EMC Filters on the Board: Layout and Return

What a Filter Has To Do

A filter on a board is there to keep noise from leaving through a cable, or to keep noise from entering a circuit from outside. It works by presenting a high impedance to the noise while passing the signal, and it does so at the frequencies where the product is measured. Two things make a filter fail: the component values are wrong for the frequency of interest, or the layout allows the noise to bypass the filter by travelling through a path the designer did not consider. The second is far more common, and it is the reason filter layout matters more than filter selection.

Differential and Common Mode

Noise on a pair of conductors exists in two forms. Differential noise is a voltage between the two conductors, and it is attenuated by a capacitor across them or by a series impedance in one or both lines. Common mode noise is a voltage that appears on both conductors relative to the reference, and it is the form that drives cables as antennas and that dominates emissions above a few tens of megahertz. A capacitor across the lines does nothing for common mode noise; what attenuates it is an impedance in both lines in the same direction, usually a common mode choke, combined with a path for the current to return to its source through the reference. A filter designed without distinguishing the two modes will suppress one and pass the other.

Component Choice

A typical board-level filter is a pi network: a capacitor to the reference at the connector, a series element, and a second capacitor at the circuit side. The series element is a ferrite bead for low current signals, an inductor for higher current, or a resistor where the impedance and the voltage drop allow it. Ferrite beads are widely used and widely misunderstood: their impedance is frequency dependent and mostly resistive at the frequencies where they are useful, and a bead that is chosen for its direct current resistance rather than its impedance at the noise frequency will pass the noise. The capacitors have to be low inductance, which means small packages and short connections, and their own self-resonance limits how high in frequency they remain effective. Where the filter has to work at hundreds of megahertz, the physical arrangement of the components matters more than their nominal values.

Placement and the Return Path

The filter belongs at the point where the noise would leave the board, which is the connector, and it belongs on the connector side of everything else. Placing it further into the board leaves a length of trace that acts as an antenna between the connector and the filter, and the noise that couples into that trace is not attenuated. The return path is the second half of the design: the filtered current has to return to the source through the reference, and if the reference path is long or has a high impedance, the noise will find another route. This is why a filter capacitor should connect to a ground plane with a short, wide path and several vias, and why the plane under the filter should be solid. A filter with a long ground return is a filter that appears to do nothing when the product is measured.

Grounding the Filter

Where a connector’s shell or shield is involved, the grounding arrangement becomes part of the filter. The shell should be bonded to the chassis or to a dedicated ground with a low impedance, and the filter’s reference should be the same node, otherwise the common mode current has nowhere to go. Where the board reference is not the chassis, the connection between them is a deliberate design feature, and its position relative to the connector determines whether the filter attenuates the common mode current or provides it with a path around the filter. This is the point where a design that looks correct on a schematic fails on the bench, and it is usually solved by drawing the current path rather than by changing components.

Filters at Connectors

Connectors are where most filters are needed and where they are most often compromised. The filter has to be close to the pin, which conflicts with the mechanical space the connector occupies, and the reference has to be available at that point, which conflicts with the layout of the connector footprint. The practical arrangements are a filter array integrated into the connector footprint, a row of components immediately behind the pins on a ground plane, and where the connector is shielded, a filtering element that is bonded to the shell. Whichever is used, the connection lengths should be kept to a minimum and the ground should be as close to the signal as the footprint allows.

Measuring the Result

Filters are verified by measuring the emission or the immunity of the product, not by measuring the components. A near field probe on the cable and a spectrum analyser will show whether the noise is reduced at the frequencies of interest, and the measurement should be made with the product in its normal operating mode and with its normal cables, since the cables are part of the antenna. Where the filter is not working, the useful diagnostic is to bypass it deliberately and see whether the emission changes: if it does not, the noise is not travelling through the path the filter was meant to block, and the design should look for the path it is using instead.

Filtering Power Lines

The same principles apply to a power line, with one addition: the filter has to work in the presence of the direct current, so the series element is a component that can carry it, and the current rating has to be respected without saturating the magnetic element. A ferrite bead that saturates under the operating current offers almost no impedance to the noise, which is a common reason a power filter that measured well on the bench does nothing in the product. The capacitor side of the filter is the same as for a signal line, with one further consideration: the impedance of the power source and of the cable feeding it can form a resonance with the filter, which can amplify the noise at one frequency instead of attenuating it. Where the power line is long, the filter should be damped, usually with a small resistor or a lossy capacitor, so that the resonance does not become a problem. The most reliable way to check the result is to measure the emission with the product’s real power supply rather than a bench supply.

FAQ

What is the most common filter layout mistake? A long ground return, which gives the noise a path around the filter and makes the filter ineffective.

Does a capacitor fix common mode noise? No. Common mode noise needs an impedance in both lines, usually a common mode choke, and a return path to the source.

Where should the filter go? At the connector, on the connector side of everything else, so that no unfiltered trace runs from the connector into the board.

Why do ferrite beads sometimes not work? Because they are selected for direct current resistance rather than for impedance at the noise frequency.

How is a filter verified? By measuring the emission or immunity of the complete product with its cables, not by measuring the components on the bench.

Conclusion

A filter is a layout structure as much as a set of components, and it works only if the noise is forced through it and its return path is short and solid. Distinguish the modes, choose components for the noise frequency, place the filter at the connector and check the ground return path. EMC layout is part of PCB design and layout, the assembly of the filter components belongs to SMT PCB assembly, and the verification is part of PCBA testing. Emissions testing is normally planned alongside prototype PCB assembly in 2026.