![Preface to ”Convergence of Intelligent Data
Acquisition and Advanced Computing Systems”
This preface briefly outlines the objectives of the Special Issue on “Convergence of Intelligent
Data Acquisition and Advanced Computing Systems” published between September 2019 and
September 2020 in the journal Sensors. This Special Issue welcomed submissions as extended versions
of conference articles published in the proceedings of the “10th IEEE International Conference
on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications
(IDAACS’2019)”, allowing the authors to expand on their theoretical and experimental contributions.
These were complemented with external submissions from interested researchers working in this
timely area. We highlight the main contributions of the published articles, grouped by key topics:
intelligent data acquisition, learning methods and optimization for data processing, and advanced
computing systems to support data processing.
As modern data-acquisition solutions increasingly integrate on-board computing in the case of
intelligent sensors, in-network computing, or in the case of distributed systems and sensor networks,
a need to provide a joint view of these previously divergent topics has been observed. We define
the convergence of intelligent data acquisition and advanced computing systems at the interface of
theory and applications for industrial manufacturing, scientific computing, precision agriculture, and
energy systems such as smart building and smart city systems. This book bridges specific topics for
instrumentation such as accuracy, sampling, time synchronization, sensor selection and calibration
with algorithm design, statistical and machine learning, computational efficiency, and reconfigurable
computing that supports conventional engineering tasks. Computing-as-a-service is another relevant
approach that can be leveraged to improve measurements and instrumentation in the future.
The potential topics of this Special Issue that were initially defined were mapped onto
the core areas and special streams of the IDAACS’2019 conference and included but were not
limited to advanced instrumentation and data-acquisition systems; advanced mathematical methods
for data acquisition and computing systems; computational intelligence for instrumentation and
data-acquisition systems; data analysis and modeling; intelligent distributed systems and remote
control; intelligent information systems, data mining, and ontology; Internet of Things; pattern
recognition, digital image, and signal processing; intelligent instrumentation and data acquisition
systems in advanced manufacturing for Industry 4; intelligent robotics and sensors; machine learning;
and application for smart buildings and smart cities.
This Special Issue received 25 submissions, and after a thorough and competitive review process,
only nine articles were published. The accepted articles were co-authored by 41 authors in total from
14 countries, with good geographical diversity. Performing an overview of the accepted articles,
we grouped them into three main categories focused on data acquisition, modern data-processing
algorithms, and computing architectures that support data processing.
Intelligent data acquisiton and data collection platforms: this section highlights three articles
that focus on data acqusition from sensors and sensor platforms in electrical vehicles, precision
agriculture, and educational robotics applications.
In [1], the authors provided an experimental evaluation of time delay estimation methods for
current measurements in electrical vehicle powertrains. These include linear regression (LR), variance
minimization (VM), and adaptive filters (AF). The methods were benchmarked in terms of Root Mean
Squared Error (RMSE) and average run-time on data collected from realistic driving profiles. Given
ix](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-14-320.jpg)
![its efficiency, the VM method was recommended for this application considering noise resistance
and computational efficiency. This study is highly relevant as it also considers the computational
constraints of distributed electronic control units (ECU) in modern automobiles.
Ref. [2] presented a system that combines scalar, ground-level measurements from sensor
networks with aerial robotic platforms (UAVs) as a data-collection and communication-relay
infrastructure. The main innovation lies in the network data aggregation by consensus in order to
reduce the need for data transmissions. The flight path of the UAV was optimized using spline
functions in order to maximize the flight time and data gathering in a given area from the cluster
heads. A symbolic aggregation approximation (SAX) method encoded the raw sensor measurements
into text strings for each of the monitored parameters in the precision agriculture application. The
results quantify both the data reduction as well as the UAV performance improvement using the
trajectory control scheme.
A sensor-intensive robotic platform for education was introduced by [3]. A thorough review
of existing educational robotic platforms was carried out alongside the argument for the Robotic
Operating System (ROS) as a framework of choice for the underlying implementation and new
developments. The system was built around a Raspberry Pi-embedded development board with
dedicated function specific sensors such as ultrasonic sensors, wheel encoders, LIDAR, and a camera.
It can be controlled over a WiFi communication interface using a PC or a mobile phone. Its
performance was experimentally validated through a series of tests as well as in qualitative studies
using various groups of students.
Learning from data and optimization: this section focuses on four articles that cover statistical
and machine learning and optimization techniques that make use of collected sensor data together
with advanced computing algorithms for implementation.
In [4], a detailed study was provided for smart building system identification combining both
classical, such as nonlinear autoregressive (NARX) models, and data-driven machine learning-based
approaches, such as multi-layer perceptrons (MLP). The goal was to achieve an accurate model of
building thermal dynamics for control using collected data under various conditions. The data
were collected through an existing building automation system (BAS) infrastructure with wireless
components for sensing, room temperature sensors, and controlled motor valves for the heating
elements. The study was carried out on a real building at the University Paris-Est Creteil (UPEC). The
main results showed good performance in terms of Mean Squared Error (MSE) and Mean Absolute
Error (MAE), less than 0.2C compared to the ground truth.
Reference [5] introduced an improved deep learning method for the evaluation and classification
of road quality based on the U-Net deep autoencoder. The main innovation stemmed from
the addition of residual connections, atrous spatial pyramid pooling, and attention gates to
increase performance. An evaluation was performed on multiple reference benchmarking datasets:
CrackForest, Crack500, and GAPs384. The Dice score was used to compare various architectures
and parametrization options of the deep learning architectures with a robust improvement in the
proposed ResU-NET + ASPP + AG network over the baseline specfication. Testing was performed
on dedicated graphical processing units on each dataset, while mixed dataset training did not yield
consistent results.
A mixed integer linear programming optimization problem for smart parking EV charging was
formulated by [6]. The goals were two-fold: to maximize the parking lot revenue by accommodating
charging EVs as efficiently as possible and by minimizing the cost of power consumption through
x](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-15-320.jpg)
![participation in the utility-level demand response (DR) program. The study was validated in a
simulation using predefined schedules and model EV charging characteristics. A simplified convex
relaxation technique was introduced to ensure the feasibility of the optimization problem. The
solution was compared against a standard variable charging power approach and showed consistent
improvements in terms of power consumption cost and percentage savings.
The authors of [7] proposed a new scheme for data classification by combining sparse
auto-encoders (SAE) with data postprocessing using a nature-inspired particle swarm optimization
(PSO) algorithm. The postprocessing layer improved the classification performance of the deep
neural network through a parameter optimized linear model. The labeled datasets used for practical
evaluation stemmed from the medical field and included epileptic seizures, SPECTF, and cardiac
arrhythmias while experimenting with multiple parameters of both the neural network and the PSO
algorithm. The adjustment layer improved the performance of the models, as illustrated through
other documented studies from the literature, achieving for example a 99.27% accuracy on the cardiac
arrhythmia dataset.
Advanced computing systems for data processing: this section includes two articles that present
improvements to data processing through optimized architectures for solving differential equations
and reconfigurable computing with FPGAs.
Reference [8] approached the problem of increasing the performance when solving ordinary
differential equations (ODE) on multi-core embedded systems, which can describe the system model
of certain physical phenomena. The authors introduced an adaptive algorithm, PAMCL, based on
the Adams–Moulton and Parareal methods and provided a comparison with existing approaches.
The implementation-wise OpenCL platform was used with optimized solvers for both CPU and GPU
systems. Quantitative results were reported, which include the CPU run time, GPU speed-up, and the
memory footprints of the reference algorithms. This method showed good results and achieved full
convergence to the exact solutions. A potential extension of the PAMCL method for partial derivative
equations was described.
Power-oriented monitoring of clock signals in FPGA systems was described by [9]. The
argumentation of the work lays out the need to reduce the power consumption and checkability
of reconfigurable computing platforms. The study included two types of power-monitoring:
the detection of synchronization failures, and the dissipation of power using temperature and
current sensors. The experiments were carried out on typical computing tasks, e.g., digital
filter implementations, using standardized tools for monitoring and data collection. The thermal
and power dissipation data were associated with fault conditions in the synchronization. Such
improvements to the evaluation of FPGA systems are highly relevant for critical and highly reliable
applications.
References:
1. Pfeiffer, J.; Wu, X.; Ayadi, A. Evaluation of Three Different Approaches for Automated Time
Delay Estimation for Distributed Sensor Systems of Electric Vehicles. Sensors 2020, 20, 351. [Google
Scholar] [CrossRef] [PubMed]
2. Popescu, D.; Stoican, F.; Stamatescu, G.; Ichim, L.; Dragana, C. Advanced UAV–WSN System
for Intelligent Monitoring in Precision Agriculture. Sensors 2020, 20, 817. [Google Scholar] [CrossRef]
[PubMed]
3. Karalekas, G.; Vologiannidis, S.; Kalomiros, J. EUROPA: A Case Study for Teaching Sensors,
Data Acquisition and Robotics via a ROS-Based Educational Robot. Sensors 2020, 20, 2469. [Google
xi](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-16-320.jpg)
![Scholar] [CrossRef] [PubMed]
4. Sadeghian Broujeny, R.; Madani, K.; Chebira, A.; Amarger, V.; Hurtard, L. Data-Driven
Living Spaces’Heating Dynamics Modeling in Smart Buildings using Machine Learning-Based
Identification. Sensors 2020, 20, 1071. [Google Scholar] [CrossRef] [PubMed]
5. Augustauskas, R.; Lipnickas, A. Improved Pixel-Level Pavement-Defect Segmentation Using
a Deep Autoencoder. Sensors 2020, 20, 2557. [Google Scholar] [CrossRef] [PubMed]
6. Jawad, M.; Qureshi, M.B.; Ali, S.M.; Shabbir, N.; Khan, M.U.S.; Aloraini, A.; Nawaz, R. A
Cost-Effective Electric Vehicle Intelligent Charge Scheduling Method for Commercial Smart Parking
Lots Using a Simplified Convex Relaxation Technique. Sensors 2020, 20, 4842. [Google Scholar]
[CrossRef] [PubMed]
7. Karim, A.M.; Kaya, H.; Güzel, M.S.; Tolun, M.R.; Çelebi, F.V.; Mishra, A. A Novel Framework
Using Deep Auto-Encoders Based Linear Model for Data Classification. Sensors 2020, 20, 6378.
[Google Scholar] [CrossRef] [PubMed]
8. Tavakkoli, V.; Mohsenzadegan, K.; Chedjou, J.C.; Kyamakya, K. Contribution to Speeding-Up
the Solving of Nonlinear Ordinary Differential Equations on Parallel/Multi-Core Platforms for
Sensing Systems. Sensors 2020, 20, 6130. [Google Scholar] [CrossRef] [PubMed]
9. Drozd, O.; Nowakowski, G.; Sachenko, A.; Antoniuk, V.; Kochan, V.; Drozd, M.
Power-Oriented Monitoring of Clock Signals in FPGA Systems for Critical Application. Sensors 2021,
21, 792. [Google Scholar] [CrossRef] [PubMed]
Grigore Stamatescu, Anatoliy Sachenko, Dan Popescu
Editors
xii](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-17-320.jpg)
![sensors
Article
Evaluation of Three Different Approaches for
Automated Time Delay Estimation for Distributed
Sensor Systems of Electric Vehicles
Jakob Pfeiffer 1,2,*, Xuyi Wu 2 and Ahmed Ayadi 2
1 BMW Group, Petuelring 130, 80788 Munich, Germany
2 Department of Electrical and Computer Engineering, Technical University of Munich, Arcisstr. 21,
80333 Munich, Germany; Xuyi.Wu@tum.de (X.W.); Ahmed.Ayadi@tum.de (A.A.)
* Correspondence: Jakob.J.Pfeiffer@bmwgroup.com
Received: 3 December 2019; Accepted: 26 December 2019; Published: 8 January 2020
Abstract: Deviations between High Voltage (HV) current measurements and the corresponding real
values provoke serious problems in the power trains of Electric Vehicles (EVs). Examples for these
problems have inaccurate performance coordinations and unnecessary power limitations during
driving or charging. The main reason for the deviations are time delays. By correcting these delays
with accurate Time Delay Estimation (TDE), our data shows that we can reduce the measurement
deviations from 25% of the maximum current to below 5%. In this paper, we present three different
approaches for TDE. We evaluate all approaches with real data from power trains of EVs. To enable
an execution on automotive Electronic Control Units (ECUs), the focus of our evaluation lies not
only on the accuracy of the TDE, but also on the computational efficiency. The proposed Linear
Regression (LR) approach suffers even from small noise and offsets in the measurement data and
is unsuited for our purpose. A better alternative is the Variance Minimization (VM) approach. It is
not only more noise-resistant but also very efficient after the first execution. Another interesting
approach are Adaptive Filters (AFs), introduced by Emadzadeh et al. Unfortunately, AFs do not
reach the accuracy and efficiency of VM in our experiments. Thus, we recommend VM for TDE of
HV current signals in the power train of EVs and present an additional optimization to enable its
execution on ECUs.
Keywords: automotive; current; electric power train; electric vehicle; embedded systems; delay;
detection; distributed systems; measurements; power train; sensor; signals; time delay estimation
1. Introduction
Political guidelines in various countries to decarbonize individual mobility led to an exponential
growth of Electric Vehicles (EVs) in offers and sales. However, one obstacle for the success of EVs is the
so-called range anxiety [1]. Customers are afraid that an EV is not able to provide the range they need
for all of their journeys. To combat range anxiety and increase the range of EVs, there are two different
ways. The first one is to simply increase the size of the High Voltage Battery (HVB). Unfortunately,
this means to increase the size of the most expensive component of an EV, and after all, it is not a very
sustainable way. The second way, which is our solution of choice, is to make EVs more efficient.
Kirchhoff’s current law states that the sum of all currents at a node of an electric system is equal to
0 A. However, considering measurement signals of nodes in the power trains of EVs with distributed
sensor systems, the sum of all currents can differ up to 20 % of the maximum current (see Figure 1).
If we look closer at the Root Mean Square Error (RMSE) of the sum of currents RMSE(isum) = 0.67%,
we realize that it has the same value as the mean current of the DCDC converter µiDCDC
= 0.67%.
1](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-18-320.jpg)
![Sensors 2020, 20, 351
0 1000 2000 3000 4000 5000 6000 7000
Time Step [10 ms]
-100
-80
-60
-40
-20
0
20
40
60
80
100
Relative
Current
[%]
i
HVB
i
EM
i
heat
i
cool
i
DCDC
i
sum
Figure 1. Currents of all HV components in an EV on a test drive. The sum of all currents isum is
plotted in black. According to Kirchhoff’s current law, it should be constantly 0 %. However, looking
at the measurements shows that the deviation isum is higher than the current of the DCDC converter
iDCDC. Even its noise spectrum is approximately half as high as the consumption of the heating iheat,
which is the second largest consumer in this drive.
A different value than 0 A for the sum of all currents indicates that there is a divergence between
measurements and real values. The divergence becomes problematic when the power train is operating
close to the system boundaries. For example, there are boundaries for the protection of the HVB.
The HVB is only capable of discharging or charging a restricted amount of power. Higher amounts
would threaten the HVB’s lifetime and safety [2]. To ensure a safe operation mode even for high
divergences between measurements and real values, additional protection offsets (see Figure 2) might
be added to the boundaries, although they have some drawbacks.
t
i Maximum
Battery Current
Measurement
Additional Battery
Protection Offset
Measurement
Tolerance
Figure 2. A simplified example of offsets for protection of the HVB. The measured value (black)
differs from the real value in the range of some tolerance (grey). To prevent exceeding the battery
limit (red, solid) even under the worst measurement conditions, an additional battery protection offset
(red, dashed) is introduced. The same principle is used analogously with negative currents. It can be
extended to other High Voltage (HV) components.
2](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-19-320.jpg)
![Sensors 2020, 20, 351
For example, in the charging case, most notably during recuperation, the HVB might not allow the
full power level, even though it would be capable of handling it. Thus, the amount of power charged
to the battery is restricted and the EV loses cruising range while its power consumption increases.
In the opposite case, the system might not release requested power, although the HVB could provide it
in reality. This additional restriction of power decreases the EV’s performance. As can be seen from
the two examples above, minimizing the magnitude of the protection offsets also allows increasing the
performance as the efficiency and the cruising range of EVs.
0 1000 2000 3000 4000 5000 6000 7000
Time Step [10 ms]
-100
-80
-60
-40
-20
0
20
40
60
80
100
Relative
Current
[%]
iHVB
(moved by 6 time steps) iEM
iheat
icool
iDCDC
isum
Figure 3. The same test drive as in Figure 1 but with the battery current iHVB
(green) shifted by six
time steps. The sum of all currents isum (black) is significantly closer to 0 %.
Besides measurement faults and sensor uncertainties [3], the divergence between measurements
and real values is caused by time delays. Figure 3 shows an example of the sum of all currents isum
being reduced by shifting a signal by 6 time steps. The delays result from distributed sensor systems in
the power train as plotted in Figure 4. The High Voltage (HV) components have their own Electronic
Control Unit (ECU) which is connected with the current sensors and processes the sensor information.
The ECUs exchange this information via bus systems. The buses require individual amounts of time
to send the measurement signals. Thus, from an ECU’s point of view, the sensor information from
other ECUs arrives with individual delays (see the Ego ECU in Figure 4). These individual delays
could be compensated easily with a synchronized clock and time stamps as part of each bus message.
However, this solution would have two drawbacks. First, it would increase the bus traffic as not only
the measurement information must be carried by the messages but also the time stamp. As a result,
the EV would either require a faster bus which is able to transport more information, or it would have
to reduce the information exchanged between the ECUs. Second, there exists no clock in the power
trains of modern series EVs which is synchronized with all ECUs at the same frequency as the message
exchange. Usually, the ECUs are synchronized in a longer time frame than they communicate. Thus,
the time stamp solution would require additional or higher performing hardware and increase the
costs for the production of the EV.
3](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-20-320.jpg)
![Sensors 2020, 20, 351
…
𝑆1,1
𝑆1,𝑛
𝑆1,2
…
𝑆𝑚,1
𝑆𝑚,𝑛
𝑆𝑚,2
…
Legend:
Sensor ECU
Δ𝑡
Δ𝑡 Δ𝑡
Δ𝑡
Δ𝑡
Δ𝑡 Δ𝑡
Δ𝑡
𝑆𝑖,𝑗
Data Processing
and Transmission
(e.g. Bus)
Δ𝑡
Time
Delay
…
𝑆𝐸𝑔𝑜 1
𝑆𝐸𝑔𝑜 𝑛
𝑆𝐸𝑔𝑜 2
Figure 4. A schematic example of an automotive bus system with higlighted sources of time delays.
Please note that the time delays are highly individual and not necessarily equal, but constant or only
slowly changing. The ECUs can be connected directly or indirectly via other ECUs. The Ego ECU
is not able to reconstruct the time delays, because it only knows the received measurement values
and their last sender. It has no further information about the time passed since the measurement’s
original creation.
The aim of this work is to automatically detect the time delay between measurement signals
from different sensors without additional hardware. For this purpose, we develop two different
approaches. One of them is based on Linear Regression (LR), whereas the other one optimizes the
estimated variance of the difference between several signals. We compare our approaches to other
state-of-the-art Time Delay Estimation (TDE) algorithms and evaluate them with a focus on precision
and run-time efficiency. Apart from allowing a more accurate power distribution, the automated
TDE helps to reduce the battery protection offset and thus to increase the performance, efficiency and
cruising range of EVs.
The rest of this paper is structured as follows. Section 2 states related work and the similarities
and differences to our work. Furthermore, Section 2 highlights the contributions of our work to the
state of the art. In Section 3, we explain the theory behind our work before we describe the practical
experiments in Section 4. The experiments’ results, stated in Section 5, show us the performance of
the algorithms for our use case. Based on this evaluation, we take the best performing algorithm
and optimize it further. The optimization steps can be taken from Section 3.4 and their impacts to
the results from Section 5.4. In Section 6, we discuss the advantages and drawbacks of all proposed
concepts. Finally, we draw our conclusions and give a short outlook in Section 7.
2. State of the Art
There exists plenty of literature about TDE, although—to the best of our knowledge—none
of them is tailored to the specific problem of TDE of current signals in EVs. In the following, we
present several publications about TDE from different fields of application, such as embedded systems,
acoustics, medicine, positioning, aeronautics, process technology, and robotics.
An approach which also deals with EVs and time delays is the one by Guo et al. [4]. However,
their approach is similar to ours only at the first look. Their goal is to stabilize a grid of electric sources
and sinks with EVs. For the stabilization of the grid, they propose time delay resistent control strategies
4](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-21-320.jpg)
![Sensors 2020, 20, 351
of smart grids with EVs. The EVs are able to charge bidirectionally. The bidirectional charging is used
to smooth disturbances and respond rapidly to fast occurring changes in the power distribution of
the grid. An example for such a rapidly occurring change in the times of renewable energies is the
power output of wind turbines when a strong wind occurs. Compared to our approach, Guo’s focus
is rather on the control strategy than on the TDE. Another difference with our work is that Guo’s
system is rather macroscopic with lots of different elements and many EVs in the grid. Our system
is instead quite microscopic. We consider a single EV with a power train of around five sources and
sinks. Our communication network might be smaller than the number of HV components as some
consumers might share the same ECU. For example, the heating and the cooling component of an EV
use both the climate control ECU for bus communication.
Kali et al. [5] propose a controller with TDE for Electric Machines (EMs). The TDE is executed
state-based with the help of a model of the EM. The model design demands expert knowledge about
the physical principles of an EM. This is justified for Kali et al. as they require the same knowledge
for their controller. However, for our case, we want to be able to estimate the time delays without
further knowledge about the HV components. Our TDE shall be executable with nothing else than the
available measurement data.
Zeng et al. [6] introduce a statistical approach to predict the delay of a bus message. The content
of the messages does not need to be known to achieve high accuracy. This is different from our scenario
where we want to make use of the information carried by the message. In contrast to Zeng et al., we do
not require predicting the time delay accurately to milliseconds. For our purposes, an estimation of
the number of delayed discrete time steps is sufficient.
Not from the field of electric power trains or bus communication, but from acoustics is the
approach shown by Lourtie and Moura [7]. They use a stochastic approach to model time delays in
an acoustic path environment. Like ours, their environment consists of several sources. However,
in contrast to our scenario, the delay they want to estimate varies with time. In our case, we assume
the time delay to be constant in a short time frame. For longer periods, it might change slowly.
The reason for the slowly changing time delay is that it is caused during the wake up procedure of
the EV. The ECUs wake up in an unsynchronized way. Afterwards, the ECUs are synchronized on a
relatively large time frame (e.g., 1 s), but work based on short time steps (e.g., 10 ms).
Another acoustics application for TDE is shown by He et alii [8]. They use the so-called
Multichannel Cross-Correlation Coefficient algorithm to estimate time delays of speech sources in
noisy and reverberant environments.
Svilainis et al. [9] present another interesting approach. Their goal is to estimate the time passed
between emitting an ultrasonic signal and absorbing its reflection. Like Zeng et al., they require high
precision. Another difference to our approach is that their algorithms make use of the pulse form of
ultrasonic signals. Our signal as plotted in Figure 1 can vary in a large range and does not necessarily
contain pulses (e.g., after time step 5,000).
Mirzaei et al. expand TDE for ultrasonic signals to the field of medicine [10]. The authors introduce
a window-based TDE approach to estimate the time passed between two frames of radio-frequency
data. They compare the results of the new window-based approach to their previously developed,
optimization-based method [11] and to Normalized Cross-Correlation.
Recently, Garcez et al. published their work on a similar problem to ours, but in a completely
different field of application [12]. Like bus systems of EVs, Global Navigation Satellite Systemss
(GNSSs) systems have real-time requirements. Their goal is to minimize deviations between
measurements and real position data. The time delays are caused during the transmission of GNSS
messages, when the signals do not take straight lines of sight, but are reflected on their way or suffer
from noise. The authors propose a tensor-based subspace tracking algorithm to efficiently estimate
time delays of received GNSS signals.
A similar approach is presented by Xie et al. for an indoor positioning sensing system [13].
They sense positions of mobile devices based on the signal strength and the signal’s time delay since
5](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-22-320.jpg)
![Sensors 2020, 20, 351
its transmission from a base station. For the TDE, Xie et al. combine Cross-Correlation with Quadratic
Fitting. This is similar to our LR approach (see Section 3.2), where we try to fit the signals with
quadratic functions to retrieve the delay between them. Like Garcez et al., they have to deal with
the problem that the signals are often reflected and do not take direct lines of sight. Different to
Garcez et al., Xie’s approach takes the strength of the signal into account for retrieving a more exact
position estimation. For our work, we cannot take advantage of this information, because in wire-based
bus systems all signals are equally strong.
Schmidhammer et al. estimate positions of moving, non-cooperative objects in vehicular
environments [14]. Their idea is to estimate the position of an object based on time delays in a
network of distributed receiving and transmitting nodes. In contrast to our work, the networking
nodes of Schmidhammer et al. are not necessarily on-board the vehicle, but can also be mounted on
the road infrastructure.
Emadzadeh et al. [15] show an inspiring approach for detecting the relative position of spacecrafts.
For retrieving the position, they examine an X-ray signal received by two spacecrafts and determine
the time delay between them. For the TDE, they use Adaptive Filters (AFs). This approach seems
very promising to us. We implement the algorithms of Emadzadeh et al. and compare them to ours in
order to find out if their approach can be transferred from X-ray signals to current measurements in
the power train of EVs.
Like Emadzadeh et al., Liu et al. focus on AFs [16]. Compared to our problem of fixed or only
slowly changing time delays, the difference in Liu et al. is that they deal with time-varying time delays.
That makes further processing steps necessary. For example, they require a transition probability matrix
and an initial probability distribution vector to model the time delay changes with a Markov chain.
Park et al. analyze time series data with Autoencoders and Long Short-Term Memory Neural
Networks (LSTMs) to detect faults in industrial processes [17]. The authors emphasize the importance
of TDE for correct fault detection. However, they focus only on time delays caused by their own
fault detection system. Our focus lies on earlier steps in the processing chain. We want to detect time
delays between the input signals before they are passed to other computation processes. Furthermore,
we want to implement algorithms which are able to learn on-board the automotive ECUs and adapt
themselves to new data. As the training of Neural Networks is quite memory intensive and demands
high computational power, they do not belong to our methods of choice.
Close to the application field of industrial processes is the approach of Srinivasa Rao et al. [18].
In their recent article, the authors propose fuzzy parametric uncertainty to mathematically model
systems with time delays. Their goal is to enable a robust controller design. For this purpose, they first
approximate the time delay system as an interval system. After retrieving the intervals, they design an
optimal controller for these. Like Guo et al., Srinivasa Rao et al. focus on how to retrieve an optimal
controller, which is not part of our work. Although they focus on the control of industrial processes,
their article is very general. Besides industrial plants, they also mention potential fields of application,
such as EMs or robot manipulators.
Time delay compensation for robots is the focus of Shen et al [19]. Their focus is on teleoperating
robots which require knowledge about the time delay between the master and the slave robot for
stable operation. The robots and their communication channels are modeled as extended dynamical
system. For this system, Shen et al. develop a cascade observer which is able to control it in a stable
way. The authors assume that a sufficiently accurate value for the TDE is given and concentrate on its
compensation. This is different to our work here. We explicitly want to estimate the time delay.
You et al. develop a proportional multiple integral observer for fuzzy systems [20]. The goal
of their work is the same as ours. They want to minimize deviations between measurements and
real values caused by time delays and measurement inaccuracies. Their time delays are also varying.
Unlike the varying time delays presented before, the ones of You et al. do not vary with time but rather
with states. Their focus is also on industrial processes and not on electric power trains. However, the
6](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-23-320.jpg)
![Sensors 2020, 20, 351
main difference between our works is that You et al. want to minimize time delays and measurement
inaccuracies with the same system.
Our approach follows the divide and conquer strategy and faces the two problems separately.
We focus on the problem of measurement deviations caused by measurement inaccuracies in our
previous work [3]. However, measurement inaccuracies are not part of this work. Here, we assume that
the measurements are appropriately accurate and that the main deviations are caused by time delays
as shown in Figure 1 and Figure 3. Thus, TDE is our solution of choice to minimize the deviations.
Our contribution in this article is the development of a regression-based approach and an
algorithm based on Variance Minimization (VM) for TDE as first presented in [21]. We transfer the
ideas introduced by Emadzadeh et al. to the domain of currents in the HV system of EVs and compare
the results to our approaches in matters of accuracy and computational performance. Our TDE works
only with the data available in modern series EVs and does not require an additional clock. In addition
to [21], we introduce an optimization of the most accurate and efficient of our evaluated approaches.
We further evaluate the optimization both on artificially created data with known ground truth as well
as real drive data with unknown ground truth.
3. Concepts
In this section, we introduce the algorithms and shortly explain the concepts from other authors
which we implement and compare for TDE. From now on, for the sake of easier understanding,
we focus on the current of the EM iEM and the HVB iHVB (without other consumers than the EM)
as examples. Nevertheless, the proposed methods can be extended to every current signal in the
HV system of an EV. Furthermore, we inverse the sign of iHVB from now on to make its shape similar
to the one of the EM. Thus, we can treat the HVB current signal as a delayed or preceded version of
the EM, respectively.
Our goal is to find the time delay td in a bus system which can be described as
x1(t) = i1(t) + n1(t)
x2(t) = i2(t − td) + n2(t − td),
(1)
where t stands for the time step, x1(t) is the measurement signal of the faster component, x2(t) describes
the slower component’s signal, i1(t) and i2(t) describe the corresponding currents and n1(t) and n2(t)
are noise terms [15]. As we cannot retrieve the currents i1(t) and i2(t) directly, we cannot minimize
the difference between i1(t) and i2(t). Instead, we directly minimize the difference between the two
measurement signals x1(t) and x2(t).
3.1. Adaptive Filter
The idea of Emadzadeh et al. is to model the time delay as Finite Impulse Response (FIR) filter.
They define x1(t) to be the faster signal. For each measurement x2(ti) at time step ti, they collect a
row of the last M measurements of the other signal
x1(ti − M + 1 : ti) = [x1(ti − M + 1), x1(ti − M + 2), . . . x1(ti − 1), x1(ti)] . (2)
Then, the authors search for an optimal channel impulse response vector ω∗ such that the deviation
between x2(ti) and x1(ti − M + 1 : ti)ω∗ becomes minimal. Mathematically, this can be expressed by
the minimization of the expectation value of the Mean Squared Error (MSE) between the measurement
value of the slower signal and the filtered measurement row of the faster signal. It results in the formula
ω∗
= argmin
ω
E
h
(x2(ti) − x1(ti − M + 1 : ti)ω)2
i
. (3)
7](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-24-320.jpg)
![This ebook is for the use of anyone anywhere in the United States
and most other parts of the world at no cost and with almost no
restrictions whatsoever. You may copy it, give it away or re-use it
under the terms of the Project Gutenberg License included with this
ebook or online at www.gutenberg.org. If you are not located in the
United States, you will have to check the laws of the country where
you are located before using this eBook.
Title: Index of the Project Gutenberg Works of Jack London
Author: Jack London
Editor: David Widger
Release date: May 8, 2019 [eBook #59461]
Most recently updated: July 7, 2019
Language: English
Credits: Produced by David Widger
*** START OF THE PROJECT GUTENBERG EBOOK INDEX OF THE
PROJECT GUTENBERG WORKS OF JACK LONDON ***](https://image.slidesharecdn.com/25327293-250524122405-6514e2db/85/Convergence-Of-Intelligent-Data-Acquisition-And-Advanced-Computing-Systems-Grigore-Stamatescu-30-320.jpg)
Convergence Of Intelligent Data Acquisition And Advanced Computing Systems Grigore Stamatescu
- 1. Convergence Of Intelligent Data Acquisition And Advanced Computing Systems Grigore Stamatescu download https://ebookbell.com/product/convergence-of-intelligent-data- acquisition-and-advanced-computing-systems-grigore- stamatescu-50654586 Explore and download more ebooks at ebookbell.com
- 2. Here are some recommended products that we believe you will be interested in. You can click the link to download. Convergence Of Big Data Technologies And Computational Intelligent Techniques Govind P Gupta https://ebookbell.com/product/convergence-of-big-data-technologies- and-computational-intelligent-techniques-govind-p-gupta-47573698 International Conference On Intelligent Data Communication Technologies And Internet Of Things Icici 2018 1st Ed Jude Hemanth https://ebookbell.com/product/international-conference-on-intelligent- data-communication-technologies-and-internet-of-things-icici-2018-1st- ed-jude-hemanth-7321164 Intelligent Computation In Big Data Era International Conference Of Young Computer Scientists Engineers And Educators Icycsee 2015 Harbin China January 1012 2015 Proceedings 1st Edition Hongzhi Wang https://ebookbell.com/product/intelligent-computation-in-big-data-era- international-conference-of-young-computer-scientists-engineers-and- educators-icycsee-2015-harbin-china-january-1012-2015-proceedings-1st- edition-hongzhi-wang-4976076 Proceedings Of The Fourth Eurochina Conference On Intelligent Data Analysis And Applications Alba https://ebookbell.com/product/proceedings-of-the-fourth-eurochina- conference-on-intelligent-data-analysis-and-applications-alba-6753986
- 3. Proceedings Of The Fifth Eurochina Conference On Intelligent Data Analysis And Applications 1st Ed Pavel Krmer https://ebookbell.com/product/proceedings-of-the-fifth-eurochina- conference-on-intelligent-data-analysis-and-applications-1st-ed-pavel- krmer-7321374 Intelligent Data Engineering And Analytics Proceedings Of The 10th International Conference On Frontiers In Intelligent Computing Theory And Applications Ficta 2022 Vikrant Bhateja https://ebookbell.com/product/intelligent-data-engineering-and- analytics-proceedings-of-the-10th-international-conference-on- frontiers-in-intelligent-computing-theory-and-applications- ficta-2022-vikrant-bhateja-49171398 Intelligent Data Analysis And Applications Proceedings Of The Second Eurochina Conference On Intelligent Data Analysis And Applications Ecc 2015 Abraham https://ebookbell.com/product/intelligent-data-analysis-and- applications-proceedings-of-the-second-eurochina-conference-on- intelligent-data-analysis-and-applications-ecc-2015-abraham-22090462 Intelligent Data Analysis And Applications Proceedings Of The Third Eurochina Conference On Intelligent Data Analysis And Applications Ecc 2016 1st Edition Jengshyang Pan https://ebookbell.com/product/intelligent-data-analysis-and- applications-proceedings-of-the-third-eurochina-conference-on- intelligent-data-analysis-and-applications-ecc-2016-1st-edition- jengshyang-pan-5675896 Intelligent Data Analysis And Its Applications Volume Ii Proceeding Of The First Eurochina Conference On Intelligent Data Analysis And Applications June 1315 2014 Shenzhen China 1st Edition Jengshyang Pan https://ebookbell.com/product/intelligent-data-analysis-and-its- applications-volume-ii-proceeding-of-the-first-eurochina-conference- on-intelligent-data-analysis-and-applications-june-1315-2014-shenzhen- china-1st-edition-jengshyang-pan-4697408
- 5. Edited by Convergence of Intelligent Data Acquisition and Advanced Computing Systems Grigore Stamatescu, Anatoliy Sachenko and Dan Popescu Printed Edition of the Special Issue Published in Sensors www.mdpi.com/journal/sensors
- 6. Convergence of Intelligent Data Acquisition and Advanced Computing Systems
- 8. Convergence of Intelligent Data Acquisition and Advanced Computing Systems Editors Grigore Stamatescu Anatoliy Sachenko Dan Popescu MDPI • Basel • Beijing • Wuhan • Barcelona • Belgrade • Manchester • Tokyo • Cluj • Tianjin
- 9. Editors Grigore Stamatescu Automation and Industrial Informatics University Politehnica of Bucharest Bucharest Romania Anatoliy Sachenko Information Computer Systems and Control West Ukrainian National University Ternopil Ukraine Dan Popescu Automation and Industrial Informatics University Politehnica of Bucharest Bucharest Romania Editorial Office MDPI St. Alban-Anlage 66 4052 Basel, Switzerland This is a reprint of articles from the Special Issue published online in the open access journal Sensors (ISSN 1424-8220) (available at: www.mdpi.com/journal/sensors/special issues/IDAACS2019). For citation purposes, cite each article independently as indicated on the article page online and as indicated below: LastName, A.A.; LastName, B.B.; LastName, C.C. Article Title. Journal Name Year, Volume Number, Page Range. ISBN 978-3-0365-1656-1 (Hbk) ISBN 978-3-0365-1655-4 (PDF) © 2021 by the authors. Articles in this book are Open Access and distributed under the Creative Commons Attribution (CC BY) license, which allows users to download, copy and build upon published articles, as long as the author and publisher are properly credited, which ensures maximum dissemination and a wider impact of our publications. The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commons license CC BY-NC-ND.
- 10. Contents About the Editors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii Preface to ”Convergence of Intelligent Data Acquisition and Advanced Computing Systems” ix Jakob Pfeiffer, Xuyi Wu and Ahmed Ayadi Evaluation of Three Different Approaches for Automated Time Delay Estimation for Distributed Sensor Systems of Electric Vehicles Reprinted from: Sensors 2020, 20, 351, doi:10.3390/s20020351 . . . . . . . . . . . . . . . . . . . . . 1 Dan Popescu, Florin Stoican, Grigore Stamatescu, Loretta Ichim and Cristian Dragana Advanced UAV–WSN System for Intelligent Monitoring in Precision Agriculture Reprinted from: Sensors 2020, 20, 817, doi:10.3390/s20030817 . . . . . . . . . . . . . . . . . . . . . 19 Georgios Karalekas, Stavros Vologiannidis and John Kalomiros EUROPA: A Case Study for Teaching Sensors, Data Acquisition and Robotics via a ROS-Based Educational Robot Reprinted from: Sensors 2020, 20, 2469, doi:10.3390/s20092469 . . . . . . . . . . . . . . . . . . . . 45 Roozbeh Sadeghian Broujeny, Kurosh Madani, Abdennasser Chebira, Veronique Amarger and Laurent Hurtard Data-Driven Living Spaces’ Heating Dynamics Modeling in Smart Buildings using Machine Learning-Based Identification Reprinted from: Sensors 2020, 20, 1071, doi:10.3390/s20041071 . . . . . . . . . . . . . . . . . . . . 63 Rytis Augustauskas and Arūnas Lipnickas Improved Pixel-Level Pavement-Defect Segmentation Using a Deep Autoencoder Reprinted from: Sensors 2020, 20, 2557, doi:10.3390/s20092557 . . . . . . . . . . . . . . . . . . . . 79 Muhammad Jawad, Muhammad Bilal Qureshi, Sahibzada Muhammad Ali, Noman Shabbir, Muhammad Usman Shahid Khan, Afnan Aloraini and Raheel Nawaz A Cost-Effective Electric Vehicle Intelligent Charge Scheduling Method for Commercial Smart Parking Lots Using a Simplified Convex Relaxation Technique Reprinted from: Sensors 2020, 20, 4842, doi:10.3390/s20174842 . . . . . . . . . . . . . . . . . . . . 101 Ahmad M. Karim, Hilal Kaya, Mehmet Serdar Güzel, Mehmet R. Tolun, Fatih V. Çelebi and Alok Mishra A Novel Framework Using Deep Auto-Encoders Based Linear Model for Data Classification Reprinted from: Sensors 2020, 20, 6378, doi:10.3390/s20216378 . . . . . . . . . . . . . . . . . . . . 121 Vahid Tavakkoli, Kabeh Mohsenzadegan, Jean Chamberlain Chedjou and Kyandoghere Kyamakya Contribution to Speeding-Up the Solving of Nonlinear Ordinary Differential Equations on Parallel/Multi-Core Platforms for Sensing Systems Reprinted from: Sensors 2020, 20, 6130, doi:10.3390/s20216130 . . . . . . . . . . . . . . . . . . . . 143 Oleksandr Drozd, Grzegorz Nowakowski, Anatoliy Sachenko, Viktor Antoniuk, Volodymyr Kochan and Myroslav Drozd Power-Oriented Monitoring of Clock Signals in FPGA Systems for Critical Application Reprinted from: Sensors 2021, 21, 792, doi:10.3390/s21030792 . . . . . . . . . . . . . . . . . . . . . 161 v
- 12. About the Editors Grigore Stamatescu Grigore Stamatescu graduated from the University Politehnica of Bucharest (UPB) in 2009 and holds a PhD degree (2012) from the same university. He is currently an Associate Professor (Habil. 2019) with the Department of Automation and Industrial Informatics, Faculty of Automatic Control and Computers, UPB. His research interests include networked embedded sensing, the Internet of Things and distributed information processing in the industry, the built environment, and smart city applications. His recent research focused on statistical learning methods for load forecasting and anomaly detection in building energy traces, and data-driven modelling of large-scale manufacturing systems, with a focus on energy efficiency. His research has been published in over 130 articles. Dr. Stamatescu was a Fulbright Visiting Scholar in 2015–2016 at the University California, Merced, and a JESH Scholar of the Austrian Academy of Sciences in 2019. He is a Senior Member of IEEE. Anatoliy Sachenko Anatoliy Sachenko is a Professor and the Head of the Department of Information Computing Systems and Control, and a research advisor of the Research Institute for Intelligent Computer Systems, West Ukrainian National University. He earned his B.Eng. degree in Electrical Engineering at L’viv Polytechnic Institute in 1968, his PhD degree in Electrical Engineering at L’viv Physics and Mechanics Institute in 1978, his Doctor of Technical Sciences Degree in Electrical and Computer Engineering at Leningrad Electrotechnic Institute in 1988. Since 1991, he has been an Honored Inventor of Ukraine, and since 1993, he has been an IEEE Senior Member. His main areas of research interest are the implementation of artificial neural networks, distributed systems and networks, parallel computing, and intelligent controllers for automated and robotics systems. He has published over 450 papers in the areas listed above. Dan Popescu Dan Popescu was born in Bucharest, Romania, in 1950. He received his BS and MS equivalent (five-year engineering) degrees in Automatic Control and Computers from the University Politehnica of Bucharest in 1974; his BS and MS equivalent degrees (five-years) in mathematics from the University of Bucharest, Romania; and his PhD degree in automation and remote control from the University Politehnica of Bucharest in 1987. Since 2003, he has been a Full Professor with the Faculty of Automatic Control and Computers; a PhD supervisor and the Head of the Laboratory of Innovative Products and Processes for Increasing Quality of Life PRECIS Center, since 2016; and the Vice-Dean of Research Activity with the University Politehnica of Bucharest from 2012 to 2016. His main research interests include image processing and interpretation, neural networks, wireless sensor networks, control systems in industrial and robotic applications, and environmental monitoring. vii
- 14. Preface to ”Convergence of Intelligent Data Acquisition and Advanced Computing Systems” This preface briefly outlines the objectives of the Special Issue on “Convergence of Intelligent Data Acquisition and Advanced Computing Systems” published between September 2019 and September 2020 in the journal Sensors. This Special Issue welcomed submissions as extended versions of conference articles published in the proceedings of the “10th IEEE International Conference on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications (IDAACS’2019)”, allowing the authors to expand on their theoretical and experimental contributions. These were complemented with external submissions from interested researchers working in this timely area. We highlight the main contributions of the published articles, grouped by key topics: intelligent data acquisition, learning methods and optimization for data processing, and advanced computing systems to support data processing. As modern data-acquisition solutions increasingly integrate on-board computing in the case of intelligent sensors, in-network computing, or in the case of distributed systems and sensor networks, a need to provide a joint view of these previously divergent topics has been observed. We define the convergence of intelligent data acquisition and advanced computing systems at the interface of theory and applications for industrial manufacturing, scientific computing, precision agriculture, and energy systems such as smart building and smart city systems. This book bridges specific topics for instrumentation such as accuracy, sampling, time synchronization, sensor selection and calibration with algorithm design, statistical and machine learning, computational efficiency, and reconfigurable computing that supports conventional engineering tasks. Computing-as-a-service is another relevant approach that can be leveraged to improve measurements and instrumentation in the future. The potential topics of this Special Issue that were initially defined were mapped onto the core areas and special streams of the IDAACS’2019 conference and included but were not limited to advanced instrumentation and data-acquisition systems; advanced mathematical methods for data acquisition and computing systems; computational intelligence for instrumentation and data-acquisition systems; data analysis and modeling; intelligent distributed systems and remote control; intelligent information systems, data mining, and ontology; Internet of Things; pattern recognition, digital image, and signal processing; intelligent instrumentation and data acquisition systems in advanced manufacturing for Industry 4; intelligent robotics and sensors; machine learning; and application for smart buildings and smart cities. This Special Issue received 25 submissions, and after a thorough and competitive review process, only nine articles were published. The accepted articles were co-authored by 41 authors in total from 14 countries, with good geographical diversity. Performing an overview of the accepted articles, we grouped them into three main categories focused on data acquisition, modern data-processing algorithms, and computing architectures that support data processing. Intelligent data acquisiton and data collection platforms: this section highlights three articles that focus on data acqusition from sensors and sensor platforms in electrical vehicles, precision agriculture, and educational robotics applications. In [1], the authors provided an experimental evaluation of time delay estimation methods for current measurements in electrical vehicle powertrains. These include linear regression (LR), variance minimization (VM), and adaptive filters (AF). The methods were benchmarked in terms of Root Mean Squared Error (RMSE) and average run-time on data collected from realistic driving profiles. Given ix
- 15. its efficiency, the VM method was recommended for this application considering noise resistance and computational efficiency. This study is highly relevant as it also considers the computational constraints of distributed electronic control units (ECU) in modern automobiles. Ref. [2] presented a system that combines scalar, ground-level measurements from sensor networks with aerial robotic platforms (UAVs) as a data-collection and communication-relay infrastructure. The main innovation lies in the network data aggregation by consensus in order to reduce the need for data transmissions. The flight path of the UAV was optimized using spline functions in order to maximize the flight time and data gathering in a given area from the cluster heads. A symbolic aggregation approximation (SAX) method encoded the raw sensor measurements into text strings for each of the monitored parameters in the precision agriculture application. The results quantify both the data reduction as well as the UAV performance improvement using the trajectory control scheme. A sensor-intensive robotic platform for education was introduced by [3]. A thorough review of existing educational robotic platforms was carried out alongside the argument for the Robotic Operating System (ROS) as a framework of choice for the underlying implementation and new developments. The system was built around a Raspberry Pi-embedded development board with dedicated function specific sensors such as ultrasonic sensors, wheel encoders, LIDAR, and a camera. It can be controlled over a WiFi communication interface using a PC or a mobile phone. Its performance was experimentally validated through a series of tests as well as in qualitative studies using various groups of students. Learning from data and optimization: this section focuses on four articles that cover statistical and machine learning and optimization techniques that make use of collected sensor data together with advanced computing algorithms for implementation. In [4], a detailed study was provided for smart building system identification combining both classical, such as nonlinear autoregressive (NARX) models, and data-driven machine learning-based approaches, such as multi-layer perceptrons (MLP). The goal was to achieve an accurate model of building thermal dynamics for control using collected data under various conditions. The data were collected through an existing building automation system (BAS) infrastructure with wireless components for sensing, room temperature sensors, and controlled motor valves for the heating elements. The study was carried out on a real building at the University Paris-Est Creteil (UPEC). The main results showed good performance in terms of Mean Squared Error (MSE) and Mean Absolute Error (MAE), less than 0.2C compared to the ground truth. Reference [5] introduced an improved deep learning method for the evaluation and classification of road quality based on the U-Net deep autoencoder. The main innovation stemmed from the addition of residual connections, atrous spatial pyramid pooling, and attention gates to increase performance. An evaluation was performed on multiple reference benchmarking datasets: CrackForest, Crack500, and GAPs384. The Dice score was used to compare various architectures and parametrization options of the deep learning architectures with a robust improvement in the proposed ResU-NET + ASPP + AG network over the baseline specfication. Testing was performed on dedicated graphical processing units on each dataset, while mixed dataset training did not yield consistent results. A mixed integer linear programming optimization problem for smart parking EV charging was formulated by [6]. The goals were two-fold: to maximize the parking lot revenue by accommodating charging EVs as efficiently as possible and by minimizing the cost of power consumption through x
- 16. participation in the utility-level demand response (DR) program. The study was validated in a simulation using predefined schedules and model EV charging characteristics. A simplified convex relaxation technique was introduced to ensure the feasibility of the optimization problem. The solution was compared against a standard variable charging power approach and showed consistent improvements in terms of power consumption cost and percentage savings. The authors of [7] proposed a new scheme for data classification by combining sparse auto-encoders (SAE) with data postprocessing using a nature-inspired particle swarm optimization (PSO) algorithm. The postprocessing layer improved the classification performance of the deep neural network through a parameter optimized linear model. The labeled datasets used for practical evaluation stemmed from the medical field and included epileptic seizures, SPECTF, and cardiac arrhythmias while experimenting with multiple parameters of both the neural network and the PSO algorithm. The adjustment layer improved the performance of the models, as illustrated through other documented studies from the literature, achieving for example a 99.27% accuracy on the cardiac arrhythmia dataset. Advanced computing systems for data processing: this section includes two articles that present improvements to data processing through optimized architectures for solving differential equations and reconfigurable computing with FPGAs. Reference [8] approached the problem of increasing the performance when solving ordinary differential equations (ODE) on multi-core embedded systems, which can describe the system model of certain physical phenomena. The authors introduced an adaptive algorithm, PAMCL, based on the Adams–Moulton and Parareal methods and provided a comparison with existing approaches. The implementation-wise OpenCL platform was used with optimized solvers for both CPU and GPU systems. Quantitative results were reported, which include the CPU run time, GPU speed-up, and the memory footprints of the reference algorithms. This method showed good results and achieved full convergence to the exact solutions. A potential extension of the PAMCL method for partial derivative equations was described. Power-oriented monitoring of clock signals in FPGA systems was described by [9]. The argumentation of the work lays out the need to reduce the power consumption and checkability of reconfigurable computing platforms. The study included two types of power-monitoring: the detection of synchronization failures, and the dissipation of power using temperature and current sensors. The experiments were carried out on typical computing tasks, e.g., digital filter implementations, using standardized tools for monitoring and data collection. The thermal and power dissipation data were associated with fault conditions in the synchronization. Such improvements to the evaluation of FPGA systems are highly relevant for critical and highly reliable applications. References: 1. Pfeiffer, J.; Wu, X.; Ayadi, A. Evaluation of Three Different Approaches for Automated Time Delay Estimation for Distributed Sensor Systems of Electric Vehicles. Sensors 2020, 20, 351. [Google Scholar] [CrossRef] [PubMed] 2. Popescu, D.; Stoican, F.; Stamatescu, G.; Ichim, L.; Dragana, C. Advanced UAV–WSN System for Intelligent Monitoring in Precision Agriculture. Sensors 2020, 20, 817. [Google Scholar] [CrossRef] [PubMed] 3. Karalekas, G.; Vologiannidis, S.; Kalomiros, J. EUROPA: A Case Study for Teaching Sensors, Data Acquisition and Robotics via a ROS-Based Educational Robot. Sensors 2020, 20, 2469. [Google xi
- 17. Scholar] [CrossRef] [PubMed] 4. Sadeghian Broujeny, R.; Madani, K.; Chebira, A.; Amarger, V.; Hurtard, L. Data-Driven Living Spaces’Heating Dynamics Modeling in Smart Buildings using Machine Learning-Based Identification. Sensors 2020, 20, 1071. [Google Scholar] [CrossRef] [PubMed] 5. Augustauskas, R.; Lipnickas, A. Improved Pixel-Level Pavement-Defect Segmentation Using a Deep Autoencoder. Sensors 2020, 20, 2557. [Google Scholar] [CrossRef] [PubMed] 6. Jawad, M.; Qureshi, M.B.; Ali, S.M.; Shabbir, N.; Khan, M.U.S.; Aloraini, A.; Nawaz, R. A Cost-Effective Electric Vehicle Intelligent Charge Scheduling Method for Commercial Smart Parking Lots Using a Simplified Convex Relaxation Technique. Sensors 2020, 20, 4842. [Google Scholar] [CrossRef] [PubMed] 7. Karim, A.M.; Kaya, H.; Güzel, M.S.; Tolun, M.R.; Çelebi, F.V.; Mishra, A. A Novel Framework Using Deep Auto-Encoders Based Linear Model for Data Classification. Sensors 2020, 20, 6378. [Google Scholar] [CrossRef] [PubMed] 8. Tavakkoli, V.; Mohsenzadegan, K.; Chedjou, J.C.; Kyamakya, K. Contribution to Speeding-Up the Solving of Nonlinear Ordinary Differential Equations on Parallel/Multi-Core Platforms for Sensing Systems. Sensors 2020, 20, 6130. [Google Scholar] [CrossRef] [PubMed] 9. Drozd, O.; Nowakowski, G.; Sachenko, A.; Antoniuk, V.; Kochan, V.; Drozd, M. Power-Oriented Monitoring of Clock Signals in FPGA Systems for Critical Application. Sensors 2021, 21, 792. [Google Scholar] [CrossRef] [PubMed] Grigore Stamatescu, Anatoliy Sachenko, Dan Popescu Editors xii
- 18. sensors Article Evaluation of Three Different Approaches for Automated Time Delay Estimation for Distributed Sensor Systems of Electric Vehicles Jakob Pfeiffer 1,2,*, Xuyi Wu 2 and Ahmed Ayadi 2 1 BMW Group, Petuelring 130, 80788 Munich, Germany 2 Department of Electrical and Computer Engineering, Technical University of Munich, Arcisstr. 21, 80333 Munich, Germany; Xuyi.Wu@tum.de (X.W.); Ahmed.Ayadi@tum.de (A.A.) * Correspondence: Jakob.J.Pfeiffer@bmwgroup.com Received: 3 December 2019; Accepted: 26 December 2019; Published: 8 January 2020 Abstract: Deviations between High Voltage (HV) current measurements and the corresponding real values provoke serious problems in the power trains of Electric Vehicles (EVs). Examples for these problems have inaccurate performance coordinations and unnecessary power limitations during driving or charging. The main reason for the deviations are time delays. By correcting these delays with accurate Time Delay Estimation (TDE), our data shows that we can reduce the measurement deviations from 25% of the maximum current to below 5%. In this paper, we present three different approaches for TDE. We evaluate all approaches with real data from power trains of EVs. To enable an execution on automotive Electronic Control Units (ECUs), the focus of our evaluation lies not only on the accuracy of the TDE, but also on the computational efficiency. The proposed Linear Regression (LR) approach suffers even from small noise and offsets in the measurement data and is unsuited for our purpose. A better alternative is the Variance Minimization (VM) approach. It is not only more noise-resistant but also very efficient after the first execution. Another interesting approach are Adaptive Filters (AFs), introduced by Emadzadeh et al. Unfortunately, AFs do not reach the accuracy and efficiency of VM in our experiments. Thus, we recommend VM for TDE of HV current signals in the power train of EVs and present an additional optimization to enable its execution on ECUs. Keywords: automotive; current; electric power train; electric vehicle; embedded systems; delay; detection; distributed systems; measurements; power train; sensor; signals; time delay estimation 1. Introduction Political guidelines in various countries to decarbonize individual mobility led to an exponential growth of Electric Vehicles (EVs) in offers and sales. However, one obstacle for the success of EVs is the so-called range anxiety [1]. Customers are afraid that an EV is not able to provide the range they need for all of their journeys. To combat range anxiety and increase the range of EVs, there are two different ways. The first one is to simply increase the size of the High Voltage Battery (HVB). Unfortunately, this means to increase the size of the most expensive component of an EV, and after all, it is not a very sustainable way. The second way, which is our solution of choice, is to make EVs more efficient. Kirchhoff’s current law states that the sum of all currents at a node of an electric system is equal to 0 A. However, considering measurement signals of nodes in the power trains of EVs with distributed sensor systems, the sum of all currents can differ up to 20 % of the maximum current (see Figure 1). If we look closer at the Root Mean Square Error (RMSE) of the sum of currents RMSE(isum) = 0.67%, we realize that it has the same value as the mean current of the DCDC converter µiDCDC = 0.67%. 1
- 19. Sensors 2020, 20, 351 0 1000 2000 3000 4000 5000 6000 7000 Time Step [10 ms] -100 -80 -60 -40 -20 0 20 40 60 80 100 Relative Current [%] i HVB i EM i heat i cool i DCDC i sum Figure 1. Currents of all HV components in an EV on a test drive. The sum of all currents isum is plotted in black. According to Kirchhoff’s current law, it should be constantly 0 %. However, looking at the measurements shows that the deviation isum is higher than the current of the DCDC converter iDCDC. Even its noise spectrum is approximately half as high as the consumption of the heating iheat, which is the second largest consumer in this drive. A different value than 0 A for the sum of all currents indicates that there is a divergence between measurements and real values. The divergence becomes problematic when the power train is operating close to the system boundaries. For example, there are boundaries for the protection of the HVB. The HVB is only capable of discharging or charging a restricted amount of power. Higher amounts would threaten the HVB’s lifetime and safety [2]. To ensure a safe operation mode even for high divergences between measurements and real values, additional protection offsets (see Figure 2) might be added to the boundaries, although they have some drawbacks. t i Maximum Battery Current Measurement Additional Battery Protection Offset Measurement Tolerance Figure 2. A simplified example of offsets for protection of the HVB. The measured value (black) differs from the real value in the range of some tolerance (grey). To prevent exceeding the battery limit (red, solid) even under the worst measurement conditions, an additional battery protection offset (red, dashed) is introduced. The same principle is used analogously with negative currents. It can be extended to other High Voltage (HV) components. 2
- 20. Sensors 2020, 20, 351 For example, in the charging case, most notably during recuperation, the HVB might not allow the full power level, even though it would be capable of handling it. Thus, the amount of power charged to the battery is restricted and the EV loses cruising range while its power consumption increases. In the opposite case, the system might not release requested power, although the HVB could provide it in reality. This additional restriction of power decreases the EV’s performance. As can be seen from the two examples above, minimizing the magnitude of the protection offsets also allows increasing the performance as the efficiency and the cruising range of EVs. 0 1000 2000 3000 4000 5000 6000 7000 Time Step [10 ms] -100 -80 -60 -40 -20 0 20 40 60 80 100 Relative Current [%] iHVB (moved by 6 time steps) iEM iheat icool iDCDC isum Figure 3. The same test drive as in Figure 1 but with the battery current iHVB (green) shifted by six time steps. The sum of all currents isum (black) is significantly closer to 0 %. Besides measurement faults and sensor uncertainties [3], the divergence between measurements and real values is caused by time delays. Figure 3 shows an example of the sum of all currents isum being reduced by shifting a signal by 6 time steps. The delays result from distributed sensor systems in the power train as plotted in Figure 4. The High Voltage (HV) components have their own Electronic Control Unit (ECU) which is connected with the current sensors and processes the sensor information. The ECUs exchange this information via bus systems. The buses require individual amounts of time to send the measurement signals. Thus, from an ECU’s point of view, the sensor information from other ECUs arrives with individual delays (see the Ego ECU in Figure 4). These individual delays could be compensated easily with a synchronized clock and time stamps as part of each bus message. However, this solution would have two drawbacks. First, it would increase the bus traffic as not only the measurement information must be carried by the messages but also the time stamp. As a result, the EV would either require a faster bus which is able to transport more information, or it would have to reduce the information exchanged between the ECUs. Second, there exists no clock in the power trains of modern series EVs which is synchronized with all ECUs at the same frequency as the message exchange. Usually, the ECUs are synchronized in a longer time frame than they communicate. Thus, the time stamp solution would require additional or higher performing hardware and increase the costs for the production of the EV. 3
- 21. Sensors 2020, 20, 351 … 𝑆1,1 𝑆1,𝑛 𝑆1,2 … 𝑆𝑚,1 𝑆𝑚,𝑛 𝑆𝑚,2 … Legend: Sensor ECU Δ𝑡 Δ𝑡 Δ𝑡 Δ𝑡 Δ𝑡 Δ𝑡 Δ𝑡 Δ𝑡 𝑆𝑖,𝑗 Data Processing and Transmission (e.g. Bus) Δ𝑡 Time Delay … 𝑆𝐸𝑔𝑜 1 𝑆𝐸𝑔𝑜 𝑛 𝑆𝐸𝑔𝑜 2 Figure 4. A schematic example of an automotive bus system with higlighted sources of time delays. Please note that the time delays are highly individual and not necessarily equal, but constant or only slowly changing. The ECUs can be connected directly or indirectly via other ECUs. The Ego ECU is not able to reconstruct the time delays, because it only knows the received measurement values and their last sender. It has no further information about the time passed since the measurement’s original creation. The aim of this work is to automatically detect the time delay between measurement signals from different sensors without additional hardware. For this purpose, we develop two different approaches. One of them is based on Linear Regression (LR), whereas the other one optimizes the estimated variance of the difference between several signals. We compare our approaches to other state-of-the-art Time Delay Estimation (TDE) algorithms and evaluate them with a focus on precision and run-time efficiency. Apart from allowing a more accurate power distribution, the automated TDE helps to reduce the battery protection offset and thus to increase the performance, efficiency and cruising range of EVs. The rest of this paper is structured as follows. Section 2 states related work and the similarities and differences to our work. Furthermore, Section 2 highlights the contributions of our work to the state of the art. In Section 3, we explain the theory behind our work before we describe the practical experiments in Section 4. The experiments’ results, stated in Section 5, show us the performance of the algorithms for our use case. Based on this evaluation, we take the best performing algorithm and optimize it further. The optimization steps can be taken from Section 3.4 and their impacts to the results from Section 5.4. In Section 6, we discuss the advantages and drawbacks of all proposed concepts. Finally, we draw our conclusions and give a short outlook in Section 7. 2. State of the Art There exists plenty of literature about TDE, although—to the best of our knowledge—none of them is tailored to the specific problem of TDE of current signals in EVs. In the following, we present several publications about TDE from different fields of application, such as embedded systems, acoustics, medicine, positioning, aeronautics, process technology, and robotics. An approach which also deals with EVs and time delays is the one by Guo et al. [4]. However, their approach is similar to ours only at the first look. Their goal is to stabilize a grid of electric sources and sinks with EVs. For the stabilization of the grid, they propose time delay resistent control strategies 4
- 22. Sensors 2020, 20, 351 of smart grids with EVs. The EVs are able to charge bidirectionally. The bidirectional charging is used to smooth disturbances and respond rapidly to fast occurring changes in the power distribution of the grid. An example for such a rapidly occurring change in the times of renewable energies is the power output of wind turbines when a strong wind occurs. Compared to our approach, Guo’s focus is rather on the control strategy than on the TDE. Another difference with our work is that Guo’s system is rather macroscopic with lots of different elements and many EVs in the grid. Our system is instead quite microscopic. We consider a single EV with a power train of around five sources and sinks. Our communication network might be smaller than the number of HV components as some consumers might share the same ECU. For example, the heating and the cooling component of an EV use both the climate control ECU for bus communication. Kali et al. [5] propose a controller with TDE for Electric Machines (EMs). The TDE is executed state-based with the help of a model of the EM. The model design demands expert knowledge about the physical principles of an EM. This is justified for Kali et al. as they require the same knowledge for their controller. However, for our case, we want to be able to estimate the time delays without further knowledge about the HV components. Our TDE shall be executable with nothing else than the available measurement data. Zeng et al. [6] introduce a statistical approach to predict the delay of a bus message. The content of the messages does not need to be known to achieve high accuracy. This is different from our scenario where we want to make use of the information carried by the message. In contrast to Zeng et al., we do not require predicting the time delay accurately to milliseconds. For our purposes, an estimation of the number of delayed discrete time steps is sufficient. Not from the field of electric power trains or bus communication, but from acoustics is the approach shown by Lourtie and Moura [7]. They use a stochastic approach to model time delays in an acoustic path environment. Like ours, their environment consists of several sources. However, in contrast to our scenario, the delay they want to estimate varies with time. In our case, we assume the time delay to be constant in a short time frame. For longer periods, it might change slowly. The reason for the slowly changing time delay is that it is caused during the wake up procedure of the EV. The ECUs wake up in an unsynchronized way. Afterwards, the ECUs are synchronized on a relatively large time frame (e.g., 1 s), but work based on short time steps (e.g., 10 ms). Another acoustics application for TDE is shown by He et alii [8]. They use the so-called Multichannel Cross-Correlation Coefficient algorithm to estimate time delays of speech sources in noisy and reverberant environments. Svilainis et al. [9] present another interesting approach. Their goal is to estimate the time passed between emitting an ultrasonic signal and absorbing its reflection. Like Zeng et al., they require high precision. Another difference to our approach is that their algorithms make use of the pulse form of ultrasonic signals. Our signal as plotted in Figure 1 can vary in a large range and does not necessarily contain pulses (e.g., after time step 5,000). Mirzaei et al. expand TDE for ultrasonic signals to the field of medicine [10]. The authors introduce a window-based TDE approach to estimate the time passed between two frames of radio-frequency data. They compare the results of the new window-based approach to their previously developed, optimization-based method [11] and to Normalized Cross-Correlation. Recently, Garcez et al. published their work on a similar problem to ours, but in a completely different field of application [12]. Like bus systems of EVs, Global Navigation Satellite Systemss (GNSSs) systems have real-time requirements. Their goal is to minimize deviations between measurements and real position data. The time delays are caused during the transmission of GNSS messages, when the signals do not take straight lines of sight, but are reflected on their way or suffer from noise. The authors propose a tensor-based subspace tracking algorithm to efficiently estimate time delays of received GNSS signals. A similar approach is presented by Xie et al. for an indoor positioning sensing system [13]. They sense positions of mobile devices based on the signal strength and the signal’s time delay since 5
- 23. Sensors 2020, 20, 351 its transmission from a base station. For the TDE, Xie et al. combine Cross-Correlation with Quadratic Fitting. This is similar to our LR approach (see Section 3.2), where we try to fit the signals with quadratic functions to retrieve the delay between them. Like Garcez et al., they have to deal with the problem that the signals are often reflected and do not take direct lines of sight. Different to Garcez et al., Xie’s approach takes the strength of the signal into account for retrieving a more exact position estimation. For our work, we cannot take advantage of this information, because in wire-based bus systems all signals are equally strong. Schmidhammer et al. estimate positions of moving, non-cooperative objects in vehicular environments [14]. Their idea is to estimate the position of an object based on time delays in a network of distributed receiving and transmitting nodes. In contrast to our work, the networking nodes of Schmidhammer et al. are not necessarily on-board the vehicle, but can also be mounted on the road infrastructure. Emadzadeh et al. [15] show an inspiring approach for detecting the relative position of spacecrafts. For retrieving the position, they examine an X-ray signal received by two spacecrafts and determine the time delay between them. For the TDE, they use Adaptive Filters (AFs). This approach seems very promising to us. We implement the algorithms of Emadzadeh et al. and compare them to ours in order to find out if their approach can be transferred from X-ray signals to current measurements in the power train of EVs. Like Emadzadeh et al., Liu et al. focus on AFs [16]. Compared to our problem of fixed or only slowly changing time delays, the difference in Liu et al. is that they deal with time-varying time delays. That makes further processing steps necessary. For example, they require a transition probability matrix and an initial probability distribution vector to model the time delay changes with a Markov chain. Park et al. analyze time series data with Autoencoders and Long Short-Term Memory Neural Networks (LSTMs) to detect faults in industrial processes [17]. The authors emphasize the importance of TDE for correct fault detection. However, they focus only on time delays caused by their own fault detection system. Our focus lies on earlier steps in the processing chain. We want to detect time delays between the input signals before they are passed to other computation processes. Furthermore, we want to implement algorithms which are able to learn on-board the automotive ECUs and adapt themselves to new data. As the training of Neural Networks is quite memory intensive and demands high computational power, they do not belong to our methods of choice. Close to the application field of industrial processes is the approach of Srinivasa Rao et al. [18]. In their recent article, the authors propose fuzzy parametric uncertainty to mathematically model systems with time delays. Their goal is to enable a robust controller design. For this purpose, they first approximate the time delay system as an interval system. After retrieving the intervals, they design an optimal controller for these. Like Guo et al., Srinivasa Rao et al. focus on how to retrieve an optimal controller, which is not part of our work. Although they focus on the control of industrial processes, their article is very general. Besides industrial plants, they also mention potential fields of application, such as EMs or robot manipulators. Time delay compensation for robots is the focus of Shen et al [19]. Their focus is on teleoperating robots which require knowledge about the time delay between the master and the slave robot for stable operation. The robots and their communication channels are modeled as extended dynamical system. For this system, Shen et al. develop a cascade observer which is able to control it in a stable way. The authors assume that a sufficiently accurate value for the TDE is given and concentrate on its compensation. This is different to our work here. We explicitly want to estimate the time delay. You et al. develop a proportional multiple integral observer for fuzzy systems [20]. The goal of their work is the same as ours. They want to minimize deviations between measurements and real values caused by time delays and measurement inaccuracies. Their time delays are also varying. Unlike the varying time delays presented before, the ones of You et al. do not vary with time but rather with states. Their focus is also on industrial processes and not on electric power trains. However, the 6
- 24. Sensors 2020, 20, 351 main difference between our works is that You et al. want to minimize time delays and measurement inaccuracies with the same system. Our approach follows the divide and conquer strategy and faces the two problems separately. We focus on the problem of measurement deviations caused by measurement inaccuracies in our previous work [3]. However, measurement inaccuracies are not part of this work. Here, we assume that the measurements are appropriately accurate and that the main deviations are caused by time delays as shown in Figure 1 and Figure 3. Thus, TDE is our solution of choice to minimize the deviations. Our contribution in this article is the development of a regression-based approach and an algorithm based on Variance Minimization (VM) for TDE as first presented in [21]. We transfer the ideas introduced by Emadzadeh et al. to the domain of currents in the HV system of EVs and compare the results to our approaches in matters of accuracy and computational performance. Our TDE works only with the data available in modern series EVs and does not require an additional clock. In addition to [21], we introduce an optimization of the most accurate and efficient of our evaluated approaches. We further evaluate the optimization both on artificially created data with known ground truth as well as real drive data with unknown ground truth. 3. Concepts In this section, we introduce the algorithms and shortly explain the concepts from other authors which we implement and compare for TDE. From now on, for the sake of easier understanding, we focus on the current of the EM iEM and the HVB iHVB (without other consumers than the EM) as examples. Nevertheless, the proposed methods can be extended to every current signal in the HV system of an EV. Furthermore, we inverse the sign of iHVB from now on to make its shape similar to the one of the EM. Thus, we can treat the HVB current signal as a delayed or preceded version of the EM, respectively. Our goal is to find the time delay td in a bus system which can be described as x1(t) = i1(t) + n1(t) x2(t) = i2(t − td) + n2(t − td), (1) where t stands for the time step, x1(t) is the measurement signal of the faster component, x2(t) describes the slower component’s signal, i1(t) and i2(t) describe the corresponding currents and n1(t) and n2(t) are noise terms [15]. As we cannot retrieve the currents i1(t) and i2(t) directly, we cannot minimize the difference between i1(t) and i2(t). Instead, we directly minimize the difference between the two measurement signals x1(t) and x2(t). 3.1. Adaptive Filter The idea of Emadzadeh et al. is to model the time delay as Finite Impulse Response (FIR) filter. They define x1(t) to be the faster signal. For each measurement x2(ti) at time step ti, they collect a row of the last M measurements of the other signal x1(ti − M + 1 : ti) = [x1(ti − M + 1), x1(ti − M + 2), . . . x1(ti − 1), x1(ti)] . (2) Then, the authors search for an optimal channel impulse response vector ω∗ such that the deviation between x2(ti) and x1(ti − M + 1 : ti)ω∗ becomes minimal. Mathematically, this can be expressed by the minimization of the expectation value of the Mean Squared Error (MSE) between the measurement value of the slower signal and the filtered measurement row of the faster signal. It results in the formula ω∗ = argmin ω E h (x2(ti) − x1(ti − M + 1 : ti)ω)2 i . (3) 7
- 25. Random documents with unrelated content Scribd suggests to you:
- 29. The Project Gutenberg eBook of Index of the Project Gutenberg Works of Jack London
- 30. This ebook is for the use of anyone anywhere in the United States and most other parts of the world at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this ebook or online at www.gutenberg.org. If you are not located in the United States, you will have to check the laws of the country where you are located before using this eBook. Title: Index of the Project Gutenberg Works of Jack London Author: Jack London Editor: David Widger Release date: May 8, 2019 [eBook #59461] Most recently updated: July 7, 2019 Language: English Credits: Produced by David Widger *** START OF THE PROJECT GUTENBERG EBOOK INDEX OF THE PROJECT GUTENBERG WORKS OF JACK LONDON ***
- 31. INDEX OF THE PROJECT GUTENBERG WORKS OF JACK LONDON Compiled by David Widger
- 33. CONTENTS Click on the ## before many of the titles to view a linked table of contents for that volume. Click on the title itself to open the original online file. ## THE CALL OF THE WILD ## BEFORE ADAM ## JOHN BARLEYCORN ## BURNING DAYLIGHT ## THE RED ONE ## THE NIGHT-BORN ## THE STRENGTH OF THE STRONG ## MOON-FACE AND OTHER STORIES ## THE IRON HEEL
- 34. ## SOUTH SEA TALES ## THE VALLEY OF THE MOON ## THE SON OF THE WOLF ## LOST FACE ## THE CRUISE OF THE SNARK ## WHEN GOD LAUGHS AND OTHERS ## SMOKE BELLEW ## CHILDREN OF THE FROST ## THE CRUISE OF THE DAZZLER ## DUTCH COURAGE AND OTHERS ## THE ROAD ## THE TURTLES OF TASMAN ## STORIES OF SHIPS AND THE SEA ## THEFT ## THE SCARLET PLAGUE ## A SON OF THE SUN
- 35. ## THE ACORN-PLANTER ## TALES OF THE FISH PATROL ## SCORN OF WOMEN ## BROWN WOLF et al. EBOOKS WITHOUT TABLES OF CONTENTS ADVENTURE HEARTS OF THREE JERRY OF THE ISLANDS MARTIN EDEN MICHAEL, BROTHER OF JERRY REVOLUTION AND OTHER ESSAYS SEA WOLF TALES OF THE FISH PATROL THE ABYSMAL BRUTE THE FAITH OF MEN THE GAME
- 36. THE GOD OF HIS FATHERS THE HOUSE OF PRIDE THE HUMAN DRIFT THE JACKET (STAR-ROVER) THE KEMPTON-WACE LETTERS THE LITTLE LADY OF THE BIG HOUSE THE MUTINY OF THE ELSINORE THE PEOPLE OF THE ABYSS WAR OF THE CLASSES WHITE FANG DAUGHTER OF SNOWS LOVE OF LIFE et al.
- 37. TABLES OF CONTENTS OF VOLUMES THE CALL OF THE WILD
- 38. by Jack London
- 39. CONTENTS Chapter I. Into the Primitive Chapter II. The Law of Club and Fang Chapter III. The Dominant Primordial Beast Chapter IV. Who Has Won to Mastership Chapter V. The Toil of Trace and Trail Chapter VI. For the Love of a Man Chapter VII. The Sounding of the Call BEFORE ADAM
- 40. by Jack London CONTENTS CHAPTER I CHAPTER II CHAPTER III CHAPTER IV CHAPTER V CHAPTER VI CHAPTER VII CHAPTER VIII CHAPTER IX CHAPTER X CHAPTER XI CHAPTER XII CHAPTER XIII CHAPTER XIV CHAPTER XV CHAPTER XVI CHAPTER XVII CHAPTER XVIII JOHN BARLEYCORN
- 41. By Jack London CONTENTS CHAPTER I CHAPTER II CHAPTER III CHAPTER IV CHAPTER V CHAPTER VI CHAPTER VII CHAPTER VIII CHAPTER IX CHAPTER X CHAPTER XI CHAPTER XII CHAPTER XIII CHAPTER XIV CHAPTER XV CHAPTER XVI CHAPTER XVII CHAPTER XVIII CHAPTER XIX CHAPTER XX CHAPTER XXI CHAPTER XXII CHAPTER XXIII CHAPTER XXIV CHAPTER XXV CHAPTER XXVI CHAPTER XXVII CHAPTER XXVIII CHAPTER XXIX CHAPTER XXX CHAPTER XXXI CHAPTER XXXII CHAPTER XXXIII CHAPTER XXXIV CHAPTER XXXV CHAPTER XXXVI CHAPTER XXXVII CHAPTER XXXVIII CHAPTER XXXIX BURNING DAYLIGHT
- 43. PART I CHAPTER I CHAPTER II CHAPTER III CHAPTER IV CHAPTER V CHAPTER VI CHAPTER VII CHAPTER VIII CHAPTER IX CHAPTER X CHAPTER XI CHAPTER XII CHAPTER XIII
- 44. PART II CHAPTER I CHAPTER II CHAPTER III CHAPTER IV CHAPTER V CHAPTER VI CHAPTER VII CHAPTER VIII CHAPTER IX CHAPTER X CHAPTER XI CHAPTER XII CHAPTER XIII CHAPTER XIV CHAPTER XV CHAPTER XVI CHAPTER XVII CHAPTER XVIII CHAPTER XIX CHAPTER XX CHAPTER XXI CHAPTER XXII CHAPTER XXIII CHAPTER XXIV CHAPTER XXV CHAPTER XXVI CHAPTER XXVII THE RED ONE
- 45. By Jack London CONTENTS PAGE The Red One 11 The Hussy 57 Like Argus of the Ancient Times 93 The Princess 141 THE NIGHT-BORN
- 46. By Jack London
- 47. Contents THE NIGHT-BORN THE MADNESS OF JOHN HARNED WHEN THE WORLD WAS YOUNG THE BENEFIT OF THE DOUBT WINGED BLACKMAIL BUNCHES OF KNUCKLES WAR UNDER THE DECK AWNINGS TO KILL A MAN THE MEXICAN THE STRENGTH OF THE STRONG
- 48. By By Jack London
- 49. CONTENTS PAGE The Strength of the Strong 11 South of the Slot 34 The Unparalleled Invasion 60 The Enemy of All the World 81 The Dream of Debs 104 The Sea-Farmer 134 Samuel 161 MOON-FACE AND OTHER STORIES
- 50. By Jack London CONTENTS MOON-FACE THE LEOPARD MAN’S STORY LOCAL COLOR AMATEUR NIGHT THE MINIONS OF MIDAS THE SHADOW AND THE FLASH ALL GOLD CANYON PLANCHETTE THE IRON HEEL
- 51. by Jack London CONTENTS FOREWORD THE IRON HEEL CHAPTER I -- MY EAGLE CHAPTER II -- CHALLENGES. CHAPTER III -- JACKSON'S ARM. CHAPTER IV -- SLAVES OF THE MACHINE CHAPTER V -- THE PHILOMATHS CHAPTER VI -- ADUMBRATIONS CHAPTER VII -- THE BISHOP'S VISION CHAPTER VIII -- THE MACHINE BREAKERS CHAPTER IX -- THE MATHEMATICS OF A DREAM CHAPTER X -- THE VORTEX CHAPTER XI -- THE GREAT ADVENTURE CHAPTER XII -- THE BISHOP CHAPTER XIII -- THE GENERAL STRIKE CHAPTER XIV -- THE BEGINNING OF THE END CHAPTER XV -- LAST DAYS CHAPTER XVI -- THE END CHAPTER XVII -- THE SCARLET LIVERY CHAPTER XVIII -- IN THE SHADOW OF SONOMA CHAPTER XIX -- TRANSFORMATION CHAPTER XX -- A LOST OLIGARCH CHAPTER XXI -- THE ROARING ABYSMAL BEAST
- 52. CHAPTER XXII -- THE CHICAGO COMMUNE CHAPTER XXIII -- THE PEOPLE OF THE ABYSS CHAPTER XXIV -- NIGHTMARE CHAPTER XXV -- THE TERRORISTS SOUTH SEA TALES
- 53. By Jack London CONTENTS THE HOUSE OF MAPUHI THE WHALE TOOTH MAUKI “YAH! YAH! YAH!” THE HEATHEN THE TERRIBLE SOLOMONS THE INEVITABLE WHITE MAN THE SEED OF McCOY THE VALLEY OF THE MOON
- 54. By Jack London CONTENTS BOOK I CHAPTER 1 CHAPTER II CHAPTER III CHAPTER IV CHAPTER V CHAPTER VI CHAPTER VII CHAPTER VIII CHAPTER IX CHAPTER X CHAPTER XI CHAPTER XII CHAPTER XIII CHAPTER XIV CHAPTER XV BOOK II CHAPTER I CHAPTER II CHAPTER III CHAPTER IV CHAPTER V CHAPTER VI CHAPTER VII CHAPTER VIII CHAPTER IX CHAPTER X CHAPTER XI CHAPTER XII CHAPTER XIII CHAPTER XIV CHAPTER XV CHAPTER XVI CHAPTER XVII CHAPTER XVIII CHAPTER XIX BOOK III CHAPTER I CHAPTER II CHAPTER III CHAPTER IV CHAPTER V CHAPTER VI CHAPTER VII CHAPTER VIII CHAPTER IX CHAPTER X CHAPTER XI CHAPTER XII CHAPTER XIII CHAPTER XIV CHAPTER XV CHAPTER XVI CHAPTER XVII CHAPTER XVIII CHAPTER XIX CHAPTER XX
- 55. CHAPTER XXI CHAPTER XXII The Son of the Wolf
- 56. By Jack London CONTENTS The White Silence The Son of the Wolf The Men of Forty Mile In a Far Country To the Man on the Trail The Priestly Prerogative The Wisdom of the Trail The Wife of a King An Odyssey of the North LOST FACE
- 57. By Jack London CONTENTS page Lost Face 11 Trust 29 To Build a Fire 47 That Spot 71 Flush of Gold 85 The Passing of Marcus O’Brien 106 The Wit of Porportuk 124 THE CRUISE OF THE SNARK
- 58. By Jack London CONTENTS CHAPTER PAGE I. Foreword 13 II. The Inconceivable and Monstrous 27 III. Adventure 47 IV. Finding One’s Way About 58 V. The First Landfall 72 VI. A Royal Sport 82 VII. The Lepers of Molokai 97 VIII. The House of the Sun 116 IX. A Pacific Traverse 134 X. Typee 156 XI. The Nature Man 175 XII. The High Seat of Abundance 193 XIII. The Stone-fishing of Bora Bora 214 XIV. The Amateur Navigator 223
- 59. XV. Cruising in the Solomons 244 XVI. Bêche de Mer English 270 XVII. The Amateur M.D. 280 Backword 303 WHEN GOD LAUGHS, AND OTHER STORIES
- 60. By Jack London CONTENTS WHEN GOD LAUGHS THE APOSTATE A WICKED WOMAN JUST MEAT CREATED HE THEM THE CHINAGO MAKE WESTING SEMPER IDEM A NOSE FOR THE KING THE “FRANCIS SPAIGHT” A CURIOUS FRAGMENT A PIECE OF STEAK SMOKE BELLEW
- 61. 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