Overview
Broadcast production is moving from SDI-based workflows to fully IP-based architectures. We supports this transition with SONiC-based switches for two typical environments: OB vans for mobile live production, and live studios built to SMPTE ST 2110.
Studio environments bring particular requirements: multi-camera synchronization, high-frequency real-time switching and deterministic signal distribution. This guide describes how an network is designed to meet them, layer by layer: timing, multicast control, redundancy and multi-site scaling.
Reference architectures
Both environments use a redundant Spine-Leaf fabric with Grandmaster timing. The OB van is a two-tier design. The studio adds a Sub-Spine layer for 25G production endpoints.
OB van | Live studio (SMPTE ST 2110) | |
|---|---|---|
Spine | 2 × CX732Q-N-V2 | 2 × CX532P-M-H (32-port 100GbE QSFP28) |
Aggregation | None: two-tier Spine-Leaf | CX308P-48Y-M-H (48 × 25GbE, 8 × 100G uplinks), with 100G links to the Spine |
Leaf | CX732Q-N-V2 in pairs, with 100G/400G links to the Spine and 25G/100G links to endpoints | CX206Y / CX204Y series, with 1G links to management, intercom and playback controller |
Where endpoints attach | Production endpoints connect to the Leaf | Production endpoints connect to the Sub-Spine at 25G |
Timing | Grandmaster clock(s) as the system-wide reference | Grandmaster clock(s) as the system-wide reference |
OB van. The machine room is the central hub of the production system. Video, audio and production equipment are integrated through modular segmentation, and Spine-Leaf interconnects all functional zones. The design is optimized for very limited vehicle space, reduces cabling and enables plug-and-play scalability. Leaf switches connect vision mixers, replay controllers, CCUs (which feed the cameras over SMPTE hybrid fiber), FPGA processing units, recording servers, the broadcast control system and monitoring servers (syslog, Grafana).
Studio. The Sub-Spine layer aggregates 25G production endpoints and provides low-latency transport for high-bandwidth video streams. A physically isolated 1G network carries configuration management, software upgrades, control signaling, intercom and tally traffic, separating control traffic from production video streams.
Typical endpoints
Device | Role |
|---|---|
CCU (Camera Control Unit) | Powers cameras and remotely controls color and exposure parameters |
Vision Mixer | Core of the production workflow: real-time video aggregation, transitions, keying and effects |
Replay Controller | Subscribes to specific multicast streams for slow-motion replay |
FPGA card / processing unit | Real-time audio/video encoding, decoding and format conversion |
Large screen controller / matrix | Distributes video to studio monitors or on-site displays |
Recording server | Stores broadcast-grade program content for post-production |
Broadcast Control System | Centralized control of PIM multicast flow direction, mapping director commands to physical signal paths |
Monitoring servers | Syslog and Grafana servers for network visibility |
Timing (PTP)
With SMPTE ST 2110, video, audio and ancillary data are separated into independent IP streams, each transmitted independently over the network. Network latency and jitter can cause these streams to become unsynchronized, so a unified time reference is needed. PTP (IEEE 1588) provides it, letting all devices share the same clock across the network.
Our PTP switches support:
IEEE 1588v2, the SMPTE ST 2059-2 profile, AES67 and ITU-T G.8275.x profiles
Grandmaster, Boundary Clock, Transparent Clock and Ordinary Clock roles
Class C PTP synchronization with accuracy below 10 ns
For M&E networks with many distributed endpoints, we recommend selecting devices with Boundary Clock functionality and enabling it across the entire IP fabric.
In both the OB van and the studio design, Grandmaster clocks provide the system-wide timing reference, and the spine switches use hardware-optimized forwarding paths for nanosecond-level PTP synchronization to every endpoint.
Engineering track record: port instability. In IP production networks, an unstable switch port can trigger cascading synchronization loss across the network, visible as video flicker and audio-video desynchronization. Our engineering team has resolved this class of problem through SDK-level optimization of the Marvell platform, eliminating abnormal synchronization resets caused by port flapping. A redundancy mechanism was also introduced to mitigate port flap events, with timing stabilized at tens-of-nanoseconds accuracy in the studio deployment.
Multicast control
Live production needs the network to follow the production. Asterfusion supports a customized PIM multicast routing model aligned with live production workflows.
Multicast routing state is directly programmable through a writable SONiC Redis control-plane interface. The Broadcast Control System drives multicast state updates at runtime, creating a joint control model between production logic and network behavior. Each switching action from the production control system is translated into an immediate multicast route recalculation, enabling frame-accurate signal switching based on dynamic PIM state updates.
Studio signal flow
Acquisition: cameras connect to the CCU over fiber. The CCU converts the raw camera feed into a standard IP stream (SMPTE ST 2110) and injects it into the network.
Production: the vision mixer retrieves multiple camera feeds from the switching fabric, performs real-time switching and effects compositing, and sends the program (PGM) stream back into the fabric.
Replay: replay servers subscribe to selected camera multicast streams for buffering. When key events occur, they generate slow-motion video, encapsulate it as an IP multicast stream and re-inject it into the network, where the switches distribute it to PGM outputs or monitoring devices.
Distribution: switches deliver PGM and replay streams, in coordination with the Broadcast Control System, to FPGA cards for compression, to display controllers and to recording servers.
Command delivery: the signaling network carries vision mixer control commands, CCU control traffic, intercom, tally and PTP timing signals over dedicated physical links.
Redundancy
The designs above are redundant at the fabric level:
The OB van uses a redundant dual-plane Spine-Leaf architecture with a dual-spine backbone.
The studio uses a dual-spine redundant backbone.
Scaling across sites
When production spans several locations, the platform extends beyond a single facility. A multi-site design uses a Leaf-Spine architecture with a BGP EVPN control plane and VXLAN multicast for media distribution. One example connects a primary and a backup data center with three edge campus sites, for geographic redundancy and lower latency.
Spine layer: high-performance switches such as the CX732Q-N-V2 serve as boundary clocks supporting Class C PTP (<10 ns) across multiple nodes.
Leaf layer: 1G and 10G devices handle burst traffic during live streaming and support bandwidth needs from standard office applications to high-bitrate 4K and 8K live video.
VXLAN multicast BUM forwarding: handles non-unicast traffic such as live stream multicast in the overlay, using the underlay to replicate packets and reducing repeated copies at the source VTEP.
Heterogeneous endpoints
FPGA cards and CCUs can show timing behavior that differs from standard NIC-based systems, which creates synchronization alignment challenges. Asterfusion addresses this through iterative co-debugging, protocol analysis and system-level adaptation, moving from a purely standardized approach to a scenario-driven engineering model, with on-site and remote support. This supports a smooth transition from SDI-based workflows to a fully IP-based production architecture.
Why open SONiC for broadcast
Asteraix combines an open SONiC platform with broadcast-specific engineering:
Programmable control plane: direct Redis access to multicast state, integrated with the customer's own broadcast control system.
SDK-level optimization: Marvell SDK tuning to improve clock synchronization stability.
Engineering support: remote collaboration and on-site support for rapid issue resolution.
