IP contribution for VAR: ultra-low-latency camera-to-review networks

How it works

The VAR pipeline is a multi-segment chain. Each segment introduces its own latency, synchronisation, and fidelity constraints. The sections below walk through the functional stages and the concrete engineering parameters that govern them.

Capture and local ingest

VAR workflows typically ingest high-definition or UHD feeds carried as SMPTE ST 292/424 SDI signals or as native ST 2110-20/30/31 IP streams. Where image integrity is paramount for frame-level analysis — offside lines, ball-crossing-line measurements — uncompressed 10-bit YCbCr 4:2:2 or 4:4:4 colour sampling is preferred.

Frame rates are dictated by the broadcast standard and the desired replay quality. Live sport commonly uses 50 Hz or 60 Hz capture; dedicated slow-motion cameras run at 100 Hz or 120 Hz for ultra-smooth replays. A single frame at 50 Hz occupies a 20 ms window; at 60 Hz, 16.667 ms. Sub-frame alignment across cameras must therefore be measured in fractions of these intervals — typically less than 1 ms — to guarantee that multi-angle replay frames correspond to the same instant of play.

At the camera or encoder edge, genlock and timecode insertion should occur before transport. Embedding SMPTE linear timecode (LTC) or RTP-level timestamps at the point of capture preserves the original timing reference throughout the downstream chain.

Timing, synchronisation, and reference distribution

Precise timing underpins every aspect of VAR. The de-facto method for synchronising packetised IP media to an absolute timeline is the Precision Time Protocol (PTP), specifically the SMPTE ST 2059 media profile of IEEE 1588-2008. PTP should be distributed through boundary clocks or transparent clocks engineered to limit end-to-end synchronisation error to sub-microsecond levels.

Where cameras still require analogue genlock (blackburst or tri-level sync), PTP-to-genlock gateways at network edges bridge the two timing domains. Continuous timecode — whether SMPTE TC or embedded via ST 2110-40 ancillary data — must be preserved across ingest, transport, and replay to label every frame unambiguously.

The synchronisation budget for VAR has two tiers. For critical multi-camera compositing and measurement overlays, inter-feed skew should be held below one video line — about 18 µs for 1080p50 (20 ms divided across 1,125 total lines), or about 36 µs for 1080i25. For the broader operational requirement of robust frame selection and replay alignment, a practical target of inter-feed skew below 1 ms is sufficient.

Contribution transport (camera → edge → VOR)

Transport design is the segment where bandwidth, latency, and resilience trade-offs are most visible. Three broad categories apply:

• Uncompressed ST 2110 over 10G, 25G, or 40G links delivers bit-perfect imagery with sub-frame network latency. Overall transport latency in a well-designed local-area network is often below 10 ms.

• Low-latency visually lossless codecs such as JPEG XS or JPEG 2000 in low-delay mode reduce bandwidth by a factor of four to ten while adding only 1–15 ms of encode-plus-decode latency per leg, depending on implementation and resolution.

• Encapsulated compressed streams using HEVC or AVC wrapped in SRT or RIST traverse unmanaged or bandwidth-constrained wide-area networks. Low-delay HEVC modes may add 20–100+ ms depending on GOP structure and encoder configuration.

Latency accumulates across discrete stages: capture, encode (if present), encapsulation, network transit, decode, and rendering. Each stage must be budgeted individually and validated end-to-end.

Redundancy is non-negotiable. SMPTE ST 2022-7 seamless protection switching provides single-link redundancy by sending identical RTP flows over two diverse paths and switching hitlessly on loss detection. For WAN links, application-layer forward error correction (FEC) per SMPTE ST 2022-1/2 or the built-in retransmit mechanisms of SRT and RIST recover burst losses without adding significant latency in normal conditions.

Multi-feed synchronised replay and slow-motion

Replay servers sit at the heart of the VAR decision process. They receive timecode-aligned, genlocked streams and buffer multiple seconds of content at full frame granularity. Each ingested frame must carry a globally consistent timestamp — either a PTP-derived value or an embedded SMPTE timecode — so that the replay operator can select the same temporal instant across every camera angle.

For frame-accurate slow-motion, buffer alignment depends entirely on the accuracy of these global timestamps. Variance in timestamp jitter must remain below 1 ms to avoid mispaired frames across angles. Original frame identifiers — timecode values and PTP timestamps — should persist in replay metadata and decision logs, creating an immutable audit trail for post-event review and league reporting.

Return paths and on-pitch display

Once the VOR reaches a decision, the reviewed clip or freeze frame must travel back to the on-pitch referee display with deterministic, sufficiently low latency to avoid undue delay in match restart. While the overall round-trip from VOR decision to on-pitch display is often kept under 3–4 seconds in practice, engineering targets are substantially lower to preserve fairness and game flow.

When re-inserting reviewed clips into the broadcast output or sending freeze frames to the touchline monitor, consistent timebase and format conversion — whether to SDI or ST 2110 — must be maintained with minimal additional processing latency. Any format or frame-rate conversion on the return path should be accounted for in the overall latency budget.

Why it matters in broadcast

VAR decisions hinge on frame accuracy and image fidelity. A single compression artefact near a player’s boot or a timing misalignment of even two frames can change the interpretation of an offside call or a ball crossing the goal line. The engineering quality of the contribution network directly determines decision integrity.

Time to decision affects the flow of the match and viewer perception. Every millisecond saved in buffering and processing latency translates into faster reviews without compromising accuracy — a critical balance for leagues under pressure to minimise stoppages.

Timestamps and preserved frame identities provide an evidentiary trail. Appeals, post-match audits, and league compliance reporting all depend on the ability to trace a reviewed frame back to its original capture instant with certainty.

Centralised VAR operations — where a single VOR supports multiple venues simultaneously — offer significant savings in staffing and equipment duplication. However, centralisation raises the bar for contribution and return infrastructure: every remote venue must achieve the same synchronisation, latency, and fidelity guarantees as an on-site setup, across potentially long and variable network paths. The general engineering framework for venue-to-facility centralisation is covered in the remote production and REMI decision guide; VAR adds the frame-accuracy and audit constraints described here.

Technical specifications and trade-offs

The table below compares common transport and codec alternatives for VAR contribution across key engineering dimensions. All figures are indicative and implementation-dependent.

Alternative	Typical one-way transport latency	Image integrity	Bandwidth per 1080p50 stream (approx.)	WAN resilience
Uncompressed ST 2110 (10G)	< 5–10 ms	Bit-perfect	~2.1–2.5 Gbit/s ST 2110-20 essence (~3 Gbit/s SDI-mapped via ST 2022-6)	High on managed L2/L3 with PTP; requires redundant paths
JPEG XS (low-delay visually lossless) over RTP	1–15 ms (encode + decode)	Visually lossless at proper profiles	~210–520 Mbit/s	Good on managed networks; benefits from FEC and path redundancy
JPEG 2000 (low-delay)	5–50 ms	High quality; mathematically lossless possible	300–1,500 Mbit/s	Good; often used with ST 2110 wrappers
HEVC (low-delay)	20–150+ ms	High perceived quality; some motion artefacts possible	5–50 Mbit/s	Suitable on constrained WAN; sensitive to loss without SRT/RIST or FEC
SRT/RIST encapsulated compressed streams	Network-dependent: adds 10–200+ ms for WAN	Depends on inner codec	Depends on codec profile	Designed for lossy/unmanaged networks with retransmit and FEC

Selection follows a clear hierarchy:

• Choose uncompressed ST 2110 when bandwidth and a managed 10G+ network are available and absolute image integrity with the lowest possible latency is required.

• Choose JPEG XS when near-lossless quality is needed with a four-to-ten-fold bandwidth saving and sub-frame to low-tens-of-milliseconds latency is acceptable.

• Choose HEVC/AVC with SRT or RIST for long-haul or public-internet contribution where bandwidth is constrained; accept higher latency and carefully tune GOP size and low-delay encoder presets.

• For any WAN link, prefer path diversity, packet-level FEC, or ST 2022-7 redundant paths to bound packet loss while preserving low jitter for timestamp accuracy.

A practical latency budget should allocate targets per segment and validate the total:

• Capture/ingest: 0–10 ms

• Encode (if present): 0–20 ms

• Network transit: 5–50 ms (local) or 20–200 ms (WAN)

• Decode/render: 5–20 ms

• Replay processing: 10–100+ ms depending on slow-motion complexity

Always verify link capacity per camera including protocol overhead, FEC, and headroom — a margin of 25–40 % above nominal bitrate is prudent.

Related approaches and standards

• SMPTE ST 2110 — the suite for professional media over IP, separating video (ST 2110-20), audio (ST 2110-30/31), and ancillary data (ST 2110-40) into independent essence streams.

• SMPTE ST 2059 — the PTP profile for synchronous media timing, used alongside ST 2110 to align all streams to a common clock.

• SMPTE ST 2022-7 — seamless protection switching for RTP flows, enabling hitless failover between redundant network paths.

• SMPTE ST 2022-1/2 — forward error correction for RTP, recovering packet loss without retransmission delay.

• JPEG XS (ISO/IEC 21122) — a low-latency, visually lossless codec purpose-built for contribution and live production.

• JPEG 2000 (ISO/IEC 15444) — a high-quality contribution codec with low-delay operational modes.

• SRT and RIST — resilient streaming protocols that add retransmission and congestion control for transport over unreliable networks.

• NMOS (AMWA IS-04/05) — discovery and connection management for IP media systems, enabling automated orchestration of flows across devices and facilities.

How Appear addresses this

Appear’s X Platform covers the venue, mid-point, and core roles in a VAR chain. X10 in 1RU and X20 in 2RU are modular chassis taking encode, decode, and SDI/IP gateway modules in any combination, so a single chassis can carry the contribution feeds out and the reviewed return feed back without adding a second box. X5 is the compact edge encoder and decoder for a venue or OB van, where rack space is the binding constraint rather than channel count.

On codecs, the module library spans JPEG XS and JPEG 2000 on the SIx family for near-lossless contribution, and AVC and HEVC on the ECx modules — ECx110 for SDI I/O, ECx210 for ST 2110 and ST 2022-6 — with selectable latency modes down to ultra-low delay. That range lets a venue pick its point on the bandwidth-against-fidelity curve described above without changing chassis. Modules lock to PTP, and ST 2022-7 seamless protection is available on supported modules for the dual-path redundancy this workflow demands.

Video alignment is the feature that matters most for multi-angle review. The X Platform aligns feeds across modules within a chassis so that replay operators can step through synchronised angles from a common instant, rather than relying on the replay server to reconcile independently arriving streams. It is what makes super-slow-motion review across several cameras behave predictably.

At the core, an X20 takes up to twelve module slots in 2RU, and an SWx400 module provides two QSFP28 ports at 10, 25, 40, or 100G optical for uplinks — the density and aggregation a centralised VOR serving several venues needs. Standards support across the platform covers ST 2110, ST 2022-6 and ST 2022-7, and PTP; check the datasheet for the specific module, since coverage varies by card. XM adds estate-level visibility across deployed chassis: inventory, alarm state, software and licence lifecycle, and an API for teams automating their own operational workflows.

Appear has delivered VAR contribution links in production, including national top-flight football deployments. Deployment accounts are published on the Appear blog and serve as reference architectures for the patterns described here.