ITEM METADATA RECORD
Title: Black-box Modelling of Analogue and Mixed-Signal Circuits in the Time Domain (Zwarte-doos modellering van analoge en gemengd analoog-digitale schakelingen in het tijdsdomein)
Other Titles: Black-box Modelling of Analogue and Mixed-Signal Circuits in the Time Domain
Authors: Ceperic, Vladimir; S0210511
Issue Date: 28-Jun-2013
Abstract: Circuit simulations at the system level are one of the most complex tasks that engineers encounter in the field of electronics and microelectronics due to two main reasons: the low speed and limited accuracy of circuit simulations. The black-box approach to behavioural modelling of electronic circuits, although very challenging, is particularly interesting for fast and relatively accurate simulations of analogue and mixed-signal integrated circuits. Today’s most prevalent approach is manual generation of the black-box models, which is heuristic, inconvenient, computationally expensive and error-prone. Automated model-building has many potential benefits. It is however very difficult to develop a fully automated model generation procedure because the model has to cover a wide range of circuits and devices. In this thesis several methods are proposed that can be regarded as essential blocks needed for the automated generation of black-box models. The new methods of functional approximation (ALSVR, TASVR and MK-ALSVR) suitable for the black-box modelling of electronic circuits are proposed. The new algorithm FTSR is proposed for model inputs selection and ranking as well as training data points selection and ranking, designed specifically for black-box modelling of electronic circuits. Also, a new method for checking and improving the stability of black-box electronic circuit models (CISB) is presented. It enables models built from the proposed behavioural modelling procedures to be effectively implemented in common circuit simulation tools. A new machine learning approach to modelling of conducted electromagnetic emissions and conducted electromagnetic immunity is proposed. Finally, the framework for behavioural modelling of electronic circuits based on the methods proposed is proposed in this thesis.
Table of Contents: Abstract iii
List of Figures xxi
List of Tables xxvii
1 Introduction 1
1.1 Introduction and motivations . . . . . . . . . . . . . . . . . . . . 1
1.2 Behavioural modelling procedure . . . . . . . . . . . . . . . . . 3
1.2.1 Brief comparison of basic behavioural modelling approaches 3
1.2.1.1 Black-box modelling . . . . . . . . . . . . . . . 4
1.3 Main obstacles and challenges . . . . . . . . . . . . . . . . . . . 7
1.4 The scope and the objectives of the dissertation . . . . . . . . . 9
1.5 Outline of the dissertation . . . . . . . . . . . . . . . . . . . . . 10
1.6 Major contributions of the dissertation . . . . . . . . . . . . . . . 11
2 New function approximation methods suitable for the modelling of
electronic circuits 15
2.1 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
xv
2.2 An overview of the state-of-the-art function approximation
methods for the black-box modelling of electronic circuits . . . 17
2.3 Support vector regression machines . . . . . . . . . . . . . . . . 19
2.3.1 Optimisation of the SVM hyper-parameters . . . . . . . 23
2.3.2 SVMs for the black-box modelling of electronic circuits 23
2.3.3 Sparse support vector regression by active learning . . . 24
2.3.3.1 Sparse support vector regression models . . . . 25
2.3.3.2 Sparse support vector regression by using active
learning principle . . . . . . . . . . . . . . . . 26
2.3.3.3 Training time requirement . . . . . . . . . . . 30
2.3.3.4 Modelling experiments . . . . . . . . . . . . . 32
2.3.3.4.1 Boston housing data set. . . . . . . . 33
2.3.3.4.2 Concrete compressive strength data set. 36
2.3.3.5 Discussion - ALSVR . . . . . . . . . . . . . . . 40
2.3.4 Sparse multikernel support vector regression machines
trained by active learning . . . . . . . . . . . . . . . . . . 41
2.3.4.1 Illustrative example of the model training
algorithm . . . . . . . . . . . . . . . . . . . . . 45
2.3.4.2 Training time requirement . . . . . . . . . . . 47
2.3.4.3 Experiments - comparison with the state-of-the-art 48
2.3.4.4 Discussion - MK-ALSVR . . . . . . . . . . . . 50
2.3.5 Recurrent sparse support vector regression machines
trained by active learning in the time domain . . . . . . . 51
2.3.5.1 Training time requirement . . . . . . . . . . . 56
2.3.5.2 Modelling experiments . . . . . . . . . . . . . 57
2.3.5.2.1 Multiple superimposed oscillations. . 57
2.3.5.2.2 10th-order nonlinear autoregressive mov-
ing average system benchmark task. . 60
2.3.5.3 Discussion - TASVR . . . . . . . . . . . . . . . 64
2.4 Application of the proposed function approximation method to
black-box circuit modelling . . . . . . . . . . . . . . . . . . . . 65
2.4.1 Test case - voltage bandgap reference with offset compen-
sation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
2.4.1.1 TASVR model vs. SVR model . . . . . . . . . 66
2.4.1.2 ALSVR and MK-ALSVR model vs. SVR model 69
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3 Automated generation and selection of the electronic circuit black-
box model inputs 75
3.1 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2 Stimulating the circuit to obtain an optimal training data set . 77
3.3 Pre-processing of the model inputs and outputs . . . . . . . . . 78
3.4 Selection of the model inputs and training data points . . . . . . 81
3.4.1 Brief survey of the literature . . . . . . . . . . . . . . . 82
3.4.2 Feature and training data points selection and ranking
algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.3 Application of the proposed algorithm for the automated
generation and selection of the electronic circuit black-box
model inputs . . . . . . . . . . . . . . . . . . . . . . . . 89
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4 Stability of behavioural electronic circuit models in circuit simulators 95
4.1 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2 Method for checking and improving the stability of black-box
models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.3 Application of the proposed method for checking and improving
the stability of black-box models within electronic circuit simulators 101
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5 Black-box modelling by classification of conducted electromagnetic
immunity and emissions using support vector machines 105
5.1 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.2 EMC Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.2.1 Conducted EMC measurement methods . . . . . . . . . 108
5.2.1.1 Conducted EMI measurements according to
IEC 62132-4 . . . . . . . . . . . . . . . . . . . 108
5.2.1.2 Conducted EME measurements according to
IEC 61967-4 . . . . . . . . . . . . . . . . . . . 108
5.3 Support vector machines for classification . . . . . . . . . . . . 110
5.3.1 SVM classifier with soft margin . . . . . . . . . . . . . . 112
5.3.2 Dual formulation . . . . . . . . . . . . . . . . . . . . . . 113
5.3.3 Nonlinear SVM classifier . . . . . . . . . . . . . . . . . . 114
5.4 Classification-based model of IC-level conducted EMI and EME
by SVMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.4.1 Time-domain modelling approach (TDMA) . . . . . . . 116
5.4.1.1 TDMA-EMI-LOGIC model - conducted EMI
modelling . . . . . . . . . . . . . . . . . . . . . 117
5.4.1.1.1 Test case – high-voltage LIN interface. 118
5.4.1.1.2 Training and testing cases for the
TDMA-EMI-LOGIC model. . . . . . 119
5.4.1.1.3 Modelling results - TDMA-EMI-LOGIC
model. . . . . . . . . . . . . . . . . . 122
5.4.1.2 TDMA-EMI-FAILURE model - conducted EMI
modelling . . . . . . . . . . . . . . . . . . . . . 125
5.4.1.2.1 Training and testing cases for the
TDMA-EMI-FAILURE model. . . . . 125
5.4.1.2.2 Modelling results - TDMA-EMI-FAILURE
model. . . . . . . . . . . . . . . . . . 127
5.4.1.2.3 Modelling of an analogue circuit using
the TDMA-EMI-FAILURE model. . . 132
5.4.1.3 TDMA-EME model - conducted EME modelling 134
5.4.1.3.1 Training and testing cases for TDMA-
EME model. . . . . . . . . . . . . . . 135
5.4.1.3.2 Modelling results - TDMA-EME model.137
5.4.2 Performance modelling approach (PMA) . . . . . . . . . 139
5.4.2.1 PMA-EMI model - conducted EMI modelling . 140
5.4.2.1.1 Training and testing cases for the PMA-
EMI model. . . . . . . . . . . . . . . . 140
5.4.2.1.2 Modelling results - PMA-EMI model. . 141
5.4.2.2 PMA-EME model - conducted EME modelling 142
5.4.2.2.1 Training and testing cases for PMA-
EME model. . . . . . . . . . . . . . . 143
5.4.2.2.2 Modelling results - PMA-EME model. 144
5.4.3 Simulation speed-up (in comparison to circuit-level
simulations) . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6 Framework for the improved method for the black-box modelling
of electronic circuits in the time domain 151
6.1 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.2 Framework for the black-box modelling of electronic circuits . . 152
6.2.1 Application of the proposed framework for the black-box
modelling of electronic circuits . . . . . . . . . . . . . . 155
6.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
7 Conclusions and suggestions for future work 159
A US patent application 13,353,701 “System, Method and Computer
Program Product for Modeling Electronic Circuits” 163
Bibliography 205
Curriculum vitae 223
Životopis 225
List of publications 227
ISBN: 978-94-6018-693-6
Publication status: published
KU Leuven publication type: TH
Appears in Collections:ESAT - MICAS, Microelectronics and Sensors

Files in This Item:
File Status SizeFormat
thesis.pdf Published 3857KbAdobe PDFView/Open Request a copy

These files are only available to some KU Leuven Association staff members

 




All items in Lirias are protected by copyright, with all rights reserved.