Stoatworks Labs

Reference

Reference signals, and why the shape of the edge matters

A genlock receiver does one thing: it decides, as precisely as it can, the instant an edge arrived. Everything that distinguishes black burst from tri-level sync comes down to how steeply the signal is moving at the moment that decision is made.

Noise on a reference line moves the signal vertically. The receiver converts that into a horizontal error — a timing error — and the exchange rate is the slew rate at the decision point. A slow edge turns a small voltage error into a large timing error; a fast one does not. That single relationship is the entire argument for tri-level sync, and this page computes it rather than asserting it: on the figures below, tri-level is about 9.3 times more accurate for the same noise.

It is also why termination is not a detail here. An unterminated reference line does not fail cleanly — it rings, and ringing near a decision threshold is indistinguishable from jitter.

The signals

SignalStandardAmplitudeSyncBurstTimes fromLocks
Bi-level (PAL)Black burst, colour black, blackburstITU-R BT.470 / SMPTE ST 3181 V p-p−300 mV4.43361875 MHzA threshold part-way down one edge525/625-line SD, and most HD equipment via its reference input
Bi-level (NTSC)Black burst, color blackSMPTE ST 170 / SMPTE ST 3181 V p-p−286 mV (40 IRE)3.579545 MHzA threshold part-way down one edge525-line SD, and HD equipment via its reference input
Tri-levelHD sync, tri-level blackSMPTE ST 274 / ST 296600 mV p-p−300 mV then +300 mVNoneThe zero crossing, mid-transitionHD and UHD equipment
TTL syncLogic-level sync, triggerNo video standard — manufacturer convention0–5 V0 V active low, or 5 V active highNoneA logic threshold, typically 1.4 VMachine vision, scientific and industrial cameras; some genlock and trigger inputs

All three video signals are 75 Ω, and TTL sync usually is not — which is the first thing to establish before connecting one to anything. Every video reference here rides on a BNC and looks identical on the patch panel.

  • Bi-level (PAL) — Composite video with a black picture. Still the most widely accepted reference input on the back of professional equipment, thirty years after anything was actually shot on it.
  • Bi-level (NTSC) — The 7.5 IRE setup pedestal puts black at +54 mV rather than at blanking, which is the origin of a great many level arguments and no timing ones.
  • Tri-level — Negative then positive about zero, so it carries no DC and the timing reference sits on a zero crossing rather than on a level. Both of those are deliberate.
  • TTL sync — Not a broadcast signal and not governed by SMPTE. Electrically excellent and interoperably hopeless — polarity, active level and termination are all whatever the manufacturer chose.

The shape of the edge

Bi-level and tri-level sync edges comparedTwo waveforms. On the left, a bi-level sync pulse falls from the blanking level to minus 300 millivolts over a relatively slow edge, and the receiver's decision threshold sits part-way down that single edge. On the right, a tri-level pulse falls to minus 300 millivolts and then rises through zero to plus 300 millivolts over a much shorter transition, and the receiver times from the zero crossing in the middle of that steep rise.Bi-levelTri-level−300 mV0 mV (blanking)decision on a slow edge−300 mV+300 mV0 mVdecision at the zero crossing
The tri-level pulse crosses zero in the middle of a 600 mV transition. The bi-level pulse is timed part-way down a 300 mV one that takes several times as long — so the same millivolt of noise buys far more nanoseconds of error.
SignalExcursionTransitionSlew rateError from 10 mV of noise
Bi-level (PAL)300 mV250 ns1.20 mV/ns8.33 ns
Bi-level (NTSC)286 mV250 ns1.14 mV/ns8.74 ns
Tri-level600 mV54 ns11.11 mV/ns0.90 ns
TTL sync5000 mV10 ns500.00 mV/ns0.02 ns

timing error = noise ÷ slew rate. The absolute figures depend on how much noise you assume; the ratio between the rows does not, and that ratio is the point. TTL's excellent showing here is real electrically and is exactly why it is used for triggering — it just has no interoperability story.

Termination decides whether any of it works

StateWhat it isReceiver seesSymptom
Correctly terminatedOne 75 Ω termination, at the far end of the lineFull amplitude, no reflection
UnterminatedA high-impedance input with nothing terminating the lineUp to double amplitude, with ringingIntermittent double-triggering, sync that locks and drops, worse as the cable gets longer
Double terminatedTwo 75 Ω loads on one line — usually a terminator left in alongside a terminating inputHalf amplitudeFails to lock at all, or locks only on short cables. The classic loop-through mistake
Wrong impedanceA 50 Ω terminator or connector in a 75 Ω lineReflections proportional to the mismatchMarginal locking that changes when anything on the line is moved

The loop-through convention is where this goes wrong. A reference input with a loop-through is high impedance and does not terminate the line; the last device in the chain has to, either with a switch set to 75 Ω or with a terminator on its spare socket. Set two devices to terminate and the amplitude halves. Set none and it rings.

A distribution amplifier avoids the argument entirely by giving every device its own terminated line, and on anything larger than a couple of devices that is the right answer — daisy-chained reference is a single point of failure that takes the whole chain down when one device is unplugged.

The things that reach site

  • Most HD equipment still accepts black burst, and it is usually the right choice. A bi-level reference will lock an HD or UHD device perfectly well through its reference input, and a facility with one legacy SD device in it is simpler on black burst throughout than on two reference standards. The timing advantage of tri-level is real and is rarely the constraint.
  • Reference and video look identical on a patch panel. Both are 75 Ω on BNC. A reference line patched into a video input shows black; a video line patched into a reference input may lock to something almost right. Label them, and keep them on separate panels if you can.
  • TTL sync into a 75 Ω terminated input is not a plan. A logic output driving 75 Ω is being asked for around 65 mA, which most cannot supply. It will either be loaded down to an unrecognisable amplitude or, on a source that can drive it, work perfectly and mislead you about the next one.
  • PTP is the modern answer and is a different discipline.SMPTE ST 2059 carries the reference over the network with IEEE 1588 Precision Time Protocol, which removes the reference cabling and replaces it with a set of questions about switch support, boundary clocks and grandmaster redundancy. It is better; it is not simpler.
  • A reference generator failing over is not free. Changeover between two generators produces a phase discontinuity unless they were locked to each other, and everything downstream re-locks — which on a live output looks like a glitch on every source at once.

Sources

  • How genlock extracts timing from a composite referenceSkyworks AN377, Timing and Synchronization in Broadcast Video, read directly. Its Figure 4 is the sync-separator and timing-generator chain described above, and §5 covers the SMPTE timing and alignment jitter limits.
  • Black burst as a reference standard — SMPTE ST 318, which specifies composite black burst with optional VITC for use as a studio reference.
  • Tri-level sync — SMPTE ST 274 (1080-line) and ST 296 (720-line). Both are paywalled and were not read. The ±300 mV levels and the 54 ns transition are cross-checked between manufacturer documentation and reference sources, and the computed jitter comparison rests on that transition figure — treat the ratio as sound and the absolute nanoseconds as nominal.
  • PTP for video — SMPTE ST 2059-1 and ST 2059-2, over IEEE 1588.

Assembled 7 September 2026 with AI assistance. The jitter comparison iscomputed from declared excursions and transition times and guarded at build time so that it cannot silently invert, but it is a first-principles model of one effect — noise converted to timing error at the decision point. Real genlock performance also depends on the receiver's PLL, its loop bandwidth and the noise it contributes itself, none of which is modelled here. Nothing on this page has been measured on hardware.

Companion page: SDI over coax — the same 75 Ω infrastructure, and the jitter limits that reference feeds into.