6 dB Hybrid Combiner

Clifton Laboratories 7236 Clifton Road Clifton VA 20124 tel: (703) 830 0368 fax: (703) 830 0711

E-mail: Jack.Smith@cliftonlaboratories.com

Home
Updates
Prior Products - no longer available
Documents
Book
Modulation
Loads
Lattice Crystal Filter
IMD Measurements
Using LP100 Coupler
Prototyping
Software Updates
K2 Measurements
Oscillator Noise Measurements
Bypassing
Capacitor Voltage Change
K2 Freq Stability
Cohn Crystal Filter
Receiver AGC Curves
K2 RX Sensitivity
Canned Osc Phase Noise
K2 Interface
K2 Filter
Surface Mount Assembly
TL750L Low Dropout Regulator
Swordfish DDS
Swordfish GLCD Module
Bessel Nulls
AM Modulation
Z10000 with FT-920
Z100 Tuning Aid
Dayton 2007
Softrock Lite 6.2
Header Adapter
Carbon Composition Resistors
Thermometers
Hakko FT-800 Thermal Wire Stripper
Heat Sinks
Diode Turn-On Time
Bill Hewlett and his Magic Lamp
Tektronix P6022 Current Probe
1N400x Diode Family Forward Voltage
Temperature Chamber
Diode Vf vs If
Ferrite Transformers
6 dB Hybrid Combiner
Type 43 Ferrite B-H Curve
K3 IF Bandpass Filter
Estimating Q of Ferrite Cores
Z10000 Buffer Amp
Z10010 Bandpass Filter
Using Softrock as a Panadapter for the K2
Signal Generator Phase Noise & Elecraft K2
Audio Transformer Data and Modeling
Measuring 60 Hz Frequency
Compact Fluorescent Lamp
Z10000-U Buffer Amp and FT-1000MP
WJ-8617B Receiver Impressions
Weather in Clifton VA
Radio Intelligence Example
Diodes for RF Probes
PIC Development Boards and Programmers
Elecraft K3 and Panadapters
Elecraft K3 AGC and S-Meter
Elecraft K3 Noise Blanker and Crystal/DSP Filtering
Jackson Harbor Press VLF Converter
Elecraft K3 Receive Audio
Headphone Impedance

6 dB Hybrid Coupler Experiments [last revised 08 March 2008] Table of Contents Introduction 6_dB_Coupler_Schematic 6_dB_Coupler_Design_2 6_dB_Coupler_Design_3 6_dB_Coupler_Design_4 6_dB_Coupler_Design_5 6_dB_Coupler_Design__6 6_dB_Coupler_Design_7 6_dB_Coupler_Design_8 Minicircuits_ZFRSC-2050 Minicircuit_ZFSC-2-1W RCA_VH476_CATV_Splitter Wilkinson_Lumped_Element_Combiner 15_MHz_Low_Pass_Filter_ IP3_Performance_of_Combiners Click on any of the lines above to jump to that section of this page.
Introduction I've mentioned working on intermodulation testing of a 10 watt 40 meter amplifier built around a 2SC1945 transistor in a single-ended configuration. In order to perform intermodulation testing, I inject two single signals into the amplifier's input, combining the two signals with a hybrid coupler. This is the standard approach to generating two-tone test signals for receiver and low power amplifier testing and is widely covered in the amateur literature. My current combiner is a Minicircuits device with 3 dB nominal loss. However, I wondered if a 6-dB combiner might provide increased port-to-port isolation. I should also mention that these devices are known by various names, including "hybrid combiner," "hybrid coupler" or "hybrid splitter." The term "hybrid" derives from telephone circuitry where a special transformer and associated RC network was developed many years ago to convert a two-wire subscriber loop into a four-wire circuit for the telephone instrument. A two-wire circuit means that communications can be transmitted in both directions (talking and receiving) over a single pair of wires. A four-wire circuit, in contrast, uses one pair of wires in each direction. The traditional twisted pair telephone circuit is a two-wire circuit, but inside a telephone instrument, separate paths are provided for the microphone and ear piece. The hybrid coupler unit inside a traditional telephone instrument combines the separate talk and receive circuits and places them onto the two-wire circuit to the serving wire center. However, the hybrid combiner isolates the instrument's talk and receive signals from each other, as otherwise there would be a feedback howl between the microphone and ear piece. Passive hybrids are generally reversible, i.e., you may use one to split a single signal into two equal paths or you may combine two signals into one. Port-to-port isolation is an important measure of combiner performance, as is the loss between input and output.

The following figure shows the hybrid combiner use we'll concentrate on -- combining two signal sources for intermodulation testing. The two most common hybrid coupler designs have either 3 dB or 6 dB combining loss. This page discusses both 3 and 6 dB versions.


I should mention that you can't—or at least should not—combine two signal generators with a simple coaxial "T" connector or two series resistors. The reason is that the output of Generator A couples into Generator B and vice versa. Almost without exception, the result will be that Generator A creates unwanted output signals or intermodulation from its output mixing with Generator B's frequency and vice versa. In some cases and with some generators you might get away with a direct "T" connection but normally we wish to isolate Generator A's output from that of Generator B and vice versa. The hybrid coupler, perhaps backed up with attenuators, provides that isolation. 6 dB Coupler Schematic The 6 dB couplers I built follow the schematic below. It can be found in many places, such as the ARRL Handbook for Radio Amateurs (1999 ed.), at page 26.40. The Handbook suggests T1 be wound as 10 turns no. 30 enamel magnet wire, bifilar wound on a FT23-77 core for 1-50 MHz coverage. The design is said to provide 40 to 50 dB isolation between ports.

6 dB Coupler - Design 1 I prefer binocular cores (more properly known as "multi-aperture" cores) to toroidial forms and my junkbox only had Type 43 and type 61 binocular cores, so my versions slightly vary from the ARRL's description.
Version 1. Built into the lid of a Hammond die-cast box.
Version 1, exterior view.

The first coupler I built is with a BN43-2402 core, wound with 6 turns no. 34 magnet wire, bifilar. It uses 5%, 51 ohm resistors and is built into the top of a Hammond die-cast box. (Fair-Rite, the core manufacturer, calls this core a model 2843002402.) The data below is taken with an HP 8752B vector network analyzer. Over almost the complete range 300 KHz - 100 MHz (log frequency scale) the splitting loss is within 0.02 dB.

The coupler also exhibits more than acceptable isolation, running between 45 and 55 dB over the HF amateur bands. Common Port Terminated with 50 Ohm Precision Load

Common Port Terminated with Short Circuit

Common Port Terminated with Open Circuit

Common Port Terminated with Precision 75 Ohm Load

Common Port Terminated with Precision 25 Ohm Load

6 dB Coupler Design 2 I built a second hybrid into a small piece of 2"x2" aluminum square tubing to see whether these results could be duplicated when built with 49.9 ohm, 1% resistors.

The above version differs by using 49.9 ohm, 1%, 1/8 watt resistors as well as being more compact. The transformer is also wound with No. 34 enamel wire, bifilar wound, but with four turns. The core, a BN43-2402, is the same. With fewer turns, we expect this design to have worse low frequency performance and the data confirms that expectation. Over the range 1.8 - 30 MHz, the splitting loss varies only 0.02 dB or so, which is more than acceptably flat. (Note the vertical scale differs between this plot and the earlier one.)

The second version's port-to-port isolation is a bit disappointing when compared with the first version. It's still quite usable, but it's clearly not as good as the first example.

The figure below compares the first coupler (green trace) with the second coupler (black trace). I assume the first coupler's performance is related to having more turns on the core and happenstance of core and parts placement.

One measure of isolation is to look at one output port when the other is terminated in a short circuit or an open circuit. The reference condition is to terminate the unused port with a precision 50 ohm load. The figure below shows the hybrid combiner's output on one port when the other port is open circuited. As expected, over the useful frequency range, the output changes little between proper termination (green trace) and an open circuit (black trace).

Likewise, placing a short circuit on one port whilst looking at the second port (black trace), shows little change in output compared with the precision termination (green trace) case.


6 dB Coupler Design 3 With the thought that a larger core would provide improved low frequency performance, I wound a new transformer, consisting of three bifilar turns of no. 22 enamel wire on a Fair-Rite 284300302 core. The 302 core is about three times the size of the 2401 core in all linear dimensions and hence will have quite a bit more inductance. The splitting loss is slightly better than the 2402 cores, but the frequency response is not nearly as flat. Indeed, the expected improvement in low frequency response is not to be seen in the data. Rather, at the low frequency end, the splitting loss reduces to around 4 dB.

The '302 core version's isolation is, however, terrible, compared with the smaller core case. The green trace is for the first prototype, with a 2402 core. The black trace is the second prototype with the '302 core. The 302 core version is 15 to 20 dB inferior than the 2402 core prototype, a result I did not expect.

I'm at a loss to explain the relatively poor performance of the hybrid combiner with the '302 core. It's possible that the windings were damaged (easy to do when winding on a ferrite core, as the edges are sharp and can easily scrape the enamel insulation off the wire.)
6 dB Coupler Design 4 I've rewound the BN43-302 core with 4 bifilar turns of #26 AWG Polythermaleze wire. Polythermaleze insulation is not subject to nearly the same risk of abrasion as normal enamel wire and it's also heat-resistant. It cannot be melted with a soldering iron.
BN43-302 core (left) and BN43-2402 core (right). The paper has a grid 0.2" x 0.2."

BN43-302 core installed.
The coupling or insertion loss makes much more sense after rewinding the BN43-302 core. It's close to the expected 6 dB value although it still drops off more than the smaller core at lower frequencies.

The larger BN43-302 core after rewinding shows inferior isolation compared with either the first or second prototypes constructed with the small '2402 core. The second '302 core exhibits about 5 dB better isolation than the first core, but it's not as good overall as either of the '2402 designs.

6 dB Coupler Design 5 With the thought that perhaps the BN43-302 tests had too many windings, I tried a two winding version. The windings are two bifilar turns of #26 AWG Polythermaleze wire. The wires are tightly twisted, perhaps 12-14 turns/inch. First, the coupling plot. The data shows that reducing the turns from four to two reduces the low frequency performance, even at 1.8 MHz.

The two turn version exhibits considerably less isolation than the four turn sample, with 10-12 dB less port-to-port loss.

6 dB Coupler Design 6 I next tried four bifilar wound turns of #26 AWG Polythermaleze wire on a ferrite toroid core, FT50-77. 77 Material has a higher permeability than 43 Material, 2000 versus 800 at low frequencies. The power split results are quite acceptable, with around 0.1 dB variation between 1.8 and 30 MHz. The splitting loss is slightly greater than the theoretical 6 dB, but still well within the acceptable range.

The isolation is remarkably flat over the range 1.8 - 30 MHz, although at around 32 dB it's noticeably inferior to the best binocular core results.

6 dB Coupler Design 7 The next transformer is wound with 5 bifilar turns, no. 28 enameled magnet wire on a Fair-Rite part number 2643002402 toroid core. This core has the same outer and inner diameter as an FT37-43 part, but is about 50% thicker, which increases the inductance per turn. Compared with the 77 Material core, we see increased loss, nearly 1 dB over the theoretical 6 dB splitting factor. Within the range 1.8 - 30 MHz, flatness is acceptable, but not as good as with the 77 Material core.

Isolation leaves something to be desired as well, at least at frequencies below 10 MHz or so.


6 dB Coupler Design 8 Snelling (Soft ferrites: properties and applications (2nd Ed.) by E. C Snelling) says that the optimum transformer core for frequencies above 5 MHz or so is either a binocular core or a toroid core that looks like a bead, i.e., tall with a small diameter hole. I tried a three turn bifilar wound transformer on a Fair-Rite part 2643021801 ferrite bead. This part has a length of 0.470" and outer diameter of 0.200" with a 0.062" hole. As the part number suggests, it's Type 43 material. The winding is three turns, no. 28 enamel magnet wire, tightly twisted. Incidentally, Snelling's book is well worth reading if you have more than a passing interest in ferrites, but the price tag is in the stratosphere. Amazon has one copy of the first edition listed at $395 and another copy of the first edition for $450, but no 2nd editions. I obtained a copy by inter-library loan a few years ago and photocopied a few key references.

The split data is reasonable, although not as flat as some of the other samples.

The port-to-port isolation is decent and quite usable in most applications over the 1.8 - 30 MHz range. At 3.5 - 30 MHz, the isolation is at least 35 dB, which is adequate for many purposes. I suspect another turn or so would improve this transformer, but with 28 AWG wire, the core is fully filled with three bifilar turns. 50 ohm precision termination on common port

The following plots show port-to-port isolation under conditions of other than 50 ohm termination on the common port. The worst case isolation (short or open) drops to around 12 dB, although with moderate mismatches (25 ohms or 75 ohms) the isolation remains in the 20-24 dB range. Short circuit on common port

Open circuit on common port

75 Ohm Precision Termination on Common Port

25 Ohm Precision Termination on Common Port

Minicircuits ZFRSC-2050 The ZFRSC-2050 is a simple resistive splitter/combiner, along the lines shown in the schematic below. A careful study of the schematic shows that it is fully symmetrical, i.e., any port may be used as the common and the other two ports used as the inputs/outputs. Having only passive resistive components, this design is largely immune from the intermodulation concerns arising from ferrite core transformers. The price paid for that immunity, however, is poor isolation, both port-to-port and splitting/combing. Another advantage of the resistive combiner is that it is frequency flat; down to DC as a matter of fact, with only circuit strays associated with the layout and resistors to upset its high frequency performance. In addition, both amplitude and phase will remain relatively uniform over the frequency range. In fact, Minicircuits rates the ZFRSC-2050 as meeting specifications from DC to 2 GHz.
Resistive Splitter
Inside the ZFRSC-2050. The three surface mount resistors are 49.9 ohm, 1% parts.
Indeed, as Minicircuits data sheet says, this device has very flat splitting loss, changing less than 0.02 dB over the 1.8 - 30 MHz range. Splitting Loss - 50 Ohm Termination on Unused Port

One problem with the resistive splitter is that when the unused port is not terminated with 50 ohms, there's a considerable shift in power delivered to the other port, as seen in the following plots. Whether this is a problem depends on your use of the splitter/combiner and whether the unused port is ever terminated by something other than a precision 50 ohm load. Open Circuit on Unused Port

Short Circuit on Unused Port

The port-to-port isolation is, in fact, equal to the splitting loss, which should be evident upon examination of the schematic diagram. The device is perfectly symmetrical, so the port-to-port isolation equals the splitting loss. The plot below shows port-to-port isolation, with the common port terminated with a 50 ohm precision load. I won't bother to show what happens when the common port is terminated with a short or open circuit, because the plot is identical with the splitting loss under the same termination.

Minicircuit ZFSC-2-1W is a hybrid transformer based splitter/combiner, with a useful frequency range from 1 MHz to 750 MHz. I thought it would be interesting to see the size and type of ferrite core used in the ZFSC-2-1W, but when I opened the cover, I received a surprise—the printed circuit board is coated with silicon rubber.
Interior of ZFSC-2-1W
The splitting loss is quite flat over the 1.8 - 30 MHz range and well within the specification. Splitting Loss with Unused Port Terminated with 50 Ohm Precision Load

If the unused port is terminated with a short or open, the splitting power stays close to 3 dB, unlike the resistive splitter. Splitting Loss with Unused Port Terminated with Open Circuit

Splitting Loss with Unused Port Terminated with Short Circuit

Likewise, the ZFSC-2-1W hybrid transformer splitter has much better isolation than the resistive combiner and the isolation is less sensitive to how the common port is terminated, at least as long as the termination is not a short or open circuit. Common Port Terminated with Precision 50 Ohm Load

Common Port Terminated with Open Circuit

Common Port Terminated with Short Circuit

Common Port Terminated with Precision 75 Ohm Load

Common Port Terminated with Precision 25 Ohm Load

RCA VH476 CATV Splitter I also looked at an inexpensive splitter intended for cable television use. The particular one I examined has the RCA label and is said to be usable from 5 MHz to 900 MHz, model VH476. Although used in 75 ohm applications, the splitter can be used in a 50 ohm system, but with a somewhat reduced bandwidth.
Interior of VH476 CATV splitter. The core is a single hole ferrite bead.
I'll just show two plots for the VH476. Splitting loss is not terribly bad, 3.6 dB plus or minus .01 dB over the HF amateur bands. However, if the unused port is not terminated with 50 ohms, the low frequency splitting loss increases significantly. This suggests the VH476 would not be a good choice for an HF receiver multi-coupler.

Port-to-port isolation also leave a great deal to be desired, particularly at lower frequencies. The most useful part of the VH476 might be the enclosure, which could be used to hold a different core. Of course, this assumes you like Type F connectors.

Wilkinson Lumped Element Combiner Although the 6 dB hybrid couplers described above have more than decent isolation and bandwidth performance, at least as long as a small binocular core is used, the question has been raised by Geoff, GM4ESD, as to whether the ferrite core will be a contributor to intermodulation distortion and whether the level of IMD will prevent properly measuring very high IP3 devices, such as the Mode H mixer developed by G3SBI. A Mode H mixer can have an IP3 in the +40 to 50 dBm range and hence requires a test signal with an IP3 at least 6 dB and preferably 10 dB greater. The IP3 data above suggests that a small binocular core combiner will be adequate to measure a device with an IP3 of +40 dBm or less, at least at 14 MHz. We have reason to expect that at lower frequencies, IP3 generation in the ferrite core will increase. Although larger ferrite cores will certainly alleviate the IP3 problem, a different approach is to remove the ferrite totally and go with a lumped Wilkinson splitter/combiner using air core inductors. The main concerns a Wilkinson combiner present are (a) relatively poor port-to-port isolation; and (b) narrow bandwidth. The narrow bandwidth issue can be resolved by building a separate coupler for each amateur band. I found close to 30 dB isolation in the Wilkinson combiner I built today for the 14 MHz band, which is 10 dB or more above my expectation. The Wilkinson coupler I built is shown schematically below. You can compute the L and C values or you may wish to use the very handy on-line Wilkinson coupler designer at http://my.athenet.net/~multiplx/cgi-bin/wilk.main.cgi. The 800 nH inductors are 10.5 turns of No. 14 enamel magnet wire, wound around a 0.5" mandrel, close spaced. I then stretched the inductors to measure 800 nH, measured at 25 MHz with an HP 4342A Q-meter. The three capacitors are paralleled polystyrene fixed components with 8-60 pF trimmers for fine adjustment. (120 pF plus the trimmer for the 161 pF caps, and a 270 pF plus the trimmer for the 320 pF.) The 100 ohm resistor is two 49.9 ohm, 1/8th watt, 1% resistors in series.

Wilkinson Combiner - component side view

Connector side.
The power split is quite close to the theoretical 3 dB. In the two plots below, the unused port is terminated with a 50 ohm precision load. Both ports are quite close to each other in level. The data below is taken with an HP8752B vector network analyzer, using the combiner as a splitter. Since the device can be operated in either direction, i.e., as a way to combine two sources into one or to split one source into two outputs, the data is directly applicable to the combining mode, although taken in the splitter configuration.


When the unused port is terminated by an open circuit or a short circuit, the target port shows more of an effect than for the ferrite core splitters. Still, however, the loss stays reasonably close to 3 dB, for either an open or short termination. Open circuit on un-used port

Short circuit on un-used port

The final plot shows the port-to-port isolation with the common port terminated in 50 ohms. The data shows a more than respectable 29 dB isolation, with 25 dB or greater isolation over a bandwidth of ±600 KHz.

25 to 30 dB port-to-port isolation is unlikely to be adequate for high IP3 applications. However, if the two ports are further isolated with, say, 20 dB pads, then the net port-to-port return loss is 85 dB, which is more than respectable. Of course, the extra 20 dB pads may not be acceptable, depending on the available power and the desired input power to the device under test.

Minicircuits ZHL-3A Amplifier Conversion Loss Geoff, GM4ESD, has asked about the intermodulation performance of these combiners, in particular those constructed with the small BN43-2402 core. Considering the typical IMD setup as illustrated below, usable to test an amplifier. (A similar setup is used for receivers and mixers, but the frequency spacing and how the output intermodulation is measured must be modified.) Mini-Circuits publication "Modern Amplifier Terms Defined" note http://www.minicircuits.com/pages/pdfs/amp3-4.pdf provides useful guidance in this regard, as does Mini-Circuits publication Improve Two-tone, Third Order Testing, http://www.minicircuits.com/pages/pdfs/mxr1-18.pdf. The four main potential sources of intermodulation in this test setup are: The device under test The combiner The ZHL-3A amplifiers The R3463 spectrum analyzer In theory any part of the setup is susceptible to IMD product generation, down to and including the connectors and cables. (Nickel plated connectors are bad, for example.) However, at levels normally measured in amateur radio applications, we are concerned with the four sources identified above. The R3463 spectrum analyzer can be removed from the list of potential IMD concerns if we pay careful attention to its input signal level. When intermodulation products are observed on the spectrum analyzer, add 10 dB attenuation. If the IMD products are generated within the spectrum analyzer, the IMD products will decrease 30 dB; if they are generated elsewhere in the test setup, they will drop 10 dB. Ultimately, of course, the spectrum analyzer may be the limiting factor in measuring IMD, but we can normally identify this limit by observing the change in IMD products when the input attenuation is altered.

In this web page, there is no separate "device under test' as we are concerned with measuring the IMD contribution of the combiner itself. This leaves, therefore, one potential incidental contributor to analyze; IMD created in the two ZHL-3A amplifiers. Looking at the test setup, it should be apparent that the signal from one generator/amplifier chain appears at the output of the second ZHL-3A amplifier, reduced in level by the combiner's port-to-port isolation. All amplifiers can act as a mixer, creating an intermodulation product between its output and any signal coupled back into its output port, as shown conceptually below. The ratio between the incident signal B and the output IMD product is the "conversion loss." If the conversion loss and the directional coupler's port-to-port isolation are known, it is a simple matter to compute the expected IMD level and IP3 intercept. In general, the conversion loss is not constant with frequency and applied levels, so caution must be exercised in applying data taken at one level and frequency to other levels and frequencies. To measure the conversion loss, we inject a known level test signal into the amplifier's output port, while operating the amplifier at its normal power level. Obviously some care must be taken to avoid damaging the test equipment or amplifier during these measurements and the test setup should also avoid introducing additional non-linearity. The block diagram below show the test setup I used to measure the ZHL-3A's conversion loss. The ZFDC-20-2 directional coupler provides a 20 dB isolated connection to the amplifier under test. Additional protection of the test probe signal is provided with the 6 dB attenuator, which guarantees the probe amplifier sees at least a 12 dB return loss (1.67 :1 VSWR). We can also measure the amplifier under test's output return loss with this setup. Chain A and B both are set to provide 1 watt (+30 dBm) at the output of the associate ZHL-3A amplifiers. At Amplifier A's output port, Chain B's signal level is reduced by 26 dB to a level of +4 dBm.

The spectrum analyzer plot below shows the result. First, we note that the probe signal level is -47 dBm. To convert this to the amplifier output port level, we subtract the 40 dB attenuation between the amplifier output and the spectrum analyzere, and find the level is -7 dBm. Since the applied signal level is +4 dBm, we compute the ZHL-3A's output return loss as 11 dB, or an output VSWR of 1.78. Mini-Circuits specification for the ZHL-3A is a VSWR of 1.65, or a return loss of 12.2 dB. An intermodulation product can be seen a few dB above the noise level on the high side. It is approximately 40 dB below the output probe signal, or 51 dB below the output probe signal applied to the amplifier under test, after we add the 11 dB measured return loss.

Hence, we measure the ZHL-3A's conversion loss as -51 dB with respect to the applied unwanted signal level, measured at the ZHL-3A's output port. As mentioned, the conversion loss can be used to estimate the intermodulation level and IP3 contribution of mixing in the ZHL-3A amplifier. Suppose the combiner has the following specifications: • Coupling loss = 3 dB • Isolation = 25 dB Assuming both amplifiers operate at 1 watt (+30 dBm) output, the unwanted signal at one amplifier deriving from the second amplifier is +5 dBm. With a conversion loss of 51 dB, each amplifier will have an IMD product output of -46 dBm at its output port, or -49 dBm at the combiner output port. The desired signal at the combiner output is +27 dBm, so the IMD product at -49 dBm is 76 dB below a single tone. Recast in the form of IP3, we may compute it as +27 dBm + 76dB/2 = +65 dBm. There's a complicating factor in this, in that both amplifiers generate an IMD product which then add based at the combiner's output based their relative amplitudes and phases. If they add in the worst case, the IP3 will be reduced 6 dB; if they happen to be of the correct amplitude and phase to cancel, then the IP3 can be improved by 20 dB or more. For a worst case estimate, therefore, we can say that this test case will have an IP3 of around +59 dBm or better. The IP3 figure can be improved by improving the combiner isolation, either with a better combiner or with attenuation between each amplifier's output port and the combiner input.
15 MHz Low Pass Filter I should add that the low pass filters following each amplifier are necessary as otherwise harmonics generated by the signal generator and/or the ZHL-3A amplifiers will mix and create erroneous IMD products with potentially large errors. The amplifiers I use are Chebyshev, 15 MHz cutoff frequency, built on the "universal filter board" sold by W8DIZ at www.kitsandparts.com. I designed the filters with AADE Filter Designer software, available free at http://www.aade.com/filter32/download.htm. I wound the inductors on T50-6 cores and adjusted their values to the design target by adjusting the turn spacing, measuring the values at 25 MHz with an HP4342A Q-meter. After adjustment, I applied a liberal coating of Q-dope to secure the windings in place. The capacitors are polystyrene. In theory, the powdered iron cores can also generate IMD, although they are less of a concern than ferrite cores in this regard. The resulting filter design as shown in the AADE design page is:

Completed filter mounted in the lid of a Hammond die-cast enclosure
Filters - outside view
The figure below shows both filters response over the range 1-100 MHz. With the exception of about 8 dB in ultimate rejection, the two filters track closely.

The theoretical response matches the measured response until the component imperfections and stray leakage level is reached.

IP3 Performance of Combiners I used the test setup below to collect the combiner IP3 data.

Wilkinson Combiner The Wilkinson combiner should contribute no excess intermodulation products, as it has only air-core inductors. (At some level, air is non-linear, as seen in the Luxembourg effect, but we are way below those power levels in this case.) The spectrum analyzer screen capture below shows very small IMD products visible at about -72 dB from a single tone. The single tone level at the spectrum analyzer is -14.5 dBm, or +25.5 dBm at the combiner output. The IP3 is thus +25.5 dBm + 72/2 = +61.5 dBm. This number is essentially identical with the estimate we made of the IP3 level resulting from the ZHL-3A amplifier IMD, as the Wilkinson combiner numbers were the ones used in that calculation. Hence, we can say that the Wilkinson combiner with air core coils contributes no measurable intermodulation at the level of our ability to resolve with this test setup. Hence our initial expectation of no excess IMD is confirmed.

Mini-Circuits ZFSC-2-1W Combiner The spectrum analyzer screen capture below shows an IMD product about 72 dB below a single tone. The single tone power at the combiner output is +27 dBm, so the IP3 is +27 dBm + 72/2 = +63 dBm. At 14 MHz, our earlier measurements show the ZFSC-2-1W combiner has about 36 dB port-to-port isolation, so we estimate the ZHL-3A amplifier contribution to IP3 will be around +70 dBm. Hence, the +63 dBm IP3 for the ZFSC-2-1W combiner is plausible.

6 dB Coupler Design 8 (Ferrite bead core) The spectrum analyzer screen capture shows no resolvable IMD product greater than 74 dB below a single tone. The output power at the combiner common port is +24 dBm, so the IP3 is greater than +24 dBm + 74dB/2 = 61 dBm.

6 dB Coupler Design 1 (Ferrite binocular core) The spectrum analyzer screen capture shows no resolvable IMD product greater than 74 dB below a single tone. The output power at the combiner common port is +24 dBm, so the IP3 is greater than +24 dBm + 74/2 = 61 dBm.

6 dB Pad Combiner Since the ZHL-3A amplifiers have rather good conversion loss, I tried connecting the two amplifier output ports together with a simple BNC "tee" connector, with a 6 dB pad at the output of each amplifier. As the image below shows, there are no IMD products visible at 70 dB below a single tone output. At the combiner output, the power level is +20 dBm, so the IP3 is greater than +20dBm + 70dB/2 = +55 dBm.

3 dB Pad Combiner Taking the resistive combiner one step further, I replaced the 6 dB pads with 3 dB pads. As the spectrum analyzer image shows, IMD products are clearly visible at (worst case) -56 dB from the single tone level. The power out of the combiner is +23 dBm, so the IP3 level is +23dBm +56/2 = +51 dBm.