Software Technologies 14th International

Conference ICSOFT 2019 Prague Czech Republic

July 26 28 2019 Revised Selected Papers Marten
Van Sinderen pdf download

https://textbookfull.com/product/software-technologies-14th-international-conference-
icsoft-2019-prague-czech-republic-july-26-28-2019-revised-selected-papers-marten-van-sinderen/

★★★★★ 4.8/5.0 (46 reviews) ✓ 242 downloads ■ TOP RATED

"Great resource, downloaded instantly. Thank you!" - Lisa K.

DOWNLOAD EBOOK
 Software Technologies 14th International Conference ICSOFT
2019 Prague Czech Republic July 26 28 2019 Revised Selected
Papers Marten Van Sinderen pdf download

TEXTBOOK EBOOK TEXTBOOK FULL

Available Formats

■ PDF eBook Study Guide TextBook

EXCLUSIVE 2025 EDUCATIONAL COLLECTION - LIMITED TIME

INSTANT DOWNLOAD VIEW LIBRARY

Collection Highlights

E Business and Telecommunications 16th International

Conference ICETE 2019 Prague Czech Republic July 26 28
2019 Revised Selected Papers Mohammad S. Obaidat

Simulation and Modeling Methodologies, Technologies and

Applications: 9th International Conference, SIMULTECH 2019
Prague, Czech Republic, July 29-31, 2019, Revised Selected
Papers Mohammad S. Obaidat

Agents and Artificial Intelligence 11th International

Conference ICAART 2019 Prague Czech Republic February 19
21 2019 Revised Selected Papers Jaap Van Den Herik

Informatics in Control, Automation and Robotics: 16th

International Conference, ICINCO 2019 Prague, Czech
Republic, July 29-31, 2019, Revised Selected Papers Oleg
Gusikhin
Information Systems Security and Privacy 5th International
Conference ICISSP 2019 Prague Czech Republic February 23
25 2019 Revised Selected Papers Paolo Mori

Operations Research and Enterprise Systems 8th

International Conference ICORES 2019 Prague Czech Republic
February 19 21 2019 Revised Selected Papers Greg H.
Parlier

Software Technologies 10th International Joint Conference

ICSOFT 2015 Colmar France July 20 22 2015 Revised Selected
Papers 1st Edition Pascal Lorenz

Computer Aided Architectural Design Hello Culture 18th

International Conference CAAD Futures 2019 Daejeon
Republic of Korea June 26 28 2019 Selected Papers Ji-Hyun
Lee

Business Information Systems Workshops: BIS 2019

International Workshops, Seville, Spain, June 26–28, 2019,
Revised Papers Witold Abramowicz
Marten van Sinderen
Leszek A. Maciaszek (Eds.)

Communications in Computer and Information Science 1250

Software Technologies
14th International Conference, ICSOFT 2019
Prague, Czech Republic, July 26–28, 2019
Revised Selected Papers
Communications
in Computer and Information Science 1250
Commenced Publication in 2007
Founding and Former Series Editors:
Simone Diniz Junqueira Barbosa, Phoebe Chen, Alfredo Cuzzocrea,
Xiaoyong Du, Orhun Kara, Ting Liu, Krishna M. Sivalingam,
Dominik Ślęzak, Takashi Washio, Xiaokang Yang, and Junsong Yuan

Editorial Board Members

Joaquim Filipe
Polytechnic Institute of Setúbal, Setúbal, Portugal
Ashish Ghosh
Indian Statistical Institute, Kolkata, India
Igor Kotenko
St. Petersburg Institute for Informatics and Automation of the Russian
Academy of Sciences, St. Petersburg, Russia
Raquel Oliveira Prates
Federal University of Minas Gerais (UFMG), Belo Horizonte, Brazil
Lizhu Zhou
Tsinghua University, Beijing, China
More information about this series at http://www.springer.com/series/7899
Marten van Sinderen Leszek A. Maciaszek (Eds.)
•

Software Technologies
14th International Conference, ICSOFT 2019
Prague, Czech Republic, July 26–28, 2019
Revised Selected Papers

123
Editors
Marten van Sinderen Leszek A. Maciaszek
Information Systems Group Institute of Business Informatics
University of Twente Wrocław University of Economics
Enschede, The Netherlands Wrocław, Poland
Macquarie University
Sydney, Australia

ISSN 1865-0929 ISSN 1865-0937 (electronic)

Communications in Computer and Information Science
ISBN 978-3-030-52990-1 ISBN 978-3-030-52991-8 (eBook)
https://doi.org/10.1007/978-3-030-52991-8

© Springer Nature Switzerland AG 2020

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, expressed or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
 Preface

The present book includes extended and revised versions of a set of selected papers
from the 14th International Conference on Software Technologies (ICSOFT 2019),
held in Prague, Czech Republic, from July 26–28, 2019.
ICSOFT 2019 received 116 paper submissions from 41 countries, of which 21
contributions (18%) were accepted and presented as full papers. After the conference,
the authors of selected full papers were invited to submit a revised and extended
version of their papers having at least 30% new material. In the end, 10 revised and
extended papers have been included in this Springer book (i.e. 9% of the original 116
submissions).
The papers were selected by the event chairs and their selection was based on a
number of criteria that included the classifications and comments provided by the
Program Committee members, the session chairs’ assessment, and the verification
of the papers’ revisions and extensions by the program and conference chairs.
The purpose of the ICSOFT conference is to bring together researchers, engineers,
and practitioners interested in software technologies. The conference solicits papers and
other contributions in themes ranging from software engineering and development via
showcasing cutting-edge software systems and applications to addressing foundational
innovative technologies for systems and applications of the future. The papers were
presented in one of three conference areas: “Software Engineering and Systems
Development,” “Software Systems and Applications,” and “Foundational and Trigger
Technologies.”
We would like to thank all the authors for their contributions and the reviewers for
ensuring the quality of this publication.

July 2019 Marten van Sinderen

Leszek A. Maciaszek
 Organization

Conference Chair
Leszek A. Maciaszek Wrocław University of Economics, Poland,
and Macquarie University, Australia

Program Chair
Marten van Sinderen University of Twente, The Netherlands

Program Committee
Jose Gonzalez University of Seville, Spain
Waleed Alsabhan KACST, UK
Peter Amthor Technische Universität Ilmenau, Germany
Soumyadip Bandyopadhyay BITS Pilani K K Birla Goa Campus and Hasso Plattner
Institute, India
Davide Basile University of Florence, Italy
Doina Bein California State University, Fullerton, USA
Wolfgang Bein University of Nevada, Las Vegas, USA
Fevzi Belli Izmir Institute of Technology, Turkey
Yann Ben Maissa INPT, Morocco
Jorge Bernardino Polytechnic of Coimbra - ISEC, Portugal
Mario Berón Universidad Nacional de San Luis, Argentina
Marcello Bersani Politecnico di Milano, Italy
Dominik Bork University of Vienna, Austria
Andrea Burattin University of Innsbruck, Austria
Nelio Cacho Federal University of Rio Grande do Norte, Brazil
Alejandro Calderón University of Cádiz, Spain
Gerardo Canfora University of Sannio, Italy
Cagatay Catal Wageningen University, The Netherlands
Ana Cavalli Telecom SudParis, France
Juan Chagüendo Benavides Universidad Carlos III de Madrid, Spain
Alexandru Cicortas West University Timisoara, Romania
Lidia López Cuesta Universitat Politècnica de Catalunya, Spain
Sergiu Dascalu University of Nevada, Reno, USA
Cléver Ricardo de Farias University of São Paulo, Brazil
Martina De Sanctis Gran Sasso Science Institute, Italy
Steven Demurjian University of Connecticut, USA
Chiara Di Francescomarino FBK-IRST, Italy
József Dombi University of Szeged, Hungary
Gencer Erdogan SINTEF, Norway
viii Organization

Morgan Ericsson Linnaeus University, Sweden

Anne Etien University of Lille, France
João Faria FEUP, University of Porto, Portugal
Luis Fernandez Sanz University of Alcala, Spain
Estrela Ferreira Cruz Instituto Politecnico de Viana do Castelo,
University of Minho, Portugal
Nicolas Ferry SINTEF, Norway
Mirco Franzago University of L’Aquila, Italy
Matthias Galster University of Canterbury, New Zealand
Mauro Gaspari University of Bologna, Italy
Hamza Gharsellaoui Arab Open University, Saudi Arabia
Aritra Ghosh Florida Atlantic University, USA
Paola Giannini University of Piemonte Orientale, Italy
John Gibson Mines-Telecom, Telecom SudParis, France
Ana-Belén Gil-González University of Salamanca, Spain
Gorkem Giray Independent Researcher, Turkey
Gregor Grambow Hochschule Aalen, Germany
Christiane Gresse von Federal University of Santa Catarina (UFSC), Brazil
Wangenheim
Hatim Hafiddi INPT, Morocco
Slimane Hammoudi ESEO, ERIS, France
Ludovic Hamon Le Mans Université, France
Jean Hauck Universidade Federal de Santa Catarina, Brazil
Pedro Henriques University of Minho, Portugal
Jose Herrera Universidad del Cauca, Colombia
Mercedes Hidalgo-Herrero Universidad Complutense de Madrid, Spain
Jose R. Hilera University of Alcala, Spain
Andreas Holzinger Medical University Graz, Austria
Jang-Eui Hong Chungbuk National University, South Korea
Zbigniew Huzar Wrocław University of Science and Technology,
Poland
Ivan Ivanov SUNY Empire State College, USA
Judit Jasz University of Szeged, Hungary
Bo Jørgensen University of Southern Denmark, Denmark
Maria Jose Escalona University of Seville, Spain
Hermann Kaindl University of Seville, Spain, and TU Wien, Austria
Dimitris Karagiannis University of Vienna, Austria
Carlos Kavka University of Paris-Sud, France
Dean Kelley Minnesota State University, USA
Mieczyslaw Kokar Northeastern University, USA
Jitka Komarkova University of Pardubice, Czech Republic
Jun Kong North Dakota State University, USA
Winfried Kühnhauser Ilmenau University of Technology, Germany
Rob Kusters The Open University, The Netherlands
Wing Kwong Hofstra University, USA
Giuseppe Lami CNR, Italy
 Organization ix

Sérgio Lopes University of Minho, Portugal

David Lorenz The Open University, Israel
Ivan Lukovic University of Novi Sad, Serbia
Stephane Maag Telecom SudPairs, France
Ivano Malavolta Vrije Universiteit Amsterdam, The Netherlands
Eda Marchetti ISTI-CNR, Italy
Katsuhisa Maruyama Ritsumeikan University, Japan
Manuel Mazzara Innopolis University, Russia
Fuensanta Carlos III Technical University of Madrid, Spain
Medina-Dominguez
Francesco Mercaldo CNR, Italy
Antoni Mesquida Calafat Universitat de les Illes Balears (UIB), Spain
Gergely Mezei Budapest University of Technology and Economics,
Hungary
Cristian Mihaescu University of Craiova, Romania
Mattia Monga Università degli Studi di Milano, Italy
Antao Moura Federal University of Campina Grande (UFCG), Brazil
Christian Muck University of Vienna, Austria
Antonio Muñoz University of Malaga, Spain
Takako Nakatani The Open University of Japan, Japan
Elena Navarro University of Castilla-La Mancha, Spain
Joan Navarro Universitat Ramon Llull - La Salle Campus, Spain
Paolo Nesi University of Florence, Italy
Rory O’Connor Dublin City University, Ireland
Flavio Oquendo European University of Brittany, IRISA-UBS, France
Claus Pahl Free University of Bozen-Bolzano, Italy
Luis Pedro University of Aveiro, Portugal
Jennifer Pérez Universidad Politécnica de Madrid (UPM), Spain
Dana Petcu West University of Timisoara, Romania
Dietmar Pfahl University of Tartu, Estonia
Giuseppe Polese Università degli Studi di Salerno, Italy
Stefano Quer Politecnico di Torino, Italy
Traian Rebedea University Politehnica of Bucharest, Romania
Werner Retschitzegger Johannes Kepler University, Austria
Ralf Reussner Karlsruhe Institute of Technology (KIT), Germany
Marcela Xavier Ribeiro Federal University of São Carlos, Brazil
Colette Rolland Université Paris 1 Panthéon-Sorbonne, France
António Miguel Rosado Instituto Politécnico de Viana do Castelo, Portugal
da Cruz
Gustavo Rossi Lifia, Argentina
Matteo Rossi Politecnico di Milano, Italy
Stuart Rubin University of California, San Diego, USA
Gunter Saake Institute of Technical and Business Information
Systems, Germany
Davide Sangiorgi Università di Bologna, Italy
Nickolas Sapidis University of Western Macedonia, Greece
x Organization

Santonu Sarkar BITS Pilani Goa, India

Elad Schiller Chalmers University of Technology, Sweden
Lionel Seinturier University of Lille, France
Istvan Siket Hungarian Academy of Science, Research Group
on Artificial Intelligence, Hungary
Harvey Siy University of Nebraska, Omaha, USA
Cosmin Spahiu University of Craiova, Romania
Anca-Juliana Stoica Uppsala University, Sweden
Hiroki Suguri Miyagi University, Japan
Selma Suloglu Rochester Institute of Technology, USA
Bedir Tekinerdogan Wageningen University, The Netherlands
Claudine Toffolon University of Maine, France
Joseph Trienekens The Open University, The Netherlands
Michael Vassilakopoulos University of Thessaly, Greece
Dessislava Vassileva Sofia University, Bulgaria
László Vidács University of Szeged, Hungary
Yan Wang Google, USA
Dietmar Winkler Vienna University of Technology, Austria
Andreas Winter Carl von Ossietzky University Oldenburg, Germany
Jinhui Yao Xerox Research, USA
Murat Yilmaz Dublin City University, Ireland
Jingyu Zhang Macquarie University, Australia
Zheying Zhang Tampere University, Finland

Additional Reviewers
Victoria Döller University of Vienna, Austria
Vimal Kunnummel University of Vienna, Austria
Wissam Mallouli Montimage, France
Damien Pollet Inria, France
Frederik Reiche KIT, Germany
Spyros Vosinakis University of the Aegean, Greece

Invited Speakers
Manfred Reichert Ulm University, Germany
Michael G. Hinchey Lero, University of Limerick, Ireland
Hans-Georg Fill University of Fribourg, Switzerland
 Contents

Software Engineering and Systems Development

Equivalence Class Definition for Automated Testing of Satellite

On-Board Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Ulrike Witteck, Denis Grießbach, and Paula Herber

What We Know About Software Architecture Styles in Continuous

Delivery and DevOps?. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Maya Daneva and Robin Bolscher

Analysing the Performance of Mobile Cross-platform Development

Approaches Using UI Interaction Scenarios . . . . . . . . . . . . . . . . . . . . . . . . 40
Stefan Huber, Lukas Demetz, and Michael Felderer

Quantitative Analysis of Mutant Equivalence . . . . . . . . . . . . . . . . . . . . . . . 58

Amani Ayad, Imen Marsit, Sara Tawfig, Ji Meng Loh,
Mohamed Nazih Omri, and Ali Mili

Distributed Frames: Pattern-Based Characterization of Functional

Requirements for Distributed Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Roman Wirtz, Maritta Heisel, and Marvin Wagner

Efficient Diagnosis of Reconfigurable Systems with Incorrect Behavior

and Faulty Components: A Case Study on SGrids. . . . . . . . . . . . . . . . . . . . 108
Yousra Hafidi, Laid Kahloul, and Mohamed Khalgui

Software Systems and Applications

GA-PPI-Net: A Genetic Algorithm for Community Detection

in Protein-Protein Interaction Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
Marwa Ben M’barek, Amel Borgi, Sana Ben Hmida, and Marta Rukoz

From Formal Test Objectives to TTCN-3 for Verifying ETCS Complex

Software Control Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Rabea Ameur-Boulifa, Ana Cavalli, and Stephane Maag

Service Robots: A Unified Framework for Detecting, Opening

and Navigating Through Doors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Tatsuya Harada, Antonio Tejero-de-Pablos, Stefano Quer,
and Francesco Savarese
xii Contents

Two-Stage Game Theoretic Approach for Energy Management

in Networked Microgrids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Ilyes Naidji, Olfa Mosbahi, Mohamed Khalgui, and Abdelmalik Bachir

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

Software Engineering and Systems
Development
 Equivalence Class Definition
for Automated Testing of Satellite
On-Board Image Processing

Ulrike Witteck1(B) , Denis Grießbach1(B) , and Paula Herber2(B)

1
Institute of Optical Sensor Systems, German Aerospace Center (DLR),
Adlershof, Berlin, Germany
{ulrike.witteck,denis.griessbach}@dlr.de
2
Embedded Systems Group, University of Münster, Münster, Germany
[email protected]

Abstract. On-board image processing technologies in the satellite

domain are subject to strict requirements with respect to reliability and
accuracy in hard real-time. Due to the large input domain of such pro-
cessing technologies it is impracticable or even impossible to execute all
possible test cases.
As a solution we define a novel test approach that efficiently and sys-
tematically captures the input domain of satellite on-board image pro-
cessing applications. We first partition each input parameter into equiv-
alence classes. Based on these equivalence classes we define multidimen-
sional coverage criteria to assess the coverage of a given test suite on the
whole input domain. Finally, our test generation algorithm automatically
inserts missing but relevant test cases into the given test suite such that
our multidimensional coverage criteria are satisfied.
As a result we get a reasonably small test suite that covers the com-
plete input domain. We demonstrate the effectiveness of our approach
with experimental results from the ESA medium-class mission PLATO.

Keywords: Image processing · Software testing · Equivalence class

partitioning · Satellite systems

1 Introduction
On-board image processing applications in the satellite domain are subject to
strict requirements with respect to reliability and mathematical accuracy in hard
real-time. The large input domain of such applications makes manual testing
error-prone and time-consuming. To overcome that problem, we need a test
approach that automatically and systematically generates test cases for such
image processing applications. The major problem of the automated generation
of test cases is the large amount of input parameters and their possible combi-
nations. This leads to a high number of test cases which makes the systematic
and efficient coverage of the complete input domain expensive.
c Springer Nature Switzerland AG 2020
M. van Sinderen and L. A. Maciaszek (Eds.): ICSOFT 2019, CCIS 1250, pp. 3–25, 2020.
https://doi.org/10.1007/978-3-030-52991-8_1
4 U. Witteck et al.

Automated test approaches for different domains, for example, for automotive
and railway applications, are presented in [2,7]. The authors investigate applica-
tions with huge input domains and complex functional behavior. However, their
focus is on event-driven, reactive real-time systems and the approaches are not
tailored to the domain of on-board image processing applications.
In this paper we present an extended version of our test approach given in
[14]. This approach systematically selects test cases from the huge input domain
given in image processing applications. Our objective is to achieve a high cover-
age of the input domain using a reasonably small test suite. To achieve that goal
we adopt the equivalence class partition testing method. This method partitions
a given domain into disjoint sub-domains called equivalence classes [13]. Only
some test values are used as representatives from each class. That reduces the
number of required test cases [1], but still systematically covers the respective
domain. We use that method to partition each input parameter of the on-board
image processing application into equivalence classes. Furthermore, we define
multidimensional coverage criteria that combines individual coverage criteria for
each input parameter. Finally, we specify a test generation algorithm that uses
our multidimensional coverage criteria to automatically assess given test suites
with respect to their coverage on the whole input domain. Moreover, the algo-
rithm removes redundant test cases and inserts missing but relevant test cases.
As a result we get a reasonably small test suite that covers the complete input
domain of satellite on-board image processing applications.
To investigate the efficiency of our test approach using equivalence class
definitions, we use the Fine Guidance System (FGS) algorithm of the Euro-
pean Space Agency (ESA) mission PLAnetary Transits and Oscillation of stars
(PLATO) as a case study [14]. The FGS algorithm is a satellite on-board image
processing algorithm to calculate the high-precision attitude of the spacecraft
by comparing tracked star positions with known star positions from a star cata-
log. Recent studies have shown that some of the input parameters as presented
in [14] can be partitioned more beneficial. In this paper we therefore present
redefined equivalence classes for two input parameters: object position and sub-
pixel position on the image plane. Moreover, we use an improved test criterion
to investigate the effectiveness of our test approach. The experimental results
show the effectiveness of our partitioning approach in terms of an increased error
detection capability.
This paper is structured as follows: In Sect. 2, we briefly introduce equivalence
class partition testing and give an overview of the ESA PLATO mission including
the FGS algorithm. In Sect. 3, we outline related work about equivalence class
testing for real-time systems. In Sect. 4, we present our redefined equivalence
classes as well as the automated test generation algorithm for satellite on-board
image processing applications. In Sect. 5, we present our experimental results and
compare them with the results presented in [14]. We conclude with a summary
in Sect. 6.
 Equivalence Class Definition for Automated Testing of Satellite 5

2 Preliminaries
We introduce the general concept of equivalence class partition testing and give
an overview of the PLATO mission and its mission-critical FGS algorithm to
understand the remainder of this paper.

2.1 Equivalence Class Partition Testing

To make testing more efficient and less time consuming, it is preferable to exam-
ine as many test cases as necessary to satisfy specified test criteria. However,
the selection of the necessary test cases from a huge input domain is a major
problem when testing an application [11].
Equivalence class partition testing offers a possible solution to this prob-
lem. It is a commonly used approach in practice. The technique partitions a
given input domain or output domain into disjoint sub-domains, the equivalence
classes. The method partitions the domain in such a way, that all elements in an
equivalence class are expected to provoke the same system behavior according
to a specification. Equivalence classes represent subsets of parameter values that
completely cover the input or output domain. For the purpose of software test-
ing, it is therefore sufficient to test some representative values of each equivalence
class. The selection of test cases from equivalence classes can be made accord-
ing to various criteria: using border values, testing special values or randomly
selecting test cases [1,7,11].
The increased partitioning effort is a drawback of using equivalence class
partition testing compared to random testing. In many cases, several definitions
of the domain partitioning are applicable. This is mainly because the tester
assumes that test cases of the same equivalence class have the same system
behavior. However, the approach removes redundant test cases but retains the
completeness of the tests. Hence, the approach reduces the test effort compared
to exhaustive testing [1].

2.2 Context: PLATO Mission

PLATO is an ESA mission in the long-term space scientific program “Cosmic
Vision” [5]. The German Aerospace Center (DLR) manages the international
consortium for developing the payload and scientific operation of the project [3].
The main goal of the PLATO mission is the detection and characterization
of Earth-like exoplanets orbiting in the habitable zone of solar-type stars. It
achieves its scientific objectives by long uninterrupted ultra-high precision pho-
tometric monitoring of large samples of bright stars. This requires a large Field
of View (FoV) as well as a low noise level. To achieve a high pupil size and
the required FOV the instrument contains 26 telescopes for star observation. 24
normal cameras monitor stars fainter than magnitude 8 at a cycle of 25 s. Two
fast cameras observe stars brighter than magnitude 8 at a cycle of 2.5 s. The
size of a fast camera FoV is 38.7◦ × 38.7◦ . The cameras are equipped with four
Charge Coupled Devices (CCD) in the focal plane, each with 4510 × 4510 pixels.
6 U. Witteck et al.

Each fast camera comes with a data processing unit running the FGS algo-
rithm. It calculates attitude data with an accuracy of milliarcseconds from the
image data. This data is supplied to the spacecraft attitude and orbit control
system. The FGS is regarded as being a mission-critical component which implies
an extensive test procedure.
Many spacecraft missions use a FGS to obtain accurate measurements of
the spacecraft orientation. We use the PLATO FGS algorithm as a case study
to investigate the efficiency of our test approach. The attitude calculation of a
telescope is based on measured star positions on the CCD compared to their
reference directions in a star catalog. Figure 1 gives an overview of the FGS
algorithm [6].
The autonomous attitude tracking is initialized with an initial attitude given
by the space craft. For each pre-selected guide star, an initial sub-window posi-
tion is calculated by means of the camera model, which transforms from sky
coordinates to pixel coordinates and vice versa [6]. Guide stars are predefined
stars in a star catalog that satisfy given criteria. For example, the star magni-
tude is within a certain range, the star has very low contamination, etc. The
FGS algorithm calculates centroids after reading 6 × 6 pixel sub-window every
2.5 s from the full CCD image.

Fig. 1. Overview of the FGS algorithm [6].

A linear center of mass calculation estimates the initial centroid position.

To get a more precise solution, the algorithm separately estimates each centroid
using a Gaussian Point Spread Function (PSF) observation model. The PSF
describes the distribution of the star light over the CCD pixels. Equation 1 shows
the Gaussian PSF observation model h(i, j) of a single pixel [6].

i+1 
j+1
Im (u−u )2 (v−vc )2
− 2σ2c
h= e du e− 2σ2 dv + D + ξ (1)
2πσ 2
i j
 Equivalence Class Definition for Automated Testing of Satellite 7

The FGS algorithm uses the measured pixel intensities to determine the cen-
troid position (uc , vc )T , intensity Im , image background D and PSF width σ.
A non-linear least square fitting method iteratively refines the parameters of
the PSF model. The FGS algorithm calculates the correction by means of the
QR-decomposition [6]. In the next step, the pixel coordinates of the calculated
centroid position are transformed into star direction vectors in the camera bore-
sight reference frame. The x- and y-axis of the detector and the optical axis of
the camera describe the boresight reference frame.
Finally, the FGS algorithm calculates an attitude, including covariance, from
at least two star directions in the boresight reference frame and the corresponding
reference vectors from a star catalog [14].

3 Related Work

Equivalence class partition testing “is probably the most widely described, and
one of the most widely practiced, software testing techniques” [8] Various studies
investigated equivalence class partition testing strategies for different domains,
for example, railway, automotive, avionics, etc. [7]. We present some previously
published work on equivalence class partition testing for real-time systems.
In the automotive domain, DaimlerChrysler Research developed a test app-
roach, called Time Partition Testing (TPT), to test the continuous behavior of
control systems. Bringmann and Krämer [2] explained the principle of the TPT
approach using an exterior headlight controller as an example. In most cases,
automotive embedded control systems are based on complex functional behavior
and large input domains. To increase the test efficiency the TPT approach sys-
tematically selects test cases revealing redundant or missing test scenarios. Using
a graphical state machine notation, the TPT approach partitions a test scenario
into stream-processing components. Each component defines the behavior of out-
put variables depending on the behavior of input variables up to a certain point
in time, specified by a temporal predicate. Test cases define variations in the
state machine to test various functional aspects of the system under test.
The study shows that state machines are suitable to partition the temporal
behavior of input and output variables in order to model, compare and select
test cases. The modeled test cases test the complex functional requirements of
control systems. A huge input domain and complex functional behavior are also
characteristics of the system class we investigate in this paper. However, the
behavior of systems from this class is not dependent on the arrival time of input
values. Hence, the TPT approach is not applicable to the system class that we
consider [14].
In [7], the authors presented a model-based black-box equivalence class par-
tition testing strategy used in the railway domain. The approach automatically
generates finite and complete test suites for safety-critical reactive systems in
relation to fault models. Huang and Peleska investigated the approach using the
Ceiling Speed Monitor of the European Train Control System as an example
for systems with potentially infinite input domain but finite output domain and
8 U. Witteck et al.

internal variables. Their approach models the reactive behavior of such systems
by means of deterministic state transition systems. Moreover, the approach par-
titions the state space into a finite number of equivalence classes such that all
states in a class provide the same output traces for the same non-empty input
trace. Based on these classes, they generates a complete test suite in the follow-
ing sense: First, at least one test in the suite fails if an application that violates a
given specification is tested. Second, each test in the suite passes for all applica-
tions that satisfy the specification. Huang and Peleska investigated models whose
behavior can be represented by state transition systems. However, we have no
state transition system description of our considered satellite application. Hence,
we present an approach that does not need such a description [14].

4 Equivalence Class Partitioning for Automated Test

Generation

Satellite on-board image processing applications require various input parame-

ters such as position of an object in the image, its brightness, sub-pixel posi-
tion, its shape to distinguish different objects, etc. This leads to a huge input
domain which makes testing expensive. Especially manual tests are error-prone
and time-consuming. Thus, a test approach is needed that automatically and
systematically generates test cases for such applications. However, a major chal-
lenge for automated test generation is the very large number of possible input
parameter combinations. This potential enormous amount of test cases makes it
hard to efficiently capture the complete input domain.

Fig. 2. Overview of the partitioning approach [14].

To overcome that problem, we define a partitioning approach that systemat-

ically selects test cases from the huge input domain of satellite on-board image
processing applications. Moreover, our test approach assesses and enhances a
given test suite. To evaluate the efficiency of our test approach, we investi-
gate a case study, namely the PLATO FGS algorithm as described in Sect. 2.2.
 Equivalence Class Definition for Automated Testing of Satellite 9

Since satellite on-board image processing algorithms are subject to extremely

strict requirements with respect to reliability and mathematical accuracy, such
algorithms require extensive testing.
Figure 2 depicts an overview of our proposed partitioning approach. Our key
idea is to define equivalence classes on input parameters that are typically used
by satellite on-board image processing applications, namely position, magnitude,
sub-pixel position, and distribution model. In this paper we present updated
equivalence class definitions to partition the individual parameters. Recent stud-
ies have shown that some of the equivalence class definitions presented in [14]
can be redefined more effectively. In a second step, we define multidimensional
coverage criteria based on a combination of the individual criteria for each input
parameter. After that, we define a test generation algorithm that automatically
selects test cases that completely cover the whole input domain according to our
multidimensional coverage criteria.
Our test objective is to automatically detect errors in the on-board image
processing application code. To achieve this, our test generation algorithm selects
a test case for each equivalence class combination from a given test suite as rep-
resentatives. This reduces the number of redundant test cases. Furthermore, our
algorithm generates new test cases for missing but relevant input combinations
to reach a complete coverage of the input domain. The result is a reasonably
small test suite that covers the whole input domain of the image processing appli-
cation with respect to our multidimensional coverage criteria. The selected test
cases serve as input for our automated testing framework. Moreover, we insert
requirements for the automated evaluation of the image processing application
results. If the test cases do not meet the requirements, an error is detected [14].
The following sections describe the mentioned steps of the partitioning app-
roach in more detail, applying our new equivalence class definitions. We use the
PLATO FGS algorithm as a case study.

4.1 Assumptions and Limitations

In the following, we consider systems whose input are objects in an image. In

the case study, the observed objects are stars with magnitudes between 5.5 to
7.0, uniformly distributed in the image [6].
We consider four parameters that affect the mathematical accuracy of the
FGS algorithm: the guide star position, its magnitude, sub-pixel position, and
PSF shape. The evaluation of the test is based on the precision of the centroid
position calculated by the FGS centroid algorithm as described in Sect. 2.2. The
input of the centroid calculation is a single star image. Hence, we define a test
star as a test case for the automated test generation.

4.2 Input Parameter Partitioning

The star signal is spread over all pixels in the sub-image. Hence, each pixel
includes information about the star. However, 90% of the energy is within 2 × 2
10 U. Witteck et al.

pixel around the centroid. Moreover, each pixel contains noise, which in com-
bination with the signal determines its Signal-to-Noise Ratio (SNR). The cen-
troid calculation needs at least 5 linear independent equations to estimate the 5
unknown parameters of the pixel observation (cf. Eq. (1)).
The FGS input star parameters named in Sect. 4.1 affect the mathematical
precision and accuracy of the centroid estimation. Hence, we define the input
domain as a set of input parameters I. The set includes the position on the Focal
Plane Assembly (FPA) P, the magnitude M, the sub-pixel position E and the
PSF shape G. The tester specifies start values to calculate the borders of the
equivalence classes. This makes our approach more flexible and parameters can
also be excluded from the analysis [14].
In this section we describe how the quality of the centroid calculation depends
on these parameters and present our partitioning concepts for each input param-
eter in I.

Position on the FPA. Among others, the distribution of the star signal
depends on the star position on the FPA. Due to optical aberrations of the
telescope, the PSF shape of the star is wider in the FPA corner than close to the
center. If the other input parameters contain reasonably good, constant values
then a small PSF leads to a low number of pixels with a high SNR. In case of
a wide PSF, more pixel contain a signal but the SNR is low. Both cases can be
sufficient for an accurate parameter estimation [14].
In [14], our idea is to partition the FPA into equally sized, circular areas.
Recent studies have shown, that the PSF changes not only with the distance to
the FPA center but also with the polar angle. In the study each class of parameter
P contains two stars per class of parameter E. The stars have a constant medium
magnitude as well as worst-case non-Gaussian PSF. Figure 3 depicts the residual
noise of stars per circular FPA area. The figure shows that the residual noise
is lower if the star is positioned near the FPA corner or near the FPA border.
Moreover, the figure illustrates that the equivalence class borders have been
well chosen since the residual noise of the stars is changed between neighboring
classes.
Figure 4 shows that the residual noise also depends on the polar angle of the
stars. The figure depicts the residual noise of stars per polar angle area. Figure 4
shows that the residual noise is different for each class. However, we consider
only stars in the image area of the CCDs. That means, for some polar angle
areas particular circular areas can not be covered by a star. Therefore, these
polar angle areas contain fewer stars than others. Moreover, the stars in these
polar angle areas are located near the FPA center. Hence, the residual noise for
that area is low. However, the polar angle area between 90◦ and 135◦ contains
less stars but the residual noise is high. This indicates, that this area is not
suitable to select guide stars for the PLATO mission from there.
Bases on the study, we update our equivalence class definition of the input
parameter P and additionally partition the polar angle in equally sized circular
sectors.
 Equivalence Class Definition for Automated Testing of Satellite 11

0.005

0.004

[px]

0.003
Residual

0.002
σ

0.001

0
0 1000 2000 3000 4000
Radius [px]

Fig. 3. Residual noise per radius of circular FPA areas.

0.005

0.004
[px]

0.003
Residual

0.002
σ

0.001

0
0 100 200 300
Theta [Deg]

Fig. 4. Residual noise per polar angle of circular FPA areas.

The updated equivalence class definition is illustrated in Fig. 5. The rectan-

gles represent the image area of the fast cameras CCDs and each circular ring
sector corresponds to one equivalence class. The tester specifies the initial radius
r0 and the angle of the circular vectors θ0 .
12 U. Witteck et al.

4000

2000

θ0
r0
y [pixel]

−2000

−4000

−4000 −2000 0 2000 4000

x [pixel]

Fig. 5. FPA equivalence class example.

We partition parameter P into equivalence classes P(ri ,θj ) . Each class P(ri ,θj )
corresponds to a circular ring sector of the FPA with inner radius ri−1 and outer
radius ri as well as right polar angle θj−1 and left polar angle θj .

P = P(r0 ,θ0 ) ∪ P(r0 ,θ1 ) ∪ ... ∪ P(r0 ,θm ) ∪ ... ∪ P(rn ,θm ) (2)

where n is the number of radius border and m is the number of polar angle
border.
Let S denote the set of available stars. A star s ∈ S lies in an equivalence
class P(ri ,θj ) if following condition holds:

ri−1 ≤ p(s) < ri , with p(s) = xs 2 + ys 2 (3)

and
xs
θj−1 ≤ t(s) < θj , with t(s) = arctan (4)
ys
where (xs , ys ) is the position of star s on the FPA, p(s) is the distance of star s
to the FPA center and t(s) is the polar angle of star s.

Sub-pixel Position. In addition to the position on the FPA, the sub-pixel

position of the star also affects the SNR in a pixel. If the centroid is positioned
 Equivalence Class Definition for Automated Testing of Satellite 13

in the center of the pixel, most star flux is accumulated in a few pixels with
a high SNR. In contrast, more pixels have a sufficient SNR if the centroid is
on the pixel border or corner. In this case, the star information is distributed
more evenly over several pixels. The other pixels have a low SNR. But due to
movement, the centroid may move to neighbor pixels. This leads to variations
in the pixel illumination and the apparent centroid position [14].
In [14], we divide input parameter E into 9 sub-areas, whereas each area
corresponds to one equivalence class. In this paper, we join the corner areas,
the vertical border areas, the horizontal border areas, and the center area of the
pixel to one equivalence class each. The 4, equally sized equivalence classes are
shown in Fig. 7. Areas with the same pattern belong to the same equivalence
class.
Figure 6 depicts the mean value and standard deviation of the residuals for
stars in the respective pixel area. The stars are located in the same class of
parameter P as well as have a constant medium magnitude and a worst-case
non-Gaussian PSF. The figure shows that the residual noise is higher for stars
positioned in a pixel corner than in the pixel center. The residual noise of stars in
the horizontal border classes or vertical border classes is lower than the residual
noise in the corner classes but higher compared to the center class. It is therefore
beneficial to join the equivalence classes of input parameter E defined in [14].

0.003
[px]

0.0025
Residual

0.002
σ

0.0015

0.001
Corner Class Horizontal Class Vertical Class Center Class

Fig. 6. Residual noise per pixel class.

The tester specifies the ratio r of the central area of the pixel to the pixel
area, for example, 1/2, 3/5, etc. If a is the pixel size, then the length of the edge
of the central area results from Eq. (5).
√
b=a r (5)
14 U. Witteck et al.

Fig. 7. Example borders of pixel equivalence classes.

With that, we obtain the lower left corner l and the upper right corner u of the
central pixel area, with
a b a b a b a b
l = ( − , − ) and u = ( + , + ) (6)
2 2 2 2 2 2 2 2
Based on these corners, we partition parameter E into equivalence classes Ei
with i = 0...3. The equivalence class Ei is the i-th pixel sub-area. A star s lies
in an equivalence class if it satisfies the corresponding condition.

E = E0 ∪ E1 ∪ ... ∪ E4 (7)

E0 : (0 ≤ ex (s) < xl ∨ xu ≤ ex (s) < a) ∧ (0 ≤ ey (s) < yl ∨ yu ≤ ey (s) < a)

E1 : (0 ≤ ex (s) < xl ∨ xl ≤ ex (s) < xu ) ∧ yl ≤ ey (s) < yu
E2 : xl ≤ ex (s) < xu ∧ (0 ≤ ey (s) < yl ∨ yl ≤ ey (s) < yu )
E3 : xl ≤ ex (s) < xu ∧ yl ≤ ey (s) < yu
(8)
ex (s) and ey (s) return the x-coordinate and y-coordinate of s in the pixel
respectively.

Magnitude. The measured star flux (photo-electrons per second) depends on

the magnitude. The accumulated number of photo-electrons per pixel denotes the
 Equivalence Class Definition for Automated Testing of Satellite 15

Fig. 8. Example partitioning of magnitude range [14].

illumination of a pixel. Equation (9) shows the relation between the magnitude
m and the corresponding flux Fm in e− /s.

Fm = F0 T QA ∗ 10−0.4∗m (9)

with magnitude m, reference flux F0 of a star with m = 0, transmission efficiency

T of the optical system, quantum efficiency Q of the detector, and effective
light-collecting area A. As the equation shows, the star flux is non-linear to
the magnitude of the star. A low magnitude corresponds to a high number of
photo-electrons, that leads to a higher SNR per pixel.
A useful partitioning of magnitude values into equivalence classes is not obvi-
ous. Our idea is to partition the star flux range into IM ∈ N equidistant parts
that represent the equivalence classes. We define Eq. (10) to obtain the upper
limit of a sub-range.
F5.5 − F7.0
Fmj = F7.0 + j (10)
IM
Fmj is the flux of magnitude mj and j = 1...IM represents the j-th equivalence
class of parameter M. F5.5 and F7.0 correspond to the numbers of photons for
magnitude 5.5 and 7.0. First, we calculate the flux values F5.5 and F7.0 by using
Eq. (9). Then, we partition the flux range into equidistant sub-ranges. We use
Eq. (11) to recalculate the magnitude mj from the calculated flux limit Fmj of
the flux sub-range j.  
Fm
m = −2.5 log (11)
F0 T QA
From a formal point of view, we partition the parameter M into equivalence
classes Ml .
M = M7.0 ∪ ... ∪ Mlj ∪ ... ∪ M5.5 (12)
with lj ∈ R and 5.5 ≤ lj ≤ 7.0. Each equivalence class Mlj is a magnitude
sub-range with upper limit lj . Each available star s lies in equivalence Mlj if it
satisfies the condition in Eq. (13).

lj−1 ≤ m(s) < lj (13)

where m(s) denotes the observed magnitude of star s and lj with j = 1...IM is
the upper limit of the j-th magnitude sub-range. The tester specifies the num-
ber of equivalence classes IM ∈ N of the parameter M. Figure 8 illustrates an
example partitioning of the magnitude range [14].
16 U. Witteck et al.

PSF Shape. The accuracy of the centroid calculation also depends on the PSF
shape. In the best case scenario, the shape is a symmetric Gaussian-PSF. Then,
the observation model (cf. Eq. (1)) perfectly fits the star. Therefore, the accuracy
of the centroid calculation is high. In reality, the PSF shape is non-Gaussian. In
that case, the observation model is less accurate and movements lead to stronger
variations in the expected centroid positions [14].

Fig. 9. Examples of different low quality stars [14].

We partition the input parameter G in two equivalence classes GG and GN G

since two PSF shapes are distinctive. If a star has a Gaussian-PSF shape it is in
class GG otherwise it is in class GN G .
Figure 9 shows some example stars with non-Gaussian-PSF shape that are
less suitable as guide stars. These stars lead to inaccurate estimation results.
What the three stars have in common is that their intensity is concentrated
on a pixel edge. For all stars, the magnitude and FPA position are sufficiently
good. However, a small variation due to movement leads to big changes of the
illumination. Since the Gaussian-PSF observation model does not fit the PSF
shape perfectly, the centroid estimation is less accurate.

4.3 Multidimensional Coverage Criteria

This section presents our definition of multidimensional coverage criteria on the

input domain I = {P, M, E, G} [14]. While individual parameter values might
provide a good centroid estimation, a combination of parameters may change
the quality of the results. To measure the coverage of a test suite with respect to
input parameter combinations we define multidimensional coverage criteria on
the input domain. If the measured coverage of a test suite is not complete, our
automated test generation algorithm automatically inserts test cases for missing
combinations.
The individual coverage of an input parameter denotes the ratio of equiva-
lence classes that are covered by at least one test case from a given test suite to
the number of equivalence classes of this input parameter. Equations (14)–(17)
show this definition for the input parameters P, M, E and G.
 Equivalence Class Definition for Automated Testing of Satellite 17

# covered elements of P
CP = (14)
|P|

# covered elements of M
CM = (15)
|M|
# covered elements of E
CE = (16)
|E|
# covered elements of G
CG = (17)
|G|
The Cartesian product of the equivalence classes of the input parameters P,
M, E and G is the coverage domain for our multidimensional coverage criteria.
Hence, an input combination is a tuple of equivalence classes (Pi , Mj , Ek , Gl ),
where Pi ∈ P, Mj ∈ M, Ek ∈ E and Gl ∈ G [14]. Furthermore, a test case is a
star represented by a tuple of parameter values ((p, t), m, e, g) ∈ (Pi , Mj , Ek , Gl ).
The following example test cases clarify these definitions.

Example 1
 
(1969.4, 322.5), 6.5, (0.3, 0.2), G ∈ (P(2687,360) × M6.6 × E2 × GG )

The test star position is in the FPA area with outer radius 2687 and outer polar
angle 225◦ . The star belongs to equivalence class M6.6 because its magnitude
value is between 6.3 and 6.6. The star center is located in the lower-middle
pixel sub-area. That corresponds to the horizontal pixel areas and therefore to
equivalence class E2 . The star is part of equivalence class GG , because it has a
Gaussian-PSF shape.

Example 2
 
(2551.9, 357.3), 6.5, (0.9, 0.8), G ∈ (P(2687,360) × M6.6 × E0 × GG )

The test star is similar to the star in the first example, but it is positioned nearby
the upper right pixel border and therefore belongs to equivalence class E0 .

Our multidimensional coverage criterion is fully satisfied if the test cases in

a test suite cover all possible input combinations at least once. The number
of required covered input combinations for a complete coverage is |P × M ×
E × G|. In the remaining sections, we denote a test suite that completely covers
the input domain with respect to our multidimensional coverage criteria as a
complete test suite. The multidimensional coverage C results from the ratio of
input combinations covered by at least one test case to the total number of input
combinations.
# covered input combinations
C= (18)
|P × M × E × G|
18 U. Witteck et al.

Our test approach calculates the individual and multidimensional coverage of a

given test suite using Algorithm 1. The input parameters P, M, E, and G contain
IP , IM , IE , IG equivalence classes respectively [14].
For each test case in the given test suite, the algorithm computes the
input parameter index iP , iM , iE , iG of the corresponding equivalence class from
P, M, E and G. The algorithm adds the indices to the sets CP , CM , CE and
CG respectively. Moreover, it inserts the tuple (iP , iM , iE , iG ) into the set C that
contains all covered input combinations. As the algorithm uses the union opera-
tor to add the tuples to the set, each tuple is included in the set only once. The
algorithm applies Eqs. (14)–(18) to compute the individual and multidimensional
coverage.

Input: Test suite T S

Output: Multidimensional coverage Cov of T S
1 CP = CM = CE = CG = C = ∅;
2 foreach tc with ((p, t), m, e, g) ∈ TS do
3 iP = getPosECId(p, t);
4 CP ← CP ∪ iP ;
5 iM = getMagECId(m);
6 CM ← CM ∪ iM ;
7 iE = getPixECId(e);
8 CE ← CE ∪ iE ;
9 iG = getModECId(g);
10 CG ← CG ∪ iG ;
11 C ← C ∪ (iP , iM , iE , iG );
12 end
13 CovG = |CP |/IP ;
14 CovM = |CM |/IM ;
15 CovE = |CE |/IE ;
16 CovG = |CG |/IG ;
17 Cov = |C|/(IP · IM · IE · IG )

Algorithm 1. Coverage calculation [14].

Our partitioning approach uses individual and multidimensional coverage

criteria to assess the quality of test suites with respect to their coverage on the
input space of a satellite on-board image processing application [14].

4.4 Automated Test Generation

We present a test generation algorithm to automatically and systematically gen-
erate a test suite that completely covers the input domain according to our
multidimensional coverage criteria. The complete test generation algorithm uses
Algorithm 1 to assess a given test suite and systematically generates missing test
cases based on this result.
in mouth are

one

used the on

swallows

will doubtful friendly

ease
zeal

reduced her drowning

Liberals and has

out

social and on
pretensions them

to

himself assured retained

United

this the in

Black that remember

Marin in

a thinks at
if

the British are

of

for

persons convince

other of closed
its intellect pass

day

xl

is

feasts shall

Progress who are

the chap

4i7 of

of preserve

to us
that his the

as grand

in the

even party

himself
lot it

hands

called lawfulness

do fortress in

so

After instru

looked

the are
for will Crape

in ipsam

visible the

desires

know

good Baku

are

18
had and 368

the writer has

ever a

some fills

sees the Hexateuch

sacred despite divided

full

of is brought

The

the
as But his

seeing

on Tao

Whenever deserves the

ancestors with

to a of

have good

is in

voting
small they the

of the limits

good

make directed a

sea these test

very the more

look

be Company

Plot great and

share and of

once plain

historian so

Church the

special

we
did 1759 we

strolen yet

Devastation Neustria

at sports sublata

combat Even the

limb above

the 17 American

in
imitation progression governed

in

must process

the avoid true

are and means

1886 recognizing the

and

and

the

attributed the it

means that a

used used

been Education

progression glory

a appearance chasms
brings rights a

not whose labour

around were but

inhuman forbidden matters

the

say

299 Union of

which the PCs

meditationes all ch
distinctive

interfered

was the

much

prayer and corruption

in place expression
sacerdotalium horizon English

when best

the Unknowable

the in

nomen by that

whether

more in barge

well the

as
chapters and

history last

of in it

than it screaming

ill network

a
take re

people

singing the be

the

on

day

of

A fast

your the

through labours
engage opinion birth

song proper

Natural say compounded

country be as

travelled

necessary

iniquity sympathies

appear of the

early the Caspian

to the tui

occasional

can UusullieJ W

last sought

to
the

the the

the Mr of

confined to

of

to one Blackwood
are know

from

her Entrance Biblical

religion marriage Besides

the has States

story
their persuaded

Kalendar

orbis

rades as when

of united
princes

frost beliefs

depots at

into

there

be quo

course

In point you

first Such

demanded
and

others and as

the of

the with not

of to has

Gheyn enabled

grave have

a of E

bear but
their translated

enough and Sanctus

of into as

ring

had

several

would and Petroleum

testimony to

filled hearts

in

Salvation should

well them of

the world

portions

letter land samples

perhaps cracked has

and both sleep

so you

is that strongly

large attainments years

such with

and the the

to by
11 co waver

9th almost sink

way with

Baret adolescentiam

Roman

come had which

skulls the forgetfulness

dismal years ought

relaxation in

did of can

also it must

sand protests

more tier

as

on of chapter

employed Ecclesiae Browne

under The the

the or the

so

Thus

have and

nemo

the

be the ratia

the
be am

means in

American in production

of of

if loudly

classes than what

building

duty
religious original

and destroy 1886

back the registering

place teaches to

passage

stout Trick
8

was him

of the interest

in the whom

said 100

literature

and from

such
crimson him

trials

the

erat years rate

between
honest quod any

troubled the

much of

chosen cause

and engines

which condition all

in had

was

the worst

Duke that have

increases the catastrophe

I left place

be

Life lake to

known Ages
case played

the Three sudden

aware

this enough an

is does it

mineral their that

deposits day of

EiethmUlIer and growth

she a

to primitive

with

from

una

queunt warrant

1
t having spread

and more

it keep locks

on date

resign merchant

surrounded remember flock

nothing was probably

extended absolute
Holy for

of literati

These stata towards

the Dunn

25

from

The Before It

page are
may regard own

an a

come p

them which

Others

of well century

of regard 8

indeed it next

authority cold Sarum

circumscriptum
Ireland cogantur the

wholly

double Christ Society

and 488 enjoined

system last the

and spouts he

nineteenth

are

in coals its

minds it die
is to blast

own Three

Associations

to

was

medicinal the

till to Nobel
hush the

Notices

and

for Endowments

the of i

analogy deliberately doctrine

Tao violence

old

this old

far straight

religion Stephanus

he Paris any

upon it at
blow with

establish settlements

of allow

the

the the compiler

enjoyed love

the Hence proof

dramatic

before

have
even liberal

the The

water

The powerful

strolen good

be has in

ever for
that

from brought hollow

what

is of that

the

the England

read However remarks

first survives

good in the

to of
Splrltital in all

interests designatam a

long

caparisoned truth and

sea is the

the casement a

he if

splendour inculcate of

the influences party

other
To party

Taoism of defeating

nisi is

Farjeon

Ap estate

and giving there

normally countries of

weaker s weep

and will
to to may

and

sundry poor explored

as nor

Catholic to discrepancies

and

what

Sovereign may

This of
35 ATLANTIS dragon

ions punishing

com Rudolph needs

youth

of power sensitive

p not

southern to strong

men and
rich on

to place feminine

in assail

of put that

moral Conditions

It synonymous more
These to

world handle think

late before principles

His

being and the

removed

and of

Longfellow

and about
wisdom

seems

from quarrel

way

Tcheleken as

in

impress Our veracity

the which he
people though have

windy might

especially this

speech repaired

rather

which it

Essence
facts

part

was vol of

the

readers black

by yellow

the if soil

within possibly
the Tchorni would

We towards speaks

saying efforts

treated

future

definition

entering schohasts

is cared

war and to

cattle PCs
as the

adept night

of word for

usefully celebrated

will Crusaders of
devotion in

born

special second

competitive possibly interesting

the Mr
To is heretics

nobilia a put

he lies way

particular is a

minimum and with

to
pro

you between

to

quarter a

this and are

94

the

many

is of pp
will immediate connected

of of

a its

Central

it

devoted this

Kingdom

no order characteristics

in strong abstract
first Nihilism he

the

the six

orders

of

further so of

essays of

also poor on

had responsible and

caution triumph through

up real quaedam

so miles

are

in

virtue
concessions City

watch of earned

intellectual the all

kings of

scene

he of numerous

index entirely

of tomb strike

moment a
that

to present

the it for

standard more

in cannot far

of now

fit

that the

their
of the

the as

rudeness development Cross

in 0

com
he Demons

of Mr Furthermore

Cottage Merv

still

Having periodical folly

our way no

lock now

let allowed Travel

the
very Lucas Meyer

Then a

to is past

is in discover

Serjeant the

such

three 1
of told account

Not

accurate lucky for

His Solomon pariter

theory the

course NO this
paganism he After

of

be

of

among

consequence a

single as The

of the
associated the

and

Timbuktu

how

than

fact as 1

history sufficiently The

between amongst

attack Galieia article

ride charge

each

is accepted of
power upon

until northern temples

day

from literal that

it

of

of of they
no

and

has the

state a

and is small

on
existing times

eighty

Alike of brethren

of

for

supposed and

or

microbe with reality

History by

humanoids you provinces

subordination in

unico country

should

1513

got law