Most power amplifiers (PAs) currently used in point-to-point microwave link applications are based on GaAs PHEMT technology, but GaN is now also a contender for this application. GaN technology has a well-proven capability for PAs, especially those requiring very high levels of saturated output power. As the technology has matured it is increasingly used in communications applications requiring high linearity. In order for GaN solutions to be adopted, they need to provide a performance advantage at a comparable cost.
A GaN PA was designed for the 15GHz line-of-sight link band, targeting the outline performance specification in Table 1. The initial design was optimized for IP3 — the traditional microwave measure of linearity for amplifiers. The performance when amplifying QAM 256 modulated signals was then analyzed. Spectral re-growth and Adjacent Channel Power Ratio (ACPR), distortion of the constellation and Error Vector Magnitude (EVM) were all evaluated.
The design used Cree’s 0.25µm GaN-on-SiC process biased at 28V Vds, which is capable of delivering 4W of RF output power per mm of gate width. Preliminary simulations were made to select the preferred transistor size and bias point, and to determine the most appropriate amplifier topology. A 4x250µm transistor (1mm total gate width) was selected with the Ids bias initially set to 100mA. For this application, linearity is more important than saturated output power performance. The optimum load impedance for intermodulation performance was determined using large-signal analysis with two input tones of a specified level. With each tone set to +18dBm at the output, this resulted in a third-order product of -50.2dBc giving an OIP3 of +43.1dBm.
An inductive matching component was used near the transistor’s drain to transform the optimum impedance to a purely real impedance. The remaining real impedance is 140 for optimum linearity, compared to 48 for optimum Psat and 70 for optimum PAE.
The transistor models in the Cree Process Design Kit (PDK) calculate the junction temperature and account for the thermal effects on RF performance, given the baseplate temperature and the thermal resistance of the transistor. The designer needs to ensure the overall thermal impedance is good enough to give adequate thermal behavior.
Detailed Amplifier Design and Layout
Based on this preliminary simulation work, it was decided to implement a two-stage amplifier with an output stage based on two power combined transistors, each of 1mm gate width. It is important that the input stage is large enough to drive the output stage without exhibiting excessive compression. GaN transistors have high gain, but it is a mistake to use too small a driver stage as the soft compression characteristics of GaN mean that the driver stage and output stage will compress simultaneously, causing the amplifier linearity to be compromised.
Figure 1 compares the layout of the complete 15GHz amplifier with a commercially available GaAs part currently used in 15GHz point to point applications. The thick blue box indicates the outline of the GaAs part. Not only is the GaN die smaller, its linearity is also 2dB higher. This clearly demonstrates the potential benefit of GaN technology for line-of-sight applications.
Simulated RF Performance
Small signal s-parameters of the amplifier show a gain of 22.1dB ± 0.4dB from 14GHz to 16GHz, providing a healthy guard band on the 14.5 to 15.35GHz operating band. The input return loss is better than 15dB from 14.5 to 16.2GHz and the output return loss better than 14dB to 15.6GHz. The simulated output power is better than 38dBm across the band, and power added efficiency at 3dB compression is better than 36%. Further increases in both output power and PAE are possible if a higher level of compression can be tolerated.
The OIP3 versus output tone power was simulated at 15GHz with two tones10MHz apart, and indicates an OIP3 of 46dBm at 22dBm per output tone. It is possible to further improve the OIP3 by increasing the quiescent bias current, while paying attention to ensure adequate thermal performance.
Higher Order Modulation Simulations
The modulation schemes used in point-to-point microwave radio links are chosen to be spectrally efficient, so a large amount of data can be passed through a single channel. This is because such links are typically used in high data rate applications, such as cellular back-haul, with required data rates of the order of hundreds of megabits, or even gigabits, per second.
Figure 2 shows the constellation diagram of four such schemes — QAM 64, 128, 256 and 512 — where each data symbol represents 6, 7, 8 and 9 bits respectively.
The corresponding eye diagrams in the I (in-phase) plane for these schemes (Figure 3) show the point in the received transmission where the signal has to be sampled to correctly determine its identity. The corresponding eye diagram for the Q (quadrature) plane is not shown. The receiver symbol timing recovery loop must ensure the receiver samples the signal at the correct sampling point — the middle of the eye. All four modulation schemes occupy almost the same amount of spectrum, but the linearity required to preserve modulation fidelity increases with the order of the modulation.
The spectrum of an undistorted QAM 256 referenced to a 1Hz bandwidth is given in Figure 4. The almost rectangular spectrum demonstrates that it will be well confined in frequency to an allocated channel, which is a result of root-raised cosine filtering of the digital modulation. In this case, the application is for a 15GHz Class 4H transmission with a 56MHz channel spacing and a symbol rate of 46MBd. The class 4H transmission is one of a set of transmission standards applied to point-to-point links as specified in ETSI standard EN 302217. This standard specifies a transmit mask within which the modulated spectrum must lie, which is also depicted in Figure 4.
A disadvantage of these spectrally efficient modulation schemes is that they are corrupted by non-linearities in the transmission system, and one of the largest contributors is non-linearity due to the PA when driven towards its maximum output. It is important not to overdrive the PA in an attempt to achieve more output power, because this will corrupt the modulation.
Corruption can initially be observed as a distortion of the modulation constellation leading to eye closure, which in the worst case could mean one transmitted data symbol being read as a different symbol at the receiver, leading to transmission errors. Error Vector Magnitude (EVM) is one measure of characterizing this effect, and specifications for radio links will have an EVM figure that the system must meet.
A second manifestation of PA distortion is the generation of intermodulation products (also known as spectral regrowth) in nearby channels. This is when the modulated spectrum spills over into channels occupied by other users, outside of the spectral mask, and must be kept below the required limits.
The impact of PA distortion on the link modulation can be predicted using a radio system simulator such as Matlab/Simulink. Random data streams can be generated and modulated, for instance using a QAM 256 modulator, root-raised cosine filtered, and then applied to a non-linear model of the PA. The output of the PA can then be observed and the spectrum, eye diagram, constellation diagram, and EVM compared against the various mandatory and operational requirements. Figure 5 shows the major blocks of this system simulation model.
When the PA was analyzed in this way the resulting output constellation, eye, and spectral diagrams initially showed some eye closure and constellation corruption, particularly at the extremes where the instantaneous power of the modulated signal is highest. The spectrum was just contained within the mask, but only by 1 or 2dB, and a greater margin was desired. The system simulation was re-run at an increased PA bias current of 160mA/mm and with the drive level adjusted to give the same 29.4dBm output power as in the original simulation. The results are presented in Figure 6, showing well-defined constellation and eye plots, with a corresponding EVM of -37.8dB (approx 1.3%). The spectral plot now shows greater than 7dB margin with respect to the spectral mask.
The suitability of a GaN MMIC PA for use in high data-rate point-to-point link applications has been demonstrated, showing good RF performance for a PA targeting the 14.5 to 15.35GHz band. Small signal gain is 22.1dB ±0.4dB, and output power at 3dB compression is better than 38dBm with corresponding PAE better than 36%. At 130mA/mm quiescent bias the OIP3 is 46.0dBm at 22dBm per output tone, increasing to 48.5dBm for 160mA/mm bias. The die area of the GaN part is smaller than that of a commercially available, lower linearity, GaAs part targeting the same application. A behavioral model of the amplifier was developed for use in a system simulator, allowing analysis of key performance metrics when driven by typical high order modulation signals such as 256 QAM. The PA met the appropriate ETSI spectral mask, and achieved a 1.3% EVM with a mean output power of 29.4dBm when operated at a quiescent bias current of 160mA/mm.