UC San Diego
Behavioral Modeling of Nonlinearities and Memory Effects in Power Amplifiers
- Author(s): Draxler, Paul J.
- et al.
High data rates and tight spectral limitations in wireless communication systems require high fidelity waveform transmission with minimal distortion. The waveforms currently being used, or envisioned for the future, coupled with the need for higher efficiency radio frequency (RF) power amplifiers (PA), are driving transmitters to include predistortion and to adopt advanced architectures. PA behavioral models are required for system simulations of these advanced architectures and digital predistortion (DPD) algorithms, to quickly evaluate the component impact on system performance. One advanced technology of particular interest in this work is envelope tracking (ET) where the supply voltage is modulated at the envelope rate, keeping the RF signal operating rail to rail, which changes the performance of the RFPA. This dissertation focuses on accurate estimation of nonlinear behavioral models for PAs: both memoryless and with memory effects. Memory effects are the byproduct of physics-based or circuit-based changes within the amplifier with responses on a baseband time scale (rather than at the carrier frequency). When these memory effects become pronounced, they create a shift in the memoryless model. In order to accurately predict system performance, especially when the system includes a DPD block, it is critical to generate behavioral models for RFPAs which capture the shifts in the memoryless model. The thesis presents a new model, termed the Blackbox augmented behavioral characteristics (ABC) model, which includes the response of the internal states of the circuit. Its extraction is demonstrated from measured data using pulsed or modulated waveforms. The generation of modulated waveforms for instrumentation systems has length restrictions and input sequences are used repetitively, so they should be circular. Techniques that synthesize circular waveforms suitable for measurement systems, directly from long modulated waveforms are presented. An "expected gain" model is developed in this work, and a methodology for extracting waveform specific, expected gain models efficiently from measurements of modulated signals (such as WCDMA) is presented. These expected gain models are then applied to DPD. A technique that leverages the circular stationary measurements, memory mitigation, is presented. The DPD algorithm identifies and compensates systematic distortions in the waveform, generating input signals that achieve optimal outputs, compensating for all deterministic memory of the system and quantifying the measurement limits of the system. A number of systematic measurement impairments are encountered in these experiments. Techniques that compensate for these impairments, including phase drift and time alignment are presented. The thesis also describes behavioral modeling include demonstrating DPD using new techniques that stem from truncating and thresholding the Volterra series. When compared with two other published truncated Volterra series forms on standard Class AB PAs, all three memory models perform equally well. When applied to modeling ETPAs, however, the new thresholded Volterra series model is more efficient, using less than a third of the coefficients to describe the ETPA nonlinear memory effects. In this work, techniques are applied experimentally to RFPAs, from handset PAs to base station PAs, built with a wide variety of materials: HBTs on GaAs, LDMOS devices on Si, and HFET devices on GaN, and GaN HFET devices on Si