Vijaya K. Devabhaktuni

Neural Networks for Microwave Modeling: Model Development Issues and Nonlinear Modeling Techniques

() ABSTRACT: Artificial neural networks ANN recently gained attention as a fast and flexible vehicle to microwave modeling and design. Fast neural models trained from measured simulated microwave data can be used during microwave design to provide instant answers to the task they have learned. We review two important aspects of neural-network-based microwave modeling, namely, model development issues and nonlin- ear modeling. A systematic description of key issues in neural modeling approach such as data generation, range and distribution of samples in model input parameter space, data scaling, etc., is presented. Techniques that pave the way for automation of neural model development could be of immense interest to microwave engineers, whose knowledge about ANN is limited. As such, recent techniques that could lead to automatic neural model development, e.g., adaptive controller and adaptive sampling, are discussed. Neural model- ing of nonlinear device circuit characteristics has emerged as an important research area. An overview of nonlinear techniques including small large signal neural modeling of () transistors and dynamic recurrent neural network RNN modeling of circuits is presented. Practical microwave examples are used to illustrate the reviewed techniques. 2001 John Wiley & Sons, Inc. Int J RF and Microwave CAE 11: 421, 2001.

Neural Network Structures and Training Algorithms for RF and Microwave Applications

Neural networks recently gained attention as fast and flexible vehicles to microwave modeling, simulation, and optimization. After learning and abstracting from microwave data, through a process called training, neural network models are used during microwave design to provide instant answers to the task learned. Appropriate neural network structure and suitable training algorithm are two of the major issues in developing neural network models for microwave applications. Together, they decide amount of training data required, accuracy that could possibly be achieved, and more importantly developmen- tal cost of neural models. A review of the current status of this emerging technology is presented, with emphasis on neural network structures and training algorithms suitable for microwave applications. Present challenges and future directions of the area are discussed. Q 1999 John Wiley & Sons, Inc. Int J RF and Microwave CAE 9: 216)240, 1999.

A hierarchical neural network approach to the development of library of neural models for microwave design

Neural networks recently gained attention as a fast and flexible vehicle to microwave modeling simulation and optimization. This paper addresses a new challenge in this area, i.e., development of libraries of microwave neutral models. A hierarchical neural network framework is presented utilizing the knowledge of basic relationships common to all library components. The proposed method improves the reliability of neural models, while significantly reducing the cost of library development through reduced need for data collection and shortened time of training.

A Neural Network Approach to the Modeling of Heterojunction Bipolar Transistors from S-Parameter Data

Artificial neural networks have gained attention as a fast, efficient, flexible and accurate tool in the areas of microwave modeling, simulation and optimization. In this paper, a novel neural network approach is proposed for the modeling of Heterojunction Bipolar Transistors (HBT) directly from their S-Parameter data. The neural network structure incorporates bias current and bias voltage as inputs. This enables us to use the same neural model under different bias conditions. The proposed technique provides reliable neural transistor models, while significantly reducing the cost effort and complexity involved in the modeling of HBT.

Robust training of microwave neural models

Neural networks have recently gained attention as a fast and flexible vehicle for microwave modeling, simulation and optimization. A new training algorithm based on Huber-norm and quasi-Newton optimization is proposed. The Huber quasi Newton (HQN) algorithm can robustly train a neural network in the presence of large errors in training data. A multi-stage training algorithm that incorporates the HQN technique and an adaptive macro-training process, is proposed to address highly nonlinear and non-smooth modeling problems. The advantages of the proposed microwave-oriented neural network techniques are demonstrated through examples.

An adjoint dynamic neural network technique for exact sensitivities in nonlinear transient modeling and high-speed interconnect design

We propose a new adjoint dynamic neural network (ADNN) technique aimed at enhancing computer-aided design (CAD) of high-speed VLSI modules. A novel formulation for exact sensitivities is derived employing the Lagrange functions approach, and by defining an adjoint of a dynamic neural network (DNN), for the first time. The proposed ADNN is a dynamic model that we solve using integration backwards through time. One ADNN solution can be used to efficiently compute exact sensitivities of the corresponding DNN with respect to all its parameters. Using these sensitivities, we developed a training algorithm that facilitates DNN learning of nonlinear transients directly from continuous time-domain waveform data. Resulting accurate and fast DNN models can be straightaway used for carrying out high-speed VLSI CAD in SPICE-like time-domain environment. The technique can also speed-up physics-based nonlinear circuit CAD through faster sensitivity computations. Applications of the proposed ADNN technique in transient modeling and nonlinear design are demonstrated through high-speed interconnect driver examples.

Neural Network Training-Driven Adaptive Sampling Algorithm for Microwave Modeling

We present a neural network training-driven adaptive sampling algorithm for efficient generation of training and test data. The proposed approach makes microwave data generation an integral part of model development/training. For user-specified model accuracy, the algorithm periodically communicates with the neural network training process and automatically determines the number of samples required and their distribution in the model input space. The algorithm has an inherent ability to distinguish nonlinear and smooth regions of model behavior. Consequently, more samples are generated in nonlinear regions improving model accuracy, and redundant data is avoided in smooth regions reducing model development cost.

A Hybrid Neural and Circuit-Based Model Structure for Microwave Modeling

Neural networks have recently gained attention as powerful vehicles to microwave modeling, simulation, and optimization. A hybrid neural network structure incorporating prior circuit knowledge is proposed for modeling microwave components. In the proposed structure, a sub neural network establishes the mapping between original model input space and approximate circuit model input space. The neural network can learn such complicated space-mapping by training with EM simulation data. The hybrid neural models are computationally efficient and have an accuracy that is comparable to EM simulation. The proposed methodology is demonstrated through practical microwave modeling examples.

An iterative multi-stage algorithm for robust training of RF/microwave neural models

Neural networks recently gained attention as a fast and flexible vehicle to microwave modeling and design. In this paper, we propose an iterative multi-stage (IMS) algorithm to address certain key challenges in neural network based microwave modeling. The IMS decomposes the original complicated microwave behavior into several simpler portions. Each of these simpler portions is modeled separately in a different stage, by training a suitable neural network structure. Neural models from different stages are combined iteratively to produce the overall neural model that represents the original microwave behavior. The proposed technique is demonstrated through examples.

Huber optimization of neural networks: a robust training method [microwave modeling]

Neural networks as an emerging modeling technique have gained much attention in the microwave area. Due to the convergence difficulty of simulators or equipment limits where parameters are sampled at extremes, the simulated or measured training data often have both gross errors and small errors. A new training method is presented in this paper which incorporates the Huber concept into a quasi-Newton method. The proposed method can recognize the gross errors and small errors and treat them differently. Therefore this Huber training method is much more robust than traditional least-square l/sub 2/ methods, which is demonstrated through two examples, modeling of a quadratic function and transmission lines.