Experimental characterization and nonlinear modeling of microwave devices by using an LSNA

Resumo

The planar transmission line technology development led to the introduction, during the 1960s, of the microwave integrated circuits (MICs). Their much lower size and fabrication cost in comparison to the conventional waveguides/tubes microwave technology, made them ideal to be widely utilized, in conjunction with semiconductor device technology, in modern high-frequency communication transceivers. This technology opened the door for new communication uses, which over the years have evolved to present applications such as the 4th generation (4G) mobile telecommunications technology, Wireless Local Area Networks (WLANs), Global Positioning Systems (GPS), Digital Terrestrial Television (TDT), radar systems, Unmanned Aerial Vehicles (UAVs) guidance, etc. Nowadays, energy efficiency has become a priority in the design of many microwave wireless systems. Since the power amplifier (PA) stage is typically responsible for the largest percentage of the total system power consumption, its design deserves special attention, among other transceiver nonlinear components. Moreover, microwave PA stages are usually required to fulfill other demanding specifications, such as increased bandwidth, high linearity, high gain, etc., and, generally, trade-offs have to be considered. Furthermore, signals to be processed must be bandwidth efficient, presenting complex modulations and a high Peak Average Power Ratio (PAPR), as for example the WIFI or WIMAX standards, occasionally involving high frequencies and/or powers. Such demanding performances required from the nonlinear circuits are only achievable with the use of efficient Computer Aided Design (CAD) tools, i.e. device/circuit models and circuit simulators, besides an adequate fabrication technology. Hence, precise active device nonlinear models and reliable measurement tools, able to perform the required electrical characterizations under large-signal regime, are mandatory. Active device nonlinear models must predict, not only the direct current (DC) and small-signal device response, but also the large-signal behavior, single-tone and multi-tone. Furthermore, in the case of bipolar transistors and large periphery Field Effect Transistors (FETs), the prediction of their thermal behavior, static and dynamic, is important. Nowadays, the most popular transistor models for nonlinear Monolithic MICs (MMICs) CAD are empirical nonlinear analytic models, also known as compact models, like the VBIC model. They offer good computational efficiency and are easily implemented in commercial circuit simulators. Unfortunately, they entail the extraction of numerous parameters, from structured sets of device measurements, involving complex and time-consuming optimization procedures. Other modelling alternatives have been proposed. For instance, """"black box"""" model approaches, like the time-domain table-based and frequency-domain behavioral ones, allow for direct extraction but require more exhaustive experimental characterizations. For the extraction of any of these nonlinear models, not only the adequate microwave measurement system, but also the corresponding control software, are mandatory to perform the necessary devices/circuits nonlinear characterizations. Conventional Vector Network Analyzers (VNAs) allow for linear devices characterizations by means of scattering parameters (S-parameters), but are not adequate for nonlinear characterization, unless suitably upgraded. To overcome this limitation, several nonlinear measurement systems have been developed in recent years, under the denomination of Nonlinear Vector Network Analyzers (NVNAs). Among the NVNAs (LSNA, PNA-X, VTD SWAP-X402, to cite a few), the Large Signal Network Analyzer (LSNA) was the first commercially available receiver allowing fully nonlinear characterizations up to 20 GHz. In this context, the main objective of this thesis work has been to implement, through the use of a LSNA-based nonlinear measurement system, nonlinear characterization and model extraction techniques of microwave active devices, for CAD design of nonlinear microwave circuits (oscillators and RF power detectors, among others). In particular, in this work it has been used an LSNA model MTT4463B by Maury-NMDG, with a microwave bandwidth from 600 MHz to 50 GHz, and upgraded during this thesis to include a low-frequency bandwidth from 10 KHz to 24 MHz. Different measurement systems configurations have been set up for the different type of device characterizations required in this work. The corresponding hardware/software configurations implemented are described, as well as the different transistor and diode characterizations developed: DC, small-signal, single and multi-tone CW large-signal, hybrid load-pull measurements, X-parameters measurements, base-band measurements, etc. The first step has been the setting up of the LSNA-based large-signal measurement system, for active device nonlinear characterization and model extraction, and to study the consistency of the large-signal measurements performed in different NVNAs based systems. Measurements from two LSNA-based systems and one sampling oscilloscope based, enhanced with active load-pull capabilities, were compared. From the study it was concluded the importance of matching the system impedances, at fundamental frequency and harmonics, of the three systems to guarantee the consistency of the measurements for the same active device. Efficient extraction procedures have been developed of both compact and black-box nonlinear models, specifically of microwave bipolar transistors and diodes from different semiconductor technologies. III-Vs and SiGe heterojunction bipolar transistors (HBTs), both on-die and packaged, as well as packaged Schottky diodes have been used in this work. On this context, two new techniques have been proposed in this work to enhance VBIC model extraction for HBTs. The first proposed method is based on applying sensitivity analysis techniques to guide the model parameters extraction, through selective sweeps of the dominant parameters. This method improves the accuracy of the extracted model with respect to conventional extraction procedures, and forces the consistency of the obtained set of model parameters. The second proposed technique makes it possible to extract the VBIC current and thermal model parameters exclusively from DC measurements at a single ambient temperature, hence, avoiding the use of the expensive thermally controlled measurement facilities, usually required with the conventional method.