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

E-mail: Jack.Smith@cliftonlaboratories.com
 

What is IP3? IP3 is the abbreviation for "intermodulation intercept, 3rd order." I've described what IP3 is at Z10040A Norton Amplifier and will not repeat it here.

This page is a companion to my IP2 measurement page.

Table of Contents (click to jump to the topic)
Test_Setup_for_Low_Performance_Amplifiers
Setup_for_High_Performance_Amplifiers
Z10000-U_Buffer_Amplifier_
Mini-Circuits_ZFL-500LN_Amplifier
Z10040A_Norton_Amplifier
Mini-Circuits_GaLi-74+_Amplifier
BFQ19_Feedback_Amplifier
Norton_Amplifier_IP3_versus_Supply_Voltage

The classical method of measuring IP3, or odd order intermodulation distortion in  general, is to inject two test signals at f1 and f2 (usually called "tones" although in this case they are in the MHz  range) and examine the level of the intermodulation products, found at 2f1-f2 and 2f2-f1. Other methods exist, such as a notched noise test common in analog microwave and wire carrier systems, the two tone test is almost exclusively used for IP3 measurements in the amateur radio world.

Rather than jump to the arrangement I settled upon, it's useful to discuss what didn't work and why it didn't work, or, at least didn't work as well as I desired.

The setup illustrated below is one I've used for a couple of years for general purpose IP3 measurements. The 6 dB hybrid combiner is the one described at my 6 dB Hybrid Combiner page. It's the first one discussed, built into Hammond die-cast box, employing a BN43-2402 core, wound with 6 turns no. 34 magnet wire, bifilar. It turns out for frequencies between 1 and 30 MHz my simple home brew coupler has better port-to-port isolation than any of the three Mini-Circuits combiners I own.
 

One additional note on the frequency selection. I first started with nice, even frequencies—12.100 and 11.900 MHz, which yields 3rd order intermodulation products at 11.700 and 12.300 MHz. However, like all synthesized generators, the HP8657A's produce spurious signals at various frequencies. Spurious signals can be found at ±100 KHz and ±200 KHz from the center frequency. This means that the ±200 KHz spurious signals from the two test tones fall on top of the 3rd order intermodulation products. This is undesirable, to say the least, as the spurious signals interfere with the 3rd order intermodulation products when the intermodulation products are weak. (The 8657A's spurious signals are down 80 dB or more, but that's still a significant signal under certain circumstances.) Hence, the obvious easy fix is to select tone frequencies such that the synthesizer spurious signals do not fall on top of the 3rd order intermodulation product frequencies. Hence 12.120 and 11.180 MHz as  tone frequencies. (A similar issue exists with respect to the R3463 spectrum analyzer, it has discrete spurious  responses at certain offset frequencies as well.)

Instead of a single IP3 number, I wanted to look at a more detailed view of the Z10040A's intermodulation performance, as illustrated in the  plot below, extracted from the ARRL's 2006 Radio Amateur's Handbook.

Conceptually, therefore, one simply varies the level of the two signal generators and uses the spectrum analyzer to measure the individual tone levels and the intermodulation products. The resulting data is plotted and with a bit of work a nicely formatted plot  resembling the ARRL's sample is produced.

Wrong.

Or, at least over simplistic. The purpose of the exercise is to determine the IP3 performance of the device under test or DUT, not the IP3 performance of the signal generators, or the combiner or the spectrum analyzer, or the combination thereof. To some degree, imperfections in the test equipment can be worked around by adjusting the spectrum analyzer's input attenuator so as  to maintain the combined signals (two test tones) within the spectrum analyzer's spurious free range. However, the range of data that may be presented in the plot, particularly for a preamplifier with very high performance is limited. The spectrum analyzer's finite dynamic range limits the data to input power that results in rather high levels of intermodulation distortion.  (The Advantest R3463 spectrum analyzer I use has a spurious free dynamic range greater than 70 dB, which is reasonable for most purposes. When new, in the mid 1990's, the R3463 carried a price tag in the $25K range.)

So, the question becomes how to extend the spectrum analyzer's dynamic range. The obvious answer is to filter out the two strong fundamental tones, passing only the third order intermodulation products. Ideally, this would be accomplished with a band reject filter, reducing the two test tones 40 or  50 dB and passing the 2f1-f2 and 2f2-f1 intermodulation products without significant loss. In practice, it is easier to use a bandpass filter centered on either 2f1-f2 or 2f2-f1.

As a practical matter, the necessary bandpass filter will be much easier to build if  the tone spacing is increased from the 240 KHz. I have on hand a variety of coupled resonator filters, the Z10010 bandpass design, that I built as prototypes or for test purposes. One filter is centered at 10.700 MHz with a 3 dB bandwidth of 350 KHz or so.

The second improvement that can be made is to add a high quality amplifier after the signal generator and before the combiner. I use Mini-Circuits ZFL-3A broadband (100 KHz - 150 MHz) amplifiers, capable of over 1 watt output power. (Since each amplifier handles only one tone, the amplifier's IMD performance is not of major concern.)

The figure below shows the resulting test setup.
 

The two frequencies I settled on, 12.733 and 14.713 MHz, yield a lower 3rd order intermodulation product of 10.753 MHz, which is free of spurious signals from either signal generator or the spectrum analyzer.

One remaining subtle point. The 10.7 MHz bandpass filter presents a reflective load at rejection frequencies. This means that the amplifier being tested would operate into an undetermined impedance load at the two test tone frequencies, not the designed 50 ohm load impedance. The easy way to ensure that the amplifier being  tested sees a reasonable load impedance is to add a 10 dB pad between the device under  test and the 10.7 MHz bandpass filter input. The worst case seen by the amplifier under test is therefore a return loss of 20 dB, or an VSWR of 1.22:1, representing a quite reasonable termination.

The 10.7 MHz bandpass filter provides sufficient selectivity to attenuate the 12.733 and 14.733 MHz  frequencies by at least 40 dB. This greatly reduces the demand for high dynamic range in the measuring device. (I've written this page based on an Advantest R3463 spectrum analyzer as the measuring device. I've also used an HP 3586B selective voltmeter as the measuring device. With 20 Hz selectivity, and a better noise figure, it permits measuring signals below -120 dBm, whilst -100 dBm or so is about the limit of the R3463 with 300 Hz bandwidth and 20 sweep averaging.)

I've previously discussed the reason to use attenuators after the buffer amplifiers and will not repeat it. See 6 dB Hybrid Combiner for details.

Two final points before looking at some output data.

First, I set the signal generator levels to provide equal amplitude outputs at the hybrid output. A half dB or so difference between the two generators is required to compensate for different losses in the low pass filters and differences in generator attenuator accuracy.

Second, a perhaps more subtle point. To vary the input level to the amplifier under test, I adjust the HP 355 C/D step attenuator. The signal generators are set to provide +15 dBm (each tone) out of the combiner when the net effect of the ZHL-3 gain (approximately 25 dB) the 9 dB attenuator loss, the low pass filter loss and the 6 dB hybrid combing loss are all considered. This requires signal generator outputs between +4 and +5 dBm. (The ZHL-3A amplifiers operate at about 1 watt output power in this setup.) Why use this method instead of varying the signal generator levels to alter the DUT input level? Two reasons:

First, the simple one is that once the two tone levels are matched, they will stay matched and it's much faster and less error prone to just dial in different attenuation levels using the 355 C/D step attenuator.

The more important reason is that it allows is to distinguish between intermodulation  generated in the amplifier from that originating in the hybrid or in the low pass filters or in the ZHL-3A amplifiers or, for that matter, in the 8657A generators. A 1 dB change in attenuator level will result in a 1 dB change in measured 3rd order intermodulation product level if  the source of the 3rd order product is ahead of the step attenuator. If the intermodulation source is after the step attenuator, a 3 dB change in intermodulation product level will be seen for a 1 dB change in the step attenuator. From earlier measurements, I know the 10.7 MHz bandpass filter is quite clean (it uses powdered iron toroid cores) so a 3:1 change must result from IMD in either the amplifier or in the spectrum analyzer.

IMD products generated in the spectrum analyzer (unlikely given the beneficial effect upon dynamic range of the 10.7 MHz bandpass filter) can be identified by changing the spectrum analyzer's input attenuator. If changing the spectrum analyzer's attenuator by 10 dB causes no change in displayed signal level, the spectrum analyzer is not the source of the IMD. The  R3463 automatically adjusts the displayed signal level to compensate for input attenuation. Older spectrum analyzers would show a change in displayed level as input attenuation changes. Hence if changing the R3463's attenuation level causes a change in displayed signal level, the source of the IMD is within the R3463 and the attenuation should be increased until no change in signal level is seen as additional attenuation is applied.

Z10000-U Buffer Amplifier

Let's start with my Z10000-U buffer amplifier. The amplifier I tested is set for +14 dB net gain and operates with 12V on the AD8007, not the 9V as in a stock Z10000. Increasing  the supply voltage to 12V, the maximum permitted voltage, improves the intermodulation performance by a dB or two.

The resulting data is plotted below. It's similar to the ARRL's sample plot, but with some differences worthy of comment.

Second, note that all the low level IMD products fit the 3 dB out for 1 dB in change line almost perfectly. This means that none of the IMD products measured have their source in the test equipment.

Third, we see the onset of compression in the two tone outputs at around -2 dBm. Keep in mind that the input level is for each tone, but the test signal is two tones. Hence the amplifier actually sees 6 dB greater input than the X axis scale indicates. Since 1 dB compression is traditionally measured with a single tone input, the correct 1 dB compression point for the Z10000-U operating with 14 dB net gain, is actually around +4 dBm.

Fourth, what's going on with the IMD product around the same input level for compression onset? The answer is that the Z10000-U, like an audio op-amp design, is quite clean until the output reaches clipping. At clipping, the odd harmonic content and odd order intermodulation dramatically increases, as evidenced by the plot. (At gross overdrive, the output begins to resemble a square wave, which contains only odd order harmonics.)

A simple cross-check can be made, of course. Remember that the Z10000-U has a 49.9 ohm series resistor in the output for stability when driving capacitive loads such as coaxial cable. The tabular data shows the break point to be at +16 dBm PEP output, corresponding to 2.00 V RMS, or 5.6V peak-to-peak. The AD8007 has to output twice this voltage, of course, due to the 49.9 ohm series resistor at the amplifier output, corresponding to 11.2 volts PP. The supply voltage is 12V, so the AD8007 is swinging within ±0.4V of the supply rails, which correlates nicely.

Finally, note that the "third order intercept" or IP3 is a fictitious point, being the intersection of two extrapolated straight line fits  to the measured data. Although fictitious, the IP3 value does have  real meaning and can be useful in evaluating amplifiers and their relative performance.

However, it must always be remembered that these predicted interference levels are based on the extrapolated data. So long as  the input and output signals stay relatively close to the straight line fits, the predicted interference level will be accurate. But, at some level, there's a strong divergence from the straight line fit and the extrapolation, and hence predicted interference levels are no longer accurately predicted. In the case of the Z10000-U, and similar op-amp type devices, the divergence from linear prediction can be sharp, with as little as 1 dB difference making a huge difference in actual intermodulation product levels.

Let's look at a different amplifier, a Mini-Circuits ZFL-500LN. This is a small, low noise amplifier with nearly 30 dB gain at +15V, with 1 dB compression at 7.74 dBm output and IP3 of +14 dBm (unspecified input voltage, might be 12V.) according to the data sheet.

One point of difference worthy of note is that the 3rd order IMD products require measuring at the -90 dBm level. There are no indications of intermodulation from the test equipment at the -90 dBm level.

The data table shows the 1 dB compression point is +4.0 dBm output (single tone) or +10 dBm when PEP is considered. This is a bit above the +8 dBm specification.

Likewise, the measured IP3 is +22 dBm, considerably above the +14 dB specification. However the data sheet fails to provide the test voltage for the IP3 specification. Most data is provided for 12, 15 and 16 volts, but not IP3.

It's also worth noting the absolute signal levels involved here; the Z10040A 3rd order IMD product at 0 dBm input is around -70 dBm. That is much better than either the Z10000-U or the ZFL-500LN amplifiers. And 0 dBm is a very strong input signal for most conditions; I've seldom seen a total RMS input level stronger than -20 dBm with a log periodic antenna at my location near Washington DC.

At the -1 dB compression point, as a matter of fact, the Z10040A has almost 1 watt PEP output.

Mini-Circuits makes a line of MMIC (monolithic microwave integrated circuits) often used where a simple, reasonable performance gain block is needed. My Z90 panadapter uses the GaLi-74 as a matter of fact. As part of designing the Z90, I laid out a printed circuit board containing just a GaLi.

The Gali-74's details can be found at http://www.minicircuits.com/pdfs/GALI-74+.pdf from which I've extracted the more useful specifications for this discussion:

The data sheet figures are based upon feeding DC power to the device through an RF choke and resistor combination. I've used a simple design, as illustrated in the data sheet; DC power is supplied through a series resistor without an RF choke. The experimental Gali-74 design used for the IP3 tests has an effective 43 ohm resistor in parallel with the output, which reduces the gain by 4 dB to around 21 dB. It also will reduce IP3 and output power compared with the series choke feed.
 

It's also notable that the test equipment 3rd order IMD floor can be seen for low level inputs. At low input levels, the Gali-74's intermodulation products are lower than the amplified test equipment 3rd order intermodulation product ("upstream of the amplifier") and can be seen in the first two or three measurement points before the Gali-74's internally generated intermodulation products become stronger than those generated in the test equipment.

The test equipment intermodulation floor is thus around -135 dBm, measured at the output of the combiner / amplifier input.

I've also been working on a higher gain broadband preamplifier using the BFQ19 transistor. It's not a finished product but so far the results are promising. 3 dB bandwidth from 10 KHz to 100 MHz, 20 dB gain and +44 dBm IP3, as reflected in the data plot below.

Experience shows that the IP3 value of an amplifier changes with the DC supply voltage and device collector current. I looked at how the third order intercept of the Z10040A Norton amplifier changes with DC supply voltage over the range 9V to 16V, with the results shown below.

The collector current ranges from around 40 mA per transistor at 9V to 60 mA at 16V. I do not recommend operating the Z10040A at 16V for extended periods, particularly if the amplifier is to be used outdoors in a hot climate.

At very strong input levels, 0 and +5 dBm (these signal levels are way above the levels one normally experiences on an antenna unless located in the shadow of a transmitting station) there's a clear monotonic improvement as DC supply voltage increases. This is likely due to allowing the 2N5109 transistors to operate further from saturation, although at a point on their characteristic curves above the optimum linearity point, which is important for lower input levels.

My recommendation remains to operate the Z10040A at 13.8 V supply, or at least in the range 13 to 14.5 volts. In extraordinary cases, where the Z10040A is continuously exposed to very strong input signals, increasing the supply voltage to 15 to 15.5 V may be useful, but these circumstances should be relatively uncommon and the operating temperature of the 2N5109 devices should be carefully monitored for ensure operation is within the safe temperature range.