Enabling Fast Characterization of PA Performance with Modulated Signals
By Andy Howard, Agilent Technologies
Power amplifiers (PAs) play a key role in modern communication systems—contributing significantly to power consumption, and in turn, overall system performance—making their accurate characterization all the more critical. Through the years, however, PA design and characterization has become exceedingly more difficult, complicated by the complexity of modern digital signals and the continually evolving nature of wireless standards. Simply characterizing amplifier performance with sinusoids is no longer sufficient. Quickly and accurately designing a PA that meets adjacent channel power ratio (ACPR) and error vector magnitude (EVM) specifications, while delivering a specified output power, now requires that designers have access to a range of information. Besides ACPR, EVM and output power, for example, they must know how performance varies versus power and at specific output powers, and how they can adjust design parameters to improve performance. Moreover, they need to know statistical distributions of performances.
Two possible solutions to this PA characterization dilemma include use of swept-power harmonic balance (HB) simulation or Ptolemy co-simulation. Both techniques are part of the Agilent Advanced Design System (ADS) software framework. While neither technique is new, recent advances now make each much more effective. Let’s take a closer look.
Swept-Power HB Simulation
Swept-power HB simulation provides a fast way of computing ACPR, modulated output power and/or EVM of an amplifier versus input power. It simulates PA nonlinearity using a 1-tone HB power sweep and applies a modulated signal to the nonlinearity in post-processing. The results are then interpolated to obtain data at a specific output power. Thanks to recent enhancements, the technique now permits inclusion of extra sweeps, for example, of a parameter value or a Monte Carlo analysis. Additionally, it allows designers to see correlations between statistical variables and results, and makes data available at user-specified output power(s), without the designer needing to re-simulate or run an optimization.
Using this technique requires modulated signal or baseband I and Q data, and the main, adjacent and alternate channel frequency limits for the ACPR calculation. An example from which to copy the setup is available in ADS. The technique involves seven basic steps, many of which are carried out by measurement expressions:
Step 1: Run a swept input-power HB simulation of the amplifier.
Step 2: From the simulation, determine the amplitude-out-versus-amplitude-in (AM-to-AM) and phase-out-versus-phase-in (AM-to-PM) transfer functions of the amplifier.
Step 3: Obtain the envelope-versus-time waveform of the modulated input signal, either from a separate simulation or from some file. This will be the magnitude and phase of the signal.
Step 4: Apply the modulated signal to the transfer functions from Step 2 (e.g., to the simulated amplifier’s nonlinearity) and compute the envelope-versus-time of the output signal.
Step 5: Use a time-to-frequency transform (fs() function) to compute the spectrum of the modulated output signal.
Step 6: From this spectrum, compute the powers in the main, adjacent and alternate channels. Next, compute the adjacent- and alternate-channel power ratios from these values.
Step 7: Compute the EVM. This is the “raw” EVM, computed by comparing the output signal to the input signal at each time point, after correcting for average gain and phase shift. While not a specification-compliant EVM, it is a good proxy for a more accurate EVM calculation that would require much more data.
Once the calculations are complete, swept power (from scaling input signal amplitude) and interpolated results can be displayed. The interpolation provides data at a specified output power (Figure 1).
Designers also want to know how the ACPR, EVM, gain, and gain compression depend on different design parameters to determine whether performance improvements can be made.
This can be done by sweeping a parameter in the PA subcircuit. However, sweeping one parameter at a time is inefficient. What if the designer wants to investigate performance variation while varying multiple design variables simultaneously? This can be done by running a Monte Carlo simulation and specifying the design variables to be uniformly distributed across whatever range is desired, although to avoid convergence problems the range should be limited.
To use the swept-power HB simulation technique effectively, there are a few key things to remember. Designers can use as many design variables as they want, but the overall simulation time will be determined by the number of Monte Carlo trials. As the specified output power is changed, the best parameter values will be automatically updated. Monte Carlo analysis can then be rerun after the parameter value ranges have been adjusted based on the best set of parameter values.
Monte Carlo analysis can be run to simulate statistical variation. Note that via interpolation, the statistical variation can be obtained while the amplifier is delivering a user-specified output power. Once the statistical variation is known, designers must determine which statistical variables matter. Here, scatter plots of ACPR and EVM prove extremely valuable, showing any correlations between statistical variables and results, and in turn highlighting critical variables (Figure 2).
A more accurate technique for characterizing PA performance uses Ptolemy co-simulation and automatic verification modeling, and then interpolates to obtain data at specific output power(s). As with the previous technique, this process is repeated for each Monte Carlo trial or swept-parameter value. Thanks to recent enhancements, the technique’s power sweeps run much more efficiently, correlations between statistical variables and results can be seen, and data is now available at user-specified output power(s), without having to re-simulate or run optimizations.
The Ptolemy co-simulation technique requires an envelope controller in the PA subcircuit and involves a few key steps (Figure 3). First, a specification-compliant source is used to generate a modulated signal that is supplied to a PA subcircuit-under-test. The output from the PA is then measured using specification-compliant sinks to determine the adjacent- and alternate-channel power ratios, EVM and mean transmitted RF carrier power. Additionally, bias voltages and currents in the amplifier subcircuit are measured to determine the average DC power consumption, enabling efficient calculations.
To speed up the simulation, Automatic Verification Modeling (or Fast Cosimulation) is used to create (via harmonic balance power sweep) and simulate a behavioral model (which can include frequency variation across the modulation bandwidth) of the PA subcircuit. This model is only regenerated if something in the circuit changes, such as if sweeping a parameter or running Monte Carlo. The resulting speed increase is orders of magnitude faster than a full transistor-level model. Figure 4 shows results from this type of simulation.
There are various advantages to using the two techniques detailed. Ptolemy co-simulation, for example, provides specification-compliant measurements and, if required, can include receive-side filtering. Additionally, it allows PAE to be computed. By comparison, swept-power harmonic balance simulation is faster, although less accurate, for short time sequences and can be used with any source. While which method is employed will depend in part on the designer’s specific needs, both allow engineers to better understand the statistical variation of their design and understand which variables matter most—all key factors in being able to quickly and accurately design today’s power amplifiers.
About The Author
Andy Howard is a senior applications engineer with Agilent Technologies. Over the years, he developed the ADS Amplifier DesignGuide, worked on the Mixer DesignGuide and has extensively updated the ADS Load Pull DesignGuide. He also designed a high-speed prescaler IC using Agilent’s RFIC Dynamic Link that was fabricated using IBM’s SiGe process. His more recent work focuses on power amplifier simulation applications. He holds a B.S. E.E and M.S. E.E. from the University of California, Berkeley.
this article to a friend!