PhD seminar - Dylan Bespalko

Monday, November 30, 2015 2:00 pm - 2:00 pm EST (GMT -05:00)


Dylan Bespalko


Modular Nonlinear Characterization System and Large-Signal Behavioral Modeling of Unmatched Transistors for Streamlined Power Amplifier Design


Slim Boumaiza


This thesis provides a comprehensive approach to the characterization and modelling of large-signal nonlinear RF/microwave devices, circuits and systems. This research is motivated by the increased linearity and power-efficiency requirements of modern power amplifier technology for wireless communications. For instance, maximizing the power amplifier’s efficiency can only be achieved by operating RF transistors under strong nonlinear conditions, however this is contradictory to maximizing PA linearity. Simultaneously designing for efficiency and linearity is a challenging trade-off in today’s fragmented design process, therefore the advancement of computer-aided design (CAD) tools is essential for achieving an optimal solution. The successful and effective CAD tool based PA design relies on the availability of accurate nonlinear models to mimic the electro-thermal behaviour of RF transistors. The accuracy of these models depends on three factors:

  1. The formulation of the model.
  2. The model extraction procedure.
  3. The accuracy of the measurement data.

While prior work focuses separately on the improved model formulations or improving characterization accuracy, this thesis provides a comprehensive analysis of all three factors. This thesis proposes a modular large-signal RF device characterization system, and a nonlinear behavioral model capable of handling strongly nonlinear unmatched RF transistors, each necessary to streamline the design process and achieve a first-pass PA design.

As a first step, a large-signal characterization system has been developed to measure the multi-harmonic frequency response of RF transistors and has the ability to i) Perform high-power measurements, ii) Characterize unmatched transistors, iii) Operate the DUT under any possible operating condition, iv) Synthesize any multi-harmonic stimulus, and v) Reconstruct the time-domain I/V waveforms at the ports of the DUT. The proposed characterization system eliminates fragmentation between measurement and simulation environments by providing seamless integration with Harmonic Balance simulations. This provides a common framework that integrates all steps of the PA design process from device-level characterization, to circuit-level measurement and validation. This system is implemented using modular instruments consisting of mixer-based receivers, arbitrary waveform generators, impedance tuners, and a multi-harmonic phase-coherent reference source. It also integrates sequential calibration routines to provide receiver, port match, and source-power corrections to the DUT measurement plane and measurement routines for automated data collection.

The second part of the thesis researches black-box frequency-domain behavioral models that can approximate strongly nonlinear, unmatched devices. Our investigation yielded two complimentary solutions to ensure the targeted modelling accuracy. First, improving the accuracy of a first-order expansion-based Poly-Harmonic Distortion (PHD) by 5dB, in terms of Normalized Mean-Squared Error (NMSE), by minimizing multi-harmonic reflections that artificially increase the order of the nonlinear system. While this addresses the fictitious need for higher-order models due to the deficiencies in the model extraction procedure, strongly nonlinear devices will require high-order models to achieve the targeted accuracy over a larger measurement distribution. Hence, a variable order Multi-Harmonic Volterra (MHV) model is proposed to extend the PHD model formulation to strong nonlinear devices. This model is extracted by utilizing the proposed characterization system to extract higher-order multi-variate model coefficients not included in the PHD model. The resulting model improves DC drain current prediction by 5dB and improves fundamental output-power prediction by 2dB.

Finally, a concurrent dual-band PA design is studied as an example of how the proposed nonlinear characterization system and behavioural modelling approach can be used to enable complex PA designs. First, a 10W Class-AB PA is designed using dual-band matching-network theory, however it is difficult to implement because the design technique does not control the matching fractional bandwidth as a design parameter. Therefore, an alternative Class-J 45W dual-band PA was designed using a low-impedance matching network, combined with a trans-impedance dual-band filter. Although the dual-band PA can achieve comparable performance to an equivalent single-band PA at each separate frequency, further development of characterization, modeling, and circuit design techniques is needed to achieve high-efficiency during concurrent operation.