Switch-Mode RF PAs Using Chireix Outphasing

Transcription

1 Switch-Mode RF PAs Using Chireix Outphasing (Simplified Theory and Practical Application Notes by Robin Wesson, Base Station System Architect, RF Power Innovation, NXP Semiconductors Mark van der Heijden, Senior Scientist, RF Power Innovation, NXP Semiconductors Demand for Energy Efficiency Drives Power Amplifier Innovation Rapidly increasing consumer wireless data usage is driving the requirement for increased mobile network capacity. Combined with rising energy and environmental costs, this is resulting in higher network operating costs. It is well known in the wireless infrastructure domain that the final stage RF power amplifier consumes a significant proportion of the total radio base station power budget. Further, studies have shown that base stations consume up to 90% of the power required by the entire mobile network, user equipment included. Perhaps not surprisingly then, the focus for improvements in infrastructure power efficiency has been the RF power amplifier (PA subsystem. Today, two-way LDMOS Doherty amplifiers are widely used in conjunction with digital pre-distortion (DPD to provide greater than 40% drain efficiency for the final stage of the power amplifier. The Doherty amplifier architecture is part of a wider class of load-modulation architectures. The Doherty architecture, first proposed in 1936, consists of two amplifiers that modulate each other s load depending on the required output power. There is another PA architecture that consists of two amplifiers, known as an outphasing amplifier. The simplest version of an outphasing architecture is known as a LINC amplifier (Linear Amplification using Non-linear Components. A LINC amplifier uses an isolating combiner with saturated amplifier stages to achieve high linearity with good peak efficiency. Due to the isolating combiner there is no load modulation effect in a LINC amplifier. With this amplifier, high peak to average (PAR signals suffer from reduced average efficiency since the individual amplifiers operate at a constant power even when the required signal output power is low. In the 1930 s Henri Chireix resolved this drawback by combining outphasing with a non-isolating combiner design to enable load modulation resulting in improved average efficiency for signals with amplitude modulation. Load Modulation Basics The basic concept of load modulation is a simple yet powerful one. In essence, two amplifiers share a common load which modifies the current and voltage waveforms within each device to maintain high efficiency for signals at less than peak output power. Much has been published on this topic, and it is beyond the scope of this discussion to go into the mathematics in detail. Fortunately, this detailed treatment is not necessary in order to demonstrate some of the key differences between Chireix and Doherty. For a more detailed but still practical examination of these concepts, the reader may wish to refer to Steve C. Cripps reference work RF Power Amplifiers for Wireless Communications (ISBN

2 Figure 1 illustrates how two trans-impedance amplifiers (as a typical field effect transistor behaves in its linear mode of operation can be used to apply RF currents into a single load. If both devices instantaneously provide the same current, these RF currents add in phase, doubling the voltage across the load. Impedance is defined as the ratio of the voltage and current waveforms. By this definition, the load impedance seen from one device is modified, or modulated, by the current from the other device. Of course, the physical load is unchanged, but the apparent load as defined by the relationship between the I,V waveforms seen at the output plane of the first active device is modified by the other. However, considering again the basic load modulation diagram in figure 1, it can be seen that with both devices conducting, the voltage at the load is increased and therefore the apparent load is increased instead of decreased. This is the opposite sense to what is required in our Doherty amplifier. Ideally, when more power is required, the load impedance of the main amplifier should decrease, allowing more current and hence more power to be supplied. This problem is solved in a Doherty amplifier by including an impedance inverter on the main device output, as shown in figure 2. With the inclusion of a quarter wave transmission line, or its lumped element equivalent, the operational sense of the load modulation is corrected. Device 1 Current from device 1, (i 1 Current from device 2, (i 2 Device 2 Main Stage Current from device 1, (i 1 Current from device 2, (i 2 Peaking Stage RF Load R When i 1 =i 2 (ie the two devices are the same size The effect of the load modulation is that both devices see the load resistance 2R V(load =R (i 1 + i 2 Device 1 sees only the current i 1 and the increased load voltage V=R (i 1 + i 2 R >> R(i 1 + i 2 / i 1 Reducing apparent load Quarter Wave impedance inverter Increasing apparent load RF Load R Figure 1: Basic load modulation concept Load modulation in the Doherty Architecture The Doherty amplifier is an architecture that makes use of this simple concept to deliver increased efficiency. In a Doherty amplifier the two devices are biased differently, such that at lower signal amplitudes only the main stage is operational. The load seen by the main amplifier at low power levels is R. As the amplitude increases, this stage moves into compression and its efficiency increases, as with a conventional class AB amplifier. Further increases in amplitude will begin to switch on the peaking amplifier which is biased typically in class C. This is the point at which load modulation begins. By peak power, both devices are being fully driven and the effective load for both main and peaking amplifiers is R/2. Figure 2: The Doherty arrangement for correcting the sense of the load modulation Load modulation in the Chireix Architecture Although Chireix and Doherty are both referred to as load modulation PA architectures, it is evident when comparing Figure 2 and Figure 3 that the load modulation is achieved in a very different way, and is based on a different set of assumptions about device behavior. In the Chireix arrangement, the load is modulated by controlling the phase of the voltages at the outputs of each active device. When the voltage waveforms are in phase, the voltage across the load will be zero and no current will flow. When the voltage waveforms are completely out of phase, the voltage across the load is double the voltage at each device output. 2

3 Device 1 Current (i 1 RF Load R Current (i 2 Device 2 Device 1 Current (i 1 Current (i 2 RF Load R Device 2 devices are assumed to act as voltage sources / ideal switches V 1 V(load =V 1 -V 2 Ii 1 I=Ii 2 I V 2 - jx + jx V 1 V(load V 2 =V 1 -V 2 a When V1 = V2, i=0, R= b When V1 = V2, i=2i, R=R/2 c Between these two extremes the load current is a function of both voltage waveforms, and has a phase shift as well as an intermediate impedance. Ii 1 I=Ii 2 I Figure 3: The Chireix representation of the load modulation concept. Several important differences between the two architectures can be inferred from these basic representations. a The Doherty architecture relies on the active devices acting as current sources. The Chireix architecture relies on the devices acting as voltage sources. b The currents in a Doherty PA are combined in a single ended load, whereas the voltages in a Chireix combiner are applied, conceptually, to a floating load. c The Doherty amplifier requires an impedance inverter to change the sense of the load modulation to achieve the desired goal of increased efficiency, whereas the load modulation in the Chireix amplifier is achieved purely with phase control of the output voltage waveforms. d In the Doherty amplifier the load impedance of the amplifier is modulated between R/2 and R, while that of the peaking amplifier is modulated between and R/2. In the Chireix, the range of load modulation for both amplifiers extends from R/2 to. e Since the current is a function of the phases of the two RF voltages driving the load, which are in turn a function of the outphasing angle, we can infer that there will be a reactive part to the load modulation which can be modified with the addition of compensating reactive elements as shown in figure 4. Figure 4: Chireix compensating reactances are used to modify the load modulation trajectory to enhance efficiency. After this brief examination we can distinguish between the two types of outphasing architecture. The LINC amplifier uses an isolating combiner which is fundamentally lossy at anything below peak power as the output power from Path 1 is nulling power from Path 2 through vector power cancellation. By contrast, in the Chireix amplifier a non-isolating combiner is used to enable load modulation, which preserves high efficiency at power back-off. At lower output powers the effective load impedance increases thereby reducing the dissipated power losses. Although LINC and Chireix differ in the detail of design, and to some extent in the principle of operation, there are many similarities between the two techniques. Like the LINC approach, Chireix outphasing involves the use of an additional signal processing step known as signal decomposition. This stage transforms a signal with both amplitude and phase modulation into two different phase modulated signals which are transmitted through the separate paths in the amplifier. A drawback of Chireix compared to LINC is that the non-isolating combiner reduces the linearity of the amplifier. Fortunately this is a familiar problem to the RF systems engineer, resolved in today s 3G and 4G base stations through the use of digital pre-distortion. 3

4 Outphasing places additional technical requirements on base station design. Similar to Doherty PAs, the Chireix outphasing concept uses two power transistors and a non-isolating power combiner to enable load modulation based efficiency enhancement. However, unlike the Doherty method, the signal required in each path is different. Thus, a base station designed for a Chireix outphasing amplifier must provide two complete transmitter paths for each PA. (It should be noted at this point that some Doherty designs also use dual path transmitters to provide ultimate control over the behavior of the PA. Further, the base station baseband design must incorporate the signal decomposition stage which increases the signal bandwidth in each path of the transmitter. This may require higher speed data converters and different reconstruction filter designs. The signal decomposition function creates constant envelope signals in each amplifier path, which enables high efficiency operation with amplifiers operated in compression. Whilst it is true that the Chireix PA arrangement has been known for nearly eighty years, this aspect of its operation in particular suggested that it warranted renewed consideration. The use of constant envelope signals through each path of the architecture not only allows the use of compressed amplifiers but also enables the use of Class D, E and F switched mode branch amplifiers. towards digital switched mode operation can be swift and pervasive as suitable processes become available. An RF transistor today, as typically used in a Doherty configuration, is biased slightly in the on-state if there is no input signal. There is a small current flowing that is called the quiescent current. The trans-impedance nature of a PA biased this way means that an input voltage signal causes a variation in output current. A range of power amplifier classes arise from this scenario. A PA that is biased to conduct 50% of its maximum current under quiescent conditions is called class A, and in normal operation this class will conduct current at all times. Since at all times the active device supports both a current and a voltage drop, there is significant power dissipation in a class A amplifier, and efficiency is correspondingly low. A PA that is biased at a lower quiescent current will only conduct over part of the input signal cycle. Class AB is defined as a linear amplifier which conducts over slightly more than half of the input signal cycle. Class B conducts over exactly half, and class C less than half. With all these classes, the phase relationship of the voltage and current waveforms has long been known to be critical for efficiency. Classes AB, B and C can be more efficient, but are less linear than the class A amplifier. The reduction in linearity has two effects. Firstly, a signal with amplitude variation will be distorted by the amplifier, and secondly, power will be wasted in unwanted harmonic emissions. Origins of Switch Mode PA Classes Switched Mode power amplifiers have existed for many years in non RF applications, and for nearly as long, they have been a topic of research for RF applications. The first class D audio amplifiers were developed in the 1960 s, and the class E RF PA was granted a patent in the mid 1970 s. Today, nearly every audio device uses switched mode amplifiers. Of course, there is a big difference between being able to operate a transistor as a switch for RF waveforms at 2 GHz compared to audio at a few tens of kilohertz, however the message from history is clear. Progress One could infer from these considerations that the most highly efficient amplifiers would have no requirement for linearity, would control the phase relationship of the voltage and current waveforms to minimize overlap, and would limit the amount of power in the harmonics. Research shows that these requirements can be fulfilled if ideal switches can be used as amplifiers. Ideal switches do not dissipate power in the on or off state, so automatically the efficiency is very high and voltage and current waveforms do not overlap. [1]. 4

5 Innovation in Switched Mode Chireix PA s One problem arising from the requirement to keep the current and voltage waveforms orthogonal is that the phases of the I,V waveforms are strongly influenced by the reactive components in the power amplifier. If a circuit provides an ideal phase relationship at a particular frequency, this relationship and therefore the efficiency - will be degraded at another frequency. To solve this, NXP has developed and patented a new driver technology to control the phase relationship across a wide frequency range. This technique enables the Class E amplifier to achieve optimum efficiency across a proportional frequency range of around 25%. [3] Figure 5: Example current and voltage waveforms characteristic of Class-E operation In a practical implementation, switch-mode amplifiers use field-effect power transistors driven into saturation so that they operate as non-ideal switches. By avoiding the linear mode of transistor operation as used in typical Class-AB circuit structures, switchmode or digital amplifiers have less overlap between voltage and current waveforms and therefore naturally achieve relatively lower thermal dissipation and higher electrical efficiency. It is immediately apparent that, by using transistors as switches, there is no easy way of generating an amplitude modulated signal at the PA s output. As we have seen, a more complex architecture like outphasing, using two switch-mode amplifiers is needed to enable output signal amplitude variations. Outphasing provides a method for controlling a pair of switch-mode RF amplifiers in such fashion as to enable them to operate efficiently with high crest factor modulated signals, whilst providing the power efficiency merits (opex and CO 2 reduction that Mobile Network Operators want to see in next generation radio base station transmit subsystems. To achieve this, NXP is combining Chireix Outphasing with GaN HEMT Class-E switch-mode transistor amplifiers. The class-e amplifiers used by NXP in the Chireix Outphasing system have been made possible in part by the development of GaN RF power transistors which have lower parasitic capacitance than LDMOS and have higher maximum frequency of operation. These devices can be driven into saturation more quickly and with less loss than a typical LDMOS device as used today. The ongoing development and commercialization of GaN RF transistor technology is enabling operation in these new classes at frequencies of interest to base station design engineers. With these innovations, NXP s Chireix outphasing RF PAs can combine high power efficiency with a wide range of carrier frequency reconfigurability. Reconfigurability is a commercial advantage for wireless infrastructure OEMs and the mobile network operators they serve, as it allows them to have one PA subsystem in stock that can be reconfigured to meet the carrier frequency requirements of base sites operating across a number of bands. In addition to reducing the number of PA variants, reconfigurability brings another potential benefit. In future generation cognitive radio wireless network deployments, carrier frequencies may be switched dynamically, increasing the commercial value of this unique feature of the NXP Chireix outphasing switchmode RF PA design. 5

6 Although reconfigurable mobile networks may be some years from deployment, they represent a significant field of industrial and academic research, and many papers already exist discussing the benefits that the reconfigurable base station may bring in the future. Developing reconfigurable hardware is one of the key challenges in realizing the benefits extolled. R In FPGA implementations, this transformation function can be performed using a CORDIC implementation, as shown in figure 9. The two separate steps of coordinate transformation are required to modify the conventional in-phase and quadrature information channels into, and back from, the magnitude and phase form. This facilitates the implementation of the phase adaptation functions used to encode the amplitude of the signal in the phase domain. R Σ Next in the chain the baseband signals are upconverted typically using high speed digital to analog converters in conjunction with a standard IQ modulator as found in base station designs today. The key difference with an Outphasing base station is of course that four DACs and two IQ modulators are required for a single transmitter. STAGE 1: CMOS predriver with reconfigurability control function STAGE 2: High voltage CMOS driver STAGE 3: Class E GaN Final stage with Chirex Combiner Figure 6: The three stage PA line up forming the NXP outphasing PA module Outphasing Transmitter Architecture Whilst at first glance an outphasing amplifier seems to require a completely different architecture for the base station transmitter, closer inspection reveals that most of the line up and signal processing blocks are unchanged. Figure 7 outlines one architecture suitable for an outphasing base station for multicarrier 2G, 3G or 4G mobile communications signals. As in a base station today, the in-phase and quadrature (I and Q digital signals from the digital baseband processor pass through the typical Digital Front End (DFE processing algorithms, including filtering, interpolation, crest factor reduction (CFR and digital pre-distortion (DPD. An outphasing-specific signal processing block, signal decomposition, is next in the TX signal chain. The signal decomposition block is also known as the signal splitter or signal component separator. Finally the two phase modulated RF signals are amplified and conditioned in a way that would be recognizable to a base station designer today, and the signals enter the two input - one output Chireix PA module at a power level in the range of 0 to +10 dbm, depending on partitioning choices. Although it is a potential drawback of the outphasing architecture that it requires two separate transmit chains per power amplifier, simulations show that the channel to channel isolation requirements for this architecture are significantly relaxed compared to the requirements for separate channels in a base station today. Broadly, this is due to the nature of the cross channel interference. Signals from one transmit chain appearing at the output of another channel can be considered spurious emissions, which lead to isolation specifications of the order of 80 db. In the case of two channels serving a single power amplifier, the requirement is purely a functional one. This distinction is an important one when considering the ideal architecture for an outphasing transmitter. It is anticipated that the reduced channel to channel isolation specification will enable use of quad DAC s, dual path IQ modulators and VGA amplifiers. 6

7 A greater level of integration, as well as eliminating the requirement for a separate screening cavity for each path, will enable these additional transmit chains to be implemented with a reduced impact on cost and PCB real estate. phase modulation component. The common-phase modulation carries the phase information of the original signal. The differential phase modulation contains the information about the original amplitude modulation. The textbook version of Outphasing is to use an inverse cosine function to convert the instantaneous signal amplitude into an outphasing angle. This is easier to visualize with the aid of few equations and a diagram. The input and output signals of the digital signal decomposition function are given by: j( t + (t S IN (t = A(te j( t + (t + (t S 1 (t = e Figure 7: Architecture of the Chireix Outphasing Switch-Mode RF PA portion of NXP s Prototype Design Signal Decomposition As mentioned, 3G and 4G wireless air interfaces are both Amplitude and Phase Modulated to achieve high downlink data rates. Also previously discussed, switch-mode PAs are either "off" or "on", and are not capable of preserving AM information. The signal decomposition function is the key to understanding how outphasing overcomes this challenge. Various academic papers have been published on this topic, but we present the basics here. In short, Chireix outphasing converts the AM part of the complex waveform into a phase shift between the two transmit branches. After the RF power has been amplified by the two final stages of the amplifier, the Chireix power combiner restores the original AM information. All types of outphasing amplifier require a signal splitting or signal decomposition function. This is typically done in the digital domain. In the signal splitting operation, the signal is transformed from one that is both amplitude and phase modulated into two signals with constant amplitude that have a common phase modulation component and a differential j( t + (t - (t S 2 (t = e (where (t is the time varying Outphasing angle There is no time varying amplitude component in the signals S 1 and S. The two constant envelope signals 2 when recombined recreate the original signal with both amplitude and phase information. In practical implementations the outphasing angle must be calculated for every sample within the waveform. The outphasing angle is then added to the phase of the original signal in one path, and subtracted from the phase of the original signal in the other. It can be intuitively seen that when the outphasing angle is zero, the two paths combine in-phase, creating an output signal with maximum magnitude. When the outphasing angle is ninety degrees, the two vectors theoretically cancel and the resulting magnitude will be zero. The continuum of outphasing angles from zero to ninety degrees provides the mechanism for recreating the amplitude content of a complex signal whilst enabling the individual amplifiers to operate at maximum efficiency at all times. Figure 8 illustrates the recombination of the two outphasing signals at an instant in time. 7

8 The basic outphasing function described works unit circle Q (t S 1 (t vector magnitude = 1 (t = - A(t MAX S IN (t vector magnitude = 1.31 S 2 (t vector magnitude = 1 well for a LINC outphasing system with isolating combiners. The phase to amplitude transfer function, as seen as a solid line in figure 10, is symmetrical. However, when testing a real Chireix outphasing amplifier it quickly becomes evident that the phase to amplitude transfer function is modified with use of the efficiency enhancing Chireix combiner. complex plane example τ = 40 Figure 8: Arrangement of Input signal vector and two outphased vectors for an outphasing angle of 40 degrees Figure 10: Comparing the ideal mathematical outphasing function with the function required for Chireix. Figure 9: representation of a CORDIC based FPGA outphasing arrangement. Outphasing Functions, from Ideal to Real Rather than calculating the resultant magnitude from an outphasing angle, when starting with a waveform, the engineer must calculate the outphasing angle required for a given sample magnitude. In a pleasing application of high school trigonometry, we can show that the effect of the outphasing angle theta is to reduce the magnitude to A = A max cos(. Turning this around, it is possible to calculate the required outphasing angle theta from the desired instantaneous signal amplitude A. Textbook Outphasing Function Outphasing angle (t = acos ( A (t A Max Several aspects of Chireix Outphasing signal decomposition are useful to consider. Firstly, the true Chireix outphasing function is asymmetrical, which is not the case for the ideal arccosine based outphasing function. The asymmetry in the phase to magnitude transfer function implies that there is a right and a wrong sense in which to apply outphasing. An outphasing angle applied positively to path one and negatively to path two will generate a different magnitude than if it were applied with the opposite sense. This asymmetry is evident from the shape of the function, in that the angular distance from peak to null, expressed in degrees of outphasing angle, is different going one way compared to the other. The engineer will find, when working with an arccosine function on a Chireix PA, that the null is not ninety degrees from the peak, and the effective dynamic range of the amplifier is reduced unless new outphasing functions are developed. 8

9 Figure 11: Simulation results, using ideal power devices, illustrating the good and bad sense of Chireix load modulation. It can be seen from simulations of efficiency versus phase that the desired load modulation effect (which is the key reason to use a Chireix combiner is working optimally only on one side of the outphasing function. The long side with the extended phase range from peak to null is the side on which the efficiency enhancement is occurring. Operating a Chireix amplifier on the short side of its phase versus magnitude trajectory results in very poor efficiency since the phases of the signals will be causing load modulation in the opposite sense to that required. Designing an optimal outphasing decomposition function requires first understanding the phase range expansion and the efficiency enhancement mechanisms to ensure that the architecture achieves its full promise. The Chireix Power Combiner We have seen how the signal decomposition function is vital in building an amplifier that can deliver the full promise of the Chireix architecture. The same is true for the all-important power combiner. The combiner design process is complicated by the fact that it must satisfy a diverse set of requirements including: The combiner must be low loss, to maintain the efficiency achieved through implementing the new architecture. a It must have a wide operating bandwidth to accommodate the bandwidth expansion of the outphasing transformation and still achieve a wide frequency band of operation. b The combiner must be non-isolating to allow for the load modulation efficiency enhancement. c The combiner must perform the balanced load to single ended load transformation that is needed for conventional RF antenna structures. d The combiner must provide the right harmonic terminations to both amplifiers to operate at high efficiency. e The combiner must incorporate the Chireix reactance corrections required to keep the load impedances seen by the device real at the chosen back off point of operation. f In a typical common source amplifier configuration, the combiner may also supply the bias current for each device drain terminal. Figures 2 and 3 compare the load modulation concepts as applied to a Doherty and a Chireix amplifier, and showed how the Chireix amplifier had a theoretical range of load modulation from R/2 to infinity. In normal operation, especially with high peak to average signals, neither extreme state is visited often. The PA spends the majority of its time operating neither fully out-phased or fully in-phased. In these cases, the phase of the current though the load is influenced by both devices. Since the current through the load is the current through each device, we can see that in normal operation each device will be subject to non-orthogonal current and voltage waveforms. This, we have discussed, is a key source of inefficiency in amplifiers of all classes. A practical solution, and an integral part of Chireix s original recipe in 1936, is to select an operating point where the PA should be most efficient, and add compensating reactances to optimize the load impedance seen by each device at this point. In todays basestation, this would usually be around the 8dB power back-off point associated with 3G peak to average power ratios. 9

10 Figure 13, based on a simple circuit model of the Chireix combiner, shows the complex load trajectory for each amplifier path across outphasing angle. The magnitude of the load impedance increases to infinity as the outphasing angle applied to each path in opposing directions increases from zero to 90 degrees. However the reactive part of the load is not optimal for efficient operation of each amplifier. Correcting for the reactive part of the load with a shunt capacitance on the advancing phase path and a shunt inductance on the retarding phase path ensures that the efficiency is maximized at the chosen back off point. These device terminations are a key part of the Chireix combiner design as shown in figure 12. [2], [4] Vdd 1 Device 1 Device 2 + jx - jx Vdd 2 RF Load Figure 12: Simple representation of a wideband dual device Chireix combiner showing the additional requirements for bias network, Chireix compensation and balun. Due to the number of sometimes conflicting requirements placed on the Chireix combiner, in practice it can be difficult to successfully implement. NXP has developed several circuit innovations to maximize the performance of the power combiner, and continues to develop and optimize its own outphasing demonstration PA s. Figure 13: Outphasing Load Pull Contours with a purely resistive load (above and with compensating Chireix reactances (below: Red dots indicate the impedances seen by each path for an outphasing angle of degrees corresponding to normalised output magnitude of 0.4 / -8 db. Various arrangements for the Chireix combiner form part of the roadmap for the NXP development activity, including versions suitable for integration within a device package, and others suitable for use in a module with power devices, drivers and PCB combiner included. In NXP's current prototype PA design, the combiner is a broadband component, based on a transformer topology rather than a /4 transmission line topology. The transformer is built up using two asymmetric (l1 l2 coupled-line sections. Figure 14 illustrates this concept. 10

11 Figure 14: Duty-Cycle Controlled Class-E Chireix Outphasing Concept (Digital Chireix PA NXP's current switch-mode base station prototype power amplifier is designed to operate across a bandwidth from 1.8 to 2.2 GHz (approximately 25% of FMAX, and handle signals with a PAR up to 10 db, which can occur in multi-carrier UMTS and LTE systems. NXP s design goal is to maintain high average power efficiency (with a goal of greater than 50% over the full frequency band at an 8 to 10 db back-off power level. Figure 15: Measured drain efficiency versus power back-off showing 60% drain efficiency at 8dB back-off Calibration and DPD Key to Meeting Base station Specifications In any outphasing amplifier, amplitude and phase calibration is essential for meeting EVM and ACPR. In a Chireix amplifier using a non isolating combiner to achieve efficiency enhancements with load modulation, digital predistortion is also required to correct the nonlinearity arising from the interaction. The loaded quality factor QL of the PA output matching filter has a significant impact on drain efficiency. PA bandwidth trades off for back-off efficiency: a broadband PA needs a higher-order matching filter that provides a flat in-band power and efficiency response for the fundamental while providing rejection to reduce out-of-band emissions at harmonic frequencies. NXP's switch-mode outphasing PA design innovations include a broadband matching power combiner and a dedicated CMOS RF driver amplifier. Using NXP s patented driver concept it is possible to preserve high average efficiency at any operating frequency within the band of interest by adapting the pulse width with which the two amplifier final stages are driven. In this fashion, NXP's PA can be optimized for efficiency at 8 to 10 db backoff, over the full frequency range (1.8 to 2.2 GHz. Phase calibration can be applied directly from the baseband signal, however amplitude calibration is complicated by the use of switched mode amplifiers in each path of the system. These amplifiers are largely insensitive to adjustments to input drive level, which means that amplitude calibration must be achieved in another way. The simplest way is to make the drain voltages independently adjustable over a small range. This adjustment is a static adjustment to suit a particular operating frequency and is not dynamic. This means that the complexity associated with wideband power supply modulators that arises with envelope tracking systems is not required for the outphasing power supply. Many base station power supplies already have static power supply control, although finer resolution will typically be required. 11

12 An alternative way to achieve amplitude calibration has been employed in the NXP outphasing system which makes use of the patented duty cycle controlled driver technology to create a differential amplitude offset between the two paths of the PA. This approach bypasses the requirement for any modification of the PA power supply. The digital predistortion used for the outphasing system test bench is a multi-segment amplitudeamplitude and amplitude-phase polynomial based algorithm with no memory compensation. This algorithm achieves good performance. This result, combined with memory modeling tests have demonstrated that the outphasing system has low memory content compared to typical Doherty amplifiers for the same instantaneous signal bandwidth. including both amplitude and phase modulation. This approach has demonstrated improved drain efficiency compared to the Doherty amplifiers which represent the typical solution used in base stations today. Operation of active devices as switches places a range of different requirements on device technology. In addition to developing GaN processes and devices for conventional linear power amplifiers and Doherty designs, NXP is also optimizing GaN technology for switching applications. In addition to the different requirements for the active devices in the PA, some modifications are required to the architecture of a base station designed for Outphasing. Many of these topics have been highlighted in this paper to provide an informative starting point for any designer, architect or team leader considering an outphasing product development. NXP has been developing the class E Chireix Outphasing technology discussed in this article for more than four years in house, and for many years before that in collaboration with academic research partners. The Gallium Nitride High Electron Mobility Transistor technology which enables switch-mode operation at frequencies in the multi-gigahertz range also represents many years of research and development. Figure 16: Measured Spectral Output Power of a Class-E Outphasing Transmitter with and without Phase and Amplitude Offset Calibration for a WCDMA Signal at F0 = 2.14 GHz Conclusion Efficiency Across a Broad Frequency Range Switched mode power amplifier classes promise to deliver significant increases in efficiency for the critical final stage amplifier in future mobile communications infrastructure. However, these classes require additional measures to support WCDMA and LTE waveforms. Often overlooked because of the poor average efficiency of the simpler LINC variant, Chireix outphasing provides an architecture suitable for use with class E amplifiers in conjunction with load modulation to maintain high efficiency at back off, providing compatibility with complex waveforms Figure 17: The Class-E GaN Chireix Outphasing PA Module 12

13 Acknowledgements References The authors gratefully acknowledge the contributions of the following NXP innovators who assisted in the development of this technology and the preparation of this white paper. Rik Jos Innovation Manager, BL RF Power Mustafa Acar RFIC Design Specialist / Senior Scientist, RF ADT Melina Apostolidou RFIC Design Specialist / Senior Scientist, RF ADT Further, we acknowledge David Calvillo and Leo de Vreede from Delft University of Technology for their contributions [2,4] [1] M. Acar, A.J. Annema, and B. Nauta, Generalized Analytical Design Equations for Variable Slope Class-E Power Amplifiers, Proc. IEEE ICECS, pp , Dec [2] M.P. van der Heijden, M. Acar, J.S. Vromans, and D.A. Calvillo-Cortes A 19W High-Efficiency Wide-Band CMOS-GaN Class-E Chireix RF Outphasing Power Amplifier, in IEEE MTT-S Digest, June [3] M. Acar, M.P. van der Heijden, and D.M.W. Leenaerts, 0.75 Watt and 5 Watt Drivers in Standard 65nm CMOS Technology for High Power RF Applications, in IEEE RFIC Digest, June [4] David A. Calvillo-Cortes, Mark P. van der Heijden, Mustafa Acar, Michel de Langen, Fred van Rijs, Robin Wesson, and Leo C. N. de Vreede, A Package- Integrated Chireix Outphasing RF Switch-Mode High-Power Amplifier submitted for IEEE Transaction on MTT-S NXP Semiconductors N.V. All rights reserved. Reproduction in whole or in part is prohibited without the prior written consent of the copyright owner. The information presented in this document does not form part of any quotation or contract, is believed to be accurate and reliable and may be changed without notice. No liability will be accepted by the publisher for any consequence of its use. Publication thereof does not convey nor imply any license under patent- or other industrial or intellectual property rights. Date of release: May 2013 Document order number: Printed in the Netherlands