This paper demonstrates the Cardiff behavioral model based on the large-signal response of a single pole, double throw switch (SPDT) to realize a tunable reactive termination for advanced power amplifier architectures operating in the Ka-band up to 3 watts. The model was developed using a comprehensive load-pull (LP) characterization dataset comprising 2176 data points. Subsequently, it was trained to replicate this dataset with only 44 input data points, achieving an acceptable accuracy reflected by an NMSE of -33.8 dB.

The Cardiff behavioral model is a robust nonlinear tool for guiding nonlinear circuit design. This paper proposes a methodology to generate Cardiff AB models suitable for microwave frequency doubler optimization. Active Load-Pull data from both measurement and simulations are processed through a self-developed MATLAB application to extract and evaluate Cardiff AB models formulated with different mixing orders and number of parameters. Moreover, the role of harmonic source and load impedances in model extraction for frequency doubler design space prediction is studied.

This paper presents a novel experimental technique for automatically identifying the complexity and coefficients of a Cardiff behavioral model of a microwave transistor using a conventional, narrowband active load-pull system. The method ensures the accuracy of the extracted model while eliminating the need for expert human judgment/intervention. The paper details the solutions adopted to overcome the technical challenges of implementing A-pull using a narrowband vector network analyzer-based load-pull system. Specifically, to ensure that the A-pull grid is achieved quickly and accurately, and that it covers a meaningful and safe operating space for the device under test. A gallium nitride (GaN) microwave transistor is characterized and modeled to demonstrate the technique at 2.45 GHz. Results clearly show how the model complexity is automatically identified and accurate coefficients extracted. In addition, the paper demonstrates how to use this approach to allow for a systematic reduction in the number of measured load points without compromising model accuracy, further improving the process’s speed.

A previously introduced Iterative Envelope Simulator is expanded upon. The new implementation uses a Matlab-controlled ADS circuit simulator, allowing for the simulation of manufacturer provided nonlinear transistor models. The model is validated with 2-tone and 256-Quadrature Amplitude Modulation (256-QAM) stimuli applied to a power amplifier with a resonant BB network. Its output is compared against 2-tone HB simulations and ADS envelope simulations. The 2-tone results show that the improved simulation captures the same trend as the HB simulator. The new model introduces the ability to predict signal linearity for complex signal modulation schemes. It is capable of predicting the issues related to the resonant networks for the 256-QAM signal, showing a significant Error Vector Magnitude (EVM) degradation when the signal envelope starts overlapping the resonant frequency, while the traditional envelope simulator does not capture this issue.

This article investigates sources of calibration error inherent in the load-pull (LP) measurement of millimeter-wave transistors when high reflection coefficient loads are applied. In these conditions, the calibration error on relative metrics such as power gain and power added efficiency (PAE) can be significant and strongly dependent on the reflection coefficient. Using both simulations and experiments, the article tries to discriminate the source of uncertainty between noise and "operator" errors. While the former can be minimized by maximizing the dynamic range and improving the filtering on the receiver, the latter, linked to the small mechanical perturbations that occur when changing the system configuration from calibration to measurement, is unavoidable in most LP systems. To address this issue, the article proposes a method, based on the TRL calibration, that consists of load-pulling the thru and line calibration standards using the measurement system in its measurement configuration and calculating their scattering parameters using least-squares on the measured dataset. This enables a recalculation of the calibration coefficients previously obtained when the measurement system was configured for calibration and allows for a significant reduction in calibration error. This method has been tested and applied to measurements of mm-wave devices in both active and passive LP systems, demonstrating a significant impact on the measured performance metrics. Calibration error, at 82.5 GHz, is shown to have resulted in a power gain inaccuracy as high as 0.7 dB at a reflection coefficient of 0.7, which led to an underestimation of PAE of 8.9% for a gallium arsenide pHEMT.

A new baseband (BB) impedance cancellation technique is presented for broadband multitransistor amplifier designs. By placing a BB transformer in between the Doherty main and auxiliary amplifier, the BB component of a two-tone signal can be canceled out, resulting in improved efficiency, output power, and linearity over two-tone bandwidth. A 20-W, 6-dB back-off 3.5-GHz Doherty is designed and its performance is measured with and without this cancellation technique. The prototype shows significant improvements for all metrics for tone spacing above 130 MHz when the transformer is used, and it has an efficiency enhancement up to 15% points at 300-MHz tone spacing.

In this work, hybrid GaN-based single-ended and balanced active frequency doubler prototypes, from 2.45 to 4.9 GHz, have been designed. To minimize the balanced topology area, the Moore space-filling curve has been used in the design of the 180 degrees hybrid coupler. Conversion gain exceeds 10 dB in all the measured single-ended prototypes and no amplification stage was required. Balanced design obtained in simulation 11.3 dB of conversion gain and 26.3 dBm of output power whereas the fundamental frequency suppression exceeds 38 dBc.

This letter presents an advanced Cardiff Model (CM) extraction method based on artificial neural network (ANN) techniques that ensures reasonable extrapolation behavior. The nonphysical extrapolation behavior occurring with CMs using a high, user-defined, mixing order, i.e., false output optima point and erroneous efficiency behavior, when extracted using measured load-pull datasets can be avoided with the proposed method. The method proposed maintains the accuracy within the load-pull measurements, design-relevant, impedance space. The method was verified by modeling the measured load-pull data of a Wolfspeed 10-W gallium nitride (GaN) packaged device and a WIN Semiconductors GaN on-wafer device. With both the devices, the extrapolation issues shown when using the high-order CM are removed by the novel extracted CM coefficients.

This paper shows the ability of the Cardiff Behavioural Model to precisely reproduce the response of an advanced power ampli.er architecture. The investigation involved varying the phase of the complex load on the isolated port of a 50W orthogonal load-modulated balanced ampli.er operating at 2.5GHz. Leveraging a limited set of data points, the Cardiff model effectively estimates the PA's performance across a wide range, from approximately 10 dB output backoff to around 1-2 dB of gain compression, with a maximum normalised error of only -20 dB. This novel application of the Cardiff Model offers an ef.cient means of reducing measurement and simulation time for designing and optimizing this particular type of power ampli.er.

This article introduces a novel artificial neural network (ANN) structure determination process based on the Cardiff model (CM), to determine ANN-based transistor nonlinear behavioral models. By relating the CM formulation and coefficients to the Taylor series expansion of the ANN model, a novel approach for determining the required values of a fully connected cascaded (FCC) ANN structure has been formulated. The proposed method provides the chance to escape from the possible time-consuming ANN determination process. Experiments proved that the proposed ANN models using the determination method can provide accurate prediction for the behavior acquired from load-pull characterizations of a Wolfspeed 10-W packaged gallium nitride (GaN) high electron mobility transistor (HEMT) simulation at 3.5 GHz, and a dense load-pull measurement of WIN NP12 $4$ $\times$ $75$ $\mu$ m GaN HEMT at 20 GHz, with normalized mean square error (NMSE) levels lower than $-$ 40 dB.

This paper presents a simplified simulation method for estimating the impact of baseband terminations in advanced power amplifier architectures when used with wideband signals. Previous work has introduced the idea of using the baseband signal as a separate domain from the RF signal in order to estimate continuous wideband power amplifier performance with different baseband terminations. In this paper, this idea is expanded, the underpinning equations are generalized, and a new circuit simulator is designed utilizing the idea of baseband, or envelope, domain. The results are compared to 2-tone simulations using a commercial harmonic balance solver. The new envelope model appears to be able to predict issues with baseband impedance matching for wideband signals. It allows for efficient simulations of complex wideband signals without the escalation of harmonics as required for N-tone harmonic balance solvers, and allows for more complex amplifier architectures, such a Doherty power amplifier, to be studied.

This paper describes a large-signal single-sweep characterization system based on vector network analyzer receivers for on-wafer harmonic load/source pull measurements up to 110 GHz using passive tuners, and waveform measurements up to 100 GHz using an oscilloscope as a phase meter. The calibration and measurement procedures are described and validated with thru structures and on GaAs HEMTs at Ka-band demonstrating the capability to offer an important insight for both technology developers and designers of millimetre-wave transistors and amplifiers.

This paper presents a novel way to extract the Cardiff model coefficients. Analysis shows that the Cardiff model is easier for implementation in commercial RF design related software, while the Artificial Neural Network (ANN) based behavioral model provides more flexibility for extracting the coefficients of a model, the two methods have been combined. The load-pull setup, implemented in Keysight Advanced Design System (ADS), is used to acquire the pseudo-wave data of the Cree 10W transistor. By using the feedforward ANN structure together with a novel Cardiff model related backpropagation algorithm, a set of accurate coefficients can be extracted, whose Normalized Mean Squared Error (NMSE) is -54.73 dB as shown in this paper.

Recent advances in RFPA design make extensive use of active load modulation techniques. These architectures can be made more effective by the insertion of electronically tuneable reactive elements. This work describes the measurement and modeling of microwave switches aimed at this application. Active open-loop load-pull measurements were performed on microwave switches with different biasing voltages and power levels. Using load-pull measurement data, the devices are modeled based on Cardiff behavioral model’s mathematics expressing good accuracy in reproducing the switch response in the linear and compression regions.

This paper explores the use of the Cardiff non-linear behavioral model to characterize the response of multiple-input power amplifiers. In particular, a case study is presented on a 300 W load modulated balanced amplifier operating at 2.1 GHz. The model mathematical formulation is presented, and the comparison between original data and model shows an error below 3%. More importantly, it is shown that the model can accurately interpolate between characterization points allowing a reduction of up to 96% of the points needed to accurately predict the model behavior. This significantly reduces the simulation and measurement time for multiple-input PA's whilst attempting to determine the optimal driving conditions.

Breakdown mechanism in 0.25- ${\mu } \text{m}$ gate length AlGaN/GaN-on-SiC iron doped high electron mobility transistors (HEMTs) with background carbon is investigated through the drain current injection technique. The measurement results reveal that it can be divided into two distinct stages according to the gate voltage levels. The first stage of the measured drain injected breakdown is mainly due to the initiation of the punchthrough process under the gate, and the second stage of breakdown is associated with the potential barrier between the unintentionally doped (UID) GaN and the Fe doped p-type GaN buffer layer which also has a higher carbon density. The electroluminescence (EL) results suggest that the first stage shows uniform punchthrough current flow, but localized leakage current flow associated with a snapback breakdown mechanism replaces the uniform punchthrough current flow and dominates the second stage. A 2D-TCAD simulation has been implemented and shows the current paths under uniform flow conditions.

In the framework of Power Amplifier (PA) design for communications, frequency domain non-linear behavioural models have shown their potential as efficient complementary modelling tools when Field Effect Transistor compact models are not available or sufficiently accurate. The Admittance behavioural model, formulated in the V-I domain, is especially suitable for device size and fundamental frequency scaling. It is important to note that the direct extraction of this model, from the Nonlinear Vector Network Analyser (NVNA) load-pull (LP) measurements, requires some extra processing since it necessitates a Look-up-Table indexed to |V-11| rather than |A(11)|. When using such models in PA design, there is the need for the user to select the necessary model complexity. To address this requirement, in this paper, a systematic analysis methodology, to guide the user, is presented and validated in different PA design scenarios. The methodology was tested using NVNA LP measurements of GaN Heterostructure FETs. A fifth order Admittance model formulation showed good accuracy in the studied PA design scenarios.

This paper presents an analysis of load-pull measurement data of a GaN on SiC high electron mobility transistor (HEMT) at 3.6 GHz, oriented to aid the design of Doherty power amplifiers with multiple DC supply voltages. Fundamental load-pull data in class AB configuration are analyzed to realize the optimum load modulation for the design of the "Main" amplifier at DC bias levels of 28V and 50V.

This paper presents analysis of Fourier transferred dataset to accurately extract a DC-dependent Cardiff nonlinear behavioral model. The analysis is based on simulation data of a 10 W GaN high electron mobility transistor (HEMT) at 3.5 GHz. Tailored load-pull data were collected over a modulated gate and drain voltages ranging from −3.0 V to −2.0 V and 15 V to 30 V, respectively. The extracted model coefficients are capable of accurately predicting the load-pull data with normalized-mean-square-error (NMSE) value of less than −40 dB.

This paper analyzes the dependence vs. gate bias voltage of the coefficients of the Cardiff model in the admittance form. The load-pull measurement data used to extract the model, inclusive of input power sweep, is taken on a GaAs pseudomorphic high electron mobility transistor (pHEMT) at the frequency of 36 G Hz. The gate bias is swept in class C and in class AB and a different set of coefficients is extracted at each bias point. It is observed that the model coefficients can be fitted vs. bias using a linear function within the class C and class AB ranges. This allows to predict the model coefficients within a range of bias voltages with load-pull measurements at only a few bias points, significantly reducing the measurement effort. Using the predicted coefficients, the model shows a global error lower than -31 dB for the DC and fundamental output current.