Above a Few Hundred Megahertz, Traces Stop Being Wires

In a digital design a trace is a connection: either it goes from A to B or it does not. At 2.4 GHz a trace is a transmission line with a characteristic impedance, and every discontinuity along it reflects energy back toward the source. A 0.2 in (5 mm) stub that would be invisible at 10 MHz can cost you several dB of transmit power. A ground via placed 0.1 in (2.5 mm) too far from a component pad can turn a matching network into a mistuned filter.

RF layout is not harder because the components are exotic. It is harder because geometry is now electrically significant and you cannot tell by looking whether it is right. A board radiating 6 dB below spec looks exactly like one that works, until range testing.

Controlled Impedance Is a Manufacturing Specification

RF signal paths run at a defined impedance, almost always 50 ohms single-ended and 100 ohms differential for antenna feeds and high-speed pairs. That impedance is set by trace width, dielectric thickness to the reference plane, dielectric constant, and copper thickness, so it is a stackup decision before it is a routing decision.

Practical consequences that catch teams out:

  • You must specify controlled impedance on the fab drawing with target values and tolerance, typically plus or minus 10 percent. A fab house that is not told will build to nominal thicknesses and your impedance may land anywhere.
  • The reference plane must be continuous under the entire RF trace. A split, a slot, or a routing channel in the plane under a 50-ohm line changes the impedance and creates a radiating loop.
  • Layer count usually goes up. A two-layer board rarely supports clean RF because the dielectric is too thick for reasonable trace widths. Four layers with the RF layer referenced to an adjacent solid ground is the practical minimum, and the cost implications are laid out in how many PCB layers your product needs.
  • Keep RF traces short, straight, and free of stubs. Use smooth arcs or 45-degree segments rather than right angles, and never route an unused test point as a dangling branch off the feed line.

Substrate: When FR-4 Stops Being Good Enough

Standard FR-4 works fine through roughly 3 GHz for short traces, which covers Bluetooth, Wi-Fi in the 2.4 GHz band, most sub-GHz radios, and LTE bands. Its problems are a dielectric constant that varies by lot and with frequency, typically 4.2 to 4.6, and a loss tangent around 0.02 that becomes expensive above a few gigahertz.

Above that, mid-loss laminates and PTFE-based materials with stable, tightly specified dielectric constants and loss tangents near 0.002 to 0.004 become necessary. They cost several times FR-4, need different drilling and lamination processes, and not every fab runs them. For 5 GHz Wi-Fi, mmWave, or radar, plan on hybrid stackups that put an RF laminate only on the outer layer that carries the radio and keep FR-4 for the rest.

Two details worth specifying: ENIG's nickel layer is lossy at high frequency, so immersion silver or OSP is often preferred on critical RF surfaces, and low-profile copper reduces conductor loss versus standard electrodeposited foil.

Grounding, Isolation, and Keepouts

Most RF failures in real products are not matching problems. They are noise problems.

Give the radio a solid, continuous ground reference with a dense via fence, vias every eighth of a wavelength or closer along the edges of RF sections and around any shielding can footprint. Partition the layout physically: radio section in its own area, digital and switching supplies well away from it, and no digital traces routed under the antenna or the matching network. Switching regulators are the usual culprits, injecting spurs into the receive band; if sensitivity is 8 dB worse on your board than in the module datasheet, look at the buck converter first.

The antenna needs a keepout region with no copper on any layer, on the order of 0.2 to 0.4 in (5 to 10 mm) depending on the antenna and band, and it also needs a defined ground plane size because a chip antenna's ground plane is part of the radiator. Enclosure material, battery position, and the user's hand all detune it. That whole set of interactions is the subject of antenna design for wireless products, and it should be settled before mechanical design freezes, not after.

Certified Module or Discrete Radio

This is the biggest budget decision in the project.

A pre-certified module contains the radio, the matching network, sometimes the antenna, and carries its own modular grant. You inherit that grant, reducing your certification effort to unintentional radiator testing plus limited retesting. Modules cost 3 to 15 USD more per unit than a discrete implementation and save four to eight weeks plus a large chunk of the certification budget. Below roughly 50,000 units per year, modules almost always win.

A discrete design with a bare transceiver and your own matching network saves that per-unit cost and gives full form-factor control, but it means full intentional radiator certification, RF lab time, at least one iteration on the matching network, and genuine RF layout skill on the team. What the certification side costs either way is set out in FCC certification for electronic products, and the emissions work that applies regardless is described in EMC testing cost. The protocol choice that sits upstream of both, and drives which module family you are shopping in, is compared in Bluetooth vs Wi-Fi for a product.

You Measure RF, You Do Not Estimate It

Simulation gets you a starting point. Only measurement tells you where you are. The minimum useful set:

  • Vector network analyzer for return loss at the antenna feed. Target better than 10 dB, meaning under 10 percent of power reflected, and tune the match on real boards inside the real enclosure.
  • Spectrum analyzer for output power, harmonics, and spurious emissions. Catching a harmonic 5 dB over the limit on your bench costs nothing; catching it at the lab costs a retest fee and a schedule slip.
  • Radiated range tests with the product in its final enclosure, battery installed, and a hand on it.

Lay out the matching network as a tunable pi with three placeholder positions from the first spin, and expect to iterate values. Budget for at least one RF-driven respin. Long-range low-power designs add their own link-budget constraints, covered in LoRa and LoRaWAN in a product.

Get the Radio Right the First Time

Projects House designs RF boards and wireless products end to end, including stackup definition, antenna integration, matching network tuning, and pre-compliance measurement before you book lab time. Send your frequency band, range target, and enclosure concept through our contact form and we will scope the RF work.