Wi-Fi or Ethernet: the network behind a live broadcast

The venue's fibre is not your connection: it is three hundred people checking their email. What is actually guaranteed to you, and what happens when it stops holding.

The network is the one link in a live production nobody discusses during prep and everybody discusses during the incident. Cameras get debated, audio gets debated, the encoder gets debated, and then somebody plugs the encoder into the nearest wall socket on the assumption that the venue has fibre.

The venue's fibre is not your connection. It is three hundred people checking their email. The question is never whether the building is connected, but what is actually guaranteed to you, over what length of cable, and what happens when the guarantee stops holding.

How much do you actually need

The starting point is not the venue's capacity but what the destination platform requires, and the spread between platforms is wider than most people expect.

YouTube publishes its H.264 tiers: 10 Mb/s for 1080p at 30 frames per second, 12 Mb/s at 60, 30 Mb/s for 4K at 30 and 35 Mb/s at 60. It also notes that the low-latency option is unavailable at 2160p, which settles a trade-off before it can even be argued.

LinkedIn, in its integrator documentation, caps things far lower: 1080p maximum, 30 frames per second maximum, 6 Mb/s of video, AAC audio at 128 kb/s and 48 kHz, a keyframe every two seconds, H.264 over RTMP, and a four-hour ceiling. There is no 4K on LinkedIn, and pushing a heavier stream at it accomplishes nothing.

Vimeo takes a different approach, indexing recommended quality to measured upload speed: below 4 Mb/s, 720p at 2,000 kb/s; between 10 and 25 Mb/s, 1080p at 4,000 kb/s; above 25 Mb/s, 1080p at 5,000 kb/s. For 4K delivery it asks for 25 to 40 Mb/s of upload.

Twitch no longer publishes its recommendations on a page that can be reliably consulted, so we will not quote figures for it rather than repeat unverifiable numbers.

Keep the essential distinction in view: these values describe the stream, not the connection carrying it. Confusing the two is the founding error behind most live incidents.

What a cable guarantees, and over what distance

A copper link is standardised at one hundred metres of channel, made up of ninety metres of permanent link and ten metres of patch cords. Past that, nothing fails abruptly: performance degrades, errors accumulate, and the problem becomes hard to diagnose precisely because it is gradual.

Category decides the rate sustained over that distance. Cat 5e carries a gigabit per second over the full hundred metres. Unshielded Cat 6, by contrast, only guarantees ten gigabits over roughly fifty-five metres, while Cat 6A holds ten gigabits across the whole run.

For a live production the conclusion is reassuring: a gigabit covers every case above, including 4K with redundancy. The cable is never the limiting factor in a broadcast. The internet connection, and the fact that you share it, are.

Why Wi-Fi cannot be tested in an empty room

Wi-Fi's weakness in live production is not throughput, which is often excellent. It is the nature of the medium: spectrum is shared and, as a matter of regulation, unprotected.

In the European Union, the 5945–6425 MHz band was opened to wireless local networks on a non-exclusive basis, without interference protection — the French implementing decision of October 2021 spells this out in exactly those terms. Read the phrase slowly: there is no regulatory guarantee that the medium will be available. If somebody transmits next to you, you have no recourse.

The same framework separates two power regimes: low-power indoor access points capped at 23 dBm EIRP and restricted to indoor use, and very low power equipment capped at 14 dBm, permitted indoors and outdoors. National implementations differ in detail — the exact limits and the list of prohibited contexts are worth checking for your own country — but the principle of an unprotected shared band is common to all of them, and it rules out a number of outdoor event configurations.

Wi-Fi 7 brings genuine improvements, notably multi-link operation and 320 MHz channels wherever the 6 GHz band is available, which raise throughput and lower average latency. None of that changes the shared nature of the medium, and no official source documents a throughput guarantee for an associated client.

Hence the practical rule: a speed test run the evening before in an empty room does not measure what you will have tomorrow with three hundred associated devices. It measures the best case, not the nominal one. You will also encounter claims along the lines of "beyond N clients throughput drops by X per cent" — none of them is backed by a primary source, and we will not repeat them.

Cellular bonding, and what nobody publishes

When wired is impossible or untrustworthy, cellular bonding is the industry's answer. The principle is documented by the manufacturers: distribution happens at packet level, which lets a single session use the combined bandwidth of every available link, whether cellular, wired, Wi-Fi or satellite.

The supporting mechanisms are documented too: retransmission of lost packets, adaptive forward error correction, dynamic adjustment of encoding bitrate to network conditions, and immediate failover of traffic to healthy links when one fails, without dropping the session.

Two numbers are missing, and it is better to say so than to invent them. No manufacturer publishes the latency added by the bonding buffer, even though that trade-off between margin and delay sits at the heart of the technology. None publishes a maximum number of modems presented as a general figure either. Both are values to measure on your own configuration, not to quote.

Separating production from guests

The principle is uncontroversial and stated officially: Vimeo recommends a wired connection on a dedicated, non-shared network, and setting the encoder's output rate well below the available upload speed to absorb fluctuation.

What does not exist is an official numeric margin. The rules you hear — double it, keep fifty per cent — appear in no platform documentation. Vimeo's published tiers do imply a ratio well below half of available upload, but that ratio is never stated as a rule, and presenting it as one would be extrapolation.

VLAN isolation and traffic prioritisation follow the same logic: documented good practice from equipment vendors, with no normative reservation threshold. The reasoning is simple — your stream must never depend on the audience's behaviour — and it is implemented by asking the venue for a distinct connection or VLAN, ideally with its own uplink.

Measuring properly

A consumer speed test measures what a browser obtains from a commercial server at one moment. To prepare a broadcast you want a measurement between two points you control, which is what iperf3 provides.

The principle is two commands: a server at one end, a client at the other. By default the client transmits toward the server, which measures exactly the direction that matters — upload; a dedicated flag reverses it. Another opens several parallel streams, which reflects a real delivery chain better than a single one.

The choice between TCP and UDP changes the nature of the measurement, and it is worth understanding. TCP measures the application throughput actually achievable, which is the right question for sizing an encoder. UDP generates a constant-rate stream that the documentation itself calls very artificial, but which has one merit: it exposes packet loss and jitter, which is precisely what breaks a broadcast on a connection whose raw throughput was perfectly adequate. Watch one trap: in UDP mode the default target bitrate is one megabit per second, and forgetting to set it produces a measurement unrelated to reality.

The configuration that holds

All of the above points to a reference setup, to be adapted but not worked around. A dedicated wired connection obtained from the venue and not shared with the public, run in Cat 5e or better within the hundred-metre limit, through a managed switch that isolates production traffic. An encoder with two network interfaces, so that failover does not require a human. Cellular bonding as backup, with SIMs from at least two different operators, because two SIMs on the same network fail together. A UPS covering the entire network path, including the switch and the venue's equipment, which everyone forgets in favour of the cameras. And a real upload measurement, taken with a tool that measures what you ask it to.

Wi-Fi appears in that list only as a last resort, for a broadcast with nothing riding on it, and in full awareness that it cannot be verified in advance.

Frequently asked questions

How much upload do I need for a 1080p broadcast?

The stream itself wants between 4 and 12 Mb/s depending on platform and frame rate — LinkedIn caps at 6, YouTube recommends 10 to 12, Vimeo 4 to 5. The connection needs comfortable headroom above that figure, though no platform publishes an official margin.

Do I need Cat 6?

Not for a broadcast. Cat 5e carries a gigabit over a hundred metres, far beyond any contribution stream. Cat 6A earns its place for ten-gigabit runs, which is a different problem.

Can I stream over Wi-Fi if the room has no network socket?

Technically yes, but with no way to verify in advance what you will get on the day, because the degradation comes from the audience being present. As soon as anything is riding on it, cellular bonding is more predictable than shared Wi-Fi, precisely because it does not share a medium with the spectators.

Are two SIM cards enough as backup?

Only if they are on two different networks. Two subscriptions with the same operator share the same cell site and the same outages, and provide no real redundancy at a given venue.

Sources

TIA 568-series standards update (90 m + 10 m channel) and public IEEE 802.3 documents for category reach.
YouTube Help, Choose live encoder settings, bitrates, and resolutions; LinkedIn Live ingest documentation; Vimeo Help, network configuration for live events and 4K streaming.
French regulator decision 2021-2184 of 14 October 2021 on RLAN in the 6 GHz band, implementing EU Decision 2021/1067; Wi-Fi Alliance announcement of Wi-Fi CERTIFIED 7.
Vendor documentation on link bonding (Peplink SpeedFusion, Haivision SST, LiveU LRT).
Official iperf3 documentation.