Load pull for power devices

Updated March 31, 2006

This page is sponsored by our friends at Auriga Measurement Systems, a microwave test services company that also offers turnkey load-pull test equipment seamlessly integrated with Agilent sources and analyzers, pulsed I-V measurement equipment, noise parameter extraction systems, and associated software and support. Auriga has contributed their expertise in explaining how load-pull works, and provided us with some example measurements.

Click here to go to our main measurements page

Click here to visit our Smith chart page

Click here to go to our impedance tuner page

Click here to go to our page on efficiency of microwave devices

On this page you will learn how load pull measurements are done... but without an integrated system and software, you'll waste a lot of time trying to set up this measurement system yourself. Why do that when Auriga already can do the heavy lifting for you? Visit their shop which is conveniently near the Brewery Exchange in beautiful Lowell MA, or visit their web site, or give them a call at 978-441-1117 and tell them the Unknown Editor sent you!

Attention other forward-thinking microwave vendors: join our team and sponsor your own page on Microwaves101... find out details here!

Here is an index to the page you are on now:

Definitions of load pull versus source pull

Why load-pull a power device?

Load-pull bench block diagram

Calibration and verification procedures

Example 1: measuring a power transistor

Definitions of load pull versus source pull

Microwaves101 convention: When we're talking about the process of load pull or source pull as nouns (or adjectives like "load pull bench") we won't hyphenate. When we're telling you to load-pull or source-pull something (verbs), we'll hyphenate. Just so you know.

Load pull: the process of varying the impedance seen by the output of an active device to other than 50 ohms in order to measure performance parameters, in the simplest case, gain. In the case of a power device, a load pull power bench is used to evaluate large signal parameters such as compression characteristics, saturated power, efficiency and linearity as the output load is varied across the Smith chart.

Harmonic load pull: the process of varying the impedance at the output of a device, with separate control of the impedances at F0, 2F0, 3F0, etc. A very tricky measurement!

Source pull: the process of varying the impedance seen by the input of an active device to other than 50 ohms in order to measure performance parameters. In the case of a low noise device, source pull is used in a noise parameter extraction setup to evaluate how signal-to-noise ratio (noise figure) varies with source impedance.

In practice, power benches employ both load and source pull, and noise parameter setups employ both as well, but the emphasis is on one or the other. In power measurements, the input is source-pulled to a single location that provides good power gain, while the output is swept all over the Smith chart. In noise parameter extraction, the output is load-pulled to an impedance that provides good gain, then the input is swept all over the Smith chart.

Microwave impedance tuners are the "engines" that let you drive all over the Smith Chart.

Why load pull a power device?

The performance of an active device is a function of many things:

• Frequency
• Bias point
• Temperature
• Source/load impedance at fundamental frequency
• Load impedance at harmonic frequencies
• Power level

When you measure active devices on a network analyzer, you are looking at the small-signal response in a fifty-ohm system, as a function of frequency and bias point, and perhaps temperature if you are fortunate enough to have a temperature controller. Using linear CAD software you can accurately predict the small-signal response if the device sees impedances other than fifty ohms. It's more difficult to predict performance under large-signal conditions. Perhaps you can obtain a large signal model of your power device, or use Steve Cripps method for predicting saturated power performance. But there are limitations to each of these methods; large-signal models are notoriously inaccurate. This is where load pull comes in--it can be employed to empirically gather all of data you need to design a power amplifier and predict its large-signal responses, including compression characteristics, efficiency, harmonics and intermodulation products.

To review large-signal data, most of the plots that are of interest will be on Smith charts. Typically you examine data one frequency at a time, by plotting contours of constant output power, gain, efficiency, etc. The contours look like potatoes, a minor irritation if you are on a low-carb diet. Of course, you can also plot Pin versus Pout on Cartesian coordinates, and now you can do it at the "sweet spot" where power is maximized.

Load pull bench block diagram

Below is a simplified system block diagram that shows most of the necessary components of a load pull system that can measure both CW and two-tone (intermodulation) signals on a transistor. The DUT resides in the middle, surrounded by microwave tuners, then bias tees (this assumes that the tuners provide a DC path from one port to the other, which is not always the case). One component that is absent from the block diagram is an RF probe station (probes plus coax cables) which is used to make contact to the device. Note that the loss of all components between the DUT and the tuners will affect the maximum reflection coefficient to which you can load-pull, so you want to carefully choose low-loss components in this case.

A pair of tuners are used. The input-side tuner allows the source match to be "pulled" to an impedance where the device has appreciable gain, then it is generally left at this fixed impedance for all measurements on a device at a fixed frequency. The output-side tuner is the one that gets a work-out!

The test-set extender interfaces the DUT to a suite of measurement equipment that makes up the rest of the test gear (this is where Auriga's expertise shines). A signal generator (or perhaps two for two-tone measurements), and quite often a power amplifier, a coupling network and power meter, comprise the input network. The output interface of the test set extender may include a high-power attenuator depending on the available power of the transistor, then perhaps a power meter and/or spectrum analyzer. A network analyzer is also used to measure input/output response. All of the test gear inside of the test set is hooked up with electrically-controlled RF switches which are extremely repeatable. For convenience, system-controlled power supplies and DC current meters are part of the setup; without these you would have to calculate efficiency by hand, and who wants to do that for 1000 measurements that you can make in one hour?

Of course, the entire bench is controlled by easy-to-use software running on a dedicated computer. Once the setup is calibrated and the device installed, the operator interface is entirely by computer. Grab a donut and let's go!

Calibration and verification procedures

Calibrating the load pull station is far more complicated than a normal S-parameter measurement, and can take several hours. A vector network analyzer must be dedicated to the setup, because the S-parameters of all of the components must be be recorded, over the frequency band of interest. This is not that big a deal for the cables, bias tees, probes and the stuff behind the test-set extender, but for the tuners, hundreds of tuner positions must be measured and stored. No question, the tuners are the heart of the system, and the accuracy of the measurement is most affected by the repeatability of the tuners.

Calibrating a load pull system starts with some basic setup information: what frequency band (start/stop/step size?) What area of the Smith chart? What input power levels?

Once a load pull system is configured and calibrated, its performance should be verified. This can be accomplished by measuring something that is linear. A "thru" turns out to be a very good verification standard and by measuring it we can verify that we can make accurate measurements of device gain. Absolute power is verified by putting a power meter directly onto the output of the input tuner to verify input power at the DUT, and directly on the output of the DUT to verify output power. This is not practical for a device that is on-wafer, but demonstrating such verifications at the tuner's coax (or waveguide) interface shows that the methodology is sound.

The plot below show gain circles on power contours for "State 1" (a fifty-ohm system). This is the result of measuring a thru-line at a fixed power level. As expected, the maximum output power is obtained at fifty ohms (the center of the Smith Chart). the concentric circles surrounding fifty ohms show how power drops off with load impedance, here each circle represents a 0.5 dB drop in power. Click here to learn more about maximum power transfer!

The plot below shows all of the data that was taken to crunch out these curves. Taking this amount of data "by hand" is unimaginable!

Editor's warning: from here down, consider this page under construction!

This plot shows something to do with errors???

Show this linearly with error

These will show that we can measure gain versus tuner state with magnitude accuracy of +/- 0.06 with a few peaks greater than this. This is a typical calibration and can get slightly better is we increase network analyzer averaging, and use the best calibration techniques and temperature controlled lab.

Now make a measurement versus input power to verify the errors with this measurement.

Here we see the effect of noise as we go below measured output power of -30 dBm. The system can measure output powers greater than 45 dBm and this show the limits of the VNA dynamic range.

This plot shows and even lower output power and the signal to noise ratio of the receiver the network analyzer.

These will show that we can measure gain versus input power. So now we have confidence in the measurements of gain but to verify the absolute power we are in trouble for we are on wafer and do not have a on wafer power meter. What has been done is repeated coaxial calibration and power verification to prove the methodology utilized.

Once we have faith in the basics power measuring capably we can verify the two tone dynamic range.

We will now show a measurement of a thru using two tones, and discuss the system performance. The COI stays constant due to the system automatically controlling the power level presented to the SA mixer. The system corrects for any attention. This gives an instantaneous dynamic range of greater than 75 dB over a controlled power range of 60 dB. Here are the attenuator and amplifier block diagram details. As the measured output power increases the attenuation in front of the SA is increased to keep the mixer linear and at the power where it has the largest dynamic range. These measurements are on a thru and the power levels are limited to the power amplifiers available. System performance contuse as we see and we have and effective system TOI of greater that 65 dBm of course due to the nature of added attenuation as the power increases there is no real TOI.

We can measure 90 dBc with two tone of 5 watts each with proper spectrum analyzer and power amplifiers.

Spectrum power control block diagram

Example 1: measuring a power transistor

Unlike a network analyzer measurements, load pull allows you to make a lot of measurements at a single frequency, then move on to another frequency. You can vary input power, source match and load match. You might try different bias voltages while you are at it. These four variables will keep you busy for a while, before you change frequency.

Like fifty-ohm large-signal test stations, you can make swept power measurements at a single frequency. But now you can make swept power measurements at any frequency, at any of the reflection coefficient you previously calibrated.

Let's start by choosing a device. Here's a power transistor with RF probes down (describe...)

An evaluation of any active device should start with IV curves. Here they are:

After you choose a bias point, next it's time to measure device S-parameters versus frequency, so see how stable it is. From the S-parameters you need to carefully examine the stability of the device (maximum available or stable gain, K-factor, stability circles). You should do this for every bias condition you are considering (class AB, class A, etc.). Hey, this bad boy is unpotentially unstable (K<1) everywhere!

A typical measurement, a scattering of reflection coefficients are made on the input, with the output fixed at fifty ohms, and the input power somewhere near where you believe the output will be in 2 dB compression once the source and load match are correct. This will allow you to construct a contour plot to find a good input-side tuner location:

Show circles for 0.5 90 show conjugate match plot_two.all

Pin pout max gain problems of stability for max gain set device as close to simultaneous conjugate match as we can. Because it's a potentially unstable device this can not be done so we can just try S11 conjugate and S22 conjugate.

Pin pout tuned for power

Pin pout tuned for efficiency

Now show optimums

Contour for power and efficiency versus input power

Show the optimum moving with input power for both

Contour with both power and efficiency

Do a two tone measurement at both points?