Professional Documents
Culture Documents
Adaptive MV ARMA Identification Under The Presence of Noise As Published in Springer
Adaptive MV ARMA Identification Under The Presence of Noise As Published in Springer
2
Lectue Notes in Electrical Engineering
3
4 Volume 27
5
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46 Lecture Notes in Electrical Engineering
47
48
49
Proceedings of the European Computing Conference: Volume 2
50 Mastorakis, Nikos, Mladenov, Valeri (Eds.)
51 978-0-387-84818-1, Vol. 28
52
Proceedings of the European Computing Conference: Volume 1
53
Mastorakis, Nikos, Mladenov, Valeri (Eds.)
54 978-0-387-84813-6, Vol. 27
55
Electronics System Design Techniques for Safety Critical Applications
56
Sterpone, Luca
57
978-1-4020-8978-7, Vol. 26
58
Data Mining and Applications in Genomics
59
Ao, Sio-Iong
60
978-1-4020-8974-9, Vol. 25
61
Informatics in Control, Automation and Robotics: Selected Papers from the
62
International Conference on Informatics in Control, Automation and Robotics 2007
63
Filipe, J.B.; Ferrier, Jean-Louis; Andrade-Cetto, Juan (Eds.)
64
978-3-540-85639-9, Vol. 24
65
Digital Terrestrial Broadcasting Networks
66
Beutler, Roland
67
ISBN 978-0-387-09634-6, Vol. 23
68
69
Logic Synthesis for Compositional Microprogram Control Units
Barkalov, Alexander, Titarenko, Larysa
70
ISBN: 978-3-540-69283-6, Vol. 22
71
90
91
92
Nikos Mastorakis Valeri Mladenov
l l
93 Vassiliki T. Kontargyri
94
Editors
95
96
97
98
99
100
103
104
105
106
107
Conference
108
109
110
111
Volume 1
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
13
132
133
134
135
136 Editors
137 Nikos Mastorakis Valeri Mladenov
138 Hellenic Naval Academy Technical University of Sofia
139
Military Institutes of University Sofia, Bulgaria
Education valerim@tu-sofia.bg
140
Piraeus, Greece
141 mastor@hna.gr
142
143
144
Vassiliki T. Kontargyri
145
National Technical
146 University of Athens
147 Athens, Greece
148 vkont@central.ntua.gr
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
ISSN 1876-1100 e-ISSN 1876-1119
164
ISBN 978-0-387-84813-6 e-ISBN 978-0-387-84814-3
165 DOI 10.1007/978-0-387-84814-3
166
Library of Congress Control Number: 2008937179
167
180 springer.com
181
182
Contents
183
184
185
186
187
188
189
190
191
192
v
vi Contents
320
34 Building Moderately Open Multi-Agent Systems:
321
The HABA Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
322
A. Garcés, R. Quirós, M. Chover, J. Huerta, and E. Camahort
323
324
325
Part V Educational Software and E-Learning Systems
326
327
35 Using Mathematics for Data Traffic Modeling Within
328
an E-Learning Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
329
Marian Cristian Mihăescu
330
335
37 Distance Teaching in the Technology Course in Senior
336
High School . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
337
Shi-Jer Lou, Tzai-Hung Huang, Chia-Hung Yen, Mei-Huang
338
Huang, and Yi-Hui Lui
339
340
38 Electronic Exams for the 21st Century. . . . . . . . . . . . . . . . . . . . . . . . 383
341
Antonios S. Andreatos
342
343
344
39 Effects of the Orff Music Teaching Method on Creative
345
Thinking Abilities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
346
Rong - Jyue Fang, Hung-Jen Yang, C. Ray Diez,
347
Hua-Lin Tsai, Chi-Jen Lee, and Tien-Sheng Tsai
348
353
41 Collaborative Online Network and Cultural Exchange Project . . . . . 409
354
Rong - Jyue Fang, Hung-Jen Yang, C. Ray Diez, Hua-Lin Tsai,
355
Chi-Jen Lee, and Tien-Sheng Tsai
356
357
42 A Study of the Project on Mobile Devices in Education . . . . . . . . . . . 421
358
Rong - Jyue Fang, Hung-Jen Yang, Hua-Lin Tsai, Chi-Jen Lee,
359
Chung-Ping Lee, and Pofen Wang
360
Contents ix
371 46 New Software for the Study of the Classic Surfaces from
372 Differential Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
373 Anca Iordan, George Savii, Manuela Pănoiu, and Caius Pănoiu
374
405
x Contents
409 56 Numerical Simulation for the Hopf Bifurcation in TCP Systems . . . . 567
410 Gabriela Mircea and Mihaela Muntean
411
542
Contributors
543
544
545
546
547
548
549
550
551
552
xiii
xiv Contributors
590
Mohamed Salim Bouhlel Research Unit, Sciences and Technologies of Image
591
and Telecommunications (SETIT), Higher Institute of Biotechnology of Sfax,
592
Tunisia, medsalim.bouhlel@enis.rnu.tn
593
S. Bunluechokchai Department of Industrial Physics and Medical
594
Instrumentation, Faculty of Applied Science, King Mongkut’s Institute
595
of Technology, North Bangkok, 1518 Pibulsongkram Road, Bangsue,
596
Bangkok 10800 Thailand
597
598 Rui Camacho DEEC & LIACC, Universidade do Porto, R. Dr Roberto Frias,
599 s/n, 4200-465 Porto, Portugal, rcamacho@fe.up.pt
600
601
E. Camahort Department of Computer Systems, Polytechnic University
602
of Valencia, Valencia, Spain
603
Leonarda Carnimeo Dipartimento di Elettrotecnica ed Elettronica, Politecnico
604
di Bari, Via Orabona, 4 70125 BARI – Italy, carnimeo@deemail.poliba.it
605
609
Hassan Charaf Department of Automation and Applied Informatics,
610
Budapest University of Technology and Economics, H-1111 Budapest,
611 Goldmann György tér 3., Hungary
612
Chokri Chemak Research Unit, Sciences and Technologies of Image and
613
Telecommunications (SETIT), Higher Institute of Biotechnology of Sfax,
614
Tunisia, cchemak@univ-fcomte.fr
615
634
Monica Dascălu IMT Bucharest, Erou Iancu Nicolae 32B Street, Romania
635
Nor Ridzuan Daud Faculty of Computer Science and Information Technology,
636
University of Malaya, 50603 Kuala Lumpur, Malaysia
637
638 James DeArmon The MITRE Corporation, 202 Burlington Road, Bedford,
639 MA 01730-1420, USA
640
C. Ray Diez Department of Technology, The University of North Dakota,
641
Grand Forks, North Dakota, USA, Clayton.Diez@mail.business.und.edu
642
646
Daniel Dumitriu Institute of Solid Mechanics, Romanian Academy, Bucures° ti,
647
Romania, dumitri04@yahoo.com
648
Lambros Ekonomou Hellenic American University, 12 Kaplanon Str., 106 80
649
Athens, Greece, leekonom@gmail.com
650
661
Miguel A. Ferrer Centro Tecnológico para la Innovación en Comunicaciones
662
(CETIC), Dpto. de Señales y Comunicaciones, Universidad de Las Palmas de
663
Gran Canaria, Campus de Tafira, E35017, Las Palmas de Gran Canaria, Spain,
664
mferrer@dsc.ulpgc.es
665
Bertalan Forstner Department of Automation and Applied Informatics,
666
Budapest University of Technology and Economics, H-1111 Budapest,
667
Goldmann György tér 3., Hungary, bertalan.forstner@aut.bme.hu
668
680
A. G. Ravelo-Garcı́a Dpto. de Señales y Comunicaciones, Universidad de Las
681
Palmas de Gran Canaria, Campus de Tafira, E35017, Las Palmas de Gran
682
Canaria, Spain, aravelo@dsc.ulpgc.es
683
G. Gargano Università degli Studi dell’Aquila, Centro di Eccellenza DEWS,
684
67100 Poggio di Roio (AQ), Italy, gino.gargano@email.it
685
686 S.A.M. Gilani Faculty of Computer Science and Engineering, Ghulam Ishaq
687 Khan Institute of Engineering Sciences and Technology, Topi, Pakistan,
688 asif@giki.edu.pk
689
E. Glavas A.T.E.I. of Epirus, Department of Teleinformatics and
690
Management, GR-471 00 Arta, Greece, eglavas@teiep.gr
691
696
F. Graziosi Università degli Studi dell’Aquila, Centro di Eccellenza DEWS,
697
67100 Poggio di Roio (AQ), Italy, graziosi@ing.univaq.it
698
Eugene Grichuk Moscow Engineering Physics Institute, Kashirskoe Schosse,
699
115409 Moscow, Russia, es@t-25.ru
700
726
01730-1420, USA
727 A. Jraifi Groupe CPR, Canal, Radio & Propagation, Lab/UFR-PHE, Faculté
728 des Sciences de Rabat, Morocco
729
730
Arbana Kadriu CST Department, SEE-University, b.b. 1200 Tetovo,
731
FYROM, a.kadriu@seeu.edu.mk
732 G. Karangelis Oncology Systems Ltd., Shrewsbury, SY1 3AF, UK
733
AQ5
734
Alireza Kargar Department of Electronics and Engineering, School of
735
Engineering, Shiraz University, Shiraz, Iran; Shiraz Nanotechnology Research
736
Institute, Shiraz, Iran
737 V.N. Kasyanov A.P. Ershov Institute of Informatics Systems/Novosibirsk
738 State University, Novosibirsk, 630090, Russia
739
740
E.V. Kasyanova A.P. Ershov Institute of Informatics Systems/Novosibirsk
741
State University, Novosibirsk, 630090, Russia
742
Sokratis K. Katsikas Department of Technology Education and Digital
743
Systems, University of Piraeus, 150 Androutsou St., Piraeus, 18532, Greece,
744
ska@unipi.gr
745
746
Imre Kelényi Department of Automation and Applied Informatics, Budapest
747
University of Technology and Economics, H-1111 Budapest, Goldmann
748 György tér 3., Hungary
749
A. El Khafaji Lab d’Electronique et de Télécommunication, EMI, Rabat,
750
Morocco
751
769
Jean Christophe Lapayre Computer Laboratory of Franche-Comte (L.I.F.C.),
770
Franche-Compte University of Sciences and Techniques, France, Jean
AQ6
771
christophe.lapayre@univ-fcomte.fr
772
Chi-Jen Lee Department of Industrial Technology Education, National
773
Kaohsiung Normal University, Taiwan, mickeylee@gmail.com
774
785
Shi-Jer Lou Institute of Technological and Vocational Education, National
786
Pingtung University of Science and Technology, Taiwan, ROC,
787
lou@mail.npust.edu.tw
788
Yi-Hui Lui Institute of Technological and Vocational Education, National
789
Pingtung University of Science and Technology, Taiwan
790
791 Demétrio Renó Magalhães Depto de Exatas & CSI, UnilesteMG, Av Pres.
792 Tancredo Neves, 3500, 35170-056, C. Fabriciano, MG, Brasil,
793 reno@uniliestemg.br
794
Paula Mahoney The MITRE Corporation, 202 Burlington Road, Bedford,
795
MA 01730-1420, USA
796
797 G.A. Mallah National University of Computer and Emerging Sciences, ST-4,
798 Sector-17/D, Shah Latif Town, Karachi, Pakistan
799
Eduard Manykin Russian RC ‘‘Kurchatov Institute’’, Kurchatov Sq. 1, 123182
800
Moscow, Russia, edmany@isssph.kiae.ru
801
811 Nisar Ahmed Memon Faculty of Computer Science and Engineering, Ghulam
812 Ishaq Khan Institute of Engineering Sciences and Technology, Topi, Pakistan,
813 memon_nisar@yahoo.com
814
815
Marian Cristian Mihăescu Software Engineering Department, University
816
of Craiova, Craiova, Romania
817
Jamie D. Mills Department of Educational Studies, University of Alabama,
818
316-A Carmichael Hall, Box 870231, Tuscaloosa, AL 35487-0231, USA,
819
jmills@bamaed.ua.edu
820
831
Payman Moallem Faculty of Engineering, University of Isfahan, Isfahan, Iran,
832
p_moallem@eng.ui.ac.ir
833
K.S. Mohammed Electronic Department, National Telecommunication
834
Institute, Cairo, Egypt
835
845
Mihaela Muntean Faculty of Economic Sciences, West University
846 of Timisoara, Romania, mihaela.muntean@fse.uvt.ro
847
Angelos Nakulas National and Kapodistrian University of Athens, 11 Asklipiu
848
Str., 153 54 Athens, Greece, aaatos@gmail.com
849
854
Adrian-Ioan Niculescu Institute of Solid Mechanics, Romanian Academy,
855
Bucures° ti, Romania, adrian_ioan_niculescu@yahoo.com
xx Contributors
860
Ali Olfat Electrical and Computer Engineering Department, University
861
of Tehran, Campus #2, North Kargar Ave., Tehran, Iran, aolfat@ut.ac.ir
862
R. Orzechowski Chair of Wireless Communication, Poznan University
863
of Technology, Piotrowo 3a, 60-965 Poznan, Poland
864
874
Sonja Petrovic-Lazarevic Department of Management, Monash University, 26
875
Sir John Monash Drive, Caulfield East, VIC 3145, Australia,
876
sonja.petrovic-lazarevic@buseco.monash.edu.au
877
L. Pomante Università degli Studi dell’Aquila, Centro di Eccellenza DEWS,
878
67100 Poggio di Roio (AQ), Italy, pomante@ing.univaq.it
879
889
Mircea Preda Computer Science Department, University of Craiova,
890
Romania
891
R. Quirós Department of Computer Systems, Jaume I University, Castellón,
892
Spain
893
899 Luı́s P. Reis DEEC and LIACC, Universidade do Porto, R. Dr Roberto Frias,
900 s/n, 4200-465 Porto, Portugal, lpreis@fe.up.pt
Contributors xxi
909
Carlos F. Romero Centro Tecnológico para la Innovación en Comunicaciones
910
(CETIC), Dpto. de Señales y Comunicaciones, Universidad de Las Palmas de
911
Gran Canaria, Campus de Tafira, E35017, Las Palmas de Gran Canaria, Spain,
912
fabian_romero@ciudad.com.ar
913 András Rövid Department of Chassis and Lightweight Structures, Budapest
914 University of Technology and Economics, Bertalan Lajos u. 2. 7. em., 1111
915 Budapest, Hungary
916
917
Hajar Sadeghi Faculty of Engineering, University of Isfahan, Isfahan, Iran,
918
hsadeghi@eng.ui.ac.ir
919
Lakhdar Sahbi Computer Science Department, University of Oum El
920
Bouaghi, 04000 Oum El Bouaghi, Algeria, sahbi_lak@yahoo.fr
921
922
E.H Saidi VACBT, Virtual African Center for Basic Science and Technology,
923
Focal Point Lab/UFR-PHE, Fac Sciences, Rabat, Morocco
924
Farhat Saleemi Lahore College for Women University, Lahore, Pakistan
925
926
Ahmed Sameh Department of Computer Science, The American University
927
in Cairo, 113 Kaser Al Aini Street, P.O. Box 2511, Cairo, Egypt,
928 sameh@aucegypt.edu
929
Vianey Guadalupe Cruz Sánchez Centro Nacional de Investigación y
930
Desarrollo Tecnológico (CENIDET), Interior Internado Palmira s/n, Col.
931
Palmira, Cuernavaca, Morelos, México, vianey@cenidet.edu.mx
932
938 N.A. Shaikh National University of Computer and Emerging Sciences, ST-4,
939 Sector-17/D, Shah Latif Town, Karachi, Pakistan
940
Z.A. Shaikh National University of Computer and Emerging Sciences, ST-4,
941
Sector-17/D, Shah Latif Town, Karachi, Pakistan
942
950
Tudor Sireteanu Institute of Solid Mechanics, Romanian Academy, Bucures° ti,
951
Romania, siret@imsar.bu.edu.ro
952
Tomasz Śliwiński Institute of Control and Computation Engineering, Warsaw
953
University of Technology, 00-665 Warsaw, Poland, tsliwins@ia.pw.edu.pl
954
960 Huma Tauseef Lahore College for Women University, Lahore, Pakistan
961
962
Z. Cihan Taysi Department of Computer Engineering, Yldz Technical
963
University, 34349 Besiktas / Istanbul, Turkey, cihan@ce.yildiz.edu.tr
964
Virgil Teodor Department of Manufacturing Engineering, ‘‘Dunărea de Jos’’
965
University of Galat° i, Galat° i, 111 Domnească Street, zip code 800201,
966
Romania, virgil.teodor@ugal.ro
967
979
Philippe Trigano Heudiasyc UMR CNRS 6599, Université de Technologie de
980
Compiègne, France
981
Hua-Lin Tsai Department of Industrial Technology Education, National
982
Kaohsiung Normal University, Taiwan, kittyhl@gmail.com
983
994
W. Ussawawongaraya Department of Industrial Physics and Medical
995
Instrumentation, Faculty of Applied Science, King Mongkut’s Institute of
996
Technology, North Bangkok, 1518 Pibulsongkram Road, Bangsue, Bangkok
997
10800 Thailand
998
I. O. Vardiambasis Division of Telecommunications, Microwave
999
Communications and Electromagnetic Applications Laboratory, Department
1000
of Electronics, Technological Educational Institute (T.E.I.) of Crete, Chania
1001
Branch, Romanou 3, Chalepa, 73133 Chania, Crete, Greece,
1002
ivardia@chania.teicrete.gr
1003
1008
D.C. Vasiliadis Department of Computer Science and Technology, University
1009
of Peloponnese, GR-221 00 Tripolis, Greece, dvas@uop.gr
1010
Andreas Veglis Department of Journalism and MC, Aristotle University
1011
of Thessaloniki, 54006, Thessaloniki, Greece
1012
1017
Pofen Wang Department of Information Management, Southern Taiwan
1018
University of Technology, Taiwan, rxf26@mail.stut.edu.tw,
1019
pohfen@ma.ks.edu.tw
1020
P. Woraratsoontorn Department of Industrial Physics and Medical
1021
Instrumentation, Faculty of Applied Science, King Mongkut’s Institute of
1022
Technology, North Bangkok, 1518 Pibulsongkram Road, Bangsue, Bangkok
1023
10800, Thailand
1024
1027
Hsieh-Hua Yang Department of Health Care Administration, Chang Jung
1028
Christian University, 396 Chang Jung Rd., Sec. 1, Kway Jen, Tainan, Taiwan,
1029
ROC, yansnow@gmail.com, yangsnow@mail.cju.edu.tw
1030
Hung-Jen Yang Department of Industrial Technology Education, National
1031
Kaohsiung Normal University, Kaohsiung County, Taiwan, ROC,
1032
hjyang@nknucc.nknu.edu.tw, Hungjen.yang@gmail.com
1033
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1
Chapter 17
2
3
Adaptive MV ARMA Identification Under
4
the Presence of Noise
5
7
Stylianos Sp. Pappas, Vassilios C. Moussas, and Sokratis K. Katsikas
8
10
11
12
22
23
24
17.1 Introduction
25
26
The problem of fitting a multivariate ARMA model to a given time series is an
27
essential one in speech analysis, biomedical applications, hydrology, electric
28
power systems, and many other applications [1–4].
29
In this paper, a new method for multivariate ARMA model order
30
selection and parameter identification is presented, as an extension to the
31
one proposed in [5] for MV AR models. The method is based on the well
32
known adaptive multimodel partitioning theory [6, 7]. It is not restricted to
33
the Gaussian case, it is applicable to online/adaptive operation, and it is
34 computationally efficient. Furthermore, it identifies the correct model
35 order very quickly.
36 An m-variate ARMA model of order (p, q) [ARMA (p, q)] for a stationary
37 time series of vectors y observed at equally spaced instants k ¼ 1, 2, . . . , n is
38 defined as:
39
40
41
S.Sp. Pappas (*)
42
Department of Information and Communication Systems Engineering, University
43 of the Aegean, 83200 Karlovassi Samos, Greece
44 e-mail: spappas@aegean.gr
45 X
p X
q
46 yk ¼ Ai yki þ Bj vkj þ vk; E½vk vTk ¼ R (17:1)
47
i¼1 j¼1
48
49
where the m-dimensional vector vk is uncorrelated random noise, not necessarily
50
Gaussian, with zero mean and covariance matrix R, ¼ (p, q) is the order of the
51
predictor, and A1, . . . , Ap, B1, . . . , Bq are the m m coefficient matrices of the
52
MV ARMA model.
53
It is obvious that the problem is twofold. The first task, which is the most
54
important for the problem under consideration, is the successful determination
55
of the predictor’s order ¼ (p, q). Once the model order selection task is
56
completed, one proceeds with the second task, i.e., the computation of the
57
predictor’s matrix coefficients {Ai, Bj }.
58
Determining the order of the ARMA process is usually the most delicate and
59
crucial part of the problem. Over the past years substantial literature has been
60
produced on this problem and various different criteria, such as Akaike’s [8],
61
Rissanen’s [9, 10], Schwarz’s [11], and Wax’s [12], have been proposed to
62
implement the order selection process.
63
The above mentioned criteria are not optimal and are also known to suffer
64
from deficiencies; for example, Akaike’s information criterion suffers from
65
overfit [13]. Also, their performance depends on the assumption that the data
66
are Gaussian and upon asymptotic results. In addition to this, their applicabil-
67
ity is justified only for large samples; furthermore, they are two-pass methods,
68
so they cannot be used in an online or adaptive fashion.
69
The paper is organized as follows. In Section 17.2, the MV ARMA model
70
order selection problem is reformulated so that it can be fitted into the state
71
space under the uncertainty estimation problem framework. In the same section
72
the multi-model partitioning filter (MMPF) is briefly described and its applica-
73
tion to the specific problem is discussed. In Section 17.3, simulation examples
74
are presented which demonstrate the performance of our method in comparison
75
to previously reported ones. Finally, Section 17.4 summarizes the conclusions.
76
77
80 If we assume that the model order fitting the data is known and is equal to
81 ¼ (p, q), we can rewrite equation (17.1) in standard state-space form as:
82
6 .. .. .. 7
93
6
Ap is6 . . . 7 and
7
94
4 am1p
p
amm 5
95
96
2 3
97 bq11 ::: bq1m
98 6 .. .. .. 7
6 . . . 7
99
Bq is6
6
7 than
7
100 4 q
bm1 q
bmm 5
101
102
103
. . . .
104 xðkÞ ¼ ½111 121 1m1 .. 112 122 1m2 .. 1mm .. pmm ..
. . .
105
. . .
HðkÞ ¼ ½y1 ðk 1Þ I ym ðk 1Þ I .. .. y1 ðk pÞ I ym ðk pÞ I ..
108
109
110 . .
111 v1 ðk 1Þ I vm ðk 1Þ I .. .. v1 ðk qÞ I vm ðk qÞ I
112
113 where I is the m m identity matrix and ¼ (p, q) is the model order.
114 If the system model and its statistics were completely known, the Kalman
115 filter (KF) in its various forms would be the optimal estimator in the minimum
116 variance sense.
117 In the case where the prediction coefficients are subject to random perturba-
118 tions, (17.2) becomes
119
xðk þ 1Þ ¼ xðkÞ þ wðkÞ (17:4)
120
121 v(k), w(k) are independent, zero-mean, white processes, not necessarily
122 Gaussian.
123
. . . .
124 wðkÞ ¼ ½w111 w121 w1m1 .. w112 w122 w1m2 .. w1mm .. wpmm ..
. . .
125
128 A complete system description requires the value assignments of the var-
129 iances of the random processes w(k) and v(k). We adopt the usual assumption
130 that w(k) and v(k) at least are wide sense stationary processes, hence their
131 variances; and Q and R respectively are time invariant. Obtaining these values
132 is not always trivial. If Q and R are not known, they can be estimated by using a
133 method such as the one described in [14]. In the case of coefficients constant in
134 time, or slowly varying, Q is assumed to be zero (just as in equation (17.4)).
168 S.Sp. Pappas et al.
135 It is also necessary to assume an a priori mean and variance for each {Ai, Bi }.
136 The a priori mean of the Ai(0)’s and Bi(0)’s can be set to zero if no knowledge
137 about their values is available before any measurements are taken (the most
138 likely case). On the other hand, the usual choice of the initial variance of the Ai’s
139 and Bi’s, denoted by P0, is P0 ¼ nI, where n is a large integer.
140 Let us now consider the case where the system model is not completely known.
141 The adaptive multi-model partitioning filter (MMPF) is one of the most widely
142 used approaches for similar problems. This approach was introduced by Lainio-
143 tis [6, 7] and summarizes the parametric model uncertainty into an unknown,
144 finite-dimensional parameter vector whose values are assumed to lie within a
145 known set of finite cardinality. A non-exhaustive list of the reformulation,
146 extension, and application of the MMPF approach, as well as its application to
147 a variety of problems by many authors, can be found in [15] and [16–19]. In our
148 problem, assume that the model uncertainty is the lack of knowledge of the model
149 order . Let us further assume that the model order lies within a known set of
150 finite cardinality: 1 q M, where ¼ (p, q), is the model order.
151 The MMPF operates on the following discrete model:
152
155
158 where ¼ (p, q) is the unknown parameter, the model order in this case. A block
159 diagram of the MMPF is presented in Fig. 17.1.
160 In the Gaussian case, the optimal MMSE estimate of x(k) is given by
161
X
M
162
^ðk=kÞ ¼
x x^ðk=k; j Þ pðj =kÞ: (17:7)
163
j¼1
164
165 A finite set of models is designed, each matching one value of the parameter
166 vector. If the prior probabilities p(j / k) for each model are already known,
167 these are assigned to each model. In the absence of any prior knowledge, these
168 are set to pðj =kÞ ¼ 1=M where M is the cardinality of the model set.
169 A bank of conventional elemental filters (non-adaptive, e.g., Kalman) is then
170 applied, one for each model, which can be run in parallel. At each iteration the
171 MMPF selects the model which corresponds to the maximum posteriori prob-
172 ability as the correct one. This probability tends to one, while the others tend to
173 zero. The overall optimal estimate can be taken to be either the individual
174 estimate of the elemental filter exhibiting the highest posterior probability,
175 called the maximum a posteriori (MAP) estimate,in [20], which is the case
176 used in this paper; or the weighted average of the estimates produced by the
177 elemental filters, as described in equation (17.7). The weights are determined by
178 the posterior probability that each model in the model set is in fact the true
179 model.
17 Adaptive MV ARMA Identification Under the Presence of Noise 169
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
Fig. 17.1 MMPF block diagram
209
210
The posterior probabilities are calculated online in a recursive manner as
211
follows:
212
Lðk=k; j Þ
pðj =kÞ ¼ pðj =k 1Þ
213
(17:8)
214 P
M
215
Lðk=k; j Þ pðj =k 1Þ
j¼1
216
217
1=2
218 Lðk=k; j Þ ¼ P~y ðk=k 1; j Þ
219
1 (17:9)
220 exp½ y~T ðk=k 1; j Þ P1
~y ðk=k 1; j Þ ~
y ðk=k 1; j Þ
2
221
222
where the innovation process
223
224
~yðk=k 1; j Þ ¼ yðkÞ Hðk; j Þ^
xðk=k 1; j Þ (17:10)
170 S.Sp. Pappas et al.
227
Py~ ðk=k 1; j Þ ¼ Hðk; j ÞPðk=k; j ÞHT ðk; j Þ þ R (17:11)
228
For equations (17.8), (17.9), (17.10), and (17.11), j = 1, 2, . . . , M.
229
An important feature of the MMPF is that all the Kalman filters needed to
230
implement can be independently realized. This enables us to implement them in
231
parallel, thus saving us enormous computational time [20].
232
Equations (17.7) and (17.8) refer to our case where the sample space is
233
naturally discrete. However, in real world applications, ’s probability density
234
function (pdf) is continuous and an infinite number of Kalman filters have to be
235
applied for the exact realization of the optimal estimator. The usual approx-
236
imation considered to overcome this difficulty is to somehow approximate ’s
237
pdf by a finite sum. Many discretization strategies have been proposed at times,
238
and some of them are presented in [21–22].
239
When the true parameter value lies outside the assumed sample space, the
240
adaptive estimator converges to the value that is in the sample space, which is
241
closer (i.e., minimizes the Kullback information measure) to the true value [23].
242
This means that the value of the unknown parameter cannot be exactly defined.
243
The application of a variable structure MMPF is able to overcome this diffi-
244
culty [17].
245
246
247
248
17.3 Examples
249
250
In order to assess the performance of our method, several simulation experi-
251
ments were conducted. All of these experiments were conducted 100 times
252
(100 Monte Carlo runs). The model used was that of (17.2) and (17.3), with
253
cardinality M ¼ 10.
254
255
263 Figure 17.2 depicts the posterior probabilities associated with each value of
264 . Figure 17.3 shows the criteria comparison for two data sets, one relatively
265 small (50 samples) and one larger (100 samples), and Table 17.1 shows the
266 estimated ARMA parameter coefficients.
267 From Fig. 17.2, it is obvious that the MMPF identifies the correct ¼ (1, 1) ¼ 2
268 very quickly, in just 17 steps. Convergence is taken to occur when the posterior
269 probability of the model exceeds 0.9.
17 Adaptive MV ARMA Identification Under the Presence of Noise 171
270
271
272
273
274
275
277
278
279
280
281
282
283
284
285
286
287
288
Fig. 17.2 Example 1, posterior probabilities
289
290
100
291
100 100 100
292
293
294 95
295
89
300
80
301
302
100
75 Samples
303
MMPF 50
304
AIC
305
BIC
306 Criteria
307
Fig. 17.3 Example 1, criteria comparison—correct model order identification
308
309
310 From Fig. 17.3, we deduce that MMPF is 100% successful in selecting the
311 correct model order for both data sets, while only BIC matches its performance
312 for the larger data set.
313 Also, Table 17.1 shows that the parameter coefficient estimation is very
314 accurate. (RMSE—root mean square error—is very small).
172 S.Sp. Pappas et al.
325
17.3.2 Example 2
326
328
2 3
329 2 3 1 0:15 0:09
1 0:2 0:23 6 0:1 0:1 0:05 7
6 7 6 7
330
338
339
340
341
342
343
344
345
346
This figure will be printed in b/w
347
348
349
350
351
352
353
354
355
356
357
358
360
100
361
100 100 97
362
89
90
363
80 82 89
364
70
379
380
381
small (50 samples) and one larger (100 samples), and Table 17.2 shows the
382
estimated ARMA parameter coefficients.
383
From Figure 17.4, it is obvious that the MMPF identifies the correct ¼
384
(1, 1) ¼ 2 very quickly, in just 18 steps. Convergence is taken to occur when the
385
posterior probability of the model exceeds 0.9.
386
From Figure 17.5, we deduce that MMPF is 100% successful in selecting the
387 correct model order for both data sets, while none of the two other criteria
388 achieve a similar performance for either data set.
389 As Table 17.2 clearly shows, the parameter estimation is again accurate
390 because the root mean square error (RMSE) is very small.
391
392
393
417
Figure 17.6 depicts the posterior probabilities associated with each value of
418
. Figure 17.7 shows the criteria comparison for two data sets, one relatively
419
small (50 samples) and one larger (100 samples), and Table 17.3 shows the
420
421
estimated ARMA parameter coefficients.
422
From Fig. 17.6, it is obvious that the MMPF identifies the correct ¼ (1, 1) ¼ 2
423
very quickly, in just 24 steps. Convergence is taken to occur when the posterior
424
probability of the model exceeds 0.9.
425
426
427
428
429
430
431
432
433
434
435
436
This figure will be printed in b/w
437
438
439
440
441
442
443
444
445
446
447
448
450
100
451
100 100 100
452
90
86
453 92
454 80
455 70 74
456
460
30
461 20
462 10
100
463 0 Samples
464
MMPF 50
465
AIC
466
BIC
467
Criteria
468
Fig. 17.7 Example 2, criteria comparison—correct model order identification
469
470
471
Table 17.3 Example 3, estimated ARMA coefficient parameters
472
Estimated parameters RMS error
473
0.1691 0.4535 0.0047 0.0094
474
0.1896 0.3260 0.0054 0.0109
475 0.1458 0.5244 0.0137 0.0090
476 0.0899 0.6931 0.0161 0.0108
477 0.1982 0.8441 0.0064 0.0122
478 0.2234 0.6407 0.0073 0.0147
479 0.1154 0.7508 0.0094 0.0086
480
0.2573 0.5471 0.0114 0.0094
481
482
From Fig. 17.7, we deduce that MMPF is 100% successful in selecting the
483
correct model order for both data sets, while only BIC matches its performance
484
for the larger data set.
485
As Table 17.3 clearly shows, the parameter estimation is again accurate
486
because the root mean square error (RMSE) is very small.
487
488
489
492 A new method for simultaneously selecting the order and estimating the para-
493 meters of a MV ARMA model has been developed, as an extension of the
494 method proposed for the MV AR case. The proposed method successfully
176 S.Sp. Pappas et al.
495 selects the correct model order in very few steps and identifies very accurately
496 the ARMA parameters. Comparison with other established order selection
497 criteria (AIC, BIC) shows that the proposed method needs only the shortest
498 data set for successful order identification and accurate parameter estimation
499 for all the simulated models; whereas the other criteria require longer data sets
500 as the model order increases. The method performs equally well when the
501 complexity of the MV ARMA model is increased.
502
503 Acknowledgment This paper is dedicated to the memory of Prof. Dimitrios G. Lainiotis, the
504 founder of the multi-model partitioning theory, who passed away suddenly in 2006.
505
506
507 References
508
509 1. Lu S, Ju KH, Chon KH (2001) A new algorithm for linear and nonlinear ARMA model
510 parameter estimation using affine geometry [and application to blood flow/pressure
511
data]. IEEE T Biomed Eng 48(10):1116–1124
2. Derk J, Weber S, Weber C (2007) Extended ARMA models for estimating price devel-
512
opments on day-ahead electricity markets. Elec Power Syst Res 77(5–6):583–593
513 3. Chen CP, Bilmes JA (2007) MVA processing of speech features. IEEE T Audio Speech
514 15(1):257–270
515 4. Kourosh M, Eslami HR, Kahawita R (2006) Parameter estimation of an ARMA model
516
for river flow forecasting using goal programming. J Hydrol 331(1–2):293–299
5. Pappas Sp S, Leros AK, Katsikas SK (2006) Joint order and parameter estimation of
517
multivariate autoregressive models using multi-model partitioning theory. Dig Signal
518 Process 16(6):782–795
519 6. Lainiotis DG (1976) Partitioning: a unifying framework for adaptive systems I: Estima-
520 tion. Proc IEEE 64:1126–1143
521
7. Lainiotis DG (1976) Partitioning: a unifying framework for adaptive systems I: Control.
Proc IEEE 64:1182–1198
522
8. Akaike H (1969) Fitting autoregressive models for prediction. Ann Inst Stat Math
523 21:243–247
524 9. Rissanen J (1978) Modelling by shortest data description. Automatica 14:465–471
525 10. Rissanen J (1986) A predictive least squares principle. IMA J Math Contr Inform
526
3:211–222
11. Schwarz G (1978) Estimation of the dimension of the model. Ann Statist 6:461–464
527
12. Wax M (1988) Order selection for AR models by predictive least squares. IEEE T Acoust
528 Speech Signal Process 36:581–588
529 13. Lutkepohl H (1985) Comparison of criteria for estimating the order of a vector AR
530 process J Time Ser Anal 6:35–52
531
14. Sage AP, Husa GW (1969) Adaptive filtering with unknown prior statistics. In: Proceed-
ings of the Joint Automatic Control Conference, Boulder, Colorado, pp 760–769
532
15. Watanabe K (1992) Adaptive estimation and control: partitioning approach.:Prentice
533 Hall, Englewood Cliffs, NJ
534 16. Katsikas SK, Leros AK, Lainiotis DG (1994) Passive tracking of a maneuvering target:
535 an adaptive approach. IEEE T Signal Process 42(7):1820–1825
536
17. Katsikas SK, Likothanassis SD, Beligiannis GN, Berkeris KG, Fotakis DA (2001)
Genetically determined variable structure multiple model estimation. IEEE T Signal
537
Process 49(10):2253–2261
538 18. Lainiotis DG, Papaparaskeva PA (1998) Partitioned adaptive approach to nonlinear
539 channel equalization. IEEE T Commun 46(10):1325–1336
17 Adaptive MV ARMA Identification Under the Presence of Noise 177
540 19. Moussas VC, Likothanassis SD, Katsikas SK, Leros AK (2005) Adaptive online multiple
541
source detection. In: IEEE International Conference on Acoustics Speech and Signal
Processing Proceedings (ICASSP ’05,), 4:1029–1032
542
20. Lainiotis DG, Katsikas SK, Likothanassis SD (1988) Adaptive deconvolution of seismic
543
signals: performance, computational analysis parallelism. IEEE T Acoust Speech Signal
544 Process 36(11):1715–1734
545 21. Sengbush RL, Lainiotis DG (1969) Simplified parameter quantization procedure for
546
adaptive estimation. IEEE T Automat Contr AC-14:424–425
22. Anderson BDO, Brinsmead TS, Bruyne FDe, Hespanha J, Liberzon D, Morse AS (2000)
547
Multiple model adaptive control part 1: finite controller coverings. Int J Robust Non-
548
linear Contr 10:909–929
549 23. Hawks RM, Moore JB (1976) Performance of Bayesian parameter estimators for linear
550 signal models. IEEE T Automat Contr AC–21:523–527
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584