Design Technologies for Silicon-Based High-Efficiency RF Power Amplifiers: A Brief Overview

1 Introduction
    Silicon-based single-chip transceiver products for GSM, WLANs, Bluetooth, and digital enhanced cordless telecommunications (DECT) have become commercially available in the past decade. Recently, WLAN/Bluetooth integrated circuit (IC) vendors have successfully integrated less demanding silicon-based monolithic PAs into their transceiver ICs. A critical challenge for wireless transmitter design for mobile products is how to maximize the power-added efficiency (PAE) of battery-operated devices. The peak and average PAE of an RF transmitter heavily impacts the size of the battery and heat dissipation as well as reliability, yield, and cost. On the other hand, many modern wireless transmission protocols, such as WCDMA, LTE, and WiMAX use non-constant envelope modulation with high peak-to-average power ratios (PAPRs) to be spectrally efficient. This causes rapid changes in magnitude/phase of the modulated signals. Therefore, RF PAs for these modern wireless communication systems require very high linearity. The mobile WiMAX (802.16e) standard, for example, has a 1-75 Mb/s gross data rate and up to 20 MHz signal bandwidth using a 64 QAM-1/2 modulation format with a 10-12 dB peak-to-average ratio (PAR). This standard requires an error vector magnitude (EVM) of below -26 dB (5%) with stringent transmission [1]. Future 4G LTE-A systems will require even larger bandwidths, with signal bandwidth extending to 100 MHz or possibly even greater.

    Numerous transmitter configurations have been developed for better PAE, high linearity, and broadband operation [1]-[25]. Switching-mode PA typologies (classes D, E, and S) can increase PAE considerably more than  typologies of traditional linear PAs (classes A, AB, B, and C) by operating devices as switches to minimize overlapping of current and voltage waveforms [26]. With relatively easier on-chip integration and high efficiency at RF, silicon-based class-E PAs are very attractive for use in mobile devices [2]. Because of their nonlinear characteristics, class-E PAs are quite suitable for constant envelope modulation schemes; however, they can also be employed in spectrally-efficient modulation systems that require non-constant envelope signals if appropriate linearization techniques are also used.

 
    It is desirable to design highly efficient silicon-based RF PAs for mobile broadband applications because they can be cheaper than III-V compound semiconductor PAs and can also achieve higher-level integration with good thermal conductivity. However, III-V compound semiconductor PAs still dominate the RF handset PA market. SiGe PAs and complementary metal oxide semiconductor (CMOS) PAs are very popular in the WLAN market and are becoming serious contenders for the Watt-level handset market. The ruggedness and reliability of silicon-based PAs needs to be carefully tested under different voltage standing wave ratio (VSWR) mismatched conditions at the output for handset applications (for example, 20:1 VSWR). The latest results suggest that standalone SiGe PAs may be very suitable for Watt-level 2G/2.5G handset PA applications. CMOS PAs have already been successfully demonstrated in GSM PA production.

    Another key issue is keeping the RF PA and transmitter efficient not only at the Watt-level peak output power but also at the back-off low-power regions. In this way, average PAE for the transmitter system is excellent. To achieve this goal, slow power supply modulation and/or transistor size switching can be used to increase PAE of an RF PA at low output power levels. However, there is a thornier problem to solve for mobile broadband handset applications. PAR in 3G/4G communications is inherently high, and the PA output can change rapidly (in μs) from 1-3.4 V. This makes the PA mostly inefficient if it is designed to operate with a peak PAE at the high supply level of 3.4 V. Therefore, for a most power-efficient transmitter (TX) system, a fast envelope modulator should be used to track the PA supply voltage in order to somehow modulate its output power and maximize PAE at both high and low output power (so that average PAE is maximized). We discuss this envelope tracking (ET) technique later.

    The following is a discussion on high-efficiency RF PA and transmitter design with GSM. A TX architecture with attractive PAE and enhanced linearity uses polar modulation of nonlinear PAs. The baseband signal is modulated in the amplitude/phase domain rather than the in-phase/quadrature (I/Q) domain. In the past, polar transmitters were mostly used for high-power base station applications to reduce heat dissipation. However, they have recently become widely used in wireless handset TX design for mass production because they are cheaper and significantly more efficient [3]. Recent research indicates that polar TXs using either envelope tracking (ET) or envelope elimination and restoration (EER) are capable of excellent system efficiency and linearity in several 3G/4G wireless applications. ET-based polar TXs often outperform EER-based polar TXs because of the relaxed requirements on the envelope-tracking amplifier bandwidth. ET-based polar TXs are also not sensitive to timing misalignment between the AM and RF paths [1]. Fig. 1 shows a simplified block diagram of an ET-based RF TX system. The signal bandwidth increases considerably from GSM/EDGE to W-CDMA/LTE, and the bandwidth requirement on the envelope modulator becomes too high and critical to the overall efficiency of the TX system for the EER system to be very attractive. Timing alignment requirements also make it significantly more difficult to meet the linearity specifications. Digital predistortion may be inevitable in this case. We have repeatedly shown that ET-based TX systems outperform EER-based systems in terms of enhanced data rate for GSM evolution (EDGE), WiMAX, WLAN, and LTE applications, especially when SiGe PAs are used [1], [3], [9], [16].

    Doherty PA topology [26] is also effective in satisfying the stringent linearity requirement for large output power (Watt level). The topology can be used to achieve high average efficiency. Doherty power combining has been frequently used in base station transmitter applications; however, because it has better linearity and efficiency, it has been seriously proposed for handset application as well. Recent research has shown that it can be integrated with lumped element using CMOS technology.

    A new mode of operation — class-J PAs [26] — shows the theoretical potential of obtaining linearity with as much broadband efficiency as conventional narrowband class-AB designs. However, class-J PAs do not require a band-limiting harmonic short [4]. Class-J PAs use the 2nd harmonic voltage to realize a phase shift between the output current and voltage waveforms and to provide the reactive termination for the 2nd harmonics. This allows the transmitter to achieve broadband behavior and cover a wide range of frequencies in order to satisfy modern multiband multistandard mobile communication applications.

    In section 2, we discuss two considerations when deriving the class-E operation mode, covering the mathematical derivations with a silicon PA design example. In section 3, we discuss envelope tracking (ET) — a very promising technique for enhancing efficiency and linearity in broadband mobile handset applications. In section 4, we discuss the design concept and recent research into Doherty PAs for handset applications. In section 5, we discuss class-J mode PA, and we conclude in section 6.

2 Class-E PA Operation
    Among conventional class-A, AB, B, C, and switching mode PAs, Si-based class-E PAs are quite suitable for portable devices. Its simple topology allows easier integration on-chip and high efficiency at RF. The definition of class-E PA operation proposed in [5] addresses three specific objectives for the collector voltage and current waveforms: 1) the transistor is off when voltage begins to rise across the transistor; 2) the collector voltage goes all the way back to zero when the transistor is turned on; and 3) the slope of the collector voltage should be zero when the transistor is turned on. According to these three conditions, several types of derivations have been reported [6], [7], [26].

2.1 Traditional Class-E Configuration
    The classical derivation introduced in [5] assumes that choke inductance is infinite; all the passive elements are ideal (with no parasitics) and the transistor is operated as an ideal switch; and the amplifier is 100% efficient. For a given Vcc, Pout, operating frequency, and 50% duty cycle, all the component values in Fig. 2(a) can be defined by (1) and (2) from Raab’s derivations:

    Co and Lo can be determined by a desired loaded Q value of the circuit.
Vs,max is the peak voltage of the collector/drain of the transistor (switch).

2.2 Class-E PA with Parallel Element Configuration
    For modern silicon-based IC design, a very large choke inductor is not suitable for integration on-chip because it occupies a large area and has a very low Q factor. Instead, small inductance or bondwire inductance of 1-2nH can be used for the choke inductor. In this case, the equations in section 2.1 become inaccurate. Therefore, a subclass-E PA (also called a parallel circuit class-E PA) design analysis and derivation is provided, as in [7]. Fig. 2(b) shows the simplified circuit diagram for this PA configuration. The difference in this configuration is that a finite choke inductance is used, but no Lx (Fig. 2(a)) is needed. For a given Vcc, Pout, and frequency, the component values in Fig. 2(b) can be defined by (3) and (4):

    Compared with the classical class-E configuration and derivations, subclass-E PA has the following advantages: 

• Small finite choke inductance is used and no additional reactive element is connected in series to the filter. This reduces the number and size of components to enable on-chip integration.
• The load value R in (3) is larger than that in (1) when calculated for this network. This reduces the impedance transformation ratio to the 50 Ω output.
• Theoretically, a slightly higher maximum peak switch voltage can be achieved in the subclass-E circuit, and slightly higher output power is available (provided that device breakdown is not an issue).

2.3 Highly Efficient Silicon-Based Monolithic Class-E PA Design 
    A highly efficient silicon-based monolithic class-E PA design was recently reported in [9]. Fig. 3 shows a simplified schematic for this monolithic one-stage class-E PA designed with IBM 7HP 0.18 μm BiCMOS SiGe technology. Simulation and measurement using a realistic simulation program with integrated circuit emphasis (SPICE) shows that, to achieve best PAE and Pout of -15-22 dBm, a  suitable size for the last-stage RF PA (based on a high breakdown device) can be close to 200-280 μm2.  This gives a simulated maximum collector current density of <0.5 mA/μm2, well within the measured HBT safe operation area (SOA). The parasitic effect of bondwire inductance at the emitter node to ground is a little more significant at 2.4 GHz compared to 900 MHz because there is a large increase in effective reactance to ground that acts as strong negative feedback for the bipolar junction transistor (BJT) (by a factor of 2.67). The problem can be avoided by adding more down-bonds to considerably improve the gain and PAE. Our recent study suggests that at least 4-5 down-bonds should be used in a 2.4 GHz SiGe monolithic RF PA design for 20+ dBm Pout. PAE is improved by 20% just by adding the down-bonds [10]. Using these design procedures, we achieved 65% peak PAE and 70% collector efficiency (CE) for a one-stage class-E SiGe RF PA. These excellent results were achieved without using any off-chip matching components; however, the radio frequency choke (RFC) is still kept off-chip to reduce loss. For a two-stage SiGe PA, close to 20 dB gain and 50-55% PAE have been achieved [2].

 
3 Envelope-Tracking-Based RF Polar Transmitter

3.1 Considerations for the System Design of a Polar Transmitter
    In practice, RF PAs for mobile applications present many technical challenges. Polar TX systems are known to be sensitive to timing mismatches between the AM and PM paths [12]. The group delay of the two signal paths must be matched to minimize PA distortion, which is difficult to control across all process-voltage-temperature (PVT) corners [26]. The other major obstacle is the larger bandwidth required for the circuits in the polar TX system for broadband wireless. The I/Q to polar transformation at baseband is a nonlinear operation, which inevitably expands the bandwidths of both the AM and PM output signals. Depending on the specific modulation scheme and system specifications, the constant-amplitude phase signal path for the EER-based polar TX may need roughly ten times larger bandwidth than the input signal in order to pass the TX mask requirement and/or the error-vector-magnitude (EVM) specifications [11]. However, these issues can be significantly addressed by using an ET-based polar TX architecture (also known as hybrid-EER or H-EER architecture), as shown in Fig.1 [1], [12]-[14].

    Compared to an EER-based large-signal polar TX system, an ET-based polar TX system has the following benefits [3]:

• Gain at low output power is higher because the ET-based PA can be operated near saturation but is not always fully saturated, as in EER.
• Sensitivity to timing mismatch (between the RF and amplitude paths) is lower than in EER [14], [15]. The RF path for ET-based TX contains the amplitude modulated signal as well; therefore, the signals in the RF and amplitude paths are very similar. This is not the case in EER.
• For the same linearity performance, ET has lower bandwidth requirement for the envelope amplifier design than in EER [12]. This can be critical because efficiency of the envelope amplifier can limit the overall composite PAE of an ET/EER system, as shown in (8). The higher bandwidth requirement in the amplitude modulator design for EER translates into lower modulator efficiency and lower overall TX system PAE if the PAE of the RF PA alone is kept the same for both ET and EER.
• When the amplitude of the input signal is fixed, it is difficult to keep the gain of an RF PA  constant while modulating its collector/drain voltage to adjust the output power, as in EER. Therefore, EER is inherently less linear than ET because its power gain inevitably decreases when the supply voltage is decreased, especially for bipolar devices. ET is superior in this regard because both the DC bias and input signal level can be adjusted to help keep the gain more constant [1]. The high-bandwidth RF limiter required for EER can be power hungry because it needs to operate at high RF.
• ET has less RF feed-forward signal that can appear as distortion in the TX output. Because the drive signal is hard-limited in EER, the RF feed-forward can cause significant distortions (AM-AM, AM-PM) by the large gate-drain or base-collector Miller capacitance in the final RF power device. These distortions couple to the output to cause linearity/EVM issues.

3.2 Design of Envelope Amplifier for ET-Based Polar Transmitter
    The overall efficiency of the ET-based polar TX system is the product of the envelope amplifier efficiency and the PA collector efficiency (CE), which is expressed by

Therefore, the design of a high-efficiency envelope amplifier is critical to overall system efficiency in a polar TX system using either the EER or ET. Because a wide envelope signal bandwidth above 20 MHz considerably reduces the efficiency of traditional switching direct current-direct current (DC-DC) converters, a linear-assisted switch-mode envelope ampli?er (or “split-band” envelope ampli?er) proposed in [11] can be used. This system helps attain wideband tracking and maintain high overall system TX efficiency. The simplified discrete circuit schematic is shown in Fig. 4.

    The linear-assisted envelope amplifier circuit (often called envelope modulator) has three different modes of operation [3], [11]:

 
    1) linear operation for small-signal envelope (small-signal operation). This occurs when the average slew rate of the switcher current is much larger than the average slew rate of load current. The buck converter can fully support the load current; that is, the switcher stage can provide both DC and AC components of the envelope signal.

    2) large-signal operation. This occurs when the average slew rate of the switcher current is much smaller than the average slew rate of the load current. The switcher stage can only provide the DC component of the envelope, and the AC component is provided by the much faster linear stage. The average switching frequency of the buck converter is almost the same as the signal frequency that the current sensing resistor can detect.

    3) matched slew-rate point. In this case, the average slew rate of the switcher current is the same as the average slew rate of the load current.

    Fig. 5 shows envelope amplifier efficiency for supply voltage at several load resistances with WiMAX 64 QAM modulated signal of 8.75 MHz.

    As the supply voltage decreases, the efficiency always increases. However, lower supply voltage can clip the output envelope waveform and cause more TX distortion. This suggests that if the clipping can be avoided by applying some decresting algorithms to reduce PAR, the supply voltage of the envelope amplifier can be lowered to considerably increase the envelope amplifier’s efficiency. The efficiency of the envelope amplifier also increases along with the reduction of its load resistance. For an RF PA output power of 23-28 dBm, the collector impedance seen by the envelope ampli?er can be less than 10 Ω, making the envelope amplifier’s efficiency 84% at VDD = 3.8 V.

3.3 Polar Transmitters Using ET Technique
    Recent research has shown successful ET-based polar TX system designs for modern high-PAR mobile applications such as WiMAX and 3GPP LTE [1], [8], [16], [18].

    Our recent work shows that the test bench setup for the polar TX system characterization is as shown in Fig. 6. Fig. 7(a) shows the measured TX output EVM, gain, and overall PAE of the entire ET-based polar TX system with a discrete envelope modulator and a monolithic PA applying the WiMAX 64 QAM 8.75 MHz signal. The overall system PAE is 30.5% at 17 dBm average output power, and the EVM is 4.4%. The output spectrum of our ET-based polar TX also passed the stringent WiMAX 64QAM mask defined by European Telecommunication Standards Institute (ETSI), as shown in   Fig. 7(b). With the ET scheme used, the SiGe PA can operate at its P2dB compression point (17 dBm) without violating WiMAX linearity specifications. However, the standalone PA (fixed Vcc without ET) needs 4-5 dB back-off to satisfy both EVM specifications and spectral mask (not shown here).

    The ET-based polar transmitter system designed in [18] performs well for 3GPP LTE application. Fig. 8(a) shows the TX output EVM, gain, and overall PAE of the entire ET-based polar system based on a 16QAM LTE 5 MHz modulation signal. The overall TX system PAE, including the monolithic envelope amplifier and the differential cascode PA, is 33.6% when the average output power is 21 dBm. The EVM is only 7.0%. These results, when compared with those of the standalone PA, show that the output of the ET-based PA can have lower EVM, as shown in Table 1.

Table 1. EVM and PAE of the SiGe PA before and after using ET with 16QAM LTE signals at

         1.42 GHz No predistortion applied [19].

    Fig. 8(b) compares the TX output spectra of the ET-based PA with the standalone fixed supply PA at an average output power of 21 dBm. The ET technique reduced the out-of-band emission of the TX spectrum (ACPR increased by ~10 dB at the 3.5 MHz offset frequency), allowing the polar TX to pass the LTE transmit emission mask. On the other hand, even though we intentionally increased Vcc to 4.2 V for the standalone fixed-supply PA, the PA still had very high TX out-of-band emission. Furthermore, the standalone fixed-supply PA was also tested under a 6 dB back-off mode from Pout  = 21 dBm (with only 15 dBm output power). As seen in Fig. 8(b), even with Pout = 15 dBm for the standalone fixed-supply PA, the output spectrum failed once again to pass the LTE transmit spectrum mask. This measurement suggests that ET operation not only keeps the PA efficient but also somehow linearizes it compared to the standalone fixed-supply PA. This is not the first time we have observed this linearization in ET operation; we have also seen this in common-emitter and cascode SiGe PAs. Studies are ongoing to see if similar effects can be observed in CMOS PAs as well.

4 Doherty Power Amplifier
    The Doherty amplifier was first proposed by W. H. Doherty in 1936 [19]. The efficiency and output power of an RF PA can be increased by using the Doherty power combining technique [26]. A Doherty PA consists of a main amplifier and a peaking amplifier. The output load is connected to the main amplifier through an impedance inverter that is usually a quarter-wave transmission line. The Doherty amplifier configuration is shown in Fig. 9(a). Assuming the output current of each PA is linearly proportional to the input voltage with harmonic short terminations, the efficiency of a Doherty amplifier can be analyzed using the fundamental and DC components only.

    In traditional Doherty operation, the peaking PA is turned on when the main PA, which is at half the maximum input voltage, begins to compress. A load modulation takes place for the PAs according to (6) and (7). Z main and Z peaking are the impedances seen by the output of the main PA and peaking PA, respectively. Im and Ip are the currents flowing through the two PAs. In typical operation, at the low power-delivering region, the peaking PA is off, and ZL is up-converted to the load of the main PA, which enables the linear main PA to be more efficient, as in Fig. 9(b). At the high power-delivering region, the peaking PA is turned on to combine the power with the main PA while also enhancing the linearity, as shown in Fig. 9(c).
      

    Doherty PAs have been used in base stations for many years, but it is difficult to integrate the transmission line on chip for handset applications. Voltage standing wave ratio (VSWR) mismatch for handsets, and inherent narrow-band operation of the impedance converter are serious barriers for Doherty PA in handsets. However, recent literature has reported impressive fully integrated CMOS Doherty PAs, as summarized in Table 2. In some of the works, lumped elements are used to realize the impedance inverter for the loading of the main PA.

5 Class-J Mode Operation With Broadband Amplification 
    The class-J PA topology consists of a slightly inductive load at the fundamental frequency and the only capacitive load at the 2nd harmonic (Fig. 10). This circuit has no explicit harmonic trapping network, which makes the class-J PA potentially capable of broadband operation [4].

    According to [4] and [5], a class-J PA is an unconventional class-AB PA with 2nd harmonic enhancement. The key to a class-J PA design is the 2nd harmonic element contained in the voltage waveform. The class-E-like waveform in [5] makes the PA more efficient, and the non-switching mode makes it linear. Furthermore, the inductive load at the fundamental frequency, and (only) the capacitive load at the 2nd harmonic keeps the drain RF voltage above the knee region [4], which makes class-J operation highly efficient and linear.

    The class-J approach uses a phase shift between the output current and voltage waveforms to render the 2nd harmonic termination purely reactive. This implies significant possibilities for the bandwidth efficiency of the class-J mode of an RF PA. For class-J operation, the load impedance must satisfy the following equation:

    When RL is defined, the harmonic load impedance is formed. Load-pull techniques should be applied to find the value of RL according to the desired high efficiency or high power. Broadband input matching needs to be applied to realize a broadband PA.

    In [5], a class-J PA with a commercially available 10 W GaN high-electron mobility transistor is shown. This PA design has near-rated output power of 39 dBm and very high efficiency of more than 60% across a bandwidth of 1.4-2.6 GHz, centred at 2 GHz. Not many papers have been written about using class-J PAs for handset applications, but we believe class-J is promising for modern multiband mobile TX handset design.

6 Conclusions
    In this paper, several highly efficient RF PA design techniques for silicon-based mobile broadband PAs and transmitters have been discussed. Two derivations of highly efficient class-E PAs based on different assumptions are shown in order to provide design insights. However, RFIC still needs to be carefully designed because of parasitics and non-ideal components in practical application. Characteristics and advantages of ET-based polar TX design have been discussed, and a comparison has been made with the EER technique. State-of-the-art ET-based polar TX performance for WiMAX and 3GPP LTE has also been demonstrated. Doherty PA design for high power, linear, and high efficiency operation as well as class-J PA design for broadband handset applications has also been briefly discussed. It is our hope and expectation that some of the promising techniques for high-efficiency silicon-based RF PA design can, someday, revolutionize the cellular RF PA market for future broadband wireless communication.

References
[1] Yan Li,? Po-Hsing Wu,?J. Lopez, R. Wu, D.Y.C? Lie, K. Chen, S. Wu, Tzu-Yi Yang, “A highly-efficient RF polar transmitter using SiGe power amplifier and CMOS envelope-tracking amplifier for mobile WiMAX,” in Proc. IEEE VLSI Design, Automation and Test,Taiwan, Apr. 2011,?pp. 1-4.
[2] D.Y.C. Lie, J. Lopez, J.,??J.D. Popp, J.F.?Rowland, Guogong Wang,?Guoxuan Qin,?Zhenqiang Ma, “Highly efficient monolithic class-E SiGe power ampli?er design at

[Abstract] This paper presents a brief overview of several promising design technologies for high efficiency silicon-based radio frequency (RF) power amplifiers (PAs) as well as the use of these technologies in mobile broadband wireless communications. Four important aspects of PA design are addressed in this paper. First, we look at class-E PA design equations and provide an example of a class-E PA that achieves efficiency of 65-70% at 2.4 GHz. Then, we discuss state-of-the-art envelope tracking (ET) design for monolithic wideband RF mobile transmitter applications. A brief overview of Doherty PA design for the next-generation wireless handset applications is then given. Towards the end of the paper, we discuss an inherently broadband and highly efficient class-J PA design targeting future multi-band multi-standard wireless communication protocols.

[Keywords] radio frequency power amplifier; silicon-based power amplifier; envelope tracking; class-E amplifier; broadband PA; class-J; Doherty power amplifier