An LNA noise-figure result belongs to a defined pair of reference planes. A measurement at a board’s input connector includes the loss ahead of the amplifier unless that loss has already been removed. Before blaming the IC or changing the laminate, establish which network the reported number describes.
This procedure separates a connector-plane result from a device-plane estimate for a civilian receiver. The numerical example is hypothetical, not measured hardware performance. Use the RF test and measurement reference to organize the fixture documentation.
Draw the measurement boundary first
Sketch the chain from the noise source or VNA source plane through adapters, input launch, PCB route, amplifier, output route and receiver. Mark the calibration planes and desired reporting planes on that sketch. Decide whether input matching components, DC blocks and bias structures belong to the product being specified or to a removable test fixture.
A complete receiver module specified at its connector still includes its real input-path penalty. A separate device-plane estimate can help diagnose that penalty, but it does not replace the module result. Likewise, do not subtract a PCB correction from a trace that the instrument has already de-embedded.
Record frequency, supply voltage, bias current, board temperature and operating mode. Analog Devices’ LNA selection note explains that noise-figure dependence on bias and supply varies by device. Compare with the actual part’s documented conditions rather than assuming a typical front-page value applies to every setup.
Characterize the fixture that was actually assembled
Save calibrated two-port S-parameters for the removable input path across the measurement band. Include its launch, soldered connector and any intervening component. Use a representative coupon or a validated fixture extraction when the device-side plane cannot be contacted directly. Document the extraction assumptions and check repeatability after reconnection.
For a well-matched passive path, positive insertion loss in dB can be estimated as −20 log10(|S21|). Record S11 and S22 as well: a single transmission magnitude is not a complete correction for a mismatched network. Do not simply divide a back-to-back coupon’s dB loss by two unless its halves and their interaction support that approximation.
Analog Devices’ RF layout guidance relates microstrip impedance to trace width, dielectric spacing and dielectric type, with a continuous reference plane. Therefore, identify the ordered construction of a RO4350B fixture, including copper, finish and mask, rather than assigning a correction from the laminate name alone.
The dielectric-loss estimator can help explore a material contribution; it is not a measurement of connector, conductor, component and mismatch effects together. Review the edge-launch footprint checklist when the test board differs from the connector’s reference stackup.
Apply a limited passive-loss calculation
Use linear noise factor F = 10^(NF/10), and positive linear loss L = 10^(loss_dB/10). The cascade equations in Rohde & Schwarz’s cold-source noise-figure application note use available gain and a standard reference temperature T0 of 290 K. Under matched conditions, a passive input fixture at T0 has noise factor L and gain 1/L.
For that fixture followed by the amplifier, Ftotal = L + (FA − 1)/(1/L) = L × FA. Therefore, NFtotal = loss_dB + NFA. Subtracting the input loss in dB is valid for this limited model, with output-fixture and receiver contributions already accounted for.
| Input to the calculation | Assumed value | Meaning |
|---|---|---|
| Connector-plane noise figure | 2.10 dB | Includes the input fixture; receiver and output path already corrected |
| Input fixture loss | 0.35 dB | Well matched, passive, at 290 K |
| Estimated amplifier noise figure | 1.75 dB | 2.10 − 0.35; applies at the newly defined device plane |
| Alternative loss assumption | 0.30 / 0.40 dB | Changes the estimate to 1.80 / 1.70 dB |
The last row is a sensitivity exercise, not a measured uncertainty interval. Repeat the correction frequency by frequency using the actual fixture data. If its physical temperature differs from 290 K, use temperature-aware loss compensation or the appropriate passive-noise model; simple dB subtraction is no longer the exact correction.
Do not treat output loss like input loss
In the same matched, 290 K model, an output loss Lout after an amplifier with available gain GA contributes (Lout − 1)/GA to the amplifier-plus-output-path noise factor. Including an input loss Lin gives Ftotal = Lin × FA + Lin × (Lout − 1)/GA, before adding any receiver contribution. These are linear quantities.
Thus, subtracting both input and output insertion losses directly from the measured noise figure is generally wrong. Record the instrument’s input-loss, output-loss and receiver-calibration settings separately. If a pad or cable is added after calibration, follow the selected instrument’s procedure to include it correctly.
Check mismatch, interference and uncertainty
A source mismatch can change the noise delivered to the amplifier and the result inferred from it. Keysight’s noise-figure application guidance distinguishes scalar assumptions from vector correction using multiple source impedances. Use a method appropriate to the DUT and fixture; a loss-only spreadsheet cannot reconstruct missing noise parameters.
The same guidance identifies external interference as a cause of spikes and possible receiver compression. Keep gain and noise measurements in their valid operating ranges, investigate nonrepeatable features, and check a shielded setup where appropriate. Averaging a contaminated trace does not establish the amplifier’s intrinsic noise figure.
| Contribution | Record or check | Reason |
|---|---|---|
| Noise-source calibration | Source identity, ENR data and applicable uncertainty | Required for a defensible Y-factor measurement |
| Fixture correction | Loss versus frequency, temperature, match and extraction method | A correction has uncertainty of its own |
| Measurement repeatability | Repeated sweeps and controlled disconnect/reconnect trials | Separates trace jitter from connection variability |
| Receiver and gain | Calibration state, gain data and linear operating range | Prevents a receiver limitation becoming an apparent DUT penalty |
Keysight’s Y-factor measurement documentation identifies ENR, mismatch, jitter and user-calibration uncertainty contributions. Build a budget for your own method and frequency. Repeated sweeps alone do not quantify systematic uncertainty.
As another hypothetical check, suppose the 2.10 dB result is bounded by 1.98–2.22 dB and the input loss by 0.30–0.40 dB. Endpoint subtraction gives 1.58–1.92 dB for the limited corrected model. This is interval arithmetic, not a confidence interval or a complete uncertainty budget. Do not combine unrelated error limits statistically without knowing their definitions and correlations.
Report both planes and request a fixture review
Retain the connector-plane trace, corrected estimate, applied corrections and uncertainty statement together. State which matching network remains inside the reported DUT. Use the qualification-testing guide to turn this record into a repeatable acceptance procedure.
Send fixture artwork, stackup, calibration method and planes, measured insertion loss and S-parameters, noise-figure traces, bias conditions and temperature. Include a screenshot or export of enabled corrections. Request an LNA fixture PCB review, or assemble the fabrication requirements with the RFQ checklist builder.