Bring-Up, Correlation & Tuning

A better match erases the experiment

Three component swaps, a moved cable and a deeper S11 dip. What can you still infer?

The first node reaches the bench. An engineer changes two matching capacitors and an inductor, moves the programming cable, then saves the improved return-loss plot. The original trace, part identities and enclosure state were never recorded. The radio now transmits, but the team cannot reproduce the comparison that made it look better.

The present configuration can still be characterized. What is lost is the controlled difference: there is no evidence that assigns the change to a particular component, cable current or measurement-plane shift. Even a plausible simulation fit cannot recover a missing observation. Preserve the current state as a new baseline, record what is unknown, and choose a reversible experiment from that point.

This lesson follows the fictional condition-monitoring node and gateway, teaching-node-v1. Its generic 2.450 GHz QPSK waveform remains 20 kbit/s, 10 ksymbol/s and RRC roll-off 0.35. It is not automatically Bluetooth, Wi-Fi or an authorized operating mode. We will separate the node’s antenna-tuning illustration from a gateway idle-current fixture so one convenient number cannot quietly become the other product’s specification.

Bring 10.2’s model contracts, 10.3’s physical findings and the measurement foundations: 08.1 uncertainty, 08.2 lab practice, 08.3 spectrum analysis, 08.4 calibration and fixtures and 08.5 TX/RX measurements. Here the new decision is how those methods enter a first-article sequence without destroying their diagnostic value.

Identify exactly what arrived at the bench

Does the hardware match the model, or merely the name on the project folder?

Two boards can share a revision label yet have different substitutions, assembly repairs or firmware. A substituted capacitor changes more than nominal capacitance: package, bias dependence and mounting can alter the network. A different finished stackup changes the line structure even if the CAD drawing is unchanged. The intake record must connect a serialled specimen to what was actually built.

Carry the schematic, BOM and substitution list, fabrication/stackup record, assembly inspection and rework history together. Name the radio mode, register image, region configuration and concurrency state. Then identify the fixture, instruments, calibration and reference planes used to observe it. A trace without those conditions is evidence of an indication, but may be inapplicable to the claim under review.

The portfolio planes remain semantic boundaries: D0 is application payload; R1 is a component RF port; R2 is the product antenna feed; S0 is a defined OTA geometry; R3 is the receiver detector/decision boundary, not the receiver RF input. Add P0-GW for gateway input power, keep P0-SUPPLY for the node, and name CTRL and T0 control/thermal interfaces. Only the declared RF ports use a real 50 Ω reference.

Inspect before applying power

Which observations must exist before an energized measurement is meaningful?

Start with polarity, population, solder joints, connector damage, contamination and obvious bridges. Use X-ray when the package, assembly risk or symptom makes hidden joints relevant; a photograph of a QFN outline does not inspect its underside. Check continuity and resistance under the correct de-energized and discharged condition. A low reading across a rail may reflect a parallel path or charging behavior, so compare the method and expected circuit state rather than declaring a universal resistance threshold.

Fluke’s power-off measurement precautions distinguish resistance/continuity from energized current measurement, including the connection, function and fuse state. The E5052B manual’s RF-port DC caution shows why an RF connector label alone does not establish DC tolerance. Account for residual charge, bias tees, DC blocks and source states at the actual port before connecting a test chain.

Prepare the supply sequence with the selected device and board instructions: setpoints, current limiting, reset/enable order, expected transient and steady states, observation windows, temperature checks and the person who owns a stop. A bench current limit is a protection setting with its own transient behavior; it is not proof that every device node is protected. USB, a programmer or an RF path may introduce another energy source.

The ADF4351 data sheet separates operating specifications from absolute stress ratings and specifies thermal/ESD conditions. UG-435’s reference-board connections demonstrate that even supply and unused-output arrangements belong to a particular assembly. Use those documents as examples of the information to locate; their values do not become limits for this fictional product.

Establish rails, controls and thermal behavior

Is unexpected current an RF fault, a state error, or an observation made on a different basis?

Correlate supply rails and sequencing with reset, clock enable, communications and thermal behavior. A controller that repeatedly resets may change both mean current and RF activity. A rail that reads correctly on a slow meter may droop during a load step. A warm part may be dissipating unexpected power, but the temperature observation also depends on the heat path, location and elapsed state. Preserve aligned observations before assigning an RF cause.

The editable fixture is gateway mean idle current at P0-GW, CP-BU-IDLE-v1, over a fixed synthetic 1 s interval. Its engineering expectation is [120,160] mA. The indication y and expanded uncertainty U define [y−U,y+U]. A separate exercise screen latches at H=200 mA when the upper endpoint reaches it. None of these numbers is a real device, instrument or battery limit.

JCGM 100’s expanded-uncertainty definition and NIST TN 1297’s reporting guidance require a coverage meaning and a clear measurement basis. Here U is supplied with k=2 and an approximate 95% normal/large-degrees-of-freedom premise. Manufacturing tolerance is a different interval describing expected design behavior. We deliberately compare the two intervals; we do not add tolerance and U or combine them by an unjustified root-sum-square.

Containment accepts equality at an expectation boundary: 155±5 gives [150,160]. At 160±5, only part of the interval fits, so the result is indeterminate. At 180±5, the intervals are disjoint: investigate a discrepancy even though the stop is not reached. If U is unknown at y=140, hold for missing evidence. If y=205 with unknown U, every allowed U≥0 already reaches H: latch the known stop and retain the missing-U reason.

Check the reference and LO beyond lock

The PLL is locked. Why can the clock still make the radio unusable?

Frequency and lock answer narrower questions than spectral purity. Examine the reference and register state, resolved spurs, phase-noise behavior and possible unintended oscillation under the intended load and supply. A lock indicator cannot express all of these observations. Equally, a spectrum peak is not automatically produced by the PLL: the measurement chain and a coupled digital source are competing explanations.

The SUPPLY-SPUR packet keeps lock=true but supplies a discrete tone at +1 MHz from the 2.450 GHz carrier. The statistic is a carrier-relative tone ratio with 1 kHz RBW/VBW, RMS detector and fixed sweeps, not a phase-noise density in dBc/Hz. The local upper criterion is −55 dBc. An observation −45±1 dBc gives [−46,−44], wholly above the criterion, so BU3 is held and BU4 cannot advance.

ADI’s PLL debugging guidance identifies supply, reference and digital coupling among possible spur mechanisms and treats register/lock inspection separately. In our case a peak at the switching offset first motivates a hypothesis. A supply disturbance can modulate a sensitive oscillator path; reference feedthrough or analyzer-generated products remain alternatives until a controlled intervention discriminates them.

Partition passive and active RF paths

Where does the model first stop agreeing with a controlled observation?

Partition the physical path so an active device, passive network and fixture do not become one indistinguishable failure. A passive coupon test has a de-energized isolated state and a defined extraction plane. An active gain or loopback measurement has a different bias and excitation contract. Carry those states through the comparison; “the VNA was calibrated” does not identify the DUT boundary.

Keysight’s error-correction guidance distinguishes systematic, random and drift errors. Correction at a known plane is useful only with the corresponding standards, connections and environment. A moved cable or changed fixture model can make an old correction inapplicable. Independent verification of a check artifact helps distinguish a setup problem from a DUT discrepancy.

Before that independent geometry observation, plausible alternatives include an extraction-window error, wrong calibration/fixture model, a nonrepresentative coupon, material variation and incorrect fabrication. Those alternatives call for different checks. Retuning the antenna until S11 improves neither verifies the coupon method nor changes the board’s finished stackup.

Build receiver understanding from known signals

Can you account for a known signal before trying to find the weakest one?

Begin receiver characterization with an identifiable signal inside the documented safe and linear input envelope. Work forward from the generator through attenuators, cables and injection access to the receiver RF input at R1. Then relate the receiver’s gain/control state to R3 decisions and the application observation at D0. “Strong” means useful for this comparison, not an arbitrary high power setting.

If the signal is absent at a downstream point, investigate the intervening block, configuration and fixture before lowering the generator toward sensitivity. A missing gain stage, disabled receive path or wrong demodulation state can make a sensitivity sweep look like a noise problem. Conversely, a highly compressed input can conceal a gain error that becomes obvious at lower levels.

Once signal transfer and gain state are understood, characterize noise and the agreed sensitivity criterion using the measurement owner’s waveform, bandwidth and packet-denominator rules. Preserve the input plane, source noise, attenuation uncertainty and synchronization assumptions. Do not substitute the 13.5 kHz ideal pulse-shape support for a receiver equivalent noise bandwidth.

BU5’s supplied packet documents the controlled progression; it does not invent a product sensitivity or a passed 100 m service test. CP-CONC-v1’s separate gateway backhaul aggressor remains a distinct configuration. When a blocker is eventually introduced, compare on/off states while preserving wanted level, receiver gain, frequency and packet count, and check that the test sources did not create the observed product.

Progress transmitter stress deliberately

Which behavior changes when the transmitter is asked to work harder?

Establish the load and protection chain before the low-power transmitter observation. Carry each attenuator, coupler, cable and sensor’s frequency, DC, peak, average and duration ratings, plus its correction at the measurement plane. The nearest part to the transmitter may dissipate most of the power. A suitable analyzer input level does not by itself establish that the intervening pad or load is within its own envelope.

The 8478B manual’s input discussion gives a concrete manufacturer example in which average power, pulse energy and peak power are separate constraints, and depend on the instrument combination. Our fictional LOAD-BU-v1 and PAD-BU-v1 have no borrowed ratings. Their complete metadata setting represents a supplied teaching prerequisite only.

BU6 first reviews low-power spectrum and power into a controlled load. BU7 then examines modulation, duty, supply and temperature through separately identified stress states. Supply droop can change PA operating behavior; heating can change gain or matching over a burst sequence. A scalar power meter integrates information that a spectral or demodulated measurement may need to retain.

Control antenna and measurement configuration

Did the antenna improve, or did the measurement configuration change?

At R2, a one-port measurement describes how the antenna system loads that plane. At S0, the observation depends on radiation, direction, polarization, geometry and the complete product. Changing a cable route, enclosure, mounting plate or nearby conductor can alter current paths. The cable can also participate in common-mode radiation. A stable R2 match does not establish a stable pattern, and a better match does not guarantee better service.

Follow 06.3’s configuration and plane control: preserve antenna/feed/tuner identity, battery/enclosure/shield state, cable route and termination, mount, object/user state and orientation. State where the calibration plane lies relative to any matching element. Rotating the product for a directional comparison is a different change from attaching metal to the radiator’s environment.

BU8 therefore accepts only OTA configuration and fixture readiness. Its packet names CP-NOM-v1’s geometry and a proposed method. It contains no measured S0 performance. This makes NORMAL-THROUGH-OTA a useful preparation for the tuning record without treating a completed fixture plan as a coverage or product release.

Tune, correlate, retain or revert

What evidence would make you retain this change—and what would make you revert it?

Choose one declared variable for this causal teaching experiment. Predict the direction of the response, keep the untouched controls and acquire the original complex observation before rework. Inspect the physical change, repeat the same measurement and test an alternative explanation. More advanced designed experiments can vary multiple factors deliberately, with controlled assignments and an analysis model; uncontrolled simultaneous swaps do not have that structure.

TUNE-BU-v1 supplies before/after complex points at 2.450 GHz: Z=25−j25 Ω and Z=50−j0 Ω, both at the real 50 Ω R2 plane. The nominal component change is an illustrative record field, not a network synthesis that predicts those points. Do not derive a realizable capacitor or inductor recipe from two supplied observations without a validated network model.

These exact analytic anchors use 10⁻¹⁰ absolute comparison tolerance in dimensionless/dB quantities; stage IDs and mA boundaries use exact integers. A perfect mathematical match has infinite return loss because −20 log₁₀|Γ| diverges as |Γ| approaches zero. A real instrument has finite uncertainty and dynamic range, so the fixture does not assert an infinitely precise measurement.

The console distinguishes reading this illustration from committing rework. A matching change creates a BU4 event and makes BU4–BU9 current uses stale or pending. BU9 cannot then be accepted until compatible upstream reacquisition/review exists. Reversion is another restoration event with inspected configuration and fresh specified checks; it is never a shortcut that silently revives old accepted flags.

Work through the evidence sequence

1. Load STARTUP, commit BU2 at 140±5 mA and observe that only BU3 becomes ready. Select a later stage to see prerequisite lockout.
2. Load CURRENT-STOP. Attempt BU3, then try a lower current at BU2. The stop remains. Supply the recovery packets in the order below, using 140±5 mA for BU2 reacquisition.
3. Load WRONG-STACKUP. Commit the same-board interpretation review at BU4 and reacquire: the physical discrepancy still holds. A new corrected PCB instead returns to BU0 intake.
4. Load SUPPLY-SPUR and inspect lock=true with the held spectral interval. Commit a controlled BU3 supply-coupling intervention (or BU2 if the power plan changes); select the corrected packet and CT-SUPPLY-v1, then reacquire at the required stage.
5. Load NORMAL-THROUGH-OTA. Compare the tuning points without changing the event count. S0 performance is still absent. Commit the separate matching-rework event at BU4 and observe six affected current uses.
6. Record restoration only with its required premise, then reacquire the affected stages. Reset starts a new fictional exercise with canonical facts; it does not resume stopped hardware.

Recovery is a new chain of evidence

1. ISO-STOP-v1: document de-energized/isolation premise; latch remains.
2. PLAN-RECOVERY-v1: disposition the fault/configuration and inspect the recovery plan; create a new scoped manifest.
3. ROLE-RECOVERY-v1: supply the authorized-review-role premise for this fictional case.
4. Reacquire compatible BU1 inspection/plan evidence under the new recovery scope.
5. Reacquire compatible BU2 containment evidence under the recovery plan; latch still visible.
6. Release only after the five preceding steps; preserve the stop and recovery chronology. BU3 becomes ready.