Journal of Engineering Science and Technology Review 9 (4) (2016) 74 - 76
Research Article
A Novel Modeling Technique for Operational Amplifier Using RBF-ELM
B.Shivalal Patro
1,
* and Sushanta K. Mandal
2
1
School of Electronics Engineering, KIIT University, India
2
Department of Electronics and Telecommunication, CUTM University, India
Received 22 June 2016; Accepted 9 September 2016
___________________________________________________________________________________________
Abstract
This paper formulates a new modeling technique for the analog circuits like OPAMP (Operational Amplifier). One of
the fastest learning techniques, extreme learning machine along with best suitable kernel Radial Basis Function has been
implemented for modeling. The simulated result shows that the training speed is very high and can handle a large set of
data without compromising accuracy.
Keywords: Analog Circuit, Gaussian Kernel, Machine learning, Feed-Forward Network, Mean Square Error.
__________________________________________________________________________________________
1. Introduction
In recent years, researchers have shown greater interest in
the modeling of circuits as the designing of circuits has
come into a level of saturation. So, designers are taking the
help of traditional design systems to integrate as well as
modified forms to enhance the performance of the circuits.
The modeling of analog, digital and mixed signal circuits
has been carried out since several decades ago and presently
the designers are depending on these modeling techniques to
formulate some new designs [1, 2]. These formulations will
help the designers in scaling the designs to its lower level by
compromising the specifications into very less margin and
hence can improve the efficiency of the circuit into a greater
extent. The modeling of the circuits can be done on various
aspects like behavioral form, analytical form, symbolic
form, etc., without going into the details how each
component is affecting the behavior of the circuit. That
means the designer need not to be an expert in circuit design,
he can analyze the relation between the output variations in
accordance with the input.
So, on the basis of mapping of the analog circuits of
various models, many modeling techniques have been
formulated [3]. To them various machine-learning
algorithms are available which are able to map these circuits
with reliable models which work as similar to the circuits [4,
5]. Machine learning is an area of artificial intelligence
where various algorithms are formulated to develop various
systems which can be able to perform classification,
clustering, regression and many other activities. The
prerequisites for these learning algorithms are the data which
will help the system to develop a model which can be able to
make predictions. So, in the conceptual modeling of analog
circuits, RBF-ELM can play one of the best modeling
techniques.
2. Proposed Technique
Huang et al. [6] in 2004 proposed Extreme Learning
Machine to provide a speedier approach to the neural
networks for analyzing big data. In recent years, ELM
(Extreme Learning Machine) has become one of the most
preferable alternatives in machine learning algorithms.
These machines are working on the principle SLFNN
(Single Layer Feedforward Neural Network). SLFNN is an
extension of neural network, which can be used to estimate a
function from a given set of inputs and their corresponding
outputs. It relates the inputs by choosing random weights
with the outputs within an error
β results to a trained
network. For learning and tuning the input parameters slow
gradient method is used. An activation function is provided
to tune the network accordingly on the finite number of
inputs and number of hidden neurons [7, 8].
ELM is mainly used for the problems where
classification, clustering, regression are the main objectives
of the systems. As they contain a single layer of hidden
nodes or neurons where the weights connecting inputs to
hidden neurons are randomly assigned. These assigned
nodes are never trained or updated unlike other machine
learning techniques, especially ANN (Artificial Neural
Network) [9]. So, these hidden neurons, which are
connected to the outputs, are learned in a single step. The
fast training algorithm proceeded from non-iterative method
is an aisle to the domain of large data set problems.
It has been proven that when the sample of data
approaches infinity the error between the estimated and
actual result becomes zero [7]. So it becomes a very
important factor regarding the variables which are used as
inputs and their respective estimated outputs. So the
designer who is training the ELM must preanalyze the data,
whether there is any aberration or not. Adding RBF kernel to
ELM will help the circuit designers achieve some vital
design parameters to formulate new specification based
circuits. If these things are not taken into consideration then
the model may approach to over-fitting condition which will
Jestr
JOURNAL OF
Engineering Science and
Technology Review
www.jestr.org
______________
* E-mail address: shivalalpatro@gmail.com
ISSN: 1791-2377 © 2016 Eastern Macedonia and Thrace Institute of Technology. All rights reserved.