Tracking Advances In High-Power GaN HEMTs
Improvements in solid-state power amplifiers depend on advances in transistors. Fortunately, evolving gallium nitride (GaN) high-electron-mobility-transistor (HEMT) technology is bringing many benefits to high-frequency amplifier designers. A key advantage of GaN HEMT devices over other transistor technologies is the high power density possible from relatively small transistor cells. For example, HEMT devices from Cree are capable of operating at RF power densities as high as 8 W/mm of gate periphery due to the superior thermal properties of their silicon carbide substrates. Such power density has made possible a line of 120-W power transistors for a wide range of commercial and military applications.
One member of the transistor line, model CGH40120F, consists of a single unmatched GaN HEMT device in a small, industry-standard ceramic-metal package. It provides 120 W saturated output power for general- purpose military and industrial applications including digital video broadcast (DVB), homeland security, tactical communications, and radar systems. The transistor has been demonstrated in various amplifier applications, including a reference amplifier with 1200-to-1400-MHz instantaneous bandwidth, 100 W CW typical output power, 16-dB typical small-signal gain, and 75-percent typical power-added efficiency over the full band. Addi- tional amplifier designs operating at +28 VDC have been demonstrated from 800 to 1300 MHz with 90-W output power and 65-percent efficiency for tactical data links as well as amplifiers for 1450-MHz DVB with 40-W average power, 19-dB gain, and backed-off-power efficiency of 40 percent when used with 16 QAM orthogonal-frequency-division-multiplex (OFDM) signals.
As examples of impedance-matched transistors in the GaN HEMT family, models CGH21120F and CGH25120F consist of single, input impedance-matched GaN HEMT devices with 120 W saturated output power housed in small, industry-standard ceramic-metal packages. Suitable for modulated signal amplification in W-CDMA, LTE, and WiMAX systems, the devices feature convenient matching to 50-Ohm environments over 30-percent instantaneous bandwidths. The CGH21120F is designed for 1800 to 2300 MHz while the CGH25120F is optimized for 2300 to 2700 MHz. Demonstration amplifiers have been developed for each transistor. Model CGH21120F achieves 110 W peak CW power at 70-percent efficiency and 16-dB gain. With W-CDMA 3GPP signals, it provides 20 W average power and 35-percent efficiency under Class AB operation.
The same 28.8-mm-gate-width device die are used in all of these new GaN transistors, based on a 0.72-mm unit cell, which also forms the basic building block for the company's large-signal models. The model for this die is scaled by a factor of 40:1. For such scaling to be effective, the measured and modeled data for the unit cell must be in close agreement. By developing an accurate and scalable large-signal model at the cell level, scaling will provide meaningful results in the design of much larger power transistors.
Another critical aspect of power transistor modeling involves package modeling. Cree has developed a physically derived modeling approach that includes package interconnection parasitic effects. The approach is based on S-parameter measurements of package elements, electromagnetic (EM) package simulations, and quasi-static wire-bond models. Since the reference planes for such package models are usually defined at the package body, any subsequent circuit modeling must take into account any physical distances between printed-circuit-board (PCB) traces and the package body. The effect of any ground-plane discontinuities should also be taken into account when building the model. This becomes particularly important when using large power transistors with low input and output impedances at frequencies greater than 2 GHz.
The general design technique for all of these high-power amplifiers is to first load-pull the transistor model within a harmonic-balance simulator such as Microwave Office from AWR. When performing source and load pulling of a device for modeling, it is essential that it remains stable in this unmatched environment in order to provide valid data. The source and load impedances of the device form the basis for initial circuit design.1 Once the initial circuits are synthesized, a complete amplifier is simulated, optimized, and refined. Finally a completely model-driven layout with all required electromagnetic (EM) blocks is used to generate a PCB layout.
A number of CGH40120F-based amplifiers have been designed using a common PCB approach. A single frequency match to the transistor's source and load impedances was determined using Smith chart matching to provide a starting point for the optimization of the input and output matching networks.1 The networks were then optimized to give better than 20-dB return loss across the desired bandwidth. This was achieved by allowing the capacitors to "slide" along the fixed transmission lines with varying values as required. No more than six elements were optimized simultaneously to ensure that the optimizer could rapidly converge to a solution. Once the initial networks were designed, the complete amplifier was optimized to achieve the desired performance goals. Any optimized capacitor values were reset to standard values to ensure that the amplifier could be assembled using standard capacitors.
Figure 1 shows full circuit simulations for both the small-signal (a) and large-signal (b) performance of an 800-to-1300-MHz amplifier design. Amplifiers for telecommunications applications based on the CGH21120F and the CGH25120F devices were designed using a distributed matching technique with transmission lines to implement the input and output networks. This approach was used since these designs operate at higher frequencies, albeit at slightly reduced bandwidths, where the parasitic effects of via inductance and capacitor resonances are much more difficult to accurately model.The amplifier designs for the two transistors are able to cover many different applications due to the inherent high power bandwidth of these GaN HEMTs.
These amplifiers were also designed using layout-driven simulation. When designing high-quality-factor (high-Q), effectively "narrowband" amplifiers with GaN HEMTs, it is important to perform sufficient stability analyses to prove that any designs will not oscillate once assembled. Typically, there are three key techniques (as apparent in the layouts) used to stabilize GaN HEMT devices in narrow-band amplifier designs: the use of a high-pass, low-frequency stabilization network in series with the input match, a series resistive gate feed, or series stabilization resistance in the RF input path close to the device.
These designs were optimized for maximum peak power and efficiency while maintaining high gain at typical average power levels. Since these amplifiers must meet stringent linearity requirements (high peak-to-average-power ratios), digital predistortion was employed to achieve the required linearity without compromising efficiency at backed-off output-power levels. Although amplifier linearity may not be a key parameter for some designs, careful attention was paid to minimizing absolute distortion levels. This ensures that any commercially available digital predistortion solution may be applied and provide compliance to spectral masks and error-vector-magnitude (EVM) requirements.
Continued on page 2
Page Title
One important design aspect for wide bandwidth linear systems is to ensure that the bias feeds to the amplifier provide low impedance, resonant-free responses at video frequencies.2-4 Video bandwidth is broadly defined as including frequency spectrum to three times the signal bandwidth. When dealing with instantaneous signal bandwidths of 20 MHz, the 60-MHz video bandwidth becomes a challenging design task.5Figure 2 shows the difference in response between a simple quarter-wave feed and a foreshortened feed. It is important to note that although the foreshortened feed has better response at video frequencies, it does not provide optimal performance at microwave frequencies. It follows that the foreshortened feed becomes an integral part of the output network design for these amplifiers. Figures 3(a) and 3(b) and Figs. 4(a) and 4(b) show final simulated performance for amplifiers designed around the CGH21120F and the CGH25120F HEMTs.
To validate the quality of the models, the performance parameters of various amplifiers based on the CGH21120F and the CGH25120F devices were measured, with those results in all cases agreeing closely with the simulated parameters. Figure 5, for example, shows the instantaneous broadband power performance of the CGH40120F transistor in an amplifier designed for use from 800 to 1300 MHz. The results indicate outstanding power-added efficiency (PAE) of better than 60 percent across an almost 50-percent bandwidth.
Another amplifier developed with the CGH40120F transistor was optimized for DVB applications at 1.45 GHz. This transmission format features high peak-to-average power ratio with quadrature amplitude modulation (QAM), requiring good dynamic performance from a power amplifier. The DVB amplifier developed with the CGH40120F delivers 40 W average output power with 16-state QAM (16 QAM) OFDM signals. At this power level, the amplifier's power-added efficiency was 40 percent and the adjacent-channel power (ACP) was -33 dBc (Fig. 6).
The measured results for an amplifier built with the CGH21120F transistor are shown in Figs. 7(a) and 7(b) and Figs. 8(a) and 8(b) under pulsed conditions and with digital predistortion for single-carrier and multicarrier applications, respectively. The amplifier was designed for operation under W-CDMA modulation conditions in the UMTS frequency band of 2.11 to 2.17 GHz. For the single-carrier measurements, at 25-W average output power, the amplifier uses digital predistortion for greater than 25-dB correction with efficiency of 40 percent, with that level of correction maintained under four-carrier excitation. The 20-MHz-wide multicarrier signal places great emphasis on the video bandwidth of the amplifier.5
Under more stringent WiMAX modulation, the CGH25120F-based amplifier also performs admirably with digital predistortion (Fig. 9). For these measurements, the signal personality used in conjunction with the Optichron DPD system is fully compliant to 802.16e WiMAX with a peak-to-average power ratio of 11.5 dB. At 20 W average output power, the amplifier corrects beyond the -45 dBc spectral emissions mask (SEM) point at 1.5 MHz removed from the carrier edge (6.5 MHz from the center on this 10 MHz signal). At this average power level, the efficiency is 30 percent. Finally, Figure 10 shows the results for digital predistortion correction at other frequencies for amplifiers based on both the CGH21120F-TB and CGH25120F-TB transistors.
This new family of high-power GaN HEMT devices provides high performance from 800 to 2700 MHz whether with CW, pulsed, or modulated signals. A key to effective amplifier design with these transistors is the accuracy of the device models, and inclusion of accurate package models.
References
1. Steve C. Cripps, RF Power Amplifiers for Wireless Communications, second edition, Artech House, Norwood, MA, 2006.
2. Peter H. Aaen, Jaime A. Pl, and Constantine A. Blanais, "On the Development of CAD Techniques Suitable for the Design of High-Power RF Transistors," IEEE Transactions on Microwave Theory and Techniques, Vol. 53, No. 10, October 2005.
3. Antoine Rabany, "Memory Effect Reduction for LDMOS Bias Circuits," Microwave Journal, Vol. 53, No. 2, February 2003.
4. Ahmad Khanifar, Nikolai Maslennikov, and Bill Vassilakis, "Bias Circuit Topologies for Minimization of RF Amplifier Memory Effects," 33rd European Microwave Conference, Munich, 2003.
5. Marco Franco, Allan Guida, Allen Katz, and Peter Herczfeld, "Minimization of Bias-Induced Memory Effects in UHF Radio Frequency High Power Amplifiers with Broadband Signals," Radio and Wireless Symposium, 2007.