Engineering resource
Portable diagnostic PCBA: low-noise assembly considerations
A portable diagnostic PCBA can pass a connectivity test and still produce unstable measurements when the finished instrument wakes its display, charges its battery, or communicates with a host. The assembly release therefore needs more than a schematic and a request for a low-noise build. It needs a reproducible definition of the measurement, a controlled physical configuration, and records that distinguish a board problem from a test-station problem.
This article proposes a manufacturing handoff for a small instrument with an analog sensor interface and an ADC. Its examples concern bench electrical verification using simulated inputs. They do not establish diagnostic accuracy, patient suitability, or certification. The product owner must separately define those requirements. Start by documenting the particular application requirements, because an optical reader and an electrode-based instrument do not have interchangeable sensor interfaces or acceptance limits.
Freeze the measurement before ordering the pilot build
Write a measurement configuration sheet alongside the assembly drawing. Identify the input connector, channel, input range, gain setting, conversion rate, digital filter, firmware revision, and output units. State whether the recorded values are raw converter codes, calibrated engineering values, or processed application results. A production technician should not need to infer any of these choices from a screen capture supplied by a development engineer.
Describe the operating state as carefully as the signal path. Specify battery or external supply operation, display brightness, radio activity, charger state, enclosure condition, and the interval between power-up and acquisition. Give each approved configuration an identifier. If the design has several legitimate measurement modes, create separate records instead of allowing the operator to choose whichever mode produces the smallest number.
Texas Instruments explains that shorted-input and driven-signal measurements answer different ADC questions, and that gain, filtering, and other signal-chain components affect noise results. Its precision ADC noise discussion supports a practical distinction: a quiet converter test does not characterize every operating condition of the assembled instrument. Use component documentation to define valid input biasing and electrical limits before designing either fixture configuration.
Mark sensitive assembly areas explicitly
The drawing package should identify the analog input region, reference circuit, sensitive connectors, shield attachments, and areas subject to special handling. Use reference designators and drawing coordinates rather than descriptions such as near the small amplifier. A marked inspection image can help, but it should carry the same revision identifier as the assembly instruction and should not be the only source of dimensional requirements.
Analog Devices describes how leakage into high-impedance nodes and thermoelectric voltages at dissimilar-metal junctions can compromise precision measurements. Its precision amplifier layout guidance also discusses cleaning and circuit-dependent guarding. These mechanisms justify reviewing the actual input region; they do not justify adding a generic ground ring to every design. Any guard connection, exposed copper requirement, or layout change belongs in an approved engineering revision.
For assembly planning, convert that review into explicit handling instructions. Identify where fixture contacts may touch, where labels may be attached, and which surfaces must remain accessible for inspection. Ask the designer to resolve ambiguous shield or guard connections before the pilot lot. An operator should never decide that an unlabeled copper feature is a convenient grounding point simply because it is easy to reach.
Turn cleaning into a controlled process decision
Prepare a process compatibility matrix using the actual BOM. For each sensitive sensor, connector, display, switch, and packaged component, record the manufacturer’s applicable cleaning restrictions and the proposed process. Mark unresolved entries for engineering review. A general description of the board as washable is not a substitute for checking whether the installed parts and mechanical subassemblies tolerate that process.
Agree the evidence needed for process approval before evaluating trial boards. This may include inspection records, relevant cleanliness measurements, electrical results, and controlled environmental evaluation selected by the product engineer. Keep the criteria and methods distinct: a photograph can document appearance, while an electrical test answers a particular performance question. Neither should be silently substituted for the other when a lot is released.
Define the sequence around shielding, coating, and final closure. If a shield hides a region that needs inspection, require the inspection at a documented earlier stage. If coating is specified, record the material, masking drawing, cure procedure, and acceptance method. Do not add coating as an improvised cure for an unexplained test failure. Include these decisions in the project’s quality and traceability planning so that a later build uses the same approved process.
Qualify the fixture independently of the board
Assign a fixture revision and identify its cables, input simulator, power source, switching arrangement, software, and measurement equipment. Record the reference plane at which the specified input is applied. If a cable or relay sits between the source and the board, include it in the configuration record. A technician replacing that item should be able to determine which checks establish that the station remains suitable.
Use an approved station verification routine at the start of a run and after relevant maintenance. The routine should detect incorrect connections, configuration errors, and changes in the measurement path. Keep a reference assembly where useful, but document its purpose and limitations. Comparing every unit with a single aging board cannot by itself establish measurement accuracy or traceability.
Before rejecting a production unit, preserve its original result and repeat the defined station checks. Then use a controlled comparison to separate unit and fixture behavior. For example, test the suspect unit on a second qualified station without changing its firmware, and test a reference unit on the original station. Record both outcomes. Repeatedly moving cables until a passing result appears destroys the evidence needed to understand the problem.
Use a worked noise record with explicit units
Consider an illustrative bench test in which converted samples at the amplifier output have a standard deviation of 80 microvolts and the configured signal gain is 20. Referring that variation to the amplifier input gives 4 microvolts RMS. This is an arithmetic example, not a suggested product limit or a claimed factory result. It assumes the conversion from codes to output voltage and the gain definition are already established for that test.
Now suppose a display-active acquisition produces 120 microvolts RMS under the otherwise identical setup. The corresponding input-referred value is 6 microvolts RMS. Record the difference as an operating-state observation. Do not immediately attribute it to display routing: the changed state may also alter supply loading, firmware timing, or the acquisition sequence. The next experiment should isolate a specific suspected cause while retaining the original records.
The record should include the sample count, acquisition duration, units, mean, standard deviation convention, and any rejected samples. State whether startup samples were discarded and why. Preserve raw samples so the calculation can be checked. A peak-to-peak result from a short capture should not be relabeled RMS, and a comparison using a different acquisition interval should be flagged rather than treated as an equivalent repeat.
Acceptance requires limits approved for the actual product and test uncertainty. If a result is close to a limit, use the agreed disposition procedure rather than rounding it down. Record a retest as a new result linked to the first attempt. This prevents a borderline unit from appearing indistinguishable from a unit that passed its original test without intervention.
Separate characterization from the production screen
Use the pilot stage to decide which conditions warrant engineering characterization and which can be screened efficiently on every assembly. Characterization may investigate multiple operating modes, environmental conditions, and configuration changes. The production screen should retain the checks needed to detect the agreed manufacturing failures. A shorter screen is justified by documented coverage, not simply by choosing the fastest test from a successful prototype demonstration.
Create a coverage table linking each relevant failure mode to an inspection or electrical check. For example, a missing filter component may require component verification plus an electrical response check, whereas an incorrect firmware setting requires configuration read-back. Keep uncovered risks visible. Label a test as sampling where it is not performed on every unit, and identify the sampling rationale and owner rather than implying universal coverage.
For each test, document the action after failure. Specify whether the unit is quarantined, referred to engineering, or eligible for a controlled diagnostic sequence. If the same symptom recurs across several units, preserve their lot and station information. The investigation may need to consider shared materials or process changes instead of repeatedly repairing individual boards without identifying a common cause.
Rework must restore the evidence chain
A reworked assembly should carry the original serial number and a clear record of the operation, replaced parts, operator, and authorization. Identify which earlier inspections and measurements are invalidated by that operation. Replacing an input component, removing a shield, and updating firmware can require different verification steps. The repair instruction should point to those steps explicitly instead of relying on the technician’s memory.
Record any cleaning or thermal processing performed during repair using the applicable approved instructions. Keep engineering deviations visible until they are formally resolved. If a repair introduces a permanent change that later becomes the standard build, update the design package, BOM, process instructions, and fixture configuration together. Otherwise, the next lot may reproduce the original fault while the team believes the problem has already been fixed.
Release a package that another team can reproduce
Before ordering, use the release-file checklist to reconcile the BOM, placement files, drawings, and test attachments. Add the measurement configuration sheet, fixture revision, source simulator definition, and acceptance procedure to that package. Assign an owner to each unresolved input. A purchasing note requesting low noise cannot resolve missing gain settings, an undefined connector state, or an unspecified test bandwidth.
The release review should end with a reproducible sequence: build the approved revision, inspect the identified regions, configure the instrument, verify the station, acquire the specified data, and disposition the recorded result. Retain the underlying evidence with the unit and lot identifiers. This makes low-noise performance an observable manufacturing requirement that can be investigated and repeated across builds, while keeping overall diagnostic validation with the responsible product team.