SRT Latency Buffer Explained: How to Tune SRT for Cellular, Satellite, and Public Internet

SRT latency is not one number

SRT works because it gives packets time to arrive, recover, and play in order.

That time comes from the latency buffer.

Set the buffer too low, and packet loss turns into glitches. Set it too high, and the feed feels slow. The right value depends on the network, bitrate, route, encoder, decoder, and production need.

Do not copy a number from a forum. Measure the link.

What the latency buffer does

SRT sends live media over UDP and adds reliability above it. When packets go missing, the receiver asks for retransmission. The latency buffer gives SRT a short window to recover those packets before playback reaches that point in the stream.

Think of it as a deadline.

If a lost packet comes back before the deadline, the viewer sees clean video. If it comes back after the deadline, SRT drops it and moves on.

That behavior matters for live production. TCP waits for missing packets and can stall the stream. SRT uses a time budget and keeps moving.

Start with network type

Use these ranges as a starting point, not a rulebook.

These numbers only help after you test. A clean 5G link can beat bad hotel fiber. A venue network can look fine until doors open.

Tune for packet loss first

Packet loss drives buffer decisions more than raw bandwidth.

At 0.1 percent loss, a low buffer may work. At 1 percent, you need more room. At 3 percent, you may need to lower bitrate, increase latency, add bonding, or switch networks.

A common field pattern:

• 1080p60 at 12 Mbps
• Public venue internet
• 1 to 2 percent packet loss during crowd arrival
• 120 ms SRT latency

That setup will likely show hits. Move to 300 or 500 ms and the link may clean up. If it still breaks, lower bitrate before you blame SRT.

Bitrate and buffer work together

A higher bitrate pushes more packets through the same link. That gives packet loss more chances to hurt the picture.

Do not set bitrate at the measured upload speed.

If the site tests at 20 Mbps upstream, do not send a 20 Mbps stream. Start around 10 to 12 Mbps for the main feed and keep room for bursts, retransmission, control traffic, and other production tools.

If you need two feeds, budget both. Two 8 Mbps streams plus overhead can strain a 20 Mbps uplink when the network gets busy.

Cellular needs extra respect

Cellular links change constantly. Signal strength, tower load, handoff, weather, and crowd density can all hit the stream.

For a single cellular modem, start with a conservative bitrate and a larger buffer. For bonded cellular, test the bonding system under movement and crowd load. Bonding does not remove loss. It spreads risk across links.

Practical starting point:

• 720p60 or 1080p30 for rough cellular paths
• 4 to 8 Mbps
• 500 to 1,000 ms SRT latency
• Local recording at the camera or encoder

If the show needs 1080p60 at high bitrate, bring more network than you think you need.

Satellite changes the math

Satellite can carry excellent video, but delay and loss patterns differ from terrestrial links.

LEO satellite can work well for field contribution, but it can still show handoff behavior, weather effects, and congestion. Start with a larger buffer than you would use on fiber.

GEO satellite adds long round-trip time. Retransmission takes time. You may need a much larger latency setting, and you should test the exact route before the show.

Do not sell GEO satellite as sub-second contribution unless the full chain proves it.

Watch the right metrics

Video preview tells you what happened. Metrics tell you why.

Track:

• Packet loss
• RTT
• Jitter
• SRT retransmitted packets
• SRT dropped packets
• Encoder bitrate
• Encoder queue
• Decoder buffer
• CPU and GPU load

If retransmissions rise but dropped packets stay low, the buffer does its job. If dropped packets rise, increase latency, lower bitrate, or change the path.

Caller, listener, and rendezvous

SRT connection mode affects firewall behavior.

Listener waits for a connection. Caller initiates a connection. Rendezvous lets both sides attempt connection when NAT complicates the path.

In field production, caller mode often helps because the venue initiates outbound traffic to a known receiver. Many venue networks block inbound connections. Outbound connections face fewer roadblocks.

Do not leave this to show day. Confirm the mode, port, passphrase, and firewall rule in advance.

Common tuning mistakes

The "lowest latency wins" mistake:
Low latency only wins if the picture holds. A clean 400 ms feed beats a broken 120 ms feed.

The "speed test says we are fine" mistake:
A speed test does not prove low jitter or low packet loss. Test with the real encoder and real bitrate.

The "one stream worked" mistake:
One feed may work. Four feeds may fail. Test the full load.

The "no local record" mistake:
If the network dies, local recording saves post-production and highlight packages.

The "same settings everywhere" mistake:
Each venue needs its own profile. Save known-good settings per site.

A simple tuning workflow

1. Start with the target format and bitrate.
2. Set a conservative latency buffer.
3. Stream for at least 30 minutes.
4. Add real venue load if you can.
5. Watch packet recovery and dropped packets.
6. Lower latency until errors appear.
7. Move one step back.
8. Save the profile.

That process gives operators a real number, not a guess.

Where Medialooks fits

Medialooks Video Transport uses SRT for low-latency video transport and adds production-friendly connection handling for SDI and NDI workflows.

Medialooks Video SDK gives developers SRT input and output inside custom video applications, alongside NDI, WebRTC, RTMP, hardware I/O, recording, and processing.

SRT gives you control. Use it. Tune the buffer for the network in front of you.