George Stantchev

Generalized Transmitter Compensation of Frequency Dependent I/Q Imbalance

This paper considers the formulation of a digital pre-distorter to eliminate distortions that are generated by a frequency-dependent imbalance between the analog in-phase (I) and quadrature (Q) channels of a direct conversion transmitter. Estimation of the imbalance occurs in the frequency domain, while compensation occurs in the time domain. We demonstrate the efficacy of the approach by estimating and eliminating transmitter distortions whose digital-to-analog converters (DACs) and filters have differing impulse responses, as well as a phase and amplitude imbalance between the local oscillators (LOs). The estimator developed includes a separation of the linear and nonlinear transmitter response, facilitating the application of nonlinear pre-distortion without architectural restriction. Furthermore, the estimator not only compensates for I/Q imbalances, but fully characterizes the frequency response and identifies the source of the imbalance. An interesting outcome of the analysis is that it may not be possible with baseband processing to compensate for an imbalance that occurs after up-conversion but prior to I/Q summing.

I/O HSMM: Learning behavioral dynamics of a cognitive wireless network node from spectrum sensing

We introduce a generative model, dubbed I/O HSMM, for learning the bi-modal behavioral dynamics of a network of cognitive radios (CRs). Each of the two modes of the CRs is represented as a Hidden Semi-Markov model (HSMM), where the states, state durations and emissions, transition probabilities between states, and transitions between modes are uncovered based solely on RF spectrum sensing. The learning of the CR dynamics is non-parametric and derived from the Hierarchical Dirichlet Process (HDP), with the switching between the two modes modeled as a latent variable that is estimated as a part of the learning process. The non-parametric model provides flexibility in handling unknown communication protocols. We evaluate the quality of learning against ground truth, and demonstrate that this approach is promising and merits extension to more complex models.

High performance parametric design optimization of RF devices

We present an integrated environment for large scale multi-parameter design optimization of RF devices based on AFRLs Galaxy Simulation Builder productivity tool for distributed high-performance computing, Sandia National Labs DAKOTA optimization library, and a suite of highly efficient GPU-based Electromagnetic codes developed at NRL in collaboration with Leidos, Inc. The environment allows for an end-to-end optimization cycle of an RF device to be set up, deployed, carried out, monitored and analyzed in a quick, user-friendly, robust, and flexible fashion using a diverse variety of high-end parallel computing resources.

Developments in parallelization and the user environment of the MICHELLE charged particle beam optics code

The next generation of the MICHELLE ES PIC code is to improve its parallelization and leverages a number of existing and emerging DOD HPC architectures and software including distributed memory clusters, multicore, and computational accelerators such as GPUs and Intel Xeon Phi co-processors. The ongoing project supported by the DOD HASI program also aims to build interfaces between MICHELLE and existing HPC tools such as CAPSTONE, GSB, ParaView, and VisIt for efficient design and optimization workflow. This paper reports on the latest progress and discusses applicable algorithms and implementations.

A high-performance distributed computing framework for parametric design optimization of RF devices

The design cycle of RF devices is greatly facilitated by the use of the “virtual prototyping” methodology based on high-fidelity computer simulations that are capable of predicting the RF devices performance in response to changes in its physical parameters. In particular, critical dimensions of the structure, or quantitative properties of the various electromagnetic components are routinely utilized in sensitivity analyses coupled with performance optimization. This type of process is well suited to semi-supervised global optimization. To this end we have integrated our RF simulation codes with several existing software tools for optimization and distributed high-performance computing code deployment and management, such as the DAKOTA toolkit [1], and AFRLs Galaxy Simulation Builder (GSB).

Modeling and simulation of millimeter wave vacuum electronic devices at the naval research laboratory

Recent advances in the development of millimeter wave vacuum electronic devices have been made possible by powerful, specialized design tools. We summarize them here and point out areas where improved models and codes will be needed.

Vacuum electronic device design using 3D EM-PIC

We present new capabilities introduced in the 3D Electromagnetic Particle-in-Cell code Neptune to directly support rapid simulation-based design of a broad range of vacuum electronic devices.

Parallel parametric design optimization for RF amplifiers with 3D EM-PIC

Three-dimensional electromagnetic Particle-in-Cell (3D EM-PIC) codes offer an advantage in the process of RF amplifier design since they are capable of highly-accurate, device-independent simulation and prediction of the amplifiers performance and physical characteristics. Achieving the desired amplifier performance generally requires optimization of devices geometry and operating parameters with respect to a specific set of criteria. The RF amplifier design cycle typically involves many simulations with varying parameters that are carried out until such criteria are met. Traditionally, the computational intensity and prohibitively long runtimes of 3D EM-PIC codes have limited their practical application to the case of manually controlled parametric optimization of RF amplifiers.

RF amplifier design using 3D EM-PIC

Summary form only given. The 3D Electromagnetic Particle-in-Cell (EM-PIC) algorithm provides a powerful method to predict performance of RF amplifiers based on actual device geometry and defined operating parameters. Until recently its utility as a primary design tool has been restricted by long simulation times or limited accuracy, due to computational and/or algorithmic constraints of available implementations. Advances in the representation of geometry using conformal techniques, coupled with highly parallel implementations of the EM-PIC algorithm, have made EM-PIC viable as a primary design tool. The EM-PIC code Neptune1, 2 exploits highly parallel GPU hardware in conjunction with conformal geometry representation to perform fast, accurate amplifier simulations. It is now a primary tool at NRL for the design and simulation of single- and multi-stage TWT amplifiers (serpentine waveguide, folded waveguide, helix, transverse TWT), klystrons and EIK amplifiers, and has enabled new research into advanced device concepts including cascaded multi-beam TWTs and transverse-interaction TWT amplifiers. We report on recent additions to Neptunes algorithms that directly support the needs of RF amplifier design. First, a new method to improve the discretization accuracy for dielectric regions will be demonstrated, based on a novel subcell averaging scheme. Second, we report on a range of new features supporting device design. New particle beam and magnetic field import facilities allow interfacing with complementary design tools, including the gun/collector code MICHELLE3, while new diagnostics for RF current and beam interception allow detailed monitoring of device operation.

Hybrid time-domain measurement and pre-distortion of broadband complex waveforms in a Ka-band TWT amplifier

We present a technique for time-domain measurement and predistortion of complex wideband waveforms in a Ka-band helix TWT power amplifier. The measurement technique relies on a hybrid calibration procedure that uses both single tone and wideband multitone calibration for characterizing the amplifier response to simultaneous excitation across its operating frequency band. We compare our measurements with simulation data from a well-established physics-based, large signal, multi-frequency spectral code, CHRISTINE [1].