Position Navigation and Timing Technologies in

the 21st Century Integrated Satellite Navigation

Sensor Systems and Civil Applications Volume 2
1st Edition Y. Jade Morton (Editor) pdf download
https://ebookgate.com/product/position-navigation-and-timing-
technologies-in-the-21st-century-integrated-satellite-navigation-
sensor-systems-and-civil-applications-volume-2-1st-edition-y-
jade-morton-editor/

Get Instant Ebook Downloads – Browse at https://ebookgate.com

Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Position Navigation and Timing Technologies in the 21st

Century Integrated Satellite Navigation Sensor Systems and
Civil Applications Volume 1 1st Edition Y. Jade Morton
(Editor)
https://ebookgate.com/product/position-navigation-and-timing-
technologies-in-the-21st-century-integrated-satellite-navigation-
sensor-systems-and-civil-applications-volume-1-1st-edition-y-jade-
morton-editor/
ebookgate.com

Global Positioning Systems Inertial Navigation and

Integration Second Edition Mohinder S. Grewal

https://ebookgate.com/product/global-positioning-systems-inertial-
navigation-and-integration-second-edition-mohinder-s-grewal/

ebookgate.com

Aircraft communications and navigation systems principles

operation and maintenance 1st Edition Michael H. Tooley

https://ebookgate.com/product/aircraft-communications-and-navigation-
systems-principles-operation-and-maintenance-1st-edition-michael-h-
tooley/
ebookgate.com

After the Map Cartography Navigation and the

Transformation of Territory in the Twentieth Century 1st
Edition William Rankin
https://ebookgate.com/product/after-the-map-cartography-navigation-
and-the-transformation-of-territory-in-the-twentieth-century-1st-
edition-william-rankin/
ebookgate.com
Global positioning systems inertial navigation and
integration 2nd ed Edition Mohinder S Grewal

https://ebookgate.com/product/global-positioning-systems-inertial-
navigation-and-integration-2nd-ed-edition-mohinder-s-grewal/

ebookgate.com

The Career Navigation Handbook 1st Edition Christopher

Hunt

https://ebookgate.com/product/the-career-navigation-handbook-1st-
edition-christopher-hunt/

ebookgate.com

Innovative Graphene Technologies Evaluation and

Applications Volume 2 1st Edition Tiwari

https://ebookgate.com/product/innovative-graphene-technologies-
evaluation-and-applications-volume-2-1st-edition-tiwari/

ebookgate.com

RYA Navigation Handbook First Edition Tim Bartlett

https://ebookgate.com/product/rya-navigation-handbook-first-edition-
tim-bartlett/

ebookgate.com

Pre trial detention in 20th and 21st Century Common Law

and Civil Law Systems 1st Edition Marion Charret-Del Bove

https://ebookgate.com/product/pre-trial-detention-in-20th-and-21st-
century-common-law-and-civil-law-systems-1st-edition-marion-charret-
del-bove/
ebookgate.com
Position, Navigation, and Timing
Technologies in the 21st Century
 IEEE Press
445 Hoes Lane
Piscataway, NJ 08854

IEEE Press Editorial Board

Ekram Hossain, Editor in Chief

Jón Atli Benediktsson David Alan Grier Elya B. Joffe

Xiaoou Li Peter Lian Andreas Molisch
Saeid Nahavandi Jeffrey Reed Diomidis Spinellis
Sarah Spurgeon Ahmet Murat Tekalp
Position, Navigation, and Timing Technologies
in the 21st Century

Integrated Satellite Navigation, Sensor Systems, and Civil Applications

Volume 2

Edited by
Y. T. Jade Morton, University of Colorado Boulder
Frank van Diggelen, Google
James J. Spilker, Jr., Stanford University
Bradford W. Parkinson, Stanford University

Associate Editors: Sherman Lo, Stanford University

Grace Gao, Stanford University
Copyright © 2021 by The Institute of Electrical and Electronics Engineers, Inc. All rights reserved.

Published by John Wiley & Sons, Inc., Hoboken, New Jersey.

Published simultaneously in Canada.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical,
photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without
either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470, or on the web at www.copyright.com. Requests to the
Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201)
748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permissions.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no
representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied
warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales
materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where
appropriate. Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to
special, incidental, consequential, or other damages.

For general information on our other products and services or for technical support, please contact our Customer Care Department within the
United States at (800) 762-2974, outside the United States at (317) 572-3993 or fax (317) 572-4002.

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For
more information about Wiley products, visit our web site at www.wiley.com.

Library of Congress Cataloging-in-Publication Data is available.

The CiP data for ISBN 9781119458340 has been applied.

Cover Design: Wiley

Cover Images: Global telecommunication network
© NicoElNino/Getty Images, GPS Satellite
© BlackJack3D/iStockphoto

Set in 9.5/12.5pt STIXTwoText by SPi Global, Pondicherry, India

10 9 8 7 6 5 4 3 2 1
In Memory of:
Ronald L. Beard
Per Enge
Ronald Hatch
David Last
James J. Spilker, Jr.
James B. Y. Tsui
 vii

Contents

Preface xiii
Contributors xv

Part A Satellite Navigation Systems 1

1 Introduction, Early History, and Assuring PNT (PTA) 3

Bradford W. Parkinson, Y.T. Jade Morton, Frank van Diggelen, and James J. Spilker Jr.

2 Fundamentals of Satellite-Based Navigation and Timing 43

John W. Betz

3 The Navstar Global Positioning System 65

John W. Betz

4 GLONASS 87
S. Karutin, N. Testoedov, A. Tyulin, and A. Bolkunov

5 GALILEO 105
José Ángel Ávila Rodríguez, Jörg Hahn, Miguel Manteiga Bautista, and Eric Chatre

6 BeiDou Navigation Satellite System 143

Mingquan Lu and Zheng Yao

7 IRNSS 171
Vyasaraj Rao

8 Quasi-Zenith Satellite System 187

Satoshi Kogure, Yasuhiko Kawazu, and Takeyasu Sakai

9 GNSS Interoperability 205

Thomas A. Stansell, Jr.

10 GNSS Signal Quality Monitoring 215

Frank van Graas and Sabrina Ugazio

11 GNSS Orbit Determination and Time Synchronization 233

Oliver Montenbruck and Peter Steigenberger

12 Ground-Based Augmentation System 259

Boris Pervan

ftoc.3d 7 15/12/2020 6:59:11 PM

 viii Contents

13 Satellite-Based Augmentation Systems (SBASs) 277

Todd Walter

Part B Satellite Navigation Technologies 307

14 Fundamentals and Overview of GNSS Receivers 309

Sanjeev Gunawardena and Y.T. Jade Morton

15 GNSS Receiver Signal Tracking 339

Y.T. Jade Morton, R. Yang, and B. Breitsch

16 Vector Processing 377

Matthew V. Lashley, Scott Martin, and James Sennott

17 Assisted GNSS 419

Frank van Diggelen

18 High-Sensitivity GNSS 445

Frank van Diggelen

19 Relative Positioning and Real-Time Kinematic (RTK) 481

Sunil Bisnath

20 GNSS Precise Point Positioning 503

Peter J.G. Teunissen

21 Direct Position Estimation 529

Pau Closas and Grace Gao

22 Robust Positioning in the Presence of Multipath and NLOS GNSS Signals 551
Gary A. McGraw, Paul D. Groves, and Benjamin W. Ashman

23 GNSS Integrity and Receiver Autonomous Integrity Monitoring (RAIM) 591

Sam Pullen and Mathieu Joerger

24 Interference: Origins, Effects, and Mitigation 619

Logan Scott

25 Civilian GNSS Spoofing, Detection, and Recovery 655

Mark Psiaki and Todd Humphreys

26 GNSS Receiver Antennas and Antenna Array Signal Processing 681

Andrew O’Brien, Chi-Chih Chen and Inder J. Gupta

Part C Satellite Navigation for Engineering and Scientific Applications 717

27 Global Geodesy and Reference Frames 719

Chris Rizos, Zuheir Altamimi, and Gary Johnston

28 GNSS Geodesy in Geophysics, Natural Hazards, Climate‚ and the Environment 741
Yehuda Bock and Shimon Wdowinski

ftoc.3d 8 15/12/2020 6:59:24 PM

 Contents ix

29 Distributing Time and Frequency Information 821

Judah Levine

30 GNSS for Neutral Atmosphere and Severe Weather Monitoring 849

Hugues Brenot

31 Ionospheric Effects, Monitoring, and Mitigation Techniques 879

Y. Jade Morton, Zhe Yang, Brian Breitsch, Harrison Bourne, and Charles Rino

32 GNSS Observation for Detection, Monitoring, and Forecasting Natural and Man-Made Hazardous Events 939
Panagiotis Vergados, Attila Komjathy, and Xing Meng

33 GNSS Radio Occultation 971

Anthony J. Mannucci, Chi O. Ao, and Walton Williamson

34 GNSS Reflectometry for Earth Remote Sensing 1015

James Garrison, Valery U. Zavorotny, Alejandro Egido, Kristine M. Larson, Felipe Nievinski,
Antonio Mollfulleda, Giulio Ruffini, Francisco Martin, and Christine Gommenginger

Part D Position, Navigation, and Timing Using Radio Signals-of-Opportunity 1115

35 Overview of Volume 2: Integrated PNT Technologies and Applications 1117

John F. Raquet

36 Nonlinear Recursive Estimation for Integrated Navigation Systems 1121

Michael J. Veth

37 Overview of Indoor Navigation Techniques 1141

Sudeep Pasricha

38 Navigation with Cellular Signals of Opportunity 1171

Zaher (Zak) M. Kassas

39 Position, Navigation and Timing with Dedicated Metropolitan Beacon Systems 1225
Subbu Meiyappan, Arun Raghupathy, and Ganesh Pattabiraman

40 Navigation with Terrestrial Digital Broadcasting Signals 1243

Chun Yang

41 Navigation with Low-Frequency Radio Signals 1281

Wouter Pelgrum and Charles Schue

42 Adaptive Radar Navigation 1335

Kyle Kauffman

43 Navigation from Low Earth Orbit 1359

Part 1: Concept, Current Capability, and Future Promise
Tyler G.R. Reid, Todd Walter, Per K. Enge, David Lawrence, H. Stewart Cobb, Greg Gutt,
Michael O’Conner, and David Whelan

43 Navigation from Low-Earth Orbit 1381

Part 2: Models, Implementation, and Performance
Zaher (Zak) M. Kassas

ftoc.3d 9 15/12/2020 6:59:25 PM

 x Contents

Part E Position, Navigation, and Timing Using Non-Radio signals of Opportunity 1413

44 Inertial Navigation Sensors 1415

Stephen P. Smith

45 MEMS Inertial Sensors 1435

Alissa M. Fitzgerald

46 GNSS-INS Integration 1447

Part 1: Fundamentals of GNSS-INS Integration
Andrey Soloviev

46 GNSS-INS Integration 1481

Part 2: GNSS/IMU Integration Using a Segmented Approach
James Farrell and Maarten Uijt Haag

47 Atomic Clocks for GNSS 1497

Leo Hollberg

48 Positioning Using Magnetic Fields 1521

Aaron Canciani and John F. Raquet

49 Laser-Based Navigation 1541

Maarten Uijt de Haag, Zhen Zhu, and Jacob Campbell

50 Image-Aided Navigation – Concepts and Applications 1571

Michael J. Veth and John F. Raquet

51 Digital Photogrammetry 1597

Charles Toth and Zoltan Koppanyi

52 Navigation Using Pulsars and Other Variable Celestial Sources 1635

Suneel Sheikh

53 Neuroscience of Navigation 1669

Meredith E. Minear and Tesalee K. Sensibaugh

54 Orientation and Navigation in the Animal World 1689

Gillian Durieux and Miriam Liedvogel

Part F Position, Navigation, and Timing for Consumer and Commercial Applications 1711

55 GNSS Applications in Surveying and Mobile Mapping 1713

Naser El-Sheimy and Zahra Lari

56 Precision Agriculture 1735

Arthur F. Lange and John Peake

57 Wearables 1749
Mark Gretton and Peter Frans Pauwels

58 Navigation in Advanced Driver Assistance Systems and Automated Driving 1769

David Bevly and Scott Martin

ftoc.3d 10 15/12/2020 6:59:25 PM

 Contents xi

59 Train Control and Rail Traffic Management Systems 1811

Alessandro Neri

60 Commercial Unmanned Aircraft Systems (UAS) 1839

Maarten Uijt de Haag, Evan Dill, Steven Young, and Mathieu Joerger

61 Navigation for Aviation 1871

Sherman Lo

62 Orbit Determination with GNSS 1893

Yoaz Bar-Sever

63 Satellite Formation Flying and Rendezvous 1921

Simone D’Amico and J. Russell Carpenter

64 Navigation in the Arctic 1947

Tyler G. R. Reid, Todd Walter, Robert Guinness, Sarang Thombre, Heidi Kuusniemi, and Norvald Kjerstad

Glossary, Definitions, and Notation Conventions 1971

Index I1

ftoc.3d 11 15/12/2020 6:59:25 PM

 xiii

Preface

The ability to navigate has been an essential skill for applications. It starts with a historical perspective of GPS
survival throughout human history. As navigation has and other related PNT developments. Part A consists of
advanced, it has become almost inseparable from the ability 12 chapters that describe the fundamental principles and
to tell time. Today, position, navigation, and timing (PNT) latest developments of all global and regional navigation
technologies play an essential role in our modern society. satellite systems (GNSSs and RNSSs), design strategies that
Much of our reliance on PNT is the result of the availability enable their coexistence and mutual benefits, their signal
of the Global Positioning System (GPS) and the growing quality monitoring, satellite orbit and time synchroniza-
family of Global Navigation Satellite Systems (GNSSs). Sat- tion, and satellite- and ground-based systems that provide
ellite-based navigation and other PNT technologies are augmentation information to improve the accuracy of
being used in the many fast-growing, widespread, civilian navigation solutions. Part B contains 13 chapters. These
applications worldwide. A report sponsored by the US provide a comprehensive review of recent progress in satel-
National Institute of Standards and Technology (NIST) lite navigation receiver technologies such as receiver archi-
on the economic benefits of GPS indicated that GPS alone tecture, signal tracking, vector processing, assisted and
has generated a $1.4 trillion economic benefit in the private high-sensitivity GNSS, precise point positioning and real-
sector by 2019, and that the loss of GPS service would have a time kinematic (RTK) systems, direct position estimation
$1 billion per-day negative impact.1 PNT has become a techniques, and GNSS antennas and array signal processing.
pillar of our modern society. Knowledge and education Also covered are topics on the challenges of multipath-rich
are essential for the continued advancement of PNT urban environments, in handling spoofing and interference,
technologies to meet the increasing demand from society. and in ensuring PNT integrity. Part C finishes the volume
That is the rationale that led to the creation of this book. with 8 chapters on satellite navigation for engineering and
While there are many publications and several outstand- scientific applications. A review of global geodesy and refer-
ing books on satellite navigation technologies and related ence frames sets the stage for discussions on the broad field
subjects, this two-volume set offers a uniquely comprehen- of geodetic sciences, followed by a chapter on the important
sive coverage of the latest developments in the broad field of subject of GNSS-based time and frequency distribution.
PNT and has been written by world-renowned experts in GNSS signals have provided a popular passive sensing tool
each chapter’s subject area. It is written for researchers, for troposphere, ionosphere, and Earth surface monitoring.
engineers, scientists, and students who are interested in Three chapters are dedicated to severe weather, ionospheric
learning about the latest developments in satellite-based effects, and hazardous event monitoring. Finally, a compre-
PNT technologies and civilian applications. It also exam- hensive treatment of GNSS radio occultation and reflectom-
ines alternative navigation technologies based on other sig- etry is provided.
nals and sensors and offers a comprehensive treatment of The three parts in Volume 2 address PNT using alterna-
integrated PNT systems for consumer and commercial tive signals and sensors and integrated PNT technologies
applications. for consumer and commercial applications. An overview
The two-volume set contains 64 chapters organized into chapter provides the motivation and organization of the
six parts. Each volume contains three parts. Volume 1 volume, followed by a chapter on nonlinear estimation
focuses on satellite navigation systems, technologies, and methods which are often employed in navigation system
modeling and sensor integration. Part D devotes 7 chapters
to using various radio signals transmitted from sources on
1 RTI International Final Report, Sponsored by the US National
Institute of Standards and Technology, “Economic Benefits of the the ground, from aircraft, or from low Earth orbit (LEO)
Global Positioning System (GPS),” June 2019. satellites for PNT purposes. Many of these signals were
xiv Preface

intended for other functions, such as broadcasting, net- provided valuable input and comments to other chapters
working, and imaging and surveillance. In Part E, there in the book. We also sought input from graduate students
are 8 chapters covering a broad range of non-radio fre- and postdocs in the field as they will be the primary users
quency sensors operating in both passive and active modes and represent the future of the field. We want to acknowl-
to produce navigation solutions, including MEMS inertial edge the following individuals who have supported or
sensors, advances in clock technologies, magnetometers, encouraged the effort and/or helped to improve the con-
imaging, LiDAR, digital photogrammetry, and signals tents of the set: Michael Armatys, Penina Axelrad, John
received from celestial bodies. A tutorial-style chapter on Betz, Rebecca Bishop, Michael Brassch, Brian Breitsch,
multiple approaches to GNSS/INS integration methods Phil Brunner, Russell Carpenter, Charles Carrano, Ian
is included in Part E. Also included in Part E are chapters Collett, Anthea Coster, Mark Crews, Patricia Doherty,
on the neuroscience of navigation and animal navigation. Chip Eschenfelder, Hugo Fruehauf, Gaylord Green,
Finally, Part F presents a collection of work on contem- Richard Greenspan, Yu Jiao, Kyle Kauffman, Tom
porary PNT applications such as surveying and mobile Langenstein, Gerard Lachapelle, Richard Langley, Robert
mapping, precision agriculture, wearable systems, auto- Lutwak, Jake Mashburn, James J. Miller, Mikel Miller,
mated driving, train control, commercial unmanned air- Pratap Misra, Oliver Montenbruck, Sam Pullen, Stuart
craft systems, aviation, satellite orbit determination and Riley, Chuck Schue, Logan Scott, Steve Taylor, Peter Teu-
formation flying, and navigation in the unique Arctic nissen, Jim Torley, A. J. van Dierendonck, Eric Vinande,
environment. Jun Wang, Pai Wang, Yang Wang, Phil Ward, Dongyang
The chapters in this book were written by 131 authors Xu, Rong Yang, and Zhe Yang. The Wiley-IEEE Press
from 18 countries over a period of 5 years. Because of the team has demonstrated great patience and flexibility
diverse nature of the authorship and the topics covered throughout the five-year gestation period of this project.
in the two volumes, the chapters were written in a variety And our families have shown great understanding, gener-
of styles. Some are presented as high-level reviews of prog- ously allowing us to spend a seemingly endless amount of
ress in specific subject areas, while others are tutorials with time to complete the set.
detailed quantitative analysis. A few chapters include links This project was the brainchild of Dr. James Spilker, Jr. He
to MATLAB or Python example code as well as test data for remained a fervent supporter until his passing in October
those readers who desire to have hands-on practice. The 2019. A pioneer of GPS civil signal structure and receiver
collective goal is to appeal to industry professionals, technologies, Dr. Spilker was truly the inspiration behind
researchers, and academics involved with the science, engi- this effort. During the writing of this book set, several pio-
neering, and application of PNT technologies. A website, neers in the field of GNSS and PNT, including Ronald Beard,
pnt21book.com, provides chapter summaries; downloada- Per Enge, Ronald Hatch, David Last, and James Tsui also
ble code examples, data, worked homework examples, passed away. This set is dedicated to these heroes and all
select high-resolution figures, errata, and a way for readers those who laid the foundation for the field of PNT.
to provide feedback.
A comprehensive project of this scale would not be pos- Jade Morton
sible without the collective efforts of the GNSS and PNT Frank van Diggelen
community. We appreciate the leading experts in the field Bradford Parkinson
taking time from their busy schedules to answer the call in Sherman Lo
contributing to this book. Some of the authors also Grace Gao
 xv

Contributors

Zuheir Altamimi Jacob Campbell

Institut National de l’Information Géographique et Air Force Research Laboratory, United States
Forestière, France
Aaron Canciani
Chi O. Ao Air Force Institute of Technology, United States
Jet Propulsion Laboratory, United States
J. Russell Carpenter
Benjamin W. Ashman National Aeronautics and Space Administration,
National Aeronautics and Space Administration, United States
United States
Eric Châtre
Yoaz Bar-Sever European Commission, Belgium
Jet Propulsion Lab, United States
Chi-Chih Chen
Miguel Manteiga Bautista The Ohio State University, United States
European Space Agency, the Netherlands
Pau Closas
John W. Betz Northeastern University, United States
The MITRE Corporation, United States
H. Stewart Cobb
David Bevly Satelles, United States
Auburn University, United States
Simone D’Amico
Sunil Bisnath Stanford University, United States
York University, Canada
Evan Dill
Yehuda Bock National Aeronautics and Space Administration,
Scripps Institution of Oceanography, United States United States

Alexei Bolkunov Gillian Durieux

PNT Center, Russia Max Plank Institute for Evolutionary Biology, Germany

Harrison Bourne Alejandro Egido

University of Colorado Boulder, United States Starlab, Spain

Brian Breitsch Naser El-Sheimy

University of Colorado Boulder, United States University of Calgary, Canada

Hugues Brenot Per K. Enge

Royal Belgian Institute for Space Aeronomy, Belgium Stanford University, United States
xvi Contributors

James Farrell Zaher (Zak) M. Kassas

Vigil Inc., United States University of California Irvine, United States

Alissa M. Fitzgerald Kyle Kauffman

A.M. Fitzgerald & Associates, LLC, United States Integrated Solutions for Systems, United States

Grace Gao Yasuhiko Kawazu

Stanford University, United States National Space Policy Secretariat, Japan

James Garrison Norvald Kjerstad

Purdue University, United States Norwegian University of Science and Technology,
Norway
Christine Gommenginger
National Oceanography Centre, United Kingdom Satoshi Kogure
National Space Policy Secretariat, Japan
Mark Gretton
TomTom, United Kingdom Attila Komjathy
Jet Propulsion Laboratory, United States
Paul D. Groves
University College London, United Kingdom Zoltan Koppanyi
The Ohio State University, United States
Robert Guinness
Finnish Geospatial Research Institute, Finland Heidi Kuusniemi
Finnish Geospatial Research Institute, Finland
Sanjeev Gunawardena
Air Force Institute of Technology, United States Arthur F. Lange
Trimble Navigation, United States
Inder J. Gupta
The Ohio State University, United States Zahra Lari
Leica Geosystems Inc., Canada
Greg Gutt
Satelles, United States Kristine M. Larson
University of Colorado Boulder, United States
Maarten Uijt de Haag
Technische Universität Berlin, Germany Matthew V. Lashley
Georgia Tech Research Institute, United States
Jörg Hahn
European Space Agency, the Netherlands David Lawrence
Satelles, United States
Leo Hollberg
Stanford University, United States Judah Levine
National Institute of Standard and Technology,
Todd Humphreys United States
University of Texas–Austin, United States
Miriam Liedvogel
Mathieu Joerger Max Plank Institute for Evolutionary Biology,
Virginia Tech, United States Germany

Gary Johnson Sherman Lo

Geoscience Australia, Australia Stanford University, United States

Sergey Karutin Mingquan Lu

PNT Center, Russia Tsinghua University, China
 Contributors xvii

Anthony J. Mannucci Peter Frans Pauwels

Jet Propulsion Laboratory, United States TomTom, the Netherlands

Francisco Martin John Peake

Starlab, Spain Trimble Navigation, United States

Scott Martin Wouter Pelgrum

Auburn University, United States Blue Origin LLC, United States

Boris Pervan
Gary A. McGraw
Illinois Institute of Technology, United States
Collins Aerospace, United States
Mark Psiaki
Subbu Meiyappan Virginia Tech, United States
NextNav LLC, United States
Sam Pullen
Xing Meng Stanford University, United States
Jet Propulsion Laboratory, United States
Arun Raghupathy
Meredith E. Minear NextNav LLC, United States
University of Wyoming, United States
Vyasaraj Rao
Antonio Mollfulleda Accord Software and Systems, India
Starlab, Spain
John F. Raquet
Oliver Montenbruck Integrated Solutions for Systems, United States
German Aerospace Center, Germany
Tyler G. R. Reid
Y.T. Jade Morton Stanford University, United States
University of Colorado Boulder, United States
Charles Rino
Alessandro Neri University of Colorado Boulder, United States
University of Roma TRE, Italy
Chris Rizos
Felipe Nievinski University of New South Wales, Australia
UFRGS, Brazil
José Ángel Ávila Rodríguez
Andrew O’Brien European Space Agency, the Netherlands
The Ohio State University, United States
Giulio Ruffini
Michael O’Conner Starlab, Spain
Satelles, United States
Takeyasu Sakai
Bradford W. Parkinson National Institute of Maritime, Port, and Aviation
Stanford University, United States Technology, Japan

Sudeep Pasricha Charles Schue, III

Colorado State University, United States UrsaNav, Inc., United States

Ganesh Pattabiraman Logan Scott

NextNav LLC, United States LS Consulting, United States
xviii Contributors

James Sennott Frank van Graas

Tracking and Imaging Systems, United States Ohio University, United States

Tesalee K. Sensibaugh Panagiotis Vergados

University of Wyoming, United States Jet Propulsion Laboratory, United States

Suneel Sheikh Michael J. Veth

ASTER Labs, Inc., United States Veth Research Associates, United States

Todd Walter
Stephen P. Smith
Stanford University, United States
The Charles Stark Draper Laboratory Inc., United States
Shimon Wdowinski
Andrey Soloviev Florida International University, United States
QuNav, United States
David Whelan
James J. Spilker Jr. University of California San Diego, United States
Stanford University, United States
Walton Williamson
Thomas A. Stansell, Jr. Jet Propulsion Laboratory, United States
Stansell Consulting, United States
Chun Yang
Peter Steigenberger Sigtem Technology Inc., United States
German Aerospace Center, Germany
Rong Yang
Nikolai Testoedov University of Colorado Boulder, United States
PNT Center, Russia
Zhe Yang
University of Colorado Boulder,
Peter J. G. Teunissen
United States
Curtin University, Australia and Delft University of
Technology, The Netherlands
Zheng Yao
Tsinghua University, China
Sarang Thombre
Finnish Geospatial Research Institute, Finland
Steven Young
National Aeronautics and Space Administration,
Charles Toth
United States
The Ohio State University, United States
Valery U. Zavorotny
Andrei Tyulin
National Oceanic and Atmospheric Administration,
PNT Center, Russia
United States;
Sabrina Ugazio University of Colorado Boulder, United States
Ohio University, United States
Zhen Zhu
Frank van Diggelen East Carolina University,
Google, United States United States
 1115

Part D

Position, Navigation, and Timing Using Radio Signals-of-Opportunity

 1117

35

Overview of Volume 2: Integrated PNT Technologies and Applications

John F. Raquet
Integrated Solutions for Systems, United States

There is little doubt that global navigation satellite systems 35.1 Generalized Navigation
(GNSS) have changed the way that we think about and use Framework
navigation systems. Prior to GPS and other GNSSs, the use
of systems which could automatically (without human Fundamentally, virtually every navigation system oper-
intervention) determine their own position was generally ates the same way. This can be expressed as a predict–
limited to large, expensive platforms such as aircraft or observe–compare cycle, as shown in Figure 35.1. The
ships, and even these types of vehicles often required “Navigation State” at the lower right represents the user’s
human navigators to assist in the task of navigation. This
current navigation state, or all of the information about
has all changed with the advent of GNSS, however.
the user’s position, velocity, and so on, as well as esti-
Thanks to GNSS, most people have now become accus-
mates of that information’s quality. This can be thought
tomed to their smartphone or vehicle knowing exactly
of as the system’s best guess of the user’s position as well
where it is as a part of their everyday lives, and this capa-
as how accurate the system thinks the guess is. As
bility has been built into our expectations. Just as we expect
depicted in the “Sensor” box on the left, the system takes
the lights to come on when we turn on a light switch, we
also expect a GNSS position fix whenever we turn on a a measurement or makes an observation which gives
smartphone or other navigation device. This reliance on some insight into the user’s navigation state. For GPS,
GNSS goes well beyond obvious navigation devices – we perhaps the system observes the range to a satellite.
very much depend on many systems which heavily use The system also uses a model of the real world, depicted
GNSS for timing purposes, such as banking, communica- with the “World Model” box in the upper right. In the case
tions, and our power grid. of GPS, the world model might consist of the locations
Some have said that navigation is addictive – no matter (orbits) of the GPS satellites.
how much accuracy or availability you have, you always During the predict phase, the prediction algorithm deter-
want more. The extreme success of GNSS has, ironically, mines what the system expects to observe based upon the
led to a desire to complement GNSS with other types of sen- world model and the current navigation state, annotated
sors for situations in which GNSS is not available, in order as the “Prediction Algorithm” box in Figure 35.1. During
to guarantee (as much as is possible) the ability to deter- the observe phase, the system receives a noise-corrupted
mine time or position. measurement from the real world. During the compare
Volume 2 focuses in on many of these complementary phase, the predicted measurement is compared to the actual
navigation systems and methods and how they are inte- measurement. Any discrepancies are used to improve the
grated together to obtain the desired performance. Before navigation state and possibly the model of the world.
diving into the details, it can be helpful to step back and Consider a simplified example in which a user attempts
look at the big picture of what is really happening within to determine the distance to a wall. Perhaps the user pre-
navigation systems, in order to better understand how dicts the distance to the wall is about 30 feet based upon
the various approaches relate to each other. To do this, it mere eyesight to judge the distance. (The navigation state
is helpful to develop a “navigation framework.” is 30 feet with much uncertainty.) Then, suppose a precise

Position, Navigation, and Timing Technologies in the 21st Century: Integrated Satellite Navigation, Sensor Systems, and Civil Applications, Volume 2,
First Edition. Edited by Y. T. Jade Morton, Frank van Diggelen, James J. Spilker Jr., and Bradford W. Parkinson.
© 2021 The Institute of Electrical and Electronics Engineers, Inc. Published 2021 by John Wiley & Sons, Inc.
1118 35 Overview of Volume 2: Integrated PNT Technologies and Applications

World Model Updates

The Real world
World Model

“Predicted”
Data/ Data/
Sensor “Prediction”
Comparison Algorithm
Information Information

NavState Correction
Goodness of fit,
Navigation
State
Likelihood,
Residuals,etc.

Figure 35.1 General navigation framework.

laser range finder is used to measure, or observe, the dis- model. As a result, the user simply obtains the most recent
tance as 31.2 feet. Next, the prediction is compared to the ephemeris and satellite clock terms and uses them for
observation. The user quickly dismisses the prediction positioning. In this way, the user is completely uninvolved
and trusts the observation, because the user observation in the updating of the world model, which is helpful,
was viewed as being a more reliable estimate of distance because it greatly reduces the complexity of the system
than the prediction. Likewise, examples could be drawn for the user.
which highlight the prediction heavily outweighing an Unlike man-made signals, natural signals do not gener-
observation. ally have a dedicated part of the system that is continually
The most interesting applications involve a blending of updating a concise world model which describes how
the prediction with the observation. Typical GPS applica- sensed measurements relate to the real world. As a result,
tions use a Kalman filter to perform the predict–observe– the challenge with such systems is often to determine a usa-
compare cycle. The world model consists of GPS satellite ble world model. For example, it is very easy to obtain
locations. Based upon some prior information, the receiver images of the nearby environment using a camera. How-
predicts the user’s location. The observations might consist ever, in order to determine position and/or attitude from
of ranges to each satellite in view. These observations are this kind of measurement, the user must have knowledge
compared to a prediction of what the ranges should be of what the world looks like as a function of position and
based upon the receiver’s estimate of position (and assumed attitude (the world model).
knowledge of the world). The system conducts a blended
comparison based upon the relative quality of the predicted
35.1.1 What Is a Navigation Sensor?
navigation state and the observations.
In Figure 35.1, the arrow labeled “world model updates” The physical sensor, depicted as the yellow block in
indicates that the world model can be changed based upon Figure 35.1, is a critical part of any navigation system,
the measurements that have been taken. Some navigation and selection of the right sensor or combination of sensors
systems, particularly those which are designed and is one of the most important decisions a navigation system
deployed specifically for navigation, do not require the designer can make. What comprises a navigation sensor?
end user of the system to be involved in this part of the At a basic level, any physical sensor that measures some-
process. For example, in GPS, the world model consists thing which changes when the sensor is moved is a poten-
of information about the satellite orbits (ephemeris), the tial navigation sensor. Additionally, since clocks are an
satellite clock errors, and details that are given in the sig- integral part of many navigation systems, we also consider
nal specification (frequency, chipping rate, etc.). The GPS clocks in this section as well. In contrast to a navigation sen-
system uses its own receiver network on the ground to sor‚ which measures something that changes when the sen-
estimate satellite orbits and clock errors and to monitor sor is moved in some way, a clock is a sensor that measures
the signals coming from space, and measurements from how time “moves.” A summary of the major sensors cov-
this network are used to continually update the GPS world ered in Volume 2 is given in Table 35.1.
 35.2 Summary of Content of Volume 2 1119

Table 35.1 Sensors covered in Volume 2

Sensor Sensed phenomenon World model required Other considerations

Cellular RF Cellular phone RF Positions of cell towers, signal timing Example of signal of opportunity (SoOP),
receiver signals reference receiver sometimes required
Terrestrial beacon Navigation signals Beacon locations, signal structure, Requires dedicated infrastructure, more design
receiver from terrestrial signal timing flexibility than SoOP
beacons
Digital TV Digital TV signals Transmitter locations, signal timing Example of SoOP, reference receiver sometimes
receiver required
Low-frequency Low-frequency RF Transmitter location or direction of Susceptible to local distortions, generally less
receiver signals arrival, local distortion effects accurate than higher frequency/wider
bandwidth signals
Radar RF signals Locations of identifiable RF reflectors Generally larger/higher power than receiver-
for absolute positioning based systems
Low-Earth orbit Signals from LEO LEO satellite position/velocity, signal Greater geometric/signal diversity and higher
(LEO) satellite satellites timing (in some cases), atmospheric received power than GNSS
receiver models
Inertial Rotation and Gravitational field Dead-reckoning only – drift normally requires
specific force update
GNSS RF signals from Satellite ephemeris and clock errors, Ideal for updating inertial
satellites atmospheric models
Magnetometer Magnetic field Magnetic field map Local (vehicle) effects calibration may be
(including required
variations)
LiDAR Range and intensity Shape/location of objects being sensed Can be used in dead-reckoning or absolute
of laser returns modes
Cameras Intensity of light as Map of image features or three- Can be used in dead-reckoning or absolute
a function of dimensional image model for absolute modes
direction positioning
X-ray detector X-ray signals Knowledge of pulsar directions and Positioning is based on signal time of arrival
coming from signal characteristics (including
pulsars timing)
Clock Varies by clock type Perhaps calibration parameters Measures rate of time passage (frequency), and
if initialized and integrated, absolute time

35.2 Summary of Content of Volume 2 (Chapter 40), low-frequency systems (Chapter 41), radar (-
Chapter 42), and RF signals from low-Earth orbiting (LEO)
Volume 2 begins with an overview of nonlinear estimation satellites (Chapter 43).
techniques (Chapter 36), which are often required when There are two chapters that describe inertial technology:
integrating complementary navigation sensors. This chap- a general introduction to INS (Chapter 44) and MEMS iner-
ter also lays the groundwork for the estimation strategies tial systems (Chapter 45). The introduction chapter pro-
that are described in subsequent chapters. vides an overview of inertial systems. It describes the
The next group of chapters covers a variety of RF-based fundamental mechanisms of various accelerometers and
complementary navigation techniques. Many of the princi- gyroscopes that are the building blocks of INS, their error
ples and algorithmic approaches for indoor navigation are characteristics and performances, and outlook of technol-
summarized in Chapter 37, as well as a survey of different ogy advancement. The focus of MEMS inertial sensors is
types of indoor navigation sensors and phenomenologies. to reduce the cost, size, weight, and power when compared
This is followed by several chapters which describe in detail to existing inertial sensors. Doing so would expand the
a variety of RF signals, including cellular (Chapter 38), ter- applications in which it is feasible to leverage inertial
restrial navigation beacons (Chapter 39), digital television technology.
1120 35 Overview of Volume 2: Integrated PNT Technologies and Applications

It is important to recognize that inertial systems cannot cameras. The vision navigation and photogrammetry chap-
operate without aiding from additional sensors, other than ters can be thought of as opposite sides of the same coin.
for short time periods. The primary reason for this is that With vision navigation, the desire is to figure out where
inertial systems are unstable in the vertical channel, so at the camera is, based on some knowledge of the scene. With
a minimum they need some sort of aiding of the vertical photogrammetry, the desire is to figure out information
channel (such as a barometric altimeter or terrain height about the scene, based on some knowledge of the camera
aiding). Even if the vertical channel is aided, the horizontal position (and perhaps orientation).
directions will drift in an inertial system, with the rate of As mentioned earlier in this introductory chapter, any
drift determined by the quality of the system and the accu- measurement that changes when the sensor position
racy of the initialization of the attitude and position of the changes can potentially be used as a navigation source.
system. (Even if an INS had perfect gyroscopes and acceler- A good example of this is X-ray pulsar-based navigation
ometers, there would still be growing error due to imperfec- which is described in Chapter 52, along with other variable
tions in our knowledge of gravity). celestial sources for navigation. The fundamental premise
Probably the most common sensor used to aid an inertial here is that if we can accurately measure the time of arrive
is a GNSS receiver. Chapter 46 describes classic approaches of the periodic signal coming from several X-ray-emitting
for integrating GPS with INS, including loose and tight inte- pulsars, we can use this information to determine our loca-
gration. It also describes a different way of thinking about tion. Additionally, methods for performing X-ray pulsar-
the GPS/INS integration problem, in which there is more based attitude determination are given.
emphasis on using carrier-phase measurements to provide In contrast to all of the technology-based approaches
velocity-like updates to the INS, with additional correction describe thus far, Chapter 53 focuses on brain neural pro-
from the pseudorange measurements. cessing in order to perform various navigation tasks. While
Clock has been an essential sensor for navigation since these neurological approaches are quite difference from the
ancient times. The accuracy and stability of clocks continue approaches that engineers normally take to develop naviga-
the improve in recent decades. Chapter 47 provides an tion systems, the way in which navigation is done by the
overview of recent technology development in atomic brain suggest possibilities that we can attempt to imple-
clocks for GNSS. ment with various forms of computing. Chapter 54 further
An approach for using knowledge of the variation in Earth’s describes various ways in which animals are able to navi-
magnetic field for absolute positioning using a magnetometer gate and orient without the use of the modern sensors
is described in Chapter 48. This method works indoors, on a described elsewhere in this volume.
ground vehicle, and in an aircraft, and this chapter describes Volume 2 then concludes with several chapters that
the differences between these different environments and describe specific applications that make heavy use of nav-
shows examples of working systems in each case. igation systems. Many of these applications did not exist
Next, the use of LiDAR for navigation is described in prior to the arrival of GNSS, and those that did exist have
Chapter 49. Various types of LiDARs are considered, as well seen large increases in capabilities by leveraging both GNSS
as different ways in which LiDAR data can be leveraged for and complementary navigation approaches.
navigation purposes. This chapter also describes features The applications covered include survey and mobile map-
that can be identified using LiDAR data, and how those fea- ping (Chapter 55), precision agriculture (Chapter 56), wear-
tures can be incorporated into an integrated navigation sys- able navigation technology (Chapter 57), driverless vehicles
tem. Both dead-reckoning and absolute positioning/ (Chapter 58), train control (Chapter 59), unmanned aerial
attitude approaches are considered. systems (Chapter 60), aviation (Chapter 61), spacecraft nav-
Chapter 50 describes the many ways in which cameras igation and orbit determination (Chapter 62), spacecraft for-
can be used for navigation. Initially, a mathematical model mation flying and rendezvous (Chapter 63), and finally
of a camera is provided, as well as methods for camera cal- Arctic navigation (Chapter 64).
ibration. Image features are described as well as algorithms Taken together, Volume 2 shows the incredible value of
for using these features to relate camera images to position navigation systems and the variety of approaches that are
and rotation of the camera. Several methods for image nav- available in cases where GNSS is not sufficient. Whether
igation are described, and as with LiDAR, both dead- we realize it or not, our day-to-day lives are heavily depend-
reckoning and absolute positioning/attitude approaches ent on the ability of many systems that interact with (or that
are considered. Another chapter (51) is dedicated to the are behind the scenes) to determine time and position, and
topic of photogrammetry, which also uses a camera, but there is an increasing number of creative options and
lays more emphasis on using the camera in order to develop opportunities for precise navigation and time that can meet
knowledge about the scene that is viewed by one or more the needs of current and future applications.
 1121

36

Nonlinear Recursive Estimation for Integrated Navigation Systems

Michael J. Veth
Veth Research Associates, United States

36.1 Introduction include recursive estimation frameworks, the implicit

assumptions and limitations of traditional estimators, and
Almost immediately following its introduction in 1960, the deleterious effects on performance when these assump-
the Kalman filter and the extended Kalman filter have tions are not satisfied. Next an overview of nonlinear esti-
served as the primary algorithms used to solve navigation mation theory is presented with the goal of demonstrating
problems [1–3]. The optimal, recursive, and online charac- and deriving three main classes of nonlinear recursive
teristics of the algorithm are perfectly suited to serve a estimators. These include Gaussian sum filters, grid particle
wide range of applications requiring real-time navigation filters, and sampling particle filters. Each of these classes
solutions. of nonlinear recursive estimators is demonstrated and eval-
The traditional Kalman filter and extended Kalman filter uated using a simple navigation example. The chapter is
are based on the following assumptions: concluded with a discussion regarding the strengths and

•
weaknesses of the approaches discussed with an emphasis
Linear (or nearly linear) system dynamics and obser- on helping navigation engineers decide which estimation
vations.

•
algorithm to apply to a given problem of interest.
All noise and error sources are Gaussian.
36.1.1 Notation
While these assumptions are valid in many cases, there is
increasing interest in incorporating sensors and systems The following notation is used in this chapter:

•
that are non-Gaussian, nonlinear, or both. Because these
characteristics inherently violate the fundamental assump- State vector: The state vector at time k is represented by
the vector xk.

•
tions of the Kalman filter, when Kalman filters are used,
performance suffers. More specifically, this can result in fil- State estimate: An estimated quantity is represented
ter estimates that are inaccurate, inconsistent, or unstable. using the hat operator. For example, the estimated state
vector at time k is xk .

•
To address this limitation, researchers have developed a
number of algorithms designed to provide improved perfor- A priori/a posteriori estimates: A priori and a posteriori
mance for nonlinear and non-Gaussian problems [4–6]. estimates are represented using the + and – superscript
In this chapter, we provide an overview of some of the notation. For example, the a priori state estimate at time
k is xk− , and the a posteriori state estimate at time k is xk+ .

•
most common and useful classes of nonlinear recursive
estimators. The goal is to introduce the fundamental the- State error covariance estimates: The state error covari-
ories supporting the algorithms, identify their associated ance matrix is represented using the matrix P with super-
performance characteristics, and finally present their scripts and subscripts as required. For example, the a
respective applicability from a navigation perspective. priori state error covariance matrix at time k is given
by Pk− .

•
The chapter is organized as follows. First, an overview of
the notation and essential concepts related to estimation State transition matrix: The state transition matrix from
and probability theory are presented as a foundation for time k – 1 to k is given by Φkk − 1. Note that the time indices
nonlinear filtering development. Some of the concepts may be omitted when they are explained contextually.

Position, Navigation, and Timing Technologies in the 21st Century: Integrated Satellite Navigation, Sensor Systems, and Civil Applications, Volume 2,
First Edition. Edited by Y. T. Jade Morton, Frank van Diggelen, James J. Spilker Jr., and Bradford W. Parkinson.
© 2021 The Institute of Electrical and Electronics Engineers, Inc. Published 2021 by John Wiley & Sons, Inc.

c36.3d 1121 15/12/2020 10:54:42 PM

 1122 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

• Process noise vector and covariance: The process noise

vector at time k is wk. The process noise covariance
observation models. The process model represents the
internal dynamics of the system and can be expressed as
matrix at time k is Qk. a nonlinear, stochastic difference equation of the form

• Observation vector: The observation vector at time k is

given by zk.
xk = f xk − 1 , wk − 1 36 5

• Observation influence matrix: The observation influence

matrix at time k is given by Hk. Note that the time index
where xk is the state vector at time k N, and wk − 1 is the
process noise random vector at time k – 1. External obser-
may be omitted when contextually unnecessary.

•
vations regarding the system state are represented by an
Measurement noise vector and covariance: The measure- observation model. The generalized observation model is
ment noise vector at time k is represented by vk. The a function of both the system state and a random vector
measurement noise covariance is represented by Rk.

•
representing the observation errors:
Probability density function: Probability density func-
tions are expressed as p( ). zk = h x k , v k 36 6

In the above equation, zk is the observation at time k, and

vk is the random observation error vector at time k. The
objective of the recursive estimator is to estimate the poste-
36.2 Linear Estimation Foundations rior pdf of the state vector, conditioned on the observations
The goal of any estimator is to estimate one (or more) para- p x k Zk 36 7
meters of interest based on a model of the system, observa-
tions from sensors, or both. Because the parameters are, by where Zk is the collection of observations up to, and includ-
definition, random vectors, they can be completely charac- ing, time k. This is accomplished by performing two types of
terized by their associated probability density function transformations on the state pdf, propagation and updates.
(pdf ). If we define our parameter vector and observation The result is a filter cycle given by
vectors at time k as xk and zk, respectively, the overarching p x k − 1 Zk − 1 p x k Zk − 1 p x k Zk
propagate update
objective of a recursive estimator is to estimate the pdf of
all of the previous state vector epochs, conditioned on all 36 8
observations received up to the current epoch. Mathemat-
Note the introduction of the a priori pdf given by
ically, this is expressed as the following pdf:
p x k Zk − 1 36 9
p X k Zk 36 1
where Further examination of the propagation and update cycle
in Eq. 36.8 provides insights into how our system knowl-
Xk ≜ x0 , x1 , , xk 36 2 edge and observations are incorporated into our under-
and standing of the state vector. To begin, we consider the
propagation step from epoch k – 1 to k. Time propagation
Z k ≜ z 0 , z1 , , zk 36 3 begins with the posterior pdf p(xk − 1| Zk − 1). The process
model defined in Eq. 36.5 is used to define the transition pdf
While this is the most general case, it should be noted
p(xk| xk − 1), which can then be used to calculate the a priori
that most online algorithms would only be concerned with
pdf at time k via the Chapman–Kolmogorov equation [2]:
the conditional state estimate at the current epoch. For this
situation, Eq. 36.1 would be represented as
p x k Zk − 1 = p xk xk − 1 , Zk − 1 p xk − 1 Zk − 1 dxk − 1
p x k Zk 36 4
36 10
In the next section, we will present the typical recursive
estimation framework which will serve as the foundations Examination of the process model (Eq. 36.5) shows that
for developing the forthcoming nonlinear recursive estima- the propagated state vector is a first-order Gauss–Markov
tion strategies to follow. random process and is dependent only on the previous state
vector and the process noise vector. As a result, we can
36.2.1 Typical Recursive Estimation Framework express the transition probability, which is independent
of the prior observation, as
In a typical recursive estimation framework, the system
is represented using a process model and one (or more) p x k x k − 1 , Zk − 1 = p x k x k − 1 36 11

c36.3d 1122 15/12/2020 10:54:42 PM

 36.3 Nonlinear Filtering Concepts 1123

Substituting Eqs. 36.11 into 36.10 results in the propaga- where μ, Λ represents a Gaussian density with μ mean
tion relationship and Λ covariance. In addition, the plus and minus super-
scripts are used to express an a priori or a posteriori quan-
p x k Zk − 1 = p xk xk − 1 p xk − 1 Zk − 1 dxk − 1 tity, respectively. Substituting the linear process model
36 12 (Eq. 36.17) into our propagation relationship (Eq. 36.12)
results in the linear Kalman filter propagation equations
An observation at time k can be incorporated by consid-
ering the posterior pdf p(xk| Zk), which, given the definition xk− = Φkk − 1 xk+− 1 36 21
of our observation sequence in Eq. 36.3, can be expressed T
equivalently as Pk− = Φkk − 1 Pk+− 1 Φkk − 1 + Qk − 1 36 22

p x k Zk = p x k z k , Zk − 1 36 13 Furthermore, substituting the linear observation model

(Eq. 36.18) into our update relationship (Eq. 36.16) results
Applying Bayes’ rule to Eq. 36.13 yields
in the linear Kalman filter update equations:
p zk x k , Z k − 1 p x k Z k − 1
p x k Zk = 36 14 xk+ = xk− + Kk zk − Hk xk− 36 23
p z k Zk − 1
Observing the form of the previously defined observation, Pk+ = Pk− − Kk Hk Pk− 36 24
Eq. 36.6 shows that zk is independent of Zk − 1, and thus
where zk is the realized measurement observation‚ and Kk
Eq. 36.14 can be simplified to
is the Kalman gain at time k:
p zk x k p x k Z k − 1
p x k Zk = 36 15 Kk = Pk− HTk Sk− 1 36 25
p zk Z k − 1
and Sk is the residual covariance matrix, given by
As a final note, we observe that the normalizing term in
the denominator, known as the evidence, can be expressed Sk = HPk− HT + Rk 36 26
in a more directly obvious form by de-marginalizing about
In many cases, systems can be accurately represented by
the state vector as follows:
linear Gaussian models. Unfortunately, there are a number
p zk x k p x k Z k − 1 of systems where these models are not adequate. This moti-
p x k Zk = 36 16
p zk xk p xk Zk − 1 dxk vates the development of various algorithms that attempt to
solve these equations for various classes of problems.
Thus, we have presented the mathematical form of both
In the next section, we will present the fundamental con-
the propagation (Eq. 36.12) and update (Eq. 36.16) actions
cepts which will be used to derive various recursive nonlin-
on the pdf representing the state random vector.
ear estimators.
For a specific class of problems (e.g. linear Gaussian sys-
tems), the above equations can be solved in closed form. In
this case, the generalized process model (Eq. 36.5) simpli-
fies to
36.3 Nonlinear Filtering Concepts
xk = Φkk − 1 xk − 1 + w k − 1 36 17 In the previous sections (Section 36.2.1), we have developed
the generalized theory for recursive estimation problems.
where Φkk − 1 is the state transition matrix from time k − 1 to
The theory is based on the fundamental need to determine
k, and wk − 1 is a zero-mean, white Gaussian sequence the pdf of the state vector at an epoch of interest, condi-
with covariance Qk. Similarly, the generalized observation tioned on the observations up to, and including, the current
model 36.6 simplifies to epoch. Complete knowledge of the conditional state pdf
zk = Hk xk + v k 36 18 represents maximum possible knowledge of the system.
This is, in fact, the normal state of affairs for Gaussian sys-
where Hk is the observation influence matrix at time k, and
tems, as the pdf can be completely described by a mean and
vk is a zero-mean, white Gaussian sequence with covari-
covariance.
ance Rk.
Thus, both the a priori and posterior pdfs can be repre-
36.3.1 Effects of Nonlinear Operations on
sented as the following Gaussian densities, respectively: Random Processes – Breaking Up with Gauss
p x k Zk − 1 ≜ xk− , Pk− 36 19 Consider a Gaussian random vector x with mean and
p x k Zk ≜ xk+ , Pk+ 36 20 covariance x and Px, respectively. As mentioned previously,

c36.3d 1123 15/12/2020 10:54:43 PM

 1124 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

for Gaussian densities, these two parameters are sufficient 1.8

to completely describe the full pdf of the random vector. 1.6
Next consider a linear transformation from x to y which 1.4 p(x1) p(x2)
is governed by the transformation matrix H. The resulting
1.2
equation for y is given by
1

p(.)
y = Hx 36 27 p(x3)
0.8 p(xsum)
In this case, the transformed random vector, y, can be 0.6
shown to be a Gaussian random vector with mean and 0.4
covariance 0.2

y = Hx 36 28 0
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 2.5
T
Py = HPx H 36 29 X

This preservation of Gaussian nature when transformed Figure 36.1 Gaussian sum illustration. The random variable xsum
is represented by a weighted sum of three individual Gaussian
via linear operations is an important property of Gaussian
densities. In this example, xsum = 0.25x1 + 0.5x2 + 0.25x3.
densities that makes the linear Kalman filter relatively sim-
ple to implement.
Now consider a generalized nonlinear transformation to address the situation where unknown or uncertain para-
meters exist within the system model. Some examples of
y=h x 36 30
these types of situations include modeling discrete failure
In this case, the density of y can become difficult to cal- modes, unknown structural parameters, or processes
culate exactly. While we will address this issue in more with multiple discrete modes of operation (e.g. “jump”
detail later in the chapter, generally speaking, the resulting processes).
density function is clearly non-Gaussian, thus limiting Consider our standard linear Gaussian process and
the performance of the linear Kalman filter algorithm. observation models, repeated from Eqs. 36.17 and 36.18
Nonlinear estimators attempt to maintain a higher-fidelity for clarity:
estimate of the overall density function as it transforms
over time. xk = Φkk − 1 xk − 1 + w k − 1 36 32
In the next section, we present our first class of estimators
designed to support systems with non-Gaussian pdfs. zk = Hk xk + v k 36 33

In the previous development, it was assumed that the

36.3.2 Gaussian Sum Filters
system model parameters (i.e. Φkk − 1 , Qk − 1 , Hk , Rk) were
One approach for modeling systems with non-Gaussian known. Let us now consider the situation where some of
pdfs is the use of composite random variables expressed the system model parameters are unknown.
as a sum of Gaussian random variables. The generalized To address this situation, we can define a vector of the
Gaussian sum can be expressed as unknown system parameters, a, and jointly estimate these
parameters along with the state vector. In other words, we
J
x= w j yj 36 31 must now solve for the following density:
j=1
p xk , a Zk 36 34
where w[j] is a scalar weighting factor, yj is a Gaussian
random variable with mean y j , and covariance Py j . These which, after applying Bayes’ rule, can be expressed as
individually weighted Gaussian random variables can rep-
p xk , a Zk = p xk a, Zk p a Zk 36 35
resent the overall distribution of the state vector. An exam-
ple of a density function created using a sum of Gaussian It is important to note that this expression is the product
random variables is shown in Figure 36.1. of the “known-system model” pdf, p(xk| a, Zk), and a new
density function, p(a| Zk), which is the pdf of the unknown
36.3.2.1 Multiple Model Adaptive Estimation system parameters, conditioned on the observation set.
One implementation of the Gaussian sum filtering Assuming a Rn, the parameter density can be written as
approach is known as multiple model adaptive estimation
(MMAE). The MMAE filter uses a weighted Gaussian sum p a Z k = p a zk , Z k − 1 36 36

c36.3d 1124 15/12/2020 10:54:43 PM

 36.3 Nonlinear Filtering Concepts 1125

Applying Bayes’ rule yields At this point, we have established the posterior pdf of
the parameter vector as a finite weighted set. Revisiting
p zk a, Zk − 1 p a Zk − 1
p a Zk = 36 37 our system parameter pdf, now defined at time k
p zk Z k − 1
J
Marginalizing the denominator about the parameter vec- p a Zk =
j
wk δ a − a j
36 44
tor results in a more familiar form: j=1

p zk a, Zk − 1 p a Zk − 1
p a Zk = 36 38 and substituting into Eq. 36.43 yields the parameter density
p zk a, Zk − 1 p a Zk − 1 da update relationship
where p(zk| a, Zk−1) is the measurement prediction density, J
j
which, given our linear observation model, is expressed as wk δ a − a j
j=1
the following normal distribution:

p zk a, Zk − 1 = Hk xk− , Sk 36 39 J p zk a j , Z k − 1 j
= J
wk − 1 δ a − a j
Unfortunately, the integral in the denominator is intrac- j=1 n
wk − 1 p zk an , Zk − 1
table in general, which requires an additional constraint. n=1
If the system parameters can be chosen from a finite set 36 45
(e.g. a {a[1], a[2], , a[j]}), the parameter density can be
expressed as the sum of the individual probabilities of the In the above equation, the predicted measurement pdf,
finite set. This results in a system parameter pdf defined as p(zk| a[j], Zk − 1), is evaluated at the measurement realiza-
tion at time k, which yields the likelihood of realizing
J
p a Zk − 1 =
j
wk − 1 δ a − a j
36 40 the current measurement, conditioned on the parameter
j=1 set j. As mentioned previously, these likelihood values
j
are based on the following evaluation of a normal density
where wk − 1 is the probability of the j-th parameter vector at function:
time k-1, and δ( ) is the delta function. It can be observed
− j j
that the sum of the weights must be unity in order to rep- p zk = zk a j , Zk − 1 ≔ zk ; Hk x^ k , Sk
resent a probability density. Substituting Eq. 36.40 into
Eq. 36.38: 36 46

J
j where zk is the measurement realization at time k.
p zk a, Zk − 1 wk − 1 δ a − a j
This likelihood is equivalent to the likelihood of the
j=1
p a Zk = J residual from a Kalman filter tuned to the j-th parameter
n
p zk a, Zk − 1 wk − 1 δ a − a n da vector, a[j].
n=1
Practically speaking, the parameter pdf consists of the
36 41
discrete (fixed) parameter set and the associated weights
Moving the position of the summation operators and (likelihood) at each epoch. The parameter density update
parameter weight vector: shown in Eq. 36.45 shows the evolution of each parameter
J weight as a function of time, which can be rewritten as
j
wk − 1 p zk a, Zk − 1 δ a − a j

p a Zk =
j=1
j p zk = zk a j , Zk − 1 j
J wk = J
wk − 1 j
n
wk − 1 p zk a, Zk − 1 δ a − a n da n
wk − 1 p zk = zk a , Zk − 1
n
n=1
n=1
36 42 1, 2, ,J
The properties of the delta function can be exploited 36 47
to rewrite the numerator and eliminate the integral from
the denominator: Our final task is to determine the overall posterior joint
pdf of the system. Substituting Eq. 36.44 into Eq. 36.35,
we obtain
J
p zk a j , Z k − 1 j
p a Zk = J
wk − 1 δ a − a j J
j
j=1 n
wk − 1 p zk an , Zk − 1 p xk , a Zk = p xk a, Zk wk δ a − a j

n=1 j=1

36 43 36 48

c36.3d 1125 15/12/2020 10:54:43 PM

 1126 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

which, when combined with knowledge of the delta func- 36.3.3 MMAE Example – Integer Ambiguity
tion and implementing a straightforward rearrangement Resolution
of terms produces the joint posterior density function
The benefits of the Gaussian sum filter can be illustrated
J using a simple example. Consider the following one-
j
p xk , a Zk = wk p xk a j , Zk δ a − a j dimensional navigation scenario. A radio transmitter broad-
j=1 casts a ranging signal from a fixed location, xt. A ranging
36 49 receiver is mounted on a vehicle that is free to move in
the x-direction. The vehicle motion can be represented using
This pdf is clearly a weighted sum of Gaussian densities, a first-order Gauss–Markov velocity model [2] with uncer-
each of these densities corresponding to the posterior state tainty σ v and time constant τv. The resulting state vector is
estimate of an individual Kalman filter, tuned to the param- given by
eter vector a[j]. The blended posterior state estimate and
pk
covariance are given by xk = 36 52
vk
J
j + j where pk and vk are the position and velocity of the vehicle
xk+ = wk xk 36 50
j=1
at time k. The dynamics of the vehicle are given by

J xk + 1 = Φkk + 1 xk + w k 36 53
T
j + j + j + j
Pk+ = wk xk − xk+ xk − xk+ + Pk
where
j=1

36 51 1 Δt
Φkk + 1 = 36 54
0 exp − Δt τv
The MMAE filter can be visualized in block diagram form
in Figure 36.2. and wk is a zero-mean Gaussian random vector with
Additional forms that are very similar conceptually to the 0 0
MMAE filter are known as interactive mixture model E w j w Tk = δjk
0 σ v 1 − exp − 2Δt τv
2
(IMM) estimators [8] and Rao-Blackwellized particle filters
36 55
(RB-PFs) [9, 10], to name a few.
In the next section, we present a simple example to The ranging signal consists of both a noise-corrupted
illustrate a potential application for Gaussian sum filters measurement of the true range along with a measurement
derived in this section. of the integrated carrier phase. The integrated carrier phase

+[1]
xˆ k
Kalman Filter X
[1]
Based on a[1] rk
[1]
wk
+[2]
xˆ k xˆ k+
Kalman Filter X +
[2]
Based on a[2] rk
Zk
[2]
wk

+[J]
xˆ k
Kalman Filter X
[J]
Based on a[J] rk
[J]
wk

Conditional probability
computation

Figure 36.2 MMAE filter implementation. The MMAE filter constructs the state estimate by combining results from individual Kalman
filters tuned to a parameter realization [7].

c36.3d 1126 15/12/2020 10:54:43 PM

 36.3 Nonlinear Filtering Concepts 1127

is a high-precision measurement, but is corrupted by an Table 36.1 Simulation parameters

unknown integer ambiguity. The observation model is
Parameter Value Units
ρk = x k − x t + v ρ k 36 56
ϕk = λ − 1 x k − x t + N + v ϕ k 36 57 σρ 0.5 m
σϕ 0.1 cycles
where λ is the carrier wavelength, and N is the integer ambi-
λ 0.2 m
guity. Both observations are corrupted by zero-mean white
σv 0.2 m/s
Gaussian noise sequences with
τv 500 s
E v ρ j v ρk = σ 2ρ δjk 36 58 xt 0 m
Δt 1.0 s
E v ϕ j v ϕk = σ 2ϕ δjk 36 59

E v ρ j v ϕk = 0 36 60
The resulting trajectory, range observations, and phase
Our goal is to use the MMAE estimator to accurately rep- observations are shown in Figure 36.3.
resent the (non-Gaussian) posterior pdf, thus maintaining a The MMAE global state estimate and density function
consistent overall state estimate and uncertainty, while of position after one observation (t = 1 s) are shown in
incorporating all available information. Figure 36.4. The probability density function is clearly
In this example, the integer ambiguity is the unknown multi-modal, which accurately represents the range of solu-
parameter set, which in the previous development we tions associated with the phase observation. As expected,
designated as the vector a. We choose a range of J plausible the peaks are located at integer multiples of the carrier
integers based upon any a priori knowledge or even the ini- wavelength which corresponds to the most likely values
tial range observation itself, which results in the following of the unknown integer ambiguity. These peaks indirectly
unknown parameter vector: indicate the relative likelihood of the associated ambiguity
being correct by exhibiting influence on the overall position
a = N 1 ,N 2 , ,N J 36 61
density.
with overall joint probability density After 22 cycles, the position density shows a reduced
J
number of peaks (see Figure 36.5). This indicates that the
j filter is incorporating sensor observations and the statistical
p x k , a Zk = w k p x k a j , Zk δ a − a j

j=1 dynamics model to effectively eliminate a number of poten-

36 62 tial ambiguity possibilities.
After 100 cycles (Figure 36.6), the filter has converged to a
From this point forward, the implementation proceeds as
single ambiguity.
outlined in the previous section. A total of J weighted Kal-
The global state estimate and associated standard devia-
man filters are constructed, each with the assumption that
tion result for this simulation are shown in Figure 36.7.
N[j] is the correct integer ambiguity. The joint posterior den-
The shape of the uncertainty bound clearly shows the
sity is given by
effects described above. As the likelihood of each integer
J
j ambiguity realization changes, the overall uncertainty
p x k , a Zk = w k p x k a j , Zk δ a − a j
changes and eventually collapses to the centimeter level.
j=1
Finally, the associated normalized filter weights for a
36 63
subset of the integer ambiguity realizations are shown in
In order to demonstrate the performance of the Gaussian Figure 36.8. As expected, the highly unlikely edge integers
sum filter, the above scenario was implemented in a simu- quickly collapse. The integers closer to the mean take
lation environment. A trajectory and measurement set is longer to resolve. It is important to note that the resulting
randomly generated using the parameters specified in uncertainty is dependent on the actual measurement reali-
Table 36.1. zation sequence received; thus‚ each realization would pro-
Note the carrier phase wavelength is 0.2 m, and the car- duce a different uncertainty (Table 36.2). This is a notable
rier phase measurement uncertainty is 0.1 cycles, which difference from the standard linear Kalman filter‚ where
results in a measurement precision of 0.02 m, which is the uncertainty is independent of the observed measure-
an improvement of 50 times over the pseudorange meas- ments. Finally, it is important to note that, in this example,
urement errors. the state estimate and uncertainty of the MMAE filter are

c36.3d 1127 15/12/2020 10:54:44 PM

 1128 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

Table 36.2 Summary of filter classes

Linear and extended Kalman filter

•• • •
Strengths Weaknesses Use case
Optimal for linear Gaussian systems Suboptimal approximation for Linear, or close-to-linear,
Computationally simple nonlinear systems, can be prone to Gaussian problems
divergence
Gaussian sum filter

• • •
Strengths Weaknesses Use case
Optimal for linear Gaussian systems with discrete If parameter vector is not discrete, the Linear, or close-to-linear,

•
parameter vector differences must be observable Gaussian problems with discrete
Conservative tuning can mask parameters
difference between models and reduce

•
performance
Increased computation requirements
over simple Kalman filter
Grid particle filter

• • •
Strengths Weaknesses Use case
Optimal solution when state space consists of Computational requirements can be Nonlinear problems with lower

• •
discrete elements excessive dimensionality
Suitable for wide range of nonlinear conditions Processing requirements scale
geometrically with the number of

•
dimensions
Discretizing continuous state space
results in suboptimal performance
Sampling particle filter

• • •
Strengths Weaknesses Use case
Can produce nearly optimal solution for Maintaining good particle distribution Nonlinear problems with higher

• ••
nonlinear problems can be difficult dimensionality
Computational requirements can be reduced over Lack of repeatability from run to run
a grid particle filter via importance sampling Computational requirements can still
strategies be large

Sample trajectory
5
True range 0.5
Range meas MMAE Mean
Phase meas
0.45 p(x|Z)
4
True
0.4

3 0.35

0.3
px (m)

2
p(x)

0.25

1 0.2

0.15
0 0.1

0.05
–1
0 10 20 30 40 50 60 70 80 90 100 0
Time (s) –1.5 –1 –0.5 0 0.5 1 1.5 2
X (m)
Figure 36.3 Sample vehicle trajectory and observations. Note
that the range observations are accurate but not precise and the Figure 36.4 MMAE initial state estimate and position density
phase observations are precise but not accurate. Our goal is to function. Note the position density function is extremely multi-
accurately estimate the joint pdf of this system. modal due to the limited information available at this point.

c36.3d 1128 15/12/2020 10:54:44 PM

 36.3 Nonlinear Filtering Concepts 1129

2 1
MMAE Mean Xerr
1.8 p(x|Z) 0.8 σx
True
1.6 0.6

1.4 0.4

Position Error (m)

1.2 0.2

0
p(x)

0.8 –0.2

–0.4
0.6
–0.6
0.4
–0.8
0.2
–1
0 0 10 20 30 40 50 60 70 80 90 100
–1 –0.5 0 0.5 1 1.5 2 2.5
Time (s)
X (m)
Figure 36.7 MMAE position error and one-sigma uncertainty.
Figure 36.5 MMAE state estimate (after 22 observations). Range
Note that the error uncertainty collapses once sufficient
observations combined with the vehicle dynamics model are
information is available to resolve the integer ambiguity.
eliminating unlikely integer ambiguity values.

3.5 accomplish this by representing density functions by using

MMAE Mean collections of discrete, weighted state vectors instances.
p(x|Z)
3 True These state vectors and associated weights are referred to
as particles.
2.5
The development of the theory related to a particle-based
representation of density functions begins by reviewing the
2
essential properties of both the probability density function
p(x)

and the cumulative distribution function. An example cdf

1.5
and pdf are shown in Figure 36.9. The cumulative distribu-
1 tion function is a monotonically increasing function which
represents the probability of a random variable realization
0.5 that is less than the operand and can be defined as the inte-
gral of the density function [11]:
0
0.5 1 1.5 2 2.5 3 3.5 4 Pr x < x a = F x a 36 64
X (m) xa
= p x dx 36 65
Figure 36.6 MMAE state estimate (after 100 observations). Note −∞
the state estimate is almost completely unimodal and has
converged to the correct integer ambiguity. Additionally, the probability of a random variable reali-
zation between a range xa and xb is expressed by

truly optimal (i.e. minimum mean square error). This Pr x a ≤ x < x b = F x b − F x a 36 66

would not be the case if the integer ambiguity were resolved xb
using a more traditional approach (e.g. float estimate with = p x dx 36 67
xa
an ad hoc fixing stage). This is an interesting property of the
Gaussian sum filter and sets the stage for us to investigate As a result, the density and cumulative distribution
additional nonlinear estimation techniques. functions must have the following properties:

F −∞ =0 36 68
36.3.4 Particle Filters
F + ∞ =1 36 69
As mentioned in Section 36.3, the key requirement of
a nonlinear filter is the ability to accurately represent ∞

arbitrary probability density functions. Particle filters f x dx = 1 36 70

−∞

c36.3d 1129 15/12/2020 10:54:46 PM

 1130 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

MMAE Integer Weights

1
N=3

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N=4

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N=5

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N=6

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N=7

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N=8

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N=9

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N = 10

0.5
0
0 10 20 30 40 50 60 70 80 90
1
N = 11

0.5
0
0 10 20 30 40 50 60 70 80 90
Time (s)

Figure 36.8 MMAE integer ambiguity particle weights (subset). The correct ambiguity particle (N=7) likelihood increases over time
while the outliers are determined to be less likely.

Probability density function Cumulative Density Function

1.4 1

0.9
1.2
0.8

1 0.7

0.6
0.8
F(x)
p(x)

0.5

0.6
0.4

0.4 0.3

0.2
0.2
0.1

0 0
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 –2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x x

Figure 36.9 Probability density function (PDF) and cumulative density function example (CDF).

c36.3d 1130 15/12/2020 10:54:48 PM

 36.3 Nonlinear Filtering Concepts 1131

f x ≥0 36 71 Based on this definition, we can easily calculate some

common expectations of the weighted particle pdf. The first
The particle filter uses a collection of weighted delta is the mean, which is defined as E[x]:
functions to represent the pdf:
+ ∞
J Ex = xp x dx 36 75
−∞
px ≈ w j δ x−x j
36 72
j=1
Substituting the pdf of x from Eq. 36.72, rearranging
the summation and integral, and then applying the sifting
where w[j] is a scalar weighting value for the j-th particle
property:
with location x[j]. As mentioned previously, the sum of
weights must be unity: + ∞ J
Ex = x w j δ x−x j
dx 36 76
N −∞ j=1
j
w =1 36 73
J + ∞
j=1
= w j
xδ x − x j
dx 36 77
j=1 −∞
An example pdf represented by a collection of weighted
particles is shown in Figure 36.10. This importance J

sampling strategy allows us to represent any pdf with a = w jx j

36 78
j=1
desired level of fidelity, given enough particles. Additional
details regarding importance sampling are provided in This shows that the mean of a weighted particle random
Section 36.3.7. variable can be calculated as the weighted sum of particles.
In addition to representing arbitrary pdfs of random vec- The above development can be applied identically to the
tors, successful nonlinear estimation requires the ability to general expectation function case with the following result:
determine the resulting pdfs after applying nonlinear trans-
formations to random vectors. In general, this can be intrac- J

table; however, representing the pdf using the collection of Egx = w jg x j

36 79
j=1
weighted particles makes the transformation relatively
straightforward. An example of the effect of some sample This can easily be extended to represent a set of sufficient
nonlinear transformations is shown in Figure 36.11. statistics for an arbitrary density function. As a result, it can
One of the most common functions necessary for filtering be shown that any density function can be represented to
applications is calculation using the expectation operator. arbitrary accuracy, given enough particles. Because we
The expectation operator is defined as seek estimation methods that are computationally feasible,
+ ∞ we are searching for methods that give us “good enough”
Egx = g x p x dx 36 74 performance (e.g. accuracy and stability) with limited
−∞
computational resources.
where E[ ] is the expectation operator, g(x) is an arbitrary In the next section, we investigate one approach, known
function of the random vector x, and p(x) is the pdf of as the grid particle filter, to representing the location of our
the random vector. particle collection.

CONSTANT INTERVAL VARIABLE INTERVAL VARIABLE INTERVAL

VARIABLE WEIGHT CONSTANT WEIGHT VARIABLE WEIGHT

Figure 36.10 Importance sampling used to represent arbitrary density functions. The density function is represented by a combination of
particle locations and weights (represented by arrows), which can be varied independently.

c36.3d 1131 15/12/2020 10:54:50 PM

 1132 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

Given: x,y : u (–1,1), u (–1,1)

2
1.5
1
0.5
0

y
–0.5
–1
–1.5
–2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x
x′ = x 4 x′ = | x |
y′ = y y′ = y
x′ = sin x
y′ = y
2 2 2
1.5 1.5 1.5
1 1 1
0.5 0.5 0.5
0 0 0
y

y
–0.5 –0.5 –0.5
–1 –1 –1
–1.5 –1.5 –1.5
–2 –2 –2
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2 –2 –1.5 –1 –0.5 0 0.5 1 1.5 2 –2 –1.5 –1 –0.5 0 0.5 1 1.5 2
x x x

Figure 36.11 Visualization of nonlinear transformation on a random variable. Given uniform random variables x, y, the effects of three
nonlinear transformations show that the density can change significantly during transformation.

36.3.5 Grid Particle Filtering collection of propagation and update steps are applied. At
this point, it is relatively straightforward to derive the prop-
One approach to addressing the generalized nonlinear esti-
agation and update relations for the collection of particles.
mation requirement to maintain the full probability density
We begin with the propagation from time k – 1 to k.
is the so-called grid particle filter. The grid particle filter
Assume that the posterior density function at time k – 1
maintains a discrete collection of possible system states
is given by
and associates a probability with each of those states (i.e.
particles). This approach is optimal given systems with J
j
the following conditions: p x k − 1 Zk − 1 = wk − 1 k − 1 δ x k − 1 − x j

j=1

1) The state vector is truly discrete or can be accurately 36 81

approximated using a discretization of the state space.
Substituting Eq. 36.81 into the Chapman–Kolmogorov
2) The number of discrete states is computationally
equation (Eq. 36.12) and simplifying:
tractable.

Given these conditions, the state density function can be p x k Zk − 1 = p xk xk − 1 p xk − 1 Zk − 1 dxk − 1

expressed as a weighted collection of particles (repeated 36 82
from Eq. 36.72)
J
j
J = p xk xk − 1 wk − 1 k − 1 δ x k − 1 − x j
dxk − 1
px = w j δ x−x j
36 80 j=1
j=1
36 83
where the particle weights, w[j], must sum to one. Because J
j
the particle locations are assumed to be static, the filtering = wk − 1 k − 1 p xk xk − 1 δ xk − 1 − x j
dxk − 1
operation is performed over the collection of weights. This j=1

allows the filter to maintain the density function as the 36 84

c36.3d 1132 15/12/2020 10:54:51 PM

 36.3 Nonlinear Filtering Concepts 1133

J J
j
=
j
wk − 1 k − 1 p xk x j
36 85 wk k − 1 p zk x k δ x k − x j
j=1
j=1 = 36 94
J
l
The rightmost density in Eq. 36.85 is the transition prob- wk k − 1 p zk xk δ xk − x l dxk
l=1
ability function‚ which can be rewritten as
J J
j
p xk x j
=
l j
p x k x k − 1 δ xk − x l 36 86 wk k − 1 p zk x j
δ xk − x j
j=1
l=1 = J
36 95
l
Substituting Eq. 36.86 into Eq. 36.85 and simplifying wk k − 1 p zk x l
l=1
J J
j
p x k Zk − 1 = wk − 1 k − 1 p xl x j
δ xk − x l Finally, recalling the grid particle filter form of the pos-
j=1 l=1
terior density function
36 87
J
J J j
j p x k Zk = wk k δ xk − x j
36 96
= wk − 1 k − 1 p x x l j
δ xk − x l
j=1
l=1 j=1

36 88 and substituting into Eq. 36.95 yields

J
l J
= wk k − 1 δ x k − x l 36 89 j
wk k δ xk − x j
l=1 j=1

where the new particle weight is given by j 36 97

j
J w k k − 1 p zk x
J = J
δ xk − x j
l j
wk k − 1 = wk − 1 k − 1 p x l x j
36 90 j=1 l
wk k − 1 p zk x l
j=1 l=1

Conceptually, this can be calculated as the sum of all The bracketed areas show the final particle weight update
posterior weights at time k – 1 multiplied by the specific tran- equations
sition probability into state l from all possible prior states. j j
The development of the measurement update function j
wk k − 1 p zk x
wk k = J
36 98
proceeds in a similar fashion. Recalling our definition of l
wk k − 1 p zk x l
the a priori density function (repeated from Eq. 36.89 for l=1
clarity):
In the next section, we illustrate a potential application of
J
j
the grid particle filter in a navigation context.
p x k Zk − 1 = wk k − 1 δ xk − x j
36 91
j=1

36.3.6 Grid Particle Filter Example Application

and substituting into the update equation (Eq. 36.16) yields
J
We return to the example presented in Section 36.3.3; how-
j
p zk x k wk k − 1 δ x k − x j
ever, in this case, we utilize a grid particle filter solution.
j=1
p x k Zk = J
The first step in the process is to determine the composition
j of the grid. In this case, there are two parameters we would
p zk x k wk k − 1 δ xk − x j dxk
j=1 like to estimate, position and velocity. Both of the para-
36 92 meters are continuous random variables, so we must quan-
tize both of the parameters.
which can be simplified by changing the order of integra-
For this example, we are interested in centimeter-level
tion and using the properties of the delta function
positioning accuracy; thus, we divide the domain into
J 5 mm by 20 mm/s grids. For simplicity, we build a grid that
j
p zk x k wk k − 1 δ xk − x j
is ±2 m in range and ±0.6 m/s in velocity. The absolute
j=1
p x k Zk = J
grid location is periodically adjusted based on the current
l
p zk x k wk k − 1 δ xk − x l dxk estimated position and velocity of the vehicle.
l=1 An identical randomly generated trajectory and measure-
36 93 ment set from the MMAE example (Section 36.3.3) is used

c36.3d 1133 15/12/2020 10:54:52 PM

 1134 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

0.08
0.3
0.06
0.2
0.04
0.1
0.02

Velocity (m/s)
Velocity (m/s)

0
0
–0.1 –0.02
–0.2 –0.04
–0.3 –0.06
–0.4 –0.08
–0.5 –0.1
–0.6 2.1 2.15 2.2 2.25 2.3 2.35 2.4 2.45
–0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1
Position (m) Position (m)

Figure 36.12 Grid particle filter state estimate and position Figure 36.14 Grid particle filter state estimate (after 100
density function after one observation. Note the density observations). Note that the state estimate is almost completely
function is extremely multi-modal due to the limited information unimodal and has converged to the correct integer ambiguity.
available at this point.

wavelength and represent the most likely values of integer

as the inputs to the grid particle filter. For reference, the ambiguity. These peaks indirectly indicate the relative
system parameters are specified in Table 36.1, and the likelihood of the associated ambiguity being correct by
resulting trajectory, range observations, and phase observa- exhibiting influence on the overall position density. In each
tions are shown in Figure 36.3. plot below, the calculated mean is represented by a white
The grid particle filter global state estimate and density “plus,” the true state is represented by a green asterisk,
function of position after one observation (t = 1 s) are and the calculated 2-sigma uncertainty is represented as
shown in Figure 36.12. In this case, we present the proba- a white ellipse.
bility density function using a two-dimensional array After 22 cycles, the density shows a reduced number of
(position vs. velocity) of probabilities. The resulting pdf peaks (see Figure 36.13). This indicates that the filter is
is clearly multi-modal, which accurately represents the incorporating sensor observations and the statistical dyna-
range of solutions associated with the phase observation. mics model to effectively eliminate a number of potential
As expected, the peaks are located as a function of the ambiguity possibilities.
After 100 cycles (Figure 36.14), the filter has converged
to a single ambiguity.
0.2 The global state estimate and associated standard devia-
tion result for this simulation are shown in Figure 36.15.
0.15
The shape of the uncertainty bound clearly shows the
0.1 effects described above. As the likelihood of each integer
0.05 ambiguity realization changes, the overall uncertainty
Velocity (m/s)

changes and eventually collapses to the centimeter level.

0
In the next section, we will move to our final nonlinear
–0.05 filter algorithm, the sampling particle filter.
–0.1

–0.15 36.3.7 Sampling Particle Filter (SIS/SIR)

–0.2 In a similar manner to the grid particle filter, the sampling
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 particle filter, also known as a the sequential Monte Carlo
Position (m) (SMC) filter, represents the state density function using
Figure 36.13 Grid particle filter state estimate (after
a weighted collection of particles. However, we seek to
22 observations). Range observations combined with the vehicle address the computational scaling problems inherent in
dynamics model are eliminating unlikely integer ambiguity values. grid-based approaches by exploiting an approach that

c36.3d 1134 15/12/2020 10:54:52 PM

 36.3 Nonlinear Filtering Concepts 1135

1 which approaches equality as

Xerr
0.8 σx
lim N ∞ IN =I 36 102
0.6

0.4 Now consider the case where the function in the inte-
Position Error (m)

grand, g(x), can be expressed as the product

0.2

0 g x =f x p x 36 103
–0.2
where p(x) is a probability density function; thus‚ p(x) ≥ 0
–0.4 and p(x)dx = 1. If N independent samples, x[i], can be
–0.6 drawn in accordance with p( ), then the integral can be esti-
–0.8 mated as the sample mean of the transformed particles:
–1
0 10 20 30 40 50 60 70 80 90 100 1 N
IN = f xi 36 104
Time (s) N i=1
Figure 36.15 Grid particle filter position error and one-sigma The resulting error in the estimate is unbiased and, most
uncertainty. Note that the error uncertainty collapses once
importantly, scales as the reciprocal of the square root of N.
sufficient information is available to resolve the integer ambiguity.
This is an important result as it indicates that the error is
independent of the dimensionality of the state, as long as
focuses computation on the regions of the state space with the particles are properly sampled from the distribution
the highest likelihood. This is accomplished by randomly of x. This is an important distinction from the grid filter‚
sampling the state space. which requires particles that increase geometrically with
The main advantage of this approach is the potential to the number of dimensions in the state vector [6].
more completely sample the important areas of the state Unfortunately, it is not always possible to generate
space, while limiting the total number of particles required. samples from arbitrary density functions. This motivates
This is a useful advantage over the grid particle filter, additional development of the concept known as impor-
which can require unreasonable numbers of particles as tance sampling.
the state dimensionality and domain increase. While sam- To further our discussion of importance sampling, it is
pling particle filtering approaches are suboptimal, their convenient to introduce the concept of a proposal density,
computational advantages make them attractive for a chosen to resemble (and provide support over) the true
larger range of applications. density of x, while retaining the ability to generate
We begin by describing the concept of Monte Carlo inte- samples. An illustration of a proposal-density sampling
gration, which is subsequently used to develop a basic approach is shown in Figure 36.16.
recursive estimation algorithm. Given a random vector with true density p(x) and parti-
The fundamental enabling concept for the sampling par- cles sampled from a proposal density, q(x), Eq. 36.103 can
ticle filter is the concept of Monte Carlo integration. Given be rewritten as
an integral in the following form: px
g x =f x qx 36 105
qx
I= g x dx 36 99
Ω

where Ω is an nx-dimensional region in Rnx with volume π (x) Desired density

q (x) Proposal density
V= dx 36 100
Ω q (x)
~ [ j] = π (x[ j])
If N independent samples are uniformly drawn from Ω, w
q (x[ j]) π (x)
that is, {x[1], x[2], , x[N]} Ω, then the integral can be
approximated as

1 N Figure 36.16 Proposal sampling illustration. In this example,

I ≈ IN = V g xi 36 101 the particles are generated using the proposal density (q) and
N i=1 subsequently weighted to represent the desired density (π).

c36.3d 1135 15/12/2020 10:54:54 PM

 1136 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

The resulting estimate of the integral, assuming N Assuming we begin with a known posterior density, p
independent particles sampled from q( ), is given by (xk − 1| Zk − 1). If N samples are drawn from an associated
proposal density,
N
px
I= f x q x dx 36 106 i
i=1 qx x k − 1 k − 1 q x k − 1 Zk − 1 i 1, …, N 36 114

1 N p xi With normalized weights given by

≈ f xi 36 107
N i=1 q xi
i
p x k − 1 Zk − 1
i
wk − 1 k − 1 =κ 36 115
where the ratio between the true density and the proposal i
q x k − 1 Zk − 1
density can be expressed as particle importance weights:
where κ is the normalization factor required such that the
p xi
wi = 36 108 sum of weights is unity, the posterior density function is
q xi
expressed by the collection of particles and weights
Substituting Eq. 36.108 into Eq. 36.107 yields N
i i
N
p x k − 1 Zk − 1 = wk − 1 k − 1 δ x − xk − 1 k − 1
1
IN = g xi wi 36 109 i=1
N i=1 36 116

Finally, the collection of particle weights can be normal- or, equivalently

ized via
p x k − 1 Zk − 1 x i , w i ; i = 1, …, N
i k−1 k−1
w
wi = N
36 110 36 117
wi
i=1 Our goal is to estimate the posterior density at time k, p
(xk| Zk), by incorporating the statistical process model and
then Eq. 36.109 becomes the observation at time k. The density function of interest
N can be written as
IN = g xi wi 36 111
p zk x k p x k x k − 1
i=1 p x k Zk = p x k − 1 Zk − 1
p zk Z k − 1
which we will exploit to develop a recursive estimator. 36 118
p zk x k p x k x k − 1 p x k − 1 Z k − 1 36 119
36.3.8 Sequential Importance Sampling
Assuming our proposal density can be factored:
Recursive Estimator
q x k Z k = q x k x k − 1 , zk q x k − 1 Z k − 1 36 120
In this section, we leverage the previously presented con-
cept of importance sampling to derive the basis for a recur- the posterior particle locations can be sampled from
sive nonlinear estimator using Monte Carlo integration [4]. j j
This type of filter is generally referred to as a recursive par- x k q x k x k − 1 , zk 36 121
ticle filter.
Thus‚ the associated particle weights at time k can be cal-
Consider the following general system model:
culated in a similar fashion as Eq. 36.108:
x k = f xk − 1 , w k − 1 36 112 j
p x k Zk
j
zk = h x k , v k 36 113 wk 36 122
j
q x k Zk
where xk is the state vector at time k, f( , ) is the process
model function at time k – 1, wk − 1 is the process noise vec- Substituting Eqs. 36.119 and 36.120 into Eq. 36.122 yields
tor, h( , ) is the observation function, and vk is the measure-
j j j j
ment noise vector at time k. The noise vectors are assumed p zk x k p x k x k − 1 p x k − 1 Zk − 1
j
to be independent of each other and in time with a known wk
j j j
q x k x k − 1 , zk q x k − 1 Zk − 1
density function. Note that Gaussian densities are not
required or assumed. 36 123

c36.3d 1136 15/12/2020 10:54:55 PM

 36.3 Nonlinear Filtering Concepts 1137

Note that this equation is a function of the posterior 0.4

weights at time k – 1; thus‚ the right-hand fraction of 0.3
Eq. 36.123 can be replaced according to Eq. 36.108, which
0.2
yields the final particle weight update equation from time
k – 1 to time k: 0.1

Velocity (m/s)
0
j j j
p zk x k p x k xk − 1
j j –0.1
wk wk − 1 36 124
j j
q x k x k − 1 , zk –0.2

which can be normalized such that the collection of weights –0.3

sums to one, thus approximating the posterior density as –0.4

N –0.5
j j
p x k Zk ≈ wk δ x k − x k 36 125
j=1 –0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1
Position (m)
In this manner, the particle locations and weights can be
continuously maintained and updated using a recursive Figure 36.17 SIR particle filter initial state estimate and position
estimation framework. density function. Note that the density function is extremely multi-
modal due to the limited information available at this point.

36.3.9 Sampling Particle Filter Demo

In this section, we apply a sequential importance sampling 0.2
particle filter design to our previous nonlinear estimation 0.15
example. As before, an identical, randomly generated tra-
0.1
jectory and measurement set from the MMAE example
Velocity (m/s)

(Section 36.3.3) are used as inputs to the filter. Once again, 0.05
for reference, the system parameters are specified in
0
Table 36.1, and the resulting trajectory, range observations,
and phase observations are shown in Figure 36.3. For this –0.05

example, we use 10 000 two-dimensional particles. Finally, –0.1

we exercise an importance resampling procedure [6] to
–0.15
ensure that the number of effective particles remains
acceptable. –0.2
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6
The SIS particle filter global state estimate and density
Position (m)
function of position after one observation (t = 1 s) are
shown in Figure 36.17. In this example, we show the loca- Figure 36.18 SIR particle filter state estimate (after
tion of the particles along with the estimated mean and 22 observations). Range observations combined with the vehicle
dynamics model are eliminating unlikely integer ambiguity values.
one-sigma standard deviation calculated using the ensem-
ble of particles. In the figures below, the estimated mean is
represented as a magenta “plus,” the true state is a green The global state estimate and associated standard devia-
asterisk, and the estimated 2-sigma error bounds as a tion result for this simulation is shown in Figure 36.20. The
dashed ellipse. Each particle location is shown as a shape of the uncertainty bound clearly shows the effects
black dot. described above. As the likelihood of each integer ambigu-
After 22 cycles, the density shows a reduced number of ity realization changes, the overall uncertainty changes and
peaks (see Figure 36.18) and is clearly multi-modal. Based eventually collapses to the centimeter level.
on our knowledge of the true density functions developed
in the previous examples, this indicates that the filter is
36.3.10 Strengths and Weaknesses
incorporating sensor observations and the statistical
of Approaches
dynamics model to effectively eliminate a number of poten-
tial ambiguity possibilities. In this chapter, we have presented three classes of nonlin-
After 100 cycles (Figure 36.19), the filter has converged to ear recursive estimation algorithms. While each algorithm
a single ambiguity. offers improved performance over the linear and extended

c36.3d 1137 15/12/2020 10:54:55 PM

 1138 36 Nonlinear Recursive Estimation for Integrated Navigation Systems

properties of multiple estimator types. While it is beyond

the scope of this chapter to explore the range of hybrid fil-
0.1
tering approaches, the interested reader is referred to the
references (e.g. [4, 5, 6, 9, 10]) for foundational concepts.
0.05
Velocity (m/s)

0
36.4 Summary and Conclusions
–0.05
In this chapter, we have presented an overview of nonlinear
estimation approaches suitable for navigation problems.
–0.1 Starting with first principles, three classes of nonlinear,
recursive estimators were derived, the performance
–0.15 was demonstrated using a common navigation example
2 2.1 2.2 2.3 2.4 2.5
Position (m)
application, and comparisons were made between the
approaches.
Figure 36.19 SIR particle filter state estimate (after 100 The growing availability of a wide range of sensors and
observations). Note that the state estimate is almost completely
improved computational resources has heralded a new
unimodal and has converged to the correct integer ambiguity.
era of multisensor navigation. Because many of these sen-
sors have nonlinear and non-Gaussian error models,
1 researchers are developing a range of recursive navigation
Xerr algorithms to meet these requirements.
0.8 σx
When used within their associated limitations, nonlinear
0.6
estimation algorithms hold enormous promise for addres-
0.4 sing the most difficult navigation problems.
Position Error (m)

0.2

–0.2 References
–0.4
1 Kalman, R.E. (1960) A new approach to linear filtering and
–0.6
prediction problems. Transactions of the ASME–Journal of
–0.8 Basic Engineering, 82 (Series D), 35–45.
–1 2 Maybeck, P.S. (1979) Stochastic Models, Estimation, and
0 10 20 30 40 50 60 70 80 90 100
Control, Vol. I, Academic Press, Inc., Orlando,
Time (s)
Florida 32887.
Figure 36.20 SIR particle filter position error and one-sigma 3 Maybeck, P.S. (1979) Stochastic Models, Estimation, and
uncertainty. Note that the error uncertainty collapses once Control, Vol II, Academic Press, Inc., Orlando,
sufficient information is available to resolve the integer ambiguity.
Florida 32887.
4 Gordon, N.J., Salmond, D.J., and Smith, A.F. (1993) Novel
Kalman filter in the presence of nonlinearities and non- approach to nonlinear/non-Gaussian Bayesian state
Gaussian systems, it is important to address the “strengths estimation, in IEE Proceedings F-Radar and Signal
and weaknesses” of each. To accomplish this, we evaluate Processing, vol. 140, IET, vol. 140, pp. 107–113.
each estimation from this perspective, starting with the tra- 5 Doucet, A., De Freitas, N., and Gordon, N. (2001)
ditional approaches. Sequential Monte Carlo Methods in Practice. Series
As expected, each approach has a set of associated Statistics for Engineering and Information Science.
strengths and weaknesses that can greatly influence the 6 Ristic, B., Arulampalam, S., and Gordon, N. (2004) Beyond
results for a given problem. Thus, the choice of estimator the Kalman Filter: Particle Filters for Tracking Applications,
must be considered carefully based on the characteristics Artech House.
of the problem at hand. In cases where the constraints of 7 Sheldon, S. and Maybeck, P. (1990) An optimizing design
the problem do not readily fit into the generalized cate- strategy for multiple model adaptive estimation and
gories above, there are many examples of hybrid estimation control, in Decision and Control, 1990, Proceedings of the
schemes that seek to synergistically combine the desirable 29th IEEE Conference on, pp. 3522–3527, Vol. 6.

c36.3d 1138 15/12/2020 10:54:57 PM

 References 1139

8 Bar-Shalom, Y., Li, X.R., and Kirubarajan, T. (2004) UAI’00, pp. 176–183. URL http://dl.acm.org/citation.cfm?
Estimation with Applications to Tracking and id=2073946.2073968.
Navigation: Theory Algorithms and Software, John 10 Mustiere, F., Bolic, M., and Bouchard, M. (2006) Rao-
Wiley & Sons. Blackwellised particle filters: Examples of applications, in
9 Doucet, A., de Freitas, N., Murphy, K., and Russell, S. Electrical and Computer Engineering, 2006. CCECE ‘06.
(2000) Rao-Blackwellised particle filtering for dynamic Canadian Conference on, pp. 1196–1200.
Bayesian networks, in Proceedings of the Sixteenth 11 Papoulis, A. and Pillai, S.U. (2002) Probability, Random
Conference on Uncertainty in Artificial Intelligence, Morgan Variables and Stochastic Processes, McGraw-Hill,
Kaufmann Publishers Inc., San Francisco, California, New York.

c36.3d 1139 15/12/2020 10:54:59 PM

 1141

37

Overview of Indoor Navigation Techniques

Sudeep Pasricha
Colorado State University, United States

37.1 Introduction navigate new buildings to reach their desired location.

Services that utilize indoor location systems can also enable
While localization and navigation in outdoor environments smart dynamic locking of sensitive rooms and resources if
can make use of global navigation satellite systems (GNSSs) an owner is not present, to improve overall safety. Ubiqui-
such as Global Positioning System (GPS), this is not a viable tous localization already plays a central role in social
solution for indoor localization because signals from GPS networking, for instance‚ to locate friends for coordinating
satellites are too weak to penetrate through buildings, joint activities or check into restaurants and other indoor
obstacles, and into underground environments. Conse- locales via various smartphone apps, and is expected to play
quently, precise localization inside a closed structure, such an even bigger role in the future. Indoor position awareness
as shopping malls, hospitals, airports, subways, and univer- is also an essential component of industrial applications,
sity campus buildings‚ require the use of alternative local- such as for robot motion guidance, robot cooperation,
ization technologies. But indoor environments present and smart factories (e.g. the ability to find tagged mainte-
unique challenges, particularly due to a diverse array of nance tools and equipment scattered all over a plant in
obstacles such as walls, doors, furniture, electronic equip- production facilities). Localization for cargo management
ment, and stationary or moving humans, all of which give systems at airports, ports, and for rail traffic enables
rise to multipath effects in wireless signals due to signal unprecedented opportunities for increasing their efficiency.
reflection, attenuation, and noise interference. As a result, Many different techniques have been proposed to enable
accurate indoor localization with wireless communication indoor localization and navigation. The interest in indoor
signals is a very complex problem. Moreover, indoor locales navigation systems is peaking because the crucial sensors
also require much higher levels of accuracy than outdoor necessary for localization have become sufficiently small
environments; for example, while a 4–6 m accuracy is and inexpensive to enable practical tracking of individuals
acceptable outdoors for vehicle navigation, it may not be (who must carry them at all times). A prime example of this
acceptable for localization in many indoor contexts‚ where is the inertial sensors that are part of inertial measurement
4–6 m may be the difference between one room and units (IMUs) found in smartphones that can aid with
the next. localization. However, activity trackers, smart cards, and
Enabling location services for indoor locales has many various types of wearable sensors can also play a crucial role
potential applications. Buildings with awareness of the to enable indoor navigation. The challenge today is to
location of occupants can use this knowledge to optimize exploit these available sensors to achieve indoor tracking
heating, lighting, and other resources toward saving energy with acceptable robustness levels, similar to that demon-
costs. In emergency scenarios such as earthquakes and strated by GNSS in outdoor locales.
hurricanes, location services can allow emergency respon- This chapter provides an overview of the state of the-art
ders to determine where people are located at any time, in the area of indoor localization and navigation. One can
potentially expediting evacuations as well as search and res- consider localization as an instantaneous process, provid-
cue efforts. Location awareness can be used as a backbone ing the location of a user or object being tracked at a specific
for smarter workplaces by allowing telephone calls to be instance of time. In contrast, navigation can be considered
routed to the nearest device in the proximity of a person, as a form of continuous localization, where location esti-
allowing colleagues to find each other, and helping guests mates must be provided frequently and periodically over

Position, Navigation, and Timing Technologies in the 21st Century: Integrated Satellite Navigation, Sensor Systems, and Civil Applications, Volume 2,
First Edition. Edited by Y. T. Jade Morton, Frank van Diggelen, James J. Spilker Jr., and Bradford W. Parkinson.
© 2021 The Institute of Electrical and Electronics Engineers, Inc. Published 2021 by John Wiley & Sons, Inc.
1142 37 Overview of Indoor Navigation Techniques

time to help a user while they navigate an indoor environ- locomotion-capable devices (e.g. drones). In general, it is
ment. Tracking can be considered to be similar to naviga- the job of the localization system to determine the (local
tion, except that the location estimates of the mobile or global) coordinates of such mobile nodes.
subject are provided not to the user but to some third party Centralized and Distributed Localization. In a cen-
that is interested in the location information. For the sake tralized localization architecture, location estimation is car-
of clarity and brevity, we will mainly use the term localiza- ried out at a central server where all anchor and mobile
tion in the rest of the chapter to represent both instantane- node locations are stored and available to an administrator.
ous point estimates and continuous estimates (navigation The benefits of centralized architectures are simplicity, uni-
or tracking), when discussing components and solutions form service to all users, and lower expansion costs as most
that are relevant for indoor location estimation. We will of the intelligence in the system is concentrated at one loca-
use the term navigation sparingly, when necessary to tion, allowing the mobile and anchor nodes to be lower cost
discuss the unique aspects of continuous estimation of and contain fewer components. In a distributed system,
location. location estimation is carried out on each mobile and
The rest of the chapter is organized as follows. anchor node based on local observations. The advantages
Section 37.2 discusses performance metrics that are neces- of a distributed architecture are good system scalability
sary to understand, in order to compare and contrast the and better guarantees of the user’s privacy (as sensitive
landscape of indoor localization approaches. Section 37.3 location information is not centrally stored, making it less
provides an easy reference to the key technical terms that susceptible to being compromised).
are used throughout the rest of the chapter. Section 37.4 Line of Sight (LOS). When a signal can travel via a
presents a review of the various signals that can be used direct straight path from an emitter to a receiver, it is
to provide tracking in indoor locales, for the purpose of referred to as LOS transmission. Several localization tech-
localization. Section 37.5 provides an overview of the vast niques rely on LOS, for example, time of arrival (ToA)-
landscape of solutions for indoor localization. Lastly, based distance measurements with radio frequency (RF)
Section 37.6 discusses open research issues and challenges signals. But due to occlusions from walls, furniture, and
that still remain to be overcome for viable indoor people, most indoor environments typically induce non-
localization. LOS (NLOS) propagation, which may cause inconsistent
time delays at a radio receiver. These delays pose a chal-
lenge that can only be tackled by few localization
techniques.
37.2 Overview of Technical Terms Multipath Environment. An environment in which a
transmitted signal propagates along multiple paths
This section provides a brief overview of some of the (echoes), each of which arrives with different path delays
commonly used technical terms that are relevant in the at the receiver, is referred to as a multipath environment.
field of indoor localization [1]. Multipath propagation of signals is particularly problem-
Absolute and Relative Location. A location deter- atic for time-based localization methods (Section 37.5.1.2)
mined within the context of a global or large area reference because signal paths from different directions degrade the
grid obtained from GNSS satellites, markers, or landmarks ability to determine the travel time of the direct path.
is referred to as an absolute location. In contrast, relative One way to distinguish the direct path from a non-LOS
positions depend on a local frame of reference, for example, path is to move the receiver or transmitter. Non-LOS paths
coordinates within a small coverage area that represent change erratically while in motion, allowing for separation
displacement with respect to a local fixed reference (e.g. a and averaging, while the direct path is directly related to the
fixed Wi-Fi access point with known global coordinates). motion of the object. Thus, averaging over time with a
Anchor and Mobile Nodes. From a networking per- motion-tracking model is one effective way to mitigate mul-
spective, nodes in indoor environments that are part of tipath. Another way to overcome multipath is to switch to
the network and have a stationary (fixed) location are different frequency channels. Alternatively, radio signals
referred to as anchor nodes. In the literature, such nodes with a large absolute frequency bandwidth such as Ultra-
may also be referred to as beacons, fixpoints, access points Wideband (UWB) have been shown to be advantageous
(APs), base stations, or reference nodes. Typically, the coor- for mitigation of multipath fading [2].
dinates of such anchor nodes are assumed to be known. In Received Signal Strength Indicator (RSSI). Signal
contrast, nodes that are part of the network and can move attenuation can be used for distance estimation during
in the indoor environment are referred to as mobile nodes. localization, based on RSSI values. RSSI are observed RSS
Such nodes could represent people, robots, or other (received signal strength) values averaged over a specific
 37.3 Performance Metrics 1143

sampling period and usually specified as received power PR and other characteristics. Therefore, a compromise
in decibels. Based on the attenuation model between adequate accuracy and other characteristics
described below is essential.
GT GR
PR PT Timeliness: The timeliness or responsiveness of a solution
4πdp
determines how quickly the location estimate of a target
the received signal power or signal strength PR can help is obtained. For simple indoor localization queries, a fast
with the estimation of the distance d of a mobile user or response to the query is important in most cases, but not
object from the transmitter. In this model, PT is the trans- crucial. However, for navigation systems, timeliness is a
mitted power at the transmitter, GT and GR are the antenna critical measure of effectiveness: if location estimates are
gains of the transmitter and receiver, and p is the path loss not updated quickly in sync with the motion profile of
exponent. The path loss factor p characterizes the rate of the subject being tracked, the system will be ineffective
attenuation with an increase in distance d. The free space for the purpose of navigation (regardless of the eventual
model does not take into account that antennas are usually accuracy of the estimates). Usually, the term location lag
set up above the ground. In fact, the ground acts as a reflec- is used to refer to the delay between a mobile subject
tor, and thus the received power differs from that of free moving to a new location and the new location of that
space. A mathematical formulation of such a path loss subject being reported by the system.
model, also known as open field model, can be found in Coverage: Any indoor localization solution must work and
[3]. Typically, in free space p = 2, whereas for environments be usable over the entire indoor environment of interest.
with NLOS multipath, p > 2. For indoor environments, the Coverage defines the area over which a localization solu-
path loss exponent typically takes higher values between 4 tion can provide estimates of sufficient accuracy, and
and 6. Theoretically, distances estimated from RSSI values possibly timeliness, to be considered useful. The physical
to multiple anchor nodes can be used to determine the environment (e.g. obstacles, walls, doors) plays a crucial
receiver position by multilateration techniques (see role in limiting the availability of signals that are used by
Section 37.5.1 for more details). However, interference, a given localization technique, consequently impacting
multipath propagation, and presence of obstacles and peo- the coverage achievable by the technique for that envi-
ple results in a complex spatial distribution of RSSI values, ronment. Intuitively, it is possible to extend coverage
which can make the estimation of distances using RSSI by altering the physical environment or supplementing
alone quite inaccurate. Therefore, fingerprinting has it with additional hardware, for example, wireless signal
become more popular than propagation modeling (see repeaters. Coverage can also be improved by enhancing
Section 37.5.2 for more details). the hardware carried by the user or object being tracked,
for example, using mobile devices with more powerful
and capable wireless radio antennas and chipsets.
37.3 Performance Metrics Adaptiveness: Often, the physical environment around
the subject to be tracked does not stay the same over
Indoor localization solutions need to meet several goals if time. For example, at different times of the day and days
they are to be considered viable candidates for use in indoor of the week, the number of people in a shopping mall
environments. Here we review some of the more relevant varies quite significantly. In some environments,
performance metrics [4] that must be satisfied by any can- machinery, goods, containers, and other equipment
didate indoor localization solution: may be repositioned constantly. Sometimes signals from
wireless transmitters are temporarily blocked in an envi-
Accuracy: The location error of a positioning system is one ronment, or some transmitters may stop functioning due
of the most important metrics used to determine the to unpredictable circumstances. These changes create a
effectiveness of a localization system. In its simplest challenge for any indoor localization solution that relies
form, localization accuracy can be reported as an error on these signals. The ability of a solution to cope with
distance between the estimated location and the actual these environmental changes represents its adaptive-
location of the user or object being tracked. For naviga- ness, or robustness. Obviously, a solution that is able
tion systems, this may take the form of a running average to adapt to environmental changes can provide better
of errors over a time period of interest, or the error could localization accuracy than solutions that cannot adapt.
be calculated using geometric principles, to estimate the An adaptive system can also prevent the need for
deviation of the predicted trajectory from the actual tra- repeated calibration of sensors used for localization.
jectory. Usually, the higher the accuracy, the better the Scalability: At a system level, solutions for localization
system, but there is often a trade-off between accuracy may require supporting requests from multiple entities.
1144 37 Overview of Indoor Navigation Techniques

For instance, a system deployed in a shopping mall needs sophistication required from their associated signal pro-
to be able to handle location queries from a few people, cessing software and hardware. While some techniques
all the way up to thousands of people simultaneously. may involve very simple hardware (e.g. inertial sensors)
The ability to “scale up” and quickly respond to multiple and software (e.g. to implement simple filtering techni-
location queries is of paramount importance in many ques), other techniques may require more complex cus-
indoor environments. Poor scalability can result in poor tom hardware (e.g. for specialized digital signal
localization performance, necessitating the reengineer- processing) and complex software (e.g. sophisticated
ing or duplication of systems, which can increase deploy- machine learning techniques). Also, if the computation
ment overheads. of the localization algorithm is performed on a centra-
Integrity: The confidence that can be placed in the output lized server, the localization can be quickly estimated
of a localization solution can be termed its integrity. due to the powerful processing capability and the suffi-
A solution with low integrity has a high probability that cient power supply; however if it is carried out on a
a malfunction will lead to an estimated position that dif- mobile device, the effects of complexity can be much
fers from the required position by more than an accept- more apparent. Inevitably, complexity impacts the cost
able amount and that the user will not be informed of the solution, and thus it is common practice to trade
within a specified period of time about the malfunction. off the complexity with the other (non-cost) metrics.
While regulatory bodies have studied and defined integ-
rity performance parameters in some sectors such as civil
aviation, for indoor localization it is more difficult to find
well-quantified integrity parameters. At the very least, an 37.4 Indoor Localization Signal
indoor localization solution must provide an indication Classification
of some integrity parameters that are related to safety
of life, economic factors, or convenience factors; thereby GPS is the most popular wireless-signal-based positioning
allowing consumers of the solution to understand its lim- system in use today, and is extremely useful in outdoor
its and capabilities under different usage scenarios. environments [5]. GPS satellites broadcast microwave sig-
Cost: An indoor localization system has costs associated nals to enable GPS receivers on or near Earth’s surface to
with it that must be as low as possible, to incentivize determine location, velocity, and time. The GPS system
widespread adoption and ease deployment overheads. itself is operated by the US Department of Defense (DoD)
These costs may include installation of localization solu- for use by both the military and the general public. Unfor-
tion-specific hardware and site survey time during the tunately, GPS signals cannot penetrate into indoor environ-
deployment period. If a positioning system can reuse ments due to obstacles that spread and attenuate the
an existing communication infrastructure (e.g. Wi-Fi electromagnetic radio signals [6]. Thus, GPS cannot be used
APs already deployed in a building), some part of the for localization in indoor environments. Fortunately, there
infrastructure, equipment, and bandwidth costs can be are many other signals available in indoor locales that can
saved. In addition to the infrastructure, there may also be leveraged by solutions intended for indoor localization.
be costs associated with the mobile devices carried by This section reviews some of the more relevant signals that
the subject being tracked. For instance, such costs could can be used for indoor localization. Figure 37.1 shows a tax-
represent monetary costs of the smartphone and any onomy of the signals that are covered in more detail in the
externally connected hardware sensors. However, the rest of this section.
cost could also be calculated by considering other
aspects, such as lifetime, weight, and energy consump-
37.4.1 Infrared Radiation (IR) and Visible Light
tion. For example, some mobile devices, such as elec-
tronic article surveillance (EAS) tags and passive radio Electromagnetic radiation at wavelengths within the visible
frequency identification (RFID) tags, are energy passive range, which extends approximately between 380 and 750
(i.e. they only respond to external fields) and thus, can nm, as well as in its lower or upper vicinity, known as ultra-
have an unlimited lifetime; however other mobile violet (UV) and infrared (IR) light, are part of some of the
devices (e.g. smartphones with rechargeable battery) most common indoor positioning systems that use wireless
have a limited lifetime of several hours without technology.
recharging. Visible light systems typically utilize general-purpose
Complexity: Indoor localization solutions inevitably cameras and have been adopted particularly for indoor
require hardware and software components that can localization of robots. One common approach is to have a
have different complexities. Solutions may differ in the robot carry a camera to capture images of the environment
Other documents randomly have
different content
fashioned to the letter “G” placed horizontally. There are two similar
guns in the Musée d’Artillerie.
1678. Traité des Armes, p. 55, Gaya.
Quoy que les Bufles ne soient proprement que les habillemens de
Cavaliers, nous pouvons neanmoins les mettre au nombre de leurs
armes deffensives, plus qu’ils peuvent aisement résister à l’Epée
lors qu’ils sont d’une peau bien choissie.
Les Bufles ... sont faits en forme de Juste-au-corps à quatre
basques qui descend jusqu’aux genoux.
Il n’y a pas un Cavalier dans les trouppes de France qui n’ait un
habillement de Bufle.
The buff coat of leather or “cuir de bœuf” was a part of the
military equipment as early as 1585 and was in common use during
the Civil War. It was worn by the Life Guards at the Coronation of
James II in 1685 and by a detachment of the Artillery Company at
the entry of George I in 1714. It ceased to be worn as part of the
uniform in the following reign.[119]
1591–5. Instructions, Observations and Orders Militarie, p. 185, Sir
John Smith.
... halbadiers ... armed with burganets and with short skirted
Ierkins of buffe with a double buffe on their breasts and the
sleeves of their doublets with stripes of maile or serecloth
aforesaide.
Here we find a return to the primitive defence of the eleventh
century, due to the increased weight of armour which was necessary
against the improved firearms which were by this time a serious
factor in war. The serecloth recommended was probably a stout
waxed or oiled canvas. In recommending sleeves of mail, which are
shown on Plate XVIII, Sir John Smith considers that they are more
convenient for the handling of the halberdier’s weapon than the
more rigid brassards worn by the cavalry. These strips of chain are
shown on one of the figures painted by Memling for the “Chasse of
S. Ursula” at Bruges, 1486, which is given on Fig. 24 of this work.
They have been re-introduced as shoulder-straps for heavy cavalry
at the present day.
 F O OT N OT E S :
[113] The Pembridge effigy in Hereford Cathedral has thigh-
pieces which apparently represent leather laced on the inside.
[114] Memoirs, Vol. I, ch. 33.
[115] Arms and Armour at Oxford, C. ffoulkes.
[116] Arms and Armour at Oxford, C. ffoulkes.
[117] Johnes’ trans., I, 739.
[118] Kündig, Geneva, 1910.
[119] Cannon, Historical Records of the Life Guards, p. 74.
 THE WEARING OF ARMOUR

T
hough perhaps the wearing and putting on of armour was not
directly part of the craft of the armourer, it was certainly a part
of his duties to be present during the process and be ready to
carry out any small alterations which might be needed on the spot.
As has been noticed in a preceding chapter, as late as 1625 we
find this insisted upon by de Pluvinel (see page 115). Shakespeare
describes the armourers as busy “accomplishing the knights” before
Agincourt (page 33), and the fact that the travelling knight took his
armourer with him shows that he was indispensable during the
operation of dressing for war or joust.
Armour of the best kind was made to measure, and for ordinary
purposes a mould or “dobble” was kept on which to make the
ordinary harness for the man-at-arms (page 28). The following
extracts show the methods employed for sending measurements,
which were often obtained by submitting the clothes of the patron to
the armourer:—
1406. In the will of Sir Ralph Bulmer, “armatura mea corpori
talliata.”[120]
1470. Archives de Bruxelles.[121]
Baltazar du Cornet, armourer at Bruges, delivers for the Duke of
Burgundy “2 cuiraches complettes faites a la mesure de
Monseigneur.”
Lazarus de St. Augustin delivers “un harnais complet fait
naguere a la mesure de Monseigneur et pour son corps.”
1512. A jacket and hose of Prince Charles (afterwards Charles V) are
sent to Conrad Seusenhofer.[122]
1520. Brit. Mus., Calig. D, VIII, 181.
16 March. Francis I asks for an “arming doublet” of Henry VIII that
he may have made a new kind of cuirass which he will send him as
a present.

PLATE XXVII

DRAWING BY JACOB TOPF, 1530–1597

FROM THE “ARMOURER’S ALBUM,” VICTORIA AND ALBERT MUSEUM
 PLATE XXVIII

ARMOUR OF SIR JOHN SMITH, BY JACOB TOPF

1564. S.P.D. Elizabeth, Jan. 30.

Warrant to the Master of the Armoury. To cause to be made one
armour complete fit for the body of our well beloved servant
 Christopher Hatton, one of our Gentlemen Pensioners, he paying
according to the just value thereof.
1667. Verney Memoirs, IV, 301. Rich. Hals to Edmond Verney.
The armour fits well enough only the man did cut away to much
just under the arme pit both of back and breast, but for the head
piece it is something heavy, yet I think it well enough if it did not
come downe so low upon my forhead as to cover all my eyes and
offend my nose when I put my head backwards to look upwards.
In the preceding chapter some
notice was taken of the part which
the linen armourer played in the
equipment of the armed man, and it
was to him that the clothing which Fig. 51. Stripping the dead
was worn under the armour was (Bayeux Tapestry).

entrusted. Under the heading of the

“Cleaning of Armour” mention has been made of Chaucer’s knight
whose “gipoun” was “besmoturyd with his haubergeon,” but this
garment was an outer garment or surcoat. In the age of plate
armour a complete dress was worn for legs, arms, body, and head to
prevent the chafing of the armour, which in spite of its own lining of
silk, velvet, cloth, leather, or other fabric would cause grave
inconvenience, if not danger to the wearer. Besides this reason there
was also a question of warmth, which was of importance, for in long
marches and expeditions there was no warmth in a suit of plate, in
fact there was an added cold which had to be counteracted by warm
garments worn underneath.
In the eleventh and twelfth
centuries we have not much in the
way of documentary evidence which
will help us as to the clothes worn
under the armour. The Bayeux
Tapestry shows us the wounded
and dead being stripped of their
Fig. 52. Knight arming (from Livre des Nobles hauberks, under which nothing was
Femmes, Bib. Nat., Paris, fourteenth century). apparently worn (Fig. 51). It should
be remembered, however, that these hauberks were probably of
quilted fabric, which therefore did not gall the body of the wearer.
The drawing from a fourteenth-century manuscript on Fig. 52 gives
some hint at the arming-doublet, which will be noted farther on in
this chapter, and shows also the laces or points that held up the
hose. Towards the end of the fourteenth century, however, we find
on the incised brasses, which are such valuable records of the
military equipment of the period, very distinct garments represented.
On the brass to Sir John de Creke at Westley Waterless, Cambs,
1325, we see the “cyclas” or outer surcoat, the “upper pourpoint,” of
fabric, studded with metal, “the hauberk,” and under all the
“haketon” or “gambeson” (Fig. 53). According to William de
Guilleville, in the Pèlerinage de l’Ame, written in the fourteenth
century, the “pourpoint” was so called because of its quiltings:—
De pontures de gambison
Pourquoi pourpoint l’appelle-t-on.

The gambeson continued in use up to the seventeenth century

under the name of “arming-doublet,” with but little change except in
shape and form, as the style of armour required. Of the
undergarments of the early fifteenth century we have little or
nothing to guide us, and we are often at a loss to know even what
armour was worn under the tight-fitting, small-waisted jupon or
surcoat which distinguishes the end of the fourteenth and the
beginning of the fifteenth century. We have, however, a valuable
record under this head in the monument at Ash, which shows
“splinted armour” of lames worn instead of a cuirass.
The illustration on Plate IV is from a wood-carving in the church of
S. William, Strasburg. It represents the travelling armourer riveting
what appear to be bands of iron on arms and legs. Whether these
are some contrivance used in arming in the fifteenth century, or
whether they are some instrument of torture used upon the saint,
Duke William of Acquitaine, it is impossible to discover, as no other
instances of the kind can be found.
 For full details of the equipment of the latter
half of the fifteenth century we cannot do better
than refer to the Hastings MS. of the fifteenth
century, which has been discussed by the late
Albert Way,[123] and more fully by Viscount
Dillon.[124] Under the heading of “The Abilment
for the Justes of Pees” we find much that is of
value in this respect. On page 122b of the
manuscript we find the following minute
directions for dressing a man for the joust, which
should be compared with those given in
Appendix C, page 173.
How a man schall be armyd at his ese when he
schal fighte on foote:
He schal have noo schirte up on him but a dowbelet
of ffustean lyned with satene cutte full of hoolis. the
dowbelet must be strongeli boude there the pointis
muste be sette aboute the greet [bend] of the arm.
and the b ste [sic] before and behynde and the
gussetis of mayle muste be sowid un to the dowbelet
Fig. 53. Brass of Sir
John de Creke,
in the bought of the arme. and undir the arme the
Westley Waterless, armynge poyntis muste ba made of fyne twyne suche
Cambs, 1325. as men make stryngys for crossebowes and they
1. Bascinet. muste be trussid small and poyntid as poyntis. Also
2. Vervelies and camail. they muste be wexid with cordeweneris coode. and
3. Cyclas or surcoat. than they will neyther recche nor breke Also a payr
4. Upper pourpoint.
5. Hauberk.
hosyn of stamyn sengill and a payre of shorte
6. Gambeson or haketon. bulwerkis of thynne blanket to put aboute his kneys
7. Poleynes. for chawfynge of his lighernes Also a payre of shone
8. Beinbergs or jambs. of thikke Cordwene and they muste be frette with
smal whipcorde thre knottis up on a corde and thre
cordis muste be faste swoid on to the hele of the shoo and fyne cordis in
the mydill of the soole of the same shoo and that ther be betwene the
frettis of the hele and the frettis of the mydill of the shoo the space of
three fvngris.
To arme a man
ffirste ye muste sette on Sabatones and tye them up on the shoo with
smale poyntes that wol breke And then griffus [greaves] & then quisses &
 he the breeche of mayle And the tonletis And the brest And he vambras
And he rerebras And then glovys And then hange his daggere upon his
right side And then his shorte swered upon the lyfte side in a rounde
rynge all nakid to pull it oute lightlie. And then putte his cote upon his
back And then his basinet pynid up on two greet staplis before the breste
with a dowbill bokill behynde up on the bak for to make the bassinet sitte
juste. And then his long swerde in his hande. And then his pensil in his
hande peyntid of seynt George or of oure lady to blesse him with as he
goeth towards the felde and in the felde.

Fig. 54. Arming-points (from the portrait of Fig. 55. Attachment of brassard by
a Navigator, Ashmolean Mus., Oxford). points (from the portrait of the Duc
de Nevers, Hampton Court).

From the above extract it will be seen that the undergarments

consisted of a thick doublet lined with silk, but with no shirt
underneath; the reason for this being one that we at the present day
can well appreciate, for when the body is hot from exertion and
exercise a shirt is apt to “ruck up,” and it would be impossible to
readjust it when fully armed. In the Paston Letters we have the
following request from Edward IV:—
Item I praye you to send me a newe vestmente off whyght damaske
ffor a Dekyn, whyche is among myn other geer, I will make an armyng
Doublet off it.
 PLATE XXIX

ARMET, MIDDLE OF XVI CENT. ARMET ENGRAVED AND GILT, END OF XVI CENT.

HELM OF SIR RICHARD PEMBRIDGE, CIRC. 1360

PARADE CASQUE, AFTER NEGROLI, SALLAD BY ONE OF THE NEGROLIS, END OF XV CENT.
MIDDLE OF XVI CENT.
 The gussets and, in the sixteenth century, the
sleeves of mail protected the bend of the arm and
armpit, and sometimes the bend of the knee,
which were not adequately covered with plate.
The two portraits of unknown noblemen by
Moroni (National Gallery) show these details of
the equipment very clearly (Plate XVIII). The
arming-points or “tresses” were used in civilian as
well as in military attire and joined the hose to
the doublet, laced sleeves, and held coats Fig. 56. Moton attached
together, much as laces are used in ladies’ by points (from Harl.
MS. 4826).
dresses at the present day (Figs. 54–57). They
are also shown tying up the hose on Fig. 52 and
the brayette on Plate VIII.
Lord Dillon explains the hose of “stamyn sengill”
as being a worsted cloth made in Norfolk. The
“bulwerkis” were pads of blanketing fastened over
the hose at the knees to prevent the chafing of the
knee-cop, and the shoes were of Cordova leather
fastened with laces. A complete underdress of this
Fig. 57. Arming-points kind, with quilted doublet and hose with gussets of
on the foot (from mail at the knees, is to be found in the Museum at
the picture of S.
Demetrius, by Munich. The arming of a man began at the feet,
Ortolano, Nat. Gall.). and as far as was possible each piece put on
overlapped that beneath it, to ensure that glancing
surface upon the utility of which such stress has been laid in the first
chapter of this book.
The arming of a man, therefore, was carried out in the following
order and his equipment put on in the following order: Sollerets or
sabatons, jambs, knee-cops, cuisses, skirt of mail, gorget, breast
and back plates, brassards with elbow-cops, pauldrons, gauntlets,
sword-belt, and helmet (Fig. 58).
The “tonlet” would appear to be a bell-shaped skirt of plate or
deep taces such as is shown on Plate XXI, and is another example of
the use of the “glancing surface,” especially in combats with axe and
sword at barriers, for in these jousts the legs were often unarmed
and were not attacked. The rerebrace, elbow-cop, and vambrace are
usually joined by rivets in which there is a certain amount of play.
Where this was not the case, each piece was separately strapped to
the arm, as may be seen in the brasses of Sir John de Creke, 1325
(Fig. 53), and of Sir Hugh Hastings, 1347. When the three pieces,
called collectively the Brassard, were joined together, they were kept
in place on the arm by arming-points fastened to the “haustement”
or doublet just below the shoulder. The operation of tying on the
brassard is shown on the portrait now labelled the “Duc de Nevers”
at Hampton Court (Fig. 55). In the list of the equipment taken by
the Earl of Northumberland to France in 1513[125] we find mention
of arming-pateletts of white satin quilted, for wearing under the
armour, trussing-bolsters to wear round the waist to keep the weight
of the cuirass from the shoulders, arming-hose, arming-doublets,
arming-shoes, garters to wear under the armour, and coffers in
which to keep the armour.
 Fig. 58. Sixteenth-century Suit of Plate.

English French German Italian Spanish

1. scull timbre scheitelstück coppo calva
2. visor visière visier visiera vista
3. ventail ventail schembart ventaglio ventalle
{ bavière
4. bevor } kinreff baviera barbote
{ mentonnière
5. crest crête kamm cresta cresteria
{ porte-plume
6. plume-holder } pennachiera penacho
{ porte-panache
7. nape-guard couvre-nuque nackenschirm gronda cubrenuca
8. gorget colletin kragen goletta gorjal
9. spring-pin piton à ressort federzapfen
 10. neck-guard garde-collet brechränder guarda-goletta bufeta
11. pauldron épaulière achseln spallaccio guardabrazo
12. rerebrace arrière-bras oberarmzeug bracciali brazali
13. lance-rest faucre rüsthaken resta restra de muelle
14. rondel or { rotellino da
} rondelle achselhöhlscheibe } luneta
besague { bracciale
15. breast plastron brust petto peto
16. back dossière rücken schiena dos
17. elbow-cop or
} coudière armkasheln cubitiera codales
coude
18. vambrace avant-bras unterarmzeug bracciali brazali
19. gauntlet gantelet handschuhe mittene manopla
20. taces bracconière bauchreisen panziera faldaje
21. loin-guard garde-reins gesassreifen falda ”
22. fald or skirt { stahlmaschen-
} brayette } braghetta
of mail { unterschutz
23. tasset tassette beintaschen fiancale escarcela
24. upper cuishe cuissard oberdiechlinge cosciali quijotes
25. cuishe ” unterdiechlinge ” ”
26. knee-cop genouillière kniebuckel ginocchielli guarda o rodillera
27. jamb or
} jambière, grève beinröhen gambiera greba
greave
28. solleret or
} soleret schuhe scarpe escarpe
sabbaton
29. fan-plate ailerons

There is no mention of the pauldron in the Hastings MS., but when

this was worn it was strapped to the neck-opening of the cuirass or
hung from spring-pins which project from the shoulder-plate of the
cuirass.
The staples mentioned in the Hastings MS. are often very
elaborate contrivances, especially in jousting-armour, and the
foremost fastening was called the “charnel.” Fig. 59 shows the
methods of attaching jousting-helms to the cuirass. No. 1 shows the
adjustable plate which fixes the front of the helm of the suit of Philip
II (Madrid, A, 16). A similar contrivance was used with the “Brocas”
helm (Fig. 12). No. 2 is the front of a helm (Mus. d’Art, Paris, G,
163) in which the lower plate is bolted to the breast and can be
released from the helm by withdrawing the hinge-pin. No. 3 shows
the back of the same helm. Fig. 60 is a larger sketch of the fixing-
hook of this helm. A is the back-plate of the helm, E the pillar hinged
at D and hooked into a lug on the back of the cuirass. B is a solid
block of steel of circular section pierced with holes and connected to
a screw in E. B can be turned by inserting a pin in the holes and the
screw tightened or loosened. Minute details as to the fastenings of
the helm will be found in Appendix D, page 178.

Fig. 59. Attachment of jousting-helms to the cuirass.

It can therefore be easily imagined that the work of arming a man

was a serious business, and it was necessary that the armourer or
an expert assistant should be present in case some portion of the
suit or its fastenings gave way.
Details of the different parts that went to make up the complete
suit, with the thickness of each plate, the laces or points, and
various fastenings and methods of attachment, will be found in the
fifteenth-century Treatise on Military Costume of which a portion is
given in Appendix D.
 The Marquis de Belleval published an interesting
monograph on this manuscript in 1866, which is now
scarce and difficult to obtain.
In the illustration on Plate XVII the squires are shown
arming their masters from horseback, which appears to
involve some gymnastic exercises.
That such agility of the armed man was by no means
an artistic licence we may gather from the fact that
Fig. 60. Froissart[126] mentions Sir John Assueton leaping fully
Side view of armed behind his page on to his war-horse. Again,
attachment on
Fig. 59, 3. Shakespeare makes Henry V (Act V, Sc. 2) say, “If I
could win a lady at leapfrog or by vaulting into my
saddle with my armour on my back,” and Oliver de la Marche states
that Galliot de Balthasin in 1446 leaped fully armed out of the saddle
as though he had on a pourpoint only. That this was no mere figure
of speech we may judge from a little book entitled The Vaulting
Master, written by W. Stokes, an Oxford riding-master, in 1641.

PLATE XXX
 ARMOUR OF FRIEDRICH DES SIEGREICHEN,
ARMOUR OF THE MIDDLE XV CENT. BY TOMASO DA MISSAGLIA, 1460

In the preface he writes: “In war the nimble avoydance of a man’s

horse if wounded or killed under him, and in like manner the ready
ascent into his enemies saddle if it be his hap to unhorse him, and
much more which the experienced souldier shall find.”
 There is an engraving on Plate I of the
work showing a cuirassier in half-armour
about to vault into the saddle without
stirrups. Stokes occasionally breaks out into
verse as follows:—
Here’s that will make a stubborne armour weare
Gentle as Persian silks and light as air,
which refers to the ease of mounting which
his prescribed exercises ensured.
On the subject of the wearing of armour Fig. 61. Armourer in the lists
(Heralds’ Coll., MS. M, 6, f. 56).
we have much valuable information from the
works of the great military reformer of the sixteenth century, Sir
John Smith, who, as has been stated previously, suffered
imprisonment for his opinions. In his Instructions and Observations
and Orders Militarie, 1591–5, he writes:—

Page 183. “No man can be conveniently armed unlesse he

be first fitly apparelled.” He states that at Tilbury he saw “but
very few of that army that had any convenience of apparel
and chieflie of doublets to arme upon, whereof it came to
passe that the most of them did weare their armors verie
uncomelie and uneasilie.... But because the collars of their
armours doe beare the chief waight of all the rest of the
armour, I would wish that the souldiers ... should have under
Collars of Fustian convenientlie bombasted to defende the
heveth weight, and poise of their armours from the paining or
hurting of their shouldiers.”
On page 193 he writes: “Also I would have them to have
pouldrons of a good compasse and size, and vambraces both
joined together, and not asunder, because that the poise of
the pouldrons and vambraces, hanging upon the pinnes and
springes of their collars, they doe not weigh so much, nor are
not so wearisome as when they are separated; and that they
weare their vambraces tied with points to their doublets
 under their pouldrons.” Here the author, who was pre-
eminently a practical soldier, saw the discomfort and
inconvenience caused by the drag of the arming-point on the
sleeve and wisely considered that the whole arm-defence
should hang from a pin or strap from the gorget or cuirass, so
that the weight might be on the shoulders and not on the
arms.

The armour for the joust in the sixteenth and seventeenth

centuries was far too heavy to allow of such vagaries. Pluvinel in his
Maneige Royale, 1625, gives an imaginary conversation between
himself and the King which bears upon the subject:—

The King.
It seems to me that such a man would have difficulty in
getting on his horse and being on to help himself.
Pluvinel.
It would be very difficult, but with this armament the case
has been provided for. In this way, at triumphs and tourneys
where lances are broken, there must be at the two ends of
the lists a small scaffold the height of the stirrup, on which
two or three persons can stand; that is to say, the rider, an
armourer to arm him, and one other to help him, as it is
necessary in these dangerous encounters that an armourer
should always be at hand and that all should be ready. Then
the rider being armed, and the horse brought near to the
stand, he easily mounts him ... for this reason the horses
must be steady.

A little pen-drawing of the sixteenth century in a manuscript

dealing with jousts (Heralds’ Coll., M, 6, 56) shows the armourer on
one of these scaffolds at the end of the lists (Fig. 61).
 In the chapter on the Proving of Armour the question of disuse on
account of weight was considered. From the sixteenth century and
even earlier we have records of the discarding of armour because it
hampered the wearer or for some equally cogent reason. The
following extracts bear upon the subject:—
1383. Chroniques de Dugesclin, line 5973 (edit. 1839).
Leurs cuissieres osterent tres tous communement
Par coi aler peussent trop plus legierement.
This refers to the action of Sir Hugh Calverly at the battle of Mont
Auray, who ordered his men to take off their cuisses in order to
move more easily.
1590. Discourses, p. 4, Sir John Smith.
But that which is more strange, these our such new fantasied men
of warre doe despise and scorne our auncient arming of ourselves
both on horseback and on foote saying that wee armed ourselves
in times past with too much armour, or peces of yron as they
terme it. And therefore their footmen piquers they doo allow for
verie well armed when they weare their burganets, their collars,
their cuirasses, and their backs, without either pouldrons,
vambraces, gauntlets or tasses.
Sir John Smith goes on to say that it was the discarding of his
cuisses that cost Sir Philip Sidney his life, for he received a wound
from a spent bullet which his armour might have deflected.
1619. The Art of Warre, Edward Davies.
[the arquebusiers were loaded] with a heavie shirt of male and a
burganet, by the time they have marched in the heat of summer or
deepe of winter ten or twelve English miles they are more apt to
rest than readie to fight.
1625. Souldiers’ Accidence, Markham.
As for the pouldron or the vant-brace they must be spared because
they are but cumbersome.
Against these extracts we must place the opinions of military
leaders who deplored the disuse of armour:—
1632. Militarie Instructions for the Cavallrie, Cruso.
 Captain Bingham in his Low Countrie exercise appointeth him [the
harquebusier] a cuirass pistoll proofe which condemneth the late
practice of our trained Harquebusiers to be erroneous which have
wholly left off their arms and think themselves safe enough in a
calf’s skin coat.
1756. Rêveries, Marshal Maurice of Saxe, p. 56.
Je ne sais pourquoi on a quitte les Armures, car rien n’est si beau
ni si avantageux. L’on dira peut-etre que c’est l’usage de la poudre
qui les a abolis; mais point du tout car du tems de Henri IV. et
depuis jusq’en l’annee 1667 on en a porter, et il y avoit deja bien
longtems que la poudre etoit en usage: mais vous verrez que c’est
la chere commodite qui les a fait quitter.
Marshal Saxe further suggests that the large proportion of wounds
are received from sword, lance, or spent bullet, and that all these
might be guarded against by wearing armour or a buff coat of his
own invention which when reinforced with steel plates weighed 30
lb.
THE WEIGHT OF ARMOUR
We have but few records in contemporary documents of the actual
weight of the different parts of the suit of armour, but we can obtain
these from examples of the sixteenth century onwards from
specimens in the different museums and collections.
That armour had become burdensome in the extreme owing to
the necessity of subjecting it to pistol and musket proof we know
from various writers on the subject.
La Noue in his Discours Politiques et Militaires, translated by “E.
A.” 1587, writes on page 185: “For where they had some reason in
respect of the violence of harquebuzes and dagges [muskets and
pistols] to make their armor thicker and of better proofe than before,
they have now so farre exceeded, that most of the have laden
themselves with stithies [anvils] in view of clothing their bodies with
armour ... neither was their armour so heavie but that they might
wel bear it 24 hours, where those that are now worne are so
waightie that the peiz [weight] of them will benumme a Gentleman’s
shoulders of 35 yeres of age.”
 PLATE XXXI

PORTRAIT MEDAL OF COLOMAN, COLMAN, 1470–1532

DESIGNS FOR ARMOUR BY ALBERT DURER, 1517

On page 196 of Sir John Smith’s Instructions, Observations, and

Orders Militarie, the author strongly objects to the discarding of the
arm and leg defences which was advised by other authorities. He
insists that these limbs are as important as the “breste, belly, and
backe,” and should be adequately protected. His opinions are also
held by Marshal Maurice of Saxe in his Rêveries, quoted above.
Edward Ludlow, at the battle of Edgehill, 1642,[127] was
dismounted in getting through a hedge, and says: “I could not
without great difficulty recover on horse-back again being loaded
with cuirassiers arms as the rest of the guard were also.”
It would be superfluous to mention the different occasions on
which unhorsed knights were captured or killed through their
inability to remount in battle. Froissart in describing the battle of
Poitiers says that when once dismounted men could not get up
again, and other historians bear equal witness of the disadvantage
of armour when unmounted; and the Sieur de Gaya, who has been
so often referred to in these pages, writing in 1678, says in his Traité
des Armes, page 60: “Ils n’avoient trop de tort à mon avis d’équiper
ainsi leurs chevaux parce qu’un Cavalier armé n’est plus propre à
rien quand il est démonté.”
Although this may be taken as a reason put forward by the writer
for more armour for man and horse, it shows at the same time that
the fully armed man was considered to be comparatively useless
when unhorsed, as the Spanish proverb ran: “Muerto el Cavallo,
perdido el hombre d’armas.”
It may be somewhat of a surprise to learn that the present-day
equipment is but little lighter than that of the fifteenth century. The
Under Secretary for War, speaking in the House of Commons on
November 28th, 1911, stated that the infantry soldier marched on
an average thirty miles a day during the manœuvres, carrying 59 lb.
11 oz. of equipment and kit. Against this we may place the weight of
some suits of foot-soldiers’ armour of the sixteenth century, which
weigh with the helmet at the outside 25 lb.; leaving therefore a wide
margin for underclothes and weapons. And this comparison of
weight carried is even more interesting when considering the cavalry
equipment, as will be seen from the annexed table on the opposite
page.
Of course all these figures represent “dead weight”; and here we
are brought back to one of those fundamental rules of good
craftsmanship—the recognition of “Convenience in Use.”
Even in the Golden Age of armour, the fifteenth century, the
armourer was hampered by material and by methods of construction
which even the most expert craftsman could not overcome; but
when we reach the period of decadence in the seventeenth century,
the excellence of craftsmanship had deteriorated to an alarming
extent and these difficulties were still greater. The secret therefore of
the weight-carrying powers of man and horse at the present day is
greater convenience in carrying, the scientific distribution of weight,
and a more adaptable material, which when taken together give
greater freedom and greater mobility, even though the actual weight
be the same as the equipment of steel.
The following table gives the weights of typical suits from the
fifteenth century onwards:—
ARMOUR FOR THE JOUST
XV-XVI.—Helms (English). lb. oz.
Barendyne, Great Haseley, Oxon 13 8
Wallace Collection, No. 78 17 0
Westminster Abbey 17 12
Brocas, Rotunda, Woolwich 17 12
Dawtrey, Petworth, Sussex 21 8
Captain Lindsay, Sutton Courtenay, Berks 24 14
1518. Madrid, A, 37 41 9
Suits.
1520. Tower, II, 28, for fighting on foot 93 0
{ man 79 0
1530 (circ.). Madrid, A, 26
{ horse 79 0
1590. Tower, II, 9, man 103 0
 WAR HARNESS
1439. Musée d’Artillerie, Paris, G, 1, man and horse 163 0
{ man 64 13
1514. Tower, II, 5
{ horse 69 3
1588. Musée d’Artillerie, G, 80, man 92 6
1590. Tower, II, 10 79 0
1590. Tower, II, 12 55 8
1612. Tower, II, 18 77 14

CAVALRY
1450 1875 1909

G, 1, Musée d’Artillerie, Paris.

Man, about 140 lb. }
Armour for man and horse, 163 }
333 lb.
lb.[128]
Arms, clothes, saddlery, etc., }
about 30 lb.

British Household Cavalry 308 lb. }

” Heavy ” 280 lb. }
246 lb.
” Medium ” 266 lb. } [130]

259 lb.
” Light ” [129] }

German Cuirassier 334 lb.

All the above are Service equipment,
including rider and saddlery.

INFANTRY
1550 1875 1911

106–8, Rotunda, Woolwich, Maltese

Suits.
Half-armour and helmet, 25 lb. } 40
Clothes and arms, about 15 lb. } lb.
British Infantry.
52 lb. 59 lb. 11 oz.
Service equipment, including arms [129] [131]
 F O OT N OT E S :
[120] Arch. Journ., LX.
[121] Archives de Bruxelles, Cat. Mus. Porte de Hal, 1885.
[122] Jahrbuch des Kunsthist. Sammlungen, II, 1032.
[123] Arch. Journ., IV.
[124] Archæologia, LVII.
[125] Antiquarian Repertory, IV.
[126] Johnes’ edition, I, 449.
[127] Ludlow’s Memoirs, Firth, I, 44.
[128] Catalogue of the Museum.
[129] Sir G. P. Colley, K.S.I., Encyc. Brit., 1875.
[130] Col. F. N. Maude, Encyc. Brit., 1910.
[131] Morning Post, December 9, 1911.
 THE ARMOURERS’ COMPANY OF THE CITY OF LONDON,
ARMOURERS’ HALL, COLEMAN STREET, E.C.

A
t the present day this Company is combined with that of the
Braziers, but this combination only dates from the beginning of
the eighteenth century, when it had ceased to deal with the
making of armour and was more concerned with other branches of
the craft of the metal-worker. The objects of the craft-gild of the
armourers were the same as all those of like nature in the Middle
Ages. Members were protected from outside piracy of methods and
trade-marks, they were cared for in body when ill or incapable of
working, and in soul by masses and religious exercises.
An important detail in the organization of
these craft-gilds and one sadly lacking in
modern trade combinations was the
examination and approval of the members’
work by the gild-masters. In this way was the
craftsman encouraged to produce good work,
and also the purchaser was protected against
inferior workmanship. A reference to the
Appendices B, K will exemplify this, for in
these two instances alone we find that
careless work is condemned by the Company.
In the document of the reign of Edward II it is
noted that “old bascute broken and false now
Fig. 62. Arms of the Armourers’
Company of London.
newly covered by men that nothing
understood of ye mystery wh. be put in pryvie
places and borne out into ye contrye out of ye said Citye to sell and in
ye same citie of wh. men may not gaine knowledge whether they be
good or ill of ye wh. thinge greate yill might fall to ye king and his
people.”
Again, under Charles I, in the appeal of the Company to the Crown,
leave to use the mark is requested “because divers cutlers, smythes,
tynkers & other botchers of arms by their unskillfulness have utterly
spoiled many armes, armours, &c.”
The Company seems to have existed during the reign of Edward II,
but was not then incorporated, and with the exception of the
document transcribed in Appendix A, there is but little evidence of
their existence before the date of 31st Henry VI, in which year a
Charter of Incorporation was granted. This deals mostly with
questions relating to religious observances, the gild-chapel and like
matters. A report to the Court of Aldermen, dated 20th Eliz. (1578), as
to right of search for armour, etc., states that “the Armourers did
shewe us that in Kinge Edward the Second his time, the Lord Maior
and his bretheren did then graunte the serche unto the Armourers.”
As has been noticed before, the fact that armour plates were
expensive and difficult to forge will account for the scarcity of
examples of the defensive equipment up to the sixteenth century.
Either the suit was remade or, having been cast aside, it was utilized
by the common soldier as well as might be. It was only when the age
of the firearm was reached that armour was left in its perfect state
and was not improved upon. We have therefore but little to show
whether the English armourers of the fourteenth and fifteenth
centuries were more or less expert than their foreign rivals, but, from
other examples of metal-work that remain to us, we are forced to the
conclusion that the foreigner was our superior. At the same time we
find on more than one occasion that the English armourer claims to
equal his foreign rival; but whether these claims were ever proved we
are unable to decide without actual examples of the craft work or
documentary evidence. In Appendix J is printed an appeal from Capt.
John Martin in 1624 for leave to import German “platers” to teach
English armourers, with the hope that this will establish a home trade
and will stop the import of foreign work. At the same time the very
fact of this request shows that the craft in England in the reign of
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookgate.com