August 28, 2019
This is the fourth blog in an ongoing series looking at the importance of GaN HEMT nonlinear models for rapid power amplifier (PA) design success.
S-parameter matching is used to maximize gain and gain flatness in simple linear RF/microwave amplifier designs. This same S-parameter data is used to develop matching networks that address amplifier stability. This blog discusses the importance of using modeling for basic S-parameter and stability analyses in the gallium nitride (GaN) power amplifier (PA) design process. It introduces the use of models and resistive stabilization techniques to help avoid device instabilities that can affect nonlinear and linear simulations.
In this blog, we focus our attention on a simple two-port stability analysis derived from linear S-parameter calculations. We will use a nonlinear Qorvo GaN power transistor model from the Modelithics Qorvo GaN Library, in combination with simulation templates and Keysight Advanced Design System (ADS) software.
Model-Based GaN PA Design Basics
Qorvo and Modelithics have teamed up to explain how nonlinear models and the Modelithics Qorvo GaN Library can improve your PA designs.
Catch up on the first blogs here:
Stability refers to a PA’s immunity from possible spurious oscillations. Oscillations can be full power, large-signal problems, or subtle spectral problems that might go unnoticed if not properly analyzed. Even unwanted signals outside your intended frequency range can cause system oscillations and gain performance degradation.
There are two types of stability and measures to analyze PA stability in your system.
Let’s begin with the well-known “k-factor” and stability measure “b” to determine frequency ranges that cause instability at a given bias. These are given by the following equations^{1}:
k = {1- |S_{11}|^{2} - |S_{22}|^{2} + |S_{11}*S_{22} - S_{12}*S_{21}|^{2}} / {2*|S_{12}*S_{21}|}
and
b = 1 + |S_{11}|^{2} - |S_{22}|^{2} - |S_{11}*S_{22} - S_{12}*S_{21}|^{2}
Unconditional stability is indicated by k > 1 and b > 0.
However, because this criterion requires two parameters to check for unconditional stability, a more compact formulation is given with the following “mu-prime” parameter^{2}:
mu_prime = {1 - |S_{22}|^{2}} / {|S_{11} - conj(S_{22})*Delta| + |S_{21}*S_{12}|}
If mu_prime > 1, it indicates unconditional (linear) stability.
As noted above, S-parameter data is used to develop matching networks to attain amplifier stability. Figure 1 shows a single-stage amplifier configuration and the key parameters that affect gain and stability. In the unconditional stability region, maximum gain is achieved by setting Γ_{s} and Γ_{L} to conditions attaining a simultaneous conjugate match at both ports.^{1}
Figure 1.
Let’s consider an example. Figure 2 shows a simulation setup for linear S-parameter analysis of the nonlinear model for Qorvo’s T2G6003028-FS GaN HEMT device, included in the Modelithics Qorvo GaN Model Library.
Figure 2.
Note: Bias condition for all simulations in this note is set to Vds = 28 V, Vgs = -3.02 V, which corresponds to a drain current of approximately 200 mA.
In the schematic above, icons represent parameters that can be calculated from device S-parameters, including stability k, b and mu_prime. The “MaxGain1” parameter is the maximum available gain. The “MaxGain1” parameter calculates the maximum available gain for frequency ranges where the device is unconditionally stable, and displays a value that is termed the maximum stable gain. This is calculated as simply |S_{21}|/|S_{12}| for regions of conditional stability.
Figure 3 shows the MaxGain1 parameter, the 50 ohm gain (S_{21} in dB) and stability factor k, measure b and mu_prime calculated from the schematic of Figure 2 (at m5). This plot shows that the stability measure b is > 0 and stability factor k > 1. The stability measurement parameters show a clear break point at about 1.85 GHz (m5). This is the transition frequency between conditional and unconditional stability regions. For 3.5 GHz the maximum gain indicated by this simulation parameter is approximately 18.4 dB (marker m3 in Figure 3). Note: The maximum available gain goes to 0 dB at about 10.4 GHz; this frequency is referred to as the maximum frequency or f_{max}. It is also a good practice to analyze stability from a very low frequency to at least f_{max}, which is why the frequency range for this example was set to sweep from 25 MHz through 12 GHz.
From this analysis, we can conclude the following:
These S-parameters produced from the schematic simulation (Figure 2) are show in Figure 4. S_{11} and S_{22} are displayed on Smith charts, while polar charts are used for S_{21} and S_{12}.
Notice the large difference between the gain for 50 ohm input and output (|S_{21}| in dB) and the MaxGain1 value. This is due to the mismatch associated with S_{11} and S_{22} in a 50 ohm system.
Figure 3.
Figure 4.
Plotting the stability circles in the input and output planes provides additional insight. Also included in the schematic of Figure 2 are the icons for "S_StabCircle” and “L_StabCircle”, which correspond to calculations of stability circles in the input and output planes.
The meanings of these circles can be described as follows. In the case of the input stability circle at 25 MHz, indicated by marker 14 in Figure 5, each point along that circle represents a Γ_{s} value that will result in a Γ_{out} value equal to 1 according the following relation.
Γ_{out} = S_{22} + S_{12}*S_{21}*{Γ_{s} / (1-S_{11}*Γ_{s})}
Eq. 1
This circle sets a boundary between Γ_{out} < 1 and Γ_{out} > 1, the significance of which is that Γ_{out} > 1 corresponds to a negative resistance at the output port, which is a condition that can lead to an oscillation. The question then becomes whether the inside or outside of the circle is the unstable (Γ_{out} > 1) region. A quick check in the case of Γ_{s} = 0, which is the 50 ohm point. Note from Eq. 1, for this case Γ_{out} = S_{22}, which is less than 1 at all frequencies being analyzed here. From this, we can conclude the outside of the circle is the stable region and the inside is the unstable region.
The explanation of the output stability circles is basically the same, except here we are plotting circles of points for Γ_{L} for which Γ_{in} = 1, according to the Eq. 2. By a similar argument, we can conclude that it is the inside of the circles plotted on the right side of Figure 5 that correspond to the unstable regions. Note - the frequency plan of Figure 2 was reduced to show fewer circles in Figure 5 for clarity.
Γ_{in} = S_{11} + S_{12}*S_{21}*{Γ_{L} / (1-S_{22}*Γ_{L})}
Eq. 2
Figure 5.
So, what if a device does not meet the requirements for unconditional stability, like in our example for frequencies below 1.85 GHz?
There are multiple matching methods to help stabilize your circuit. In this blog we describe two methods. One is resistive and the second is frequency-dependent stabilization.
Matching resistors can be employed in our example to help stabilize high-gain, low-frequency transistors in most microwave applications. These resistors can be series or shunt at the input or the output, can be in the parallel feedback loop, or included in the bias networks. For PAs, we want to maximize output power, so it’s best to avoid resistors in the output network. Feedback amplifiers are outside the scope of this post, so we will concentrate on the series and shunt resistors in the input network.
Figure 6 shows where both series and shunt resistors have been added in the input network. The values are tuned to achieve unconditional stability over the entire 0.025 to 12 GHz frequency range. The resulting stability measurements are plotted in Figure 7. These show the transistor has unconditional stability over the entire frequency range. Note, however, f_{max} dropped from 10.3 GHz to about 8.75 GHz. Comparing the maximum gain estimation in Figure 7 (design frequency of 3.5 GHz [12.3 dB]) with Figure 3 achieved without this stabilization (18.4 dB), we can see we have incurred a 6 dB degradation in maximum available gain. This is caused by adding a purely resistive input stabilization network. The S-parameters of the resistively stabilized device are displayed in Figure 8, with the overlaid S-parameters of the non-stabilized device. We can see that S_{11} and S_{12} have been affected over the entire frequency range, and S_{21} is also reduced with only minimal change in S_{22}. It is gratifying to observe in Figure 9 that with the resistive stabilization network added, the stability circles are now all outside of the Smith Chart in both the source and load-planes.
Figure 6. (Note: Analysis setup is the same as in Figure 2)
Figure 7.
Figure 8.
Figure 9.
If the design frequency is above 1.85 GHz (e.g., 3.5 GHz), we can implement a frequency-dependent resistive approach using the series-shunt stabilization network. Let’s see if we can mitigate the above gain penalty using this approach.
In Figure 10, a resistor (R1) has been incorporated into a modified gate bias network. Additionally, a capacitor (C3) has been placed across the series stabilization resistor (R1). The value of this capacitance can be tuned to adjust what frequency the series resistor (R1) is - effectively shorting it out (making it not “seen”). This can help increase the available gain.
Figure 10.
The inductor (L1) and capacitor (C1) are used to create a low-pass filter. This prevents the resistor (R1) from being seen at higher RF frequencies, or lower frequencies for stabilization. The gain, stability and S-parameter analysis for this solution is shown in Figure 11, Figure 12 and Figure 13. As shown, the frequency-dependent stability network provides unconditional stability across the full frequency range, while reducing the impact on maximum available gain at 3.5 GHz. Note the gain at 3.5 GHz is now reduced by only about 1dB compared to the non-stabilized device, and also the f_{max} is about the same as the non-stabilized device (~10.4 GHz). In examining the S-parameter comparison to the non-stabilized device as shown in Figure 12, we see that, in contrast to the resistively stabilized device, the S-parameters are not altered over the entire frequency range, but rather only at lower frequencies, as desired. Figure 13 just confirms that none of the stability circles overlap with the Smith Chart in either the source or load planes as expected for an unconditionally stable circuit.
Figure 11.
Figure 12.
Figure 13.
So what are the key findings? As shown in the below data, stability and gain are optimized when using frequency-dependent stability.
Modeling helps address common design problems such as stability prior to testing your application on the bench. By accurately modeling and implementing stability techniques, we can match and tune for optimal S-parameter performance while maintaining unconditional stability.
As a final note, the stabilization networks explored here used ideal lumped elements. In an actual microwave design, you will need to include microstrip interconnects and accurate parasitic models for all RLC components, whether you are doing a MMIC design or a board-based hybrid design with lumped elements.
Learn more about our nonlinear models for packaged and die Qorvo GaN transistors:
For those with access to the Modelithics Qorvo GaN Library, you can also email info@modelithics.com to request an example ADS workspace and/or NI AWR project related to this blog.
Have another topic that you would like Qorvo experts to cover? Email your suggestions to the Qorvo Blog team and it could be featured in an upcoming post.