Current-Injected Load-Modulated Outphasing Amplifier for Extended Power Range Operation

In this letter, the power range to be covered with maximized efficiency by a two-way outphasing power amplifier (PA) is significantly extended, thanks to a proposed architecture with the injection of an external signal. Using a reactively terminated quadrature hybrid coupler (QHC) as nonisolating combiner, the Chireix topology is transformed into a slight variation in the load-modulated balanced amplifier (LMBA) when the auxiliary branch is activated. This combined load-modulated (LM) strategy provides a nearly resistive loading of the individual outphasing PAs over a wide power range. An appropriate output network, approximating a class-E/F2 operation of the selected GaN-HEMT device under such loading condition, leads to remarkable drain efficiency figures at deep output power back off (OPBO). Values higher than 80%, 70%, and 60% have been measured at power levels 9.5, 13.3, and 15 dB below its peak (45.7 dBm), respectively. A 5-MHz Long Term Evolution (LTE) signal with a peak-to-average power ratio (PAPR) as high as 12.65 dB has been reproduced with an average efficiency above 62% and worst-case adjacent channel leakage ratio (ACLR) of −31 dBc.


I. INTRODUCTION
M ODERN wireless systems continuously advance in the use of spectrum-efficient modulation schemes to accommodate high data rates within restricted frequency bands. The resulting signals show unprecedented peak-toaverage power ratio (PAPR) values, while the linearity requirements are harder to satisfy. Linear amplifiers with high power efficiency at deep output power back off (OPBO) need to be designed, relying on supply-modulation and/or loadmodulation strategies [1]. Given the limitation for implementing wide bandwidth envelope amplifiers with high current and voltage capability, the load-modulation techniques have been receiving increased attention. Although the simplicity of its input signal splitter has made the Doherty power amplifier (PA) a more common choice than others, outphasing and particularly the load-modulated balanced amplifier (LMBA) architecture are gaining attractiveness. In the first case, an unbeatable efficiency profile may be obtained when using class-E PAs [2]. In the latter, the desired active loadmodulated (LM) operation may be guaranteed over a very wide bandwidth [3]. Individually, the abovementioned techniques use to offer a limited-efficiency enhancement at back off levels larger than 8-9 dB, reason why multiway or combined solutions [4], [5] seem to be unavoidable. Doherty-outphasing LM power amplifiers have been demonstrated [6], [7] for extremely deep OPBO operation, taking advantage of the common use both topologies make of λ/4 impedance inverters. In this letter, the principle behind the LMBA is used for extending the power range to be covered with maximized efficiency by a two-way GaN-HEMT class-E/F 2 [8] outphasing amplifier, thanks to the nonisolating combining performance offered by a reactively terminated quadrature hybrid coupler (QHC) [9], [10] and the injection of an external control signal. II. CURRENT-INJECTED LM OUTPHASING ARCHITECTURE In Fig. 1(a), a simplified diagram of the proposed architecture is presented. Amplifiers 1 and 2 operate as branches of a two-way outphasing topology, playing a sort of carrier PA role. A third amplifying branch is added as auxiliary, control, or peaking PA, aimed to extend the power range operation of the outphasing with maximized efficiency. When the auxiliary PA is at its OFF-state, an appropriate length of transmission line transforms its highly reflective output impedance into a reactive termination, Z term = − j · X term = − j · 50 , for the QHC port isolated from the output. As depicted in Fig. 1(b) and (c), under such terminating condition and given I aux = 0, the QHC performs as a Chireix combiner [10]. 1531-1309 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information. Varying the outphasing angle, θ , the input impedance values at RF in1 and RF in2 describe horizontally oriented trajectories, centered along the resistive axis. The compensating reactances, − j · X comp and j · X comp , play their typical role, reducing the imaginary part of the input impedance and adjusting the compensating angles, θ c , for which a pure resistance may be synthesized. A moderate-range outphasing operation was set in Fig. 1 The control current, I aux , injection is recommended when the outphasing topology is operating at the high-power or low-resistive loading θ HPc value Fig. 1(c). The outphasing angle needs to be slightly readjusted with the input power to the auxiliary PA or the I aux value, so that the mutual LM trajectories of the outphasing PAs stay coincident. The desired orientation along the resistive axis may be guaranteed with a fixed phase value for I aux [3]. As it may be appreciated from Fig. 1(c), the combined trajectories for the phase-coded (PhC-) and current-injected (CI-) LM modes show much smaller reactive components when compared to those due to a wide-range Chireix.

A. Device Selection and Biasing Conditions
The 700-MHz-band design of the involved PAs is based on the CGH35030F packaged GaN-HEMT from WolfSpeed. The drain dc voltage values for the outphasing and control PAs (V DSO = 16 V and V DSC = 32 V, respectively) were selected to allow an appropriate dimensioning of the required injection current while also avoiding the peak in the drainto-source voltage waveforms to reach breakdown. The gate dc voltage for the PAs integrating the outphasing topology (V GSO = −3.4 V) was initially adjusted just below the value for which some increase in the output conductance could be appreciated. In Fig. 2(a), the evolution of the S 22 parameter with frequency for the device biased at V GSO = −3.4 V and V DSO = 16 V is represented. A marker provides the value at the second harmonic, 1.4 GHz. The control PA gate-biasing voltage was set to a lower value (V GSC = −6 V), as typical in amplifiers with a peaking function.

B. Drain Terminating Network
The load-insensitive class-E/F 2 topology in Fig. 2(b) [11] was selected for both the carrier and peaking PAs. Although the auxiliary amplifier is not aimed to work under varying load condition, the minimization of conduction losses [8] may have a positive impact on the overall efficiency performance.
With the L 2 f 0 C 2 f 0 and L s C s circuits tuned at 1.4 GHz, L b1 is responsible for providing the termination required for resonating C out at 2 · f 0 (1 113.7 • ). L b2 was adjusted in order to force the drain-biasing network to act as an RF choke at 700 MHz. C f 0 and L f 0 are in charge of synthesizing an LM trajectory at the fundamental close to the optimum (the one composed by the impedance points providing maximum efficiency at each output power level). This trajectory and the 2 · f 0 termination were added to Fig. 2(a), together with the load-pull (LP) contours.

IV. EXTENDED CLASS-E/F 2 OUTPHASING OPERATION
The evolution of the load impedance at the fundamental, as seen from the drain terminals of the devices in the outphasing branches (PA 1 and PA 2 ), has been included in Fig. 3(a) over the corresponding LP contours. The extended outphasing technique, combining a moderate-range phase-coded with a current-injected LM mode, is here compared to an equivalent wide-range outphasing scheme operating up to 10 dB below the peak power value. In Fig. 3(b), the evolution of the outphasing angle and the normalized injected current as a function of output power is presented for each case. As it may be appreciated from Fig. 3(a), the combined or extended operation leads to LM trajectories closer to the optimum, giving the reason why an improved efficiency profile could be expected. Attending to the advantages of the mixed-mode approach [6] for reproducing the lowest part of a high PAPR signal envelope, the linear operation of the PAs integrating the outphasing topology in Fig. 1(a) has also been included in Fig. 3. The LM trajectories when the outphasing angle is kept equal to θ HPc , while increasing the injected current, have been finally added in order to visualize the benefits obtained when readjusting this angle with I aux , as proposed in this work.
In Fig. 4, the resulting efficiency versus output power, P out , profiles are presented for each of the cases. The power provided by the injection source has been included in the calculations. The losses in the class-E/F 2 topology selected for PA 1 and PA 2 impose a limit to the maximum efficiency to be obtained from their outphasing operation. The proposed extended or combined technique (PhC-LM + CI-LM modes) leads to an improved performance at deep OPBO, as a consequence of an LM operation of PA 1 and PA 2 closer to the optimum as well as of the entire recovery of the injected power at the output port [3], [9]. The power losses in the control amplifier, PA 3 , would impact the CI-LM mode profile at intermediate P out values.

V. IMPLEMENTATION AND CHARACTERIZATION RESULTS
In Fig. 5(a) and (b), the schematic and a photograph of the implemented architecture are shown. The 11304-3S QHC from Anaren Inc. was selected. The compensating susceptances were incorporated into the output networks of PA 1 and PA 2 (left side). High-Q air-core coils from Coilcraft and multilayer ceramic capacitors from American Technical Ceramics (ATC) were employed. The control amplifier (right side) is connected to the QHC through a 50-transmission line. A set of SubMiniature version A (SMA) transitions allowed fitting the required electrical length, but it could be substituted by a lumped-element equivalent to an integrated version.

A. Continuous Wave (CW) Characterization
The implemented three-way architecture was measured following the mixed-mode outphasing (linear + PhC-LM) and peaking (CI-LM) operating modes. The results appear in Fig. 6.
In outphasing mode, an output power of 38.9 dBm may be achieved with overall and drain efficiency values of 79.4% Fig. 6. Simulated and measured drain efficiency, η D versus output power profiles for the complete architecture. The measured overall efficiency, η ov , which includes the carrier and peaking input RF power, has been added. and 82.9%, respectively. It may be increased up to 45.7 dBm (a 6.8 dB extension), thanks to the CI-LM or peaking mode, with a worst-case drain efficiency of 60%. Although the measured profile shows differences with respect to the simulation, remarkable figures above 80%, 70%, and 60% were obtained at OPBO values of 9.5, 13.3, and 15 dB, respectively.

B. Dynamic Characterization
The hardware was tested using a 5-MHz Long Term Evolution (LTE) signal with a highly demanding 12.65-dB PAPR. An average efficiency of 62% and an adjacent channel leakage ratio (ACLR) of better than −31 dBc were measured, without applying any dedicated digital predistortion (DPD). A customized composite memory polynomial behavioral model [12] can be considered in order to satisfy the in-band and outof-band linearity requirements. Attending to the narrowband nature of the class-E/F 2 topology selected for validation, the performance is expected to degrade for signals with a bandwidth wider than 20 MHz. Table I compares the results to those from representative state-of-the-art GaN-HEMT multiway or combined LM PAs. Incorporating the reported variations in outphasing [13] and LMBA [14], the proposed three-way architecture could be made to operate from a single RF input.

VI. CONCLUSION
An architecture with injection of an external signal has been proposed for extending the power range to be covered with maximized efficiency by a two-way outphasing PA. Implemented at UHF band, based on a load-insensitive GaN-HEMT class-E/F 2 topology, efficiency figures higher than 80%, 70%, and 60% have been measured at power levels 9.5, 13.3, and 15 dB below its peak (45.7 dBm), respectively.