RF PCB Design: Impedance Control and Isolation on the Board

Radio frequency circuits are designed in two places at once. The schematic sets the topology, the matching and the gain, and the layout determines the impedance of every connection between those components. A perfect schematic on a board that ignores the transmission line behaviour of its own traces produces a circuit that works on one prototype and not on the next.

The rules that follow are not exotic. They are the ordinary rules of high speed design applied consistently, with less tolerance for improvisation.

Impedance Is a Layout Property

A trace carrying a signal at radio frequencies is a transmission line, and its characteristic impedance is set by its width, the thickness and properties of the dielectric beneath it, and the distance to the reference plane. The design value is usually fifty ohms for a single ended line and a defined differential value for a balanced one, and controlled impedance applies to every connection in the signal path: from the connector to the first component, between stages, and up to the antenna.

Two consequences follow. The controlled impedance stack-up has to be defined before the layout, because the impedance depends on the dielectric thickness and on the layer assignment. And the impedance has to be verified with a coupon on the production panel, because the calculation assumes a dielectric constant and a thickness that the process delivers within a tolerance rather than exactly.

<img src="https://www.gopcba.com/wp-content/uploads/2020/12/ptt_history.jpg" alt="RF trace with ground via fence” />

Microstrip and Stripline

A trace on an outer layer with a reference plane below it is a microstrip, and one buried between two planes is a stripline. The buried form is better shielded and radiates less; the surface form is easier to route and to tune. Both have a defined impedance, but the geometry that produces a given impedance differs between them, so a design that switches layer in the middle of a critical path changes its impedance at that point unless the width is adjusted.

That is why the layer change is either avoided or made deliberately. Where a via is unavoidable, its ground return needs to be provided adjacent to it, and the transition should be treated as a small discontinuity with an inductance of its own. On a sensitive receive path a badly placed via can cost more loss than several centimetres of correctly designed trace.

Grounding and Return Currents

At these frequencies the return current does not spread over the ground plane; it flows directly beneath the trace. That means the reference plane beneath an RF trace has to be continuous, and the ground connections of the components in the path have to be short and numerous.

The standard practice is a fence of ground vias along both sides of a trace, spaced at a fraction of the wavelength of the highest frequency of interest, which confines the field and prevents energy from leaking into neighbouring structures. The ground pads of matching components are connected with vias immediately adjacent to the pad rather than at the end of a short stub, because the stub itself is an inductor at these frequencies and it changes the match.

Isolation Between Stages

An amplifier chain can oscillate if output energy reaches the input by a path the designer did not intend. Coupling can occur through the shared ground plane, through the supply, through adjacent traces, or through the air across a board. The layout responses are physical separation, shielding structures between the stages, careful routing of the supply with its own filtering at each stage, and a ground plane that is continuous under each stage but arranged so that the high level current does not flow through the low level reference.

Where a shield is used, its connection matters as much as its presence. A shield can that is soldered at a few points around its perimeter is effective; the same can connected at a single point, or through long thin traces, behaves as an antenna for the very energy it was meant to contain. Shield pads and their via patterns should therefore be designed as part of the circuit rather than added as a mechanical detail.

Components and Their Parasitics

At radio frequencies every component is also a small reactive network. A capacitor has series inductance that limits its usefulness at the top of the band, an inductor has parasitic capacitance that creates a self resonance, and a resistor has a package capacitance that matters when its value is large. Selection therefore involves the package and the frequency range as much as the nominal value.

Because the parasitics are part of the circuit, the physical size of the components in the matching network is a design decision. Smaller packages have lower parasitics and allow the components to be placed closer together, which is why the same schematic can be tuned in one package size and not in another. The layout should keep the matching components immediately adjacent to the device pin they serve, with no intervening trace length, because every fraction of a millimetre between them is an inductance in series with the matching element.

Materials and the Practical Review

The laminate matters more than it does in digital design because loss is a direct penalty in the link budget. A low loss material reduces attenuation in the trace, though at the cost of a different dielectric constant that has to be accounted for in the geometry. Where the design has a long feed trace, the choice between a cheaper laminate with an active component and a costlier laminate with a passive one is an engineering decision with a measurable answer.

The review of an RF layout has a short list. Confirm the fabrication capability for the tolerances, that the impedance is defined for every layer that carries RF, that the reference planes under those traces are continuous, that the via fences are present and correctly spaced, that the matching networks, which belong to the layout, are compact and close to their devices, that the supply is filtered per stage, and that the ground connections of RF components are short. Then measure: a network analyser on a prototype will show whether the match and the isolation are what the simulation predicted, and that measurement, as part of the test programme, is the only reliable confirmation.

FAQ

Can RF circuits be built on a two layer board? Simple ones can, with a coplanar structure and a solid ground plane on the same layer, but the isolation and the loss are inferior to a multilayer stack.

Why must the reference plane be continuous? Because the return current follows the trace. A split forces it to detour, which changes the impedance and radiates energy.

Does the component value alone decide the matching? No. The package parasitics and the placement are part of the network at these frequencies.

How close should the ground vias be? Close enough that the spacing is a small fraction of the wavelength at the highest frequency in the design, which is a calculation rather than a habit.

Summary

RF PCB design is high speed design with less margin. Define the stack-up and the impedance before the layout, keep the reference plane continuous and fence the traces with ground vias, isolate the stages with distance and shielding rather than hoping the ground plane will do it, and select components with the package parasitics in mind as well as the value. Then verify the impedance and the match with a measurement, because the calculations are models of the board, not the board itself.