Antenna types, and how to choose one
Dipoles, monopoles, PCB inverted-F, chip, patch, Yagi and helical — what each is good for, what it costs you, and the trade nobody escapes.
The trade nobody escapes
Antenna design is governed by a relationship that is close to a physical law: size, bandwidth and efficiency. You can trade between them, but you cannot have all three. An antenna much smaller than a wavelength will be either narrowband or inefficient, and usually both.
This is why the antenna in a product is so often the last thing designed and the first thing to disappoint. Everything else on the board obeys Moore's law; the antenna obeys the wavelength, and the wavelength has not moved since the band was allocated. At 868 MHz a quarter wave is about82 mm, and no amount of clever silicon changes that.
So the sensible way to choose is not "which antenna is best" but "which compromise can I afford". What follows is what each type actually costs you.
How big is a resonant element?
Most antenna types are built around a resonant element of a half or a quarter wavelength. Those lengths are worth knowing before you commit to an enclosure size, because they are frequently the constraint.
| Band | Free-space λ | Half wave (λ/2) | Quarter wave (λ/4) |
|---|---|---|---|
| 433 MHz ISM | 690.9 mm | 328.2 mm | 164.1 mm |
| 868 MHz ISM (EU) | 345.4 mm | 164.1 mm | 82.03 mm |
| 915 MHz ISM (US) | 327.6 mm | 155.6 mm | 77.81 mm |
| 1.575 GHz GPS L1 | 190.3 mm | 90.39 mm | 45.19 mm |
| 2.45 GHz ISM | 122.4 mm | 58.12 mm | 29.06 mm |
| 5.8 GHz ISM | 51.69 mm | 24.55 mm | 12.28 mm |
Element lengths include the conventional 5% shortening for end effect — a wire resonates slightly short of a free-space half wavelength. Computed at build time; the wavelength calculatorhandles other frequencies and velocity factors.
A useful shortcut: λ in metres ≈ 300 / frequency in MHz. That is high by 0.07% because it rounds the speed of light, which is far inside any tolerance you will meet in practice.
The half-wave dipole
Two collinear quarter-wave elements fed in the middle. It is the reference antenna, and it is also genuinely good: 2.15 dBi, naturally close to 73 Ω so it matches 50 Ω coax without much help, and reasonably broadband.
Its pattern is a doughnut — omnidirectional around the element, with deep nulls off each end. That null is not a subtlety; it is a genuine hole. Point the tip of a vertical whip at your receiver and the link fails while the same distance sideways works fine. It is the commonest cause of "it works over there but not here" in indoor deployments.
Use it when you have room. Its disadvantage is simply size: at 868 MHz a dipole is about 164.1 mm end to end.
The quarter-wave monopole
Half a dipole, worked against a ground plane. The plane acts as an electrical mirror, so the missing half is supplied by the reflection. That halves the physical size, which is why most whips and wire antennas are monopoles.
The ground plane is half the antenna. This is the point that catches people out, and it catches them out repeatedly. Over a perfect, infinite ground the theoretical gain is about 5.16 dBi — better than a dipole, because all the power goes into a hemisphere rather than a sphere. Over a small ground plane on a real PCB it is nothing like that: the pattern tilts, the feed impedance moves, and the realised gain drops to around 2 dBi or worse.
A rough rule is that the plane wants to be at least a quarter wave in every direction from the feed — at 868 MHz that is a82 mm radius, which is a substantial fraction of many products. If you cannot give it that, expect to lose several dB and to have to tune the antenna on the finished assembly rather than on the bench.
Loaded whips and "rubber ducks"
A shortened monopole with inductance added — a coil at the base or the element wound as a helix — to bring a physically short radiator back to resonance.
It works, and it is not free. The loading coil has resistance, the radiation resistance of a short element is low, and the ratio between those two is the efficiency. A stubby antenna quoted at "2 dBi" frequently measures below 0 dBi in a product. The bandwidth narrows too, so it is more sensitive to being detuned by a hand or an enclosure.
Choose it when the alternative is no antenna at all, and budget for the loss rather than believing the label.
PCB antennas: the inverted-F
A quarter-wave element folded over the board, shorted to ground at one point and fed at another. It is a monopole that has been bent to fit and given an impedance adjustment for free — the position of the feed tap sets the match, which is why the shape looks arbitrary until you know what it is doing.
Its great virtue is cost: it is copper on a board you were making anyway, so it adds nothing to the bill of materials. In volume that is decisive.
Its vice is that it is not a component. Its performance is a property of your board, your ground plane, your enclosure and everything within a few millimetres of it. You cannot buy a known-good IFA; you copy a reference layout, keep the specified keep-out clear, and then tune it on the real assembly with a matching network you left space for. A design that skips that last step usually ships several dB down without anyone noticing until the field reports arrive.
Ceramic chip antennas
A few millimetres of high-permittivity ceramic that shrinks the resonant structure into something surface-mountable. When the product is genuinely tiny — a wearable, a tag — this may be the only option.
The size-bandwidth-efficiency trade applies with full force. Efficiency is often under 50%, which is more than 3 dB gone. It needs a matching network, a specified keep-out, and usually a ground plane of a particular size, all of which are in the datasheet and all of which are routinely ignored.
The honest framing is that a chip antenna buys you space at a known cost in range. That is sometimes exactly the right trade — just make it deliberately rather than discovering it later.
Patches, Yagis and helicals
These are the directional options, and the reason to reach for one is usually not raw gain but rejection — pointing away from interference is often worth more than pointing towards the signal.
A microstrip patch is a resonant copper rectangle over a ground plane, radiating from one face only. Around 7 dBi, flat, cheap to make, and narrowband. It is what is inside almost every GPS antenna.
A Yagi-Uda is a driven dipole with a reflector behind and directors in front, each parasitic element re-radiating with a phase shift that reinforces one direction. Ten to fifteen dBi is routine. It must be aimed, and its beamwidth shrinks as gain rises, so a high-gain Yagi on a mast that moves in the wind is a link that comes and goes.
An axial-mode helical radiates circular polarisation, which is why it suits satellites and anything that tumbles. Note the fixed penalty: circular into linear costs 3 dB, always, in either direction. That is not a fault, it is geometry — half the power is in the orthogonal component.
Choosing, in one table
| Type | Gain | Size | Good for | Watch out for |
|---|---|---|---|---|
| Half-wave dipole | 2.15 dBi | λ/2 | The reference antenna, and a genuinely good one when you have room | The nulls are real: point the end at your receiver and the link dies |
| Quarter-wave monopole | 2 dBi | λ/4 + ground plane | Whips and wire antennas on a board or chassis | The ground plane is half the antenna — a small one detunes it and costs gain |
| Loaded whip (rubber duck) | 0 dBi | well under λ/4 | Handhelds and anywhere a full-size element will not fit | Often quoted at 2–3 dBi and rarely measures it; shortening costs efficiency |
| PCB inverted-F (IFA) | 0 dBi | ≈ λ/4 folded | Volume products — it costs nothing but copper | Performance is set by the board and enclosure; it must be tuned in situ |
| Ceramic chip antenna | -2 dBi | a few mm | Very small products where nothing else fits | Needs a matching network and a keep-out area; efficiency is often under 50% |
| Microstrip patch | 7 dBi | ≈ λ/2 square | GPS, and anywhere you want gain pointing one way from a flat surface | Narrow bandwidth, and it only radiates from one face |
| Yagi-Uda | 12 dBi | λ/2 elements over 1–3 λ | Fixed point-to-point links where you can aim it | It must actually be aimed — and it hears interference off the back too |
| Axial-mode helical | 11 dBi | several λ long | Satellite and anything where the far end tumbles | Circular into linear costs a fixed 3 dB, always |
The decisions that matter more than the type
Having chosen a type, three things will affect your range more than that choice did.
Polarisation. Both ends should agree. A vertical antenna talking to a horizontal one loses 20 dB or more in theory, and enough in practice to ruin a link. This is a common failure when one end is a fixed gateway and the other is a device that gets put down on its side.
What is near it. Metal within a fraction of a wavelength detunes an antenna and reshapes its pattern. Batteries, screens, shielding cans, a hand, a wet wall — all of them count. This is why an antenna must be tuned in the product, not on a bench.
Height. Frequently free, and frequently worth more than any antenna upgrade. Getting above the obstruction changes the problem instead of fighting it — see planning a radio link for where that sits in the budget.
And before committing to a high-gain antenna, check theregional limits. Regulators cap radiated power, not transmitter power, so antenna gain counts against your allowance — a compliant radio with a better antenna can be a non-compliant product.