Ray Pengelly of Cree Inc. Research Triangle Park, presented a workshop talk at this year's IMS 2009 entitled, “GaN HEMT Technical Status: Transistors and MMICs for Military and Commercial Systems”. Cree’s standard 28 volt GaN on SiC HEMT COTS (0.45um gate length) process operates at 4 watts per millimeter of gate periphery at frequencies to 6 GHz while Foundry processes are capable of operating to 18 GHz. In addition, through the use of field-plated structures breakdown voltages have been substantially increased such that transistors can be operated at drain voltages over 50 volts. In the latter case power density is increased to over 8 watts per millimeter.
A 10W GaN device from Cree achieving record efficiencies in excess of 81% was also demonstrated at IMS using an active harmonic load-pull capability from Mesuro, a new measurement company with ties to Cardiff University and Tektronix. By de-embedding the small signal effects introduced by the package and extrinsic device parasitics from measured waveforms, designers were able establish the exact waveforms for optimum performance. As a result, the current and voltage peaks were directly associated with the knee- and pinch-off region of the device.
In the demonstration, a 10W GaN HFET was measured and harmonic impedances visually optimized to yield the theoretically optimum half-wave rectified sinusoidal current and squared voltage waveforms. The resulting efficiency of the device was measured to be 81.5%. Based on the recommended impedances the PA was designed and tested. The first pass design resulted in a power amp that was able to deliver 12W at 81.8% efficiency!
The complete solution developed in collaboration with Tektronix consists of 4 building blocks that can be separately purchased and represent a new class of measurement systems for enabling Waveform Engineering:
1. Tektronix instrumentation - consists of TDS Sampling Oscilloscope and AWG7122B Arbitrary Waveform Generator
2. RF test set - enables the broadband detection of all harmonics spectrum components up to 12GHz and power levels up to 100W
3. Active harmonic load and source -serves to generate any impedance at the device input and output port at the fundamental, second and third harmonic frequency
4. System software -performs absolute calibration for the accurate measurement of waveforms, PC control of the system, data display and analysis, and the conduct of automated measurements
We spoke with Ray about these impressive GaN amplifier results.
MWJ: Could you tell us about the measurement system?
Ray Pengelly: The active load pull is another method of achieving high efficiency results from so-called "waveform-engineered" amplifier design. The promise for active harmonic load pull is that it allows one to investigate design spaces that are outside the "normal" classes of operation such as A/B, C, D, E, F, inverse F etc. Where this could be particularly useful is for wideband amplifier designs (covering greater than 1 octave) for such applications as multiband telecommunications.
MWJ: Where did the modeling fail and the waveform measurements gain insight?
Ray Pengelly: The modeling did not fail but the waveform measurements allow actual device waveforms to be checked against modeling expectations
MWJ: What can be attributed to the significant increases in performance?
Ray Pengelly: Cree internally has produced Class E results with efficiencies close to 90% and there have been Class F and inverse Class F results in the mid 80%’s. The active harmonic load pull probably allows faster realization of specific waveform engineered amplifiers.
MWJ: Can these results be expected for other devices?
Ray Pengelly: Yes, provided the ft’s of the transistors are high enough for harmonic content to be significant. This is certainly true of present and next generation GaN devices.
To know more about the measurement system itself, we spoke with Wally Arceneaux of Mesuro and Darren McCarthy of Tektronix about the active load pull system used to optimize the device performance.
MWJ: What power levels are available (20w and 150 w systems) and why (the load driver amp)?
Arceneaux/McCarthy: 20 Watt and 150 Watt CW Systems are currently available and cover the power requirements for handset and basestation applications respectively for most wireless communication systems. The load amplifiers must be sufficiently broadband and linear to cover the fundamental frequency of the device under test. Since the harmonic power content is usually substantially below the fundamental power requirements, additional amplifier driver amps to drive the harmonic frequencies would require substantially less power.
Example: A 90W device with an optimum impedance of 8ohms and package leads with 11mm width would require less than 10W from the load PA when using a common broadband impedance transformer, which can be readily incorporated into a test-fixture design. The power requirement for the load PAs is actually mainly determined by the difference between the generated impedance and the impedance that is offered by the system and/or utilized test-fixture. Therefore, the active load pull does not necessarily require high-power load PAs [1].
In the Mesuro MB 20 (20W system), the driver amplifiers are included in the system. For the Mesuro MB150 (150W system), the amplifiers are purchased separate as often customers already have these expensive amplifiers, however the signal conditioning elements and system components are calibrated and rated to handle 150 Watt of CW power.
Higher power solutions can be delivered with a customized design.
MWJ: What is the Frequency range of the Active Load Pull system?
Arceneaux/McCarthy: The measurement frequency range of the RF test-set is currently limited from 500 MHz to 18 GHz due to some of the system components selected. Lower frequency (IF, baseband) measurements are an option using an IF test-set that can be multiplexed in with the RF system and goes down to approx 300 kHz. Higher frequency components are available as a custom solution, including sampling heads for the Tektronix DSA8200 that can cover from DC to 70GHz as standard product options.
On the stimulus side the frequency range is limited by the Tektronix AWG7122B, operating in two channel mode for source and load driver signals, to just less than 6 GHz. In single channel operation, the AWG7122B can operate to over 10 GHz. Additional sources can be optioned for source and load stimulus to cover higher frequency ranges through a custom design.
MWJ: How viewing the waveform provides extra insight
Arceneaux/McCarthy: To facilitate an efficient PA design it is important to identify for a given device the optimum set of fundamental and harmonic terminations. The problem is finding these impedances. At present, optimum impedances are identified through systematical sweeps where the fundamental impedance plane is scanned for a number of frequencies, drive levels, and bias conditions which can be time consuming. This process works fine in PA designs where the harmonic terminations can be ignored such as class-A or AB designs. Extending the same approach to more efficient PA modes such as class-F require the addition of the harmonic terminations into the systematic sweeps making it prohibitive.
For example: Assuming a grid of 100 impedances at the fundamental, second and third harmonic would lead to 1003 = 1 million measurements. Adding other essential sweep parameters such as input power, frequency, and bias would further extend the measurement time by orders of magnitude.
The measurement solution from Mesuro cuts down the characterization time through the simultaneous utilization of its waveform measurement and harmonic load pull capability; functionality that Mesuro terms as ‘waveform engineering’. The design approach is based on the fact that theoretical current and voltage waveforms for all amplifier modes of operation are well defined and provide an ideal goal for the PA designer. Now, it is possible to take the measured waveforms, compare these with the ideal ones, and initiate changes within the fundamental and harmonic impedances to achieve a better match. The measured waveforms and their similarity to a set of target waveforms give a PA designer a sense of direction and allow for the rapid identification of the correct set of fundamental and harmonic impedances; in case of waveform engineering these are actually the theoretically optimum waveforms and not some impedances at a local maximum.
This new design approach has been applied to two inverse class-F PA designs centered at 0.9 and 2.15 GHz. The chosen transistor was a commercially available 10W GaN device from Cree Inc. In a first step, optimum input and output bias conditions, fundamental impedance, the second and third harmonic impedances have been identified for an inverse class-F mode of operation utilizing the Mesuro system. Once, the optimum values have been obtained the design effort has shifted to replicate the identified impedance value on a circuit board designing appropriate input and output matching networks. The manufactured input and output boards were populated and mounted into a test-fixture to determine their s-parameters. If necessary the matching networks have been tuned to ensure a good agreement with the measured optimum impedances. After the board tuning the power amplifier was assembled and tested. The result was in both cases a first-pass success replicating gain, output power and efficiencies that have been obtained from the Mesuro system to almost the first decimal point when taking the losses of the input and output matching networks into account. Both realized PA designs achieved efficiencies above 80% over a relative bandwidth of 10-15%, resulting in absolute bandwidth of 100-200 MHz [2]. Similar design methodology has been applied to a class-J design using the same device resulting in a 60/60 performance, i.e. at least 60% over a 60% relative bandwidth [3].
MWJ: Compare the frequency based characterization (VNA) vs. sampling scope technology (time domain).
Arceneaux/McCarthy: There are two major measurement acquisition differences that need to be considered for VNA and sampling based solutions: frequency to time conversion, and lower frequency measurements.
The VNA is a frequency swept measurement acquisition tool. With a repetitive signal and precise phase correlated synthesis, the VNA can be calibrated to provide phase correlated information at different frequencies when measured at different points in time on a repetitive signal. This is an evolution required to extend the VNA into non-linear measurements. One must have the precision calibration tools to extend the base VNA functionality, and the signal must be repetitive.
The Tektronix DSA8200 sampling oscilloscope has a 14-bit A/D that time samples all frequency components from DC to an upper frequency limit of >70GHz. A time domain based instrument is a natural fit for calibrated harmonic measurements. The sampling scope modules are calibrated for magnitude and phase at the factory (traceable to NIST) and therefore require a simpler more traditional calibration process. All signal components; video baseband memory effects, fundamental current and voltage waveforms, and the higher order harmonics, are time-correlated by definition. There is no need for an additional 5th channel and a comb signal generator to establish the phase relationship between the spectral components, i.e. external calibration synthesis tools are needed to phase align the frequency components of the signal. They are all measured in the time-domain.
MWJ: What are the VNA limitations with terminations?
Arceneaux/McCarthy: The VNA is a 50 ohm measurement system that can be transitioned to a specific impedance when desired. However, to vary the impedance around the smith chart passive tuners are required. While passive tuners can cover much of the smith chart there are still areas which cannot be reached.
The active load pull solution allows for not only complete coverage of the smith chart but also offers negative impedances off the smith chart.
The system requirements for passive load-pull also have an effect on the receiver’s required dynamic range. During conditions of highly reflective loads at the device, the actual signal through the tuner can be quite low. This can require the VNA to need substantially higher dynamic range capability.
The active load pull solution does not require the use of tuners and places no components between the DUT and the reflectometers. This allows the dynamic range of the sampling scope to be adequate for most applications.
MWJ: How does open loop active load pull complement x-parameters, how they are different and how one replaces the other?
Arceneaux/McCarthy: This question might best be asked to compare waveform engineering with x-parameters. In their simplest form X-parameters are essentially a mathematical concept to represent waveforms. The formulation of X-parameters extends in a mathematically coherent manner the concept of s-parameters into the nonlinear domain by taking into account the nonlinear mixing terms that are produced by the device. The X-parameters are directly derived from measured waveforms with their robust mathematical definition allowing for the accurate interpolation of the measured waveforms on a finite measurement grid. The X-parameters have shown also a capability to correctly extrapolate beyond the measured grid; however, this is still an active research field. The exact waveform stimulus and measurements and the exact mathematical formulation that is used to calculate the X-parameters of a used device have not been disclosed until know. The resulting closed standard makes it difficult for the user to determine exactly or even estimate when the X-parameters are valid or not, e.g. over which fundamental and harmonic impedance regions are they valid or what is the modulation bandwidth for which they could be used?
The open-loop active load pull architecture complements X-parameters as it is capable of measuring waveforms for any combination of fundamental and harmonic terminations and optionally offers the characterization using modulated signal. These measurements can be directly imported into CAD and used if necessary in conjunction with an already available X-parameter model.
MWJ: What is the growing need for source pulling and why?
Arceneaux/McCarthy: More and more, modern designs of highly efficient power amplifier output stages involve a cascaded set of multiple amplifiers in their design. Doherty, Polar modulation, and Envelope Elimination Restoration amplifier stages are a few of these designs that involve multiple amplifiers in series or in parallel. Active load pull techniques give the designer an accurate way to define the impedances at the source side for the harmonic frequencies. This allows for easy emulation of these cascaded amplifier stages.
Metaphorically speaking, the problem is essentially how to find the ‘needle in a hay stack’ as systematic sweeps are just prohibitively time consuming due to the multi-dimensional solution space with an almost infinite number of permutations; please see also question 3 for more details.
In question 3 only two additional harmonics have been mentioned, however, the solution space is much larger than that, e.g. it could include also input harmonic terminations.
Example: To optimize the performance of the output amplifier stage, it is often needed to consider the non-linear optimization of the driver amplifier loading the input of the output stage amplifier. Characterization of this performance is commonly known as source pull. For output stage amplifier that is in a cascaded amplifier design, the true optimal design must consider a simultaneous source and load pull performance for optimized performance.
MWJ: How can the load pull information be used in design software?
Arceneaux/McCarthy: The measurement system is essentially a practical realization of a harmonic balance simulator in which the current and voltage spectra (magnitude and phase) are provided by the measurement system and not the device model.
Basing both the simulation and measurement domain on the same foundation allows to close the gap between both worlds and it becomes straightforward exchanging data. This is accomplished through by utilizing waveforms which can be readily imported into AWR MWO using the generic MDIF format. Within MWO the MDIF file can be associated with the new ‘Cardiff model’ for further use in circuit/system simulations.
The format of transferring waveforms across the different measurement and simulation platforms is completely open and transparent. More importantly the concept of current and voltage waveforms, which are used in transferring the data, is well known and readily understood.
If necessary, the designer can use the imported MDIF files for the generation and verification of other behavioral models, such as an X-parameter model.
References
[1] Z. Aboush, C. Jones, G. Knight, A. Sheikh, H. Lee, J. Lees, J. Benedikt, and P. J. Tasker, “High power active harmonic load-pull system for characterisation of high power 100Watt Transistors,” in 35th European Microwave Conference, Paris, France, October 4-6, 2005
[2] P. Wright, A. Sheikh, Ch. Roff, P. J. Tasker and J. Benedikt, “Highly Efficient Operation Modes in GaN Power Transistors Delivering Upwards of 81% Efficiency and 12W Output Power,” in 2008 IEEE MTT-S International Microwave Symposium Digest, Atlanta, Georgia, USA, June 15-20, 2008
[3] P. Wright, J. Lees, P. J. Tasker, J. Benedikt, S. C. Cripps, “An Efficient, Linear, Broadband Class-J-Mode PA Realised Using RF Waveform Engineering,” in 2009 IEEE MTT-S International Microwave Symposium Digest, Boston, Massachusetts, USA, June 7-12, 2009