Views: 0 Author: Site Editor Publish Time: 2026-11-18 Origin: Site
Sampling is a ladder, not an event. The functional sample asks "can it be built", the engineering sample "is the design settled", the pre-production sample "can the line repeat it", the golden sample "what does conforming look like", the first article "did the line actually repeat it".
Every stage needs a comparison object. A limit table, a limit sample, an approval sheet. Without one, sample review collapses into "it sounds fine and looks fine".
The expensive part of sampling is the revision, not the unit. Each loop reopens fixtures, re-runs tests and restarts the approval paperwork.
Thermal and acoustic noise have to be closed out at sample stage. The classic amplifier failure is a unit that passes on the bench at 25 °C and protects itself at 40 °C in a client rack.
A signed sample freezes one unit, not the bill of materials. Without change control attached, the signature covers nothing.
The approval package is the real deliverable. The hardware is the carrier; the documents are what production gets judged against.
The OEM amplifier sampling process has five stages — functional, engineering, pre-production, signed sample, first article — each answering one question. Skip a stage and the risk does not disappear; it moves downstream to tooling, fixtures and the line, where correcting it costs an order of magnitude more.
The vocabulary is a mess. A-sample, B-sample, PP sample, golden sample, first article — the same words mean different things at different factories, and sales and engineering inside one factory often disagree about what a "B-sample" is. Aligning on letters is pointless; align on the question each stage answers.
The ladder retires uncertainty in order. Early stages carry many variables and cheap changes; late stages carry few and charge dearly for them.
Stage | The one question it answers | Main variables | Key output | Typical misuse |
|---|---|---|---|---|
1. Functional | Can this circuit and architecture hit the numbers? | Topology, output stage, power supply | Verification report, key waveforms | Judging cosmetics on a breadboard |
2. Engineering | Is the design settled, and can it be manufactured? | Mechanics, thermal path, EMC, process margin | Full test data, DFM issue list | Quoting volume pricing off an engineering sample |
3. Pre-production | Can the real tooling and fixtures build it consistently? | Moulds, jigs, process parameters, incoming material | Pilot run records, first-pass data | Building a single hand-made "pilot" unit |
4. Signed sample | What does conforming look like, and which unit is the reference? | Finish, colour, silkscreen, accessories | Sealed sample, limit samples, approval sheet | Signing a sheet without keeping a sealed unit |
5. First article | Can the line reproduce the signed sample? | Batch material, equipment state, operators | First-article inspection and comparison | Checking electricals only, skipping cosmetics |
Stages three and four are where Chinese factories and overseas brands most often talk past each other: the buyer assumes a confirmed sample authorises production, the factory assumes a sample only proves the design is feasible. Neither is being careless — they are using one word for two things.
A functional sample proves the circuit can hit its targets. These units look rough — hand-soldered boards, an external heat sink, an off-the-shelf chassis with holes cut in it. That is normal, and cosmetics do not belong in this review.
What matters is six groups of hard numbers:
What to look at | How it is checked | Typical direction of judgment (industry practice, not a commitment) | What it tells you |
|---|---|---|---|
Rated power | 1 kHz single tone, then full 20 Hz–20 kHz band | Full-band output holds up against the 1 kHz figure | A 1 kHz-only number usually means the band sags |
Distortion | THD+N at low, mid and full power | All three points acceptable, not just 1 W | Good single-point data with a collapsing top end means thin supply margin |
Channel crosstalk | Adjacent channels, then all channels driven | Adjacent suppression clearly better than all-driven | Multichannel designs give themselves away here |
DC offset at output | No load and loaded | Ideally tens of millivolts or lower | High offset means switch-on thumps and stressed woofers |
Protection behaviour | Trigger OCP, OTP and DC protection individually | It trips when it should, recovers, and does not false-trip | Protection that exists only in the manual does not exist |
Idle noise | Input shorted, gain at maximum | No audible hiss with an ear at the tweeter | Speaks to supply design and grounding |
The usual mistake at this stage is treating a functional sample as evidence of manufacturability. A hand-soldered board has never seen a stencil, a reflow profile or a wave solder pot. An external heat sink says nothing about whether the thermal path survives inside the real chassis.
Typical construction: a multichannel power amplifier chassis with a vented top panel (illustrative only)
The engineering sample puts the working circuit inside something close to the final chassis. Mechanics, heat sink, connector layout and silkscreen should all be near-final, because the problems this stage exists to surface live in the assembled state.
Review area | What specifically | Why it cannot wait |
|---|---|---|
Mechanical fit | Board mounting, connector cut-outs, gap consistency | Once the mould is cut, a mechanical change becomes a mould change |
Thermal path | Device-to-heat-sink contact, interface material, airflow | The thermal design is locked into the chassis; afterwards you can only derate or add a fan |
Connector definition | Balanced and unbalanced input pinout, terminal spacing | A reversed pinout is a batch incident, not a defect |
Process margin | Insertion clearance, pad spacing, screw access | Assembly fixtures are designed around this; changing it means new jigs |
The step most often skipped is inviting process engineers into the review. Designers ask whether the numbers can be reached; process engineers ask how two hundred units a day get assembled identically. Review with one perspective only, and the other perspective's problems arrive during the pilot run, on schedule.
The pre-production sample is the hinge of the whole ladder. Samples were built by engineers; from here they are built by the line with production tooling. What is being proven is no longer the design but the manufacturing system.
To check whether this stage really happened: ask which fixtures, which line, how many units. A "pre-production sample" hand-picked and tuned by an engineer is an engineering sample wearing a different label.
Record to request | What it should contain | What its absence means |
|---|---|---|
Pilot quantity and batch | How many units, from one incoming lot or several | One unit tells you nothing about consistency |
Fixture and jig list | Assembly jigs, test fixtures, burn-in positions | Still hand-built; the process is not frozen |
Process parameter records | Reflow profile, soldering parameters, torque settings | Parameters live in someone's experience, so batches will drift |
First-pass and rework data | How many passed first time, what was reworked, why | "All passed" with no records usually means no records |
Packaging and drop test | Cushioning, carton, drop per the packaging standard | Packaging problems get deferred to mass production, and the damage rate transfers to the buyer |
This stage also exposes yield cost for the first time. A structurally sound design can still cost thirty extra seconds per unit because of one screw position; an indicator light in the wrong place adds a rework step at the test station. Invisible on a bench sample, obvious the moment the line runs.
A signed sample is usually read as "we agree this unit is fine". Its real job is to create an object future production is compared against.
Signing a sheet without sealing a unit is the most expensive shortcut in the process. Three actions belong together:
Action | How it is done | What happens without it |
|---|---|---|
Sealing | Seal the signed unit, note date and revision, one unit held by each side | A dispute ends up argued over photographs, and photographs do not settle colour |
Limit samples | Keep one upper and one lower acceptable boundary piece (colour, texture, silkscreen offset) | "Slight colour difference" can never be resolved |
First-article comparison | Compare each batch's first unit against the sealed sample, record the delta | Drift starts with batch one and nobody notices |
Limit samples are the most practical cosmetic control and the least requested. They turn "conforming" from an adjective into two objects: a lower bound below which goods are rejected, and an upper bound nobody needs to exceed. Arguments between the line and QA drop sharply afterwards — not because the standard got stricter, but because it got concrete.
Hardware is the carrier. Documents are the judge. A complete sample approval package normally contains the following, and whichever item is missing becomes the argument later.
Document | Function | Consequence if missing |
|---|---|---|
Sample approval sheet | Revision, date, both signatures | Nobody can say which revision was approved |
Controlled BOM with approved part numbers and substitution rules | Freezes material, defines what may be swapped | Silent swaps, drifting voicing and reliability |
Per-unit test report | Measured values with serial numbers | A pass/fail conclusion with no data behind it |
Dimension and connector drawing | Cut-outs, terminal spacing, screw positions | The unit does not fit the client's rack or the cables |
Cosmetic limit samples and colour reference | Makes cosmetic judgment physical | Every batch draws a "colour is different" complaint |
Packaging drawing and cushioning design | Inner packaging, carton, stacking | Higher transit damage and unclear liability |
Change control procedure | Who raises, who approves, who notifies | The buyer is the last to hear about a change |
Ask for one complete package from any past batch. A factory running genuine OEM work keeps these filed by model; a factory that cannot produce one is telling you the approval process lives only in conversation.
The table below pairs each test with the failure it is meant to catch. A test's value is not in passing; it is in predicting which failure mode is still open.
Test | Method and instrument (general practice) | Typical direction | Failure mode it points to |
|---|---|---|---|
Rated power | 1 kHz and full band, dummy load plus power analyser | Full band does not sag | Thin supply margin, inadequate heat sinking |
THD+N | Low, mid and full power | All three acceptable | Poor gain-stage linearity, early clipping |
Frequency response | 20 Hz–20 kHz sweep | Flat within roughly ±0.5 dB | Wrong coupling capacitor value, input network mismatch |
SNR | Input shorted, A-weighted | Noise floor stable across gain settings | Grounding layout, supply ripple |
Channel crosstalk | Adjacent, then all channels | Adjacent better than all-driven | Shared supply, dense layout |
DC offset | No load and loaded | Tens of millivolts or lower | Input stage offset, missing protection |
Protection behaviour | Trip OCP, OTP, DC individually | Trips correctly, no false trips | Thresholds designed on paper, never measured |
Temperature rise | Continuous rated power to thermal equilibrium | Junction temperature retains margin | Under-sized heat sink, high contact resistance |
High-ambient run | Environmental chamber at 40 °C and above, extended run | No derating, no nuisance protection | Design validated at room temperature only |
Standby consumption | Power meter in standby | Meets the target market's regulatory limit | Standby circuit never optimised |
Dielectric strength and insulation | Hipot tester, insulation resistance meter | Passes to the safety requirement | Creepage distance, transformer construction |
Earth continuity | Earth bond resistance test | Low enough resistance | Masking missed on a painted part, missing earth screw |
EMC pre-scan | Pre-compliance radiated and conducted scan | Adequate margin | Switching harmonics, cable routing |
Packaging drop | Face, edge and corner drops per the packaging standard | No cosmetic or functional damage | Insufficient cushioning, misaligned inserts |
Two notes. Temperature rise and high-ambient operation are separate tests: a comfortable rise at room temperature proves the heat sink works at room temperature, nothing more. And EMC pre-scanning belongs to the sample stage — failing emissions usually means changing switching frequency, adding a choke or rerouting a harness, which is cheap on a sample and a redesign on a production unit.
Typical construction: per-channel input and loudspeaker output terminals on a multichannel amplifier rear panel (illustrative only)
Two numbers lie most convincingly on an amplifier datasheet: temperature rise and noise.
The thermal number is a question of test conditions. A bench test runs at around 25 °C, unit open to the air, for a few minutes. Real installation means a closed rack, several units stacked, 30–40 °C ambient, and hours of operation. Stack those conditions and hot-spot temperature climbs well past the bench figure: a fan-equipped unit goes to full speed and the noise arrives, a fanless unit starts counting toward thermal protection.
Noise depends on how it is measured. "SNR of 100 dB" and "no hiss with an ear at the tweeter" are different claims. The first is electrical; the second also covers mechanical vibration, transformer buzz, chassis resonance and hum from ground loops. The stubborn noise usually sits in the transformer and the chassis, where no analyser looks.
Symptom | Common cause | What to do at sample stage |
|---|---|---|
Cuts out after a while at high volume | Thermal protection tripping early | Increase thermal margin, improve contact surfaces, re-derive the threshold |
Audible hum at idle | Transformer resonating with the chassis, poor mounting | Change mounting structure, add damping, adjust winding construction |
Continuous hiss at the tweeter | High noise floor, badly distributed gain | Rework grounding and supply, redistribute gain |
Thump on power-up or down | DC offset, no delay or mute circuit | Add relay delay and mute control |
Slight bleed between adjacent channels | Shared supply, dense layout | Rework supply distribution and shielding |
Failure | Real cause | Where it should have been closed |
|---|---|---|
Sample sounds right, production does not | Undisclosed substitution, or no listening comparison on alternates | Between stages 1 and 2 (controlled BOM) |
"Slight colour difference" never resolved | No limit samples, only a verbal "close enough" | Stage 4 |
Unit will not fit the client's rack | Dimensions never confirmed with the sample | Stage 2 |
Low first-pass yield on batch one | Pilot run was one unit, process unproven | Stage 3 |
Emissions fail at certification | No pre-scan, or no margin | Stage 2 |
Protects itself at high volume | Designed at room temperature, no 40 °C run | Stage 2 |
High transit damage | Drop testing deferred to mass production | Stage 3 |
Buyer learns about a change last | Change procedure absent from the approval sheet | Stage 4 |
The pattern matters more than any row: not one of the eight is a technical impossibility, all eight are stage misalignments. Sampling rarely fails because something cannot be built — it fails because the right work happened at the wrong rung.
The most common and least visible problem after signing is a material change, and it is rarely malicious: the part is out of stock, the lead time is too long, a capacitor got dearer, an engineer concluded the substitute "measures the same".
The trouble is that "measures the same" is not "behaves the same". Three component groups are especially sensitive in an amplifier:
Capacitors, particularly coupling and filtering. Same capacitance and voltage rating, different dielectric, and low-frequency behaviour plus temperature characteristics move.
Output modules and power devices. Same package, same voltage rating, different transconductance and switching behaviour — and the protection thresholds need recalculating.
Transformers and inductors. Different winding construction or shielding changes mechanical noise and stray field directly.
Three controls cover it, all belonging in the approval sheet: specify part numbers, define the substitution window, and require written notice plus a new sample before any change. The third is dropped most often, and most disputes are about a change happening silently rather than the change itself.
Change type | New sample required? | Basis |
|---|---|---|
Same part, different lot | Generally no, but keep lot records | Not a design change |
Same specification, different brand | Yes | Identical parameters do not mean identical behaviour |
Mechanical part, mould or finish | Yes | Direct effect on cosmetics and assembly |
Firmware or control logic | Yes | Affects protection and power-up behaviour |
Sub-tier supplier change | Assess, usually sample | Affects consistency and traceability |
This sheet can be used directly in a sample review meeting. Score each item 0 (missing), 1 (present but incomplete) or 2 (complete and checkable); maximum 24.
# | Item | Score 0 looks like | Score 2 looks like |
|---|---|---|---|
1 | Stage clearly identified | Nobody can say which stage this is | Stage and revision marked |
2 | Test data | A "qualified" conclusion only | Per-unit values with serial numbers |
3 | Full-band power | 1 kHz figure only | Full-band measurement |
4 | Protection verification | Text in the manual | Individual trip records |
5 | Thermal and high ambient | Short room-temperature run | Extended 40 °C run included |
6 | EMC pre-scan | Not performed | Report with margin stated |
7 | Mechanical fit and drawings | Sample with no drawing | Drawing matches the sample |
8 | Connector definition | Verbal description | Written pinout |
9 | Cosmetic limit samples | None | Upper and lower bound pieces |
10 | Sealed sample | Not sealed | Sealed both sides, dated |
11 | Approval package completeness | One signed page | BOM, drawings, reports, packaging |
12 | Change procedure | Not mentioned | Written procedure with notification duty |
Read the total simply: 18 or above, move to pre-production review; 12 to 17, close the gaps first; below 12, the stage was not actually completed.
Typical construction: rear panel layout with mains inlet, fuse holder and channel labelling (illustrative only)
Q1: Does every project need all five stages? Derivative projects can be compressed, not skipped. Where a proven platform is reused and only cosmetics or connectors change, the functional sample can fold into the engineering sample. Pre-production and signed-sample stages stay — they govern whether the line can build it and how conformity is judged.
Q2: Can a hand-built unit count as a pre-production sample? No. It is by definition built with production tooling, processes and the line itself. A hand-built unit proves the design, not the manufacturing system. Easy test: which fixtures, which line, how many units.
Q3: How should "slight colour difference" be defined? With limit samples, not adjectives: one upper and one lower physical boundary, signed by both sides, compared against production. Without them the argument cannot converge, because the problem is a missing standard rather than a quality defect.
Q4: When should EMC and safety get involved? EMC pre-scanning and a safety-oriented mechanical review — creepage, earthing, insulation — both at engineering sample stage. Neither should wait for certification, when the same change costs several times more and may drag the mould with it.
Q5: How much test data can a buyer reasonably demand? Down to per-unit, serial-numbered data with judgment limits attached. Customer details can be masked; the test values have no confidentiality rationale. A supplier sharing only conclusions usually means the data does not live in a system.
OEM amplifier sampling is a sequence of retiring uncertainty while it is still cheap. The functional sample retires "technically possible", the engineering sample "manufacturable", the pre-production sample "repeatable", the signed sample turns conformity into an object, and the first article proves the object can be reproduced. Run all five and mass production is execution. Skip any and mass production becomes sampling again.
To see one real line's insertion, assembly and burn-in stations, start with our assembly stations and test equipment in detail; earlier notes on sampling, inspection and production control sit in our amplifier engineering and manufacturing notes; the ranges these lines build are listed in the amplifiers and speakers we build today.
