Core responsibility (one sentence)

A vision lighting controller is a deterministic power + timing device that delivers repeatable pulse energy (not just “brightness”) within a known trigger window, while providing observability (current/temperature/fault/log) and protection.

Engineering note: “current looks stable” does not automatically mean “optical energy is stable.” Optical output depends on junction temperature, LED efficiency droop, wiring parasitics, and pulse waveform shaping. Chapter H2-2 defines measurement evidence to separate them.

What each light geometry forces the controller to be good at

• Strobe (point/spot): µs-class edges, high peak current, strict delay/jitter; EMI risk peaks during fast switching.
• Bar light (multi-channel, long harness): channel-to-channel skew control, cable inductance management, repeatable pulse energy across channels.
• Ring light (segmented zones): zone matching (current trim), thermal gradient handling (top hotter than bottom), stable output over time.
• Dome (large area): long-duration stability, thermal derating strategy, fault isolation by zones to avoid total blackout.

Practical framing: geometry changes the dominant bottleneck — edge speed (strobe), skew/uniformity (bar/ring), and thermal stability (dome).

Clear ownership prevents mis-troubleshooting

• Lighting controller owns: input power conditioning, DC/DC, constant-current stage, pulse shaping, programmable delay, telemetry (I/T), fault detection, event logging.
• LED head owns: LED array, optics, heatsinking, mechanical alignment. (These determine how much overdrive is physically safe.)
• Camera owns: exposure/window scheduling, ISP pipeline, image processing. Not covered here (avoid scope creep).

Rule of thumb: if the symptom is “brightness jumps” or “freeze quality varies,” prove whether it is timing or energy first (trigger/current/photodiode evidence), before blaming camera-side algorithms.

When something looks wrong, measure these first (before changing settings)

• Trigger-in waveform: edge integrity, noise, level, and timing stability at the controller input pin (same ground reference).
• LED current waveform: rise/fall time, peak current, pulse width, overshoot/ringing (captures wiring + switching reality).

These two signals already separate many root causes: timing-chain issues vs current-stage issues vs harness parasitics.

Engineering goal: measurable, repeatable, verifiable

For machine vision, “it looks bright enough” is not an acceptance criterion. A lighting controller is judged by timing determinism and optical energy repeatability. The same symptom (blur, flicker, inconsistent contrast) can come from different failure modes; metrics provide a clean discriminator.

Measurement principle: capture Trigger, LED current, and photodiode (PD) output together. This separates timing-chain issues (delay/jitter) from current-stage issues (rise/fall, overshoot) and from optical/thermal effects (energy drift).

Core metrics (with measurement evidence)

Standardization tip: always state the reference definition (10–90% for edges, 50% point for jitter/skew, threshold for pulse width). Without a definition, two teams can “measure the same thing” and still disagree.

Fast diagnosis mapping (avoid guessing)

• Motion blur / edge softness → rise time too slow, or light pulse not centered in exposure window (check delay/jitter).
• Frame-to-frame brightness jump → energy repeatability issue (PD evidence) or trigger jitter spike (TIE stats).
• Bar/ring uneven segments → channel current mismatch (error budget) or channel skew (multi-channel timing).
• Gets worse when warm → thermal derating or junction-efficiency droop (requires temperature + PD trend evidence).

Minimal evidence to separate timing vs energy issues

• Trigger + current: proves determinism and electrical pulse integrity (delay/jitter + rise/fall + overshoot).
• PD output: proves whether optical energy is stable and matches the exposure window (integrated energy, not only peak).

If current is stable but PD energy drifts, focus on thermal dynamics, LED efficiency droop, and head-level heat path. If PD energy is stable but images vary, focus on timing alignment and exposure scheduling (camera-side), without changing current stage first.

Topology selection is a return-path problem before it is a schematic problem

In machine vision strobing, the topology you choose primarily determines where the pulsed current returns, how much ground bounce / common-mode noise you inject into the system, and how reliably you can measure current and energy. Efficiency and cost matter, but they are typically second-order once the harness gets longer and channels scale.

Practical rule: long harness + multi-channel + shared cabinet grounding favors high-side or hybrid. Low-side switching can work for short cables and isolated subassemblies, but it fails fast when return paths are uncontrolled.

High-side CC vs Low-side switch vs Hybrid (slow CC loop + fast gate)

Key idea: every µs strobe system effectively needs a fast path (edge + width control) and a slow path (current regulation). The topologies mainly differ in where the fast path current returns and how much collateral noise it creates.

Why “slow loop + fast gate” is not optional for µs strobe

• Slow CC loop (buck / current regulator) stabilizes average current and supply compliance. It corrects drift and load variation over longer timescales.
• Fast gate path (MOSFET gating) defines the electrical pulse: rise/fall, pulse width, and edge placement relative to trigger.
• Separation requirement: if the slow loop tries to react within the µs pulse window, it can create energy non-repeatability (pulse-to-pulse variation) and overshoot.

Engineering test: if you change pulse width slightly and the peak/shape changes unpredictably, the slow loop is “leaking” into the fast domain.

Cable inductance turns fast strobe into a voltage spike generator

Long LED leads behave like an inductor and a resonator. When current is gated quickly, the harness can create overshoot (at turn-on) and undershoot (at turn-off), plus ringing that distorts pulse energy and increases EMI.

• TVS placement rule: protect the node you care about at the node. Controller-end TVS protects controller ports; head-end TVS protects the LED head and local wiring.
• RC snubber rule: place it as close as possible to the switching hotspot (switch node + smallest loop), not at the power entry.
• Proof: measure node voltage at both ends of the harness. If spikes are large at the far end but not the near end, you are seeing harness reflection/resonance.

Fix order (fastest wins): reduce loop area → add gate damping (R_g) → clamp (TVS) at correct endpoint → snubber at switch node → revisit topology if ground injection persists.

Quick discriminator: return-path noise vs pulse integrity

• LED current waveform: rise/fall, overshoot/ringing, pulse width stability (electrical pulse integrity).
• Ground bounce/common-mode at a sensitive reference (camera/IO ground region): confirms whether the topology is injecting noise into shared ground.

If current looks clean but ground bounce correlates with the strobe edge, topology/return-path is the root (especially low-side switch in shared-ground cabinets). If ground is quiet but current rings badly, harness + termination is the root.

Why “electrical pulse achieved” does not guarantee stable light or long life

µs-class strobing is a three-constraint problem: edge speed (freeze motion), energy repeatability (consistent contrast), and thermal envelope (keep LED junction temperature within limits). Pushing only I_peak can make the pulse appear “strong” while optical output becomes unstable and LED lifetime collapses.

Target behavior: the current pulse shape is intentional (rise/fall + width), the delay/jitter is bounded, and the optical energy distribution stays tight across pulses.

Fast edge limiting factors (in practical order)

• Gate driver + MOSFET Qg: insufficient gate current slows edges and increases switching loss.
• Loop inductance (layout): large loop area creates ringing and overshoot; it also spreads EMI broadly.
• Cable inductance (harness): long leads behave like an inductor/resonator; fast gating turns it into spikes.
• LED junction capacitance / dynamics: current-to-light response is not perfectly instantaneous; PD waveform may differ from current.
• Measurement bandwidth: probe bandwidth and grounding can fake slow rise time or hide overshoot.

Engineering discriminator: if rise time changes dramatically with probe setup, the limitation is likely measurement, not the design.

The overdrive envelope is defined by I_peak, T_j, and average power (duty/repetition)

• Red line #1 — I_peak: pulse peak current must remain within LED pulse rating and driver switch current capability.
• Red line #2 — T_j: junction temperature is the real limiter; case NTC responds slower and can lag the real peak.
• Red line #3 — average power: LED (V_f·I) and resistive losses (I²R in shunt/MOSFET) accumulate with repetition.
• Knob — duty & repetition: the same I_peak can be safe or unsafe depending on pulse width and pulse rate.

Practical acceptance: define an envelope (allowed combinations of I_peak, width, repetition) and enforce it in firmware/hardware limits.

Waveform shaping that improves repeatability without “slowing everything down”

• Pre-pulse (preheat): reduces uncertainty by stabilizing junction conditions before the main pulse; improves pulse-to-pulse energy repeatability.
• Two-level current: a small pre-level followed by a controlled high main pulse keeps edges predictable while controlling thermal stress.
• Soft turn-off: reduces undershoot/ringing and EMI by controlling the current fall (not a hard slam to zero).

Shaping is most effective when the fast gate path is explicit (hybrid approach) and the slow CC loop is kept out of the µs domain.

Measurement differences can create false conclusions

• Clamp/current probe: convenient, but bandwidth and placement affect rise/fall and overshoot visibility; calibration drift can hide energy variation.
• Shunt + differential measurement: accurate for current, but requires Kelvin routing and sufficient bandwidth; the measurement loop must be tight.
• T_j estimate vs NTC: NTC measures case/board temperature with delay; short-pulse T_j peaks can be missed.

Minimum viable proof for “stable light”: Trigger + Current + PD (with an integration window aligned to exposure), plus temperature trend during sustained operation.

Separate “timing problem” from “energy/thermal problem” quickly

• Current waveform: I_peak, rise/fall, pulse width, overshoot/ringing → proves electrical pulse integrity and edge control.
• Photodiode energy (integrated over exposure window): proves pulse-to-pulse optical repeatability; add temperature trend to confirm thermal envelope effects.

If current is stable but PD energy drifts with time, thermal dynamics or LED efficiency droop is driving variation; address derating and heat path before modifying edge speed.

Closed-loop current control is only meaningful when the measurement chain is trustworthy

A “constant-current” label is not enough for strobe lighting. The practical requirement is repeatable pulse shape and energy under fast edges, temperature drift, and multi-channel scaling. That means the sensing path (sense location, shunt, CSA, ADC) must be designed with an explicit error budget and a plan for factory + field calibration.

Minimum definition of success: the same trigger and setpoint produces the same current waveform area (and therefore stable optical energy after derating), and every channel can prove health via self-test thresholds.

High-side vs low-side sensing: choose based on what you must protect

Quick discriminator: measure sense waveform and ground bounce at a sensitive reference with the same trigger. If the reading changes dramatically with probing/grounding, the measurement chain is being fooled by common-mode and dv/dt.

Shunt is a component + routing decision (not just a resistance value)

• Resistance value: too low reduces resolution; too high adds loss and raises self-heating, shifting readings over time.
• Pulse vs average loss: I_peak drives instantaneous I²R; repetition/duty drives temperature rise and drift.
• Tempco (TCR): shunt drift directly becomes current drift unless compensated or calibrated.
• Kelvin routing: without Kelvin sense, line/connector resistance becomes part of the “shunt,” breaking calibration repeatability.

Choose CSA parameters that preserve pulse integrity, not just DC accuracy

• Bandwidth / step response: insufficient BW hides overshoot and makes rise/fall appear slower than reality.
• Offset & drift: dominates short pulses and low-current modes; must be zero-calibrated.
• CMRR & PWM rejection: determines whether dv/dt is interpreted as “fake current,” especially in high-side sensing.
• Input filtering: necessary to suppress noise, but excessive filtering adds delay/phase shift that can destabilize fast regulation loops.

Evidence-based check: if measured current changes disproportionately when edges are sped up or slowed down (with the same load), CSA/PWM-rejection and filtering are limiting fidelity.

Make channels match each other even when LEDs do not

• Per-channel zero calibration: capture offset at “off” state; apply at runtime to remove baseline drift.
• Per-channel gain calibration: factory set gain using a known reference condition; store coefficients with versioning.
• LED Vf binning & temperature compensation: equal current does not always produce equal optical output; compensate via per-channel tables if optical uniformity is required.
• Connector/contact resistance: treat as a measured variable; large changes indicate harness aging or poor crimp.

Practical acceptance target: channel-to-channel pulse area (∫I dt) stays within a tight band after applying calibration and derating.

Turn “it drifts” into an audit-friendly error table and thresholds

Thermal is slow, strobe is fast — derating must be an envelope

The goal is not “keep it cool” but keep optical energy consistent while staying inside a safe thermal envelope. The correct implementation is to translate temperature and thermal-model outputs into time-domain limits: I_peak, pulse width, and repetition rate.

Key caution: a board or heatsink NTC can lag junction temperature. Short-pulse peaks can damage lifetime even when NTC looks safe.

Use multiple temperature points to separate root causes

• LED board NTC: closest proxy to LED module heating; most correlated with optical droop.
• Heatsink temperature: shows thermal path effectiveness and ambient change sensitivity.
• Driver MOSFET temperature: indicates switching/conduction losses; rises with faster edges and higher repetition.

Interpretation: LED board rises faster than heatsink → poor module-to-sink interface. MOSFET rises faster than LED board → driver loss dominating (edges or Rds(on)).

Three limiters, one objective: stable energy without runaway

• Limit I_peak: directly caps junction stress and peak power; best for preventing short-pulse damage.
• Limit pulse width: caps per-pulse energy; effective when brightness is driven by longer pulses.
• Limit repetition: caps average power; most effective for long-run stability and enclosure heat soak.

Implementation should be an envelope, not a single threshold: as temperature increases, allowable I_peak and/or repetition decreases smoothly.

Pick the feedback method based on what you can prove

• Photodiode feedback (if available): controls optical energy directly; compensates LED efficiency droop, but requires PD calibration and drift handling.
• Temperature-model derating: cheaper and universal; depends on thermal path consistency across builds and environments.
• Hybrid: temperature envelope sets hard safety limits; PD fine-tunes energy within that envelope.

Proof requirement: regardless of method, log the derating decision (temperature, envelope limit, resulting I_peak/width/repetition) so field behavior is explainable.

Make stability measurable: long-run drift curves and fitted thermal resistance

• 1h / 8h drift test: hold a fixed trigger and strobe recipe; record temperature points and optical output proxy over time.
• Optical proxy options: PD-integrated pulse energy (preferred) or camera grayscale statistics (as a measurement signal only).
• Thermal model alignment: estimate Rθ path (junction→board→sink→ambient) and align model slope/time constant to measured curves.

Determinism is measurable: delay is calibratable, jitter is noise, skew is alignment

Deterministic strobing is not “it triggers.” It is the ability to deliver the same current rise event at a predictable time. On the lighting-controller side, determinism breaks into three metrics: delay (mean trigger-to-current latency), jitter (cycle-to-cycle timing variation), and multi-channel skew (relative alignment between channels).

Scope boundary: only the trigger input, timing chain, and pulse generation inside the lighting controller are covered here. System time-sync (PTP/1588) is out of scope.

Trigger input choices set the noise floor before the timing chain even begins

Key rule: the input conditioner must control edge rate, threshold stability, and noise rejection, otherwise jitter is injected before any delay-line can help.

A repeatable delay chain is a system of blocks, each with a timing error signature

• Capture: hardware capture reduces software-induced jitter compared to polling or heavy ISR paths.
• Conditioning: debounce, Schmitt behavior, and isolation define effective edge time and its stability.
• Timebase / PLL: the clock reference sets the achievable jitter floor.
• Programmable delay: coarse + fine delay improves range without sacrificing timing granularity.
• Pulse generation: the reference event is the current rise, not only the logic pulse edge.

Jitter is not one thing — identify whether it is load-dependent, quantized, or noise-floor

• MCU ISR jitter: scheduling, interrupt masking windows, and memory wait states translate into time variation under CPU load.
• FPGA CDC jitter: clock-domain crossing can create step-like timing uncertainty (discrete bins).
• PLL phase noise: reference noise becomes a timing-noise floor; calibration cannot remove it.
• Input threshold noise: slow edges and noisy thresholds shift the perceived trigger edge time.

Fast discriminator: load-dependent jitter points to ISR/scheduling; quantized multi-bin jitter points to CDC; environment- or noise-correlated jitter points to PLL reference or input conditioning.

Measure what matters: trigger-to-current TIE and same-screen skew

• TIE definition: TIE = t(current-rise) − t(trigger-edge) on the controller-side reference points.
• Probe points: trigger edge at controller input; current rise from current probe or shunt waveform.
• Metrics: mean delay, RMS jitter, peak-to-peak jitter, and tail behavior (P99).
• Multi-channel skew: observe multiple current rise edges on one scope screen under the same trigger.

High di/dt + fast dv/dt turn cables and chassis into antennas

A strobe channel is effectively a pulsed power transmitter. Strong current edges create large loop radiation and common-mode currents. The result is cell-level symptoms such as false triggers, I/O upsets, and power input disturbances. The controller-side job is twofold: reduce injection at the source and harden its own interfaces against ESD/EFT/surge.

Scope boundary: only controller/harness-side EMI coupling and protection are covered. Camera/network protocol specifics are out of scope.

Identify what radiates and why it couples

• Cable loop area: high di/dt loop behaves like a loop antenna; routing and return path define radiation strength.
• Fast switch node: high dv/dt edges drive displacement currents into parasitic capacitances and shields.
• Common-mode current: shield, chassis, and ground conductors carry RF energy into other equipment.

Control the return path before adding components

• Minimize high di/dt loop: keep forward and return conductors tightly coupled; avoid large harness loops.
• Partition: separate noisy power switching region from sensitive trigger/IO conditioning region.
• Single-point coupling strategy: avoid multiple uncontrolled connections that create ground loops and unpredictable common-mode paths.

Component fixes (CMC/TVS/snubber) are most effective only when the current loop and return path are already constrained.

Use the right tool for the right path

Protect at the connectors and keep the return path short

• Power input (24V/PoE front-end): surge and EFT bursts require clamping and filtering at the entry point to prevent brownout/reset cascades.
• IO/Trigger/Strobe lines: ESD and EFT energy couples through thresholds; use ESD arrays, series damping, and (when needed) isolation to keep conditioning stable.

Protection effectiveness is dominated by placement: long return inductance turns a “protector” into an RF injector.

Make EMI actionable: correlate probe peaks with fault counters

• Near-field scan: sweep harness loop, switch node area, IO connector region, and power entry filter region.
• Correlation: tie near-field peaks to counters such as false-trigger count, input fault flags, and reset events.
• A/B tests: change only one variable (edge rate, shield termination, loop area, CMC insertion) and compare the fault rate.

Hardwire does real-time; the bus does configuration, telemetry, and logs

A lighting controller has two planes by design: a real-time plane for deterministic triggering, and a control plane for configuration, telemetry, and diagnostics. The bus is not used for µs-class triggering.

Trigger and status pins are the deterministic contract

• Trigger In: edge/level semantics, minimum pulse width, conditioning and isolation determine noise immunity.
• Strobe Out: event output for external coordination (a signal, not a transport).
• Ready: indicates the controller can accept triggers (e.g., not in derating or fault lock).
• Fault: provides immediate fault visibility; fault pin behavior must map to clear fault codes in telemetry.

Best practice: hardwire pins give deterministic behavior; the control plane provides the “why” with codes and event history.

Choose the bus by installation reality, not by feature lists

The control plane is an engineering layer for “evidence delivery”: it makes strobing diagnosable without adding timing uncertainty to the real-time chain.

Telemetry is not “temperature readback” — it is the evidence chain

• Pulse counters: per-channel pulses, peak-rate windows, total run time.
• Protection events: over-temp, over-current, open/short counts and last-occurrence snapshots.
• Operating states: derating active, current-limit active, restart attempts, lockout state.
• Input health: undervoltage counts and brownout markers (to distinguish load dip vs logic upset).
• Versioning: firmware version + configuration recipe version + last-change record.

Logs should answer “what happened, on which channel, under which recipe”

• Event log: event code + channel + timestamp + key snapshot (Iset, pulse width, temperature, input state).
• Fault codes: stable mapping from fault pin behavior to a readable code table.
• Snapshots: auto-capture on fault (derating state, input dip markers, last trigger count window).
• Export: read out via the control bus without interrupting the real-time plane.

A well-designed control plane makes “bus-only debugging” possible when probing the power stage is impractical in the field.

Validation is a proof chain: metric → setup → test points → criteria → record

“Works in the demo” is not a qualification. A lighting controller is validated by measurable evidence: current waveform integrity, optical energy repeatability, timing determinism, protection behavior, and EMC robustness.

Current waveform: rise/fall, peak accuracy, width accuracy, ringing

• Rise/Fall time (TP2)

  Measure t_rise/t_fall at the current stage output; verify against the motion/exposure window requirement.
• Peak current error (TP2)

  Compare I_peak vs setpoint across channels and temperatures; record min/typ/max.
• Pulse width error (TP2)

  Verify width accuracy over the supported range; check for quantization steps if using discrete timing.
• Overshoot/ringing (TP2)

  Record overshoot and damping; A/B test snubber and harness loop changes; confirm improvement is consistent.

Timing reference is the current rise edge, not the logic pulse edge. This prevents “false determinism” caused by power-stage variation.

Optical proof: photodiode waveform and energy repeatability

• PD waveform alignment (TP3)

  Capture photodiode output aligned to TP2; verify the optical rise follows the intended pulse shaping.
• Pulse-to-pulse energy repeatability (TP3)

  Compute energy proxy (integral or peak) and report repeatability (e.g., ±% over N pulses).
• Thermal drift curve

  Run 1h/8h at a fixed recipe; record optical output vs time and temperature; verify derating holds stability.

Optical output is the real deliverable; current regulation alone is not sufficient proof.

Timing: delay, jitter (TIE), and multi-channel skew

• Delay range and resolution (TP1→TP2)

  Measure mean trigger-to-current delay across programmable settings; record effective step size.
• Jitter (TIE) statistics (TP1→TP2)

  Report RMS, peak-to-peak, and tail (P99) jitter; check if jitter increases with CPU load or EMI conditions.
• Multi-channel skew (TP2)

  Observe multiple channels on the same scope trigger; record skew before/after per-channel offset calibration.

Reliability behaviors must be observable and repeatable

• Over-temperature derating

  Force temperature ramps; verify current/width/frequency limits engage as designed and recover per policy.
• Open/short protection

  Inject open/short faults; verify detection time, safe shutdown behavior, and fault code/log entries.
• Fault recovery strategy

  Test auto-retry vs latched lockout; confirm Ready/Fault pins and telemetry remain consistent and documented.

EMC is proven by correlation: pre-scan peaks vs event records

• Conducted/radiated pre-scan

  Near-field scan harness, switch node, IO, and power entry; record hotspots and compare after mitigation A/B changes.
• ESD/EFT event logging

  Apply disturbances and record fault type matrix: false triggers, resets, channel mismatch; preserve timestamps and counters.
• Bus-loss fail-safe

  Disconnect the control bus; verify deterministic strobing continues safely with local defaults and logs resume after reconnect.