RO4350B RF Test Planning: Define Reference Planes Before Ordering
Define where the measurement begins and ends before specifying a loss limit. A connector-to-connector result includes structures that a de-embedded transmission-line result may exclude. Both can be useful, but they answer different questions. A supplier cannot reliably quote or execute an RF acceptance test from the phrase “test the RO4350B board” alone.
This guide extends the fabricated PCB testing overview with an original requirement worksheet and a reproducible fixture-loss example. It addresses passive PCB structures. Active devices, high-power tests and over-the-air antenna measurements require additional procedures and operating conditions.
Choose the measurand before choosing the instrument
State the property that must be demonstrated: connectivity, impedance, insertion loss, reflection, phase tracking or another explicitly defined quantity. Do not treat these as substitutes. A passing continuity test does not establish RF attenuation, and a good reflection result alone does not demonstrate antenna efficiency.
Attach a drawing that identifies ports, signal direction and reference planes. Mark whether connector bodies, launches, vias, transitions and matching components are inside the device under test. For a coupon, identify its relationship to the production stackup and relevant routing geometry. A convenient test structure may not represent every board feature.
Use material identity and construction revision alongside electrical requirements. Rogers' RO4000 datasheet distinguishes design and process dielectric constants and their test conditions. These material values are not acceptance results for a finished board. RO4350B is a candidate material, not an automatic solution for every low-frequency or millimeter-wave product.
Write a requirement that survives quotation and handoff
| Field | What to specify | Ambiguity to remove |
|---|---|---|
| Test object | Board/coupon ID, drawing and stackup revision, port mapping | Which physical structure does the report represent? |
| Frequency | Start/stop in Hz, frequency grid and critical frequencies | Is the requirement checked across the band or only at selected points? |
| Reference planes | Drawing locations and included launches/transitions | Does the reported loss include fixtures? |
| Calibration and correction | Method, standards, validity range and fixture treatment | Which errors are corrected and which remain? |
| Conditions | Power at the stated plane, temperature, terminations and relevant instrument settings | Can another lab reproduce the setup? |
| Acceptance | Limit, units, sign convention, uncertainty decision rule | What happens near a limit? |
| Sampling | Panel/unit selection, quantity and retest rules | Which population is supported by the evidence? |
| Deliverables | Raw complex data, plots, setup record and traceability | Can the data be reused and audited? |
Specify the sweep grid with enough resolution for the feature being assessed. A sparse grid can miss a narrow resonance. More points are not automatically better if repeatability, dynamic range or fixture stability is the limiting issue. Agree the settings and test duration with the laboratory instead of copying an unrelated instrument default.
Separate calibration from fixture removal
A calibration establishes corrected measurements at defined reference planes. If those planes are at cable ends, the PCB connectors and launches downstream can remain in the result. Moving the reporting plane into the circuit requires an appropriate calibration structure or a justified fixture-removal method. A change in a plot label does not remove a physical discontinuity.
Keysight's TRL guidance describes standards whose quality and repeatability determine calibration quality. It states that the thru and line establish the reference impedance, and that line length must be suitable for the measurement band; broad coverage can require multiple lines. Therefore, do not assume a nominally 50-ohm drawing makes an on-board calibration exactly 50 ohms at all frequencies.
Discuss the required standards before panelization. Record whether they share the intended transmission medium, and how their dimensions and definitions will be checked. Keep the calibration procedure version and a verification measurement with the test record. A correction that produces a plausible curve still needs independent validation.
Worked example: the cost of ambiguous reference planes
Assume a hypothetical passive path at 6 GHz has 0.40 dB loss in the PCB section of interest and 0.15 dB in each launch. Under an adequately matched cascade approximation, connector-to-connector loss is 0.15 + 0.40 + 0.15 = 0.70 dB. The inner PCB section remains 0.40 dB in this simplified bookkeeping model.
| Reporting boundary | Included elements | Illustrative loss |
|---|---|---|
| Connector to connector | Two launches plus PCB section | 0.70 dB |
| PCB section only | Inner section after valid fixture treatment | 0.40 dB |
A requirement of “loss no greater than 0.50 dB” gives different conclusions for those two boundaries. It is incomplete until the boundary is specified. The example does not authorize subtracting 0.30 dB from real data: reflections interact, fixtures vary with frequency, and scalar subtraction generally cannot recover a mismatched network. Use validated complex network models or a suitable calibration method.
For a matched power-wave reference, report insertion loss as −20 log10|S21| in dB when using that definition. Distinguish it from an S21 magnitude plot expressed as 20 log10|S21|, which is negative for an attenuating passive path. State the convention explicitly to avoid reversing an acceptance inequality. See the insertion-loss budgeting guide for path-level context.
Specify the report and the near-limit decision
Request raw complex S-parameter data when it is useful for model correlation, together with frequency units, port order, reference impedance and correction status. Retain the original measured dataset separately from processed results. A screenshot without these definitions is difficult to reuse reliably.
Record the fixture revision, calibration date, relevant environmental conditions and sample identity. Ask the laboratory how uncertainty and connection repeatability affect the proposed limits. Agree a decision rule before testing: near-limit results may need a guard band, further review or additional measurements under an agreed procedure. Do not choose a rule after seeing which one produces a pass.
Sampling must reflect the actual purpose. Development coupons can compare candidate constructions; production acceptance needs a defined relationship to panels and shipped units. Neither a single good coupon nor an attractive simulation establishes the distribution across an entire production lot. Preserve failed and repeated results with their disposition rather than reporting only the best trace.
Primary sources and scope
- Keysight: TRL calibration — standard quality, reference impedance and frequency coverage.
- Rogers RO4000 datasheet — material-property definitions and conditions.
Sources reviewed 16 October 2026. The worksheet and numerical example are original editorial work with hypothetical inputs. They do not represent measured results or manufacturer endorsement.