The Limits of NEC: Model the System You Actually Built
The Limits of NEC: Model the System You Actually Built
NEC can calculate currents, impedance and fields for a declared electromagnetic problem. Its answer belongs to that problem—not automatically to the wire, balcony, boat or feed system outside your model.
RF.Guru working definition: Common-mode current is the non-cancelling phasor-sum current in a specified set of conductors, evaluated at a defined cross-section and using a declared current-direction convention. In the intended differential transmission-line mode, the outgoing and return currents are equal and opposite, so their phasor sum is zero. When they do not cancel, the remaining current must close through another reference or return path—such as the outside of a coax shield, a mast, equipment chassis, station wiring, nearby structures, earth, the operator, or distributed coupling through the environment.
This broader working definition is especially useful in practical antenna systems. On transmit, non-cancelling current on the outside of the coax can make the feedline and connected structures part of the radiating antenna system unless that path is intentional, clearly defined and properly controlled—for example by providing the required return path and placing a suitable common-mode choke at the correct boundary.
My warning is aimed at false certainty, not at modelling. A beautiful far-field plot is useful only when the NEC engine, geometry, materials, ground, feed arrangement and requested output correspond closely enough to the engineering question.
NEC does not “see” or “miss” an installation by intuition. It solves the conductors, sources, loads, networks, ground model and material approximations that were actually entered. An omitted coax exterior, mast, return conductor or nearby structure is a modelling omission; a thin-wire or ground-formulation limit is a solver boundary; an uncertain dimension or material value is an input uncertainty. Keep those three failure modes separate.
Start by Naming the Engine
“NEC” is a family, not one timeless calculator. Classic NEC-2, licensed NEC-4 and NEC-5 use related but different formulations and capabilities. A program such as 4nec2 is also a modelling and display environment that can call different calculation engines. The engine name and version belong in every saved result.
The original NEC-2 theory describes a Method-of-Moments solution for currents on conducting wires and surfaces. Later versions change important parts of the numerical treatment. Lawrence Livermore’s NEC-5 documentation, for example, lists wire and conducting-surface models, homogeneous-ground capability, sources, loads, networks and transmission lines; its validation work compares versions precisely because agreement cannot be assumed for every geometry.
That distinction matters when someone says “NEC cannot model this.” The requested feature may be absent from one engine, approximated differently in another, or available only when the user constructs the missing current path explicitly.
Solver Limits, Model Limits and Input Limits
| Boundary | Typical question | Useful test |
|---|---|---|
| Numerical formulation | Is this wire radius, junction, surface, ground interaction or electrically small feature within the selected engine’s formulation? | Read that engine’s manual and validation cases; compare with another suitable formulation where possible |
| Discretised model | Are segments, sources, loads and connections representing the intended physical structure without introducing a numerical artefact? | Refine and redistribute segmentation while holding the physical geometry and reference planes fixed |
| Physical abstraction | Were the feedline exterior, return path, mast, rail, rigging, losses and important nearby objects included? | Add one credible feature at a time and run a sensitivity study |
| Input uncertainty | Are conductivity, permittivity, dimensions, bonds, soil state and load values actually known? | Sweep credible bounds instead of publishing one guessed decimal |
| Output interpretation | Is the reported quantity directivity, gain with modelled loss, input impedance, field at a declared distance, or something else? | Record normalization, power reference, coordinate system, ground choice and observation region |
A solver can be numerically correct for a physically incomplete model. It can also be fed a detailed geometry that violates its own modelling guidelines. More segments and more visible detail do not automatically produce a more accurate answer.
Thin Wires Still Need Disciplined Segmentation
The NEC-2 user guide defines wires as strings of straight segments and warns that segment length, wire radius, junctions, abrupt radius changes, sources and loads affect accuracy. Its familiar segment-length guidance is a starting point, not a certificate. Critical regions may need finer resolution, while extremely short or badly proportioned segments can create their own numerical trouble.
Run a convergence study around the quantities that drive the decision. Increase or redistribute segmentation without changing the intended conductor, source position or load placement. Input impedance, current maxima, gain and null depth should settle to engineering-relevant tolerances. A stable SWR alone does not prove that a deep null or current at a junction has converged.
Use geometry checks as well. Confirm every wire end, junction, radius, source segment and load segment. Plot the current distribution and look for discontinuities that belong to the mesh rather than the antenna. If two engines or kernels disagree, investigate the geometry and known formulation limits before choosing the prettier result.
Ground Is a Model With Parameters
Classic NEC ground choices include ideal and approximate treatments of a conducting half-space; the available formulations depend on the engine. They need frequency-appropriate conductivity and relative permittivity. “Average ground” is not a site measurement, and one homogeneous half-space is not a full description of dry soil over wet clay, concrete with rebar, a roof, a shoreline or seawater over a seabed.
This does not make every ground result useless. It defines the claim. A height sweep over a declared homogeneous ground can expose a design trend. It does not prove the exact elevation pattern at a particular garden or harbour. Run credible ground cases and compare which conclusions survive.
Near-ground wires and structures crossing an interface deserve particular care because NEC versions do not treat them identically. Use the manual for the selected engine rather than transferring a rule from another NEC version or from a graphical front end.
A Boat or Balcony Is a Scenario, Not a Single Geometry
The boat and balcony installations I describe are among the hardest kinds of everyday HF problem. A mast, stay, rail, engine, bonding conductor, coax exterior, building steel, rebar, gutter and station cable can form part of the RF current network. Some connections are intentional; others are capacitive, intermittent, corroded or simply unknown.
NEC can calculate coupling among conductors that are represented within the selected engine’s limits. It cannot infer whether a hidden bond exists. When that state is unknown, build plausible cases: bonded and floating rail, included and omitted mast, several feedline routes, bounded contact impedance and bounded ground properties. If the decision changes across credible cases, the model has found sensitivity—not certainty.
The same caution applies at VHF, UHF and microwave frequencies. Shorter physical structures can be easier to measure and reproduce, but tolerances, connectors, housings, dielectric details and nearby objects become electrically larger. Frequency alone does not make the environment simpler.
Non-Metallic Objects Need the Right Electromagnetic Model
Wet wood, foliage, concrete, fibreglass, water tanks, ropes, walls and coatings can change fields through permittivity, conductivity and loss. Classic wire-oriented NEC models do not natively turn an arbitrary three-dimensional dielectric body into a faithful volume model merely because its outline was drawn.
Sometimes a dielectric is weak enough or far enough away to omit within the required uncertainty. Sometimes an equivalent load, insulated-wire option, surface approximation or another full-wave method is appropriate. The choice must follow electrical size, field strength at the object, material data and the accuracy needed—not the slogan that only metal matters.
The Feed System Is Missing Only When You Leave It Out
An ideal source across two wire segments contains no automatic coax exterior, transformer, choke, tuner chassis or station wiring. If those branches are absent, the calculated model cannot carry common-mode current on them. That is a property of the input deck, not proof that NEC is blind to common mode.
Model the conductors that may carry unintended current when the engine and geometry permit it. Represent the outside of a coaxial feed as a conductor or validated equivalent, include the intended return branch, and give mast and station bonds their declared electrical connections. Real transformers, traps and matching networks can be represented only to the quality of their measured or justified equivalent circuits.
Then measure the installed system. A current probe along the feedline exterior, calibrated impedance at a named reference plane and reversible choke or routing changes can reveal a missing current path. If the model changes when the feedline is added and the installation changes in the same direction when that path is controlled, the comparison has engineering meaning.
Loss and Gain Need a Complete Power Boundary
NEC can include conductor loss, lumped loads, networks and ground loss within the capabilities of the chosen engine. It does not know the loss of a corroded clamp, wet trap, ferrite assembly or tuner unless that behaviour is represented. Perfect conductors and ideal loads produce results for perfect conductors and ideal loads.
Label every plotted quantity. Directivity describes angular concentration independent of loss. Gain includes the losses represented in the model. Neither automatically includes an unmodelled matching network, feedline or transmitter mismatch. “6.1 dBi” therefore means only what the saved model, power normalization and ground convention say it means.
Delete universal error budgets such as “real antennas are always 2–3 dB worse.” A simple, well-characterised structure can agree closely with calculation; an uncertain installation can differ by much more, or differ mainly in impedance, null direction or current path rather than peak gain. The uncertainty comes from the case.
What I Use Modelling For
I also write lightweight Python/NumPy Method-of-Moments code. I use it deliberately as a simplified free-space baseline: expose the assumptions, inspect current and pattern trends, and compare clean cases with a NEC engine before adding environmental complexity.
That is not a claim that two models validate an antenna in the field. Similar answers from related approximations can share the same omission. The value is diagnostic: a small model makes the current path visible, while an independent formulation, a sensitivity sweep and a measurement each test a different part of the argument.
A Model-to-Mast Workflow
- Write the engineering question. Name the frequency set, impedance plane, pattern region, polarization, loss boundary and decision tolerance.
- Record the engine. Save the NEC version, calculation engine, kernel and ground option—not only the front-end application.
- Build the minimum complete current path. Include the radiator, intentional return, feedline exterior and nearby conductors that can change the result.
- Enter bounded materials. Use measured or defensible conductor, load and ground data; sweep uncertain values.
- Check geometry and convergence. Inspect junctions, sources, loads and current, then repeat with changed segmentation.
- Test model sensitivity. Change one uncertain bond, route, height, loss or nearby object at a time.
- Cross-check important cases. Use an analytical limit, another suitable solver or a known validation problem where practical.
- Measure at matching planes. Compare calibrated R + jX, feedline-exterior current or field data using the same geometry and normalization.
- Restore and repeat. Use A/B/A changes so weather, propagation or a disturbed cable route does not become the apparent model result.
Primary NEC Sources
- Burke and Poggio — NEC-2, Part I: Program Description—Theory: the integral equations, current expansion and Method-of-Moments foundation of NEC-2.
- Burke and Poggio — NEC-2, Part III: User’s Guide: wire segmentation, radius, junction, source, ground and model-construction guidance.
- Burke, Miller and Poggio — The Numerical Electromagnetics Code: A Brief History: the original authors’ account of the NEC family and changing capabilities.
- Lawrence Livermore National Laboratory — NEC 5: current official capability, version and maintenance information for the later NEC engine.
- Arie Voors — 4nec2: the author’s description of 4nec2 as a NEC modelling, checking, sweeping, optimization and display environment.
Practical Conclusion
Do not worship the plot, and do not blame the solver for a current path you never drew. NEC is strongest when the problem is declared, the engine is named, the mesh converges, the uncertain inputs are swept and the result is compared with something independent.
For a messy HF installation, the useful answer is often a range or a sensitivity: which bond matters, which feedline route changes the pattern, which loss dominates, or which conclusion survives every credible ground case. That is not a weaker result. It is honest engineering.
Mini-FAQ
- Is NEC only useful for ideal antennas? No. It can represent complex conductor systems, sources, loads, networks and supported ground models. Accuracy depends on the engine, model, inputs and validation.
- Can NEC show feedline common-mode current? Yes, when the relevant exterior conductor and return paths are represented appropriately. An ideal feedpoint source alone contains no automatic coax exterior.
- Does adding more segments always improve accuracy? No. Segmentation must respect the engine’s guidelines and the physical geometry. Use a convergence study around each decision-driving output.
- Can I trust an exact gain number from NEC? Only within the declared model, normalization and represented losses. Installation uncertainty requires sensitivity analysis and measurement.
- Are boats and balconies impossible to model? No. Model plausible electrical states and learn which bonds, routes and nearby structures matter; do not disguise unknown inputs as one exact geometry.
- What is the best validation step? Match the test to the claim: calibrated impedance, conductor current, field or pattern data at declared reference planes, with repeatable geometry and uncertainty.