Linking two studios over SRT

Two control rooms, two cities, one programme. The signal has to cross the public internet and come out clean enough to mix with local cameras — which means treating it as a contribution link, not a stream to viewers.

Decide the direction first

Contribution is asymmetric. Write down, before touching any settings:

• which studio originates the programme feed
• whether the far end needs a return feed (programme, or clean, or just talkback)
• who cuts the show, because that side needs the lowest delay

A one-way link is one SRT path. A return feed is a second, independent path in the other direction — not a magic "bidirectional" mode. Budget for both.

Who listens, who calls

For a permanent installation, a listener on your own hardware is worth the setup effort. For a link that follows the production around — different venues, different months — a relay costs less than the firewall tickets.

Bitrate: pick the number you can hold at 20:00

Contribution quality starts around 8 Mbit/s for 1080p25 with H.264, and 12–15 Mbit/s is comfortable for material with motion. Before committing:

1. Measure the upload at both ends during the hour the show airs, not at midday.
2. Take 60–70% of the worst figure you saw.
3. If that lands under 6 Mbit/s, drop to 720p rather than starving 1080p — the picture will look better.

Encode CBR. A variable-bitrate encoder makes SRT's buffer maths unpredictable, and the peaks are exactly when the link is already struggling.

Latency for a permanent link

Between cities on wired connections, expect 10–40 ms RTT within a country, 90–150 ms across a continent. Run the numbers in the calculator, then add margin for the worst hour rather than the average one.

Practical starting points:

The return feed

If the remote presenter needs to see the programme, send a second SRT path back. Keep it cheap: 2–4 Mbit/s at 720p is plenty for a confidence monitor, and the lower bitrate survives a congested uplink that the main feed is already using.

Talkback is a different problem: at 300 ms each way plus encoder and decoder, a conversation is already awkward. Either accept the delay and instruct people to leave gaps, or run a separate low-latency audio path — a phone line, a Cleanfeed session, anything that is not carrying video.

Redundancy that actually helps

• Second path on a different uplink. The main link on fibre, a backup on 4G into the same receiver on a different port. Cut between them in the switcher. This survives the single most common failure: someone's ISP.
• Relay in the middle, two receivers. One input, two outputs: the programme machine and a backup machine. The backup costs nothing to run and is already warm when you need it.
• Do not bother with a "backup" on the same cable. Two streams down one uplink fail together and halve your usable bitrate on the way.

Before the first show

1. Point the receiving studio at the free test signal and leave it running for an hour. If the frame counter never skips, the receiving path is clean.
2. Run the real link at full bitrate for the length of the show, with someone watching the receiving end. A link that survives ten minutes is not proof of anything — congestion arrives on a schedule.
3. Write the SRT URL, port, passphrase and latency on a card and tape it to both racks. At 3 a.m. nobody remembers whether it was 4012 or 4021.

When a leg dies mid-show

Almost every mid-show recovery is "restart the receiving input first" — it is faster, and it does not lose whatever the encoder is still delivering correctly.