Improving Efficiency, Output Power with 802.11a Out-Phasing PAs
Gord Rabjohn and Jim Wight, IceFyre Semiconductor
Jan 09, 2004 (9:30 AM)
Figure 1: Phasor diagrams illustrating how constant voltage signals, alpha (α) and beta (β), combine to form a variable voltage signal, R.
The high power consumption and limited performance of traditional 802.11a orthogonal frequency division multiplexing (OFDM) systems has delayed the adoption of 802.11a and dual-band WLAN products. The physics involved in handling the multi-carrier waveforms such as OFDM and wideband CDMA (W-CDMA) waveforms fundamentally limits the efficiency, output power, and signal quality delivered by linear power amplifiers, especially those powering traditional 802.11a systems. For 802.11a-based systems to deliver their full performance potential in the widest variety of WLAN-enabled devices, including power limited, small form factor devices, a completely new modem architecture and PA design is required.
The 802.11a standard is based on OFDM modulation, in which the data is multiplexed among 52 carriers, each of which can be modulated with BPSK, QPSK, 16QAM, or 64QAM. This spreading improves immunity to multipath fading and certain forms of interference. A disadvantage of this modulation scheme is that the resulting RF signal has large peak-to-average power ratio. Furthermore, the high-level modulation schemes require amplification with minimum distortion to avoid increasing error vector magnitude (EVM).
There is traditionally a complex trade-off between output power (and therefore range), data rate, and power consumption. To achieve high data rate, excellent linearity is required, and this is usually achieved by backing-off a class AB power amplifier, resulting in lower transmitted power. Lower transmitted power results in a poorer link budget, and therefore less operating range. Higher power, and therefore range, can be achieved, but only at the expense of data rate or battery life. In other words, users would like low power requirement, high data rate, and good range, but with linear class AB amplifiers handling signal amplification duties, they can simultaneously achieve only two of the three.
Traditional class AB amplifiers can be made quite efficient when operating at their peak power (theoretical efficiency of 78.5%), but their efficiency drops quickly at lower envelope powers. When such an amplifier is used with 802.11a OFDM signals, the amplifier must be sized to handle the peak power levels, but operates, on average, 8dB below peak, and therefore operates at very low efficiency most of the time, with an average efficiency in the neighborhood of 10%. Even lower efficiency will be seen when the amplifier is backed off to support 54 Mbit/s data rates.
What's needed is a technique that will allow an amplifier to operate at its peak power and therefore peak efficiency most of the time. The answer to the problem lies in out-phasing the PA. Let's see how the out-phasing architecture is constructed and the impact it has on the development of 802.11a power amplifiers.
Linear amplification using non-linear components (LINC)1, also known as the out-phasing amplifier technique, offers an alternative for WLAN designers that can provide excellent efficiency over a broad range of output powers. In an out-phasing amplifier, two signals of constant amplitude but of varying phase (these signals are referred to as "phasor fragments") are amplified in two separate amplifiers (these amplifiers are referred to as "branch amplifiers") and combined to create a single signal of varying phase and amplitude. When these phasor fragments are in-phase, the envelope power is at a maximum. When they are out-of-phase, envelope power is at a minimum.
Figure 1 shows on vector diagrams how two constant voltage, variable phase signals, alpha (α) and beta (β) can be combined to produce a signal of arbitrary voltage, R. Figure 2 shows the architecture of a power amplifier employing the out-phasing technique.
Figure 2: Diagram of a typical out-phasing amplifier architecture.
Since the branch amplifiers are constantly operating at optimum, maximum swing, each amplifier is always operating at peak efficiency. If the combiner provides isolation between the amplifiers, the resulting system provides poor efficiency because of loss in the combiner. However, if a low-loss combiner (that cannot provide isolation) is used, the overall system can be very efficient.
A particular variant of the out-phasing amplifier technique, known as the Chireix technique,2 employs a passive combiner. This combiner imposes a load impedance on the branch amplifiers that varies with the envelope, so that the branch amplifiers are driving a high impedance load when low output power is required. This swinging impedance forces the amplifiers to draw less current when less RF power is required, allowing high efficiency to be maintained at back-off. Note that the branch amplifiers are operating at a constant voltage swing at the output, but the varying output impedance causes the current swing, and therefore DC current requirements, to change.
When using an out-phasing technique, the choice of branch amplifier is important. A class F amplifier is particularly well suited for operation in this mode. Class F amplifiers are not linear, but as the branch amplifiers in an out-phasing amplifier system operate at a fixed amplitude, this is not important. Class F amplifiers employ specific terminations at the second and third harmonic to minimize the voltage across the amplifier transistor when it is "on", thereby reducing power lost in the switching device. The peak output power from such an amplifier is proportional to the square of the drain voltage, so the Vdd supply is used to set the average output power. The Vdd can be set so that, for any average output power, the amplifier is operating at the highest possible efficiency.
Practical Realization of a 5GHz
A 5 GHz out-phasing amplifier has been developed using a pair of class "F" amplifiers fabricated on a GaAs die. In order to achieve a high efficiency class "F" amplifier, the active device must look like an ideal switch; that is, it must have minimal "on" resistance, low capacitance, and it must switch quickly from the "on" state to the "off" state. In addition, the device must support enough voltage to allow adequate power to be generated without large impedance transformations that would cause unnecessary complications to the harmonic terminations and combiner. In a 50-ohm system that is expected to produce over 1 W, load lines dictate that each amplifier must produce 5 V root mean square (RMS), or 15 V peak-to-peak. The switch must withstand peak voltage excursions well above the peak-to-peak voltage. This combination of requirements is best met with a 0.5-μm GaAs PHEMT. Power PHEMTs are available that can support over 17 V, and have an Fmax that approaches 100 GHz.
The branch amplifiers are fabricated in pairs on a GaAs monolithic IC along with the driver stage and biasing circuitry. The amplifier die does not include the combiner or termination circuitry that follows the final device. The combiner requires low-loss transmission lines that operate at 5 GHz and the class F amplifier requires low loss terminations at 10 and 15 GHz. These components cannot be implemented in a low-loss manner on the GaAs chip, so they are implemented using a combination of well-controlled wirebonds and passive elements implemented on the ceramic base of the module. The resulting module measures 8 x 8 mm and is fabricated on 15-mil alumina with a thick-film like process.
As with all amplifiers, out-phasing power amplifiers generate some distortion. This distortion arises primarily from AM-to-AM conversion (gain compression) as well as from AM-to-PM conversion, and results in an increase in the error vector magnitude (EVM) of the modulation constellation as well as an increase in out-of-band emissions.
Predistortion is one approach that can compensate for these distortions by providing an enhanced magnitude and a corrective phase for large amplitude signals. In practice, adaptive predistortion is required in order to achieve the accuracy needed to reduce the EVM and out-of-band emissions.
The adaptive predistorter compares the desired transmitter signal (before digital-to-analog conversion, up-conversion and power amplification) with a down-converted and digitized version of the actual transmitted signal. Differences between the two signals are used to update the complex coefficients of the predistorter look up table.
For applications such as 802.11a WLAN, the EVM of an out-phasing PA can be reduced to around -30 dB by incorporating adaptive predistortion. As well, the adjacent channel emission level of an out-phasing power amplifier can be reduced to levels approaching -60 dBc with the use of adaptive predistortion.
In order for the out-phasing amplifier to work, a system that generates the constant envelope phasor fragment signals is required. Any arbitrary signal can be decomposed into phasor fragments. This has been difficult in the past, but modern DSP techniques make it practical, even for complex OFDM signals. For example, a single-chip physical layer (PHY) IC has been developed that generates phasor fragments which, when amplified by the out-phasing PA, produce a fully compliant 802.11a signal. Simply, even though phasor fragments are used, the resulting output is an interoperable 802.11a signal.
Figure 3 shows a plot of output power and output stage supply current versus the phase angle of the signals driving the amplifiers. (This plot is measured data from an amplifier operating at 5.25 GHz, with a Vdd of 5 V) Recall that the amplifiers are driven heavily into saturation, and are therefore operating at a constant voltage amplitude, but note that the supply current depends heavily on the out-phasing angle. This demonstrates that indeed the impedance that each amplifier sees (looking into the combiner) increases at lower output powers, and that the supply current drops.
Figure 3: Plot of output power and output stage supply current versus phase angle performance.
Exactly how much supply current drops can be seen in Figure 4, which shows the measured drain efficiency of the out-phasing power amplifier at various output powers. It also shows the maximum theoretical efficiency of an ideal class B and ideal class A amplifiers. A practical class AB amplifier will fall between the ideal class A and ideal class B curves. Note that the practical, measured efficiency of the out-phasing power amplifier is better than a theoretically perfect class B amplifier. At full power, where the amplifiers are operating fully in-phase, a drain efficiency of about 80% is observed. As the phase of the signals into the amplifier is reduced, the output power drops, but the efficiency drops much more slowly than a typical class AB amplifier. At a power 7.8-dB backed-off from peak (which is the typical peak-to-average ratio of an 802.11a signal), the amplifier demonstrates a drain efficiency of 46%.
Figure 4: Comparison of amplifier drain efficiencies.
The driver stage contributes to the total power dissipation. The power-added efficiency (PAE) with the driver stage included is over 33% at the 7.8-dB back-off point. This backed-off efficiency can be achieved over a wide range of power supplies. Figure 5 shows the efficiency of the amplifier (when backed-off by 7.8dB) over a range of power supply voltages. Note that there are really two ways to control the instantaneous output power of the amplifier: through out-phasing and by varying the power supply voltage. The power supply voltage is generally used to vary the average output power slowly, and the phase angle is used to provide the quickly varying instantaneous envelope of the signal.
Figure 5: Efficiency and output power of the out-phasing power amplifier over a range of supply voltages.
Figure 6 shows that the power amplifier achieves very high backed-off efficiency over a broad range of output power levels. Contrasting this behavior with that of traditional class AB amplifiers whose efficiency drops off rapidly outside an optimized operating point, the out-phasing power amp maintains a superior PAE/power consumption ratio over a broad range of output power levels.
Figure 6: Efficiency of the out-phasing power amplifier over a range of output power levels.
Traditional amplifier architectures limit the range and efficiency of WLAN systems. We have demonstrated that the Chireix architecture, when implemented with very low-loss combiners, offers advantages well suited to the 802.11a WLAN standard. This implementation has demonstrated unprecedented 80% peak drain efficiency, and over 33% average power added efficiency on real OFDM signals. The baseband processor that generates the correct signals to drive the Chireix amplifier is practical and adds minimal complexity to the transmit architecture. This innovation will make 802.11a WLANs robust and energy efficient in portable and low power applications.
- Zhang, X, L. Larson, P. Asbeck, Design of Linear RF Outphasing Power Amplifiers, Boston: Artech House, 2003.
- Chireix, H., "High Power Outphasing Modulation," Proc IRE, Vol. 23, No. 11, Nov. 1935, pp.1370-1392.
About the Authors
Gord Rabjohn is manager of the power amplifier group at IceFyre Semiconductor. He holds a B. A. Sc. from the University of Waterloo and a M. Eng. degree from Carleton University. Gord can be reached at email@example.com.
Jim Wight is the principal architect and a co-founder of IceFyre Semiconductor. He is also a Full Professor at Carleton University where he has led research into radio systems and technologies since 1976. Jim can be reached at firstname.lastname@example.org.
Copyright © 2003 CMP Media, LLC | Privacy Statement