In this article, some of the requirements of power amplifiers (PA) used for massive multiple-input, multiple-output (mMIMO) applications will be discussed. These requirements are gathered from our experiences in this application. Some of the requirements are related to the PA design and some are lineup requirements. The importance of Amplitude-to-Phase modulation (AM/PM) of an amplifier and the impact of its behavior on linearizability will be presented. Likewise, the impact of the measurement setup configuration on the wideband PA linearizability problems and solutions will be discussed. The focus will be on the design of PAs using Gallium Nitride (GaN) power transistors. We will demonstrate high power 50 Ohm asymmetrical Doherty amplifier lineups with excellent wideband digital predistortion (DPD) correction up to 200 MHz.
To accommodate the immense continued growth in data rates of future wireless communication systems require, PAs should be able to transmit multi-carrier and multi-mode signals that occupy the full frequency span of a given communication spectrum, while also meeting challenging power, efficiency, and linearity requirements. Independent of the signal bandwidth the linearity of PA with the DPD system must meet the linearity requirement in the 3GPP standard. One of the PA linearity requirement is the adjacent channel power ratio (ACLR), which must be < -45dBc1. In general, if the PA is easy to linearize with DPD system, it will lead to improved system efficiency. Therefore, the PA designer must focus on linearizability of PA and the maximization of efficiency.
In these wideband applications the performance of the PA is envelope and frequency dependent. This envelope and frequency dependence are memory specific phenomenon’s, which means that the output signal of the active device is not only a function of the instantaneous input signal values, but also a function of previous values2. Increased data rates will enable more uses and fast data transfer which lead to better service quality (Figure 1).
To enable Gigabit amplifier capability the frequency of operation of the amplifier must move to higher frequencies since the fractional bandwidth for operation at high frequency is smaller than operation at low frequencies. The basic relations of data rate and bandwidth are:
Where N is the number of coding states.
In today’s sub 6 GHz mMIMO applications the required bandwidth is in the range of 200 MHz. High power levels, 5-10W average, are another amplifier requirement for such applications. GaN on Silicon Carbide (SiC) is the only technology which can provide such operation at high frequency and power.
Wolfspeed also is only supplier which has a fully vertically integrated materials and device business specifically targeted at world class GaN devices. This enables a reliable supple of GaN Front-End Modules from the bulk crystal, epitaxy, wafer die, device design and hence the resulting PA (Figure 2).
In 4G, PAs were designed for relatively narrowband applications (~60 MHz). In such applications, the linearizability problems were caused mainly by the final stage Doherty PA.
Massive multiple-input, multiple-output (mMIMO) system is a technology for 5G. It is basically the parallelization and optimization transmitters with antennas and receivers (2^N, where N=1,2,…) to provide better throughput and better spectrum efficiency. These group of transmitters and receivers are working together to give high speed, capacity, security and other benefits for 5G as shown in (Figure 3).
For 5G mMIMO applications, the required instantaneous bandwidth is 160 MHz or more. With such wideband signals, the linearizability problems are not caused by the final stage PA only; rather, they are caused by the combination of PA lineup, wideband transceiver, and DPD system. Hence, wideband linearizability is limited by different parts of the whole system.
In these 4G narrow band applications we see the system operating in more or less SISO (Single Input / Single Output configurations known as Macro systems which are typified by single antennas at the Base station and handset. Having said that, handset diversity may also be in use – switching between two handset antennas for example for best Signal-To-Noise Ratio (SNR). Its throughput is limited in general by the channel bandwidth in use and the propagation condition. Equation 1 shows the basic channel capacity relationship.
Equation 1: The traditional Macro SISO relationship as a function of Bandwidth, B and SNR.
The 3GPP standards, particularly at level 13 & 14 (LTE Advanced Pro) and 15 (5G) offer many of technologies that rely on this additional bandwidth by parallelizing channel carriers. This includes Channel Aggregation (CA) and Dynamic Spectrum Sharing (DSS) in the sub 6GHz channel space. Both technologies directly increase the throughput (Gbps) by either using extra TDD (or FDD) channels of up to 20 MHz in bandwidth. In 3GPP level 15 we also have the ability to use both LTE and 5G channels simultaneously (DSS) to increase throughput. And with this we also see increases in the complexity of the available modulation methods i.e. in LTE we may see 64 QAM versus 256 QAM at level 3GPP level 15. This then further increases in 3GPP levels 16 and above.
Further to bandwidth increases, we have seen with LTE Advanced Pro and 5G (Sub 6 GHz) standards the ability of operators to employ multiple antennas at the transmit (tower) end of the link. This leads to linear increases in data rates as the mMIMO operates in a MISO (Multiple Input – Single Output) mode, often with a pair of sub 6 GHz antenna’s in the handset also offering switched diversity to overcome hand placement and signal loss in the handset. The mMIMO tower will then adjust its phase and amplitude components for the individual customer to beamform towards that handset or at least to peak throughput as well as is possible in a heavy multipath environment. Whereas 4G LTE Advanced Pro Chipsets were operating in the 1 GSPS range (although improving) the use of mMIMO has allowed sub 6 GHz handsets to obtain data rates in the 2.5-3 GSPS range. The channel capacity relation for such contemporary 5G Systems at the time of writing is given by Equation 2.
Equation 2: The current sub 6 GHz 5G MISO relationship as a function Base Antennas Numbers (N), Bandwidth (B) and SNR. This arrangement is currently leading to possible throughputs in the range of several Gbps.
Such handset chipsets as the recently released Qualcomm X60 achieve these record throughput improvements and as a result need ever wider band PA’s for cellular systems.
In addition to the wideband requirements of the PA for mMIMO, high efficiency is very important. The required efficiency is a must-meet metric to fit the size and weight requirements of mMIMO systems (Figure 1). To achieve high efficiency power amplifiers, asymmetrical Doherty architectures are generally used. Unfortunately, this kind of architecture is highly nonlinear. Since linearity and efficiency are trade-off parameters in PAs, to maximize the efficiency, the PA should be designed with the minimum required linearity; “only as much as we need.” For example, using a Class-A amplifier in back-off as a driver is not the most efficient solution for the lineup.
In addition to the as-current MISO configuration of the base station – handset arrangement, the true throughput of mMIMO as suggested by the Qualcomm X60 (and X50 solutions paired with a 4G LTE Modem) will likely move to higher spectrum allocations to achieve the bandwidth necessary for speed improvement. The second advantage of moving to these higher spectrum allocations is the reduction in size of the necessary antennas for both handsets and base stations. However, there are some tradeoffs at the present time. One is in the choice of system configuration. In built up areas, these higher frequency allocations do not propagate nearly as well as low frequency macro base stations, particularly the very low UHF bands such 617/619 MHz. In addition, another tradeoff due to complexity is the extended period for development of 5G base station systems based on expense, size and weight that it brings to the cellular base station solution. And this must be traded against the relative simplicity of high power Macro cellular base stations. In particular, in 5G we are looking at close integration of the antenna system with front end amplification and switching. In addition, since most practical systems will use phase and amplitude control in the digital domain for each customer slot, extensive processing power is necessary to achieve this.
Wolfspeed is enabling these solutions via the naturally higher efficiency at these higher frequencies using GaN on SiC technology. LDMOS is not a viable solution at these higher frequency allocations, simply due to poor device efficiency. We have talked about 3GPP levels 14 (4G LTE Advanced Pro) and level 15 (5G) however 5G will continue in development for several additional 3GPP releases before we finally get to 6G concepts. It is worth having a look at the present 3GPP roadmap, considering that not all chip vendors of cellular modems may implement all features of any one release level. (Figure 4) from the 3G Partnership Project shows the expected development as we work towards 6G systems. At the time of writing 3GPP are working on additions to 5G via releases 16 and 17 and these will place a heavier focus on the higher frequency allocations (mm Wave).
As can be seen the standards the industry is developing at present (3GPP levels 14 and 15) have been frozen. From 2020 onwards, we will be looking through to release 18 by 2022 3.
One example where, cellular modem makers may deviate from published standards is in vehicular communications where it may be omitted in a 4G LTE Advanced Pro solution or even a 5G modem development if the target market is handsets rather than vehicles. From a marketing perspective vehicular sales opportunity may be dwarfed by those of handsets. On the other hand, modem chipset makers may make tradeoff’s in other areas of the chip design to service vehicular communications (slated for the 4.9 GHz band). One such example would be the Qualcomm 9150 4G LTE chipset that supports V2V (Vehicle to Vehicle) communications but lacks the speed perhaps of some of the later non V2V solutions. Consequently, it is fair to assume we will see a rich ecosystem in the coming years of macro and mMIMO modem chipsets to serve urban, regional and special applications.
Finally, when vendors do move to true mMIMO solutions with a multitude of antennas on both the base stations and the handsets (possible at higher frequencies due to their smaller dimensions) we will see the power of MIMO speed improvement move from a linear increase – as is the case in MISO systems to an exponential increase, theoretically edging upwards of 10 Gbps at 28.8 GHz. The mMIMO channel capacity equation then reformulates as a complex equation determined by the Channel Matrix between many users, U, and the base station antenna array size, BS. A detailed derivation of the mMIMO channel capacity equation is not possible within the scope of this paper due to the variations in implementation and whether the system is TDD or FDD based. The result however is dependent on the multiplexing gain (gain due to algorithmic addition of multipath components) and the antenna gain under those conditions (due to the number of elements in play)4.
To design wideband PAs with a given linearity requirement, the optimization of the design should not only focus on the final-stage PA alone, but also the whole amplification lineup. In addition to this, the lineup should be optimized with the transceiver system as well as the DPD system in mind. The DPD system used to linearize mMIMO power amplifiers must be simple and highly efficient itself, compared to DPD systems used in traditional macro base-station applications. This will make the design of the PA lineup more challenging.
Furthermore, based on our expertise, different causes of wideband linearizability problems have been identified (Figure 5). These include:
- Input source/ pre-driver: Nonlinearity, harmonics, noise level, and bandwidth.
- Measurement setup: Calibration, isolation, filtering, load, thermal management, and dynamic range.
- Memory effects of the amplifier: Bias network, trapping effect, mismatch of even harmonics, and power supply modulation.
- Transceiver: Nonlinearity of up/down converters, LO leakage / spurs, image / harmonics, dynamic range, and / or noise level.
- DPD: For wideband applications (> 20 MHz), the DPD should include memory. Complex algorithms will consume much power; this is not acceptable for mMIMO application.
Figure 6 shows different shapes of a static amplitude modulation to amplitude modulation (AM/AM) and amplitude modulation to phase modulation (AM/PM) characteristics as metrics for wideband linearizability. As an example, the absolute nonlinear phase value of the amplifier is proportional to the amount of the nonlinearity in the amplifier. However, this phase is a good linearizability indication only for narrowband linearizability, <20 MHz Instantaneous Bandwidth (iBW).
As shown in the Fig. 6. good indicators for wideband, > 60 MHz iBW, linearizability:
- < 1-2 degrees AM/PM inflection in any static AM/PM shape within the frequency band.
- Static AM/PM dispersion across the frequency band. Low dispersion lead to good linearizability.
- Low peak power capability of the amplifier at any frequency.
Power Amplifier Design Example
Wolfspeed’s high power multi-chip asymmetrical Doherty PA module (PAM) is designed using GaN on SiC HEMT dies for 5G massive MIMO base station applications. PA Modules are compact and in a much smaller surface mount package than discrete components. These modules require minimum external components to build fully functional high-performance Doherty PAs. The asymmetrical Doherty power amplifier Module is designed for 5G/4G/LTE standards with a 28 V supply voltage and with 50-ohm input and output matching within 6 x 10 mm Surface Mount Package. Two stage internally matched Doherty PAs can deliver up to 30 dB of the gain, and high saturated power of up to 46 dBm. This PAM covers mobile frequency band from 3.4 to 3.6 GHz. Using the ADI (Analog Devices Inc) DPD system with 200 MHz iBW and 8dB Peak to Average Ratio (PAR) signal, the PAM provides excellent linearizability of -50 dBc ACPR at average output power of 37.5dBm. (Figure 7) and (Figure 8) show respectively the DPD data for narrowband, 20 MHz, as well as wideband, 200 MHz signals measured by ADI.
Different challenges and solutions of wideband massive MIMO applications were discussed and it's clear that the linearizability problem in 5G is caused not only by the PA lineup but the whole system. Therefore, the system should be optimized with DPD in mind. Different shapes of static AM/AM and AM/PM characteristics are proposed as metrics for wideband linearizability. For mMIMO systems, higher efficiency semiconductor devices like GaN should be used to achieve size and weight limitations. As an example, Wolfspeed’s multi-chip asymmetrical Doherty 50 Ohm PAM has shown excellent wideband linearizability using the ADI DPD system.
The authors gratefully acknowledge the DPD support from Noureddinethe Outaleb and his team from Analog Devices Inc.
- 3GPP standard: https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=3389.
- J. H. K. Vuolevi, T. Rahkonen, "Measurement Technique for Characterizing Memory Effects in RF Power Amplifiers," IEEE Trans. on Microwave Theory and Techniques, vol. 49, No. 8, August 2001.
- 3GPP ROADMAP (2020). Proposed 3GPP releases and milestones. Retrieved from https://www.3gpp.org/specifications/releases
-  Erik Dahlman: "5G NR: The Next Generation Wireless Access Technology." Academic Press 2018, ISBN-13: 978-0128143230.