a click that is now part of the music, permanently. A scratch on a CD is a wrong number, and error correction can often work the right one out again, which is why a damaged disc plays perfectly and then stops dead rather than getting gradually worse. That is the same distinction as the two eras in Block 0, arriving in a different place.

The optical line is one rule applied three times. Capacity is set by how small a mark you can reliably read, and that is set by your wavelength. 780 nm infrared for CD, 650 nm red for DVD, 405 nm blue violet for Blu-ray, and the pits shrink each time. Layers, read by refocusing deeper into the same disc, multiply it again. The physical object never changed.

Blu-ray ended the line, and not because the physics stopped: networks got faster than discs. The format did not fail, the problem it solved stopped being a problem, which is the most common way a technology in this industry dies.

Flash changed the question. Density stopped coming from making the mark smaller and started coming from stacking cells upward (3D NAND, now over two hundred layers) and from storing several bits in one cell, which costs endurance. And once nothing had to move, the interface became the bottleneck: SATA was designed around a device that seeks, so NVMe exists because flash does not.

Why this matters on a show#

The number that ends the argument is not throughput, it is what happens to throughput when the work gets awkward.

A hard disk reading one file in order is perfectly respectable, and that is the figure on the box. A media server playing six layers is not reading one file in order. Judge a show drive by what it does under the awkward case, because the awkward case is the job.

Block 6: What is coming, and what to believe about it#

Two honest categories, and the difference between them matters more than either.

Already here: dedicated compute for pattern work#

Machine learning hardware, sold as an NPU or a tensor unit, is a processor built for one shape of arithmetic: many small multiplications in parallel. In production tools it is already doing unglamorous, useful jobs:

• Upscaling and frame interpolation, so an HD plate can fill a 4K wall better than a plain scaler manages
• Denoise and cleanup on camera feeds, especially at high gain
• Keying and rotoscoping without a green screen, which changes what is possible on a fast turnaround
• Deciding where the bits go inside an encoder, which is the codec quantisation argument from Class 2 with a better guess about what the eye will notice
• Speech to text for live captioning and surtitles

The pattern is worth naming: it is very good at judgement calls that used to need a person and a lot of time, and it is not deterministic. A show wants deterministic. So the sound place for it today is in preparation and in things that degrade gracefully, and not yet in the signal path where a wrong guess is a visible failure in front of an audience.

Quantum teleportation, and what it actually does#

This one is worth doing properly, because the popular version of it is wrong in a specific way you can check, and because "we will teleport the signal" is a sentence somebody will say to you.

Quantum teleportation is real. It has been demonstrated in laboratories since 1997, over 1,400 km between a ground station and a satellite, and in 2024 over a fibre that was carrying ordinary internet traffic at the same time. None of that is marketing. What it does is transfer the quantum state of one particle onto another particle somewhere else, and the original state is destroyed in the process, because quantum states cannot be copied.

Now the part that gets left out.

The protocol needs two things: a pair of entangled particles shared in advance, and a classical message of two ordinary bits, sent from the sending end to the receiving end, describing the result of a measurement. Without those two bits the receiver holds a particle in a state indistinguishable from noise. It cannot do anything with it, and it cannot tell that anything has happened.

Those two bits travel down an ordinary channel, at the speed of light or slower. So teleportation cannot deliver anything before light could have delivered it. Not "not yet". Not "until the engineering improves". The classical channel is part of the protocol, and it is the bottleneck by construction.

So what is it for? Networking quantum computers to each other, and extending quantum key distribution over long distances through repeaters. QKD is also real, and what it offers is detecting whether somebody has listened to a link, not sending data faster. Both are serious research with serious money behind them. Neither one moves a video frame across a venue any sooner, and neither is on a timescale that changes how you specify a show.

The test to apply. If somebody tells you a new technology will beat the latency of light over the same path, they are either describing a shorter path, a faster medium, or nothing.

The floor nobody moves#

Which brings us to the number that actually governs long distance transmission, and always will.

Light in a glass fibre travels at about two thirds of its vacuum speed, which works out at roughly 5 microseconds per kilometre, or 5 ms per thousand kilometres, each way.

Read the last row against the lip sync threshold. A round trip to London is more than two frames of video before any equipment is switched on. That is why remote production is designed around not asking a question and waiting for the answer, and why a remote director's talkback feels different from a local one no matter what is spent on it.

What will genuinely change transmission in your working life#

Four things, all unglamorous, all already shipping:

• Hollow core fibre. Light travels down air in the middle of the fibre rather than through glass, at about 99.7 percent of vacuum speed instead of 68. That is roughly a third less latency on the same route, and it is deployed today on financial trading links because a third of 48 ms is worth real money. It is the only item on this list that touches the physics.
• Deterministic networking. TSN, the IEEE family that grew out of AVB, brings reserved bandwidth and guaranteed delivery windows to ordinary Ethernet. This is the AVB argument from Block 1 winning the general case at last, and it matters far more to a show than any exotic physics.
• Better compression, not less of it. JPEG XS at a few lines of latency, AV1, and encoders that use machine learning to decide where the bits go. All of this buys distance and cost, and pays for it in the same currency as always.
• More lanes. 400 and 800 gigabit Ethernet, and private 5G with network slicing for wireless links that need a guarantee. More bandwidth is not more speed, and confusing the two is the most common mistake in this whole area: a wider road does not shorten the journey.

The test to apply to any new technology in this industry is the same one as everywhere else in this module: what problem did it solve, and what did it charge you for solving it? If a product cannot answer the second half, nobody has finished thinking about it yet.