Antenna Placement and Isolation in Multi-Radio Embedded Systems: The Design Decision You Cannot Undo
Every radio needs an antenna. Every antenna needs clearance. Every pair of antennas needs isolation. Here is what that means when you have four radios and one enclosure.
In a multi-radio embedded product, the RF team and the mechanical team are solving contradictory problems at the same time. The mechanical team is shrinking the enclosure to hit a target form factor.
The RF team needs clearance zones, specific antenna orientations, and physical separation between radios that the enclosure directly constrains. Both teams have valid objectives, but they are conflicting. The product has to satisfy both requirements at once, on a board that was probably laid out before anyone drew the antenna keep-out zones.
Most schedules treat antenna placement as something that gets resolved during layout. By the time layout happens, the processor position is fixed, the power tree is placed, the connectors are committed, and the antenna has whatever board space remains.
On a single-radio board, that approach works. On a board with BLE, Wi-Fi, LTE, and UWB, it produces a product that underperforms on at least one radio before it ships. And the failure is not recoverable in firmware. It is geometry, and geometry does not update over the air.
A 2026 market analysis of embedded antenna systems puts the problem plainly: antennas can no longer be treated as isolated components. They are critical system elements that influence thermal management, sensing performance, and user experience.
Early-stage co-design practices are now emerging as a consistent differentiator between teams that get isolation right the first time and teams that discover the problem from field reports.
This article works through the antenna placement problem in three parts: space, coupling, and material. Each one represents a different class of engineering decision that has to be made before the board goes to fabrication.
For the software side of the multi-radio problem, the arbitration and coexistence challenge that shares the same board, see our earlier article on RF coexistence in multi-radio embedded devices.
Space: What the Numbers Actually Mean
The first thing most teams get wrong about antenna isolation is treating it as a qualitative instruction. Keep antennas apart. Avoid coupling. Maintain clearance. In practice, isolation is a measurement with specific numerical targets, and those targets vary by radio technology in ways that matter directly for layout.
The most demanding targets in a current multi-radio design come from UWB, specifically from angle-of-arrival implementations that are now becoming standard in commercial access control, asset tracking, and proximity applications. Qorvo’s Application Note APH511, the widely referenced industry guide for UWB AoA antenna integration, sets two requirements that EDN describes as non-negotiable:
i) Inter-antenna isolation must reach at least -25 dB across the full UWB operating band
ii) Physical antenna separation should be approximately 0.45 times the signal wavelength
For UWB Channel 9, centred at 7.987 GHz, the 0.45 wavelength spacing works out to roughly 16.87 mm. That number is not a comfortable guideline with margin built in. Miss it, and the two UWB antennas corrupt each other’s phase measurements. The ranging still works. The angle of arrival does not. For a product that depends on UWB for precise location, that failure mode is not something firmware can correct.
For BLE and Wi-Fi sharing a board, the isolation target between their antenna paths is 20 dB or better. The failure mode here is different from UWB. Instead of phase corruption, the problem is receiver desensitization: the stronger radio raises the noise floor seen by the weaker one. The consequence shows up as reduced range and increased packet error rates. A field technician will notice both before any test report catches them.
For LTE alongside shorter-range radios, teams sometimes assume the isolation problem disappears because the frequency bands are different. It does not disappear. It changes into a harmonic problem, which the next section covers directly.
The reason isolation targets matter as specific numbers rather than general guidance is this: a target of -25 dB is a measurement. It can be verified with a vector network analyzer during bring-up, before the enclosure is assembled. Vague separation guidance cannot be verified against anything.
Coupling: The Problem That Distance Alone Does Not Solve
Physical separation is the most obvious tool for achieving antenna isolation. It is not the only one, and on a compact board, it is often insufficient on its own. There are three coupling mechanisms that distance alone does not fully address.
1. Harmonic Desense From LTE.
A cellular modem during a transmission burst generates harmonics that can land in the operating band of a nearby BLE or Wi-Fi receiver, raising its noise floor and degrading its sensitivity. As the analysis of self-generated interference in IoT products describes, the most important harmonic frequencies to monitor are the cellular LTE bands and the Wi-Fi band, and self-generated interference from within the product itself is the most common source of poor wireless performance in embedded multi-radio designs.
Physical separation helps. What most teams underestimate is how much orientation contributes. A cellular patch antenna placed orthogonally to a BLE inverted-F antenna achieves meaningfully better isolation than two parallel antennas at the same separation distance. Orientation is a free variable that costs nothing to optimize at the architecture stage and cannot be changed after the board is laid out.
2. RF Trace Routing as a Coupling Path.
RF design guidelines for multi-radio boards make the point directly: a 50-ohm RF trace that is not cleanly routed creates impedance discontinuities that degrade signal-to-noise ratio and contribute to desensitization across the board.
The antenna placement problem and the RF layout problem are not separate workstreams. They interact, and they need to be evaluated together under the same simulation pass before the board goes to fabrication.
Teams that simulate antenna placement and RF routing independently sometimes discover at bring-up that their individual results did not predict their combined behavior.
3. The UWB Ground Plane Problem.
This is the coupling mechanism that most consistently surprises teams adding UWB to a design for the first time. As documented in research on UWB antenna ground plane effects, the circuit ground plane can distort the antenna pattern if placed too close to the UWB radiating element.
If the UWB antenna ground plane and the RF electronics ground plane are directly connected, or placed too close without adequate separation, the circuit ground plane distorts the UWB antenna’s radiation pattern and compresses its bandwidth. The antenna still transmits. But not in the pattern the design requires, and not across the full bandwidth that accurate time-of-flight measurement depends on.
The fix requires clearing a dedicated zone of the PCB for the UWB antenna with its own isolated ground reference. That zone costs board space that, on most compact designs, was already committed to something else.
4. Edge AI adds a Fourth Coupling Consideration.
On-device inference does not directly draw on the radio front-end. Modern tinyML implementations run on dedicated inference cores or micro-NPUs that are architecturally separate from the radio hardware.
But when an inference result crosses a threshold and triggers a transmission, that event-driven burst arrives at the antenna with a timing profile that is fundamentally different from scheduled periodic telemetry. It is unpredictable, it can be high-priority, and it has to travel through the same antenna paths whose isolation was characterized under a different traffic model.
A product whose antenna layout was validated only against periodic telemetry and OTA, update traffic may behave differently once an inference layer is generating bursty, event-driven transmissions at moments of peak radio demand. This is worth modeling explicitly if the product includes any form of on-device inference, rather than discovering it during field validation.
Material: The Constraint Nobody Budgets For
The substrate and enclosure material decisions that interact with antenna placement are the ones most frequently absent from early design reviews, and the ones with the most stubborn consequences when they surface late.
i) FR4 Versus Lower-Loss Substrates for UWB.
Separating the UWB antenna from the main ground plane introduces a feed trace between the RF electronics and the radiating element. On FR4, the standard cost-effective substrate for most embedded products, that feed trace introduces tangent loss that degrades signal integrity across the UWB operating band.
As published comparisons of FR4 and Rogers materials show, FR4’s dissipation factor sits around 0.02 at microwave frequencies, while Rogers RO4350B delivers a dissipation factor as low as 0.0037.
At 10 GHz, standard FR4 introduces roughly 0.8 dB of signal loss per inch, compared to approximately 0.18 dB per inch on RO4350B. That difference compounds over the length of a feed trace in a way that matters for UWB link budget. Teams that need to preserve signal integrity across the full UWB band have two realistic options:
Rogers RO4350B: dissipation factor of 0.0037 at 10 GHz, compatible with standard FR4 fabrication processes, typically costs 3 to 5 times more than FR4 per board
Isola FR408: intermediate performance, better than FR4 above 1 GHz, lower cost than Rogers materials, suitable for designs where the full Rogers premium is not justified
The cost difference is a board-level decision that compounds at production volume. It needs to be evaluated at the architecture stage. Discovering that FR4 is insufficient during bring-up, when the substrate is already specified, and boards have been fabricated, is one of the more expensive ways to learn it.
ii) The Enclosure Effect.
Every measurement taken on a bare board changes once the enclosure is assembled. Plastic enclosures affect antenna performance less than metal ones, but every enclosure material has a dielectric effect that shifts antenna resonance and changes the radiation pattern.
A product that meets its isolation targets on a bare board may not meet them inside its own housing. Testing with the enclosure present is not optional. It is the only measurement that reflects how the product actually performs in the field.
iii) Pre-Certified Modules as a Material and Risk Decision.
For teams shipping their first or second multi-radio product, pre-certified modules trade higher unit BOM cost for a known, validated antenna design that has already passed RF characterisation.
On Zephyr RTOS-based products, where the firmware stack, driver layer, and radio hardware need to be co-validated in any case, using pre-certified modules reduces the variables in the antenna validation pass specifically. That tradeoff is almost always worth taking on a first design. It is almost never discussed at the point in the schedule when it would change the decision.
Bottom Line
Antenna placement is the point in a multi-radio design where the physical world asserts itself most completely. Board space, enclosure geometry, ground plane topology, and substrate material all interact in ways that simulation models well but cannot fully predict. The only authoritative measurement is the one taken with the enclosure assembled, under real operating conditions, with all radios active simultaneously.
What separates the teams that get this right consistently from the teams that discover the problem from field reports is not access to better tools or better components. It is the decision, made early, to treat antenna placement as an architectural input rather than a layout output. That decision costs nothing at the start of a project. However, it costs a board respin to correct at the end of one.
If your team is working through antenna placement decisions on a current multi-radio design, or planning a product that will combine BLE, Wi-Fi, LTE, and UWB on a single board for the first time, embedUR’s engineering team is glad to take a look at what you are dealing with. We have worked through this class of problem across many production designs and can help identify where the risk sits before the board goes to fabrication. Get in touch with us here.
