Menu

The Antenna That Actually Radiates

The Antenna That Actually Radiates: 2.4 GHz PCB Antenna Design

Notes · Radio and wirelessNovember 19, 202623 min readSenova Systems Research

A 2.4 GHz antenna is 31 mm of copper, folded to fit, and it only radiates if the ground plane is the right size, the keep-out is empty on every layer, and the enclosure is part of the design from day one. A battery a millimetre away costs 6–10 dB, and 6 dB halves the range.

Picture the Wi-Fi product on the bench. The radio connects, the link forms, the sensor reads, and the range is fine, across the room, down the corridor, out to the car park. Then the board goes into the enclosure, the lid goes on, and the range collapses to a few metres.

Or the board goes into the chamber, a room lined with absorber, with an antenna three metres away, and the lab fails it, and the launch date becomes the retest date. Nobody changed the radio. Nobody changed the firmware.

The antenna is the same copper it was on the bench. What changed is everything around it.

That is the antenna's secret, and the reason it is the hardest part of a wireless product to get right by accident: the antenna is not the copper trace. The antenna is the copper trace plus the ground plane it radiates against, plus the empty space around it, plus the enclosure it sits inside. Change any one of those and you have changed the antenna, and the board that worked on the bench is a different antenna in the box.

This article is about the one number behind every 2.4 GHz antenna, the ground plane that is the other half of it, the keep-out that has to be empty on every layer, the three options for getting an antenna onto a board, and the enclosure that decides the link budget before any schematic exists.

The one number: 31 mm

Every 2.4 GHz antenna starts from one number, and it is a physics number, not a design choice. The wavelength is the speed of light divided by the frequency: 299,792,458 metres per second divided by 2.4 billion hertz is 124.9 mm, about 125 mm in free space. A quarter of that is 31.2 mm.

That is the magic number: most PCB antennas are quarter-wave designs, and every shape you will meet in a reference design, the inverted-F, the meander, the monopole, is that 31 mm of trace, folded to fit a smaller footprint. The shape is packaging. The length is the physics.

The actual radiating arm is shorter than 31 mm, and the reasons are worth knowing because they are the reasons a copied length does not work on a different board. The standard inverted-F reference design from Texas Instruments (SWRU120D) is 25.7 × 7.5 mm overall, with a radiating arm of 25.58 mm, and the guide says to make an exact copy of the dimensions, because small changes have a large impact on performance.

Andwin's guide gives the same range: the typical IFA for 2.4 GHz measures 25–30 mm long and 7–8 mm wide, occupying roughly 200–250 mm² of board area, with a gain of 2.0–2.5 dBi and a bandwidth of 5–8%. The shortening comes from two things: the shorting stub, the vertical ground connection that gives the inverted-F its name, which reduces the length by 10–15%, and the dielectric, which loads the antenna and shortens it further.

The compact meandered version is shorter still: TI's small-size IFA (AN043) needs no more than 15.2 × 5.7 mm of space, with a VSWR under 2 across the 2.4 GHz ISM band, and the pipeline's figure of "roughly 12–16 mm long" is that compact meandered version, not the standard IFA. The article uses both numbers, and says which is which.

One number that is easy to confuse with the 31 mm is the microstrip guided wavelength. A wave buried in FR-4, with a dielectric constant of 4.4, has a wavelength of 59.6 mm, 125 divided by the square root of 4.4, and a quarter-wave of 14.9 mm. A wave on a microstrip line, where the field lives partly in the board and partly in the air, has a guided wavelength of about 68.6 mm and a quarter-wave of about 17 mm.

Those are the numbers for the feed line, not the radiating element. A PCB antenna radiates mostly into air, so its effective wavelength is roughly 60–70% of the free-space value, 75–87 mm, and a quarter of that is 19–22 mm, not 31 mm. The article states the 31 mm as the free-space quarter-wave, and explains that the IFA's actual arm is 25–30 mm because the shorting stub and the dielectric shorten it.

The two wavelengths are different quantities, and the feed line is dimensioned in guided wavelengths while the antenna is dimensioned in free-space wavelengths, which is why a quarter-wave stub cut to the free-space quarter wavelength resonates far below the intended frequency.

The ground plane is the other half

Here is the concept that trips up most first-time antenna designers: the ground plane is part of the antenna. It is not just a return path for the digital signals. The antenna radiates against the ground plane. Without it, or with the wrong size, the antenna does not work as designed. The ground plane is the counterpoise, the other half of the dipole. The element pushes current, and the ground plane carries the return; together they radiate.

The numbers say how much it matters. Inadequate ground-plane size reduces radiation efficiency by 40–60% and shifts the resonant frequency by 50–150 MHz (Andwin). The ground plane should extend at least a quarter wavelength, 30–35 mm at 2.4 GHz on FR-4, in all directions from the antenna feed point.

(One Andwin line says "at least 80 mm in all directions", that contradicts its own quarter-wave figure, and the article uses the quarter-wave figure, 30–35 mm, and notes the discrepancy.) Shrink the ground plane or crowd it and you have changed the antenna, not merely its surroundings, because the ground plane is the antenna, and the trace is only half of it.

This is also why the ground plane is the first thing to check when a product works on the bench and dies in the enclosure. On the bench, the board sits on a wooden desk with nothing around it, and the ground plane is the whole board. In the enclosure, the battery sits over the ground plane, the shield can sits beside it, and the metal screws go through it, and every one of those changes the counterpoise. The antenna did not move. Its other half did.

The keep-out: empty on every layer

The keep-out zone is a copper-free area surrounding the radiating element where no ground planes, no power planes, and no signal traces should exist on any PCB layer. This clearance zone typically extends 5–8 mm beyond the antenna perimeter for 2.4 GHz designs (Andwin), and the pipeline's figure of "5–10 mm of nothing around it" is the same zone, stated as a range. The most common range-killer the guides see is a ground pour, or a power pour, or signal routing, sitting under or beside a PCB trace antenna.

Copper directly beneath the antenna element creates an image current that opposes the antenna's radiation, killing efficiency: imagine holding a mirror an inch from a lightbulb, and the reflection cancels out much of the light going in that direction. Put copper in the keep-out and you are loading the antenna with capacitance it was not designed for, pulling its resonance off-band and dumping energy into the plane as heat instead of into the air.

Three numbers that look like the same rule but are not: the copper keep-out on the board is 5–8 mm beyond the antenna perimeter; the clear board area a PCB trace antenna demands is 15–25 mm of ground-plane clearance (Hubble), the area where you cannot place components, pour copper, or run traces; and the clearance Espressif recommends between the antenna and any metallic housing element inside the enclosure is at least 15 mm in all directions (via NextPCB, Elroot, and QFPCB).

The module's own antenna area on the carrier board needs no copper of any kind within 2 mm around the antenna section (SiliconWit). The article distinguishes all four, because the engineer who treats the 15 mm housing clearance as the board keep-out will leave the board's copper in place and wonder why the range collapsed, and the engineer who treats the 5–8 mm keep-out as the housing clearance will put the battery 6 mm from the antenna and lose 6–10 dB.

The rule the vendors all give, in the same words, is to copy the keep-out from the vendor's reference layout instead of estimating it. TI's guide says it plainly: make an exact copy of the antenna dimensions to obtain optimum performance, and the easiest approach is to import the antenna layout from a Gerber or DXF file. Every BLE chip vendor gives you a reference antenna layout, exact trace dimensions, keep-out zones, ground-plane rules, all clearly specified.

The reference layout is not a suggestion. It is the antenna, already tuned, and the tuning only holds if the dimensions hold.

The three options

There are three practical options for a 2.4 GHz design, and they trade cost against board area, range consistency, tuning effort, and how painful certification will be. The table is a decision, not a menu.

Antenna optionRelative costRange / consistencyBoard area neededTuning effortCertification ease
PCB trace (IFA / meandered monopole)Lowest, free in copper, no partGood if tuned; very sensitive to nearby copper, enclosure, and groundLargest, needs a clear keep-out at the board edgeHighest, must be tuned and measured for your exact stackup and enclosureHardest if self-designed; easiest if you reuse a module's pre-certified trace
Ceramic chip antennaLow part cost + a matching networkGood in a small footprint; less efficient than a well-tuned traceSmall footprint, but still needs a ground keep-out around itModerate, vendor gives a reference match, you still tune for your boardModerate, vendor app notes and reference layout help
U.FL / external (whip, dipole, IPEX)Highest, connector + antenna + assemblyMost consistent and usually best range; antenna lives outside the PCBSmallest on-board (just the connector and 50 Ω feed)Lowest, antenna is pre-tuned by its makerEasiest, pre-certified antennas simplify FCC/CE

The PCB trace is free in unit cost, copper you were going to etch anyway, but it pays rent in area: an inverted-F at 2.4 GHz is roughly 12–16 mm long in its compact meandered form (25–30 mm in the standard form) and wants a clear region around it, typically 5–10 mm of nothing on the same layer and a matching cut-out in every other copper layer beneath. Its gain, when properly tuned, is 0 to +2 dBi (Hubble) or 2.0–2.5 dBi (Andwin), and its radiation efficiency is typically 40–60% (Dolph Microwave). It is the default if you have board space and want zero antenna BOM cost, and the lowest-risk way to ship one is to use a module whose printed antenna is already part of a pre-certified design, so you inherit its tuning and its FCC/CE modular approval.

The ceramic chip is a small ceramic or multilayer SMD component, typically 2 × 1.2 mm to 5 × 2 mm, soldered onto the PCB like any other part. The BOM cost is $0.10–$0.50 per unit (Hubble), negligible at production volumes. It still requires a ground-plane keep-out zone, but it is smaller: typically 5–10 mm, compared with 15–25 mm for a PCB trace antenna. The vendors provide tested reference layouts with specific ground-plane dimensions and matching networks, which is why the chip antenna hits a sweet spot for most BLE designs: small, predictable, and vendor-supported. Many BLE modules from Raytac, Fanstel, and Laird ship with a chip antenna pre-integrated and pre-matched, and a module with an integrated chip antenna can eliminate antenna design entirely, as long as you follow the module vendor's keep-out rules.

The external antenna connects through a U.FL (IPEX MHF) connector, an SMA connector, or a permanently attached cable, and it decouples RF performance from the PCB layout entirely: the antenna is a standalone, pre-characterised component. Its range performance is the strongest of all three options, commonly 2–4× that of PCB or chip antennas (Hubble), and it can be positioned optimally within the enclosure or mounted outside a metal one. The penalty is cost, connector, cable, assembly labour, and a coax run whose routing and grounding you now have to get right. If the enclosure is metal, the unit is potted or buried, or the range must be rock-solid, go external: put a real antenna outside the metal and stop fighting physics inside it.

The module or the bare chip

Before any layout work begins, the most consequential antenna decision is already in front of you: use a pre-certified module, or design around the bare SoC. The module route, ESP32-WROOM-32E, ESP32-C3-MINI-1, ESP32-S3-WROOM-1, and their equivalents, integrates the antenna, crystal, flash memory, and RF matching network into a shielded unit that has already passed FCC and CE testing. The bare-chip route gives you the smallest possible footprint and the lowest per-unit BOM cost at high volume.

It also gives you every responsibility: antenna design, impedance matching, crystal selection and load capacitance, PCB stackup specification, and a full RF certification process from scratch.

The numbers: a module costs roughly $2.00–$3.50 at 10,000 units, against roughly $0.80–$1.50 per chip plus passives for the bare SoC (NextPCB). Full certification for a bare-chip design runs roughly $5,000–$15,000+. The WROOM-32E is 18 × 25.5 mm; the bare chip is a 5 × 5 mm or 6 × 6 mm QFN.

The recommended volume is prototypes to ~10,000 units for modules, and above 10,000 units where the BOM savings justify the engineering cost for the bare chip. For the vast majority of custom ESP32 PCBs, product prototypes, low-to-medium-volume IoT devices, one-off tools, modules are the correct answer.

The module's naming tells you the antenna type: the ESP32-WROOM-32E carries the on-board PCB antenna and is the current recommended revision; the ESP32-WROOM-32UE carries a U.FL connector for an external antenna. The "U" suffix always means a U.FL connector. The module must sit at the edge of the base board with the antenna portion overhanging the board boundary or at minimum flush with the edge, and the keep-out zone must still be clear on all sides of the antenna, with at least 15 mm of clearance to any metal inside the enclosure.

If the module must sit toward the centre of the board, Espressif's guidelines allow it, but with conditions: the keep-out zone must still be clear, and the antenna's environment is the thing that decides the range.

The matching network

The radio expects 50 Ω, and the antenna must present it across the band. The inverted-F has a built-in advantage here: the position of the ground stub acts as a built-in impedance tuning mechanism, and by adjusting the distance between the feed point and the ground stub, the designer gets the antenna close to 50 Ω without external matching components. The rest is done by a pi-network, one series element and one shunt element, placed close to the antenna feed point.

For an IFA measuring 30 + j15 Ω, the typical matching components are a 2.7 nH series inductor and a 1.5 pF shunt capacitor, transforming the impedance to 50 Ω (Andwin).

The number that tells you whether the match works is the return loss: the reflection at the feed point, which determines how much of the applied power is delivered to the antenna. The design goal is a return loss of less than −10 dB across the whole 2.4 GHz ISM band, when connected to a 50 Ω source, which is a VSWR of less than 2, and ensures that more than 90% of the available power is delivered to the antenna (TI AN043).

The arithmetic is simple: a return loss of −10 dB means 10% of the power is reflected, and the mismatch loss is 0.46 dB. A return loss of −15 dB means 3.2% reflected, and the mismatch loss is 0.14 dB. A return loss of −6 dB means 25% reflected, and the mismatch loss is 1.26 dB.

The return loss is the symptom, not the disease, and the demo shows why: the mismatch loss from a bad return loss is a decibel or two, while the near-field loss from a battery next to the antenna is 6–10 dB. The return loss tells you the antenna is detuned. It does not tell you how much range you have lost, because the detuning also wrecks the pattern and the efficiency, and neither of those shows up in the return loss.

The enclosure: plastic loads, metal blocks

The enclosure is part of the antenna from day one, and the two materials do two different things. Plastic is mostly transparent to RF but not free: a plastic wall a millimetre off the antenna has a dielectric constant of around 2–3 and will lower the resonant frequency, it "loads" the antenna. That is exactly why you tune with the plastic in place.

The shift is real but predictable, and you can design for it, as long as the plastic is in place when you tune, and as long as it stays in the same place in production.

The Flomerics simulation from 2002 is the dramatic version of the plastic story, and it is a simulation, not a measured product, the article says so. A Bluetooth antenna was simulated in free space with a return loss of 22 dB centred at 2.428 GHz, well inside the Bluetooth band of 2.4–2.4835 GHz. Then a sheet of 3 mm engineering thermoplastic was introduced, just touching the top of the antenna.

The broadcast frequency shifted to 2.158 GHz, 270 MHz, outside the Bluetooth band entirely, and the return loss dropped by 6 dB, or 37.5%, to 16 dB. The magnitude of the change, brought about by simply installing the antenna in a more realistic setting, was huge, and in a production device it would have required a significant redesign to bring the unit up to a sensitivity of −70 dBm at 10 m.

Metal is the enemy, and it does two ugly things: it detunes by coupling capacitively, or, close enough, it effectively shorts the near field and craters efficiency. The usual offenders, ranked by how often they wreck a design: the battery, a LiPo pouch or 18650 is a large grounded metal slab, and an antenna laid over or beside the cell with a millimetre of gap can lose 6–10 dB; keep the cell out of the keep-out entirely, and treat 5 mm as a floor and more as better.

The shield can over the RF or PMIC section, its wall must respect the keep-out like any other ground. The USB and connector shells, screws, springs, and metal stiffeners, the grounded USB shell sitting next to a corner antenna is a classic 3 dB thief. And the user's hand and head, tissue is lossy, high-permittivity, and shows up only when someone holds the product the way people actually hold it.

The metal-box field report makes the same point from the other side. A printed antenna on a sub-GHz meter was centred and matched at about −15 dB return loss on the open bench. Once it was fitted inside the metal box, it moved by more than 40 MHz, and the return loss at the operating frequency degraded to roughly −6 dB.

The metal wall close to the antenna changes the effective ground and adds capacitive loading, so the resonant frequency moves down and the bandwidth narrows. The fix was to treat the enclosure as part of the antenna from day one, model and measure with the real box, the real PCB position, and the real cable and battery in place, because every metal object near the antenna changes the result, and to tune in situ, aiming for a return loss better than −10 dB across the whole band with the lid closed, not just at one spot frequency.

One practical check: measure the antenna in the closed box, then open the lid and measure again. If the two measurements differ by more than a few MHz, the design is still too sensitive to the enclosure, and it needs more clearance or a different antenna topology.

The physics of the metal wall is absolute: at 433 MHz, the skin depth of aluminium is roughly 4.1 micrometres, and a 2 mm machined wall is about 490 skin depths thick. Nothing propagates through it, and no increase in transmit power recovers it. Energy leaves a metal box through apertures, and for a single opening, the classical estimate for shielding effectiveness is 20 log(λ/2d), where d is the longest dimension of the aperture: at 433 MHz, a 10 mm seam gives roughly 31 dB of attenuation and a 40 mm slot gives roughly 19 dB.

Enclosure architecture therefore decides the link budget before any schematic exists: the useful apertures are a plastic end cap, a plastic bottom plate, an IR window, a battery door, or a deliberate dielectric insert. In the handheld remote the Grinalds guide describes, the top was aluminium and the bottom was plastic, so the radiating aperture was the entire bottom face.

Example Labs is a made-up company we use for worked examples.

DEMO

The Example Labs field sensor hub is the 100 × 60 × 40 mm unit from the cluster: a 4-layer board, a BLE radio at 2.4 GHz (from #12), and a plastic enclosure (from #7). The demo works the link budget, then prices what the enclosure and the battery cost it. The arithmetic was run in Python before writing.

Step 1, the wavelength. At 2.4 GHz, the free-space wavelength is 124.9 mm and the quarter-wave is 31.2 mm. The IFA's radiating arm is 25–30 mm (TI's reference is 25.7 mm), and the compact meandered version is 15.2 mm. The hub's board is 100 × 60 mm, so the standard IFA fits at the board corner with room for its keep-out, and the demo uses the standard IFA, not the compact one, because the board has the space.

Step 2, the link budget. The Friis equation, in decibels: received power equals transmit power plus transmit antenna gain plus receive antenna gain, minus free-space path loss, minus miscellaneous losses. Free-space path loss is 20·log₁₀(4πdf/c), and at 2.4 GHz and 1 km it is 100.1 dB, at 100 m it is 80.1 dB, and doubling the distance adds 6.02 dB. The demo uses a transmit power of 0 dBm (a stated value, BLE SoCs output 0 to +8 dBm), antenna gains of +2 dBi at each end (a stated value, the IFA's gain is 2.0–2.5 dBi when properly tuned), and a receiver sensitivity of −90 dBm (a stated value, a typical 6 Mbps sensitivity). Free space, line of sight, no multipath, no fading, no interference, stated.

DistanceFSPLReceived powerLink margin
10 m60.1 dB−56.1 dBm33.9 dB
50 m74.0 dB−70.0 dBm20.0 dB
100 m80.1 dB−76.1 dBm13.9 dB
200 m86.1 dB−82.1 dBm7.9 dB
300 m89.6 dB−85.6 dBm4.4 dB
498 m94.0 dB−90.0 dBm0 dB

The maximum free-space range, at zero margin, is 498 m. That is the bench number, and it is the number the enclosure takes away from.

Step 3, what the detuning costs. Every 6 dB of loss halves the range, because doubling the distance adds 6.02 dB of path loss. A 3 dB loss, the grounded USB shell sitting next to the corner antenna, multiplies the range by 0.71, and the 498 m becomes 353 m. A 6 dB loss, the battery a millimetre from the antenna, halves it, and the 498 m becomes 250 m. A 10 dB loss, the battery hard against the antenna, divides it by 3.16, and the 498 m becomes 158 m. The mismatch loss, by comparison, is small: a return loss of −15 dB (the bench) is 0.14 dB of mismatch loss; a return loss of −6 dB (the metal box) is 1.26 dB. The return loss is the symptom. The 6–10 dB of near-field loss is the disease, and it does not show up in the return loss, because the detuning also wrecks the pattern and the efficiency.

Step 4, the enclosure. The hub's lid is plastic, and the Flomerics numbers say what happens if the plastic touches the antenna: the resonance shifts by 270 MHz, out of the Bluetooth band, and the return loss drops by 6 dB. The demo's rule is that the lid must not touch the antenna, the battery must be at least 5 mm clear, and the antenna must be tuned with the lid closed and the battery in place. If the hub's enclosure were metal instead, the demo's rule would be different: no increase in transmit power recovers a metal wall, and the antenna would have to go external, a U.FL whip outside the box, with the 2–4× range that the external option buys.

Step 5, the three options for the hub. The trace IFA is free and fits the 100 × 60 mm board, but it needs the 25–30 mm arm, the 5–8 mm keep-out on every layer, and a 15 mm clearance to any metal inside the enclosure, and it must be tuned with the lid closed. The chip antenna costs $0.10–$0.50, needs a 5–10 mm keep-out, and buys a smaller footprint at some efficiency. The external whip costs the most and buys 2–4× the range, with the antenna outside the enclosure entirely. The demo's recommendation is the module route: an ESP32-WROOM-32E at the board edge, with the antenna overhanging the board boundary, the 15 mm housing clearance respected, and the module's pre-certified trace doing the tuning, because the hub is a field sensor, not a lab demo, and the field is where the range is decided.

The decision test: five questions

A hardware team with a Wi-Fi or BLE product that works on the bench and dies in the enclosure can answer five questions in an afternoon. If it cannot answer three of them, it has an antenna problem.

  1. Which antenna is it, and why that one? If the answer is "the one that was in the reference design," the question is whether the reference design's keep-out, ground plane, and enclosure were copied too, because the antenna is the trace plus the ground plane plus the empty space plus the enclosure, and copying the trace alone is copying half the antenna.
  2. What is the keep-out, and is it empty on every layer? If the keep-out exists only on the top layer, it does not exist: the ground pour on layer two is still under the antenna, and the return loss will say so. If a trace crosses the keep-out, the antenna is loaded and the resonance has moved.
  3. What is the ground plane, and how big is it? If the ground plane is smaller than a quarter wavelength, 30–35 mm, in any direction from the feed point, the counterpoise is too small, and the efficiency is down 40–60% before the enclosure is even closed.
  4. What is the return loss with the lid closed, the battery in place, and the product held the way people hold it? If the measurement was taken on the open bench, it is a measurement of a different antenna. If the return loss is worse than −10 dB across the band with the lid closed, the antenna is detuned, and the fix is clearance, not a new matching capacitor.
  5. What is the link budget, and what does the enclosure cost it? If the team cannot say what the free-space range is and what 6 dB of near-field loss does to it, it cannot say whether the product works in the field, and the field is where the product lives.

Objections

"The radio module is pre-certified, so the antenna is done." The module's certification covers the module, tested in the module's conditions. The moment the module sits on a different board, in a different enclosure, next to a different battery, the tested conditions have changed, and the module's authorisation evaporates if any tested condition is modified (from #6 in the pipeline). The module is the start of the antenna work, not the end of it.

"We'll just turn up the transmit power." The radio's output is 0 to +8 dBm, and the legal limit is the EIRP, not the transmit power, and no increase in transmit power recovers a metal wall, because a 2 mm aluminium wall is 490 skin depths thick at 433 MHz. A 6 dB loss from the battery costs half the range, and recovering it would need four times the transmit power, which the radio does not have, and the battery (from #6) cannot afford.

"The chip antenna is small, so it doesn't need a keep-out." It does. The chip antenna still requires a ground-plane keep-out zone, typically 5–10 mm, against 15–25 mm for a PCB trace antenna, and the vendors provide tested reference layouts with specific ground-plane dimensions and matching networks for a reason. A small antenna with a violated keep-out is a small antenna that does not radiate.

"We tuned it on the bench, so it's tuned." The bench is not the product. The plastic lid loads the antenna, the battery detunes it, the metal screws short its near field, and the user's hand absorbs it. The Flomerics simulation moved the resonance by 270 MHz when a sheet of plastic touched the antenna, and the sub-GHz meter moved by more than 40 MHz when it went into its metal box. Tune with the lid closed, the battery in place, and the product held the way people hold it, or tune it twice.

"The range is fine, we tested it across the office." The office is not the field. The free-space range is 498 m at zero margin, and the link budget's margin is what the field eats: multipath, fading, interference, and the 6–10 dB the battery costs. A product that works across the office with 4 dB of margin works until the first metal shelf, the first concrete wall, or the first rainy day, and then it does not.

Sources

  1. Texas Instruments, "2.4-GHz Inverted F Antenna" (SWRU120D, revised January 2019)
    What we used

    The reference IFA is 25.7 × 7.5 mm overall (radiating arm 25.58 mm), maximum gain +3.3 dBi on the reference board; "make an exact copy of the antenna dimensions to obtain optimum performance," import the Gerber or DXF; the impedance is tuned to 50 Ω on the reference board and is affected by the ground plane's size and shape, objects in the near field, PCB thickness, and material; add a pi-network close to the feed point

  2. Texas Instruments, "Small Size 2.4 GHz PCB Antenna" (AN043, SWRA117D)
    What we used

    The meandered IFA needs no more than 15.2 × 5.7 mm, VSWR under 2 across the 2.4 GHz ISM band when connected to a 50 Ω source; a reflection of less than −10 dB ensures more than 90% of the available power is delivered; "small changes of the antenna dimensions may have large impact on the performance."

  3. Hubble, "PCB Antenna Design for BLE"
    What we used

    2.4 GHz is ~125 mm wavelength in free space, ~80–87 mm effective on a PCB (60–70% of free-space), quarter-wave ~31 mm; every IFA and meander is that 31 mm folded to fit; the ground plane is part of the antenna, and copper directly beneath the element creates an image current that opposes radiation; "every BLE chip vendor gives you a reference antenna layout."

  4. Hubble, "PCB Antenna vs Chip Antenna vs External Antenna for BLE Designs"
    What we used

    BLE SoCs output 0 to +8 dBm; PCB trace gain 0 to +2 dBi when properly tuned, demands 15–25 mm of ground-plane clearance; chip antennas 2 × 1.2 mm to 5 × 2 mm, BOM $0.10–$0.50, keep-out 5–10 mm; external antennas commonly 2–4× the range of PCB or chip antennas; modules from Raytac, Fanstel, and Laird ship with a chip antenna pre-integrated and pre-matched

  5. Andwin, "Antenna PCB Design, PCB Trace Antennas for IoT Devices"
    What we used

    Typical trace lengths 28–32 mm (quarter-wavelength adjusted for the PCB dielectric constant); IFA 25–30 mm × 7–8 mm, ~200–250 mm², gain 2.0–2.5 dBi, bandwidth 5–8%; inadequate ground plane reduces efficiency 40–60% and shifts resonance 50–150 MHz; keep-out typically 5–8 mm beyond the antenna perimeter for 2.4 GHz (10–15 mm sub-GHz); ground plane should extend at least λ/4 (30–35 mm at 2.4 GHz on FR-4) in all directions from the feed point; IFA with 30 + j15 Ω matched by a 2.7 nH series inductor and 1.5 pF shunt capacitor to 50 Ω

  6. GizanTech, "ESP32 Antenna Design and Placement"
    What we used

    The three-option comparison table (trace / chip / external) on cost, range, board area, tuning effort, and certification ease; the most common range-killer is a ground pour, power pour, or signal routing under or beside a PCB trace antenna; the keep-out is a copper-free zone on every layer, no ground plane, no power plane, no traces, no via stitching

  7. NextPCB, "ESP32 Antenna Design Guide"
    What we used

    The module route integrates antenna, crystal, flash, and matching into a shielded unit that has already passed FCC and CE; the bare chip gives the smallest footprint and every responsibility; full certification ~$5,000–$15,000+; module ~$2.00–$3.50 at 10K units vs chip ~$0.80–$1.50 plus passives; WROOM-32E 18 × 25.5 mm; Espressif's guidelines specify a mandatory keep-out, with at least 15 mm clearance in all directions from the antenna and the same 15 mm to any metallic housing element; the module must sit at the board edge with the antenna overhanging or flush

  8. HIL Electronic, "ESP32 PCB Antenna Hardware Design Guide"
    What we used

    ESP32-WROOM-32E (PCB antenna) vs -32UE (U.FL external); the "U" suffix always means a U.FL connector; a bare-chip antenna needs a pi matching network and a controlled stackup

  9. Ideambox, "Antenna keep-outs nobody mentions until the test report"
    What we used

    An inverted-F at 2.4 GHz is roughly 12–16 mm long and wants 5–10 mm of nothing around it; the ground plane is the counterpoise; plastic (dielectric constant 2–3) lowers the resonant frequency; metal detunes or shorts the near field; a battery beside the antenna with a millimetre of gap can lose 6–10 dB, keep the cell out of the keep-out entirely, treat 5 mm as a floor; the grounded USB shell is a classic 3 dB thief

  10. RF Forum, "Antenna for smart meters, metal enclosure detuning?"
    What we used

    A printed antenna matched at ~−15 dB return loss on the open bench moved by more than 40 MHz inside the metal box, return loss degraded to ~−6 dB; treat the enclosure as part of the antenna from day one; tune in situ, aiming for better than −10 dB across the whole band with the lid closed; measure with the lid closed and open, and if the two differ by more than a few MHz, the design is too sensitive to the enclosure

  11. EDN / Flomerics (2002)
    What we used

    A Bluetooth antenna simulated at 22 dB return loss centred at 2.428 GHz shifted to 2.158 GHz (270 MHz, outside the 2.4–2.4835 GHz Bluetooth band) when a 3 mm thermoplastic sheet touched it, and the return loss dropped by 6 dB (37.5%) to 16 dB; a simulation, not a measured product

  12. Grinalds Solutions, "Getting 433 MHz Out of a Machined Aluminum Enclosure"
    What we used

    Aluminium skin depth at 433.92 MHz is ~4.1 µm, so a 2 mm wall is ~490 skin depths and nothing propagates through it; energy leaves through apertures, 20 log(λ/2d): a 10 mm seam gives ~31 dB attenuation, a 40 mm slot ~19 dB; 20 log(2450/433.92) = 15.0 dB in favour of 433 MHz at equal distance; halving the RF frequency doubles free-space range

  13. RFTools, "RF Link Budget Calculator"
    What we used

    Friis: Pr = Pt + Gt + Gr − FSPL − Lmisc, FSPL = 20·log₁₀(4πdf/c); at 2.4 GHz and 1 km, FSPL = 100.05 dB; λ = 124.9 mm

  14. EngiCalHub, "FSPL Calculator Guide"
    What we used

    FSPL (dB) = 20·log₁₀(d) + 20·log₁₀(f) + 32.44 (d in km, f in MHz); doubling distance adds 6.02 dB, doubling frequency adds 6 dB; every 6 dB of additional link margin doubles achievable range; a typical 2.4 GHz Wi-Fi link at 100 m has FSPL of about 80 dB

  15. Sorumatik, "Wavelength in FR4"
    What we used

    FR-4 dielectric constant ~4.4–4.8; λ_FR4 = λ₀/√εr; at 2.4 GHz, λ₀ = 125 mm, λ_FR4 ≈ 59.6 mm

  16. NRTech Studio, "Microstrip Wavelength Calculator"
    What we used

    Effective dielectric constant 3.3249, free-space wavelength 124.91 mm, guided wavelength 68.50 mm, quarter-wave 17.13 mm at 2.4 GHz; the guided wavelength is commonly only 50–70% of the free-space value

  17. Procirel, "Antenna Basics for IoT"
    What we used

    At 2.4 GHz, λ ≈ 12.5 cm, quarter-wave 3.1 cm; the radio expects 50 Ω

  18. RayPCB, "Monopole Antenna"
    What we used

    Free-space quarter wavelength L(mm) = 74,948/f(MHz); at 2.4 GHz, 31.23 mm

  19. Duxtel, "Radio Theory and Link Planning"
    What we used

    Receiver sensitivity examples: 802.11g 6 Mbps → −92 dBm, 54 Mbps → −76 dBm; 802.11b 1 Mbps → −96 dBm

  20. Example Labs figures from the cluster: hub 100 × 60 × 40 mm, 4-layer board
    What we used

    BLE radio at 2.4 GHz (#12); plastic enclosure (#7); CR2032 230 mAh = 2,484 J (#6); sensor hub board 100 × 60 mm, 4-layer, 120 nets (#32); the #19 EMC figures (differential-mode loop, common-mode cable, slot antenna, return path)

More in design notes

Which RadioBLE, Wi-Fi, and LoRa each win one column and lose another, and the wrong choice costs battery life, range, or a redesign. BLE is the cheapest way to talk to a phone (10–15 mA transmit, months to years on a coin cell). Wi-Fi is the cheapest way to move a lot of data (5–20 Mbps practical) but its 2–4 second connection costs 1.78 joules, 51,000 times the BLE reading. LoRa is the cheapest way to cross a field (2–15 km) but its spreading factor is a 23× energy lever and the EU 1% duty cycle turns a 512 kB firmware update into 21 hours. The same 24-byte reading, once an hour for a year: BLE 0.3 J, LoRa SF7 49 J, LoRa SF12 1,144 J, Wi-Fi 15,610 J, 6.3 times the whole cell.13 minCE markingOn Wednesday 20 January 2027, Regulation (EU) 2023/1230 applies in full and Directive 2006/42/EC is repealed, a hard cut-over, unit by unit, with no grace period for new machines built to the old Directive. The Regulation adds essential requirements the Directive never had: protection against corruption, software and update integrity, self-evolving behaviour, and AI in safety functions, and for the six Annex I Part A categories, a notified body is always mandatory. Example Labs' cobot cell costs €6,625 in testing without the AI safety function and €16,625 with it, and the notified-body route lands February–May 2027, after the deadline.21 minHuman FactorsIEC 62366-1:2015+AMD1:2020 is the international standard for usability engineering of medical devices, recognised by FDA as a consensus standard, and FDA's guidance 'Applying Human Factors and Usability Engineering to Medical Devices' (2016, revised August 2026) sets what reviewers expect. The process: specify users and use environment, identify use-related hazards, design the interface to mitigate them, validate with representative users. Formative testing uses early prototypes with 5–8 participants per user group; summative validation uses the finalised, production-equivalent device with at least 15 participants per distinct user group. On average 85% of utilisation problems are found with 5 representative users, 95% with 10. Between 2012 and 2015 there were 423 medical device recalls resulting from UI software errors. Acceptance is the correct execution of a task without use errors or difficulties, a percentage of users completing a task in a time is not a pass mark, because the users who cannot complete it are the signal, not the noise.24 minComponent deratingYour board was designed at the ratings, and now the warranty claims are coming in, slow drift, not sudden death. Here is the derating table that makes the product last: the 50% rule, NASA's table, the tantalum numbers, and what MTBF actually means.19 minEMC/EMIEvery board is a radio station. The only question is whether it knows it is one. A 60 mm clock trace on a 2-layer board fails radiated emissions at its 5th harmonic by 1.7 dB; the same trace on a 4-layer board passes with 14.4 dB of margin. This piece shows the three ways a board radiates, the return path that decides most failures, and the stackup that fixes them.25 min