A NEC Plot Is Not a Measurement: The KJ6ER Primer

In the July 2026 edition of Greg Mihran, KJ6ER’s antenna primer, the PERformer’s headline efficiency is followed by model specifications, radiation graphics and then analyser sweeps. That sequence deserves a closer look. The graphs are real outputs, but they are not all evidence of the same thing.

I use models because they can expose a poor idea before copper is cut and explain currents that are awkward to probe. My objection is not to NEC, nor to building an antenna from a model. It is to letting the confidence earned by one measurement spill over into quantities that measurement never observed.

Greg’s Own Method Sets the Right Standard

Slide 9 includes modelling, backyard construction and portable operation. Slide 39 explicitly calls for field testing over real ground after calculation. That is good advice, and it deserves credit. The criticism here applies that standard to the numerical performance claims; it does not pretend Greg recommends replacing all experiments with software.

The edition matters: the slide numbers below refer to the linked July 2026 PDF. KJ6ER’s own site describes the primers as living documents. This is a discussion of the identified presentation, not a claim about everything its author has ever measured or may publish later.

Follow the Evidence from the Model to the Analyser

That is a concrete limit on the case being made, not a verdict that the PERformer cannot work. The supplied plots and match traces do not demonstrate the headline radiation quantities experimentally. An undisclosed or incompletely documented model may nevertheless be right. The way to make its predictions persuasive is to expose the calculation and compare it with the appropriate physical observation.

Prediction, Measurement and Validation Are Different Jobs

The distinction is not an argument against NEC. Lawrence Livermore National Laboratory describes NEC as a numerical tool that can calculate currents, fields and radiation patterns from a specified model. LLNL’s own NEC-5 validation manual separates numerical modelling errors from physical modelling errors caused by simplifying or omitting parts of the real structure.

That gives us four useful questions:

LLNL’s verification-and-validation guidance makes the same boundary explicit: verification assesses the numerical implementation, while validation compares predictions with experimental data. A model may converge beautifully and still omit the coax, mast or actual soil. A measurement may be perfectly real but measure only SWR when the claim concerns gain.

“Modelled in 4nec2” Is Not a Complete Method

4nec2 is a modelling environment and front end. It prepares input, invokes an engine and renders the output. Its support page distributes distinct NEC-2 engine executables; its release history documents changes involving engines, auto-segmentation, preprocessing and display.

That history also records real corrections, including a normalised-gain bug fixed in 4nec2 5.9.3 and earlier NEC-4.2 ground-preprocessing fixes. The lesson is simple: name the complete calculation path.

For the primer, “extensively modelled” tells us work was done, but not which executable and settings produced each result. Software history is a reason to identify those details—not evidence that Greg used a defective release.

• 4nec2 version;
• NEC engine, version and exact executable or build;
• frequency points and sweep method;
• thin-wire or extended-kernel choice;
• automatic segmentation and preprocessing;
• optimiser objective, constraints and settings; and
• average-gain correction or other post-processing.

Publish the Model, Not Only Its Screenshot

A reproducible numerical claim needs the files another competent modeller would use:

1. the original 4nec2 authoring file, including symbols;
2. the expanded NEC input actually sent to the engine;
3. the complete raw output, including warnings;
4. optimiser, sweep and plot settings;
5. a README mapping every published figure to a file and run; and
6. a date, version or checksum identifying the archive.

The input reveals geometry, source, loads, networks and ground cards. The output reveals segment currents, power entries, junctions and warnings. A cropped plot reveals almost none of them.

The primer itself shows why a file-to-figure map helps: its 15 m gain is 0.26 dBi in slide 42 and 0.31 dBi in slide 45. That small difference could reflect a revised model or configuration. Without the corresponding runs, it cannot be diagnosed. It is a reproducibility question, not proof of deception or poor antenna performance.

Geometry Includes the Return Structure

For every conductor, disclose coordinates, radius, material, length and segmentation. For an elevated-radial vertical, that includes the radiator, radial count and azimuth, droop, feedpoint height, source segment, mast, stakes, supports and any conductive control wiring.

The NEC-2 User’s Guide spends considerable effort on wire junctions, segment geometry, source placement and thin-wire restrictions because the solver analyses the geometry it receives—not the antenna the author had in mind.

A photograph is useful beside the model. It may reveal conductors that the model omitted. With only a few elevated radials, the outside of the coax, tripod or mast can become part of the return structure. Omitting such a conductor is an assumption, not proof that its current is zero.

Ground Is an Electrical Input

“Real ground” is not one material. A report should state:

• ground formulation;
• conductivity and relative permittivity;
• frequency dependence;
• whether the values were measured, selected from a preset or assumed;
• homogeneous, two-medium, slope or terrain assumptions;
• whether radials were explicit wires or a screen approximation; and
• the precise conductor clearance above the interface.

The primer uses a 36-inch radial-end height. That is a physical distance, not a universal electrical clearance: at 7.2 MHz it is only about 0.022 wavelength. The NEC-2 modelling guidance says that its reflection-coefficient approximation has limited accuracy for structures close to ground, while the Sommerfeld/Norton wire-ground treatment is intended for close interaction. It also warns that a real ground stake is not accurately represented by simply running a wire through the interface.

Rudy Severns, N6LF, disclosed these choices carefully in his study of elevated ground systems: software, engine, frequency, conductor assumptions and soil parameters were named, and NEC-2 close-to-ground limitations were treated as limitations rather than hidden.

Every Watt Needs a Defined Destination

A model can be a legitimate ideal benchmark and omit real losses deliberately. Trouble starts when ideal-conductor or “structural” efficiency is presented as installed-system efficiency.

State whether the model boundary includes:

• finite conductor conductivity and skin-effect loss;
• loading-coil resistance or measured Q;
• capacitor, trap, connection and insulation loss;
• matching-network or transformer loss;
• feedline loss and outside-shield current;
• complex choke impedance at every frequency; and
• soil dissipation under the selected ground solution.

The NEC-2 output documentation distinguishes source data, currents, power budget and field output. The report must be just as precise. For one declared boundary, a schematic power balance is:

Only include terms that fit the chosen boundary, and do not count the same loss twice. An omitted path is not a measured zero.

Do not assume that the printed NEC power budget is already this complete physical balance. The NEC-2 manual defines its reported radiated power by subtracting structure and network losses from input power. Over lossy ground, the definition and treatment of power coupled into the ground also matter. A displayed efficiency label needs its calculation method and boundary, not just a percentage sign.

That is why the primer’s structural-efficiency and separate transformer/unun figures must remain separate until their reference planes and loss accounting are reconciled. A model of the radiating structure can be useful without being a model of the complete station. It becomes misleading only when one is used as evidence for the other.

A Finished Run Is Not Necessarily Converged

Repeat the physically equivalent model with progressively refined segmentation. Track at least feedpoint resistance and reactance, absolute gain, the claimed peak angle, efficiency or average-gain result, and selected complex currents. Important outputs should approach stable values.

More segments are not automatically better. Very short or badly proportioned segments, unsuitable junctions and poor source placement can introduce new errors. A useful convergence test refines discretisation while preserving physical dimensions, loads and source location as closely as the method permits.

Convergence establishes stability for the chosen mathematical model. It does not validate omitted conductors, incorrect soil or an idealised loss term. Run a separate sensitivity study for plausible soil, radial height and sag, component Q, feedline route, choke impedance and nearby metal.

Currents Explain the Pattern

Publish complex current—magnitude and phase—on the radiator, each radial at comparable positions, the coax exterior, mast, nearby conductors and matching branches. This is often the fastest way to discover what the solver is actually treating as the antenna.

Current continuity at a junction is enforced by the model equations when the geometry is valid; the engineering audit is to verify that the junction represents the physical connection and that current division remains credible across segmentation and frequency. A symmetric drawing does not guarantee symmetric current after soil, feedline and environmental asymmetry are introduced.

Severns’s work is particularly valuable because it compared measured and modelled radial currents. Sparse elevated systems can be sensitive to modest asymmetry. A lobe can be pattern redistribution caused by an unintended current path rather than an increase in total radiated power.

Define the Plotted Quantity

For each radiation plot, state:

• directivity, gain or realised gain;
• dBi, dBd or normalised relative level;
• the accepted- or incident-power reference plane;
• polarisation component;
• theta/phi or elevation/azimuth convention;
• frequency and angular resolution;
• ground-wave or space-wave calculation;
• full 3D pattern or selected cut; and
• absolute peak plus the value in the comparison direction.

The active IEEE 145-2025 antenna terminology standard is the current formal reference for antenna terms. The practical point is that directivity, gain, realised gain and a normalised plot answer different questions.

A plot normalised to its own maximum sets its best direction to 0 dB regardless of absolute performance. It can compare shape, but not prove that one antenna radiates more accepted power. A 3D renderer may also auto-scale two patterns until both fill the window. Colour and visual size are presentation choices; publish the underlying absolute data.

Applied to the primer’s radial-orientation comparison, a stronger lobe in the displayed cut can be an interesting design result. It does not by itself tell us whether total efficiency improved or power moved from another direction. The useful next comparison is the absolute pattern and current distribution for the two complete arrangements, followed by a field test of the relevant direction—not a vote for whichever picture looks larger.

What the VNA Does—and Does Not—Validate

A calibrated VNA or antenna analyser measures complex reflection coefficient at its calibration plane. From that it can derive impedance, return loss and SWR. Agreement with the model across frequency is useful partial validation of input behaviour.

It does not, from S11 alone, separate radiated power from dissipation in soil, conductors or transformers, or identify power flowing along an unintended external path. Common-mode current can radiate as well as lead to dissipation; it is not itself a synonym for loss. Converting SWR into a percentage calculates mismatch efficiency:

An SWR of 1.10:1 means about 99.77% of the incident power is accepted at that plane. It does not mean 99.77% is radiated. A dummy load can be well matched and radiate almost nothing; an efficient antenna can be mismatched before a tuner is added.

The percentage box beside the primer’s analyser traces is therefore a matching calculation. It cannot validate the neighbouring radiation-efficiency values. The photographs remain useful measurements: they answer the input-match question, not every question in the model table.

Efficiency itself needs a measurement designed for efficiency. NIST’s work on reverberation-chamber antenna efficiency illustrates why reference conditions, statistical treatment and uncertainty belong to such a claim. A one-direction field-strength comparison can validate a relative field prediction in that direction; it is not automatically an absolute efficiency measurement.

Match the Test to the Claim

The active IEEE 149-2021 recommended practice for antenna measurements provides the formal measurement framework. The practical rule is to measure the same quantity the model predicts, at a declared reference plane and with an uncertainty statement.

For a relative field or pattern test, document the physical geometry, frequency, soil and surroundings. Calibrate impedance at the feedpoint; normalise to equal accepted power at the declared boundary; keep the receiving antenna’s position, height and polarisation fixed; use A–B–A or randomised switching; measure noise floor and drift; and repeat enough azimuths to reveal pattern skew.

Rudy Severns’s ground-system method documented transmission comparisons, receiving geometry, isolation, calibration checks, stray coupling and current measurements. His four-elevated-versus-64-surface-radial experiment combined relative field, impedance and current evidence and repeated the sequence on three days. He kept the conclusion within the experiment’s scope instead of relabelling one-direction relative field as absolute efficiency.

This is the constructive comparison with the primer: not “Severns used better-looking graphs,” but that he connected a physical change to a documented measured response. That is how an attractive radial-layout prediction becomes a useful engineering result someone else can test.

The Publication Checklist

Takeaways You Can Trust

• NEC is a powerful engineering tool; its output is a conditional prediction.
• Greg’s primer correctly calls for field testing; its analyser traces validate match, not its headline radiation figures.
• 4nec2 is a front end, so identify the solver, version, preprocessing and options.
• Source and output files make a result auditable; screenshots do not.
• Numerical convergence does not prove physical completeness.
• Close-to-ground wire models require careful ground formulation and geometry.
• Structural, radiation, mismatch, network and complete-system efficiency are not interchangeable.
• Complex currents can reveal imbalance, feedline radiation and omitted conductors.
• A normalised pattern does not show absolute gain.
• SWR measures input match, not efficiency, gain or radiation angle.
• Every claimed quantity needs a measurement designed for that quantity.
• Display precision is not measurement accuracy.

In Summary

Greg’s primer gives readers something useful to build and it recognises that modelling must meet the real world. I would use its model results to investigate the proposed geometry. I would not treat the neighbouring SWR traces as proof of the published gain, efficiency or radiation angle. That distinction changes what the reader can reasonably conclude from the presentation.

The practical choice is not modelling or measurement. Use the model to understand why a layout may help, then validate the claimed benefit. For the PERformer comparison that means connecting the modelled radial currents and directional field to the installed arrangement, rather than asking a match trace to certify them. Publish the files, define the losses and ground, show numerical stability, and measure the physical quantity being claimed. That closes the loop Greg’s own field-testing advice opens.

Questions, antenna-factor records or height trials to share? Contact RF.Guru.