Buck (non-isolated) fits when the LED string voltage stays well below the rectified bus with enough headroom for current regulation and transient margin, and when the output is not user-accessible as a SELV interface. Flyback becomes the safer choice when galvanic isolation is required by touch/safety boundary, when wide output-voltage compliance is needed, or when surge return paths and EMI filtering benefit from an isolated secondary. Example parts: TPS92515/LM3409 (buck), NCL30000 or LYTSwitch-5 (isolated flyback).

PWM-resolution limits typically show as discrete brightness steps: the light changes in “levels” and the LED current amplitude is stable inside each step. Control-loop instability or interaction tends to show as low-frequency modulation on the LED current (or output ripple) even when PWM duty is constant, often worsening with temperature, input ripple, or load changes. A fast split is to capture LED current and the control node: quantization looks step-like; loop interaction looks like oscillation or beat patterns riding on the current waveform.

Start by identifying which mechanism dominates: rolling-shutter aliasing (PWM edges sampled by the camera) or true light modulation (current ripple from the power stage). If the fixture uses PWM dimming, moving PWM above typical camera alias regions (often into the tens of kHz) can reduce banding, but it will not fix ripple-driven modulation in analog/CCR modes. Always capture LED current during the complaint condition: if ripple/beat exists with fixed duty, prioritize power-stage ripple and loop interaction; if current is clean but banding persists, tune PWM frequency and edge behavior.

Enclosure swaps often detune or shadow the antenna and shift the ground reference, so antenna keep-out, metal proximity, and cable routing should be checked first. However, lighting drivers can also inject broadband noise that reduces RF sensitivity, especially when the enclosure changes return paths or shielding. A fast split is to compare (a) RF link margin proxies (RSSI trend, packet retries) and (b) RF-rail integrity during TX bursts. If supply droop or spikes align with dropouts, fix power filtering/domain isolation; if RF margin collapses only with the new enclosure, fix antenna clearance and grounding.

The most common hidden standby loads are (1) the high-voltage auxiliary converter operating in a poor light-load region, (2) an always-on LDO feeding RF/MCU peripherals that were meant to be gated, and (3) sensors/AFE rails that keep bias networks alive via pull-ups or protection leakage. Domain-level debugging should isolate current by rail: measure always-on vs active rails separately, verify that enable pins truly de-assert, and remove unintended back-power paths (GPIO clamp diodes, I²C pull-ups to the wrong rail, ESD structures). A tiny reset supervisor (e.g., TPS3839) can prevent “brownout loops” that inflate average standby current.

In lighting drivers, “stable but wrong” often comes from sampling-point and phase errors before temperature drift becomes the main culprit. The current waveform is frequently non-sinusoidal and rich in harmonics; sampling at the wrong location (bus vs primary vs secondary) can miss real power flow or introduce timing skew. Phase error between voltage and current channels can bias real-power estimation even when RMS readings look consistent. Temperature drift then adds slow bias through shunt value change and analog front-end gain/offset. Lighting-level metering should prioritize consistent trends and protection thresholds rather than utility-grade absolute accuracy.

“Lighting-level enough” typically means repeatable energy and power trends (for reporting, limiting, and anomaly detection) rather than revenue-grade calibration. A minimum error budget can be built from a few dominant terms: shunt tolerance + tempco, current-sense amplifier gain/offset, ADC reference and quantization, and voltage/current channel timing skew. Practical choices often land around 0.5–1% shunt tolerance with reasonable tempco, a zero-drift current-sense amplifier (e.g., INA190) for stable gain/offset, and ADC/reference design that avoids drift under heat. If harmonics are high, phase/timing control can matter more than raw ADC bits.

If false triggers correlate with dimming transitions, switching-node activity, or surge events, EMI injection is the first suspect; if they correlate with environmental slow changes (50/60 Hz lighting, temperature drift, sensor warm-up), AFE bandwidth and filtering are more likely. PIR and photodiode front-ends are high impedance and sensitive to leakage: input protection, PCB contamination, and return-path noise can push the TIA into saturation or shift its baseline. A fast check is to scope the AFE output and input node during the trigger: injection shows sharp bursts aligned with power events; bandwidth issues show slow periodic modulation.

Capture all three if possible, but prioritize the chain that proves causality: start with the auxiliary rail and PG/RESET, then zoom into the RF rail during TX bursts. Dimming often creates load steps and ripple that momentarily collapses the aux converter or triggers brownout thresholds; PG/RESET confirms whether the MCU/SoC is being reset. If PG stays valid yet the radio drops, the RF rail may be dipping or being polluted by switching noise. Measurements should be taken at the RF supply pin (with local decoupling), not only at the regulator output, to expose layout-induced droop.

The most common field-only failure is an unintended return path created by real installation: cable routing, earth/chassis bonding, long impedance paths, or enclosure contact points that do not exist in the lab setup. A surge current that should be clamped at the entry can instead traverse logic ground, cross isolation capacitance, or inject into the always-on domain. Fixes are usually physical: place TVS/MOV at the true entry, shorten high-current loops, define a controlled chassis/earth reference, and keep sensitive returns (RF/AFE) away from surge loops. If isolation is used, verify that common-mode displacement current has a safe path that does not flow through signal ground.

Split by correlating flicker with thermal and power trends. Capacitor ESR drift often increases output ripple and worsens at high temperature, showing a clear rise in ripple amplitude over time. Magnetics heating can shift inductance and loss, changing switching behavior and creating audible noise or ripple changes near certain dimming levels. LED aging tends to show gradual lumen change and may shift the required current/voltage point, but it rarely causes sudden ripple unless the driver is near compliance limits. A quick method is to log input power, LED current ripple, and a few hotspot temperatures (MOSFET/magnetics/capacitor) under the same dimming profile.

A minimal, high-yield screen can target three failure families with few tests: (1) power integrity under dynamic load—step dimming and capture PG/RESET and aux-rail droop; (2) flicker susceptibility—measure LED current ripple at several dim levels and verify stability (no low-frequency beat); (3) interference robustness—inject a controlled disturbance (EFT-like burst or fast transient) and verify no latch-up/reset while monitoring key rails and AFE outputs. For RF dropouts, a simple enclosure-in-place margin check (RSSI trend + rail integrity during TX) can catch antenna detuning and noisy supply layouts without touching protocol details.