IP Video Infrastructure for Live Production and Contribution

How it works

Media transport and packet formats

Uncompressed professional video over IP is most commonly carried using the SMPTE ST 2110 suite, which separates video, audio, and ancillary data into discrete essence streams transported via RTP over UDP. ST 2110 packetises uncompressed YCbCr frames — typically 10-bit or 12-bit, sampled at 4:2:2 or 4:4:4 — at sustained line rates equivalent to their SDI counterparts. A 1080p50/59.94 signal requires approximately 3 Gbit/s, while a single-lane 4K stream demands around 12 Gbit/s, matching 12G-SDI capacity.

Compressed contribution workflows use RTP/UDP or tunnel RTP inside protocols such as SRT or DTLS, paired with codecs selected for specific latency and bandwidth trade-offs:

• JPEG XS and JPEG 2000 deliver visually lossless compression at low latency, making them well-suited to production-grade contribution where frame-accurate quality must be preserved.

• HEVC (H.265) and AVC (H.264) provide substantially higher compression ratios — critical for WAN and cloud-bound links — at the cost of increased encoding latency and processing complexity.

The choice of codec and transport wrapper is driven by the link characteristics: available bandwidth, acceptable latency, and the degree of packet-loss protection required.

Synchronisation and timing (PTP)

Frame-accurate switching and lip-sync demand precise, network-wide timing. IEEE 1588 Precision Time Protocol (PTP) distributes sub-microsecond clock synchronisation across all devices in an IP broadcast facility. ST 2110 workflows specifically rely on the SMPTE ST 2059 PTP profile to align video frames, audio samples, and ancillary data to a common time reference.

Achieving this in practice requires PTP-aware network infrastructure: switches with hardware timestamping and boundary clock capability ensure timing accuracy is maintained hop by hop. Redundancy is built in through primary and secondary PTP grandmaster clocks, so synchronisation persists during equipment or path failures. Without robust PTP distribution, the deterministic behaviour that broadcasters expect from baseband SDI cannot be replicated in the IP domain.

Orchestration, discovery and control (NMOS and management)

Device discovery and connection management in modern IP broadcast facilities rely on the NMOS family of specifications developed by AMWA. IS-04 handles discovery and registration — allowing devices to announce their capabilities and available resources to a central registry. IS-05 provides connection management, enabling programmatic routing of ST 2110 flows without manual patching.

Together, these APIs enable automation integration, source metadata exchange, and auditable routing changes. Beyond NMOS, centralised management systems handle inventory, configuration, firmware lifecycle, logging, and alarms. Estate management layers collect telemetry from across the infrastructure and expose APIs for real-time monitoring and automated failover — essential when managing dozens to hundreds of IP endpoints.

Addressing and logical topology

Hierarchical addressing is the foundation of a manageable broadcast IP fabric. The goal is to map organisational structure and failure domains directly onto prefix lengths so that routing tables remain compact and troubleshooting stays intuitive.

For IPv4, a common allocation model assigns a /16 or /20 block per site or region, subdivided into /24 subnets per rack, workgroup, or logical domain. For IPv6, a /48 per site with /64 subnets per VLAN follows the same principle with vastly larger address space. Contiguous block assignment per region enables route summarisation, which reduces global forwarding information base (FIB) entries and simplifies policy enforcement at aggregation boundaries.

Subnet sizing should include formulaic headroom: multiply expected hosts by 1.5 to accommodate stable growth without re-addressing. Avoid proliferating many small subnets, as each one adds a route entry and increases convergence complexity.

Service-plane separation is equally critical. Use VLANs or VxLAN Network Identifiers (VNIs) to isolate media, control, management, and timing traffic. Tie each plane to its own addressing range and route policy so that a misconfigured device on the management plane cannot inject traffic into the media plane or vice versa.

Topology and layer‑2/3 boundary

The choice between a collapsed-core and a spine-leaf topology depends on facility scale and growth trajectory.

A collapsed core — where Layer-3 routing is pushed to the access layer — is acceptable for small-to-medium facilities with tens of racks. It minimises switch tiers and simplifies initial deployment. However, it scales poorly because every new access switch increases the routing adjacency count at the core.

For datacentre-scale or large-campus fabrics, a spine-leaf topology with equal-cost multipath (ECMP) provides predictable latency and linear scale. Typical spine-leaf sizing targets a 4:1 oversubscription ratio from leaf to spine for non-blocking media designs, with 25G, 40G, or 100G uplinks selected based on media density and peak channel counts.

Regardless of topology, place media switching or gateway appliances at the edge of each site. Route between sites at Layer 3 to limit broadcast and multicast domains, and to enable independent timing domains per facility. This boundary also simplifies security policy and fault isolation.

Routing fabrics and protocols

Interior gateway protocols (IGPs) handle intra-site routing. OSPFv2/v3 and IS-IS are both viable; IS-IS is preferred for very large fabrics because its flat identifier model avoids OSPF’s area hierarchy overhead. When tuning convergence timers — for example, reducing hello/dead intervals to 50–100 ms — do so only after careful testing, since aggressive timers on a loaded fabric can cause false adjacency flaps.

For multi-site WAN connectivity, Internet peering, and cloud interconnects, BGP is the standard. Use BGP route reflectors to scale the iBGP mesh and BGP communities for granular policy control (e.g., tagging routes by site, service type, or priority).

Where EVPN-VxLAN fabrics are deployed, BGP EVPN serves as the control plane for Layer-2 overlay MAC learning and inter-site L2 extension. This approach eliminates flood-and-learn behaviour, provides integrated ARP suppression, and allows multi-tenant overlays without the operational cost of spanning tree at scale.

Network engineering and QoS

Broadcast IP networks require deterministic packet handling that goes well beyond best-effort delivery. Common techniques include:

• VLAN segmentation to isolate media, management, and PTP traffic domains.

• DiffServ/DSCP markings to classify media packets for priority treatment.

• Per-flow policing and queuing — using class-based weighted fair queuing (CBWFQ) or strict priority queuing — to bound jitter and delay for media flows.

For WAN links, where packet loss and variable latency are unavoidable, forward error correction (FEC) schemes such as SMPTE 2022-1/2022-7 add parity data that enables receiver-side recovery without retransmission. Automatic repeat request (ARQ) mechanisms — often integrated into SRT or accelerated SRT stacks — provide an additional recovery layer. Link capacity planning must account for peak uncompressed line rates or codec bitrates plus the overhead introduced by FEC, RTP headers, and network encapsulation.

Security and content protection

Content in transit is protected through multiple layers. Transport encryption — DTLS for media streams, SRTP for real-time flows, and TLS for control-plane communications — prevents interception and tampering. Secure tunnelling protocols such as SRT with AES encryption or IPsec VPNs provide end-to-end protection over unmanaged networks.

Where conditional access is required, MPEG-TS scramblers and BISS variants are applied at ingest or head-end points. Operational security measures include role-based access control, signed firmware to prevent unauthorised modifications, and hardened management APIs with authentication and audit logging.

Why it matters in broadcast

IP video infrastructure is not simply a like-for-like replacement for SDI cabling. It fundamentally changes what is operationally and economically feasible.

• Remote production (REMI) and centralised workflows. Multiple camera feeds from a venue can be transported to a central facility or cloud environment for mixing and production, dramatically reducing the on-site footprint and enabling flexible staffing and resource allocation across simultaneous events.

• Higher resolutions and multi-angle coverage. IP scales to the multi-gigabit streams required for 4K and 8K acquisition and supports multi-camera contribution without the rigid point-to-point constraints of SDI cabling.

• Hybrid and cloud workflows. Standardised IP transports and orchestration APIs allow integration with cloud-based encoders, processing chains, and CDN handoffs, enabling workflows that span on-premises and public cloud infrastructure.

• Operational flexibility and cost control. Virtualised or software-defined transport and processing allow capacity to be dynamically reallocated. Hardware resources can be consolidated, and new services can be provisioned without physical re-cabling.

Technical specifications and trade-offs

The engineering trade-offs in IP video infrastructure revolve around three axes: latency, bandwidth, and reliability.

Latency varies widely depending on the transport and codec. Uncompressed ST 2110 achieves sub-frame latency — tens of microseconds to low single-digit milliseconds within equipment. JPEG XS encoding typically adds 1–5 ms. HEVC and AVC links over WAN often yield 20–250+ ms depending on encoder lookahead settings, GOP structure, network round-trip time, and ARQ/FEC recovery windows.

Bandwidth requirements range from multi-gigabit for uncompressed to tens of megabits for highly compressed contribution. A 1080p50/59.94 uncompressed signal (10-bit, 4:2:2) consumes approximately 1.5–3 Gbit/s depending on packing. JPEG XS can reduce this by roughly 10:1 or more while maintaining production quality. HEVC achieves compression ratios of 20:1 and beyond versus uncompressed, but at higher encoder delay.

Reliability on-premises is addressed through redundant switching and ST 2022-7 seamless protection switching, achieving near-lossless operation. WAN links require FEC, path bonding, or accelerated transport protocols like SRT to approach equivalent reliability.

Comparison of transport alternatives

Alternative	Typical use case	Latency (typical)	Bandwidth (1080p50)	Synchronisation	Resilience options
ST 2110 (uncompressed)	In-facility production, OB trucks	Sub-frame to single-frame (µs–ms)	1.5–3 Gbit/s	PTP (IEEE 1588 / ST 2059)	Redundant switches, ST 2022-7
JPEG XS (visually lossless)	Low-latency contribution, intra-facility	1–5 ms (encode/decode)	~150–500 Mbit/s	PTP / timing via wrappers	FEC, link aggregation
JPEG 2000	Contribution with multicast needs	5–30 ms	~200–800 Mbit/s	PTP via wrappers/metadata	FEC, ST 2022 family
HEVC / AVC (compressed)	Long-haul WAN, cloud ingestion	20–250+ ms	5–50+ Mbit/s	PTP for production; RTP timestamps otherwise	ARQ, SRT, FEC
SDI over coax/fibre (legacy)	Traditional studio/OB links	Sub-frame (electrical)	1.5–12 Gbit/s per link	Embedded timing	Redundant SDI paths

Latency ranges depend on encoder configuration, network RTT, and protection schemes. Bandwidth figures are illustrative and should be verified against specific codec profiles and product datasheets.

Link speeds and density must be planned against media ingress requirements. As a reference point, a high-density core chassis supporting 100G bi-directional switching and up to 72 Gbit/s of aggregate SRT capacity can service very high channel counts in a single platform.

Comparison of fabric architectures

Dimension	Traditional collapsed core (Layer-3 at edge)	Spine-leaf fabric (EVPN-VxLAN)	EVPN over MPLS or L3VPN (multi-site)
Best scale	Small–medium facilities (tens of racks)	Large datacentre / campus (hundreds of racks)	Multi-site WAN spanning regions
Typical oversubscription	1:1–4:1	1:1 (non-blocking)	Depends on WAN links
East–west latency	Low within site; limited scale	Lowest predictable latency; ECMP	Variable, depends on WAN
Multicast / media handling	VLAN multicast, PIM sparse mode	EVPN-VxLAN for L2 adjacency; PIM/IGMP evasion possible	MPLS L3VPN + PIM or EVPN over MPLS for media
Operational complexity	Lower initial; higher at scale	Higher initial design; lower at scale	High (requires WAN carrier coordination)
Cloud interconnect	Via L3 routing / BGP	Native via BGP EVPN / Direct Connect integrations	Suited for managed carrier interconnects

Choose collapsed core for simplicity at small scale. Choose spine-leaf with EVPN for predictable micro-latency and high east-west throughput across many racks. Choose EVPN/MPLS or BGP L3VPN where carrier WAN connectivity and multi-region span are the primary constraints.

Related approaches and standards

• SMPTE ST 2110: The media-over-IP suite defining separate essence streams for video, audio, and ancillary data.

• SMPTE ST 2059: The PTP profile for broadcast timing alignment, built on IEEE 1588.

• SMPTE 2022-1 / 2022-7: FEC and seamless protection switching for RTP over UDP streams.

• NMOS (AMWA IS-04 / IS-05 / IS-08): Specifications for discovery, connection management, and audio channel mapping of IP media devices.

• SRT (Secure Reliable Transport): An open-source transport protocol providing packet recovery and optional AES encryption for contribution over unmanaged networks.

• RTP / RTCP and RTSP: Real-time packet protocols and control signalling used across compressed and uncompressed workflows.

• JPEG XS, JPEG 2000, HEVC, AVC: Commonly deployed production and contribution codecs, each with distinct latency, quality, and bitrate characteristics.

• EVPN-VxLAN: Uses a BGP EVPN control plane for scalable Layer-2 overlays and multi-tenant fabrics, eliminating flood-and-learn and spanning tree dependencies.

• RFC 7938: Guidance on BGP use in large-scale data centres, relevant to spine-leaf fabric design.

• RFC 7655: RTP payload formats and usage patterns for time-synchronised media.

How Appear addresses this

Appear does not build the fabric itself. The X Platform sits at its edge — the boundary where contribution traffic enters a facility network, and where facility traffic leaves for the WAN.

The X Platform hardware family — spanning the X5, X10, and X20 chassis — delivers modular media transport and real-time encoding/decoding across multiple codecs including AVC, HEVC, JPEG XS, and JPEG 2000. These platforms support MPEG-TS, RTP, and accelerated SRT transports, enabling operators to select the appropriate codec and delivery mechanism for each link in the signal chain. For the network planes described above, the platform carries ST 2110 with ST 2110-30/31 audio on X10 and X20, ST 2022-6 and ST 2022-7 dual-path protection, and hardware-accelerated SRT, with modules locking to PTP.

Capacity is a slot budget. On an X20 in 2RU, an SWx400 module provides two QSFP28 ports at 10, 25, 40, or 100G optical for uplinks into a leaf or aggregation layer. Filled instead with accelerated SRT modules, the same 12 slots carry up to 72 Gbit/s of SRT traffic — an alternative use of the budget, not an additional figure. X10 covers the same functions in 1RU, and X5 serves the compact edge site.

For synchronisation and management, the Appear platform families support PTP for frame-accurate timing alignment and include a centralised estate management and control layer (XM) that handles configuration, monitoring, and lifecycle operations across deployed equipment. XM is self-hosted and sits outside the media path, so chassis keep running if the management layer does not.

Appear also publishes solution-level guidance mapping platform capabilities to specific broadcast workflows, including remote production, interfacility connectivity, and satellite-to-IP migration. These resources are available through Appear’s contribution solutions overview.

FAQ

What is the difference between ST 2110 and SMPTE 2022?
ST 2110 defines how to packetise discrete essences — video, audio, and ancillary data — for uncompressed IP transport. SMPTE 2022-7, part of the 2022 family, provides RTP stream protection and seamless switching for IP links.

When should I use JPEG XS versus HEVC for contribution?
Use JPEG XS when ultra-low latency and visually lossless quality are required with moderate bandwidth reduction. Use HEVC when bandwidth is constrained and higher compression ratios are essential, accepting increased encoder latency and processing demands.

How much network capacity do I need for an ST 2110 studio?
Plan for sustained line rates per video essence — 1.5–3 Gbit/s for HD, approximately 12 Gbit/s for single-lane 4K — plus headroom for audio, ancillary data, FEC, and protocol overhead.

How is lip-sync maintained across IP links?
Through PTP time distribution (IEEE 1588 / SMPTE ST 2059) to timestamp frames and audio samples; receivers align playback using PTP and RTP timestamps to keep audio and video synchronised.

Can ST 2110 traverse the public internet?
Native ST 2110 requires low jitter, packet ordering, and consistent bandwidth not typically available on unmanaged internet paths. For WAN or public internet use, use low-latency compression with reliable transports such as SRT or FEC, or provision service-provider-managed IP circuits with QoS guarantees.

What redundancy models are common in IP video infrastructures?
Common strategies include dual redundant PTP grandmaster clocks, redundant switch fabrics, ST 2022-7 paired flows on separate network paths, hot-standby encoders/decoders, and multi-path WAN bonding.

How does NMOS improve routing in IP production?
NMOS IS-04 enables automatic discovery and registration of devices and their resources. IS-05 provides programmatic connection control so routing changes are made via API rather than manual patching.

What are practical QoS settings for media flows?
Mark media RTP flows with high-priority DSCP values, place them on dedicated VLANs, and configure strict priority or reserved-bandwidth queues on core switches, while reserving buffers to absorb jitter without excessive queueing latency.

How do I secure contribution links carrying premium sports content?
Use encrypted transports (SRTP/DTLS, SRT with AES), authenticated management APIs, conditional access where required, network isolation via VPNs or private circuits, and continuous monitoring.

What monitoring telemetry is important for IP video?
Key metrics include packet loss, jitter, latency, PTP offset and clock quality, stream bitrates, buffer occupancy, encoder/decoder CPU load, stream continuity counters (RTCP), and FEC/ARQ recovery statistics.

Is multicast necessary for ST 2110?
ST 2110 supports both unicast and multicast. In-facility deployments commonly use multicast for efficient one-to-many distribution; WAN or cloud scenarios often use unicast or orchestrated point-to-point connections.

How do you plan for future formats such as 4K, 8K, and HDR?
Design with scalable switch fabrics (40G, 100G, or 400G interfaces), modular processing platforms that can host updated codec and format modules, PTP timing capable of higher frame rates, and uplink headroom to accommodate higher per-stream bitrates or multiple parallel feeds.

Should I use IPv4 or IPv6 for a new broadcast fabric?
Use IPv6 for future-proof addressing wherever carrier and cloud support exists. Continue with IPv4 in a dual-stack configuration where legacy equipment or third-party services require it.

How large should site subnets be?
Allocate contiguous blocks sized for current hosts multiplied by 1.5 for headroom. Common practice assigns /16 or /20 for regional or site allocation and /24 per rack or logical workgroup for IPv4, with /64 subnets per VLAN for IPv6.

When should I choose EVPN-VxLAN?
Choose EVPN-VxLAN when you need large-scale Layer-2 adjacency, multi-tenant overlays, and predictable east-west performance across many racks or nodes.

How do I preserve PTP accuracy across WAN?
Use boundary or transparent clocks within the network where possible. Over WAN, prefer local genlock at edge sites and apply timestamp compensation in ingest if PTP cannot be reliably passed end-to-end.

How much uplink capacity do I need per leaf?
Dimension based on peak channel bitrate multiplied by simultaneous channels. Uncompressed 1080p50 at ST 2110 consumes approximately 1.5–3 Gbit/s per stream depending on sample rates and colour profiles. Sum peak channels and select uplink ports (25G, 40G, or 100G) to stay within the chosen oversubscription target — 1:1 or better for non-blocking designs.

What protection methods exist for live RTP flows?
SMPTE 2022-7 provides dual-path hitless protection. IP-FEC with appropriate budgets handles probabilistic loss. Application-level resilient transports such as SRT and RIST offer end-to-end recovery with NAT traversal. The choice depends on link characteristics and recovery-time requirements.

How should I handle address and route security?
Apply reverse-path forwarding (RPF) checks, prefix filtering, and router authentication (MD5 or TTL security where supported). Use BGP communities and policy to prevent route leaks. Separate the management plane with ACLs and role-based access control.

How do I test architecture changes safely?
Use a lab or staging environment with traffic replay and timing simulation. Validate PTP and media flows, run failover scenarios covering link, module, and protocol failures, and measure end-to-end latency and packet loss before promoting changes to production.

What are realistic jitter buffer settings?
Start with buffers sized to cover 2× RTT plus expected queuing delay. For WANs with 50–100 ms RTT, buffers of 100–200 ms are typically required. Balance buffer size against the added end-to-end latency it introduces.

When is application-level FEC preferable to network FEC?
Use application-level FEC (SRT, RIST) when end-to-end transport control and NAT traversal are needed. Use network-level FEC (IP-FEC) when the carrier or service supports it with predictable loss characteristics and the path does not cross NAT boundaries.