Deep Learning Applications Of Shortrange Radars

Avik Santra download

https://ebookbell.com/product/deep-learning-applications-of-
shortrange-radars-avik-santra-32739326

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Novel Financial Applications Of Machine Learning And Deep Learning

Algorithms Product Modelling And Applications Mohammad Zoynul Abedin

https://ebookbell.com/product/novel-financial-applications-of-machine-
learning-and-deep-learning-algorithms-product-modelling-and-
applications-mohammad-zoynul-abedin-48060318

Current Applications Of Deep Learning In Cancer Diagnostics Jyotismita

Chaki

https://ebookbell.com/product/current-applications-of-deep-learning-
in-cancer-diagnostics-jyotismita-chaki-48050224

Current Applications Of Deep Learning In Cancer Diagnostics Jyotismita

Chaki

https://ebookbell.com/product/current-applications-of-deep-learning-
in-cancer-diagnostics-jyotismita-chaki-48862674

Multidisciplinary Applications Of Deep Learningbased Artificial

Emotional Intelligence Chiranji Lal Chowdhary

https://ebookbell.com/product/multidisciplinary-applications-of-deep-
learningbased-artificial-emotional-intelligence-chiranji-lal-
chowdhary-47573630
Multidisciplinary Applications Of Deep Learningbased Artificial
Emotional Intelligence Chiranji Lal Chowdhary

https://ebookbell.com/product/multidisciplinary-applications-of-deep-
learningbased-artificial-emotional-intelligence-chiranji-lal-
chowdhary-49036524

Applications Of Deep Learning In Electromagnetics Teaching Maxwells

Equations To Machines Maokun Li

https://ebookbell.com/product/applications-of-deep-learning-in-
electromagnetics-teaching-maxwells-equations-to-machines-maokun-
li-48490672

Handbook Of Deep Learning Applications Balas Valentina Emilia Roy

https://ebookbell.com/product/handbook-of-deep-learning-applications-
balas-valentina-emilia-roy-32704804

Blockchain Of Things And Deep Learning Applications In Construction

Digital Construction Transformation Faris Elghaish

https://ebookbell.com/product/blockchain-of-things-and-deep-learning-
applications-in-construction-digital-construction-transformation-
faris-elghaish-43877436

Applications Of Machine Learning And Deep Learning On Biological Data

1st Edition Faheem Masoodi

https://ebookbell.com/product/applications-of-machine-learning-and-
deep-learning-on-biological-data-1st-edition-faheem-masoodi-49150142
 Deep Learning Applications of
Short-Range Radars

Santra: “fm” — 2020/8/21 — 10:09 — page i — #1

For a complete listing of the Radar Series, turn to the back of this book.

Santra: “fm” — 2020/8/21 — 10:09 — page ii — #2

 Deep Learning Applications of
Short-Range Radars
Avik Santra
Souvik Hazra

Santra: “fm” — 2020/8/21 — 10:09 — page iii — #3

Library of Congress Cataloging-in-Publication Data
A catalog record for this book is available from the U.S. Library of Congress.

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library.

Cover design by John Gomes

ISBN 13: 978-1-63081-746-6

© 2020 ARTECH HOUSE

685 Canton Street
Norwood, MA 02062

All rights reserved. Printed and bound in the United States of America. No part of this book
may be reproduced or utilized in any form or by any means, electronic or mechanical, including
photocopying, recording, or by any information storage and retrieval system, without permission
in writing from the publisher.
All terms mentioned in this book that are known to be trademarks or service marks have been
appropriately capitalized. Artech House cannot attest to the accuracy of this information. Use of a
term in this book should not be regarded as affecting the validity of any trademark or service mark.
Many product and company names that occur in this book are trademarks or registered
trademarks of their respective holders. They remain their property, and a mention does not imply
any affiliation with or endorsement by the respective holder.

10 9 8 7 6 5 4 3 2 1

Santra: “fm” — 2020/8/21 — 10:09 — page iv — #4

Contents

Preface xiii

1 Introduction to Radar Signal Processing 1

1.1 Types of Radar 3
1.1.1 CW Radar 3
1.1.2 Modulated CW Radar 3
1.1.3 Impulse UWB Radar 7
1.1.4 Other Short Range Radars 7
1.2 Waveform Design and Ambiguity Function 9
1.3 System Concept 13
1.4 Target Model 16
1.5 3D Data-cube Processing 20
1.5.1 1D Processing 22
1.5.2 2D Range Doppler Images 25
1.5.3 2D Range Cross-Range Images 28
1.6 Detection Strategy and Clustering 29
1.6.1 Detection Algorithm 29
1.6.2 Clustering 32
1.7 Parameter Estimation and Cramer-Rao Bound 36
1.8 Tracking 38
1.8.1 Track Management 38
1.8.2 Track Filtering 39
1.9 Applications of Short-Range Radar 48
1.10 Problems 53
References 54

Santra: “fm” — 2020/8/21 — 10:09 — page v — #5

vi Deep Learning Applications of Short-Range Radars

2 Introduction to Deep Learning 59

2.1 Perceptron 60
2.2 Multilayer Perceptron 65
2.2.1 Training 67
2.2.2 Activation Functions 70
2.2.3 Optimizers 73
2.2.4 Types of Models 75
2.3 Convolutional Neural Networks 75
2.3.1 Convolution Layer 75
2.3.2 Popular Architectures 78
2.3.3 Transfer Learning 86
2.4 Long Short Term Memory 87
2.5 Autoencoders 90
2.6 Variational Autoencoder 92
2.7 Generative Adversarial Network 95
2.8 Robust Deep Learning 100
2.9 Problems 101
References 102

3 Gesture Sensing and Recognition 105

3.1 Introduction 105
3.1.1 Related Work 106
3.2 Gesture Sensing/Detection 108
3.3 Micro-Gestures 111
3.3.1 System Parameters 111
3.3.2 Micro-Gesture Set 112
3.4 2D All CNN-LSTM 112
3.4.1 Architecture and Learning 113
3.4.2 System Evaluation 116
3.5 3D CNN and Pseudo-3D CNN 117
3.5.1 3D CNN Architecture and Learning 118
3.5.2 Pseudo-3D CNN Architecture and Learning 118

Santra: “fm” — 2020/8/21 — 10:09 — page vi — #6

 Contents vii

3.6 Meta-Learning 119

3.6.1 Architecture and Learning 121
3.6.2 System Evaluation 124
3.7 Macro-Gestures 131
3.7.1 System Parameters 131
3.7.2 Macro-Gesture Set 131
3.8 Unguided Attention 2D CNN-LSTM 133
3.8.1 Architecture and Learning 133
3.8.2 System Evaluation 135
3.9 Future Work and Direction 137
3.10 Problems 137
References 138

4 Human Activity Recognition and Elderly-Fall

Detection 141
4.1 Introduction 141
4.1.1 Related Work 142
4.2 Preprocessing for Feature Image 143
4.2.1 Fast-Time FFT 143
4.2.2 Coherent Pulse Integration 144
4.2.3 MTI Filtering 144
4.2.4 Adaptive Detection Thresholding 145
4.2.5 Euclidean Clustering 145
4.2.6 Kalman Filter 145
4.3 Input Feature Images 146
4.3.1 Range Spectrogram 147
4.3.2 Doppler Spectrogram 147
4.3.3 Video of RDI 149
4.4 Human Activity Data Set 149
4.5 DCNN Activity Classification 150
4.5.1 Architecture and Learning 150
4.5.2 Results and Discussion 153

Santra: “fm” — 2020/8/21 — 10:09 — page vii — #7

viii Deep Learning Applications of Short-Range Radars

4.6 Bayesian Classification 155

4.6.1 Integrated Classifier and Tracker 158
4.6.2 Results and Discussion 162
4.7 Fall-Motion Recognition 166
4.7.1 Architecture and Learning 168
4.7.2 Deformable CNN 169
4.7.3 Loss Function 171
4.7.4 Results and Discussion 173
4.8 Future Work and Directions 174
4.9 Problems 176
References 176

5 Air-Writing 179
5.1 Introduction 179
5.2 Radar Network Placement 181
5.3 Preprocessing 183
5.3.1 Coherent Pulse Integration 183
5.3.2 Moving Target Indication Filtering 183
5.3.3 Target Detection and Selection 184
5.3.4 Localization with Trilateration 185
5.3.5 Trajectory Smoothening Filters 187
5.4 Setup and Characters 188
5.4.1 Character Set 188
5.4.2 System Parameters 188
5.4.3 Setup and Data Acquistion 188
5.5 LSTM 190
5.5.1 Architecture 191
5.5.2 Loss Function: CTC 194
5.5.3 Design Considerations 196
5.5.4 Performance Evaluation 196
5.6 Deep Convolutional Neural Networks 197
5.6.1 Architecture 197

Santra: “fm” — 2020/8/21 — 10:09 — page viii — #8

 Contents ix

5.6.2 Weight Initialization 198

5.6.3 Learning Schedule 198
5.6.4 Data Augmentation 198
5.6.5 Performance Evaluation 199
5.7 1D CNN-LSTM 200
5.7.1 Architecture 200
5.7.2 Performance Evaluation 200
5.8 Future Work and Directions 201
5.9 Problems 203
References 203

6 Material Classification 205

6.1 Introduction 205
6.1.1 Related Work 206
6.2 Feature Images: Range Angle Images 207
6.3 Deep Convolutional Neural Networks 210
6.3.1 Architecture and Learning 212
6.3.2 Design Considerations 214
6.3.3 Results and Discussion 215
6.4 Siamese Network 217
6.4.1 Architecture and Learning 218
6.4.2 Design Considerations 219
6.4.3 Results and Discussion 220
6.5 Future Work and Directions 222
6.6 Problems 222
References 223

7 Vital Sensing and Classification 225

7.1 Introduction 225
7.2 Vital Signal Fundamentals 227
7.2.1 Preprocessing Steps 230
7.3 Heart Rate Estimation through a Deep-Learning
Approach 234

Santra: “fm” — 2020/8/21 — 10:09 — page ix — #9

x Deep Learning Applications of Short-Range Radars

7.3.1 GAN-Based Data Augmentation 236

7.3.2 Results and Discussions 238
7.4 Adaptive Signal Processing with a Tracking Approach 240
7.4.1 Algorithm 241
7.4.2 Results and Discussion 244
7.5 IQ Signal Evaluation using Deep Learning 247
7.5.1 Deep Learning Architecture 249
7.5.2 Results and Discussion 249
7.6 Future Work and Direction 250
7.7 Problems 251
References 251

8 People Sensing, Counting, and Localization 255

8.1 Introduction 255
8.2 Presence Sensing: Signal Processing Approach 260
8.2.1 Challenges 260
8.2.2 Solution 261
8.3 Presence Sensing: Deep Learning Approach 263
8.3.1 Challenges 263
8.3.2 Solution 265
8.3.3 Results and Discussion 268
8.4 People Counting: Signal Processing Approach 269
8.4.1 Challenges 269
8.4.2 Solution 271
8.5 People Counting: Deep Learning Approach 276
8.5.1 Data Preparation and Processing 279
8.5.2 Solution 281
8.5.3 Results and Discussion 285
8.6 People Detection and Localization: Signal
Processing Approach 286
8.7 People Detection and Localization: Deep Learning
Approach 290
8.7.1 Challenges 290

Santra: “fm” — 2020/8/21 — 10:09 — page x — #10

 Contents xi

8.7.2 Architecture and Learning 292

8.7.3 Results and Discussion 295
8.8 Future Work and Direction 301
8.9 Problems 302
References 302

9 Automotive In-Cabin Sensing 307

9.1 Introduction 307
9.2 Smart Trunk Opening 308
9.2.1 Challenges 308
9.2.2 Solution 309
9.2.3 Results and Discussion 314
9.3 Vehicle Occupancy Sensing 314
9.3.1 Challenges 314
9.3.2 Solution 316
9.4 Federated Learning 319
9.4.1 Challenges 319
9.4.2 Solution 321
9.5 Future Work and Direction 322
9.6 Problems 322
References 323

About the Authors 325

Index 327

Santra: “fm” — 2020/8/21 — 10:09 — page xi — #11

Santra: “fm” — 2020/8/21 — 10:09 — page xii — #12
Preface
Radar has evolved from a complex, high-end military technology into a relatively
simple, low-end solution penetrating industrial, automotive, and consumer
market segments. This rapid evolution has been driven by two main factors:
advancements in silicon and packaging technology that has led to miniaturization,
and growth of computing power that has enabled the use of deep learning
algorithms to tap the full potential of radar signals. The use and applications
of radar technology have grown multifold in recent years. From military and
defense applications, radars have been widely used in automotive applications to
increase safety and facilitate driving in medium- to premium-priced cars, and is
also an important sensor for autonomous cars. For adoption of short-range radars
for several industrial, consumer, and in-cabin automotive applications require
reliable system performance at small-form factor, low power, and low cost.
The advent of deep learning has transformed many fields and resulted
in state-of-the-art solutions in areas such as computer vision, natural language
processing, and speech processing. However, the application of deep learning
algorithms to radars is still by and large at its nascent stage. A radar system
consists of two parts: first, the radar hardware, including the RF transceiver,
waveform generator, receiver unit, antenna and system packaging. State-of-the-
art SiGe and CMOS are candidate technologies for mm-wave short-range radars
and offer flexibility for integration and smaller form-factor. The second part is the
sensing aspect, which relies on signal processing or deep learning algorithms that
parses the radar return echo into meaningful target information that facilitates a
desired application. The goal of this book is to demonstrate and highlight how
deep learning is enabling several advanced industrial, consumer, and in-cabin
applications of short-range radars, which weren’t otherwise possible. The book
illustrates various advanced applications, their respective challenges, and how they
are being addressed using different deep learning architectures and algorithms.
While several applications presented in the book are at a high product readiness
stage, some of them form the basis for further research and development for
product readiness.
The book is laid out as follows: Chapter 1 introduces radar signal processing,
the various types of short-range radars, the basics of waveform designs, target

xiii

Santra: “fm” — 2020/8/21 — 10:09 — page xiii — #13

xiv Deep Learning Applications of Short-Range Radars

models for high-resolution radars, and 3D radar data-cube processing involving

detection, parameter estimation, and tracking algorithms. The chapter concludes
with various advanced industrial, consumer and automotive applications of short-
range radars.
Chapter 2 provides an introduction to deep learning, outlining the history
of neural networks and the optimization algorithms to train them. The chapter
introduces the modern deep convolutional neural network (DCNN), popular
DCNN architectures for computer vision is a field of study dealing with processing
of images. It presents other deep learning architectures such as long-short term
memory (LSTM), autoencoders, variational autoencoders (VAE), and generative
adversarial networks (GANs).
Chapter 3 introduces the application of gesture recognition. We divide them
into macrogesture sensing and recognition, which involve major arm movements
and are performed from a distance to the sensor, as well as microgesture
sensing and recognition, which involves minute subtle finger movements and
are performed close to the sensor. While the macrogesture finds application
in projectors or TVs, the microgesture finds applications in smartphones,
smartwatches, and automotive dashboards.
Chapter 4 introduces the application of human activity recognition. The
chapter presents various DCNN architectures for enabling activity recognition
and comparing their performance. The chapter further introduces the concept
of continuous activity recognition using a combination of tracking and a deep
learning classifier such as LSTM. The chapter end by presents the topic of elderly
fall motion recognition, the various challenges, and how it can be addressed
through deep learning.
Chapter 5 presents the application of air-writing using a network of short-
range radars. Air-writing or air-drawing refers to linguistic characters drawn on
an imaginary board and the system recognizing these characters. Air-writing
finds application in human-machine interface for augmented-reality virtual reality
(AR-VR) and alternate interface mechanisms for desktops.
Chapter 6 presents the application of material classification. The objective
of short-range radar is to classify among everyday objects and materials, and
finds applications in vacuum cleaners and robots. The chapter concludes with
the challenges to be overcome for integration into deployable system solutions.
Chapter 7 presents the application of remote vital sensing using short-range
radars. The chapter presents various challenges with a pure deep learning solution
and introduces a tracking-based solution as a stable solution, and also introduces
a hybrid tracking with deep learning solution as a more reliable and scalable
solution.
Chapter 8 introduces the application of short-range radars in conjunction
with deep learning for human presence, counting, and localization applications.
The chapter presents how deep learning has and is replacing the classical

Santra: “fm” — 2020/8/21 — 10:09 — page xiv — #14

 Preface xv

signal processing pipeline to improve system performance as well as addressing

challenges that weren’t feasible with a classical signal processing pipeline.
Chapter 9 introduces the application of in-cabin sensing. Apart from
automatic cruise control (ACC), blind spot detection (BSD), and parking
assistance, radars are enabling a plethora of in-cabin sensing applications such
as smart trunk opening (STO) applications, child-left-behind applications, and
smart airbag applications. We conclude the chapter with a federated learning
framework, wherein a mechanism to automatically update a deployed deep
learning model in production is outlined.
The book is intended for graduate students, academic researchers, and
industry practioners working with mm-wave radars and radar algorithms. The
book is written for beginners to advanced researchers and assumes sufficient
knowledge of linear algebra and engineering mathematics. Each chapter has
a question section to assess the understanding of the reader. The book covers
conventional deep learning architectures, as well as the adaptations required for
product-ready solution for an application. It covers advanced concepts like meta-
learning, multimodal cross-learning, attention mechanism, federated learning,
deformable convolution, and PointNets. While each chapter is independent of
the others, it is suggested that an early researcher read the first two introductory
chapters before reading any application-specific chapter. This book also provides
an introduction to classical signal processing algorithms in Chapter 1 and basics
of deep learning algorithms in Chapter 2.
The authors would like to thank their respective families: Avik Santra would
like to thank his wife Sudarshana for her constant love, encouragement, and
motivation that helped him shape the book, and his son Abhinava who is a
perpetual source of energy and happiness. The authors would like to acknowledge
and thank their manager, Thomas Finke, for his tremendous support and
encouragement during the period of writing this book. The authors would also
like to thank their colleagues and masters students, who worked together on
various projects. In particular, the authors would like to thank Prachi Vaishnav,
Raghavendran Vagarappan Ulaganathan, Christoph Will, Yogesh Shankar, Jonas
Weiss, Michael Stephan, Muhammed Arsalan, and Rodrigo Hernangomez. The
authors would also like to thank anonymous reviewers for their encouragement,
reviews, and suggestions to improve the book.

Santra: “fm” — 2020/8/21 — 10:09 — page xv — #15

Santra: “fm” — 2020/8/21 — 10:09 — page xvi — #16
1
Introduction to Radar Signal Processing

The use and application of radar technology has grown multifold in recent years
[1]. Radar technology has migrated from military and defense applications to
being standard components in medium to premium cars. Modern automotive
radars increase safety and thus facilitate driving in both human-driven and
automotive-driven cases. Slowly, radar technology has also penetrated into the
industrial and consumer markets and is enabling several new applications.
The design of any radar system consists of two parts. The first part is radar
hardware, including a radio frequency (RF) transceiver, waveform generator,
receiver unit, antenna, and system-in-packaging. The second part is the signal
processing aspect to parse the radar return echo to extract meaningful target
information. Since the invention of integrated circuits, the operating frequency
of transistors has been steadily increasing enabling the realization of circuit blocks
that operate at frequencies up to 1 THz [2]. In parallel, transistors have been
shrinking with more advanced technology nodes, allowing for further integration
[3]. Figure 1.1 highlights the evolution of radar technology used in automotive
applications. Silicon germanium (SiGe) bipolar technology has been the preferred
silicon technology for automotive and industrial mm-wave radar over the last
few years as its performance, cost, and integration level fit superbly into the
application requirements [4,5]. The state-of-the-art SiGe technology has reached
operating frequencies beyond 300 GHz. In [6], a FE BICMOS that can be
used technology is presented with an FT of 250 GHz and a Fmax of 370 GHz
for the SiGe transistors. The present technology also provides 130-nm CMOS
that can be used for the realization of different radar building blocks like PLLs
and DSP. RF CMOS technology also has been shown to be a candidate for
mm-wave radar [7] although the RF performance is not on the same level as of

Santra: “chapter1” — 2020/8/20 — 11:20 — page 1 — #1

2 Deep Learning Applications of Short-Range Radars

Figure 1.1 Technology trend of 77-GHz radar transceivers.

SiGe. CMOS technology offers more digital integration, which is attractive for
performing signal processing on a radar chip. The high operating frequency and
advanced packaging technologies have also allowed the integration of antennas
into a package and in some cases into the silicon die itself. Antenna integration
is essential for reducing the complexity in radar design and for reducing the
overall system cost, allowing penetration of the technology into the industrial
and consumer markets. It has been shown how different antenna configurations
can be integrated into packaging for different field of views that cover various
application requirements [8]. Going with a frequency beyond 100 GHz allows
for the integration of the antenna on silicon [9], which eventually helps to reduce
the cost and size of radar solutions.
This chapter is laid out as follows: we present and introduce the
different types of short-range radars in Section 1.1. In Section 1.2, we present
several waveform design and ambiguity functions of radar waveforms and
their properties. The system concept of a short-range radar, specifically that
of a frequency modulated continuous-wave (FMCW) radar, is presented in
Section 1.3. Section 1.4 presents the radar target models and their canonical
structures of any radar target. In Section 1.5, three-dimensional (3D) radar
data-cube processing is presented. The detection strategies clustering algorithms
required for short-range radars are presented in Section 1.6, and radar target
parameter estimation, namely range, velocity, and angle, are presented in
Section 1.7. A detailed introduction of common tracking algorithms namely
extended Kalman filters and unscented Kalman filters used in short-range

Santra: “chapter1” — 2020/8/20 — 11:20 — page 2 — #2

 Introduction to Radar Signal Processing 3

radars, are presented in Section 1.8. Section 1.9 presents various applications
of short-range radars in industrial, consumer, and automotive sectors.

1.1 Types of Radar

Short- and small-range radars can typically be categorized as continuous-wave
(CW) radar, modulated CW radar, impulse ultrawideband (UWB) radars, and
noise radars. CW radars transmit and receive waveform continuously, whereas
impulse radars transmit short pulses while the receiver is not operating followed
by quiet reception. Thus CW radars require distinct transmit and receive antennas
and need good isolation. The major advantage of CW radars is that the signal
processing at the receiver is performed at low frequency, and hence the sampling
rate requirement is substantially lower compared to the transmit signal, thereby
considerably simplifying the realization of processing circuitry.

1.1.1 CW Radar
CW radar transmits an unmodulated continuous frequency tone and the received
echo signal is processed to estimate a target’s radial velocity by evaluating the
change in phase. This is caused due to a shift in Doppler frequency arising from
the reflection of a moving target. However, CW radar has the disadvantage that
the range information cannot be obtained. Figure 1.2 shows a block diagram of a
CW radar presenting the transmit and the receive chains. However, a coarse range
estimation can be obtained by either pulse-Doppler operation or transmitting two
distinct frequency tones referred as frequency shift keying (FSK).

1.1.2 Modulated CW Radar

There are several modulation patterns that can modulate the transmit signal
in frequency. One of the popular patterns is sawtooth frequency modulation,
wherein frequency is linearly increased over time (upchirp) or decreased over
time (downchirp), and thus are also referred to as linear frequency modulated
(LFM). In the case of upchirp LFM, the frequency is linearly increased as
B
f (t) = f0 + t (1.1)
Tc

Figure 1.2 Block diagram of CW radar.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 3 — #3

4 Deep Learning Applications of Short-Range Radars

where f0 is the ramp start frequency, B is the bandwidth, and Tc is the chirp ramp
time.
The LFM waveform exhibits a near flat frequency response given as

  π
Tc 2
S(f ) = exp −jπ f exp j (1.2)
B 4
Figure 1.3 presents the upchirp LFM waveform, where the transmit (Tx)
signal is an upchirp and the received (Rx) signal from two targets at different range
bins. The received signal from both targets are received as a superimposition of
the two at the reciever, the delay signal t is due to the varying range of the
targets, and the shifted frequency f arises due to the speed of the targets. The
received signal is mixed with the transmit signal followed by lowpass filtered, and
this signal is referred as intermodulation frequency or IF signal.
LFM being a constant amplitude waveform is desired since it allows
operating the power amplifier at saturation and thus with maximum efficiency.
Further, since the chirp time and the bandwidth can be chosen independently, it
offers flexibility to the radar system designer to meet different range and Doppler

Figure 1.3 Illustration of transmit (Tx) upchirp LFM modulated signal and corresponding
received (Rx) signal from two targets at different ranges.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 4 — #4

 Introduction to Radar Signal Processing 5

specifications. One of the biggest advantages of frequency modulated continuous-

wave radar is at the receiver the transmitted waveform is mixed with the received
waveform followed by lowpass filtering, in a process called deramping. The
output of the low-pass filtered intermediate frequency (IF) signal have much
lower bandwidth compared to the transmitted signal that has a large sweep
bandwidth, thus requiring a much lower analog-to-digital converters (ADCs)
and signal processing constraints. This enables linear frequency modulated radars
to be compact, low power and low-cost radar sensors.
Another commonly used waveform is triangular waveform comprising of
upchirp and downchirp in sequence. Figure 1.4 presents the transmit triangular
modulated waveform and the corresponding IF signal at the receiver. The beat
frequency in the IF signal consists of a component due to range and Doppler as
B 2R
fr = .
Tc c
2v
fd = (1.3)
λ
where R, v are the range and velocity of the detected target, respectively, and
λ is the carrier wavelength. The beat frequencies at the IF signal for upchirps and
downchirps are given as
fbu = fr − fd
fbd = fr + fd (1.4)

Figure 1.4 Illustration of transmit (Tx) triangular waveform with sequence of upchirp LFM and
downchirp LFM and corresponding IF signal.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 5 — #5

6 Deep Learning Applications of Short-Range Radars

Thus, once the beat frequencies are determined the range and velocity of
the target can be estimated as

cTc
Rest = (fbd + fbu )
4B
λ
vest = (fbd − fbu ) (1.5)
4

This shows that in the range-velocity plot, the estimates could be calculated
by the intersection of two lines given by (1.5). However, it is easy to see that
for two targets with a different range, velocity would mean four intersecting
lines, resulting in ambiguous-range velocity assignments. To solve this problem,
staggered pulse repetition times (PRTs) with varying chirp time in subsequent
waveform are transmitted, resulting in different slopes and facilitating unique
range-velocity assignments.
In the case of a frequency modulated continuous wave, there are two critical
aspects that determine the performance of such radar systems. One is the ramp
linearity; any deviation from a linear ramp results in range estimation errors.
The other critical aspect is the Tx-Rx leakage since the transmit and receiver are
operating at the same time. The Tx-Rx leakage affects the first few range bins and
also limits the radar’s maximum detectable range since the dynamic range of the
ADC is severally affected if it is not accounted for.
There are various frequency modulation patterns that are used in practice
and in most cases require different architecture design and different processing
pipelines. Apart from sawtooth and triangular frequency modulation, some of the
other common frequency modulation patterns used in modern short-range radars
are stepped frequency or interrupted frequency modulated waveforms. Stepped
frequency modulation provides a piecewise approximation of the sawtooth
linear chirp signal. The stepped frequency modulation waveform uses the same
homodyne process as sawtooth frequency modulation but has much lower phase
noise and has less constraints on ramp linearity. However, the stepped frequency
modulation codes suffer from Doppler tolerance compared to their sawtooth
equivalents. Interrupted frequency modulation tries to combine the advantage
from pulse radar and frequency modulated continuous-wave radar. A linear
sawtooth waveform is transmitted but to avoid the Tx-Rx leakage, the transmit
signal is switched off intermittently when the receiver is receiving the echoes.
However, the interrupted frequency modulated poses a strict timing constraint
of the transmitter since the transmission time should be less than the round-trip
propagation time of the maximum detectable target.
Apart from LFM, nonlinear frequency modulation (NLFM) can be used as
transmit waveforms, which offers excellent ambiguity function properties such as
lower sidelobe levels in a trade-off to complexity. Some of the nonlinear frequency

Santra: “chapter1” — 2020/8/20 — 11:20 — page 6 — #6

 Introduction to Radar Signal Processing 7

modulations are
• Sine-based NLFM

B 2t
f (t) = fc + arcsin −1 (1.6)
π Tc
• Symmetrical NLFM


7 
f (t) = fc + B a(t) + Kn sin(2π na(t))
n=1


7 
· b(t) + Kn sin(2π nb(t)) (1.7)
n=1

where a(t) = T2tc , b(t) = 1 − T2tc . K1 = −0.1145, K2 = 0.0396, K3 =

−0.0202, K4 = 0.0118, K5 = −0.0082, K6 = 0.0055, K7 = −0.0040.

1.1.3 Impulse UWB Radar

Impulse radar transmits short pulses and determines distance by measuring the
time delay between the transmitted and returned signal. Impulse radar transmits
ultrawideband, short-pulse waveforms. The time delay of the received pulse
determines the range of the target and the peak of the spectrum determines
the Doppler velocity. The pulsewidth determines the range resolution and the
Doppler resolution is typically poor in such systems. However, such systems do
not suffer from Tx-Rx leakage since the transmitter and the receiver do not operate
at the same time.
UWB radar systems transmit signals across a much wider frequency than
conventional radar systems and are usually very difficult to detect by other radio
devices in their vicinity. The transmitted waveform is typically a Gaussian pulse
or a derivative of a Gaussian pulse since a rectangular pulse waveform would
change its shape due to the spectral limited response of the electronics circuitry
apart from the target itself. The transmitted signal is significant for its very light
power spectrum, which is lower than the allowed unintentional radiated emissions
for electronics. The most common technique for generating a UWB signal is to
transmit pulses with very short durations (less than 1 nanosecond). The spectrum
of a very narrow-width pulse has a very large frequency spectrum approaching
that of white noise as the pulse becomes narrower. These very short pulses need
a wider receiver bandwidth compared to conventional radar systems.

1.1.4 Other Short Range Radars

Compared to frequency modulated and impulse radars, there are several other
radars such as orthogonal frequency modulated (OFDM) radar, which works

Santra: “chapter1” — 2020/8/20 — 11:20 — page 7 — #7

8 Deep Learning Applications of Short-Range Radars

on the principle of a communication transceiver. In the case of an OFDM

radar, the time and frequency sensitivity of OFDM waveforms are utilized to
estimate the range and velocity of the target. The cyclic prefix in the OFDM
transmission converts the linear convolution channel into circular convolution,
thus easing the receiver processing to just a scalar division operation to estimate
either the unknown communication data or unknown channel/target response.
The maximum unambiguous range in the case of OFDM radars is determined
by the cyclic prefix length. OFDM waveforms are typically beneficial for range-
extended target responses. OFDM radars can be operated in a pure radar sensing
mode, transmitting digital pseudorandom sequence as a data stream, or in a
radar-communication mode; that is, the radar sensing symbols can be embedded
within the time-frequency resource block with a communication data stream.
Another type of radar is noise radar, which transmits a pseudorandom
signal directly from a noise-generating high-frequency source. The echo detection
is based on optimal reception and correlation between the transmitted signal
and the received noise waveform radar returns. Implementation of noise radar
requires advanced components such as efficient noise waveform transmitters,
digital correlation receivers, and wideband antennas.
Other short-range radars, such as UWB noise radar, also exists, where
the transmitting waveform is a burst of noise sequence. In the case of such
radar systems, transmission of the radar cannot be determined by measurement
from the outside of the radar since the transmit pulses do not differ much from
environmental noise, and thus are useful for interference mitigations. Table 1.1
summarizes different short-range radars and their properties [10].
Further, depending on the Tx-Rx position, a radar can be classified as
monostatic if the transmit and receive antennas are located slightly apart but
from the target can be viewed as approximately the same, and bistatic if the
transmit and receive antennas are located at different locations.
Table 1.1
Different Short-Range Radar Waveforms and Their Key Properties

Radar Waveform Properties

Continuous wave Can estimate only Doppler

Frequency shift keying Can estimate only nonzero Doppler targets, coarse
range estimates
Linear FM (LFM) Compact and lowcost due to low bandwidth IF,
high peak sidelobes
Nonlinear FM (NLFM) Tradeoff range resolution for lower sidelobes
compared to LFM
UWB impulse Ultrahigh range resolution, poor Doppler resolution
OFDM High ADC requirements, faciliates radar-comm operations
Noise Complex hardware circuitry, no interference effects

Santra: “chapter1” — 2020/8/20 — 11:20 — page 8 — #8

 Introduction to Radar Signal Processing 9

Irrespective of the radar type, the radar range equation provides means to
the radar designer to theoretically calculate the maximum range, set transmit
power, and transmit/receive antenna gain for detection of a target with a given
radar cross section (RCS) through a radar link budget. Given the following radar
parameters and notations:


 Pt : transmit power




Pr : receive power
Gt : transmit antenna gain



 Gr : receive antenna gain


σ : radar cross section (RCS)

The received power at the receiver after reflection from a target with RCS
σ and at range R can be expressed as
Pt Gt σ
Pr = (1.8)
(4π R 2 )2
Gr λ2
The effective receive antenna aperture is given as Ar = 4π , thus the
effective received power can be expressed as

Pt Gt Gr σ λ2
Pr = (1.9)
(4π)3 R 4
Therefore, the maximum radar range can be provided as

1/4
Pt Gt Gr σ λ2
Rmax = (1.10)
(4π)3 Prmin

where Prmin is the minimum received power required for target detection; in
typical cases it is set 3 to 5 dB higher than the noise floor.

1.2 Waveform Design and Ambiguity Function

Assuming a point target, the received signal after reflection from the target can
be modeled as
sr (t) = ρs(t − τa ) exp j2πfa t (1.11)
where s(t) is the transmit waveform and ρ is the point target radar cross
section and scaling factors due to path loss and antenna gains, and so forth,
(τa , fa ) represents the target’s actual delay and Doppler components. Thus after
receiver convolution processing with filter response h(t), the output signal is

Santra: “chapter1” — 2020/8/20 — 11:20 — page 9 — #9

10 Deep Learning Applications of Short-Range Radars

represented as
y(t) = sr (τ )h(t − τ )d τ (1.12)

Therefore, the frequency spectrum of the output signal can be expressed as

Y (f ) = ρ H (f )S(f − fa ) exp(j2π f τa )df (1.13)

The output SNR at time t = τa assuming white noise can thus be written as
2
ρ 2 H (f )S(f )df
SNRout (t = τa ) = (1.14)
N0 |H (f )|2 df
where N0 is the noise power.
By Cauchy-Schwarz inequality, the numerator follows the following
inequality
2
H (f )S(f )df ≤ |H (f )|2 df |S(f )|2 df (1.15)

where the equality follows when H (f ) = S ∗ (f ), which means the optimal

receiver filter processing for maximizing the output SNR corresponding to a
point target with an assumption of white noise is a matched filter, where the
received signal sr (t) is mixed with the transmit waveform followed by lowpass
filtering (representing the integration function in (1.12)). The signal power at
t = τa is equal to | |S(f )|2 df |2 and for a fixed frequency bandwidth this is
maximized by a flat transmit frequency spectra, such as LFM waveform.
However it must be noted that in the case of an extended target or colored
noise interference, the conventional matched filter is no longer the optimal
receiver processing, and the flat frequency spectra transmit waveform is also not
the optimal waveform in terms of output detection SNR.
The pulse compression gain [11] denotes the ratio of signal-to-noise ratio
(SNR) improvement achieved by the receiver filter processing and is defined as
SNRout (t = τa )
G= (1.16)
SNRin
where input SNR, SNRin can be written as
ρ 2 (1/Tc ) |S(f )|2 df
SNRin = (1.17)
N0 B
On substituting SNRout and SNRin , (1.16) for compression gain can be
simplified to
Tc
G = Tc B = (1.18)
τB

Santra: “chapter1” — 2020/8/20 — 11:20 — page 10 — #10

 Introduction to Radar Signal Processing 11

where τB denotes the compressed pulsewidth at the output of the filter

matched to the transmit waveform. This shows the effect of the matched filtering
as a compression gain, which is equal to the time-bandwidth product. Thus the
output SNR can be increased by either increasing the time or the bandwidth of
the transmit waveform, while retaining the transmit power. This is an important
design consideration for the radar system designer since the transmit power is
typically limited by regulations to avoid interferences to other radio devices.
The other important aspect in radar signal processing is the Woodward
ambiguity function, χ(τ , ν), which is a two-dimensional time delay and Doppler
frequency response of the received radar signal from a point target due to a receiver
matched filter. The Woodward or self-ambiguity function is determined by the
properties of the transmitted waveform irrespective of interference, noise, or target
scenario (multitarget or spread targets). The Woodward ambiguity function of
any waveform can be defined as
∞
χ(τ , ν) = s(t)s ∗ (t + τ )e j2πνt dt (1.19)
−∞

where s(t) is the transmitted waveform, and (τ , ν) are the mismatched delay and
Doppler shifts.
In the case of basic FMCW waveform, the Woodward ambiguity function
has a closed-form expression as follows






τ τ τ
χ(τ , ν) = 1 − sinc πTp ν ∓ B 1− (1.20)
Tp Tp Tp

where sinc(x) = sin(x)/x. Figure 1.5 presents the 3D woodward ambiguity

function and the corresponding 2D contour of the ambiguity function for an
LFM waveform. The observed mainlobe is actually part of a delay-Doppler ridge
that exhibits a gradual roll-off from the peak at (0,0). The existence of this ridge is
why LFM is also referred to as a Doppler-tolerant waveform, since an appreciable
Doppler shift induces little SNR loss relative to the peak.
The Woodward ambiguity function of a rectangular pulse waveform, whose
pulse width is Tc can be expressed as




|τ | |τ |
χ(τ , ν) = 1 − sinc πνTc 1 − (1.21)
Tc Tc
In the case of a pulse train of FMCW waveform, the ambiguity function is
given as

∞ 
M 
M
χ fmcw (τ , ν) = s(t − mTp )s(t − m Tp − τ )
−∞ m=1 m =1

Santra: “chapter1” — 2020/8/20 — 11:20 — page 11 — #11

12 Deep Learning Applications of Short-Range Radars

Figure 1.5 (a) 3D Woodward ambiguity function of LFM waveform, (b) 2D contour of Woodward
ambiguity function of LFM waveform.

P(t − mTp )P(t − τ − m Tp )e i2πνt dt

where M is the train of consecutive chirps transmitted by the radar sensor. The
Woodward ambiguity function has several properties, such as

1. Maximum at (0,0); that is,

χ(τ , ν) ≤ χ (0, 0) = 1 ∀(τ , ν)

2. Constant volume; that is,

∞ ∞
χ (τ , ν)2 d τ d ν = 1
−∞ −∞

Santra: “chapter1” — 2020/8/20 — 11:20 — page 12 — #12

 Introduction to Radar Signal Processing 13

3. Symmetric at origin; that is,

χ (−τ , −ν) = χ (τ , ν)
4. Linear FM effect; that is,
2
u(t) =⇒ χu (τ , ν), then u(t)e jπkt =⇒ χu (τ , ν − kτ )
The Woodward ambiguity function has different interpretations among
radar practitioners. One school of thought is that the ideal ambiguity function
is a thumback delta dirac function, and is expressed as χ (τ , ν) = δ(τ , ν) which
means the matched filter would respond to a correct estimate of range and
Doppler. The search for an ideal thumback response for the ambiguity function
has lead to several waveform research such as Barker code, Golay complementary
pairs, Frank Zadoff Chu codes, and Gold sequence. The other school of thought
is that such response is not usually desirable since a nonzero Doppler would not
remove the target response along the range transform. Thus a range-Doppler
coupling effect, as exhibited by LFM waveform, is preferred. As a result, the
ambiguity function presents an important tool to study the waveform design and
properties suitable for radar sensing functions.
Apart from the narrowband Woodward ambiguity function, there are several
variants that exist, such as the wideband ambiguity function, spread ambiguity
function, pulse train ambiguity function, and expected ambiguity function. The
general cross ambiguity function can be defined as

χcross (τh , fh ; τa , fa ) = s(t; τa , fa ) g ∗ (t; τh , fh ) dt

= S(f ; τa , fa ) G ∗ (f ; τh , fh ) df (1.22)

for the cross ambiguity function, where (τh , fh ) are the hypothesized parameters in
the receiver and (τa , fa ) are the true range and Doppler parameters. s(t), S(f ) are
the time domain and frequency response of the transmitted waveform, g (t), G(f )
are the time and frequency response of the mismatched signal. The above equation
can be extended to derive a closed-form expression for a cross ambiguity function
in presence of a point target. It can also be extended to derive the ambiguity
function in the case of the colored-noise matched filter where receiver filter g (t)
is not matched to the transmit waveform. This will be a case when the noise is
not white, thus requiring a prewhitening operation provided by the colored-noise
matched filter.

1.3 System Concept

In this section, we describe the system concept of a typical short-range FMCW
radar. Figure 1.6 presents the block diagram of a typical short-range FMCW radar

Santra: “chapter1” — 2020/8/20 — 11:20 — page 13 — #13

14 Deep Learning Applications of Short-Range Radars

Figure 1.6 Functional block diagram of short range FMCW radar.

with 1 transmit, 1 receive antenna. The radar features power amplifiers, low-
noise amplifiers, voltage-controlled oscillators, and analog-to-digital converters
(ADCs). The voltage-controlled oscillator module generates a linear frequency-
modulated continuous-wave signal, which is referenced through a local oscillator.
The generated LFM waveform is then amplified by the power amplifier before
being transmitted from the antenna. The receive antenna receives the target-
distorted transmit waveform, which is then amplified by the low-noise amplifier.
This receiver amplified RF signal is then mixed with the transmitted signal
followed by a lowpass filter to produce the beat intermediate frequency. The
intermediate frequency signal is then passed to the ADC for subsequent
processing. As mentioned earlier, since the intermediate frequency has low
bandwidth, the requirements for ADC and signal processing are drastically
lower compared to sampling the transmit waveform with large bandwidths, thus
enabling FMCW radars to be compact, small, and low-cost solutions.
Figure 1.7(a) depicts the de-ramping operation at the receiver. sTX (t) and
sRX (t) refers to the transmit and corresponding received chirp, respectively. The
round-trip propagation delay τ = 2R/c gets translated to intermediate frequency
after mixing at the receiver. And thus spectral analysis along the chirp provides the
range estimation of the targets in the radar’s field of view. The swept bandwidth
B determines the range resolution as δR = c/2B. The maximum unambiguous
range turns out to be Rmax = Ns δR, where Ns is the number of transmit frequency
steps. The ADC output along a single chirp is referred to as fast time in the
literature.
Figure 1.7(b) shows the frame structure of FMCW radar. The chirp
duration Tc determines the maximum detectable unambiguous Doppler

Santra: “chapter1” — 2020/8/20 — 11:20 — page 14 — #14

 Introduction to Radar Signal Processing 15

Figure 1.7 (a) Deramping processing of FMCW signal, and (b) frame structure of FMCW radar.

fdmax = 1/2(Tc + Ti ), whereas Ti denotes ignored time duration, which

represents the time for resetting the start frequency, DAC, and ADC sampling.
The time duration between two chirps is referred to the PRT or pulse repetition
interval (PRI) and is given as PRT = Tc + Ti . The velocity content of the target
at a range bin causes phase change across multiple chirps at the target’s range bin.
Thus the spectral estimation across chirps provides the velocity information of
the target. The collection of consecutive chirps used for coherent integration is
referred to as the frame or dwell, and represents how often target parameters are
estimated or updated. The coherent processing interval (CPI), which is the time
duration in a frame, determines the Doppler resolution δfd = 1/2CPI. The time
samples along the chirps within the coherent processing interval is referred to as
slow time in the literature.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 15 — #15

16 Deep Learning Applications of Short-Range Radars

Correct raw data acquisition as the first step is ensured by setting the start
time and sampling frequency of the ADC so that the required NS number of
DAC/ADC samples are equally distributed within the ramp start and end. For
each frame having NRX number of Rx channel, a three-dimensional data cube
∈ C NS ×NC ×NRX containing the complex-valued baseband signals is obtained.
The first dimension contains all samples per chirp (fast-time) for range estimation,
the second dimension belongs to the different chirps per frame (slow-time) for
velocity estimation, and the third dimension corresponds to the NRX receive
antennas for angle of arrival (AoA) estimation.
In the case of multiple-input, multiple-output (MIMO) radar, it is
important to maintain orthogonality at the receive among transmit waveforms to
ensure independent data channel streams. By enabling MIMO configuration,
the virtual channels can be computed as a 2D convolution of the transmit
spatial and recieve spatial locations thus increasing the angular resolution and
probability of detection. The orthogonal transmit waveform can be achieved by
distributing chirps across time, frequency, and codes. Figure 1.8 depicts the time
division multiplexing (TDM), frequency division multiplexing (FDM) and slow-
time code division multiplexing (CDM) scheme [12]. Time-division multiplexed
MIMO radar is the simplest configuration and is achieved by transmitting the
same chirp in a round-robin fashion across all transmit antennas. However,
this simple configuration has the maximum unambiguous Doppler limited by
the inverse of the number of transmit antennas. In the frequency division
multiplexing configuration, each transmit antenna transmits a different part
of the frequency sweep. This configuration requires different voltage-controlled
oscillators for each transmit channel and although it preserves the maximum
unambiguous Doppler it suffers a loss of range resolution for each transmit
waveform. On the other hand, slow-time CDM makes use of all the time-
frequency resources, thus retaining the maximum unambiguous Doppler and
also the range resolution of a single-input, single-output (SISO) FMCW radar,
while at the same time increasing the number of virtual antennas leading to
enhanced angular resolutions and also increasing the probability of detection.

1.4 Target Model

For a long-range radar with a narrow frequency sweep, targets appear as a point
target. However, with a wide frequency sweep targets appear as a range-spread
target instead of a point target at the radar receiver. Further radar responses
from targets such as humans are in general spread across Doppler as well due
to the macro-Doppler component of the torso and associated micro-Doppler
components due to hand, shoulder, and leg movements. Thus, human targets
are perceived as doubly spread targets across range and Doppler. The simplest
extended target models are based on the weak scattering model, which utilizes

Santra: “chapter1” — 2020/8/20 — 11:20 — page 16 — #16

 Introduction to Radar Signal Processing 17

Figure 1.8 Chirp distribution across time, frequency, and transmit antennas: (a) time division
multiplexing, (b) frequency division multiplexing, and (c) slow-time code division
mutliplexing.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 17 — #17

18 Deep Learning Applications of Short-Range Radars

the idealized concept of persistent, localized scattering centers. The scattering

centers in this model are the outcome of the geometrical approximation and are
represented as


N
h(τ , ν, θ ) = an δ(τ − τn )δ(ν − νn )δ(θ − θn ) (1.23)
n=1

where h(τ , ν, θ ) is the target impulse response as a function of delay, Doppler and
azimuth angle of each scattering center represented by the parameters an , τn , νn , θn
denote the radar cross section amplitude, delay, Doppler, and angle of the nth
scattering center. The target impulse response is characterized by the super-
position of N such scattering centers. The weak scattering model has limitations
and is ill-behaved for scatterers that include edges, gentle curved plates, and
reentrant structures such as ducts and cavities. Further, the model does not capture
variations in the target impulse response arising from aspect angle changes and
other nongeometric processes. However, the weak scatterer model can be used to
understand the target response due to its geometric properties.
The radar views the space in polar coordinates, thus the 3D Cartesian
position of the kth target and its corresponding unit vector can be represented as

rk = Rk cos(θk ) sin(φk ) Rk cos(θk ) cos(φk ) Rk sin(θk )

uk = cos(θk ) sin(φk ) cos(θk ) cos(φk ) sin(θk ) (1.24)

where θk and φk are the elevation and azimuth angle, respectively, of the target
with respect to the center of the virtual radar array. For estimation of the target in
a 3D space, a MIMO configuration of at least NTX = 2 transmit elements and
NRX = 2 receive elements in an L-shaped linear array fashion with appropriate
spacing is required. This results in a virtual 2 × 2 rectangular array configuration
sufficient for estimation of target’s elevation and azimuth coordinates. As depicted
in Figure 1.9, the 3D positional coordinates of the TX element are denoted as
dmTX , m = 1, ... , NTX and the RX element as dnRX , n = 1, ... , NRX in space.
On assuming far-field conditions, the signal propagation from the TX element
dmTX to a point scatterer p and subsequently the reflection from p to RX element
dnRX can be approximated as 2Rk + dmn , where Rk is the base distance of the
kth scatterer to the center of the virtual linear array and dmn refers to the relative
position of the virtual element to the center of the array. Figure 1.9 presents
the 3D position of the target with reference to mth transmit and nth receive
antennas.
The transmit steering vector can be written as


TX dmTX u(θ , φ)
am (θ , φ) = exp −j2π ; m = 1, ... , NTX (1.25)
λ

Santra: “chapter1” — 2020/8/20 — 11:20 — page 18 — #18

 Introduction to Radar Signal Processing 19

Figure 1.9 3D position of target along with the mth transmit antenna and nth receive antenna.

while the receiving steering vector is


RX dnRX u(θ , φ)
an (θ , φ) = exp −j2π ; n = 1, ... , NRX (1.26)
λ
where λ is the wavelength of the transmit signal.
The received baseband signal from the kth target scatterer can be expressed as
−j2π2uk .rk
s̄RX (t) = ρk e λ anRx (θk , φk )am
Tx
(θk , φk )T s̄TX (t) (1.27)
where ρk represents the composite amplitude contribution due to propagation
path loss, antenna gains, and receiver gains. s̄TX (t) and s̄RX (t) are the transmitted
signals from NTX transmit antennas and the received signal at NRX receive
antennas, respectively. After having estimated the range Rk of the kth target
through spectral analysis along fast time, the angular coordinates of the target,
namely azimuth angle θk and elevation angle φk , can be estimated through
monopulse, Capon, or FFT beamforming algorithms.
Compared to the weak scattering center model, there are several rigorous
models that take the physical characteristics of the scattering center into
consideration. Some of such approximate target models are geometric theory of
diffraction (GTD) [13,14] and uniform theory of diffraction (UTD). Figure 1.10

Santra: “chapter1” — 2020/8/20 — 11:20 — page 19 — #19

20 Deep Learning Applications of Short-Range Radars

Figure 1.10 GTD primitive scattering structures.

Figure 1.11 Overall standard radar processing flow.

presents the canonical structures, such as flat plate, top hat, trihedral corner
reflector, dihedral corner reflector, cylinder, or sphere, which comprises the target
response. The RCS response for a sphere is 1, the RCS response of some of the
structures can be expressed as
S(k, φ, θ ) = (jk)α sinc(kLn sin(φ) cos(θ)) (1.28)
where k is the wavenumber, Ln = 0 if the scattering center is localized and Ln = 0
if distributed. The parameter α has a half-integer value.

1.5 3D Data-cube Processing

Figure 1.11 presents the overall radar signal processing blocks involving 3D data-
cube processing, detection mechanism, target parameter estimation and target
tracking. In this section, we present the 3D data-cube processing. The transmitted
LFM on being reflected from the target is mixed with a replica of the transmitted
signal resulting in a beat signal. The phase of the beat signal after the mixer due

Santra: “chapter1” — 2020/8/20 — 11:20 — page 20 — #20

 Introduction to Radar Signal Processing 21

to the kth point target can be expressed as


B B 2
φk (t) = 2π fc τk + tτk − τ (1.29)
Tc 2Tc k

where τk = 2(Rk +vc

k t)
is the round-trip propagation delay between the transmitted
and received signal after reflection from the kth target with range Rk and radial
velocity vk . The downconverted IF signal therefore is the superposition of received
signal from K point scatterers and thus expressed as


K


 
2fc Rk 2fc vk 2BRk
sIF (t) = exp 2π + + t (1.30)
c c cTc
k=1

after ignoring the second-order terms 2Bv k 2

Tc c t . The frequency shifts due to range
and velocity arising from multiple point targets at the IF signal are decoupled by
generating range-Doppler images (RDI) across all virtual channels. Expanding
the time index t as nk Tframe + ns TPRT + nf , where nf is the fast time index
0 < nf < Tc , and ns denotes the slow time index, TPRT is the chirp repetition
time indicating the time difference between start of two consecutive chirps in a
frame and nk denotes the frame number.
Figure 1.12 presents the 3D radar data cube that depicts data across the fast
time, slow time, and virtual channels. Now based on the processing complexity
and the specific application, the 3D data cube can be processed through a 1D
transformation, followed by detection of the targets in that domain and then

Figure 1.12 3D radar data cube.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 21 — #21

22 Deep Learning Applications of Short-Range Radars

estimation of the other target parameters. Alternately the target is detected in one
of the 2D slices and the other target parameter is estimated.

1.5.1 1D Processing
Figure 1.13(a) depicts the range processing step, wherein the fast time data is
transformed into range bins through a 1D FFT. The fast-time data is first applied
with a windowing function and then optionally zero-padded to increase the range
accuracy as
S −1
N

2πln
Rin (k) = r(l )w(l ) exp −j , 0≤n<Z (1.31)
Z
l =0

where Z is the zero pad length, NS is the number of ADC samples per chirp,
and Rin (k) is the range bin value at ith chirp and nth range bin on kth frame.
w(l ) is the window function and is applied to reduce the sidelobe level from
−13 dB (no windowing) to a much lower acceptable value. However, the window
function exhibits a trade-off of sidelobe levels with the main-lobe width. Thus
c
the theoretical range resolution, δr = 2B , is increased by a some factor, which is a
function of the windowing function used. Some of the standard window functions

Figure 1.13 (a) Range processing data, and (b) Doppler processing data.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 22 — #22

 Introduction to Radar Signal Processing 23

are Hanning window, Blackman-Harris window, Kaiser window, raised-root

cosine window function, and Chebysev window function. Note that in (1.31) the
summation runs up to NS instead of Z since beyond NS the samples are zeros.
Further it must be noted that to ensure proper scaling the amplitude spectra should
be scaled by N1S instead of Z1 for the same reason. The 1D range bins depicts the
range position of all the targets in front of the sensor. Once the target range bins
are detected, the Doppler and angle of the targets can be estimated with Doppler
FFT and mono-pulse algorithm of angle FFT, respectively. Similar preprocessing
steps are applied across slow-time and virtual channels if FFT transformation are
used.
Following the range transformation, pulse integration is applied to combine
the data over all chirps to improve the signal-to-noise ratio (SNR) facilitating
subsequent target detection. Coherent integration combines the phase and
magnitude of the range FFT data coherently over all chirps of a frame, and is
applied at frame k. Rci (k) is based on the mean of the range FFT data {Rin (k)}N C
i=1
over NC chirps:
1 
NC
Rcin (k) = · Rin (k) (1.32)
NC
i=1

In the case of additive white Gaussian noise, owing to the coherent

integration, the noise power is reduced by a factor of NC . Considering noise
zin ∼ N (0, σz2 ) to be independent and identically distributed (I.I.D.) then
1 Nc 1 2
i=1 zi ∼ N (0, Nc σz ), thus improving the signal-to-noise ratio.
n
Nc
In the case of noncoherent integration, the integration is applied just over
the range spectra, which results in a fractional signal-to-noise ratio improvement
compared to their coherent integration counterparts. The coherently integrated
range spectrum contains reflections from not only the moving objects but also
the stationary objects in its field of view. If the application is specific to detecting
moving targets such as humans, moving target indicator (MTI) filter is used. The
MTI filter can be used to suppress the contribution of these stationary objects and
also removes the static Tx-Rx leakage. One of the means of applying the MTI filter
is through a moving average filter so that at each frame the coherent integrated
range FFT spectrum, denoted as Rci (k) at kth frame, can be expressed as

Rci (k) = Rci (k) − S(k − 1)

S(k) = α · Rci (k) + (1 − α) · S(k − 1) (1.33)

where α is the forget factor, for example set to 0.01. This filtered range FFT
spectrum can then be utilized for the subsequent target detection.
Figure 1.13(b) depicts the Doppler processing step, and for the cases where
the desired signal component across Doppler is modeled as a discrete sinusoid at

Santra: “chapter1” — 2020/8/20 — 11:20 — page 23 — #23

24 Deep Learning Applications of Short-Range Radars

frequency ν, the signal across the slow time can be represented as


s=σ 1 e j2π νTPRT e j2πν2TPRT · · · e j2πνNc TPRT (1.34)

Thus the matched filter output in this case of discrete sinusoid in Doppler
amounts to a discrete Fourier transform (DFT); that is,

h = σ 1 e −j2π νTPRT e −j2πν2TPRT · · · e −j2πνNc TPRT (1.35)

However, in the case of human sensing where the Doppler component is

not just the macro-Doppler or single sinusoid but is modeled as a superposition
of a micro-Doppler signature along with the macro-Doppler component, the
DFT is no longer the matched filter operation that maximizes the output SNR.
In the case of FSK or pulsed-CW radars, the Doppler transform is first applied,
followed by range estimation.
In the case of Doppler processing when time division multiplexing of chirps
in a MIMO configuration is used the maximum unambiguous Doppler is reduced
by a factor of the number of transmit antennas. In such cases, the unambiguous
Doppler can be increased by staggered PRT concept. Say PRF1 , PRF2 , · · · PRFT
are the pulse repetition frequency of group of chirps transmitted in a frame, where
PRFt are chosen so that they are integers and relative multiples of some frequency
(i.e., PRFt = pt PRF0 ). By application of the Chinese remainder theorem, the
maximum unambiguous Doppler turns out to be fmax = p1 p2 · · · pT PRF0 ,
thus increasing the unambiguous Doppler frequency. Further, if the pulse
repetition time is increased it leads to a lower duty cycle, and therefore lower
power consumption, and is an effective processing technique in power-critical
applications.
As depicted in Figure 1.14, the phase difference of the received signal
between two antennas is provided as δφ = 2π dλ sin(θ), where d is the distance
between the two antennas, λ is the wavelength of the receive waveform, and θ
is the angle extended from the boresight with respect to the receive wavefront.
In the case where there is only one target at a range bin, the phase monopulse
algorithm is the maximum likelihood estimator for estimating the angle of arrival
and can be expressed as
 
λ ˆ
δφ
θ̂ = sin−1 (1.36)
2π d

where δφ ˆ is the estimated phase difference between received data on both

antennas. It is easily noted that the system becomes underdetermined if there is
more than one target at the range bin with two antennas. Thus, for Nt , multiple-
target angle of arrival estimation requires at least Nt + 1 virtual receive antennas
to form an overdetermined system of equations.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 24 — #24

 Introduction to Radar Signal Processing 25

Figure 1.14 Angle of arrival.

Alternately, 2D transformation are applied before detection and estimation

of parameters. Based on the application, either RDI, range-cross range images
(RCRIs), or angle-Doppler images are generated. The following sections describe
the generation of these images before further processing.

1.5.2 2D Range Doppler Images

The received signal at frame nk , sIF (t; nk ), from consecutive chirps are arranged
in the form of a 2D matrix; that is, sIF (ns , nf ; nk ). The RDI is generated for each
channel by applying a window function, zero-padding, and then a 1D fast Fourier
transform (FFT) along fast time to obtain the range transformation, followed by
applying window function, zero-padding and then 1D FFT along slow time
index. The 2D discrete Fourier transform (DFT) transforms the sampled signal
sIF (ns , nf ; nk ) into range-Doppler domain
ZNc
Z
 NTS 
−j2πpnf /ZNTS
S(p, q, nk ) = wf (nf )sIF (ns , nf ; nk )e
ns =1 nf =1

· ws (ns )e −j2πqns /ZNc (1.37)

with NTS and ZNTS being the number of transmitted samples defined by
digital to analog converter (DAC) sampling points over chirp duration and
zero-padding along fast-time, respectively. Nc and ZNc are being the number of
chirps in a frame and zero-padding along the slow-time, respectively. wf (nf ) and

Santra: “chapter1” — 2020/8/20 — 11:20 — page 25 — #25

26 Deep Learning Applications of Short-Range Radars

ws (ns ) represent the window function along fast-time and slow-time, respectively,
and for our implementation we have used the Hamming window and Kaiser
window, respectively. p, q denotes the index over range and Doppler dimensions,
respectively. It is obvious that the peaks in the range-Doppler domain occur at


2fc 2B
pk = vk + Rk
c cTc
2vk fc
qk = (1.38)
c
2f 2B
Using sawtooth FMCW with fast ramps, c c vk cTc Rk , the range peaks appear
2B
at cT R . The maximum velocity is given as
c k

c
vmax =
2fc TPRT
and the minimum velocity is
c
δv =
2fc ZNc TPRT
Figure 1.15 shows the signal processing steps to create the processed RDIs.
Following the 2D FFT to transform the data into range-Doppler domain,
background subtraction is achieved through a moving average filter as
S(p, q; nk ) = S(p, q; nk ) − SB (p, q; nk )
SB (p, q; nk + 1) = γ SB (p, q; nk ) + (1 − γ )S(p, q; nk ) (1.39)

where S(p, q; nk ) is the RDI at nkth frame, and SB (p, q; nk ) is the background RDI
at the nkth frame, and γ is the moving average coefficient.
Following the RDI generation across all virtual channels, to gain diversity
and improve signal quality, maximal ratio combining is used to combine the RDIs
from different antennas. The gains for the weighted averaging are determined
by estimating the SNR for each RDI across antennas. The effective RDI is
computed as
NRx rx
g |RDIrx |
RDIeff = rx=1 NRx rx (1.40)
rx=1 g
where RDIrx is the complex RDI of the rxth receive channel, and the gain is
adaptively calculated as

max{|RDIrx |2 }
g rx = (NTS.PN)   (1.41)
NTS PN 2
m=1 |RDI (m, l )| − max{|RDIrx |2 }
rx
l =1

Santra: “chapter1” — 2020/8/20 — 11:20 — page 26 — #26

 Introduction to Radar Signal Processing 27

Figure 1.15 (a) Range processing steps, (b) Doppler processing steps, and (c) background
subtraction and combining.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 27 — #27

28 Deep Learning Applications of Short-Range Radars

where max{.} represents the maximum value from the 2D function, and
g rx represents the estimated SNR at rth receive channel. Thus the
(PN × NTS × NRx ) RDI tensor is transformed into a (PN × NTS) RDI matrix,
which is then subsequently fed to the detection algorithm.

1.5.3 Range Cross-Range Images

In this book, range cross-range images (RCRI) and range-angle images (RAI) are
used interchangeably, since one can simply be obtained from the other through
coordinate transformation. Figure 1.16 presents the algorithm flow of how the
range transform across the virtual channels are transformed into range-angle across
all the Doppler indices, which is further combined to form a single range-angle
image used for detection and further processing. There are several approaches
to obtain the range-angle image. The simplest approach takes a FFT along the
virtual channel after applying a window function and zero-padding.
The Nt × Nr deramped beat signal can be stacked into a vector and
the Kronecker product of the steering vector of the Tx array aTx (θ) and the
steering vector of the Rx array a Rx (θ); that is, aTx (θ) ⊗ a Rx (θ) can be used
to resolve the relative angle θ of the scatterer. Subsequently, beamforming of
the MIMO array signals can be regarded as synthesizing the received signals
with the Tx and Rx steering vectors. The azimuth imaging profile for a range
bin l can be generated using the Capon spectrum from the beamformer. The
Capon beamformer is computed by minimizing the variance/power of noise while
maintaining a distortionless response toward a desired angle. The corresponding

Figure 1.16 Range-angle transformation followed by maximal ratio combining.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 28 — #28

 Introduction to Radar Signal Processing 29

quadratic optimization problem is

min w H Cw
w

s.t. w H (a Tx (θ) ⊗ a Rx (θ)) = 1 (1.42)

where C is the covariance matrix of noise, the above optimization has a closed-
C −1 a(θ )
form expression given as wcapon = aH (θ )C −1 a(θ )
, with θ being a desired angle.
On substituting wcapon in objective function of (1.42), the spatial spectrum is
given as
1
Pl (θ) =  H   (1.43)
aTx (θ) ⊗ a Rx (θ) Ĉl−1 aTx (θ) ⊗ a Rx (θ)
with l = 0, ..., L
However, estimation of noise covariance at each range bin l is difficult in
practice, and hence Ĉl is estimated, which contains the signal component as
well and can be estimated using the sample matrix inversion (SMI) technique

Ĉl = N1 Kk=1 slIF (k)slIF (k)H , where K denotes the number of snapshots
used for the signal-plus-noise covariance estimation and slIF (k) is the deramped
intermediate frequency signal at range bin l with k being the frame index.
The Capon spatial spectrum can be viewed as an adaptive signal-dependent
spatial filtered spectrum that is characteristic or defined by the material the radar is
illuminating. Capon spectra capture the spatial contour of the material intensity
reflection and thus are more robust to increase in noise power, antenna array
errors, and target aspect angle.
Alternately the range-cross range image can be generated by using simple
beamforming algorithm based on phase monopulse algorithm. The range-angle
images can be expressed as


Nv
 
Pl (θ) = aTx (θ) ⊗ a Rx (θ) slIF (k) (1.44)
i=1

Figure 1.17 presents the 2D slices of the 3D radar data cube that can be
processed for either RDI, RCRI, or angle-Doppler images based on different
applications for subsequent processing.

1.6 Detection Strategy and Clustering

1.6.1 Detection Algorithm
Following either the 1D range transformation or 2D range-Doppler/range cross-
range transformation, detection strategy is applied to detect if a cell under test

Santra: “chapter1” — 2020/8/20 — 11:20 — page 29 — #29

30 Deep Learning Applications of Short-Range Radars

Figure 1.17 3D data-cube processing with 2D slices.

(CUT) contains a target. The detection strategy can be applied over the 3D data
cube; however, it is seldom done on an embedded processor mainly due to its
computational complexity. Figure 1.18(a) presents the detection strategy applied
to a 1D data vector, which can be the range transformation or the Doppler
transformation data. Figure 1.18(b) presents the detection strategy applied to a
2D data matrix (i.e., either a range-Doppler image, the range cross-range image,
or the angle Doppler image).
The target detection problem can be expressed as

2
1 if zcut > µσcut
zcut = 2 (1.45)
0 if zcut < µσcut

Radar detection algorithms are typically based on Neyman-Pearson (NP)

criterion, wherein the probability of false alarm, PFA , is set constant and the
detector strategy is to maximize the probability of detection, PD , for a given SNR
in the presence of varying interference or noise levels. Such detectors are referred
to as a constant false-alarm rate (CFAR) detector. In an indoor environment or
automotive setting, varying clutter levels arise due to interfering objects in the
field of view and multiple targets. The NP detector assumes that the interference
is independent and identically distributed (IID) over all resolution cells and the
noise variance of the interference distribution are estimated.
One of the simplest form of CFAR detection is the cell averaging CFAR
algorithm, wherein the square value of the reference cells from the lagging cell
and leading cells are added to estimate the noise variance for the cell under test;

Santra: “chapter1” — 2020/8/20 — 11:20 — page 30 — #30

 Introduction to Radar Signal Processing 31

Figure 1.18 (a) 1D CFAR detector following 1D transformation, and (b) 2D CFAR detector
following 2D transformation.

that is,

N /2 
N /2
2
σcut (CA) = |zlagging (r)|2 + |zleading (r)|2 (1.46)
r=1 r=1

The threshold multiplier µ for the CFAR detection is set by an acceptable

probability of false alarm, which for cell-averaging CFAR (CA-CFAR) is
provided as
−1/N
µ = N (Pfa − 1) (1.47)

Santra: “chapter1” — 2020/8/20 — 11:20 — page 31 — #31

32 Deep Learning Applications of Short-Range Radars

where Pfa is the probability of false alarm and N is the window size used for noise
power estimation.
In the case of CA-CFAR, there are guard cells that are adjacent cells to the
CUT and are left out from computation of the noise variance for the threshold
for CUT. The size of the reference cells and the guard cells are hyperparameters
and are chosen to optimize target detection for a given false alarm rate. CA-CFAR
performs well for a point target; however, in the case of range or Doppler spread
targets the target scatterers are present in the reference cells and thus biasing
the noise variance estimation to higher value and missed detection. Further, CA-
CFAR elevates the noise threshold near a strong target, thus occluding or masking
nearby weaker targets leading to a low probability of detection.
The alternative CFAR detectors are smallest of CA-CFAR (SOCA-CFAR)
and greatest of CA-CFAR (GOCA-CFAR), which are given as

N /2 
N /2
2 2
σcut (SOCA) = min{ |zlagging (r)| , |zleading (r)|2 }
r=1 r=1


N /2 
N /2
2
σcut (GOCA) = max{ |zlagging (r)|2 , |zleading (r)|2 } (1.48)
r=1 r=1
SOCA-CFAR and GOCA-CFAR are computationally cheaper than CA-CFAR
and help in handling scenarios where the clutter and noise levels change abruptly.
However, like their CA-CFAR counterpart, they suffer from the target masking
and biased estimate issues in the case of extended targets.
Alternately for doubly spread targets, order-statistics CFAR (OS-CFAR) is
used to avoid issues of biased estimates, since the ordered statistic is robust to any
outliers. If the target’s spread lies in the reference cells the noise estimate is not
affected. In case of OS-CFAR, instead of the mean power in the reference cells,
the kth ordered data is selected as the estimated noise variance, σ 2 . A detailed
description of OS-CFAR can be found in [15,16]. In the case of heterogeneous
clutter where the noise parameters vary from one resolution cell to another,
an adaptive clutter map is adaptively computed from frame to frame for noise
statistics.

1.6.2 Clustering
Contrary to a point target, in the case of doubly extended targets, the output of the
detection algorithm is not a single detection in the RDI for a target but is spread
across range and Doppler. Thus, a clustering algorithm is required to group the
detections from a single target, based on its size, as a single cluster. This helps in
reducing the computational complexity for the target tracking algorithm, which
after clustering tracks a single target parameter instead of tracking, nonclustered
group of target parameters.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 32 — #32

 Introduction to Radar Signal Processing 33

One of the simplest and effective clustering algorithms is based on the

Euclidian distance metric. In the Euclidean distance-based clustering algorithm,
the strongest detection is picked and the Euclidean distance between itself and the
neighboring detections are computed. If they fall within the prior known object
size they are clustered together as reflections from a single target and assigned
a label. Subsequently, the next strongest detection is picked and the cluster is
formed again based on the Euclidean distance lying within the objects geometric
size and assigned another label. This is continued until all the detected targets
are clustered and assigned a label. The drawback of this approach is that in a
dense target reflection environment, cluster boundaries can be highly skewed
by the target reflection strengths, which can result in incorrect assignments of
detections to clusters. Further, this clustering algorithm requires the target object
size as a a priori information.
A much more robust and effective unsupervised clustering algorithm is
called the density-based spatial clustering of applications with noise (DBSCAN)
algorithm [17]. Given a set of target detections from the same and multiple targets
in the 2D space, DBSCAN groups detections that are closely packed together,
while at the same time removing detections that lie alone in low-density regions
as outliers. To do this, DBSCAN classifies each point as either a core point, edge
point, or noise. Two input parameters are needed for the DBSCAN clustering
algorithm—the neighborhood radius d , and the minimum number of neighbors,
M . A point is defined as core point if it has at least M − 1 neighbors (i.e., points
within the distance d ). An edge point has less than M − 1 neighbors, but at
least one of its neighbors is a core point. All points that have less than M − 1
neighbors and no core point as a neighbor do not belong to any cluster and will
be classified as noise [17]. Figure 1.19 outlines the process of clustering target

Figure 1.19 Illustration of DBSCAN algorithm.

Santra: “chapter1” — 2020/8/20 — 11:20 — page 33 — #33

Discovering Diverse Content Through
Random Scribd Documents
other. Each feeds and reinforces the other. Freedom of soul and
subordination of action dwell in harmony. Obedience, definite
subjection and control, detailed organization is the lesson enforced
by the rule of causal necessity in the outer world of space and time
in which action takes place. Unlimited freedom, the heightening of
consciousness for its own sake, sheer reveling in noble ideals, the
law of the inner world. What more can mortal man ask?
It would not be difficult, I imagine, to fill the three hours devoted to
these lectures with quotations from representative German authors
to the effect that supreme regard for the inner meaning of things,
reverence for inner truth in disregard of external consequences of
advantage or disadvantage, is the distinguishing mark of the German
spirit as against, say, the externality of the Latin spirit or the
utilitarianism of Anglo-Saxondom. I content myself with one
quotation, a quotation which also indicates the same inclination to
treat historic facts as symbolic of great truths which is found in
Kant's treatment of church dogmas. Speaking of the Germanic
languages, an historian of German civilization says:

"While all other Indo-European languages allow a wide

liberty in placing the accent and make external
considerations, such as the quantity of the syllables and
euphony, of deciding influence, the Germanic tribes show
a remarkable and intentional transition to an internal
principle of accentuation. . . . Of all related peoples the
Germanic alone puts the accent on the root syllable of the
word, that is, on the part that gives it its meaning. There
is hardly an ethnological fact extant which gives so much
food for thought as this. What leads these people to give
up a habit which must have been so old that it had
become instinctive, and to evolve out of their own minds a
principle which indicates a power of discrimination far in
advance of anything we are used to attribute to the lower
stages of civilization? Circumstances of which we are not
now aware must have compelled them to distinguish the
 inner essence of things from their external form, and must
have taught them to appreciate the former as of higher,
indeed as of sole, importance. It is this accentuation of
the real substance of things, the ever-powerful desire to
discover this real substance, and the ever-present impulse
to give expression to this inner reality which has become
the controlling trait of the Germanic soul. Hence the
conviction gained by countless unfruitful efforts, that
reason alone will never get at the true foundation of
things; hence the thoroughness of German science; hence
a great many of the qualities that explain Germanic
successes and failures; hence, perhaps, a certain
stubbornness and obstinacy, the unwillingness to give up
a conviction once formed; hence the tendency to
mysticism; hence that continuous struggle which marks
the history of German art,—the struggle to give to the
contents powerful and adequate expression, and to satisfy
at the same time the requirements of esthetic elegance
and beauty, a struggle in which the victory is ever on the
side of truth, though it be homely, over beauty of form
whenever it appears deceitful; hence the part played by
music as the only expression of those imponderable
vibrations of the soul for which language seems to have
no words; hence the faith of the German in his mission
among the nations as a bringer of truth, as a recognizer of
the real value of things as against the hollow shell of
beautiful form, as the doer of right deeds for their own
sake and not for any reward beyond the natural outcome
of the deed itself."

The division established between the outer realm, in which of course

acts fall, and the inner realm of consciousness explains what is
otherwise so paradoxical to a foreigner in German writings: The
constant assertion that Germany brought to the world the conscious
recognition of the principle of freedom coupled with the assertion of
the relative incompetency of the German folk en masse for political
self-direction. To one saturated by the English tradition which
identifies freedom with power to act upon one's ideas, to make one's
purposes effective in regulation of public affairs, the combination
seems self-contradictory. To the German it is natural. Readers who
have been led by newspaper quotations to regard Bernhardi as
preaching simply a gospel of superior force will find in his writings a
continual assertion that the German spirit is the spirit of freedom, of
complete intellectual self-determination; that the Germans have
"always been the standard bearers of free thought." We find him
supporting his teachings not by appeal to Nietzsche, but by the
Kantian distinction between the "empirical and rational ego."
It is Bernhardi who says:

"Two great movements were born from the German

intellectual life, on which, henceforth, all the intellectual
and moral progress of mankind must rest:—The
Reformation and the critical philosophy. The Reformation
that broke the intellectual yoke imposed by the Church,
which checked all free progress; and the Critique of Pure
Reason which put a stop to the caprice of philosophic
speculation by defining for the human mind the limitations
of its capacities for knowledge, and at the same time
pointed out the way in which knowledge is really possible.
On this substructure was developed the intellectual life of
our time, whose deepest significance consists in the
attempt to reconcile the result of free inquiry with the
religious needs of the heart, and thus to lay a foundation
for the harmonious organization of mankind. . . . The
German nation not only laid the foundations of this great
struggle for a harmonious development of humanity but
took the lead in it. We are thus incurring an obligation for
the future from which we cannot shrink. We must be
prepared to be the leader in this campaign which is being
fought for the highest stake that has been offered to
human efforts. . . . To no nation except the German has it
 been given to enjoy in its inner self 'that which is given to
mankind as a whole.' . . . It is this quality which especially
fits us for leadership in the intellectual domain and
imposes upon us the obligation to maintain that position."
[35:A]

More significant than the words themselves are their occasion and
the occupation of the one who utters them. Outside of Germany,
cavalry generals who employ philosophy to bring home practical
lessons are, I think, rare. Outside of Germany, it would be hard to
find an audience where an appeal for military preparedness would
be reinforced by allusions to the Critique of Pure Reason.
Yet only by taking such statements seriously can one understand the
temper in which opinion in Germany meets a national crisis. When
the philosopher Eucken (who received a Nobel prize for contributing
to the idealistic literature of the world) justifies the part taken by
Germany in a world war because the Germans alone do not
represent a particularistic and nationalistic spirit, but embody the
"universalism" of humanity itself, he utters a conviction bred in
German thought by the ruling interpretation of German philosophic
idealism. By the side of this motif the glorification of war as a
biologic necessity, forced by increase of population, is a secondary
detail, giving a totally false impression when isolated from its
context. The main thing is that Germany, more than any other
nation, in a sense alone of all nations, embodies the essential
principle of humanity: freedom of spirit, combined with thorough
and detailed work in the outer sphere where reigns causal law,
where obedience, discipline and subordination are the necessities of
successful organization. It is perhaps worth while to recall that Kant
lived, taught and died in Königsberg; and that Königsberg was the
chief city of east Prussia, an island still cut off in his early years from
western Prussia, a titular capital for the Prussian kings where they
went for their coronations. His lifework in philosophy coincides
essentially with the political work of Frederick the Great, the king
who combined a régime of freedom of thought and complete
religious toleration with the most extraordinary display known in
history of administrative and military efficiency. Fortunately for our
present purposes, Kant, in one of his minor essays, has touched
upon this combination and stated its philosophy in terms of his own
thought.
The essay in question is that entitled "What is the Enlightenment?"
His reply in substance is that it is the coming of age on the part of
humanity: the transition from a state of minority or infancy wherein
man does not dare to think freely to that period of majority or
maturity in which mankind dares to use its own power of
understanding. The growth of this power of free use of reason is the
sole hope of progress in human affairs. External revolutions which
are not the natural expression of an inner or intellectual revolution
are of little significance. Genuine growth is found in the slow growth
of science and philosophy and in the gradual diffusion throughout
the mass of the discoveries and conclusions of those who are
superior in intelligence. True freedom is inner freedom, freedom of
thought together with the liberty consequent upon it of teaching and
publication. To check this rational freedom "is a sin against the very
nature of man, the primary law of which consists in just the advance
in rational enlightenment."
In contrast with this realm of inner freedom stands that of civil and
political action, the principle of which is obedience or subordination
to constituted authority. Kant illustrates the nature of the two by the
position of a military subordinate who is given an order to execute
which his reason tells him is unwise. His sole duty in the realm of
practice is to obey—to do his duty. But as a member not of the State
but of the kingdom of science, he has the right of free inquiry and
publication. Later he might write upon the campaign in which this
event took place and point out, upon intellectual grounds, the
mistake involved in the order. No wonder that Kant proclaims that
the age of the enlightenment is the age of Frederick the Great. Yet
we should do injustice to Kant if we inferred that he expected this
dualism of spheres of action, with its twofold moral law of freedom
and obedience, to endure forever. By the exercise of freedom of
thought, and by its publication and the education which should make
its results permeate the whole state, the habits of a nation will
finally become elevated to rationality, and the spread of reason will
make it possible for the government to treat men, not as cogs in a
machine, but in accord with the dignity of rational creatures.
Before leaving this theme, I must point out one aspect of the work
of reason thus far passed over. Nature, the sensible world of space
and time, is, as a knowable object, constituted by the legislative
work of reason, although constituted out of a non-rational sensible
stuff. This determining work of reason forms not merely the Idealism
of the Kantian philosophy but determines its emphasis upon the a
priori. The functions of reason through which nature is rendered a
knowable object cannot be derived from experience, for they are
necessary to the existence of experience. The details of this a priori
apparatus lie far outside our present concern. Suffice it to say that
as compared with some of his successors, Kant was an economical
soul and got along with only two a priori forms and twelve a priori
categories. The mental habitudes generated by attachment to a
priori categories cannot however be entirely neglected in even such
a cursory discussion as the present.
If one were to follow the suggestion involved in the lately quoted
passage as to the significant symbolism of the place of the accent in
German speech, one might discourse upon the deep meaning of the
Capitalization of Nouns in the written form of the German language,
together with the richness of the language in abstract nouns. One
might fancy that the dignity of the common noun substantive,
expressing as it does the universal or generic, has bred an
intellectual deference. One may fancy a whole nation of readers
reverently bowing their heads at each successively capitalized word.
In such fashion one might arrive at a picture, not without its truth,
of what it means to be devoted to a priori rational principles.
A number of times during the course of the world war I have heard
someone remark that he would not so much mind what the Germans
did if it were not for the reasons assigned in its justification. But to
rationalize such a tangled skein as human experience is a difficult
task. If one is in possession of antecedent rational concepts which
are legislative for experience, the task is much simplified. It only
remains to subsume each empirical event under its proper category.
If the outsider does not see the applicability of the concept to the
event, it may be argued that his blindness shows his ineptness for
truly universal thinking. He is probably a crass empiric who thinks in
terms of material consequences instead of upon the basis of
antecedent informing principles of reason.
Thus it has come about that no moral, social or political question is
adequately discussed in Germany until the matter in hand has been
properly deduced from an exhaustive determination of its
fundamental Begriff or Wesen. Or if the material is too obviously
empirical to allow of such deduction, it must at least be placed under
its appropriate rational form. What a convenience, what a resource,
nay, what a weapon is the Kantian distinction of a priori rational form
and a posteriori empirical matter. Let the latter be as brutely
diversified, as chaotic as you please. There always exists a form of
unity under which it may be brought. If the empirical facts are
recalcitrant, so much the worse for them. It only shows how
empirical they are. To put them under a rational form is but to
subdue their irrational opposition to reason, or to invade their
lukewarm neutrality. Any violence done them is more than
indemnified by the favor of bringing them under the sway of a priori
reason, the incarnation of the Absolute on earth.
Yet there are certain disadvantages attached to a priori categories.
They have a certain rigidity, appalling to those who have not learned
to identify stiffness with force. Empirical matters are subject to
revision. The strongest belief that claims the support of experience is
subject to modification when experience testifies against it. But an a
priori conception is not open to adverse evidence. There is no court
having jurisdiction. If, then, an unfortunate mortal should happen to
be imposed upon so that he was led to regard a prejudice or
predilection as an a priori truth, contrary experience would have a
tendency to make him the more obstinate in his belief. History
proves what a dangerous thing it has been for men, when they try
to impose their will upon other men, to think of themselves as
special instruments and organs of Deity. The danger is equally great
when an a priori Reason is substituted for a Divine Providence.
Empirically grounded truths do not have a wide scope; they do not
inspire such violent loyalty to themselves as ideas supposed to
proceed directly from reason itself. But they are discussable; they
have a humane and social quality, while truths of pure reason have a
paradoxical way, in the end, of escaping from the arbitrament of
reasoning. They evade the logic of experience, only to become, in
the phrase of a recent writer, the spoil of a "logic of fanaticism."
Weapons forged in the smithy of the Absolute become brutal and
cruel when confronted by merely human resistance.
The stiffly constrained character of an a priori Reason manifests
itself in another way. A category of pure reason is suspiciously like a
pigeonhole. An American writer, speaking before the present war,
remarked with witty exaggeration that "Germany is a monstrous set
of pigeonholes, and every mother's son of a German is pigeoned in
his respective hole—tagged, labeled and ticketed. Germany is a huge
human check-room, and the government carries the checks in its
pocket." John Locke's deepest objection to the older form of the a
priori philosophy, the doctrine of innate ideas, was the readiness
with which such ideas become strongholds behind which authority
shelters itself from questioning. And John Morley pointed out long
ago the undoubted historic fact that the whole modern liberal social
and political movement has allied itself with philosophic empiricism.
It is hard here, as everywhere, to disentangle cause and effect. But
one can at least say with considerable assurance that a hierarchically
ordered and subordered State will feel an affinity for a philosophy of
fixed categories, while a flexible democratic society will, in its crude
empiricism, exhibit loose ends.
There is a story to the effect that the good townspeople of
Königsberg were accustomed to their watches by the time at which
Kant passed upon his walks—so uniform was he. Yielding to the
Teutonic temptation to find an inner meaning in the outer event, one
may wonder whether German thought has not since Kant's time set
its intellectual and spiritual clocks by the Kantian standard: the
separation of the inner and the outer, with its lesson of freedom and
idealism in one realm, and of mechanism, efficiency and organization
in the other. A German professor of philosophy has said that while
the Latins live in the present moment, the Germans live in the
infinite and ineffable. His accusation (though I am not sure he meant
it as such) is not completely justified. But it does seem to be true
that the Germans, more readily than other peoples, can withdraw
themselves from the exigencies and contingencies of life into a
region of Innerlichkeit which at least seems boundless; and which
can rarely be successfully uttered save through music, and a frail
and tender poetry, sometimes domestic, sometimes lyric, but always
full of mysterious charm. But technical ideas, ideas about means and
instruments, can readily be externalized because the outer world is
in truth their abiding home.
 II
GERMAN MORAL AND POLITICAL PHILOSOPHY

It is difficult to select sentences from Kant which are intelligible to

those not trained in his vocabulary, unless the selection is
accompanied by an almost word-by-word commentary. His writings
have proved an admirable terrain for the display of German
Gründlichkeit. But I venture upon the quotation of one sentence
which may serve the purpose of at once recalling the main lesson of
the previous lecture and furnishing a transition to the theme of the
present hour.

"Even if an immeasurable gulf is fixed between the

sensible realm of the concept of nature and the
supersensible realm of the concept of freedom, so that it
is not possible to go from the first to the second (at least
by means of the theoretical use of reason) any more than
if they were two separate worlds of which the first could
have no influence upon the second,—yet the second is
meant to have an influence upon the first. The concept of
freedom is meant to actualize in the world of sense the
purpose proposed by its laws." . . .

That is, the relation between the world of space and time where
physical causality reigns and the moral world of freedom and duty is
not a symmetrical one. The former cannot intrude into the latter. But
it is the very nature of moral legislation that it is meant to influence
the world of sense; its object is to realize the purposes of free
rational action within the sense world. This fact fixes the chief
features of Kant's philosophy of Morals and of the State.
It is a claim of the admirers of Kant that he first brought to
recognition the true and infinite nature of the principle of Personality.
On one side, the individual is homo phenomenon—a part of the
scheme of nature, governed by its laws as much as any stone or
plant. But in virtue of his citizenship in the kingdom of supersensible
Laws and Ends, he is elevated to true universality. He is no longer a
mere occurrence. He is a Person—one in whom the purpose of
Humanity is incarnate. In English and American writings the terms
subjective and subjectivism usually carry with them a disparaging
color. Quite otherwise is it in German literature. This sets the age of
subjectivism, whose commencement, roughly speaking, coincides
with the influence of Kantian thought, in sharp opposition to the age
of individualism, as well as to a prior period of subordination to
external authority. Individualism means isolation; it means external
relations of human beings with one another and with the world; it
looks at things quantitatively, in terms of wholes and parts.
Subjectivism means recognition of the principle of free personality:
the self as creative, occupied not with an external world which limits
it from without, but, through its own self-consciousness, finding a
world within itself; and having found the universal within itself,
setting to work to recreate itself in what had been the external
world, and by its own creative expansion in industry, art and politics
to transform what had been mere limiting material into a work of its
own. Free as was Kant from the sentimental, the mystic and the
romantic phases of this Subjectivism, we shall do well to bear it in
mind in thinking of his ethical theory. Personality means that man as
a rational being does not receive the end which forms the law of his
action from without, whether from Nature, the State or from God,
but from his own self. Morality is autonomous; man, humanity, is an
end in itself. Obedience to the self-imposed law will transform the
sensible world (within which falls all social ties so far as they spring
from natural instinct desire) into a form appropriate to universal
reason. Thus we may paraphrase the sentence quoted from Kant.
The gospel of duty has an invigorating ring. It is easy to present it as
the most noble and sublime of all moral doctrines. What is more
worthy of humanity, what better marks the separation of man from
brute, than the will to subordinate selfish desire and individual
inclination to the commands of stern and lofty duty? And if the idea
of command (which inevitably goes with the notion of duty) carries a
sinister suggestion of legal authority, pains and penalties and of
subservience to an external authority who issues the commands,
Kant seems to have provided a final corrective in insisting that duty
is self-imposed. Moral commands are imposed by the higher,
supranatural self upon the lower empirical self, by the rational self
upon the self of passions and inclinations. German philosophy is
attached to antitheses and their reconciliation in a higher synthesis.
The Kantian principle of Duty is a striking case of the reconciliation
of the seemingly conflicting ideas of freedom and authority.
Unfortunately, however, the balance cannot be maintained in
practice. Kant's faithful logic compels him to insist that the concept
of duty is empty and formal. It tells men that to do their duty is their
supreme law of action, but is silent as to what men's duties
specifically are. Kant, moreover, insists, as he is in logic bound to do,
that the motive which measures duty is wholly inner; it is purely a
matter of inner consciousness. To admit that consequences can be
taken into account in deciding what duty is in a particular case would
be to make concessions to the empirical and sensible world which
are fatal to the scheme. The combination of these two features of
pure internality and pure formalism leads, in a world where men's
acts take place wholly in the external and empirical region, to
serious consequences.
The dangerous character of these consequences may perhaps be
best gathered indirectly by means of a quotation.

"While the French people in savage revolt against spiritual

and secular despotism had broken their chains and
proclaimed their rights, another quite different revolution
was working in Prussia—the revolution of duty. The
assertion of the rights of the individual leads ultimately to
individual irresponsibility and to a repudiation of the State.
Immanuel Kant, the founder of the critical philosophy,
taught, in opposition to this view, the gospel of moral
 duty, and Scharnhorst grasped the idea of universal
military service. By calling upon which individual to
sacrifice property and life for the good of the community,
he gave the clearest expression to the idea of the State,
and created a sound basis on which the claims to
individual rights might rest."[52:A]

The sudden jump, by means of only a comma, from the gospel of

moral duty to universal military service is much more logical than the
shock which it gives to an American reader would indicate. I do not
mean, of course, that Kant's teaching was the cause of Prussia's
adoption of universal military service and of the thoroughgoing
subordination of individual happiness and liberty of action to that
capitalized entity, the State. But I do mean that when the practical
political situation called for universal military service in order to
support and expand the existing state, the gospel of a Duty devoid
of content naturally lent itself to the consecration and idealization of
such specific duties as the existing national order might prescribe.
The sense of duty must get its subject-matter somewhere, and
unless subjectivism was to revert to anarchic or romantic
individualism (which is hardly in the spirit of obedience to
authoritative law) its appropriate subject-matter lies in the
commands of a superior. Concretely what the State commands is the
congenial outer filling of a purely inner sense of duty. That the
despotism of Frederick the Great and of the Hohenzollerns who
remained true to his policy was at least that hitherto unknown thing,
an enlightened despotism, made the identification easier. Individuals
have at all times, in epochs of stress, offered their supreme sacrifice
to their country's good. In Germany this sacrifice in times of peace
as well as of war has been systematically reinforced by an inner
mystic sense of a Duty elevating men to the plane of the universal
and eternal.
In short, the sublime gospel of duty has its defects. Outside of the
theological and the Kantian moral traditions, men have generally
agreed that duties are relative to ends. Not the obligation, but some
purpose, some good, which the fulfillment of duty realizes, is the
principle of morals. The business of reason is to see that the end,
the good, for which one acts is a reasonable one—that is to say, as
wide and as equitable in its working out as the situation permits.
Morals which are based upon consideration of good and evil
consequences not only allow, but imperiously demand the exercise
of a discriminating intelligence. A gospel of duty separated from
empirical purposes and results tends to gag intelligence. It
substitutes for the work of reason displayed in a wide and
distributed survey of consequences in order to determine where duty
lies an inner consciousness, empty of content, which clothes with
the form of rationality the demands of existing social authorities. A
consciousness which is not based upon and checked by
consideration of actual results upon human welfare is none the less
socially irresponsible because labeled Reason.
Professor Eucken represents a type of idealistic philosophy which is
hardly acceptable to strict Kantians. Yet only where the fundamental
Kantian ideas were current would such ethical ideas as the following
flourish:

"When justice is considered as a mere means of securing

man's welfare, and is treated accordingly—whether it be
the welfare of individuals or of society as a whole makes
no essential difference—it loses all its characteristic
features. No longer can it compel us to see life from its
own standpoint; no longer can it change the existing
condition of things; no longer can it sway our hearts with
the force of a primitive passion, and oppose to all
consideration of consequences an irresistible spiritual
compulsion. It degenerates rather into the complaisant
servant of utility; it adopts herself to her demands, and in
so doing suffers inward annihilation. It can maintain itself
only when it comes as a unique revelation of the Spiritual
Life within our human world, as a lofty Presence
transcending all considerations of expediency."[55-A]
Such writing is capable of arousing emotional reverberations in the
breasts of many persons. But they are emotions which, if given
headway, smother intelligence, and undermine its responsibility for
promoting the actual goods of life. If justice loses all its
characteristic features when regarded as a means (the word "mere"
inserted before "means" speaks volumes) of the welfare of society as
a whole, then there is no objective and responsible criterion for
justice at all. A justice which, irrespective of the determination of
social well-being, proclaims itself as an irresistible spiritual impulsion
possessed of the force of a primitive passion, is nothing but a
primitive passion clothed with a spiritual title so that it is protected
from having to render an account of itself. During an ordinary course
of things, it passes for but an emotional indulgence; in a time of
stress and strain, it exhibits itself as surrender of intelligence to
passion.
The passage (from Bernhardi) quoted earlier puts the German
principle of duty in opposition to the French principle of rights—a
favorite contrast in German thought. Men like Jeremy Bentham also
found the Revolutionary Rights of Man doctrinaire and conducing to
tyranny rather than to freedom. These Rights were a priori, like
Duty, being derived from the supposed nature or essence of man,
instead of being adopted as empirical expedients to further progress
and happiness. But the conception of duty is one-sided, expressing
command on one side and obedience on the other, while rights are
at least reciprocal. Rights are social and sociable in accord with the
spirit of French philosophy. Put in a less abstract form than the
revolutionary theory stated them, they are things to be discussed
and measured. They admit of more and less, of compromise and
adjustment. So also does the characteristic moral contribution of
English thought—intelligent self-interest. This is hardly an ultimate
idea. But at least it evokes a picture of merchants bargaining, while
the categorical imperative calls up the drill sergeant. Trafficking
ethics, in which each gives up something which he wants to get
something which he wants more, is not the noblest kind of morals,
but at least it is socially responsible as far as it goes. "Give so that it
may be given to you in return" has at least some tendency to bring
men together; it promotes agreement. It requires deliberation and
discussion. This is just what the authoritative voice of a superior will
not tolerate; it is the one unforgiveable sin.
The morals of bargaining, exchange, the mutual satisfaction of
wants may be outlived in some remote future, but up to the present
they play an important part in life. To me there is something
uncanny in the scorn which German ethics, in behalf of an unsullied
moral idealism, pours upon a theory which takes cognizance of
practical motives. In a highly esthetic people one might understand
the display of contempt. But when an aggressive and commercial
nation carries on commerce and war simply from the motive of
obedience to duty, there is awakened an unpleasant suspicion of a
suppressed "psychic complex." When Nietzsche says, "Man does not
desire happiness; only the Englishman does that," we laugh at the
fair hit. But persons who profess no regard for happiness as a test of
action have an unfortunate way of living up to their principle by
making others unhappy. I should entertain some suspicion of the
complete sincerity of those who profess disregard for their own
happiness, but I should be quite certain of their sincerity when it
comes to a question of my happiness.
Within the Kantian philosophy of morals there is an idea which
conducts necessarily to a philosophy of society and the State. Leibniz
was the great German source of the philosophy of the
enlightenment. Harmony was the dominant thought of this
philosophy; the harmony of nature with itself and with intelligence;
the harmony of nature with the moral ends of humanity. Although
Kant was a true son of the enlightenment, his doctrine of the
radically dual nature of the legislation of Reason put an end to its
complacent optimism. According to Kant, morality is in no way a
work of nature. It is the achievement of the self-conscious reason of
man through conquest of nature. The ideal of a final harmony
remains, but it is an ideal to be won through a battle with the
natural forces of man. His breach with the enlightenment is nowhere
as marked as in his denial that man is by nature good. On the
contrary, man is by nature evil—that is, his philosophical rendering
of the doctrine of original sin. Not that the passions, appetites and
senses are of themselves evil, but they tend to usurp the sovereignty
of duty as the motivating force of human action. Hence morality is a
ceaseless battle to transform all the natural desires of man into
willing servants of the law and purpose of reason.
Even the kindly and sociable instincts of man, in which so many have
sought the basis of both morality and organized society, fall under
Kant's condemnation. As natural desires, they aspire to an
illegitimate control in man's motives. They are parts of human self-
love: the unlawful tendency to make happiness the controlling
purpose of action. The natural relations of man to man are those of
an unsocial sociableness. On the one hand, men are forced together
by natural ties. Only in social relations can individuals develop their
capacities. But no sooner do they come together than disintegrating
tendencies set in. Union with his fellows give a stimulus to vanity,
avarice and gaining power over others—traits which cannot show in
themselves in individuals when they are isolated. This mutual
antagonism is, however, more of a force in evolving man from
savagery to civilization than are the kindly and sociable instincts.

"Without these unlovely qualities which set man over

against man in strife, individuals would have lived on in
perfect harmony, contentment and mutual love, with all
their distinctive abilities latent and undeveloped."

In short, they would have remained in Rousseau's paradise of a

state of nature, and

"perhaps Rousseau was right when he preferred the

savage state to the state of civilization provided we leave
out of account the last stage to which our species is yet
destined to rise."

But since the condition of civilization is but an intermediary between

the natural state and the truly or rational moral condition to which
man is to rise, Rousseau was wrong.

"Thanks then be to nature for the unsociableness, the

spiteful competition of vanity, the insatiate desires for
power and gain."

These quotations, selected from Kant's little essay on an "Idea for a

Universal History," are precious for understanding two of the most
characteristic traits of subsequent German thought, the distinctions
made between Society and the State and between Civilization and
Culture. Much of the trouble which has been experienced in respect
to the recent use of Kultur might have been allayed by a knowledge
that Kultur has little in common with the English word "culture" save
a likeness in sound. Kultur is sharply antithetical to civilization in its
meaning. Civilization is a natural and largely unconscious or
involuntary growth. It is, so to speak, a by-product of the needs
engendered when people live close together. It is external, in short.
Culture, on the other, is deliberate and conscious. It is a fruit not of
men's natural motives, but of natural motives which have been
transformed by the inner spirit. Kant made the distinction when he
said that Rousseau was not so far wrong in preferring savagery to
civilization, since civilization meant simply social decencies and
elegancies and outward proprieties, while morality, that is, the rule
of the end of Reason, is necessary to culture. And the real
significance of the term "culture" becomes more obvious when he
adds that it involves the slow toil of education of the Inner Life, and
that the attainment of culture on the part of an individual depends
upon long effort by the community to which he belongs. It is not
primarily an individual trait or possession, but a conquest of the
community won through devotion to "duty."
In recent German literature, Culture has been given even a more
sharply technical distinction from Civilization and one which
emphasizes even more its collective and nationalistic character.
Civilization as external and uncontrolled by self-conscious purpose
includes such things as language in its more spontaneous colloquial
expression, trade, conventional manners or etiquette, and the police
activities of government. Kultur comprises language used for
purposes of higher literature; commerce pursued not as means of
enriching individuals but as a condition of the development of
national life; art, philosophy (especially in that untranslatable thing,
the "Welt-Anschauung"); science, religion, and the activities of the
state in the nurture and expansion of the other forms of national
genius, that is, its activities in education and the army. The
legislation of Bismarck with reference to certain Roman Catholic
orders is nicknamed Kultur-kampf, for it was conceived as
embodying a struggle between two radically different philosophies of
life, the Roman, or Italian, and the true Germanic, not simply as a
measure of political expediency. Thus it is that a trading and military
post like Kiao-Chou is officially spoken of as a "monument of
Teutonic Kultur." The war now raging is conceived of as an outer
manifestation of a great spiritual struggle, in which what is really at
stake is the supreme value of the Germanic attitude in philosophy,
science and social questions generally, the "specifically German
habits of feeling and thinking."
Very similar motives are at work in the distinction between society
and the State, which is almost a commonplace of German thought.
In English and American writings the State is almost always used to
denote society in its more organized aspects, or it may be identified
with government as a special agency operating for the collective
interests of men in association. But in German literature society is a
technical term and means something empirical and, so to speak,
external; while the State, if not avowedly something mystic and
transcendental, is at least a moral entity, the creation of self-
conscious reason operating in behalf of the spiritual and ideal
interests of its members. Its function is cultural, educative. Even
when it intervenes in material interests, as it does in regulating
lawsuits, poor laws, protective tariffs, etc., etc., its action has
ultimately an ethical significance: its purpose is the furthering of an
ideal community. The same thing is to be said of wars when they are
really national wars, and not merely dynastic or accidental.
Society is an expression of man's egoistic nature; his natural seeking
for personal advantage and profit. Its typical manifestation is in
competitive economic struggle and in the struggle for honor and
recognized social status. These have their proper place; but with
respect even to them it is the duty of the State to intervene so that
the struggle may contribute to ideal ends which alone are universal.
Hence the significance of the force or power of the State. Unlike
other forms of force, it has a sort of sacred import, for it represents
force consecrated to the assertion and expansion of final goods
which are spiritual, moral, rational. These absolute ends can be
maintained only in struggle against man's individualistic ends.
Conquest through conflict is the law of morals everywhere.
In Kant we find only the beginnings of this political philosophy. He is
still held back by the individualism of the eighteenth century.
Everything legal and political is conceived by him as external and
hence outside the strictly moral realm of inner motivation. Yet he is
not content to leave the State and its law as a wholly unmoral
matter. The natural motives of man are, according to Kant (evidently
following Hobbes), love of power, love of gain, love of glory. These
motives are egoistic; they issue in strife—in the war of all against all.
While such a state of affairs does not and cannot invade the inner
realm of duty, the realm of the moral motive, it evidently presents a
régime in which the conquest of the world of sense by the law of
reason cannot be effected. Man in his rational or universal capacity
must, therefore, will an outward order of harmony in which it is at
least possible for acts dictated by rational freedom to get a footing.
Such an outer order is the State. Its province is not to promote
moral freedom directly—only the moral will can do that. But its
business is to hinder the hindrances to freedom: to establish a social
condition of outward order in which truly moral acts may gradually
evolve a kingdom of humanity. Thus while the State does not have a
directly moral scope of action (since the coercion of motive is a
moral absurdity), it does have a moral basis and an ultimate moral
function.
It is the law of reason, "holy and inviolable," which impels man to
the institution of the State, not natural sociability, much less
considerations of expediency. And so necessary is the State to
humanity's realization of its moral purpose that there can be no right
of revolution. The overthrow and execution of the sovereign (Kant
evidently had the French Revolution and Louis XVI in mind) is "an
immortal and inexpiable sin like the sin against the Holy Ghost
spoken of by theologians, which can never be forgiven in this world
or in the next."
Kant was enough of a child of the eighteenth century to be
cosmopolitan, not nationalistic, in his feeling. Since humanity as a
whole, in its universality, alone truly corresponds to the universality
of reason, he upheld the ideal of an ultimate republican federation of
states; he was one of the first to proclaim the possibility of enduring
peace among nations on the basis of such a federated union of
mankind.
The threatened domination of Europe by Napoleon following on the
wars waged by republican France put an end, however, to
cosmopolitanism. Since Germany was the greatest sufferer from
these wars, and since it was obvious that the lack of national unity,
the division of Germany into a multitude of petty states, was the
great source of her weakness; since it was equally obvious that
Prussia, the one strong and centralized power among the German
states, was the only thing which saved them all from national
extinction, subsequent political philosophy in Germany rescued the
idea of the State from the somewhat ambiguous moral position in
which Kant had left it. Since a state which is an absolute moral
necessity and whose actions are nevertheless lacking in inherent
moral quality is an anomaly, the doctrine almost calls for a theory
which shall make the State the supreme moral entity.
Fichte marks the beginning of the transformation; and, in his
writings, it is easy to detect a marked difference of attitude toward
the nationalistic state before and after 1806, when in the battle of
Jena Germany went down to inglorious defeat. From the time of
Fichte, the German philosophy of the State blends with its
philosophy of history, so that my reservation of the latter topic for
the next section is somewhat arbitrary, and I shall not try rigidly to
maintain the division of themes.
I have already mentioned the fact that Kant relaxes the separation of
the moral realm of freedom from the sensuous realm of nature
sufficiently to assert that the former is meant to influence the latter
and finally to subjugate it. By means of the little crack thus
introduced into nature, Fichte rewrites the Kantian philosophy. The
world of sense must be regarded from the very start as material
which the free, rational, moral Ego has created in order to have
material for its own adequate realization of will. Fichte had a longing
for an absolute unity which did not afflict Kant, to whom, save for
the concession just referred to, a complete separation of the two
operations of legislative reason sufficed. Fichte was also an ardently
active soul, whose very temperament assured him of the
subordination of theoretical knowledge to moral action.
It would be as difficult to give, in short space, an adequate sketch of
Fichte's philosophy as of Kant's. To him, however, reason was the
expression of the will, not (as with Kant) the will an application of
reason to action. "Im Anfang war die That" is good Fichteanism.
While Kant continued the usual significance of the term Reason (with
only such modifications as the rationalism of his century had made
current), Fichte began the transformation which consummated in
later German idealism. If the world of nature and of human relations
is an expression of reason, then reason must be the sort of thing,
and have the sort of attributes by means of which the world may be
construed, no matter how far away this conception of reason takes
us from the usual meaning of the term. To Fichte the formula which
best described such aspects of the world and of life as he was
interested in was effort at self-realization through struggle with
difficulties and overcoming opposition. Hence his formula for reason
was a Will which, having "posited" itself, then "posited" its antithesis
in order, through further action subjugating this opposite, to conquer
its own freedom.
The doctrine of the primacy of the Deed, and of the Duty to achieve
freedom through moral self-assertion against obstacles (which, after
all, are there only to further this self-assertion) was one which could,
with more or less plausibility, be derived from Kant. More to our
present point, it was a doctrine which could be preached with noble
moral fervor in connection with the difficulties and needs of a
divided and conquered Germany. Fichte saw himself as the
continuator of the work of Luther and Kant. His final "science of
knowledge" brought the German people alone of the peoples of the
world into the possession of the idea and ideal of absolute freedom.
Hence the peculiar destiny of the German scholar and the German
State. It was the duty and mission of German science and
philosophy to contribute to the cause of the spiritual emancipation of
humanity. Kant had already taught that the acts of men were to
become gradually permeated by a spirit of rationality till there should
be an equation of inner freedom of mind and outer freedom of
action. Fichte's doctrine demanded an acceleration of the process.
Men who have attained to a consciousness of the absolute freedom
and self-activity must necessarily desire to see around them similar
free beings. The scholar who is truly a scholar not merely knows, but
he knows the nature of knowledge—its place and function as a
manifestation of the Absolute. Hence he is, in a peculiar sense, the
direct manifestation of God in the world—the true priest. And his
priestly function consists in bringing other men to recognize moral
freedom in its creative operation. Such is the dignity of education as
conducted by those who have attained true philosophic insight.
Fichte made a specific application of this idea to his own country and
time. The humiliating condition of contemporary Germany was due
to the prevalence of egoism, selfishness and particularism: to the
fact that men had lowered themselves to the plane of sensuous life.
The fall was the worse because the Germans, more than any other
people, were by nature and history conscious of the ideal and
spiritual principle, the principle of freedom, lying at the very basis of
all things. The key to the political regeneration of Germany was to
be found in a moral and spiritual regeneration effected by means of
education. The key, amid political division, to political unity was to be
sought in devotion to moral unity. In this spirit Fichte preached his
"Addresses to the German Nation." In this spirit he collaborated in
the foundation of the University of Berlin, and zealously promoted all
the educational reforms introduced by Stein and Humboldt into
Prussian life.
The conception of the State as an essential moral Being charged
with an indispensable moral function lay close to these ideas.
Education is the means of the advancement of humanity toward
realization of its divine perfection. Education is the work of the State.
The syllogism completes itself. But in order that the State may carry
on its educational or moral mission it must not only possess
organization and commensurate power, but it must also control the
conditions which secure the possibility offered to the individuals
composing it. To adopt Aristotle's phrase, men must live before they
can live nobly. The primary condition of a secure life is that everyone
be able to live by his own labor. Without this, moral self-
determination is a mockery. The business of the State, outside of its
educational mission, is concerned with property, and this business
means insuring property to everyone as well as protecting him in
what he already possesses. Moreover, property is not mere physical
possession. It has a profound moral significance, for it means the
subjugation of physical things to will. It is a necessary part of the
realization of moral personality: the conquest of the non-ego by the
ego. Since property does not mean mere appropriation, but is a right
recognized and validated by society itself, property has a social basis
and aim. It is an expression not of individual egotism but of the
universal will. Hence it is essential to the very idea of property and
of the State that all the members of society have an equal
opportunity for property. Hence it is the duty of the State to secure
to its every member the right to work and the reward of his work.
The outcome, as expressed in his essay on "The Closed Industrial
State," is State Socialism, based on moral and idealistic grounds, not
on economic considerations. In order that men may have a real
opportunity to develop their moral personalities, their right to labor
and to adequate living, in return for their labor must be assured.
This cannot happen in a competitive society. Industry must be
completely regulated by the State if these indispensable rights to
labor and resulting comfort and security of life as means to moral
volition are to be achieved. But a state engaged in unrestricted
foreign trade will leave its workingmen at the mercy of foreign
conditions. It must therefore regulate or even eliminate foreign
commerce so far as is necessary to secure its own citizens. The
ultimate goal is a universal state as wide as humanity, and a state in
which each individual will act freely, without state-secured rights and
state-imposed obligations. But before this cosmopolitan and
philosophically anarchic condition can be reached, we must pass
through a period of the nationalistic closed state. Thus at the end a
wide gulf separates Fichte from Kant. The moral individualism of the
latter has become an ethical socialism. Only in and by means of a
circle of egos or personalities does a human being attain the moral
reason and freedom which Kant bestowed upon him as his
birthright. Only through the educational activities of the State and its
complete regulation of the industrial activities of its members does
the potential moral freedom of individuals become an established
reality.
If I have devoted so much space to Fichte it is not because of his
direct influence upon affairs or even upon thought. He did not found
a school. His system was at once too personal and too formal.
Nevertheless, he expressed ideas which, removed from their special
context in his system, were taken up into the thought of cultivated
Germany. Heine, speaking of the vogue of systems of thought, says
with profound truth that "nations have an instinctive presentiment of
what they require to fulfill their mission."
And Fichte's thought infiltrated through many crevices. Rodbertus
and Lasalle, the socialists, were, for example, profoundly affected by
him. When the latter was prosecuted in a criminal suit for his
"Programme of Workingmen," his reply was that his programme was
a distinctively philosophic utterance, and hence protected by the
constitutional provision for freedom of science and its teaching. And
this is his philosophy of the State:

"The State is the unity and coöperation of individuals in a

moral whole. . . . The ultimate and intrinsic end of the
State is, therefore, to further the positive unfolding, the
progressive development of human life. Its function is to
work out the true end of man; that is to say, the full
degree of culture of which human nature is capable."

And he quotes with approval the words:

"The concept of the State must be broadened so as to

make the State the contrivance whereby all human virtue
is to be realized to the full."

And if he differs from Fichte, it is but in the assertion that since the
laboring class is the one to whom the need most directly appeals, it
is workingmen who must take the lead in the development of the
true functions of the State.
Pantheism is a philosophic nickname which should be sparingly
employed; so also should the term Monism. To call Fichte's system
an ethical pantheism and monism is not to say much that is
enlightening. But with free interpretation the designation may be
highly significant in reference to the spiritual temper of the Germany
of the first part of the nineteenth century. For it gives a key to the
presentiment of what Germany needed to fulfill its mission.
It is a commonplace of German historians that its unity and
expansion to a great state powerful externally, prosperous internally,
was wrought, unlike that of any other people, from within outward.
In Lange's words, "our national development started from the most
ideal and approximated more and more to the real." Hegel and
Heine agree that in Germany the French Revolution and the
Napoleonic career were paralleled by a philosophic revolution and an
intellectual empire. You recall the bitter word that, when Napoleon
was finally conquered and Europe partitioned, to Germany was
assigned the kingdom of the clouds. But this aërial and tenuous
kingdom became a mighty power, working with and in the statesmen
of Prussia and the scholars of Germany to found a kingdom on the
solid earth. Spiritual and ideal Germany made common cause with
realistic and practical Prussia. As says Von Sybel, the historian of the
"Founding of the German Empire":

"Germany had been ruined through its own disintegration

and had dragged Prussia with it into the abyss. It was well
known that the wild fancies of the Conqueror hovered
about the utter annihilation of Prussia; if this should take
place, then east as well as west of the Elbe, not only
political independence, but every trace of a German spirit,
the German language and customs, German art and
learning—everything would be wiped out by the
foreigners. But this fatal danger was perceived just at the
time when everybody had been looking up to Kant and
Schiller, had been admiring Faust, the world-embracing
masterpiece of Goethe's, and had recognized that
Alexander von Humboldt's cosmological studies and
Niebuhr's "Roman History" had created a new era in
European science and learning. In such intellectual
attainments the Germans felt that they were far superior
to the vanquisher of the world and his great nation; and
so the political interests of Prussia and the salvation of the
German nationality exactly coincided. Schleiermacher's
patriotic sermons, Fichte's stirring addresses to the
German people, Humboldt's glorious founding of the Berlin
University, served to augment the resisting power of
Prussia, while Scharnhorst's recruits and militia were
devoted to the defense of German honor and German
customs. Everyone felt that German nationality was lost if
Prussia did not come to its rescue, and that, too, there
was no safety possible for Prussia unless all Germany was
free.
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com