Measurements on antennas are difficult. The behaviour of an antenna is best seen by monitoring the reflection coefficient with a network analyser over a band of frequencies, and for convenience a frequency about 1 GHz is appropriate. At 1 GHz a wavelength is 30 cms; the antenna is a reasonable size and it is possible to investigate the effects of adjacent objects, and different feed lengths, without too much difficult physical manipulation. The results may safely be transferred to other frequency bands by thought and analogy.
When this is done, one rapidly appreciates that an antenna can not be considered as a closed, isolated component having well-defined properties. Nearly every electronic measurement on an antenna is grossly affected by its environment and physical mounting. One might well ask the question, "What is an antenna?", or equivalently, "Where does the antenna stop and the outside world begin?". A sensible answer to this question is to consider all objects inside the near field as contributing to the radiation.
A helpful example is a Yagi-Uda antenna. We might regard this as a simple dipole with lots of resonant rods placed in the near field. But if we just consider the properties of the driven "antenna", namely the driven dipole, we know we will be grossly in error in assessing the performance of the installation. So why should we stop considering the effects of metallic structures at the end of the boom? We should add in the scattering from the mast, guys, feed (outer coaxial shields can carry induced current) and even dielectric objects (like the chimney stack or adjacent building) in the near field.
Many people now have access to software which accurately simulates antenna behaviour. To do this it is necessary to construct a model. The process of "modelling" is critical to this enterprise as the simulation has limitations of accuracy depending on the kind of model chosen. In itself, the software is essentially accurate and useful. However, the results it returns, for simulation of real antennas, depends critically on what is built into the model. It is not usually possible, in the NEC2 and miniNEC and NEC4 software, to add in all the local effects which will affect the results. This is not just because it is too difficult; there are difficulties in principle, knowing the correct dielectric and conductivity parameters to put in for a real-world installation. Details of the feed arrangement are also difficult to get right. So it is often difficult to know if the results from the simulation of the model represent the real behaviour of the antenna it was intended to investigate. The process of running the software always returns a result, and the internal checks on validity, while possible, are subtle. Belief in the results often dissolves into a matter of opinion or faith. This can be the subject of strongly-held views, which can only be resolved by recourse to measurements.
Thus, simulation should be regarded (taking the most cautious view) as merely a rough guide to an antenna's behaviour in a real installation. Any modelling process needs careful validation by measurements. One is then presented with the choice of which to believe, if there is disagreement.
I have seen people worry about 1/10 dB in gain in a simulation. This is probably unsound, and one wonders how many hundreds of hours people spend (no doubt happily) in this kind of activity.
An anonymous committed modeling expert tempers the remarks above with the following comment.
Depending upon the conditions of use and application, NEC can be quite accurate. Under some conditions of application, accuracy deviation of dimensions has run to well under 1% in some design projects. In others, accuracy of construction guidance can be considerably off. The real question is this: what are the conditions for each kind of case? However, that question requires an application-by-application analysis, not a wholesale posture toward modeling.
Yes, but application-by-application analysis usually requires experimental measurements to be made to validate the models.
Considering this italic comment, what can be said with confidence is that an unvalidated antenna model is not necessarily a good guide to how a practical antenna may behave.
One of the thrusts of the complex-systems research reported elsewhere on these pages (see complexsim.html, for example) is that simulations are only of use if one knows they are going to be accurate before one validates them against measurements. Most people using antenna models are making extrapolations from validated simulations to applications that they believe to be similar. It is the thesis presented here that this method of proceeding is unsafe and unwise in many cases. [http://www.ee.surrey.ac.uk/Personal/D.Jefferies/antennas.html]