Schematic, Stackup & RF Layout Review

The netlist passes while the RF path fails

How will currents, fields, waves, supplies, heat and manufacturing geometry behave in the physical product?

The node’s radio connects through its matching network to the antenna feed. The continuity check finds every intended net. Yet the L2 reference has a slot directly beneath the RF route. Imagine the board arriving with an unexplained mismatch or sensitivity change: which part of the apparently correct schematic would tell you where to look?

A drawn signal line is one part of a distributed structure. Its return conductor, dielectric and neighboring metal help define the fields. A low-resistance connection somewhere else on the board does not reproduce the reference directly below the route. In this exercise the slot is a known geometry fact; a detoured return and changed fields are the suspected mechanism. There are no measured failures. F-SLOT identifies a violation of a declared continuity constraint before the consequence has been quantified.

TI AN098’s stackup and ground-plane review is a useful starting point for identifying missing physical information. Its device-family examples do not authorize copying a trace width or plane split. We will keep the distinction between a known structure, a possible mechanism and evidence that the product meets its requirement throughout the lesson.

Bring 10.2’s architecture and interface record, Path 03’s line structures, Path 04’s real component models, 06.3’s antenna integration and 08.2’s lab practice. This lesson owns the physical implementation review that connects them.

Annotate energy, modes and reference planes

What does each connection carry, in which mode, and relative to which boundary?

Annotate the node QFN escape, matching region and antenna feed before optimizing their appearance. Then separate the gateway PA transmit branch, LNA receive branch and filter/duplex region. These are functional regions in a fictional package, not selected components with hidden ratings. The clock, switching supply and request/grant controls can become aggressors even though they are outside the wanted RF chain.

The wanted signal and the disturbing signal need separate annotations. A matched feed model does not explain supply-induced receiver degradation; a service deadline does not specify a launch impedance. Carry source/load conditions and the active operating mode with each claim. A 50 Ω target exists only at the declared RF interfaces, and remains a target until supported by geometry and characterization.

SERVICE-FR-v1 still asks for at least 990 of 1000 unique complete 32-byte payloads within 1 s, including timing uncertainty. Requests are 60 s apart with one retry at +200 ms. Its 100 m S0 geometry uses 1.5 m heights, vertical polarization, fixed azimuth, dry stationary conditions and 25±2 °C. A frequency sweep in the teaching calculator creates no new service mode or legal permission.

Review bias and decoupling across frequency

When the IC asks for transient current, which complete loop supplies it?

A capacitor symbol at a supply pin suggests a local charge reservoir. Its physical loop includes the package, escape, capacitor pads, vias and plane connections. Those connections have impedance too. Moving the same nominal capacitance onto a long narrow branch changes the structure the IC sees, even if the schematic value and DC voltage remain identical.

In ADI MT-101’s decoupling model, capacitance, equivalent series resistance and inductance jointly determine the response. Below self resonance the capacitive behavior may dominate; above it the series inductance can dominate. Multiple capacitors, a bead, package inductance and the surrounding network can introduce additional resonances. A regulator’s stability conditions also belong to the selected operating point and output network. “Add more capacitance” therefore needs a model and a reason.

For P-SUPPLY the default premise is more specific than “the regulator is nearby”: its hot loop shares a declared coupling region or return segment with LNA bias. The amplitude and dominant coupling mechanism are unknown. Hold wanted RF, gain state and traffic constant while changing the supply activity, then compare bias disturbance and receiver response. A bias waveform alone does not show whether the service requirement changed.

Specify the board that will actually be made

Could a fabricator reproduce the structure that the reviewer thinks is being analyzed?

“Four-layer RF PCB” leaves most of the line undefined. The signal conductor has a finished width and thickness; the reference is separated by a finished dielectric structure. Etch shape, plating, mask, layer registration and the via/antipad region can change what was modeled. A fabrication drawing must identify those quantities and the allowed variation before a target impedance becomes a reproducible engineering requirement.

STACK-LAYOUT-v1 proposes L1 signal over L2-GND and L4 signal referenced to L3-GND, with local supply routing. Both inner references are GND in this example. The numeric thicknesses and widths remain open. This is a declared teaching proposal, not the layer assignment copied from AN098 and not a finished board order.

Rogers’ RO4000 data distinguishes process and design dielectric constants with different methods and frequency conditions. A laminate’s catalogue value is not automatically the effective permittivity of the finished line. In this lesson εeff=3.0 is a supplied approximate model parameter; no laminate brand establishes it. If that basis is unknown, suppress propagation numbers while retaining the known slot and transfer facts.

Eurocircuits’ process information also illustrates why a published service description is not an agreement for a particular order. Agree finished dimensions and tolerances with the selected process. A geometric cross-section establishes geometry; a coupon measurement establishes a different quantity under its own calibration and decision rule. A conforming coupon is useful only to the extent that it represents the product’s material, layer, process and relevant transition.

Place by energy flow and access

Can the placement support both the operating current paths and the experiment needed to debug them?

Place the energy-critical connections first: the QFN escape and matching region, the PA/filter/launch route, LNA bias delivery and each switching current loop. Then reconcile heat spreading, assembly clearance and the ability to observe or replace a component. A clean left-to-right schematic can hide a long physical loop or an inaccessible tuning component under a lid.

For the gateway, preserve a reachable bias observation and a controlled way to compare the shared relation with the prescribed separate arrangement. Name where the reference lead lands and which probe state is permitted. Moving a probe ground to “nearby ground” can change the measured loop. Preserve the measurement arrangement in the configuration record so the second observation answers the same question as the first.

Clock placement deserves the same causal treatment. Consider two CTRL waveforms with the same 1 MHz repetition: one transitions gradually and the other reaches the same logic level much faster. Their period agrees, but the transition shapes and spectral distributions differ. The faster edge can excite a different portion of the interconnect response. A period label alone cannot bound coupling; use the owning time/frequency explanation and real-line model for the signal and structure together.

TI SWRA640H’s reference-layout examples include deliberately extended RF paths. A shortest possible trace is not necessarily the intended matching structure. Review any moved component against the actual device, footprint, stackup and assembly conditions; a generic spacing ratio cannot establish isolation.

Follow lines through every transition

Where does the return go when the signal changes layer, width or connector?

Walk the route through each pad, neck-down, via, antipad and launch. Each changes the surrounding structure; a signal via alone does not define how the reference changes from L2-GND to L3-GND. P-XFER therefore requires both references and a physically appropriate transfer proposal. Here the supplied repair describes an adjacent plated GND via connecting the two assigned ground planes, together with a scoped geometry-review premise.

Samtec’s launch decomposition identifies pad, via and plane geometry as parts of the transition problem. Mechanical connector fit does not determine all of that geometry. Use Path 03’s network and port conventions when combining a device, launch and fixture model; do not include the same launch twice or move the claimed DUT plane without a model.

A long controlled line may be intentional; a short discontinuity can still need review. There is no universal phase threshold that clears F-SLOT or F-XFER. The existing Transmission Line tool covers its own stated line/network domain; it does not characterize this board’s via, shield or antenna.

Treat ground and shields as physical paths

What physical boundary does the word “ground” actually identify at this frequency?

Ground names express intended connectivity. Return current follows the frequency-dependent impedance and field structure that the conductors actually provide. A slot, plane change, seam or connector can redirect those paths. Adding a stitching via may help a particular reference connection, but location, connected conductors and surrounding geometry determine which connection has been changed.

P-SHIELD makes the evidence gap visible: a lid is drawn, but the contact continuity and attachment impedance are unknown. A DC continuity check alone would still leave the frequency behavior unresolved. CONTACT-KNOWN supplies a fictional characterization premise naming the attachment, assembly state and frequency scope in the exact manifest. Its removal of F-SHIELD says that this contact-review question has supplied support; it says nothing about measured shielding attenuation.

Separate the seam/contact question from cavity and cable behavior. A boundary that improves local containment may alter a resonance or redirect common-mode current onto a cable. SWRA640H’s shield discussion is explicitly tied to its device and board examples. It supports checking the actual boundary, not a universal via-fence pitch, shield spacing or gain in dB.

Review the whole antenna environment

Does the antenna evidence describe the object that the user will install?

Empty copper in a keepout does not define the complete antenna environment. The battery, enclosure, cable, bracket, mount and user state can participate in the electromagnetic structure. Baseline evidence in this fixture describes the actual plastic/no-service-cable arrangement. The default layout intentionally includes a separately named metal/cable variant, so that evidence cannot simply be reused.

Preserve PLATE-FR-v1: the inherited aluminum plate is 200×200×2 mm, 10 mm behind the enclosure. CABLE-LAYOUT-v1 adds a service lead whose exact length, termination and route remain unknown; it is a new variant, not an unnoticed change to the original mounting record. Its presence creates a configuration question before a new antenna measurement is available.

TI AN058’s antenna measurement discussion addresses the measurement cable and actual surroundings. The cable used to obtain a match can itself alter the object being observed. A plane correction requires a valid model; reflection magnitude is not generally invariant through a lossy or mismatched fixture. Use Path 03 for that translation and 06.3 for the R2/S0 integration decision.

ANT-BASELINE removes F-ANT by restoring the actual baseline configuration, not by assigning the word “equivalent” to a new mount. That is a narrower conclusion than proving either product meets its service target. Keep antenna tuning and cable-routing options until comparable installed-state evidence supports their use.

Integrate heat, protection and resonances

Which neighboring physical system changes when you improve one part of the board?

A power stage draws peak current, dissipates heat and occupies a region that may also contain an RF reference and shield attachment. Follow its heat path through the real package, solder, board, contacts and closed enclosure. A plausible copper shape does not establish an allowable junction temperature; the selected device’s limits, actual dissipation and measured boundary conditions are still needed.

TI’s PowerPAD discussion makes solder attachment and PCB heat removal part of the package application. It does not provide a transferable temperature rise for this fictional QFN. TH-INTAKE-v1 retains ENV-CLOSED with natural convection and the planned 23/25/27 °C correlation conditions; package ratings and the actual heat/contact path remain unresolved.

An exposed RF or control interface also needs protection that is compatible during normal operation. As a reading example, TPD1E01B04’s §6.6 specifies line capacitance with bias, frequency, temperature and package conditions, while §10 gives layout considerations. That part is not selected for this radio. A low capacitance label cannot represent the complete clamp, footprint and return path, and a component ESD rating does not certify the system.

Keep normal RF loading and transient diversion as separate questions. Check powered, unpowered and reset states against the actual protected device’s limits and control contract. A removable bypass may be a useful experiment, but does not make a bypassed product an approved service configuration. A heat spreader or shield can also change antenna currents, cavity behavior and probe access. If an absorber is proposed, its material, frequency, placement, thermal and assembly conditions need evidence; no absorber performance is modeled here.

Issue findings that can be acted on

Can another engineer act on the finding and tell what evidence would close it?

“Bad ground” is not an actionable review. A finding needs the actual path fact, possible mechanism, affected interface, disposition, owner and next check. The map below evaluates those declared facts in a fixed package. It does not infer geometry from pixels, nearest neighbors or component names. Solid shapes represent illustrative geometry; dashed paths represent hypotheses, not solved fields or measured current density.

Schema, units and graph integrity are checked first. A valid state then separates supplied facts from evidence eligibility and emits all applicable concerns. Must-fix work precedes measure work, then preserve-option work, with stable finding-ID order within each class. This is a review-work policy, not a numeric risk score. The review graph is acyclic; the real circuit may legitimately contain current loops.

1. Predict the six default concerns, then inspect the facts and ordered findings.
2. Restore continuity with the direct P-RF control and commit: five concerns remain. Change P-SUPPLY to the prescribed separate arrangement and commit again: F-XFER, F-SHIELD, F-ANT and F-TEST remain.
3. Compare 2.45 and 4.90 GHz at fixed 10.0 mm and εeff=3.0. The propagation numbers change; the concern set does not.
4. Choose unknown material: suppress unsupported propagation numbers and add F-MATERIAL. Known geometry is still reviewable.
5. Compare actual plastic/no-cable and metal/service-cable manifests. Inspect packet applicability after a configuration change.
6. Load ALL-REVIEWED. No interactive concerns remain, while four static questions and mandatory physical review remain. Reset to restore the complete canonical case.

p10-m03-record-v1 | DEFAULT
MAN-LAYOUT-METAL-v1 | BOARD-LAYOUT-v1 | STACK-LAYOUT-v1
Illustrative review map: 6 concerns: F-SLOT, F-XFER, F-SHIELD, F-SUPPLY, F-ANT, F-TEST
F-SLOT | P-RF | must fix | The L2 slot is present beneath the referenced RF line.
Owner: RF layout reviewer. Next: Reroute over a continuous reference or justify continuity restoration; review actual geometry and correlate a representative structure.
F-XFER | P-XFER | must fix | L1/L2 → L4/L3 change is known; transfer is absent.
Owner: Interconnect/EM reviewer. Next: Describe both references and an adjacent GND-via connection, review pads/antipads and validate the complete transition.
F-SHIELD | P-SHIELD | measure | A lid is drawn; contact continuity and attachment impedance are unknown.
Owner: Mechanical/RF reviewer. Next: Record assembly state and frequency-scoped contact characterization; assess cavity/connector behavior separately.
F-SUPPLY | P-SUPPLY | measure | A shared hot-loop region/return segment with LNA bias is a declared geometric fact; coupling amplitude is unknown.
Owner: Supply/RF interface reviewer. Next: Characterize bias and receiver response with controlled supply activity, partition/reroute and unchanged wanted RF conditions.
F-ANT | P-ANT | preserve option | C-METAL plate/cable state is absent from baseline antenna evidence.
Owner: Antenna/mechanical reviewer. Next: Compare exact baseline and metal/cable manifests at R2 and S0; retain antenna and cable-routing options.
F-TEST | P-TEST | preserve option | Added tee pad/stub has no applicable complete parasitic and fixture review.
Owner: RF test/fixture reviewer. Next: Characterize populated/unpopulated/probed arrangements with explicit calibration planes and a scoped parasitic/de-embedding model.
Four separate synthesis findings: S-DECOUPLING, S-THERMAL, S-FAB, S-ESD.
No confirmed physical failures or product acceptance. Mandatory human/physical review remains.
Approximate uniform-line context: 70.647043 mm; 57.774996 ps; 50.957547 degrees.

Worked counterexamples · canonical teaching snapshots

This table is a fixed set of worked examples, independent of the custom committed map above. Each named preset starts from the complete canonical state. Direct controls accumulate: slot restoration plus the prescribed supply reroute removes two concerns; loading atomic QUIET-REROUTE alone restores the default slot.

Changing the board, stackup or manifest invalidates affected review-packet uses, retaining each historical packet. Incomplete contact evidence cannot clear F-SHIELD. An unmodeled reroute becomes discovery work; absence of an edge never proves isolation. Selecting a path or filtering rows changes only the view, and cannot change the evaluated population.

Canonical review package · exactly ten findings

The following issue register belongs to p10-m03-record-v1 at its default event 30, frozen 2026-09-08. It is deliberately separate from any custom map. Evidence IDs beginning EV or RP below identify illustrative records or supplied premises. Real publications in the references provide technical context; they do not approve the fictional board.