DEV Community: lamp nex

Why Your Ceiling Light Turns Yellow and Fills With Insects (It Was Never Dirt)

lamp nex — Fri, 06 Nov 2026 04:11:50 +0000

Pull down a three-year-old ceiling light and the scene repeats itself: the diffuser's inner wall has yellowed, and the base plate underneath holds a layer of insect husks and dust. The owner's first instinct is usually that it's time to clean.

It won't wipe off, and cleaning was the wrong frame from the start. Yellowing and insect ingress are two separate problems, and neither is dirt: one is bulk polymer aging at the operating temperature inside the lamp; the other is a protection rating that, for residential fixtures, was never designed to exclude insects in the first place.

Why insects get in: the IP rating was never meant to stop them

On the packaging, the two digits after IP: the first digit covers solid foreign objects.

Rating	Blocked objects	Insects?
IP2X	12.5 mm and above — a finger	Not blocked
IP3X	2.5 mm and above — tools, thick wire	Not blocked
IP4X	1.0 mm and above — includes mosquitoes, flies, small insects	Mostly blocked

The vast majority of residential ceiling lights, downlights, and strip-light power supplies are rated IP20 or IP23 — finger-safe, not waterproof, not insect-proof. Under the standard's own terms, "insects got in" is not a defect; the rating permits it from the factory. Real insect exclusion starts at IP4X, which costs noticeably more and hurts thermal performance — and given how much of a residential fixture's budget goes into optics and the driver, most manufacturers stay at IP20.

The breathing effect: insects get sucked in

Even when the gap is only a fraction of a millimeter, insects are not crawling in by coincidence.

When the lamp is on, the LEDs and driver generate heat, the air inside the cavity expands, pressure rises above ambient, and gas pushes out through every seam. When the lamp switches off and cools, cavity pressure drops below ambient and the resulting negative pressure pulls outside air back in — along with moisture, dust, and small insects. This is the breathing effect, the same physics as PV = nRT: a fixed amount of gas in a sealed-ish cavity, cycling between heating at constant volume and cooling at constant volume, with pressure driving flow through whatever leaks exist.

This is not a hypothesis. The luminaire safety standard IEC 60598-1 (the Chinese equivalent is GB 7000.1) requires that dust and water ingress tests be run only after the lamp has been brought up to its stable operating temperature — a cold lamp doesn't leak; a hot one does. Outdoor luminaires handle the pressure differential with ePTFE breathers. Residential ceiling lights have no such part; they rely entirely on structure and gaskets.

A lamp breathes roughly seven hundred times a year. Insects are also phototactic — they fly toward the light. That is why the husks always accumulate directly above the LEDs.

Five entry paths: the diffuser-to-base overlap surface (no gasket), the cable pass-through hole in the base (unsealed), the fixing screw holes, snap-fit gaps, and the driver's wire exit.

Yellowing is not a stain; the plastic itself has changed

Diffuser plastics come in three main types, and their temperature tolerance differs sharply:

Material	Light transmittance	Temperature tolerance	Yellowing tendency	Cost
PMMA (acrylic)	92–95%, highest	Long-term max 80 °C	Best outdoor weatherability; limited inside sealed hot cavities	Upper-middle
PC (polycarbonate)	88–90%	Heat deflection ~135 °C	Unstabilized grades yellow under UV; needs a UV-protected grade	High
PS (polystyrene)	85–92%	Long-term 60–80 °C	Yellows fastest, and embrittles	Lowest, about half of PC

The mechanism is not complicated: heat and UV break and oxidize the polymer chains, forming structures that absorb blue light. Once blue is eaten out of the spectrum, the transmitted light shifts yellow-green — the material has changed color, not the surface.

Sandpaper and polishing compound only remove surface grease and smoke film. Bulk yellowing, ground down one layer, is still yellow underneath — and the freshly ground surface is matte, which collects dust faster. Most "whiten your lampshade" tutorials online are treating stains as yellowing.

PS has it worst: its long-term use ceiling is only 60–80 °C, while the cavity temperature inside the lamp tracks power draw and ambient temperature. Summer, closed windows, fixture flush against the ceiling — the material sits against its upper limit for months at a time. Yellowing is inevitable.

The consequences go beyond looks. Transmittance drops, so the same lamp looks a step dimmer two years in. The diffuser becomes a yellow filter: a 4000 K lamp reads as roughly 3500 K, and color rendering drifts with it. This is distinct from LED color binning tolerance — that is lot-to-lot variation at the LED factory; this is degradation introduced by an aging diffuser.

Dead insects are not just ugly — they can cook the driver

Insect husks and dust tend to pile up near the driver board and the terminal block — the two hottest, highest-voltage points in the whole fixture.

Dry insect matter is an insulator. Damp, it is not. Accumulated between driver pins or across terminals, it forms weak conductive paths — the industry term is creepage. Given more time at high temperature, the husks carbonize; carbon conducts, and the path becomes more solid. The result: localized short circuits, a dead driver. Replacing the LEDs at that point is wasted money — the driver is gone, the bus voltage is wrong, and new LEDs will fry too.

The more common failure is chronic: husks and dust block the heat-dissipating surfaces, driver temperature climbs, electrolytic capacitors age faster, and a year or two later the light flickers and dims. The user concludes the lamp is low quality. In reality, the thermal path was buried.

Four-step inspection, power off: confirm de-energized with a voltage tester → check the LEDs for black spots → check the driver board for husks, scorch marks, bulging capacitors → check whether the gasket has gone hard or brittle and whether the cable hole still has its plug.

Three red lines: no water (it reaches the driver and terminals), no alcohol or bleach (PC and PS undergo stress cracking in contact with alcohols and alkalis), and never wipe a powered fixture.

Three questions before you buy, two jobs at install

What is the diffuser made of? You want UV-stabilized PC or acrylic. If the seller can only say "premium plastic" without a grade, assume PS — expect yellowing within two to three years.
Is there a gasket? Silicone beats EVA foam; foam embrittles and flakes off within a couple of years, leaving a wider gap than before.
Do the cable holes have plugs, and can the diffuser be removed by hand? A diffuser you cannot remove means the insect layer inside can never be cleaned.

Two jobs at install: plug the cable pass-throughs and unused screw holes with plugs or neutral-cure silicone (do not seal the drainage or breathing points the manufacturer left open), and run a thin bead of sealant along the base-ceiling joint, leaving a removable section.

For an existing lamp, four remediation steps: remove the diffuser → dry-brush and vacuum out the husks (no wet cloth) → replace the gaskets with silicone ones → plug the holes and reassemble. A lamp that cannot go through these four steps is, in essence, disposable.

FAQ

Can a yellowed diffuser be washed back to clear?
Grease and smoke film wash off; bulk yellowing does not. Compare the color against incident light after wiping — if it is uniform, it is bulk change, not a stain.

Do mothballs keep insects out?
Not recommended inside the lamp body; the internal temperature drives continuous volatilization. The only real defense is sealing the entry paths.

Yellowing and insect ingress both look like "wear and tear." But half of the outcome was fixed the moment the lamp left the factory: which material was chosen, whether the holes were plugged, whether a gasket was installed. The user only controls the other half.

Your LED Bulbs Might Be Failing a Test You've Never Heard Of: Harmonic Current

lamp nex — Thu, 05 Nov 2026 04:11:16 +0000

On October 29, 2026, Guangdong Provincial Market Regulator released announcement No. 132 of 2026: 15 non-compliant lighting products, 6 found online (3 on Tmall, 1 on Douyin, 1 on JD, 1 on Kuaishou) and 9 offline. The single most frequent failing item? Harmonic current — it appeared 11 times, across bulb lamps, panel lights, floodlights, and recessed luminaires.

If you work in lighting, this won't surprise you. If you're a homeowner, smart-home tinkerer, or DIY electrician, it probably should — because the problem isn't confined to one provincial sampling run, and the symptoms can show up in your own walls.

Not a One-Off

In another Guangdong batch sampled the same month, two well-known-brand ceiling lamps and a pillar bulb also failed on harmonic current. Over in Shandong's provincial sampling, 3 batches of Opple LED bulbs and 1 Panasonic luminaire failed the same item. Opple attributed it to old-standard inventory, which may well be true — but it tells you how long non-compliant stock can linger on shelves and in warehouses.

There's also a quiet industry consensus worth stating plainly: budget retail and online-exclusive models carry higher harmonic risk than engineering-grade products. The cheapest listings on e-commerce platforms are exactly where you should expect the worst current waveforms.

How Can a Lamp "Work Fine" and Still Fail?

Harmonics measure how cleanly a lamp draws current from the mains. A lamp can light up, dim smoothly, and pass every casual eyeball test while its current draw looks like a mess on an oscilloscope.

Here's the mechanism. An LED driver rectifies the AC mains and buffers it with a bulk capacitor. That capacitor only charges near the voltage peaks, so instead of a smooth sinusoidal current draw, the lamp gulps narrow current pulses. Those pulses are full of odd harmonics — mainly the 3rd harmonic (150 Hz on a 50 Hz grid) — which do no useful work at all. They just flow back and forth, producing nothing but heat.

The Standard: GB 17625.1—2022

China's current harmonic limit standard, GB 17625.1—2022 (effective July 1, 2024, a modified adoption of IEC 61000-3-2:2020), classifies lighting as Class C with three power tiers:

Power tier	Requirement
> 25 W	Relatively strict: 3rd ≤ 27%, 5th ≤ 10%, 7th ≤ 7%, 9th ≤ 5% (as % of fundamental current)
5 W – 25 W	Three alternative requirements; the cheapest option only limits 3rd ≤ 86% and 5th ≤ 61% — no limits on other orders or THD
< 5 W	No emission limits at all (new in the 2022 edition)

Now here's the uncomfortable part for residential lighting: home downlights, spotlights, and bulbs are mostly 5–12 W. LED strips and cabinet lights are often under 5 W. In other words, the lighting that fills your ceiling sits exactly in the loosest tiers of the standard — where a driver with no meaningful harmonic control can still "pass."

Why Your Electrician Should Care: The Neutral Problem

The 3rd harmonic isn't just an abstract compliance checkbox. It's a triplen harmonic, which means it adds in the neutral conductor instead of cancelling out. In a three-phase system this is the classic fire-engineering headache. Homes are single-phase, so the picture is simpler but still ugly: more lamps on one circuit means more neutral current, period.

Fire-protection design code notes are blunt about this: ≤ 25 W lamps allowed an 86% 3rd harmonic can push neutral current above 2.5× the phase current. The neutral — often the same gauge as the phase conductors — becomes the hottest wire in the circuit.

Symptoms, Layer by Layer

What does excessive harmonic current actually look like in a house? It tends to announce itself in three layers:

1. Perceivable (annoying):

Audio hum or buzz from equipment on the same circuit
Touch switches acting up
Motion-sensor lights triggering by themselves

2. Systemic (worrying):

Warm terminals or socket plates on that circuit
Nuisance breaker or RCD trips with no cause found

3. Long-term (expensive):

Terminal oxidation and insulation aging from chronic overheating — the damage that shows up years later, long after you've forgotten what brand of bulb you installed

A 3-Step Self-Check

If you suspect your lighting circuit, here's a practical sequence:

Switch off the suspect circuit and see whether the symptoms vanish. Instant correlation, zero cost.
Count the lamps. 20+ non-PFC 5–12 W lamps on a single circuit means elevated risk. Spread them across 2–3 circuits.
Have an electrician measure neutral current with a clamp meter. Neutral current near or above phase current is the telltale signature of triplen harmonics. While they're at it, they should check terminal temperature.

Two Questions to Ask Before Buying

When you're shopping for downlights or bulbs — especially in bulk for a renovation — ask the seller two things:

1. What's the driver's power factor — does it have active PFC?

PF ≥ 0.9 means a serious driver with active power factor correction. A PF around 0.5 means no PFC at all. But be careful: high PF ≠ low harmonics. Simple RC filtering can raise the PF while total harmonic distortion stays sky-high. PF is necessary but not sufficient.

2. Do you have a harmonic current test report per GB 17625.1—2022?

A vendor who can produce this report on request is a different breed from one who sends you marketing PDFs.

The 3C Certificate Fine Print

One nuance that trips people up: CCC (3C) certification covers the certified model and its driver scheme. Swapping the driver, the capacitor, or the component supplier requires re-evaluation. Some of the products flagged in this round did have valid certificates — the problem was in the batch or materials, not the paperwork. A certificate is a starting point, not a guarantee.

What It Looks Like from the Factory Floor

Here's the part that explains everything else. A proper active PFC stage — PFC inductor, high-voltage film capacitor, control IC, driver MOS — costs roughly 1–2.5 RMB at volume. On a downlight retailing for twenty-something RMB, that's pure margin walking out the door.

So cost-cutters have three ways to trim it:

No PFC at all — PF below 0.5, harmonics uncontrolled.
Valley-fill pseudo-PFC — PF around 0.85, but poor THD, and it flickers at low input voltage.
PFC without the input common-mode choke — harmonics barely pass the limit, but conducted EMI (conducted disturbance voltage) fails — which, notably, was also a frequent failing item in this same round of sampling.

That last trick is the clearest proof of the point: cutting one corner pushes the failure to the next invisible item down the checklist.

The Bottom Line

Harmonic current is a cost trade-off, not a technical challenge. Solving it is cheap and well understood. But it's invisible to users, only measurable with instruments, and nobody returns a lamp because of its THD — so in a price war, it's the first thing to be sacrificed.

Two FAQs worth answering directly:

My lamp lights and dims normally — can it still fail compliance? Yes. Harmonic current, conducted disturbance, and energy efficiency are all invisible items. Normal operation proves nothing about compliance.
Doesn't buying a big brand guarantee compliance? No. Well-known brands were flagged in this round and in Shandong's sampling; Opple blamed old-standard inventory. Brand reduces risk; it doesn't eliminate it.

Key Takeaways

Harmonic current was the most frequent failing item (11 of 15 products) in Guangdong's Oct 2026 lighting sampling, hitting bulbs, panels, floods, and recessed luminaires alike.
Most home lighting (5–12 W lamps, < 5 W strips) falls in the loosest tiers of GB 17625.1—2022, where weak drivers can still pass.
3rd harmonics add in the neutral: symptoms range from audio hum and ghost-triggering sensors to warm terminals, nuisance trips, and years-later insulation aging.
Self-check: kill the circuit and watch symptoms, count your lamps (20+ non-PFC units per circuit = spread them out), and clamp-meter the neutral.
When buying: ask for active PFC (PF ≥ 0.9) and a GB 17625.1—2022 harmonic test report — and remember high PF alone doesn't mean low harmonics.
Active PFC costs roughly 1–2.5 RMB at volume; its absence is a pricing decision, not a technology gap.

WiFi vs Bluetooth Mesh Ceiling Lights: The Difference Only Shows Up When the Power Comes Back

lamp nex — Wed, 04 Nov 2026 04:11:25 +0000

This year's Double 11 shopping festival turned the ceiling light into the entry ticket to a whole-home smart setup. Home tour posts and discussion threads piled up around this one product category. But the question that showed up most in the comments wasn't "which brand is better." It was simpler and much more expensive to get wrong: the same lamp, on the same product page, comes in a WiFi version and a Bluetooth Mesh version, at almost the same price. Which one do you buy?

A lot of people grab the WiFi version without thinking. No gateway, plug it in, pair it with the app — sounds like the least hassle by far. Six months later they find out that the difference between the two versions was never about whether the lamp can talk to your phone. It's about whether the lamp comes back after the power goes out.

You can't tell in daily use. It shows up in the second the power returns

This isn't an occasional glitch. It's consistent, which is exactly why it's easy to miss: manual switching works, brightness adjustment works, everything works for the first few weeks, so there's nothing to notice.

The difference happens the moment power is restored.

A WiFi lamp being powered on is not the same thing as being online. It has to request an address, find a route to the internet, establish a connection to the cloud, and complete registration. Only after that whole sequence finishes does the cloud consider the lamp online, and only then can the app or an automation find it. When power comes back, every WiFi device in the house is walking through that same sequence at the same time. The router gets crowded, and a few minutes of dropped connections is normal — sometimes you have to redo the network setup entirely. A lighting scene needs one thing: this specific lamp is online and can receive commands. If it drops offline, the whole automation chain breaks right there.

The Bluetooth Mesh version doesn't take that path. Automation scenes live on the gateway, locally. Commands complete inside the local network and never leave the house. After power is restored, local automations run as usual. There's no "re-registration" step to fail.

The two versions, side by side

	WiFi version	Bluetooth Mesh version
How it connects	Direct to the router, no gateway needed	Connects to a gateway; lamps relay for each other
After power restore	Reconnects to cloud, often drops	Runs locally, back in seconds
Usable without internet	Basically dead	Local automation works with a central gateway
Standby power per lamp	About 2W, 10x the Mesh version	About 0.2W
Router device load	20–30 devices is the limit	Doesn't consume router channels
Group control	Each lamp talks to the cloud separately, uneven latency	Queued all-on/all-off handling is optimized

The standby power row is worth a second look. A lamp is wired to mains and sits there 24 hours a day. At roughly 2W versus roughly 0.2W, that's a 10x difference in idle draw, multiplied by every lamp in the house, running every hour of the year. For one fixture it's trivial. Across a full home, it's a line item.

The router load row matters just as much, and it's the one people hit first. A typical home router handles 20–30 devices before it starts to strain. Smart lamps are the easiest devices to accumulate — you buy one per room, then one for the hallway, then the balcony. A WiFi lamp occupies a router slot. A Mesh lamp hands that job to the gateway and leaves the router's channel budget for phones, laptops, and the TV.

About Mesh 1.0 and 2.0: the version number isn't the feature

One clarification, because this is where a lot of buying decisions go wrong.

Whether a lamp still works when the internet is down depends on the local-execution architecture, not on the version number by itself. Both generations can do it, provided a gateway is present.

What Mesh 2.0 genuinely improves is three things:

Provisioning time drops from several seconds to under 1 second.
Command response is pushed to under 100 milliseconds.
Over-the-air firmware updates are supported.

The third point has a flip side that matters more than the first two: a 1.0 device cannot be upgraded to 2.0 in software. The only path is new hardware. And these are devices that go into a ceiling and get buried behind drywall. Buy wrong once, and you tear things apart once.

Choosing by scenario: one sentence each

A single lamp that will never join an automation: the WiFi version is enough, and you save the cost of a gateway. Secondary bedrooms, storage rooms, anywhere the requirement is just "it lights up" — there's no reason to pay a Mesh premium.

A whole-home smart setup: insist on Bluetooth Mesh 2.0 plus a central gateway. The ordering of decisions matters here: define the system first, then the brand, then the specific model. People who do it in reverse almost always end up redoing the work.

Watch out for the proprietary-gateway wall. Some models from the same brand require a dedicated gateway and won't join the general Mesh ecosystem. Before ordering, ask one question: "which gateway does this lamp use?" If the answer doesn't contain the words "Bluetooth Mesh 2.0," you're probably looking at a proprietary line. You'll find out after installation, when dimming doesn't respond, and returns are a headache.

Already bought the wrong one? Three fixes that don't involve opening the ceiling

The cost of tearing down a ceiling and re-cutting holes is far higher than the problem itself. Try these three first.

1. Keep the lamp permanently powered. The root cause of the dropout is that the lamp has to reconnect to the cloud after power is cut. So don't cut its power. Keep that circuit energized and the lamp stays online.

2. Switch the smart switch to wireless mode. The switch stays in the same place and the panel stays on the wall. It just stops physically breaking the circuit and only sends a signal instead, with on/off handled by an automation. Three preconditions: the lamp supports "power-on default off," the switch itself supports a wireless mode, and you create an automation in the app that catches the button press.

3. Add a central gateway. A Mesh lamp without a gateway still has no local automation. One related oddity worth knowing: some users see a "no gateway connected" status in the Mi Home app while everything actually works fine. The cause is often that the "hide individual lamps" toggle in the lamp group management got switched on by accident. Check that setting before blaming the hardware.

Four checks before your Double 11 order arrives

1. Check luminous flux. If the product page doesn't even list a lumen figure, that's a deduction no matter how smart the lamp claims to be. A rough sizing rule: room area in square meters times 200 lumens.

2. Check the firmware. When the lamp arrives, look at the firmware ID. Only modules starting with 304 are Mesh 2.0. Don't trust the marketing copy on the product page.

3. Check the dimming floor. Ask support what the lowest brightness is as a percentage, and whether it remembers its state after a power cut. This is the dividing line between a smart lamp and a lamp that merely has an app.

4. Check protocol dependencies. For a WiFi version, count how many more devices your router can actually carry. For a Mesh version, confirm the gateway model — don't buy the lamp and only then discover the gateway doesn't support it.

FAQ

Is the WiFi version simply a bad buy?

No. For locations that won't join automations and won't ruin your day if they go offline, WiFi is fine. But a lamp is a device that participates in automation and gets its power cut. Don't gamble on WiFi there.

Does Bluetooth Mesh absolutely require a gateway?

Yes. Your phone's Bluetooth can pair with a single lamp, but "works without internet" local automation requires a central gateway. Those are two different things.

Closing

Judge the lamp by its light: color rendering, glare control, color temperature. Judge how smart it is by the protocol line on the spec sheet. Don't stop at "supports app control." Look for the words "Bluetooth Mesh." Get that right and this Double 11 order won't be wasted.

Excerpt: Same lamp, two versions, nearly the same price. The difference isn't app control — it's whether the light comes back online after a power cut, and how much it costs you in standby power and router slots.

Your New LED Lamp Buzzes - and It's Almost Never the LED

lamp nex — Tue, 03 Nov 2026 04:14:32 +0000

Your New LED Lamp Buzzes — and It's Almost Never the LED

Every November, right after Double 11, my inbox turns into a noise-complaint hotline. This year the flood started about two weeks ago, and the question is almost never about wiring anymore. It's this one: "I hung the lamp last night and it won't stop humming. Is it defective? Should I return it?"

I've spent twenty years in the lighting industry, most of it on the driver side of the business, so I've heard this a few thousand times. Here's the short answer I give everyone.

The LED chip cannot make noise.

It's a slab of semiconductor. Current in, photons out. There is no moving part anywhere inside it. Every sound coming out of that fixture originates in the one component you never look at — the driver, that small box tucked into the ceiling plate or the junction box.

And inside that box, things genuinely move. Transformers, inductors, capacitors. When current flows, the magnetic material expands and contracts by microns, thousands of times a second. That's not a defect in itself. It's physics.

The good news: because the mechanism is physical, the noise is diagnosable. Describe the sound accurately and you can usually decide whether to return the lamp without ever opening it.

Three sounds, three different diseases

What you hear	Frequency	Character	Usual source
Hum	50–200 Hz	Low, dull, fridge-like	Core magnetostriction, oscillating with 100 Hz mains ripple
Buzzing / crackle	100 Hz – 5 kHz	Fine, mosquito-like	High output ripple, loose winding, vibrating MLCC capacitors
Squeal / whistle	15–20 kHz	Sharp and piercing; young ears hear it, older ears don't	Switching or PWM dimming frequency landing inside the audible band

Once you know which row you're in, the rest of the diagnosis is mostly mechanical.

Root cause 1: the core is breathing

The transformers and inductors in a driver use ferrite cores, and ferrite has a property called magnetostriction — put it in a magnetic field and it changes dimension very slightly.

A switching supply runs somewhere in the 50–150 kHz range, far above the 20 kHz ceiling of human hearing. In theory you should hear nothing at all. And most of the time you don't.

The problem is the second-order ripple. A single-stage PFC design paired with an undersized output capacitor leaves a 100 Hz component riding on the output current. That ripple flows through the output inductor, so the core expands and contracts at 100 Hz — and 100 Hz sits right in the most sensitive region of human hearing. What reaches your ear is a low, steady "hmmmmm."

This also explains the classic complaint: "I can't hear it during the day, but at night it's unbearable."

The sound never changed. During the day it's buried under ambient noise. Close the windows in winter, wait until the house goes quiet, and the masking effect disappears. The hum was always there. You just started hearing it.

Root cause 2: the housing became a speaker

Here's a pattern I see constantly: the same driver measures perfectly quiet on the bench, then howls the moment it's installed in the fixture.

That's mechanical resonance, not electrical. The vibration energy is tiny, but the aluminum substrate, the thin-walled housing, and the mounting bracket each have their own natural frequency. When the excitation hits one of them, amplitude gets multiplied and the whole housing turns into a speaker cone.

The test is simple. Cut the power, pull the driver out of the fixture, suspend it in free air, and power it back up.

Quiet when suspended, loud when bolted back in? That's structural resonance. A gasket or some damping compound between the driver and the housing usually fixes it. You don't need to return the lamp.

Root cause 3: dimming dragged the frequency into the audio band

Dimmable fixtures fail this test far more often, for three separate reasons.

PWM frequency set too low. If the driver switches below 20 kHz, the current is being chopped at an audible rate, and the pitch will actually shift as you change brightness. Industry practice is to push the PWM frequency above 8 kHz so it climbs out of the range where the ear is most sensitive.

Mismatched dimmer and driver. Old triac dimmers were designed for incandescent loads. They dim by chopping the AC waveform. An LED driver that can't interpret that chopped waveform will buzz, flicker, and refuse to dim smoothly.

Load too far below the dimmer's minimum. Dimmers specify a minimum load. Swap incandescent for LED and the wattage can fall to a tenth of what the dimmer expects, so it operates in a non-linear region and generates garbage on the line.

The signature symptom of all three: loudest at mid-dim, quiet at full brightness. At mid-range duty cycle the current pulses are at their steepest.

Load matters too. Many drivers are essentially silent from 80% to 100% load and start signing below 20%. Noise has a load curve. That's why two identical lamps in the same room can behave completely differently.

The four-step check you can do without a multimeter

Step one: locate by ear. Turn off every other appliance in the room and run only this fixture. Is the sound coming from the driver box in the ceiling plate, or from the junction box, or from the switch panel? Driver means driver problem. Junction box means wiring problem.

Step two: swap sockets. If the noise follows the lamp, it's the integrated driver. If the new location is silent, the problem is the original socket or circuit.

Step three: bypass the dimmer. Wire the fixture to a plain switch. If the noise disappears, you have a compatibility problem, and the fix is a different dimmer.

Step four: look for companion symptoms.

What you observe	Verdict	Action
Faint hum, steady volume, no flicker, housing at room temperature	Normal electromagnetic vibration	Nothing to fix
Volume rising and falling, plus flicker and noticeable heat	Unstable driver, genuine fault	Stop using it, file a warranty claim
Sound at the junction box, burning smell	Loose terminal arcing — a real safety hazard	Cut power now, call an electrician
Sound changes with dim level, gone at full brightness	Dimmer compatibility	Replace the dimmer

That third row deserves its own paragraph. A terminal that was never tightened, or oxidized wire, produces tiny arcs under high current. It isn't just noise. Arc heating burns through insulation. If you smell something burning, kill the breaker before you do anything else.

If you do want a return, write it like an engineer

Most returns stall at "unit tested, functioning normally," because support is checking whether the lamp lights up. Give them something reproducible instead.

Describe frequency, not adjectives. "In a quiet room at night, a continuous low-frequency hum is audible at one meter" beats "it's really loud."

Describe the operating condition. "Most pronounced between 30% and 50% dim, absent at full brightness." That's a compatibility statement, not a subjective impression.

Give them a control. "A second unit of the same model on the same circuit is silent; after swapping positions, the noise follows the lamp."

For context, the quiet-operation threshold used by programs like ENERGY STAR is roughly 30 dB at one meter. That's a number you can cite.

Four questions worth asking before you buy

Noise performance is essentially locked in at the factory. It is not something you can fix after the fact, so ask before you pay.

Is the driver potted? Epoxy or silicone fill does two things at once — it damps coil vibration and it blocks the path into the housing.

What dimming protocol does it use? 0-10V and DALI are inherently cleaner than triac phase-cut.

Can the driver be replaced on its own? On an integrated fixture, a failed driver means a new fixture.

Is the housing a thin-wall spun or drawn part? Thinner walls resonate more easily.

Two questions I get constantly

Q: The lamp is brand new and already humming. Is it refurbished stock?

A: Not necessarily. A faint hum on a new fixture is within normal engineering range. What matters is whether the volume is stable, and whether flicker or heat comes with it.

Q: If the driver is potted, is it guaranteed silent?

A: No. Potting addresses vibration transmission and housing resonance. If the ripple is large, or the dimming frequency sits in the audio band, you'll still get noise — it just becomes duller and quieter.

The one-line version

The lamp isn't noisy because of the LED. It's noisy because a handful of magnetic and capacitive components inside the driver are vibrating, and something in the fixture is amplifying them.

Diagnose by frequency first. Then run the four steps with the power off. Then check for flicker and heat. Flicker plus heat means a fault. Neither means a trade-off.

Your lamp arrived two weeks ago and it's buzzing. Don't rush the return label. Ten minutes of careful listening will tell you more than the product page ever will.

Tags: lighting, hardware, smarthome, diy

Why Your LED Strip Falls Off Every Winter (And Why It's Not the Glue's Fault)

lamp nex — Mon, 02 Nov 2026 04:09:38 +0000

Every winter, without fail, the same phone calls come in: "My LED strips keep falling off the wall. They were stuck on solid when I installed them!"

After two decades in the lighting business, I know this story by heart. Strips mounted in summer hold perfectly, start curling at the edges when the cold sets in, and are dangling loose by spring. The common reaction is to blame counterfeit products or "cheap adhesive." But in my experience, roughly nine out of ten failures have nothing to do with the glue brand — the real problem is that the installation temperature was never right in the first place.

How pressure-sensitive adhesive actually works

The adhesive on the back of an LED strip is a pressure-sensitive adhesive, or PSA — 3M VHB is the most famous family in this category. It doesn't "stick" the way you might imagine. It works by flowing: under pressure, the adhesive gradually flows into the microscopic bumps and valleys of the mounting surface, filling the contact area the way water spreads over sand. The industry term for this is wetting.

Wetting is extremely temperature-sensitive. For most strip backings on the market, the minimum application temperature sits around 10–15°C. Below that, the adhesive layer hardens toward a glassy state and simply stops flowing. Press the strip against the wall at that point and it looks attached, but only a small fraction of the glue is actually in contact with the surface. The initial tack is an illusion — during the 24–72 hour cure window, real strength never develops. Add the strip's own weight plus constant thermal cycling, and failure is only a matter of time.

Winter has one more accomplice: condensation. During heating season, room air is warm while metal studs, glass, and tile stay cold. Invisible moisture films form on those surfaces, and adhesive cannot bond to water. This is why an aluminum channel that mounted fine in summer mysteriously drops off in winter.

Five failure points to check yourself

Failure point	What happens	The right move
Installing below 10°C	Adhesive stiffens and can't flow; tack is fake	Warm both the surface and the strip to 15–25°C first
Wiping with just a damp cloth	Oil film remains; bond strength can drop by as much as 80%	Wipe with isopropyl alcohol and wait 3–5 minutes for it to fully evaporate
Applying to painted drywall or plaster	The substrate itself powders and flakes; the glue holds powder	Switch to screw-mounted aluminum channels, or use a primer first
Peeling the liner slowly and repositioning	Exposed adhesive collects dust and finger oils	Peel and stick in one motion, no second chances
Loading the strip immediately	The bond gets stressed during cure and never reaches full strength	Press firmly for at least 30 seconds, then leave it unloaded for 24 hours

One extra warning for installers in a hurry: don't blast the adhesive itself with a heat gun. Technical documentation from adhesive manufacturers notes that localized high heat damages the adhesive structure and accelerates long-term aging. If you're going to heat anything, heat the mounting surface — never the backing adhesive.

The four-step winter installation routine

Warm up. Bring the mounting surface to around 20°C with a hair dryer (it should feel warm against the back of your hand). Let the whole roll of strip acclimate in a heated room first — don't unbox and install it straight from the cold.
Clean. Wipe with isopropyl alcohol to remove oils, wax, and dust, then let it evaporate completely. On metal surfaces in winter, double-check there's no condensation and keep everything dry.
Press. Peel the liner as you go, sealing one section at a time, and press each section firmly for at least 30 seconds. The more pressure you apply, the better the wetting.
Wait. For the next 24–72 hours, don't hang anything on the strip and don't keep switching it on and off to test. The final bond strength "grows" during this window, and cold temperatures slow it down.

Adhesive alone isn't a mounting system

Let me be blunt after years in this trade: backing adhesive is for temporary positioning, not structural mounting. Here's how I'd rank mounting methods by reliability:

Scenario	Recommended mounting
Dry indoor spaces, flat cabinet panels, short-term decorative use	Backing adhesive alone is acceptable
Porous surfaces (concrete, drywall, raw wood) or cold environments	Adhesive plus primer (e.g. 3M 94); test a small section first
Ceiling mounting, runs over 5 meters, cove lighting that stays on for hours	Aluminum channel with clips; adhesive is just a helper
Outdoors, facades, cold rooms and other extreme environments	Mechanical fixing with screws or rivets — don't gamble on glue

The strips themselves don't mind the cold — LEDs actually run more comfortably at low temperatures. It's the adhesive and the installation process that suffer. If you can't bring the environment up to temperature, don't make the glue carry the weight alone.

Already fallen? Here's the recovery plan

Step one: clean off the old adhesive residue and any dust film on the substrate. Step two: check the season and room temperature — if the surface is still cold, warm it up before doing anything else. Step three: re-clean, and where needed, apply new sections of 3M VHB tape at intervals. Step four: if this is the second time the same strip has fallen, stop re-doing the same thing. Add an aluminum channel or clips and go mechanical. If a spot has failed twice, that substrate is simply not suited to adhesive.

When buying strips, ask about the adhesive part number

Reputable manufacturers list their backing adhesive in the specs: 3M 9495LE (friendly to low surface energy plastics) and 9495MP (for high surface energy metals) are common on SMD strips, while COB strips typically use VHB. For winter outdoor installs or unheated spaces up north, ask specifically about a low-temperature application series — 3M VHB GPL is rated for frost-free surfaces down to 0–5°C, and 3M 56415 paper tape can be applied at 0°C. If a seller can't tell you the adhesive part number, treat that as a warning sign.

Quick FAQ

Q: It held fine at install time. Why does it let go days later?
A: PSA strength builds over the first 24–72 hours after application. In the cold, wetting was incomplete, so the bond never develops — a few days of the strip's own weight is enough to pull it apart.

Q: Can I hit it with a hair dryer afterward to strengthen the bond?
A: Not on the adhesive. Heat damages the adhesive structure and accelerates aging. If you want to use heat, warm the substrate before installing — never bake the glue afterward.

Q: What about outdoor holiday lighting installed in winter?
A: Work during the warmest part of the day, let metal channels come up to temperature first, use the adhesive only as an aid, and let clips or screws do the real holding. For outdoor accent lighting, the combination of the GPL low-temperature series plus mechanical fixing is the standard recommendation.

If you're racing the winter schedule, save the four-step routine and the mounting table. Reworking a fallen strip costs far more than the ten extra minutes it takes to do it right the first time.

Your LED Light Takes a Few Seconds to Turn On? It's the Driver Capacitor, Not the LEDs

lamp nex — Fri, 30 Oct 2026 04:07:11 +0000

Last week a customer called about the light on his parents' balcony. "It's broken," he said — flip the switch at night and you wait two or three seconds before anything happens. He wanted a quote for a replacement LED panel.

I asked him to do one thing: switch it off, then on again immediately. It came on instantly.

So the light wasn't broken. That sequence — slow after sitting off, instant right after being on — has already completed most of the diagnosis: the problem is in the driver's startup stage, nothing to do with the LED chips, and swapping the panel is money down the drain. Once winter arrives, calls like this come in waves.

Three Kinds of "Slow" — Only One Is a Fault

Soft-start: normal. Any decent driver includes a power-on soft-start to suppress inrush current and protect the LEDs. The signature is a 0.3–2 second ramp-up that's identical on day one and a year later, with no loss of brightness.

Capacitor aging: fault. The signature here is that it gets progressively slower: a fraction of a second at first, two or three seconds a year later, then flickering before it stabilizes, occasionally coming on at half brightness, and finally not lighting at all. This is the most common failure path for household LED fixtures.

Switch or wiring mischief. Smart switches, wall switches with indicator LEDs, and single-live-wire powered switches all leak a tiny current when "off," confusing the driver's power-on logic. The result: delay, faint flicker, or a light that never fully goes dark.

Telling them apart takes one move: swap in an ordinary mechanical switch and test again. If the symptom disappears, it was case three. If it stays and keeps getting worse, it's case two.

Why It's the Capacitor, of All Things

LED chips don't mind the cold — at lower temperatures their efficiency actually improves slightly. The component in the whole fixture that hates the cold the most, and ages first, is the aluminum electrolytic capacitor on the driver board.

A capacitor conducts ions through an electrolyte. At room temperature the electrolyte is thin, the ions move fast, and the equivalent series resistance (ESR) stays low, so the capacitor charges to its working voltage almost instantly at power-on. Drop the temperature and the electrolyte thickens — ESR climbs by multiples. At -20℃ it runs three to five times its room-temperature value, a commonly cited industry figure.

With high ESR, charging slows down at power-on, and the output takes longer to climb to the chip's startup threshold. That's your two or three seconds. Once the circuit is running and its own heat warms the board, the capacitor regains its liveliness and everything behaves. So the symptom is "slow the first time, fine afterward" — not slow all the time.

Heat is the other enemy. Electrolytic capacitors follow a rule of thumb: every 10℃ rise in temperature roughly halves their lifetime. Sealed inside a downlight with the ceiling plenum hitting fifty degrees in summer, one ages several times faster than in a ventilated spot — which is why, within the same batch of fixtures, the bathroom one dies first and the balcony one slows first.

Which Fixtures Fail First

Location	Main cause	Typical symptom
Balconies, porches, garages, storage rooms	Low ambient temperature in winter	Startup delay after the first cold nights
Unoccupied homes, vacation houses in cold regions	Prolonged low temperature plus long periods unpowered	Delay grows from seconds to no light at all
Sealed downlights and spotlights inside ceilings	Poor heat dissipation, long-term high-temperature aging	Flicker and delay after two or three years of use
Outdoor wall lights, garden lights	Cold, temperature swings, moisture	Cold-start delay with occasional failure to light

The pattern: the more extreme the temperature, the more sealed the housing, and the less often the light is switched on, the sooner the capacitor gives up. A fixture that stays on around the clock is ironically the most stable one — its board is always warm.

A Three-Step Self-Test, No Instruments Required

Time it when the light has been off. After at least half an hour off, switch it on with a stopwatch. Then off, and immediately on again. Slow when cold, fast when warm → capacitor problem.

Swap the switch. Wire it temporarily to an ordinary mechanical switch. Symptom vanishes: switch leakage or single-wire-power interference. It doesn't: back to the capacitor.

Inspect the driver. Power off, remove it, and look for bulging, deformation, or yellowing on the casing, and domed or leaking capacitor tops on the board. If you see bulging, skip further testing — replace it. Then check the nameplate date: past five years, don't bother repairing.

Replacing the Driver: Five Parameters That Must Match

A constant-current driver is a power supply defined by "how many milliamps." Mismatched parameters mean at best no light, at worst a burned LED board.

Parameter	How to match	Common pitfall
Output current	Equal to or slightly below the original rating	Buying by wattage instead of current, one size up, and overheating the LEDs
Output voltage range	The new driver must cover the original range	If it doesn't, low-brightness modes won't start and the light flickers
Output power	Keep 10%–20% headroom	Running at full load long-term ages the capacitor early again
Certification	3C (CCC) or CQC, with a verifiable number	An uncertified driver is the last place to save a few dollars at home
Capacitor temperature rating	Prefer 105℃; wide-temperature types run -40–105℃	Anything labeled only 85℃ is a short-lived part inside a ceiling

One more thing: if potting is an option, take it. The encapsulant both insulates and conducts heat away, keeping the capacitor noticeably cooler.

Four Questions to Ask When Buying a New Fixture

Is the driver constant-current or constant-voltage, and what's the output in milliamps?
Is the electrolytic capacitor rated 85℃ or 105℃?
Is the driver potted?
Is the driver a separate, replaceable unit, or integrated with the LED board?

The first three determine whether, three years down the road, you're swapping a cheap driver or scrapping the whole fixture. The fourth determines whether you even have the choice. Integrated designs look sleeker and are easier to waterproof, but when one part fails, the whole light goes.

Three Frequently Asked Questions

Q: Is a two- or three-second delay a safety hazard?

A: A pure delay carries no risk. But if it comes with a hot casing, a faint burnt smell, or the delay grows noticeably within weeks, stop waiting — cut the power and replace the driver. That's a capacitor approaching failure, and it can take the LED board with it.

Q: Isn't replacing the whole fixture simpler?

A: It depends on the price of the light. For a budget downlight or bulb, replacing the whole unit is genuinely easier. For a ceiling or magnetic-track light costing hundreds, with LEDs and structure still fine, scrapping it over one driver throws away most of its cost. The precondition: a driver with matching parameters is actually available.

Q: Can I freely swap the driver in a smart light?

A: No. Smart drivers carry a communication module and dimming protocols; swapping in a plain constant-current driver loses dimming, color control, and connectivity. You'd need an original or a same-protocol, same-power replacement. That's also why it pays to be picky about brands in critical fixtures: the LEDs don't fail first, the driver does, and the driver may be impossible to source.

When cold weather makes a light "sluggish," the instinct is to blame aging LED chips. In fact the chips are the hardiest part of the whole fixture. The one that truly fears the cold is that inexpensive electrolytic capacitor — and it's what decides whether this light gets a cheap new lease on life or goes in the bin.

(Cross-posted from the Nexlamp lighting engineering blog.)

Why Your Bathroom Light Dies Within a Year: The Condensation Problem

lamp nex — Thu, 29 Oct 2026 04:12:09 +0000

Your bathroom light isn't dying of humidity. It's dying of condensation — twice a day, every day.

A customer called me last month about a bathroom ceiling light that had failed for the second time in a year. Same story both times: the light flickered for a few weeks, then the brightness dropped, then it wouldn't turn on at all. Both units were rated IP44, from different brands. When I opened the second one, the LED chips were fine. The driver board was corroded green.

After twenty years in the lighting industry, I can tell you the pattern is remarkably stable. In failed bathroom fixtures, fewer than 20% have burned LEDs. Around 80% fail at the driver board first, and the failures cluster between 8 and 18 months after installation — suspiciously just past most warranty periods.

Here's what's actually happening, and how to buy a bathroom light that survives it.

The real killer: twice-daily condensation cycles

Run a shower and your bathroom goes from 18°C to over 35°C, with humidity approaching saturation. The air inside the light fixture's cavity heats up along with the room. Then you turn off the water and the exhaust fan pulls the temperature back down — but the fixture's inner walls cool faster than the air. Water vapor condenses into droplets right on the PCB and capacitor shells.

Twice a day. Roughly 700 cycles a year.

The damage comes from the thin film of water each cycle leaves on the circuit board. That film eats away creepage distances between traces and accelerates electrochemical corrosion on component pins. It's not dramatic — no puddle, no visible leak — which is exactly why it slips past both buyers and warranty terms.

And here's the uncomfortable part: IP44, if it's only implemented on the diffuser side while the back panel still has ventilation openings, lets moisture in anyway. Just more slowly. Slowly enough to outlast the warranty.

Three failure points, in order

1. The driver — first to go, ~80% of failures

Cheap bathroom lights use a built-in, non-isolated, open-frame driver. To shed heat, the back of the housing must have openings. Moisture enters through those holes and attacks the electrolytic capacitor pins first. Symptoms: flickering, dimming, occasionally refusing to start — then working again ten minutes later.

The proper fix is three measures together:

A separate waterproof enclosure for the driver
The entire board potted in thermally conductive silicone, encapsulating every component
A fully sealed aluminum housing that conducts heat outward, instead of relying on open-hole convection

Miss any one of the three, and you drop a full tier in lifespan.

2. The seals

The gasket between the diffuser and the base is either ordinary rubber (cheap) or silicone (good). A bathroom doesn't just have water — it has soap scum and cleaning agents. Ordinary rubber hardens and loses elasticity in that hot, alkaline environment, and after about a year it starts to seep.

3. The diffuser

Acrylic (PMMA) has high light transmission and low cost, but it handles heat, humidity, and UV poorly. Sustained high humidity plus cleaning agents turns it yellow and hazy. Polycarbonate (PC) with a UV-resistant coating is far more durable.

Note on anti-fog coatings: they work by making the inner surface hydrophilic, so condensation spreads into a flat film instead of beading. They're effective — until soap scum builds up. Coated diffusers need neutral cleaner only.

What IP44 actually means

The first digit covers solids: 4 means objects larger than 1mm are blocked. The second digit covers water: 4 means it withstands splashing water from any direction.

That's splash. Not jets (IPX5), and certainly not immersion (IPX7). IP44 keeps splashes off the outside of the fixture. It says nothing about water vapor seeping through back openings, and nothing about condensation inside a saturated-humidity cavity.

Bathroom zones (GB/T 16895.13, identical to IEC 60364-7-701)

Zone	Location	Minimum rating
Zone 0	Inside the bathtub or shower tray	IPX7
Zone 1	Above the tub/shower tray up to 2.25m; for showers without a tray, a 1.2m radius around the shower head	IPX4 (IPX5 for public baths)
Zone 2	0.6m beyond Zone 1	IPX4
Outside zones	Rest of the bathroom	Standard indoor fixtures

The vanity light sits in Zone 2, so IP44 is the floor. Directly above the shower head, go straight to IP65.

What to spec, per position

Position	IP rating	CCT	CRI	Wattage	Source type
Main light	IP44 minimum, enclosed & sealed	3000–4000K	Ra≥80	8–15W (4–6㎡ room)	Replaceable source preferred
Vanity light	IP44 minimum, fully sealed back	4000K neutral	Ra≥90	8–12W	Replaceable source preferred
Shower downlight	IP65	4000K	Ra≥80	5–7W	Integrated is acceptable

Vanity mounting matters as much as the fixture: the light should span at least 70% of the mirror's width. If it's ceiling-mounted above the mirror, the bottom edge should sit about 1.9m from the floor.

One more decision: integrated vs. replaceable. With an integrated fixture, one failed component means replacing the whole light. An E27 or GU10 socket costs less over the long run.

Five questions before you buy

Is the driver an open board, or potted / in a separate waterproof enclosure?
Does the housing vent through openings, or is it fully sealed aluminum?
Is the diffuser PMMA or PC — and does the PC have UV coating?
Is the gasket silicone or ordinary rubber?
What rating do they recommend directly above the shower head? (If the answer isn't IP65, keep shopping.)

Three two-minute checks on delivery

Drip test on the diffuser. Water spreading flat means the anti-fog coating works; beading means it's already failed.
Press around the diffuser edge. If you can squeeze out a gap, the seal isn't seated.
Open the wiring compartment. Look for potting compound on the driver. No potting, no deal.

Smart lights in bathrooms: a different problem

PIR sensors detect motion by temperature change — and a bathroom has the most violent temperature steps in the house. Expect false triggers and false negatives. 24GHz radar sensors handle motion better but get confused by steam and mirror reflections.

Before buying any smart bathroom light, confirm the sensor can be disabled independently and that the manual switch still works when it is.

Connectivity is its own trap. 2.4GHz attenuation is significant in bathrooms: tile walls, aluminum suspended ceilings, metal mirror cabinets. Either place your hub/repeater in a relay-friendly position, or prefer Zigbee/Thread devices over Wi-Fi.

FAQ

Q: It's rated IP44 — why did it still fail?
A: The second digit in IP44 only covers external splash water. It makes no promise about condensation forming inside the fixture under saturated humidity. What actually determines lifespan: potting on the driver, cavity sealing, and how the housing manages heat.

Q: Can I retrofit an existing ceiling light?
A: Yes, in this order: first, replace the driver or add a waterproof driver box. Second, swap the gasket for silicone and seal any rear openings with a waterproof breathable membrane. Only then consider replacing the diffuser.

The short version

Bathroom lights don't die from moisture exposure. They die from 700 condensation cycles a year chewing through an unprotected driver. Buy sealed, buy potted, buy silicone — and check the potting yourself before you throw away the box.

(Cross-posted from the Nexlamp lighting engineering blog.)

A ¥423 Floor Lamp vs a ¥5,699 One: What the Extra Money Actually Buys

lamp nex — Wed, 28 Oct 2026 04:13:39 +0000

Before Double 11, floor lamps — the tall ones parents call "sun lamps" — became the priciest item on kids' school-supply lists. In July 2026 the Shanghai Consumer Council sent 20 best-selling models into a lab. Cheapest: 423.3 RMB. Most expensive: 5,699.3 RMB — a 13x gap. The results were odd: on illuminance and uniformity, all 20 hit the national AA grade. Colour rendering passed across the board. Flicker and blue light: no risk in any of the 20.

So on the four metrics that affect eyes most, the cheapest and the priciest test the same. What does the extra five thousand buy? Take the lamp apart by its bill of materials and the money sits in three layers.

Layer one: the optical cavity decides usability — price doesn't

A floor lamp has two emitting surfaces: the upper one bounces light off the ceiling to brighten the room, the lower one lights the desk. What decides "does it glare, is the desk evenly lit" is the optical cavity — diffuser plate, prism sheet, light guide. Unglamorous parts nobody demos in a store.

What's public in the industry but no salesperson will volunteer: within the same brand, the gap between flagship and budget is often exactly this layer. A trade outlet once tore down one brand's lamps: the flagship got glare down to UGR below 9, the budget version only managed 16 — and still cost more than a no-name equivalent.

UGR is the unified glare rating. At ≤13 most people feel little glare; for a child's lamp, push for ≤12. You don't buy this number with better LEDs — you build it with cavity geometry: side-emitting plus multi-layer softening panels scatter direct light into diffuse light. Direct-lit designs are cheaper, but the light comes straight at your eyes. Uniformity works the same way: good ones land around 1.1; bad ones give you a bright blob mid-desk and dark corners.

The test is simple: sit at the desk, look up at the highest setting, and see whether you can spot bright points on the emitting surface. Soft reflection only? It passes.

Layer two: the driver decides how long it lasts — and hides the one spec nobody checks

The driver does three jobs: constant current, dimming, and keeping electromagnetic interference locked in.

You can feel the first two. Whether dimming stutters or the lamp cuts out at half brightness comes down to the driver's floor. Cheap designs chop with PWM: fine until you wave a phone camera across it and see black bands. Good designs run constant current with analogue dimming.

The third job you can't feel — until it fails. Electromagnetic compatibility, EMC. 2 of the 20 samples failed here: the lamp labelled RUIJE, model QG2358 90W, missed on disturbance voltage, radiated disturbance and harmonic current — three items. The F9 100W floor lamp labelled "好视力" (Haoshili) failed on radiated disturbance. The tiering is blunt: inflated specs and EMC failures cluster below 1,000 RMB.

These lamps interfere with routers and TVs and stress home wiring over time. You think the WiFi is lagging; it's the lamp spraying noise. From the factory side, EMC is pure cost — common-mode chokes, X capacitors, shielding, certification — and the buyer perceives none of it at unboxing. So it's the first thing cut from a low-price build.

Structure is in this layer too: the lamp stands nearly 1.8 metres tall, centre of gravity high. All 20 survived the 6° tilt test; at 15° there's tip-over risk. No featherweight bases — and once assembled, wall or cabinet behind it. Always.

Layer three: above 4,000 RMB you're buying "smart", not "light"

Price tier	What testing shows	My verdict
Under 1,000 RMB	Most basic metrics pass, but inflated specs and EMC failures concentrate here	Skip it — the risk is in what you can't see
1,000–4,000 RMB	Illuminance, colour rendering, flicker, blue light all fully adequate	The mainstream for homes; money well spent
Above 4,000 RMB	No gain on core eye-protection metrics; the extras are smart scenes, voice dimming, dynamic natural-light simulation	Paying for experience, not eye protection

One sentence: past the 4,000 RMB tier, extra spending on "eye protection" buys you nothing measurable.

Then there's this category's own premium: health claims. Legal Daily reported in August that plenty of product pages state outright "prevents myopia, controls prescription growth, suppresses axial elongation." In January, a lamp maker was fined 600,000 RMB for advertising containing disease-prevention claims. The Consumer Council's wording is unambiguous: floor lamps cannot reverse or cure myopia. An ophthalmologist at a Beijing tertiary hospital put it more bluntly — a young child doing four to six hours of high-intensity close work a day cannot be offset by even a top-spec lamp. As for "red-light boost" and "developed by an academician": one 1,499 RMB lamp was sold under an academician's name, later exposed as a mere listing of his title.

The usage mistake that costs more than the lamp

The first thing many people do is switch off the ceiling light — you paid all that money, why run the main light? Exactly backwards. Main light off, the desk might read 1,500 lux while the floor two or three metres away sits at 100 lux. Every time the child looks down, then up, the pupil squeezes and releases. That cycling tires eyes more than a uniformly dim room.

Use it as a supplement, not a replacement: main light on, floor lamp evens out the desk area. Don't switch off the upward-facing source — its whole job is flattening the room's brightness difference. This fix costs nothing and beats trading up to a model 2,000 RMB pricier.

Before you click buy

Test at the desk's highest setting. "Look up, no visible bright points" only counts at full height — desks rise. Check the desk surface for reflections too.
Prop a board upright and re-test. Tilt a board and the reflection angle shifts; glare can land straight in your eyes.
Ask what a replacement light head costs out of warranty. Some lamps are cheap whole units but the panel swap alone runs 1,500 RMB, plus 200–300 RMB per service visit.
Screenshot any health claims. Anything saying "prevents myopia" or "controls prescription" is non-compliant advertising on its face — keep the evidence.

Three questions I get asked

Q: So the cheap one's fine and everything pricier is an IQ tax?
A: No. Under 1,000 RMB is the risk zone. In the 1,000–4,000 RMB range, pick UGR ≤12, uniformity close to 1, independently adjustable upper and lower sources, Ra 95 or above — money well spent.

Q: Does full-spectrum or violet-excited LED matter?
A: It helps, but it isn't first priority. The gain is far smaller than uniformity and glare control — and those two come from the cavity's structure, not from how expensive the LEDs are.

Q: My child is already nearsighted. Will a better lamp help?
A: A lamp treats nothing. It makes the local light environment comfortable and reduces eye strain. What actually works is outdoor time and limiting close-up hours.

One last honest line: the optical metrics are a threshold, not a selling point. Once you're over the threshold, every extra yuan should come with a clear answer to which of these three layers it's buying.

Originally published on the NEXLAMP blog.

All Labeled 3000K, So Why Do They Look Different? SDCM Is the Hidden Killer of No-Main-Light Design

lamp nex — Tue, 27 Oct 2026 04:13:48 +0000

You've seen this job site before: a row of six recessed downlights in the living room ceiling, every single one labeled 3000K. Flip the switch at night — two of them lean yellow, three are clean white, and one has a faint greenish cast. The whole ceiling looks like it's been patched together from leftovers.

The homeowner's first reaction is "these lights are broken." They argue with the seller, who hands over a test report: each individual light measures 3000K ±150K, fully compliant with the standard. Nobody is lying — and yet the problem is real.

After two decades in the lighting industry, I'll give you the answer up front: the standard governs each light's color temperature floor, but it can't control consistency when multiple lights sit side by side. What actually decides whether your ceiling looks "premium" or "cheap" is a parameter most sellers never bring up: color tolerance, measured in SDCM.

Color Temperature Is a Point. SDCM Is a Circle.

Let's get the concepts straight. A color temperature rating is a point — 3000K. But every mass-produced LED has manufacturing tolerance, so the actual color drifts around that point on the chromaticity diagram. The industry measures that drift with SDCM (Standard Deviation of Color Matching, derived from MacAdam's color matching experiments), expressed in "steps."

Think of it like a shooting range: 3000K is the bullseye, and color tolerance is the area on the target where a hit still counts. The smaller that circle, the more tightly the colors of your lights cluster together.

Here's the crucial insight: your eye doesn't compare "this light vs. standard 3000K." It compares "this light vs. the light next to it." Two fixtures can each sit comfortably inside spec, but if one lands at the far left edge of the circle and the other at the far right, mounting them side by side makes the mismatch obvious — one yellow, one white.

The industry's rough tiers look like this:

SDCM Tier	What You Actually See	Where It's Acceptable
≤3 steps	Visually identical to the naked eye	Commercial displays, tight mixed layouts of spots and downlights, premium residential
3–5 steps	Slight difference visible on close side-by-side inspection	The floor of acceptability for ordinary homes
>5 steps	Color shift clearly visible — a row of lights like a paint palette	Reject the shipment outright

Three Root Causes — The Money Is Saved Where You Can't See It

When lights from the same batch end up looking mismatched, the culprit is almost always one of three links in the chain.

1. Poorly binned LED chips. Chips cut from the same wafer naturally vary in wavelength and brightness. Reputable manufacturers "bin" them — sorting chips with similar color into the same grade before packaging. Finer binning costs more, because the yield of usable chips drops and the rest have to be sold downgraded. Cheap products often skip binning entirely or bin coarsely: every single light passes its factory test, but together they look like a patchwork.

2. The driver amplifies the color shift. This is the one I most want to flag, because it's the most hidden. LED color temperature drifts with current. A driver with poor constant-current precision produces output fluctuation, which widens the color gap between fixtures from the same batch — chips that were binned beautifully get dragged out of the circle by one lousy driver. Color consistency isn't only the light source's job. The driver owns a share of the blame too.

3. Inconsistent blending curves in dual-CCT fixtures. Smart tunable-white lights mix two chip sets — warm white and cool white — in varying proportions. If two batches have different blending calibration curves, then when you set both to 4000K, one reads warm and the other cool. Same rating, same setting, still mismatched.

And there's a fourth case: a flawed lens or reflector design that makes a single fixture run warmer at the center of its beam and cooler at the edges. Per-fixture color temperature testing can't catch this — you'd need an imaging colorimeter.

Here's who's responsible for each failure mode, and how to defend against it when ordering:

Root Cause	Who's at Fault	How to Prevent It at Order Time
Coarse chip binning	The LED/chip manufacturer	Ask support: "How many SDCM steps?" If they can't answer, be cautious
Poor constant-current precision in the driver	The power supply manufacturer	Use the same brand and series throughout the house — driver and light source calibrated together
Inconsistent blending curves	The fixture manufacturer	For dual-CCT lights, use one model across the whole house; don't mix two brands
Center-vs-edge color shift from the lens	The fixture manufacturer	Established brands have mature lens designs; no-name products are the danger zone

A Three-Phase Playbook: Before Buying, On Delivery, After Installation

Before buying — get SDCM into the chat log. Before you order, ask directly: "What's your color temperature tolerance? How many SDCM steps?" A legitimate brand will give you hard numbers (a good product: within ±200K, SDCM ≤5). If they hem and haw with "well, they can't be exactly identical," they're almost certainly skipping binning. Buy every light-emitting product from the same brand and same series — magnetic track lights, downlights, spots, and strip lights bought from different shops can drift hundreds of Kelvin apart despite matching labels, and this is the number one cause of no-main-light design failures. LED strip is especially sensitive: buy the entire run from one batch — even head and tail of a single reel can differ — and always order extra.

On delivery — light them up side by side before the ceiling closes. This is your only window where returns and exchanges are still painless. Power on every fixture from the batch, lay them side by side on the floor, and look: pick out the yellow- or white-leaning outliers and send them back now. Once they're above the ceiling, the cost and emotional toll of swapping them is a different universe. Commercial projects sample-test with instruments; for home projects, a naked-eye side-by-side comparison is more than enough.

After installation — it already looks bad. Three steps to fix it. First, audit: turn on every light at night, photograph and note which ones lean white and which lean yellow. Second, replace: swap out replaceable bulbs and modules directly (a few dollars each); replacing a full downlight or spotlight runs roughly 30–50 yuan per fixture. Third, manage: pick one dominant color temperature as the baseline for the whole house, then use scene grouping to keep the mismatched zones from being lit at the same time — that alone cuts the visual chaos roughly in half. On a tight budget, prioritize the areas you use daily: living room, dining room, hallway. Nobody sees the mismatch in a bedroom with the door closed.

FAQ

Q: Do high-CRI (Ra 98) lights automatically have tighter color tolerance?

Not the same thing. CRI measures how accurately objects' colors are rendered; color tolerance measures how consistent the lights are with each other. Two independent dimensions. A high Ra score cannot hide a high SDCM — ask about both when you buy.

Q: Smart lights can adjust color temperature — doesn't that solve consistency?

Adjusting color temperature changes each light's target value, not the offset of its baseline. Two lights with far-apart baselines will still look different when both are set to 4000K. The right way to use smart lights: same model, same batch, whole house, then calibrate together.

Q: Is buying a big brand enough to be safe?

Big brands are genuinely an order of magnitude better at binning and QC, but cross-series and cross-year batches can still differ. Trust the brand and insist on "same series, same batch" — and never skip the power-on comparison on delivery.

The Bottom Line

Half of a no-main-light design's premium feel comes from the design itself; the other half comes from consistency. Color tolerance is the parameter sellers won't volunteer and single-fixture test reports won't show — yet it's the thin line between "looks expensive" and "looks cheap."

Three things to remember: ask for SDCM before you order, power up and compare on delivery, and know that the moment before the ceiling closes is your last window for returns.

Originally published on the NEXLAMP blog.

Magnetic Track Lights Look Amazing Until the Driver Fails: A 20-Year Lighting Vet's Pitfall Guide

lamp nex — Mon, 26 Oct 2026 04:10:10 +0000

Magnetic track lights have become the undisputed king of "no-main-light" interior design. At the 2026 Hong Kong Autumn Lighting Fair (October 27-30), track systems, ambient lighting, and shadowless layouts are once again the headline trends — and no-main-light design now accounts for over 60% of new residential renovations.

Yet real-world results are wildly polarized. Same floor plan, same fixture list: one home looks like a gallery, the other looks like a spaghetti disaster on the ceiling. After twenty years in the lighting trade, I'll tell you where almost every failure comes from: the driver, the pre-embed timing, and brand lock-in.

The driver is the heart of your low-voltage system

Start with the basics most homeowners miss: magnetic track lights are low-voltage products (24V or 48V) and require a driver — the "transformer" everyone forgets about. For residential use, always go 48V and never touch 220V high-voltage track. With high-voltage track, removing a lamp head can expose live contacts; if you have kids at home, that's a hazard you don't want. And brightness has nothing to do with voltage — it depends on the LED chips.

The driver is also the single most failure-prone component in the whole system. Five things you must know:

Limited lifespan. Drivers typically last 3-5 years, much shorter than the lamp heads themselves.
Leave power headroom. Multiply your total lamp-head wattage by 1.2 to size the driver. Without headroom, running many heads simultaneously causes voltage instability — flickering at best, a burned driver at worst.
Brands are not interchangeable. NVC uses 28mm contact spacing, OPPLE 30mm, Yeelight 26mm, with different current specs too. Mixing brands means poor contact and constant flickering, or lights that simply won't turn on.
Mount it externally with an access panel. Embedding the driver in the ceiling with no access panel is the classic self-destruct move — when it fails, you're tearing down drywall.
Heat kills. A driver sealed in an unventilated ceiling will flicker first and die later.

The practical checklist:

Item	Do this	Cost of getting it wrong
Power headroom	Size driver at total wattage ×1.2	Voltage sag, flicker, burned driver
Brand matching	Buy track, heads, driver from one brand	Mismatched contacts, chronic flicker
Placement	External mount near AC return or access panel	Ceiling demolition on failure
Heat	Ventilation and clearance	Flicker, shortened 3-5 year lifespan
Spare	Buy one extra driver up front	Out of stock in year five

That last row sounds paranoid until your driver dies in year four and the model is discontinued. The spare costs almost nothing today and will save you then.

Pre-embed timing and ceiling height: mistakes cost double

Track cannot be added as an afterthought. It must go in during the carpentry stage, with wiring reserved at the rough-in (plumbing/electrical) stage, and the track installed before the painter skim-coats and rolls paint. Change your plan mid-renovation and the cost roughly doubles.

Recessed track needs a ceiling pocket 55-60mm wide and 85-90mm tall, with the finished ceiling dropping at least 8cm. That makes ceiling height a hard gate: below 2.7 meters, skip recessed and choose surface-mounted ultra-slim track (5-6mm thick) — you keep the look without sacrificing headroom.

Surface mount has its own catch: it demands a flat ceiling. Track is rigid and straight, so any curvature shows up as misaligned joints or a black gap. Check flatness with a 2-meter straightedge before installation, and fix surface-mounted track to the joists on drywall ceilings — otherwise it sags in the middle over time.

One more detail people forget: corners require a dedicated conductive corner module, or current simply won't pass the bend. And keep the power feed at one end of the track, not buried mid-run.

Head mix: 60/30/10, no extremes

All-grille setups (salespeople love pushing them) concentrate the beam so hard your living room feels like an interrogation room. All-spotlight setups leave the floor dotted with hotspots while the upper walls stay dark.

A proven mix is 60% flood + 30% grille + 10% pendant. Spacing: flood strips at 60-80cm, grille downlights at 50-60cm. Keep the track at least 50cm from the wall — if your "wall washing" washes nothing, the track is probably too close.

Brand lock-in: buying track is signing a contract

People usually figure this out three years too late. Track interfaces, contact spacing, and power specs differ by brand. Buy NVC track today, and in three years your only option for additional heads is... NVC. If those heads get discontinued, you're ripping out the whole track.

So buy everything — track, heads, driver, accessories — from one brand, in one order, with the driver sized at 1.2× total wattage. You're not just buying a light; you're buying a five-year parts ecosystem.

Five questions to ask before you order

Question	Why it matters
What's the contact spacing?	28mm vs 30mm decides whether heads physically fit your track
Is the driver sold separately?	Bundled sets can force full replacement; separate drivers allow repair
Is the driver sized with headroom?	Confirm total wattage ×1.2, not bare minimum
Where's the access panel?	External mount or near-panel placement avoids ceiling demolition later
What if this head is discontinued?	Ask stock cycles, and grab a spare driver while you're at it

FAQ

48V or 220V track — which one?

For homes, 48V low-voltage, no debate. 220V high-voltage track can shock you when a head is removed, a real risk with kids. Brightness comes from the LED chips, not the voltage.

Does a dead driver mean tearing down the ceiling?

Not if you planned ahead. A driver mounted externally near the AC return or an access panel swaps out in ten minutes. One sealed inside the ceiling? That's drywall surgery.

Are magnetic track lights a gimmick?

Not if installed right. With proper circuit zoning, driver headroom, and correct pre-embed timing, the flexibility and ambience justify the price. Botched by a careless installer, they're an expensive lesson.

Originally published on the NEXLAMP blog.

Quoted 80, Billed 300: How Luminaire Installation Costs Grow Add-Ons

lamp nex — Wed, 21 Oct 2026 04:06:31 +0000

Double 11 parcels are arriving, and the small job of hanging a light is where things start to go wrong.

A homeowner in Beijing was quoted 80 on the phone for one ordinary flush-mount ceiling light. The bill at the end was 300. In Chongqing, a pendant was quoted at 80; on site the installer said the ceiling slab needed reinforcing, another 400; after the job he said two mounting brackets were missing, another 120. One fixture, 600 in labor.

The money did not go into craftsmanship. It went into a quote sheet that had one word written on it: installation.

The gap between 80 and 300 is the add-on list

There is no fixed price for hanging a light — only a base price plus add-ons. Base price means: remove the old fixture, fit the new one, keep the existing hole, stay within arm's reach. Move the hole, change the ceiling height, or touch the wiring, and you are in add-on territory.

Line item	Typical range	When it shows up
Basic install, flush-mount ceiling light	RMB 60–150 per fixture	Basic wiring, drilling at the existing hole
Pendant / crystal chandelier	RMB 100–400 per fixture	Hanging parts to assemble, many components
Downlight / spotlight	RMB 15–60 each	Bulk price negotiable at volume
Smart fixture, commissioning included	RMB 150–300 per fixture	Pairing, dimming tests, user walkthrough
Old fixture removal	RMB 20–50 per fixture	Some platforms include it — ask
Call-out fee	RMB 30–50	Some installers bill it separately
Ceiling cut-out	RMB 30–80 per hole	No pre-cut hole in the ceiling
Work at height	Above 3 m, about +50 per extra metre	Double-height rooms, stairwells
Circuit inspection / rewiring	RMB 50–150 per visit / RMB 300–800	Old buildings, undersized conductors, moving the point
Whole-flat package, three-bedroom	RMB 300–500	Cheaper per fixture; negotiate at volume

The two items most often added on the spot are cut-outs and height work — neither visible before the installer walks in. The most profitable is consumables: brackets and expansion bolts cost almost nothing, are billed per piece, and each piece is billed in the tens.

Three moves that turn 80 into 300

The first is to dismantle before quoting. Once the old fixture is down you cannot put it back, and your negotiating position is gone. Correct sequence: see the site, quote, then start.

The second is to split one job into several line items — base installation, removal, reinforcement, brackets — each reasonable alone, together double the price.

The third is to cite "aging wiring". Said before anything is opened up, it is usually a script. Worth investigating when the breaker trips and terminals run hot.

Why smart fixtures cost one tier more: the money is in commissioning

Installation and commissioning are two different trades. Installation is running cable, making connections, fixing the fixture to the ceiling. Commissioning is pairing, grouping, scenes, minimum brightness. A technician with twenty years of mains experience can crimp a terminal with his eyes closed, then get stuck on a gateway and a scene.

An electrician I know took on a hotel corridor. He installed everything and could not commission it. The manual ran to dozens of pages; he read two and put it down. He ended up paying 800 to bring someone in for half a day.

Step	Ordinary fixture	Smart fixture
Wiring	Live and neutral, then it lights	Same, plus you have to decide where the driver lives
Power-on	On or off, obvious at a glance	Lighting up is step one; flicker and minimum brightness still need checking
Network	None	Pairing, grouping, naming
Scenes	None	Switches bound to lights, scenes configured, the family taught to use them
Handover	Installer leaves	Must be commissioned and demonstrated before leaving

I get these calls regularly. A customer says "your driver is dead." I ask for a photo of the terminal block: the conductor is not pushed all the way in, the insulation is clamped under the screw. I have also seen drivers stuffed into the ceiling void until thermal protection cuts in, and pairing failures from 2.4 GHz and 5 GHz on the same SSID.

Fault finding has three layers: does it light up, does it join the network, does it dim smoothly. Do not start by suspecting the driver.

Five questions that lock the price before work starts

"Is removal of the old fixture included in the installation fee?"
"Is there a pre-cut hole in the ceiling? If not, how much per hole?"
"How high is my ceiling — is there a height charge above three metres, and how much?"
"Are you bringing the brackets, expansion bolts and terminal blocks, or am I buying them? How much per piece if you bring them?"
"The smart fixture needs pairing, minimum brightness and fade-in/fade-out set. Is that inside the quote? If it is not commissioned properly, who is responsible?"

Anyone who gives a firm all-in price over the phone, without seeing the site, usually has a reason to add money after arriving. A written list before work starts is the better sign.

What you can do yourself, and what you should not touch

Under three metres of ceiling, with a stable ladder, plasterboard or aluminium panel, and an existing hole, you can swap a flush-mount or downlight yourself. Cut the power, verify it is dead, photograph the wiring first.

Do not cut new holes yourself. Plasterboard crumbles at the edge, aluminium panel needs a hole saw, and a bad hole costs more to fix than hiring someone.

Crystal pendants and multi-tier chandeliers need bracket assembly and expansion bolts into concrete. Call someone. A fixture coming down is not a money problem.

For a whole flat with dozens of downlights, negotiate a package — RMB 300–500 for a three-bedroom flat is a realistic conversation.

Three closing lines

Add-ons are not the problem. Add-ons that are not on paper are the problem.

Half of what you pay for a smart fixture is installation; the other half is commissioning. An installer who will hang it but not commission it has sold you an ordinary light.

Run every scene on the day it is installed, not after you move in.

Three questions I get asked most

Q: The product page says "installation included." Why does the installer still want money at the door?

A: It covers the most basic tier only: existing hole, flush-mount fixture, within reach. Removal, cut-outs, height work, consumables and commissioning sit outside. Read the "not included" column first.

Q: The installer says the wiring is aged and has to be replaced. How do I tell whether that is true or a script?

A: Look at when it was said. Before anything is opened up, it is usually a script. Real signs are specific: the breaker trips, terminals are hot, insulation crumbles when bent, the building still has aluminium conductors. No evidence, no replacement.

Q: The smart light will not connect to the app after installation. Is it the fixture or the installation?

A: Three layers. Does it light up — if yes, power is fine. Can the app find it — if not, usually pairing mode, or 2.4 GHz and 5 GHz on one SSID. If it connects but dimming jumps, then it is the driver and the LEDs.

Same "Fade In / Fade Out" Claim, Three Behaviors: What Actually Happens on the Driver

lamp nex — Tue, 20 Oct 2026 04:12:26 +0000

A contractor handed me a phone video last October: a downlight turning off, brightness walking down halfway, then black in one frame. Is that a fade?

I put a current probe on the driver output. At shutdown the current steps down. It does not ramp. A buyer's post that week drew dozens of replies with the identical symptom; the answer was "that's normal." It is not. Two things were left unfinished: the power-on default state, and the curve.

Three behaviors behind one slogan

What the buyer sees	What slow motion shows	Where it comes from
Genuine fade	Follows a curve, tail closes cleanly	Curve in firmware, enough bit depth, ends terminated
Flash to full, then settle	Full brightness at power-up, then down to setting	MCU reset leaves the IO high
Cliff at the midpoint	Walks down, cuts to black	The adjustable floor sits mid-range

Same phrase, different boards.

Root cause 1: the pin state at power-on

I reproduced it. Cut power, restore it, probe the MCU output: after reset and before the program runs, that line sits high. The LEDs are lit by the gap, not the code.

If main() starts a timer and configures the IO second, the interval between those steps is full brightness. The user sees a flash; the engineer sees execution order. Zeroing the duty register first costs one line.

Root cause 2: Gamma 2.2 on a shallow PWM

Perceived brightness is not linear in duty cycle — the gap is about a 2.2 power, hence γ = 2.2 in every dimming table. On that curve, 30% brightness is 7.1% duty. 10% is 0.63%. 5% is 0.14%. 1% is under four ten-thousandths of full scale.

Everything goes wrong in those last decimals. An 8-bit PWM has 256 steps; 0.63% lands on about one and a half, and 1% rounds to zero. Mathematically zero, physically off.

PWM depth	Total steps	Pure Gamma reaches	What slow motion shows
8-bit	256	around 10%	No steps below, dies when you dim
12-bit	4096	around 5%	Low end works, tail acceptable
12-bit + segmented low-end map	4096	around 1%	Common production approach
16-bit	65536	lower	Limited by timing and chip ceiling

So "5% minimum" is not laziness; it is plain Gamma's ceiling on 12 bits. Reaching 1% needs a second, denser map for the low region; skip it and the only option is a hard floor. Two defects split here: dying on the last step is insufficient depth; cutting out at the midpoint means the floor is too high.

Root cause 3: dead band and minimum current

At 3%–5% duty the eye barely separates the level from off, but the tail still smears. A dead band fixes it: below 3%, output zero.

The other end is minimum dimming current, typically around 3% of rated — roughly 10mA on a 350mA driver. Too low and you get color shift and flicker; CCR analog dimming below 10% drifts CCT, and same-batch lamps may not strike together. Too high and the fixture "won't turn off, the lowest setting still glows."

PWM frequency matters too, typically 2k–8kHz against hearing at 20Hz–20kHz: anything in that band at low output is audible.

Why fade-out is harder than fade-in

Fade-in happens the instant power arrives and gets a free window. Fade-out does not: when the off command lands, the lamp must still be alive and the driver still running the curve. A wall switch removes the power, so fade-out needs permanent live with the switch sending only a signal — kinetic, smart, voice, automation. That is wiring topology, not brand.

The curve also has to live in the right place: frame-by-frame commands from an app or the cloud die mid-slope when the network stutters, while a ramp compiled into firmware finishes with the network down. Commercial standard drivers often spec to 5% then drop out. We build drivers and would rather hard-code the ramp.

Four checks in the field

Phone on 240fps slow motion, one capture on, one off. Ramp or staircase in the playback.
Set 1% and leave it. Dim, or off? Off means depth or dropout.
Cut power, restore it. Any flash to full?
Listen. Quiet room, lowest setting, any buzz.

Four questions before you buy

What is the minimum brightness percentage, and does it die there?
Is the ramp in the driver or in the app?
What is the power-on default state — does it jump to full first?
Wall switch, kinetic, or smart: can a wall switch fade out at all?

If none of the four gets a straight answer, nobody has built this before.

FAQ

Is fade in/out a property of the LED?

No. The LED only emits; the driver board and control channel decide how brightness travels. The same LED on different drivers gives a ramp or a cliff.

Why does the light dim, then snap to black?

Either insufficient bit depth, so the low region has no usable step, or a driver floor sitting right there with dropout below it.

Does a smart lamp need permanent live for fade-out?

Yes. The driver must keep working when the off command arrives; a wall switch that cuts power leaves no window to run the curve.

Also published at https://www.nexlamp.com/dist/blog/smart-light-fade-in-out-dimming-curve-2026.html