
September 24, 2026
Small Cell EMI Filtering: Getting Every Interface and the Ground Reference Right
A sealed small cell has more than one way out. Power arrives over the Ethernet cable, an optical feed carries the backhaul, and the antenna port leaves through the same housing. Each of those crossings is an opening for common mode current, and each one needs a capacitor that returns that current to the chassis before it becomes radiation. Filtering every port is necessary. It is not sufficient, because the filters only work if they all return current to the same reference.
This is the second article in our 5G base station series. The first covered why a small cell is not a small macro cell and what that means for where noise leaves the enclosure. This one deals with the interfaces themselves: which ones take a filter, which one cannot, and how the ground reference ties them together.
Start With the Interface List
Before choosing parts, list every conductor that crosses the housing wall. A typical pole- or ceiling-mounted radio unit has four to six, and they are not equivalent. Some carry the noise out, some merely provide a path for it, and one is a deliberate opening that must stay unfiltered.
| Interface | What crosses the wall | Filter treatment |
|---|---|---|
| Power over Ethernet | Differential data pairs plus common mode current on the pair and its shield | Feedthrough capacitor at the bulkhead, one per conductor or as an array |
| Optical backhaul | Non-conductive fibre, but a metal strength member or connector shell in many assemblies | Filter only where a metallic path exists |
| DC feed (non-PoE units) | Supply pair and its return, carrying switching harmonics | Feedthrough capacitor sized for the DC current |
| Antenna port | The wanted RF signal, at the same frequencies you are trying to suppress | No filter. Screening and layout instead |
| Chassis seams and vents | Radiated fields, not conducted current | Gaskets and mesh; a capacitor has nothing to do here |
The antenna row is the one that surprises people outside RF design. The port carries the signal you want, and a capacitor across it would attenuate that signal as readily as it attenuates the noise. The same logic applies to any interface whose function is to pass a signal in the band you are trying to suppress.
Why Filtering Every Port Separately Still Misses
Common mode current does not care which connector it leaves by. It seeks the lowest impedance path back to its source, and that path runs through the chassis. If one port has a low impedance return and another does not, the current simply shifts to the easier one. Filter four ports well and leave the fifth poorly bonded, and the unit performs like the fifth.
A parts list does not catch this. Each feedthrough capacitor is a shunt from the conductor to the chassis wall, and what matters is that all of those shunts land on the same reference potential. If they do, the current is returned locally at whichever port it reaches, and nothing propagates along the outside of the cable.
The measurement makes the failure mode visible. Conducted emissions measured at one port can pass while the unit fails radiated emissions, because the current exited by a different opening and used the cable as an antenna. The port you measured was never the problem, which is why the fix has to be argued from the current path rather than from the worst test result.
Where the Capacitor Sits on Each Interface
Power over Ethernet
PoE is the hardest of the filtered interfaces, because the same pair carries the data signal and the DC power. The capacitor has to present a low impedance at the noise frequencies while leaving the signal band untouched. The part has to be transparent in the signal band and lossy above it, which is why the capacitance value is chosen against the Ethernet signal band rather than taken from the top of the range.
Physically, the part goes at the point where the cable enters the housing, threaded into or soldered to the wall. A threaded feedthrough capacitor is the usual choice for a wall thick enough to take a thread; solder-in parts suit thinner sheet metal. Mounting it on the board several centimetres inside gives the noise a length of conductor to radiate from before it is filtered. This is the same argument that applies to any bulkhead position, but it matters more here, because the cable is both the power feed and the antenna.
Optical backhaul and DC feeds
Fibre itself carries no current, so a purely optical run needs nothing. The complication is the hardware around it. Many assemblies include a metal strength member, a metallic connector shell or a shielded hybrid cable that also carries power for a remote radio head. Wherever a conductor provides a continuous metallic path through the wall, that path needs the same treatment as a copper pair. Where the run is genuinely non-conductive, adding a capacitor achieves nothing.
On units that take a separate DC feed, the supply pair is the noise source as well as the path. The switching harmonics generated by the unit’s own converter appear as common mode current on the feed, and the feedthrough capacitor at the bulkhead returns them. Current rating becomes the sizing constraint here rather than signal integrity, because the capacitor has to carry the full supply current as well as the noise.
The Ground Reference Is One Node, Not Several
Every feedthrough capacitor works by connecting a conductor to the chassis. That connection has an impedance, and the impedance depends on the mechanical joint rather than on the capacitor. A threaded body tightened into a painted panel makes contact through the paint, and a painted joint is a poor conductor at the frequencies that matter. Bare metal against bare metal, with the coating removed at the contact area, is what the threaded mounting is designed around.
The same principle applies across the enclosure. If the ports are mounted on separate brackets that are themselves isolated from the main chassis, each bracket becomes its own reference, and currents at one port do not return locally. Bonding the brackets, or mounting the feedthroughs directly to the main housing, removes the difference. This is the kind of detail that never appears in a capacitor datasheet and decides whether the filters work as a set.
What the Enclosure Size Does to the Filter Budget
A sealed unit has no fan and no spare volume, so the filter budget is set by physics rather than by preference.
Capacitor life is temperature-limited, and a sealed enclosure running warm shortens it. A part rated for 105 °C is the safer default when there is no airflow, even if the measured internal temperature is comfortable on the bench. This is the same thermal argument that applies to the DC-link capacitors in a power conversion unit. Read the derating curve rather than the headline rating.
Space at the wall is finite, which pushes designs toward parts with a smaller mounting footprint and toward filter arrays. An array of feedthrough elements in a single body occupies one hole instead of several and keeps the individual conductor lengths short. Where the interface count is high and the wall area is not, that is usually the only way to filter all of them within the mechanical envelope.
What Comes Next in This Series
Knowing which interfaces need a filter is not the same as specifying one. The next article in this series deals with the values: capacitance and current ratings for each interface type, the insertion loss to expect at the frequencies that matter, and acceptance criteria written so they still hold up in a compliance lab.
For the general selection method behind these values, our article on selecting an EMI filter feedthrough capacitor covers capacitance, frequency and mounting in that order. The companion piece on feedthrough capacitor types and mounting styles explains why the mechanical format changes the electrical result.
Common Questions About Small Cell EMI Filtering
How many feedthrough capacitors does a small cell need?
One per conductor that provides a continuous metallic path through the housing wall. In practice that usually means the PoE pairs and their shield, or a DC feed pair where the unit is not PoE powered, plus any shielded hybrid cable. Optical runs with no metallic member need none.
Can I filter the antenna port as well?
No. The antenna port carries the wanted signal at the frequencies the filter would attenuate, so a capacitor across it degrades the radio rather than the emissions. Antenna-side suppression is a screening and layout problem, and the noise has to be dealt with at the interfaces that are not carrying RF.
Why does the unit pass conducted emissions but fail radiated?
Because the current left by an opening you did not measure, and used the attached cable as an antenna. Conducted tests look at one port at a time, while common mode current is free to choose any port with a low impedance return. Closing the gap usually means improving the bonding at the port with the worst return path rather than adding capacitance elsewhere.
Does the feedthrough capacitor need a separate ground wire?
No, and adding one defeats the purpose. The body of the capacitor is the ground connection, and it works by contacting the chassis wall directly. A wire from the capacitor to a distant ground point introduces exactly the inductance the bulkhead mounting was chosen to avoid.
Working on a small cell or radio unit design? Xuansn Capacitor manufactures threaded, solder-in and high power feedthrough capacitors for EMI filtering. Send us the interface list and the frequencies you are trying to suppress, and our engineers will reply within 24 hours with feasibility and a quote.
✉ coco@xuanxcapacitors.com | ☎ +86-769-8166 8821 | Send your specification
