Duplexer, Cavity Filter, Combiner or Multicoupler? What Each One Actually Does

Four devices sit between radios and a shared antenna. In a rack they look much the same: cylinders of silver or copper tube with adjustment rods on top and coaxial cables looping between them. They cost thousands of dollars, and buying the wrong one is a mistake you discover after the site is built.

The confusion is understandable, because all four are doing versions of the same job, which is letting equipment share one antenna without ruining each other. What separates them is which direction the problem runs and how far apart the frequencies are.

The Short Answer

A duplexer is a two port case of the combining problem with a receiver on one side. A combiner is the transmit side scaled up. A multicoupler is the receive side scaled up. Cavity filters are the components most of them are built from.

What Actually Picks the Filter

The question that decides everything is not “how many megahertz apart are they”. It is fractional separation: the gap between the frequency you are passing and the one you are rejecting, as a percentage of the passed frequency.

Five megahertz is a comfortable split at 450 MHz, which is 1.1 percent. The same five megahertz at 40 MHz is 12.5 percent, an entirely different problem. What a resonator responds to is detuning normalised against its own resonant frequency, so the percentage is the number that matters and the megahertz figure on its own tells you very little.

That gives a set of rule-of-thumb starting points. They are starting points and not a classification, because which topology is right also depends on the rejection you need, the insertion loss you can afford, the bandwidth, the resonator Q, the cavity size, the power, the tuning tolerance, the number of resonators and how much isolation the antennas already give you:

Two things are worth saying plainly about that table. No standard governs filter selection. These thresholds are engineering practice distilled from resonator theory and manufacturer application notes, not a specification anyone can cite at you. And the bottom row is a real answer rather than a failure. Below about half a percent, adding cavities buys progressively less and costs progressively more loss, and moving an antenna is usually cheaper than filtering your way out.

A Duplexer’s Rejection Is Not Its Filter Skirt

This is the part that surprises people who have used duplexers for years, and it came out of measurement rather than theory.

A pass-reject cavity, the kind almost every duplexer leg is built from, places a transmission zero, a deliberate null, at the rejected frequency through a phasing network. That null, rather than the selectivity of the resonator on its own, is where most of the rejection comes from.

Taking five duplexer legs with stated cavity counts and marker readings from swept measurements, solving the loaded Q from each leg’s measured insertion loss, and asking what the bandpass skirt alone would predict:

The skirt accounts for between 5.7 and 14.6 dB. The null accounts for between 65.2 and 79.6 dB. On every one of the five legs the null contribution exceeds the skirt contribution, by 53.4 to 65.5 dB.

Those contributions are stated in decibels deliberately, and not as a share of the total. Decibels are logarithmic, so describing a portion of an attenuation figure as a percentage of the rejection invites a reading that does not mean anything. The defensible statement is the one above: in these five measurements the rejection is dominated by the transmission zero, and the skirt is a minor term.

The solved loaded Q values run from 121 to 562 against unloaded Q values of 1,800 to 7,500. A loaded Q that low gives a broad response, and a broad response has a shallow skirt, which is exactly what the second column of that table shows. The cavity is still a filter and its network still shapes the response. What it is not doing is producing this rejection by being selective.

That has a practical consequence. Because the depth comes from a null rather than a skirt, what sets it is how deep that null can be tuned and held, which is a matter of construction quality and tuning effort rather than geometry. It is why two duplexers with identical cavity counts and identical splits can differ by 15 dB, and why a duplexer that drifts out of tune loses rejection far faster than a bandpass filter of the same size would.

Worth noting that these five are not all the same kind of part. Three are catalogued as pass-reject and two, the ZDX1317-NF-6 legs, as bandpass duplexers. The skirt still falls short of explaining the measured rejection on both, by the widest margins in the set, which makes the finding stronger rather than weaker: even on parts sold as bandpass, selectivity alone does not account for what the leg achieves.

How far the honesty goes. Across those same five legs, per cavity null depth was 23.63, 31.01, 31.39, 39.11 and 40.34 dB, at a cost of 0.606 to 1.035 dB of insertion loss per cavity. Ordered by detuning, those depths show no monotonic trend, which a genuine skirt would have to. Five points cannot tell you whether the 16.7 dB spread is detuning, cavity size, tuning effort or specification conservatism. So the right way to use them is as a bracket rather than a formula, and they only speak for the separation window actually measured, 0.177 to 0.408 percent. Outside that window it is extrapolation.

Combining Transmitters: Two Ways, and the Choice Is Not Obvious

Putting several transmitters onto one antenna has two answers, and the failure mode is different from filtering. A filter protects a receiver from a transmitter. A combiner’s failure is a power amplifier being driven backwards by its neighbours and generating products on frequencies nobody licensed.

The hybrid combiner is a tree of 3 dB quadrature hybrids. It combines without relying on frequency separation at all, including on adjacent channels, provided the channels sit inside the hybrid’s specified operating band and within its power, balance and termination ratings. That generality is paid for in power, exactly and forever:

Four transmitters on a hybrid combiner means each one puts a quarter of its power into the antenna and heats a resistor with the other three quarters. That is not an inefficiency to be engineered away. It is what the device is.

The cavity star combiner gives each transmitter its own cavity string, joined at a common point. It costs a fraction of a decibel per port instead of several, and gives far more port to port isolation. It only works when the channels are far enough apart for a cavity to separate them, and how far is far enough falls out of the cavity size, the band and how much loss the design can afford.

So the trade is straightforward to state and genuinely hard to decide. Hybrid combining is the honest answer when the channels are too close to filter, and the wrong answer whenever they are not. If your channels are comfortably spaced and you are on a hybrid, you are throwing away most of your transmit power for no reason.

The Multicoupler Trap

A receive multicoupler shares one antenna across several receivers, with an amplifier to recover the splitter loss. It does exactly that, and the datasheet is honest about it.

What the datasheet does not say is that fitting one inserts an active third order nonlinearity at the point in the chain where the signals are largest, ahead of every channel filter. In a typical installation its amplifier sees the composite signal environment after whatever preselection is installed in front of it, and that preselection is usually broad.

Every co-sited transmitter that clears it arrives at the amplifier at full strength, and they arrive together. Their composite drives the amplifier, and the products it makes land inside the passband. No filter after the amplifier can remove them, because by then they are signals like any other.

On a loaded tower, a multicoupler can leave a receiver deafer than connecting it straight to the antenna would have. The sensitivity figure improved and the site got worse.

There is a second, quieter trap. A multicoupler should come out at roughly unity gain. Once the amplifier has set the system noise figure, extra gain buys no sensitivity whatsoever, and it costs headroom at exactly the stage that has none to spare. A multicoupler advertising substantially more gain than the system needs may be advertising a liability.

Both problems are why the composite input level from the surrounding licensed environment is worth calculating rather than estimating. The licence register knows what is transmitting on the tower, and the arithmetic from there is short.

What the Catalogue Says

Figures below are from 441 real parts across our component catalogue: 159 duplexers, 211 cavity filters, 37 transmitter combiners and 34 multicouplers, from Telewave, RFI, Comprod, ZCG and Label Italy.

Duplexers. Of the 159 in the catalogue, 158 publish an isolation figure, and here is the thing worth knowing: not one of them publishes it as a bare number. Every single one ties it to a condition. The most common specification, on 39 of the 158, states it three ways at once: “TX noise at RX (min) 75 dB; RX attenuation at TX (min) 75 dB; TX-RX Isolation (min) 75 dB”. Others quote a pair, such as “23 dB / 27 dB”, for the two directions.

Most usefully, 54 of the 158 quote isolation against a stated frequency separation, and those parts show the fractional separation argument in the vendor’s own words. A Telewave TPRD-3544 publishes “75 dB (5 MHz separation) / 60 dB (3 MHz separation)”. Same duplexer, same cavities, 15 dB of difference, decided purely by how far apart you put the two frequencies.

That is why this article does not quote a median isolation figure for the catalogue. A median across 158 parts whose numbers are measured at different separations, in different directions and against different conditions would be a number with no meaning attached to it.

Power handling is published cleanly as a number and runs from 5 to 650 W, with a median of 150 W. Insertion loss is mostly published cleanly too, from 0.2 to 2.5 dB, though read the basis before comparing: some parts quote per pair and others quote the whole path.

Cavity filters split almost evenly three ways by class: 82 bandpass, 67 pass-reject and 62 notch. If you assumed the market was mostly bandpass filters, it is not.

Transmitter combiners split 20 hybrid to 15 cavity across the 37 in the catalogue, with the remaining 2 catalogued as integrated units. Both approaches are genuinely in use rather than one being legacy.

Multicouplers publish port to port isolation as a floor rather than a range, most commonly as 23 dB or better, with some parts quoting 20 dB or better and a few offering 80 dB or better as a custom option. Put that beside the 75 dB minimum that is the most commonly published duplexer figure and the difference is the point: a multicoupler isolates receivers from each other, which is a far easier job than isolating a transmitter from a receiver. They are not substitutes for one another.

Practical Rules

• Work in percent, not megahertz. Fractional separation decides which filter type is even possible. The same gap is easy at 450 MHz and impossible at 40 MHz.
• Below about half a percent, price moving the antenna before pricing more cavities. It is frequently cheaper.
• Do not read a duplexer’s rejection as a filter skirt. In the legs measured here it is dominated by the depth of a tuned null, so it depends on construction and tuning as much as on size, and it degrades faster when the tuning drifts.
• Count the loss, not just the isolation. Every cavity that buys 24 to 40 dB of depth costs 0.6 to 1.0 dB of insertion loss, and on the transmit side that loss is power you paid for.
• If your channels are widely spaced, do not buy a hybrid combiner. On four transmitters it discards 75 percent of your power for a flexibility you are not using.
• Treat a multicoupler as an amplifier you are adding to a hostile environment, not as a splitter that happens to have gain. Check the composite level reaching it from the other tenants on the tower.
• Aim a multicoupler at unity gain. Gain beyond what recovers the split buys nothing and costs headroom.

Frequently Asked Questions

What is the difference between a duplexer and a diplexer? A duplexer lets a transmitter and a receiver share one antenna while operating at the same time, on two frequencies that are usually close together in the same band, and its job is isolating the transmit path from the receive path. A diplexer separates or combines two different frequency bands onto a common port, and those bands are normally far apart. The words are one letter different and the devices are not interchangeable.

Do I need a duplexer or a cavity filter? A duplexer if one antenna has to serve a transmitter and a receiver at the same time, which is the usual repeater case. Cavity filters if the problem is between separate pieces of equipment that already have their own antennas and are interfering with each other. Duplexers are built from pass-reject cavities, so the underlying components are often the same parts.

How much isolation does a duplexer give? There is no single answer, and that is not evasion. Of the 158 duplexers in our catalogue that publish an isolation figure, not one publishes it as a bare number: every one ties it to a condition, and 54 of them tie it to a stated frequency separation. One Telewave part quotes 75 dB at 5 MHz separation and 60 dB at 3 MHz, which is the same hardware differing by 15 dB purely on how far apart the frequencies are. The most commonly published specification is a minimum of 75 dB stated across three measurements. What you actually need is set by your transmit power, your receiver’s noise floor and how much desensitisation you will accept.

Why does my duplexer have so much insertion loss? Because depth costs loss, and the two are bought together. Measured across five real duplexer legs, each cavity contributed between 23.6 and 40.3 dB of null depth at a cost of 0.6 to 1.0 dB of insertion loss. A deeper duplexer is a lossier one, and on the transmit side that loss is transmitter power you are discarding.

Is a hybrid combiner or a cavity combiner better? Neither in general. A hybrid works at any channel spacing but loses 3.01 dB every time you double the port count, so four transmitters lose 6.02 dB. A cavity combiner costs a fraction of a decibel per port but needs the channels far enough apart for a cavity to separate them. If your spacing allows a cavity combiner, using a hybrid is throwing away power for nothing.

Can a multicoupler make my receiver worse? Yes, and on a busy tower it often does. It places an active third order nonlinearity ahead of every channel filter, so the composite of every co-sited transmitter that clears the preselector drives it together. The intermodulation products it creates land in the passband and no downstream filter can remove them. The sensitivity specification improves while the site performance falls.

How many cavities do I need? It depends on the depth you need and the separation you have. As a bracket from measured parts, each pass-reject cavity supplies somewhere between 23.6 and 40.3 dB, so a 75 dB requirement lands somewhere between two and four cavities. That bracket is honest rather than precise, and it comes from parts measured over a narrow separation window of 0.177 to 0.408 percent.

Can I just move the antennas instead? Often, and frequently it is cheaper. Vertical and horizontal separation can both provide substantial isolation, though how much depends on the antenna patterns, polarisation, mounting and site geometry rather than on distance alone. Below about half a percent fractional separation, filtering becomes expensive and lossy quickly, which is what makes the comparison worth running. Our guide to how much antenna separation you need works through the geometry.

Work It Out for Your Site

Choosing between these four is what the RF Filter and Combining Designer does. It takes the frequencies at your site, works the fractional separations, selects the filter topology, sizes the cavity count against the rejection your receivers actually need, computes both hybrid and cavity combining side by side, and checks the multicoupler against the composite level from the licensed environment around you rather than an estimate.

For the underlying numbers on their own, the receiver desensitisation calculator quantifies how much a nearby transmitter is costing you, the intermodulation calculator finds the product frequencies, and the 3D antenna separation and co-siting tool works the isolation that geometry alone will give you.

Related reading: What Is Intermodulation (IM3)?, What Is Passive Intermodulation (PIM)? and What Is the Noise Floor? for the mechanisms these devices exist to manage.

Notes on Sources

• The measured pass-reject figures come from five duplexer legs that carry both a stated cavity count and marker quality readings from swept measurements: ZDX1317-NF-HP legs A and B, ZDX1317-NF-6 legs A and B, and ZXL-PRF-150-3N. The loaded Q values were solved from each leg’s published insertion loss on the assumption that the cavity is tuned at the pass frequency, which their VSWR specifications support. Five legs is a small sample and the separation window they cover, 0.177 to 0.408 percent, is narrow. Both are stated because they bound what the figures can be used for.
• The per cavity ranges are the observed extremes across those legs rather than a fitted value or a mean. A mean would imply a central tendency that five points do not establish, and the width of the range is itself the finding.
• Filter selection thresholds are engineering practice distilled from resonator theory and manufacturer application notes. No standard governs filter selection, and they should not be presented as though one did.
• Hybrid combiner losses are exact: combining N ports costs 10·log₁₀(N) dB, so 3.01 dB per doubling. The percentages in that table are computed from it.
• Catalogue figures are from 441 parts in our component catalogue, drawn from published manufacturer datasheets for Telewave, RFI, Comprod, ZCG and Label Italy. They describe what those manufacturers publish, which is not the same as measured performance in your rack.

Datasheet figures are specifications, not measurements of the unit you receive, and cavity performance depends on tuning. Where a design is marginal, measure it on site rather than trusting a catalogue number.