A NEC Plot Is Not a Measurement

A reproducibility checklist: source files, ground model, losses, convergence, currents and comparative field measurements.

A NEC model can be an excellent engineering tool. It can expose a poor idea before copper is cut, isolate one variable at a time and reveal current or pattern behaviour that is difficult to measure directly.

But a coloured polar plot is not a measurement. It is the rendered answer to a mathematical problem selected by the modeller.

That distinction matters in Greg Mihran’s July 2026 antenna primer. Slide 31 says that the PERformer was modelled extensively in 4nec2 and averages more than 90% efficiency across six bands. Slide 42 lists an average efficiency of 90.8%, average gain of 0.41 dBi, a 3.09 dB front-to-back ratio and a 24-degree peak angle. Slides 43–48 show polished 4nec2 graphics. Slide 50 presents a “radiation sweep” that is also model-derived.

Slide 51 then shows real analyser sweeps. Those are measurements—of input match. They do not measure the efficiency, gain, front-to-back ratio or radiation angle printed on the preceding slides.

The missing bridge is reproducible validation.

First, give the primer credit where it is due

Slide 39 says that 4nec2 calculations are a starting point and that testing over real ground must be performed. That is correct. Slide 9 also presents modelling as only one part of a workflow that includes backyard construction and portable operation.

The issue is not that Greg used NEC. The issue is that the evidence shown later does not validate the quantities being claimed.

Antenna operation is demonstrated by making contacts. Input match is measured by an analyser. Gain, efficiency and radiation pattern require different measurements. One kind of evidence cannot silently stand in for another.

What the primer shows—and what it does not

Primer location	What is presented	What the evidence establishes
Slide 17	An unspecified computer model gives two elevated radials at 36 inches about 4 Ω loss and therefore 90% efficiency.	No model file, solver, geometry, ground parameters, loss definition, currents or convergence data identifies how that result was obtained.
Slide 31	PERformer efficiency averages over 90%, with SWR below 1.10:1.	These are stated design/model results. No model package or matching efficiency measurement is supplied on the slide.
Slide 39	Model calculations are a starting point; field testing over real ground is required.	This is the correct methodological standard.
Slide 42	Six-band 4nec2 specifications with 88.1–93.9% efficiency, gain, front-to-back ratio, angle, bandwidth and impedance.	Conditional numerical outputs, pending disclosure of the full models, definitions and assumptions.
Slides 43–48	3D and polar radiation graphics for each band.	Pictures of computed output. Screenshots do not disclose the input deck, warnings, currents or convergence.
Slide 50	A PERformer “radiation sweep” comparing two radial orientations.	A model comparison, not a sweep made with a field-strength instrument.
Slide 51	RigExpert SWR sweeps on six bands.	Real evidence of input match over frequency. It does not establish the computed efficiency or pattern.
Slide 58	RigExpert traces beside a table explicitly headed “Computer Model Specifications”, including 85.5–87.7% efficiency and gain at 86 degrees.	The juxtaposition does not make the model columns measured quantities. The two evidence types remain separate.
Slides 95–96	“Structural efficiency”, transformer/unun efficiency, SWR and gain for other antennas.	The scope of every efficiency number must be defined. Structural, matching, network, radiation and complete-system efficiency are not interchangeable.
Slide 127	An array plot with 99.5% efficiency, 3.92 dBi gain, 8.54 dB front-to-back ratio and a 16-degree radiation angle.	High numerical resolution, but not automatically high physical accuracy.

The fair criticism is “not demonstrated by the material supplied in the primer or linked from the relevant slides”—not “the model must be wrong”. An undisclosed model may be right. It is simply not auditable.

The missing audit trail also prevents harmless differences from being resolved. For example, slide 42 lists 0.26 dBi gain and a −50.3 dB reflection coefficient on 15 metres, while slide 45 shows 0.31 dBi and −51.2 dB at the same nominal frequency. These may be different revisions or configurations. Without a file-to-figure revision map, the reader cannot tell.

Four words that must not be confused

Term	Question it answers
Reproducibility	Can another modeller run the same files with the same engine and obtain the same result?
Numerical verification	Was the chosen mathematical model solved stably, with geometry and segmentation errors under control?
Validation	Does the mathematical model predict the behaviour of the physical antenna closely enough for the stated purpose?
Measurement	What did calibrated instrumentation observe on the physical system, with stated conditions and uncertainty?

A model can be perfectly reproducible yet physically incomplete. It can be numerically converged while using the wrong soil. A measurement can be real yet irrelevant to the claimed quantity. Good engineering asks all four questions.

“Modelled in 4nec2” is not a complete method

4nec2 is a modelling environment and front end. It prepares input, invokes a calculation engine and turns numerical output into useful graphs. Its own support page lists separate NEC-2 engine executables, and its release history documents support for different engines and changes in preprocessing, auto-segmentation and gain display.

That is why a reproducible report must name both the front end and the actual solver:

• 4nec2 version
• NEC engine and exact build
• frequency or sweep definition
• thin-wire or extended-kernel setting
• automatic segmentation or preprocessing
• optimiser settings and objective function
• any average-gain correction or post-processing

This is not pedantry. The official 4nec2 history records, for example, a corrected normalised-gain fault in version 5.9.3 and an earlier correction involving NEC-4.2 ground preprocessing. Software version and calculation path can change the answer.

Checklist 1: publish the source, not only the screenshot

The minimum source package should contain:

1. the original 4nec2 authoring file, including symbols
2. the expanded NEC input deck actually sent to the engine
3. the complete raw output file, including warnings
4. all optimiser, sweep and post-processing settings
5. a README mapping each published figure to its source file and run
6. a dated archive or checksum so the analysed version is identifiable

The input deck discloses geometry, excitation, material loading and ground cards. The output file discloses segmentation, junctions, currents, power entries and warnings. A cropped polar plot discloses almost none of this.

Checklist 2: disclose the complete geometry

For every conductor, publish its endpoint coordinates, radius, material, physical length and segmentation. State the coordinate units and scaling. For an elevated-radial antenna, include:

• vertical dimensions, taper and loading
• radial count, length, azimuth, droop and height at both ends
• source type, exact source segment and feedpoint height
• all junctions and connections
• mast, feedline exterior, stakes, supports, control leads and nearby conductors—or an explicit statement that they were omitted

The NEC-2 User’s Guide devotes substantial attention to segment geometry, wire junctions, source placement and thin-wire restrictions because the code solves the geometry it receives, not the antenna the author intended to describe.

A photograph is also useful. It reveals physical conductors that may not appear in the model. In a sparse-radial installation, a coax shield, tripod or mast can become part of the return structure.

Checklist 3: ground is an input, not a background colour

“Real ground” is not one setting. A report must identify:

• the selected ground formulation
• conductivity in S/m and relative permittivity
• the frequency at which those values apply
• whether they were measured, taken from a preset or assumed
• homogeneous, two-medium, slope and terrain assumptions
• whether radials were explicit wires or an approximate radial-screen representation
• the exact wire clearance above the modelled interface

This is especially important for radials only 36 inches above soil. At 7.2 MHz that height is about 0.022 wavelength. The NEC-2 modelling guide says the fast reflection-coefficient ground is of limited accuracy close to ground, while the Sommerfeld/Norton treatment is the relevant wire-ground option for close interactions. It also states that a real ground stake cannot simply be represented as a wire crossing the interface.

Rudy Severns was unusually explicit about this. In his 2012 study of elevated ground systems, he named EZNEC Pro/4 version 5.0.45, the NEC4D engine, frequencies, conductor assumptions and soil parameters. He also warned that NEC-2 results for HF verticals close to ground require caution because of the ground-interaction limitation.

This does not make NEC-2 useless. It makes the engine, ground method and geometry part of the claim.

Checklist 4: account for every place watts can go

A high modelled efficiency can be obtained honestly from an idealised benchmark. It can also be obtained trivially by omitting loss mechanisms.

State whether the model includes:

• finite wire conductivity and skin-effect loss
• loading-coil series resistance or measured Q
• capacitor, trap and connection loss
• matching-network or transformer loss
• insulation and dielectric loss where relevant
• feedline common-mode radiation and loss
• complex choke impedance at each frequency
• soil dissipation according to the chosen ground solution

The raw NEC output can report source power, conductor loss, network loss, currents and radiated-field quantities, depending on the model and requests. Its output documentation is careful about what each category means. A modeller must be equally careful.

Do not rename all missing power “ground loss”. Do not call ideal-conductor efficiency installed-system efficiency. Do not combine a transformer’s separately estimated efficiency with a structural number without showing the complete power chain.

A useful reporting equation is:

Paccepted = Pradiated + Psoil + Pconductors + Pcomponents + Pnetwork + ...

Every term must be defined, calculated by a stated method and referred to the same port. An omitted loss path is not a demonstrated zero.

Checklist 5: demonstrate numerical convergence

A run that finishes is not necessarily a converged solution.

The NEC-2 guide recommends varying segment or patch density and observing whether the result stabilises. It explicitly says that large dependence on segmentation indicates non-convergence. A credible article should therefore publish at least three progressively refined runs while keeping physical geometry, source position and loads equivalent.

Quantity to track	What stability should show
Feedpoint impedance	Resistance and reactance approach stable values.
Absolute gain	The claimed value and angle change by much less than the claimed advantage.
Efficiency or average gain	The power metric remains stable under refinement and angular-grid changes.
Radial currents	Magnitude and phase division do not depend materially on segmentation.
Feedline and mast current	An unintended radiator is not appearing or disappearing numerically.

More segments are not automatically better. Extremely short or badly proportioned segments, poor junction geometry and source placement can create their own errors. Convergence means the physically equivalent model approaches a stable answer.

If the advertised difference is 0.1 dB but reasonable segmentation changes move the output by 0.15 dB, the model cannot support that comparison.

Checklist 6: publish complex currents

Patterns are produced by currents. The current audit is therefore one of the fastest ways to discover what the model is really simulating.

Publish magnitude and phase on:

• the radiator
• every radial at the same distance from the hub
• the coax outer conductor
• the mast and nearby conductors
• each loading or matching branch

Verify vector current continuity at junctions. Check radial balance across the band, not only at one favourable frequency. A symmetric drawing does not prove symmetric current after soil, nearby objects and the feedline are introduced.

Severns’s later work is especially instructive here: he measured and modelled individual radial currents because sparse elevated systems can become strongly unbalanced with modest asymmetry. A gain lobe in one direction may be pattern redistribution caused by an unintended current path, not an increase in total radiated power.

Checklist 7: define what the plot means

For every radiation plot, disclose:

• power gain or directive gain
• dBi, dBd or normalised relative level
• accepted-power basis
• polarisation component
• theta/phi or elevation/azimuth convention
• frequency and angular resolution
• ground-wave or space-wave option
• full 3D pattern or selected cut
• absolute peak and the value in the comparison direction

A pattern normalised to its own maximum sets the best direction to 0 dB regardless of absolute performance. It can compare shape. It cannot show that one antenna radiates more accepted power than another.

Likewise, a 3D renderer can auto-scale two patterns until both fill the window. Visual size and colour are presentation choices. The underlying absolute data must be available.

Checklist 8: test sensitivity, not only the best nominal case

Numerical convergence answers whether the solver stably solved one chosen problem. Sensitivity analysis asks whether that chosen problem represents a realistic range of installations.

Sweep plausible values for:

• soil conductivity and permittivity
• radial height, sag, length and azimuth
• radial-to-radial asymmetry
• feedline route and choke impedance
• wire diameter and conductivity
• component Q
• nearby conductive objects
• source position and segmentation

Report a range, not only the optimum. A displayed value such as 90.8% may be the exact output of one run, but the decimal is not credible physical precision if plausible soil or geometry changes move the answer by several percentage points.

A VNA sweep validates one quantity family

A calibrated analyser can measure the complex reflection coefficient at its calibration plane. From that, it can derive impedance, return loss and SWR.

It cannot separate radiated power from soil loss, wire loss, transformer loss or common-mode radiation merely by observing a good match.

Slide 58 makes the category boundary unusually visible. Its RigExpert images are physical measurements. The adjacent gain and efficiency columns are explicitly labelled computer-model specifications. The small box converting SWR into a percentage describes mismatch efficiency:

ηmismatch = 1 - |Γ|2

An SWR of 1.10:1 means that about 99.8% of the incident power is accepted at the stated reference plane. It does not mean that 99.8% is radiated. Once power enters the antenna, it may still be dissipated in soil, conductors or components.

What comparative field validation should look like

A claimed gain or efficiency difference needs a controlled A/B experiment designed for that quantity:

1. Define the exact physical antenna, ground and environment.
2. Change only the intended variable.
3. Calibrate impedance measurements at the antenna feedpoint.
4. Use equal accepted power, or measure and correct the accepted-power difference.
5. Use a stable transmitter and fixed receiving antenna at a documented distance, height, azimuth and polarisation.
6. Measure the noise floor, direct coupling and system drift.
7. Use A–B–A or randomised switching order.
8. Measure radial currents and coax common-mode current.
9. Repeat at several azimuths if pattern skew is possible.
10. Repeat on different days and publish the raw readings with uncertainty.

Rudy Severns’s ground-system measurement method is a valuable example. He documented calibrated VNA transmission comparisons, receiving geometry, feedline isolation, calibration checks, stray-coupling tests and current measurements. In his four-elevated-versus-64-surface-radial experiment, he measured both relative field and feedpoint impedance, examined current division and repeated the entire sequence on three different days.

He also kept the conclusion within the scope of the experiment. Relative A/B field strength in a defined direction was not relabelled as an absolute efficiency measurement.

The minimum package for a quantitative NEC claim

Package item	Why it is needed
Original and expanded model files	Recreates the exact geometry, variables and solver input.
Front end, engine, versions and options	Identifies the actual numerical method and preprocessing.
Ground and material table	Defines the medium and loss assumptions.
Raw output and warnings	Exposes junctions, currents, power entries and solver diagnostics.
Convergence table	Shows that the result is not a segmentation artefact.
Complex-current data	Reveals imbalance and unintended radiators.
Pattern definitions and raw values	Distinguishes absolute gain from normalised shape.
Sensitivity runs	Shows robustness to realistic soil and construction variation.
Measurement protocol, raw data and uncertainty	Tests whether the model predicts the physical antenna.

An evidence ladder

Evidence	Defensible conclusion
Rendered NEC screenshot	“This is what one run displayed.”
Source package rerun independently	“The numerical result is reproducible.”
Convergence, current and power audits	“The chosen numerical model is internally credible.”
Calibrated VNA agreement	“The model predicts input impedance reasonably here.”
Controlled comparative field/current measurements	“The model predicts the corresponding physical observables within stated uncertainty.”
Independent replication	“The result does not depend on one modeller, instrument chain or installation.”

Contacts, POTA activations and user testimonials belong elsewhere on this ladder. They demonstrate that an antenna is usable and enjoyable. They do not quantify radiation efficiency or absolute gain.

Takeaways you can trust

• NEC is a powerful engineering tool; a plot is still a conditional prediction.
• Slide 39 of the primer correctly says that modelling is only a starting point.
• The primer’s analyser sweeps measure input match, not efficiency, gain or radiation pattern.
• “4nec2” does not identify the solver, version, ground treatment or preprocessing.
• Source and raw output files are essential for reproducibility.
• Close-to-ground radial models are particularly sensitive to ground formulation and soil inputs.
• Every physical loss path must be represented or explicitly excluded.
• Segmentation refinement must show that the result has converged.
• Complex current data can expose imbalance, feedline radiation and unintended conductors.
• Absolute gain must not be confused with a pattern normalised to its own peak.
• Display precision is not measurement accuracy.
• Controlled comparative field measurements are needed to validate modelled field claims.

In Summary

Greg’s primer contains useful modelling and real analyser work. It also states the correct principle that field testing over real ground must follow calculation.

But the displayed field evidence does not close the loop on the headline quantities. SWR confirms match. It does not measure efficiency. A successful contact confirms operation. It does not measure gain. A 4nec2 graphic shows the answer produced by one numerical model. It does not reveal whether the physical antenna produced the same pattern.

The remedy is not to abandon modelling. It is to make the modelling reproducible and the validation proportional to the claim: publish the files, identify the solver, define the ground, include losses, demonstrate convergence, disclose currents, define every plotted quantity and compare the prediction with controlled field data.

Want more technical RF content? Subscribe for new deep dives and lab notes.

Have a question or field observation? Contact RF.Guru.