Broadcast is where I learned this trade, and outside broadcast is still the hardest test I know for a network. A live feed does not care that the stadium cell site is saturated or that the truck is parked behind a concrete stand. If the link falters, you are off air in front of an audience, and everyone from the gallery to the client knows it. Cellular bonding is how we take inherently unreliable mobile connections and turn them into something a production can lean on.
Bonding is not load balancing
The first thing I explain on every broadcast job is the difference between load balancing and bonding, because the two get conflated constantly and the distinction decides whether your stream survives a carrier drop.
Load balancing distributes traffic per session. That is fine for an office full of people browsing, but a live video feed is one long session tied to one link; if that link fails, the session fails with it. Bonding works at the packet level. A single video stream is split into packets and sprayed across every available modem simultaneously, then reassembled in sequence at the far end. To your encoder, four flaky cellular links look like one stable IP address. Lose a carrier mid-stream and the remaining links absorb the load; the session never resets.
A bonded OB feed has three parts: the field unit with multiple modems, the carriers in between, and a headend receiver in a data centre or cloud instance that reassembles the packets. The field router initiates the tunnel outbound, which is why the field side needs no static IPs or special SIMs; only the headend needs a fixed public address.
What SpeedFusion actually does for a live feed
SpeedFusion is Peplink's bonding protocol, and three of its mechanisms matter most in broadcast work.
WAN Smoothing duplicates packets across active links. When one carrier takes a latency spike, the copy that travelled a healthier path fills the gap in real time. Jitter, not raw bandwidth, is what usually ruins live video, and this is the tool that tames it.
Forward Error Correction adds parity data so the headend can reconstruct lost packets without asking for a retransmit. Retransmission is poison for live latency; FEC avoids it at the cost of some overhead.
Dynamic Weighted Bonding continuously measures throughput, latency and loss on every modem and shifts traffic toward the healthiest links. A tower that was clear at rig time will congest as the crowd arrives; the bond adapts in sub-second intervals rather than letting one degrading SIM drag the whole pipe down.
All of this costs bandwidth. Depending on how aggressively you run smoothing and FEC, budget roughly 15 to 20 percent overhead on top of your encoder bitrate. An HD contribution feed wanting 8 Mbps needs a bond comfortably above 10; a 4K feed at 25 to 40 Mbps needs headroom to match. Skimp on that margin and the protocol has nothing to work with when conditions turn.
Carrier diversity, or why four SIMs from one network is one SIM
The most common design mistake I see is a router stuffed with four SIMs from the same provider. It looks redundant. It is not. A regional outage or a congested mast takes all four down together, and you have built a single point of failure with extra steps.
We provision across EE, Vodafone, O2 and Three as standard, so a provider-level fault costs the bond some capacity rather than the whole feed. The same logic applies to 5G. A single 5G link can post impressive numbers on a quiet morning, but high-frequency signal penetrates poorly, and inside an older stadium built from concrete and steel it can vanish entirely. Bonding low-band 4G, which punches through crowds and structures, with high-band 5G capacity gives you both reach and speed. In my experience that mix, not any single technology, is what keeps a feed alive when thousands of phones light up around you.
Design decisions that decide the outcome
Headend placement. Cloud receivers spin up fast and sit close to your distribution, but watch the egress costs. On-premise appliances give you full control but need proper fibre and a static IP. Neither is inherently better; it depends where your gallery wants the hand-off.
The physical layer. More OB connectivity problems are antenna problems than router problems. Modems buried in a metal flight case are shielded from the very signal they need. We fit external high-gain MIMO antennas, run low-loss cabling, and think about placement relative to the radar-reflective clutter of a truck park. Every decibel matters at the edge of a cell.
Visibility. InControl 2 gives you live RSRP and RSRQ per modem, so an engineer can see a link degrading and act before it becomes a problem on air. For mobile units, the GPS overlay lets you correlate drops with locations and plan around blackspots on the next run. A feed you cannot see is a feed you cannot defend.
Where we fit in
Hardware gets you a box; configuration gets you a broadcast tool. The gap between the two is where most deployments fall short. We design and tune SpeedFusion around your encoder bitrates, your latency tolerances and the carrier conditions at your actual venues, and we build the monitoring so production staff get a simple status view while engineers get the granular telemetry. We also run hands-on training for broadcast engineers who want that capability in-house, because the best deployments are the ones your own team can defend under a live countdown.
The short version
Bond at the packet level, spread across carriers, budget the overhead, get the antennas right, and watch the telemetry. Do those five things and a cellular contribution feed stops being a gamble and becomes infrastructure. If you have an outside broadcast coming up and want a second pair of eyes on the connectivity design, get in touch for a brief scoping conversation.
Frequently asked questions
What is the difference between bonding and failover?
Failover switches to a backup link after the primary has failed, which resets the session and drops the stream. Bonding distributes packets across all links simultaneously, so losing one carrier reduces capacity without interrupting the feed.
Can I bond 4G and 5G together?
Yes. SpeedFusion is agnostic to the radio technology underneath; it weights links on measured health. Mixing low-band 4G penetration with 5G capacity is standard practice on UK deployments.
How much bandwidth does bonding overhead consume?
Typically 15 to 20 percent for tunnel headers and encryption, more if you run aggressive WAN Smoothing or FEC. We size the bond so usable throughput comfortably exceeds the encoder's peak bitrate.
Do I need static IPs in the field?
No. The field unit dials out to the headend's fixed address, so ordinary consumer SIMs work fine. Only the receiver side needs a static public IP.
Is bonded cellular reliable enough for 4K?
With enough links, proper antennas and sensible headroom above the 25 to 40 Mbps a 4K stream typically needs, yes. The engineering, not the concept, decides the result.
What happens when a SIM loses signal mid-broadcast?
Traffic is already flowing on the other links, so the bond simply loses that modem's contribution. Failover within the tunnel is typically sub-second and rarely visible to the viewer when the system has been sized with headroom.