A 1.9 GHz Adaptive Feedforward Power Amplifier

A relatively low power feedforward amplifier for 1.9 GHz has been developed to demonstrate the feasibility of the adaptive feedforward linearization technique. The saturated two-tone output power of this amplifier is approximately 4 W. The measured third-order intermodulation is -18 dBc before the linearization loop is closed; after the loop is closed, this value is suppressed to -45 dBc. Therefore, the amplifier's intermodulation cancellation performance is greater than 25 dB.

Qian Cheng, Chen Yiyuan and Zhu Xiaowei
State Key Laboratory of Millimeter Waves, Southeast University
Nanjing, People's Republic of China

Linear power amplifiers frequently are used to amplify multicarrier analog signals or nonconstant envelope digital modulation signals. For these signals with high peak-to-average ratios, nonlinearities lead to intermodulation distortions or spectrum regeneration. This regeneration of power outside the transmit channel produces interference in other channels. In addition, in-band distortion products generated by this phenomenon cause errors in the symbol vectors, thus degrading digital modulation accuracy.

Two-tone, third-order intermodulation measurements usually are conducted to evaluate a power amplifier's linearity performance. The typical value of third-order intermodulation distortions for many microwave high power class-A amplifiers when operated near saturation is approximately -22 dBc or worse. These class-A amplifiers must be operated backed off 7 to 10 dB from their 1 dB compression point to achieve more stringent specifications on the order of -40 to -55 dBc, leading to higher cost and much lower DC efficiency. In some cases, output power back off alone is unacceptable. Because of these disadvantages, many communication systems exclude the use of class-A power amplifiers.

Class-AB amplifiers, especially those using laterally diffused metal oxide semiconductor transistors, are more DC efficient compared to class-A amplifiers. The intermodulation performance of these amplifiers usually is good when they are operated backed off 3 dB or more, making them good candidates for applications requiring moderate intermodulation performance. When the intermodulation or spectrum regrowth specifications are tougher, some linearization techniques must be implemented.

Feedforward linearization techniques are promising in the implementation of high power linear amplifiers yet satisfy the low cost and high DC efficiency requirements. Several designs with a variety of complexity have been developed. In this article, a 1.9 GHz power amplifier is presented with a brief analysis and design procedures. Experimental results to show the effectiveness of this linearization technique are given. The third-order intermodulation distortion cancellation is greater than 25 dB. Furthermore, an adaptive control system has been used to guarantee performance in the presence of working environmental variations, such as temperature changes and input frequency and input power level variations. Finally, some design factors important to the amplifier's overall performance are discussed.

Feedforward Linearization System Design

Figure 1 shows a block diagram of the feedforward linearization system with a two-tone spectrum used to illustrate the principle of operation. The system consists primarily of two cancellation loops: The main loop is adjusted to cancel the carrier signals and provides a sample of the distortion products to the error amplifier. The error loop amplifies these distortion products, adjusts their amplitude and phase, and injects them into the output coupler to cancel the distortion products generated by the main power amplifier.

Usually, the main and error amplifiers' output power capabilities are similar. Although a class-AB amplifier may be a better choice for the main amplifier because of its good trade-off between linearity and DC efficiency (backed off 3 dB or more), two identical ready-made class-A GaAs MESFET amplifiers have been chosen for convenience and quick installation of the demonstration prototype system. The main amplifier is operated near saturation with a 1 dB compression output power of approximately 38.4 dBm (single tone). The error amplifier is operated class A to eliminate any extra distortion products. Therefore, it is important to guarantee a sufficient cancellation of relatively high level carrier signals in the main loop to ensure that the residual carrier signals do not overload the error amplifier in the error loop. Preferably, the residual carrier signals should be weaker than the distortion products.

A vector modulator has been inserted in each loop to adjust the carrier signal (or distortion) amplitude and phase. Several vector modulator circuits, including in-phase and quadrature modulators using Schottky or PIN diodes,6 were tested and a separate variable attenuator and voltage-controlled phase shifter were chosen. The attenuator is implemented using four PIN diodes in a p-type configuration.7 The attenuation is voltage controlled from -30 to -3.5 dB with a monotonic insertion phase-shift vs. attenuation characteristic over the full interested frequency band. The voltage-controlled phase shifter is designed using varactors configured in a p -type lowpass filter, as shown in Figure 2 . To suppress extra distortions generated by the varactors' nonlinearities, all varactors are partially coupled. Attention has been taken to avoid any in-band spurious resonance caused by the components' lead inductance and the PCB's parasitic reactance, which will lead to a sharp insertion phase shift vs. frequency characteristic and an uncontrollable insertion phase shift vs. control voltage characteristic. The phase shifter provides approximately 20° tuning range with a little insertion loss ripple over an 1800 to 1900 MHz frequency range. A cascade of two such phase shifters buffered by a 3 dB fixed attenuator produces a 40° tuning range.

Because the phase shifter's insertion loss is maintained as constant as possible with respect to the control voltage, cross-coupling or interaction between the amplitude and phase adjustment in each loop has been greatly weakened. This condition ensures the stability of the loops' operation, especially in the adaptive control system described in this article.

Adaptive Control Circuits

Feedforward linearization operation is based on the vector subtraction of two signals of equal amplitude and, therefore, is sensitive to amplitude and phase imbalance. In a practical system, temperature changes, channel frequency changes, input power level variations or component changes due to environmental effects and aging all can degrade performance significantly.5 Therefore, it is necessary to employ some adaptive control techniques.

Several adaptive control techniques have been evaluated. One technique uses a frequency-swept pilot tone and synchronous detector to ascertain the average pilot tone power remaining within the final output. Then, in response to this detection, various component parameters are adjusted to reduce the amount of pilot tone at the output.3,4

A simpler, low cost approach was chosen1 where a cascade of a quadrature branch-line hybrid and Wilkinson in-phase combiner is used in the main loop for the signal combiner instead of a typical Wilkinson combiner. Four matched detectors are placed at the two input ports and one output port of the branch-line hybrid and the output port of the Wilkinson combiner. It can be shown1 that proper linear combination of these power samples at various points in the circuit generates two error signals representing amplitude and phase imbalance, respectively. The linear combination of the square-law detectors' output voltages is generated by a properly designed operational amplifier circuit. These error signals then adjust the variable attenuator and voltage-controlled phase shifter in a feedback manner to ensure cancellation of the carrier signals.

Figure 3 shows a diagram of the nulling combiner. Four matched detectors(not shown) are placed at ports 1, 2, 3 and 4. Each detector is composed of a Schottky diode and lowpass filter (parallel capacitor and resistor combination), and is coupled to a nulling combiner through a 150 W surface-mounted resistor (size 0805). Compared to the microstrip line coupler, these resistors are more compact, coupling factors are more adjustable and degradation of the combiner's port SWR is not significant. Figure 4 shows the adaptive control circuit (with some modifications compared to a configuration described previously).1 Matching of the four detectors is achieved by adjusting the potentiometers. Four emitter followers act as buffers between the detectors and the control circuit. All parts are integrated on a single PCB constructed of epoxy with a dielectric constant of 4.3 and a height of 0.8 mm.

The system can only be considered half-adaptive since, in this case, the error loop is an open loop. Fortunately, because no element of the error loop is operated in its nonlinear range, signal conditions, such as input power level variations, have little effect on cancellation performance. The main consideration is temperature changes, especially in the class-A biased-error power amplifier with high DC power dissipation. Approximately a 2 dB amplitude change and a 10° phase change were observed within a temperature range of -10° to +50°C. Some compensation measures are necessary to properly work in a wide temperature range. A temperature compensation circuit comprising a microcontroller and temperature sensor is under development.

Experimental Results

A two-tone intermodulation measurement was conducted to evaluate the feasibility of the feedforward linearization technique. As stated previously, a class-AB amplifier is preferable (taking into account a good linearity/DC efficiency trade-off). However, two class-A amplifiers were chosen as the main and error amplifiers because of their ready availability. Each amplifier has a gain of 34 dB and a 1 dB compression output power of 38.4 dBm (single-tone test, 10 V/2.5 A). The transmission loss of the path between the main amplifier output and the system output connector is approximately 1 dB, including loss of the delay line, couplers and output isolator. Two random phase-related 1.9 GHz signals with a frequency separation of approximately 1 MHz are injected into the amplifier. In the experiment, the main amplifier is driven to an output power of 36.2 dBm (two tone).

Figure 5 shows the amplifier output (after sufficient attenuation) when the second loop is opened, representing the output spectrum of the main amplifier itself. The third-order intermodulation distortion is approximately -18 dB relative to the two carrier signals, showing that the main amplifier is operating near saturation. The spectrum at the main loop output (obtained from the coupler at the error amplifier output) when the adaptive controller is closed is shown in Figure 6 . In this case, the carrier signals have been suppressed more than 29 dB (weaker than the third-order intermodulation distortions). These distortion products then are phase and amplitude adjusted, amplified and, finally, injected into the output coupler. The output spectrum after cancellation is shown in Figure 7 . (The test point is the same as shown previously.) The third-order intermodulation distortion is approximately -45 dB relative to the carrier signal. Thus, the cancellation is greater than 25 dB. Comparing open- and closed-loop data, it can be shown that the carrier signals are barely affected by the loop.

Frequency bandwidth may be an important consideration in some cases. Two major factors affecting bandwidth are irregularities in the component's phase and amplitude response as a function of frequency, and delay mismatch in both loops.2 As the experimental results show, using components (such as couplers) with linear phase and flat amplitude vs. frequency characteristics and matching the electric signal length (or distortion) paths in each loop can broaden the bandwidth significantly. In this case, output power is reduced somewhat because of extra delay line loss.


Feedforward linearization techniques are effective in intermodulation distortion cancellation. A brief analysis and design procedures for a 1.9 GHz adaptive feedforward power amplifier have been presented. The experimental results demonstrate that this technique is promising. High DC efficiency and low cost make it attractive in modern digital microwave radio and multiple-carrier signal amplification applications. Some key factors affecting overall system performance, such as frequency bandwidth, stability vs. temperature, dynamic range and reliability, should be investigated further.


The authors wish to acknowledge China 863 High Tech Plan for its financial support of this research, and Hong Wei, Qian Fulin and Zhou Jianyi of SKLMMW, SEU.


1. Eid E. Eid and Fadhel M. Ghannouchi, "Adaptive Nulling Loop Control for 1.7 GHz Feedforward Linearization Systems," IEEE Transactions on Microwave Theory and Techniques, Vol. 45, No. 1, January 1997, pp. 83-86.

2. Sang-Gee Kang et al., "Analysis and Design of Feedforward Power Amplifier," IEEE MTT-S Digest, 1997.

3. Derek L. Tattersall, Jr. and James F. Long, "High Dynamic Range Modulation Independent Feedforward Amplifier Network," US patent 5,307,322, March 20, 1992.

4. Jack Powell et al., "Amplifier Having Feedfoward Cancellation," US patent 5,323,119, November 12, 1990.

5. Ashok K. Talwar, "Reduction of Noise and Distortion in Amplifiers Using Adaptive Cancellation," IEEE Transactions on Microwave Theory and Techniques, Vol. 42, No. 6, June 1994, pp. 1086-1087.

6. L.M. Delvin and B.J. Minnis, "A Versatile Vector Modulator Design for MMIC," 1990 MTT-S Digest, pp. 519-522.

7. Raymond W. Waugh, "A Low Cost Surface-mount PIN Diode p Attenuator," Microwave Journal, Vol. 35, No. 5, May 1992, pp. 280-284.

Qian Cheng received his MS degree in radio engineering from Southeast University in 1991. Currently, he is a project engineer at Southeast University's State Key Laboratory of Millimeter Waves. Cheng has developed several types of digital microwave radio front ends and frequency synthesizers for radar applications and is now responsible for developing a new generation of Ku-band very small aperture terminal transceivers and ultra high speed, broadband direct digital synthesis frequency synthesizers for radar environmental signal simulation with numerous modulation functions.

Chen Yiyuan received his MS degree in electromagnetic field and microwave technology from Nanjing Institute of Technology (currently Southeast University), China, and his PhD degree in electronics from the National Institute of Applied Science, France, in 1981 and 1986, respectively. Currently, he is a professor at Southeast University responsible for developing many microwave and millimeter components and systems. Yiyuan has also been also involved in numerical modeling of waveguide discontinuities, active filters and high temperature superconductor microwave filters and oscillators. He is the author or co-author of more than 30 technical papers and a member of the IEEE.

Zhu Xiaowei received his MS degree in electromagnetic field and microwave technology from Southeast University in 1996. Currently, he is an associate professor at State Key Laboratory of Millimeter Waves at Southeast University. Xiaowei's research interests include microwave and millimeter-wave components, circuits and systems, and RF front ends used in CDMA mobile communication systems.