PreprintPDF Available

Artificial Neural Network for Estimating Millimeter Wave Channel Sounding Data inside a Bus

Authors:

Abstract and Figures

Radio wave propagation in an intra-vehicular environment is markedly different from other well studied indoor scenarios such as an office or a factory oor. Millimeter Wave (mmWave) based intra-vehicular communications promises large bandwidth and can achieve ultra-high data rate with lower latency. However, exploiting the advantages of mmWave communications largely relies on proper characterization of the propagation channel. Channel characterization is most accurately done through an extensive channel sounding, but due to hardware and environmental constraints, it is impractical to test channel condition for all possible transmitter and receiver locations. In this paper, we use artificial neural network to aid channel sounding. Based on some real-world sounding data we show that it is possible to accurately estimate channel transfer function (CTF) and power delay profile (PDP) in an intra-bus scenario. Such artificially generated models can help in extrapolation in other relevant scenarios for which measurement data is unavailable. The proposed model can also be used for tapped delay line based bit-error-simulations as well.
Content may be subject to copyright.
Articial Neural Network for Estimating Millimeter
Wave Channel Sounding Data inside a Bus
Rajeev Shukla ( rs.20ec1103@phd.nitdgp.ac.in )
National Institute of Technology Durgapur https://orcid.org/0000-0002-4693-9799
Abhishek Narayan Sarkar
National Institute of Technology Durgapur
Aniruddha Chandra
National Institute of Technology Durgapur
Jan M. Kelner
Wojskowa Akademia Techniczna im Jaroslawa Dabrowskiego
Cezary Ziolkowaski
Wojskowa Akademia Techniczna im Jaroslawa Dabrowskiego
Tomas Mikulasek
Brno University of Technology: Vysoke uceni technicke v Brne
Ales Prokes
Vysoké učení technické v Brně: Vysoke uceni technicke v Brne
Research Article
Keywords: power-delay-prole, tap delay gain, intra-vehicle communication, articial neural network,
mmWave channel sounding
Posted Date: August 23rd, 2022
DOI: https://doi.org/10.21203/rs.3.rs-1880257/v1
License: This work is licensed under a Creative Commons Attribution 4.0 International License. 
Read Full License
RESEARCH
Artificial Neural Network for Estimating
Millimeter Wave Channel Sounding Data inside a
Bus
Rajeev Shukla1* , Abhishek Narayan Sarkar1, Aniruddha Chandra1, Jan M Kelner2
, Cezary Ziolkowski2, Tomas Mikulasekt3and Ales Prokes3
*Correspondence:
rajeev2504s@gmail.com
1ECE Department, National
Institute of Technology, M. G.
Avenue, 713209 Durgapur, India
Full list of author information is
available at the end of the article
Abstract
Radio wave propagation in an intra-vehicular environment is markedly different
from other well studied indoor scenarios such as an office or a factory floor.
Millimeter Wave (mmWave) based intra-vehicular communications promises large
bandwidth and can achieve ultra-high data rate with lower latency. However,
exploiting the advantages of mmWave communications largely relies on proper
characterization of the propagation channel. Channel characterization is most
accurately done through an extensive channel sounding, but due to hardware and
environmental constraints, it is impractical to test channel condition for all
possible transmitter and receiver locations. In this paper, we use artificial neural
network to aid channel sounding. Based on some real-world sounding data we
show that it is possible to accurately estimate channel transfer function (CTF)
and power delay profile (PDP) in an intra-bus scenario. Such artificially generated
models can help in extrapolation in other relevant scenarios for which
measurement data is unavailable. The proposed model can also be used for
tapped delay line based bit-error-simulations as well.
Keywords: power-delay-profile; tap delay gain; intra-vehicle communication;
artificial neural network; mmWave channel sounding
Introduction
Intra-Vehicular Communication (IVC) is categorized under two categories: wired
and wireless. With the advancement in technologies and increase in the consumer-
friendly market, number of features and technologies incorporated inside the vehi-
cle has increased considerably. In this scenario, wired connections between different
modules of communication in the vehicles get cumbersome as it raises design, man-
ufacturing, and installation issues. Wireless IVC systems have high-speed duplex
data links, which are better equipped to handle multiple tasks through one system,
thus providing more convenience, simplified design, and hassle-free installation. But
with scurrying advancements in IC fabrication more and more technologies are now
available to incorporate inside a vehicle. Ranging from GPS to centralised enter-
tainment and infotainment are now indigenous inside a vehicle.
With fuel prices, global warming on the rise, and so is the increasing number
of vehicles on the road, restrictions on private vehicles and increased use of pub-
lic vehicles can be a new normal. Such restrictions will bring a check on pollution
levels and traffic congestion. These restrictions will increase the demand for public
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 2 of 11
vehicles for daily commute. While commuting through these public vehicles, people
like to spend their time on mobile devices on various platforms like gaming, media,
entertainment, news, and infotainment in the modern-day scenario. With advance-
ments in data networks, these platforms have become more engaging, increasing the
demand for on-demand high data-rate for various data network services [1].
Among the public transport vehicles, a bus is one of the most trivial modes of
commute in urban and semi-urban areas. Considering this fact, this paper provides
an artificial neural network (ANN) based channel sounding model for evaluating
various channel parameters helpful in designing a reliable channel model inside a
bus. With the evolution of the fifth generation wireless network, millimeter wave
(mmWave) bands from (24 - 300 GHz) promise to provide the required high data
rate, high bandwidth, and low latency network[2]. Intra-vehicular mmWave com-
munication has excellent prospects in providing uninterrupted data services in a
high traffic density environment without compromising bandwidth, data rate, or
spectral efficiency [3]. Consequently, the paper comprises mmWave-based channel
sounding using ANN in the intra-bus scenario.
Personal vehicles and large vehicles were the prime focus for mmWave channel
measurement campaigns. The current article is an extension of the comprehensive
channel sounding measurement campaign by researchers at Brno University, Czech
Republic [1]. It was the first mmWave channel measurement campaign inside a bus
at 60 GHz millimeter wave band. Other measurement campaigns inside a bus are
reported in [4,5,6] at narrowband and in [7] at ultrawideband.
Contributions
We propose an ANN approach to synthetically generate channel transfer func-
tion (CTF) for a given set of input parameters. The CTF, in turn, is used to
obtain power delay profile (PDP) which can be readily used for characterizing
intra-bus mmWave propagation.
We derived a tapped delay-line (TDL) channel model from the synthetically
generated PDPs. Different error measures show that the simulated model
tallies well with the TDL model directly obtained by sampling the measured
PDP.
Background
Channel Measurement
As per our knowledge, till 2015, the intra-bus channel measurement campaign was
concentrated basically in the sub-6 GHz range. In 2016, an extensive 60 GHz mea-
surement campaign inside the bus was carried out at Brno University, Czech Re-
public, and was reported in [1,8,9].The bus being the most common and cheap
mode of public transport for intercity and intra-city travel,it is the primary focus of
our channel sounding case study. As illustrated in 1, the basic idea behind channel
measurement was to enable the commuting public with onboard gigabyte wire-
less networks using a common vehicular access point (AP) inside the bus, which
is wirelessly connected to roadside units or base stations through the vehicle-to-
infrastructure (V2I) mobile networks.The AP will act as a transceiver and connects
the passenger devices to outside world giving access to various on-demand high data
rate services over static intra-vehicular channels.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 3 of 11
Measurement Set-up
Here the experimental setup consists of an analog signal generator (model: Agilent
E8257D), a scalar signal analyzer (model: Rohde and Schwarz FSUP50), a pair of
mmWave antennas, a power amplifier (PA), a mixer, a DC power supply, adapters,
and connecting cables(Fig.1). The signal generator sweeps all the frequencies from
55 GHz to 65 GHz in the transmitter section. This frequency bandwidth keeps
a step size of 10MHz, generating 1001 data points in each sweep. The generator
output power is set to 13 dBm. The output signal is carried through a 2.5m long
phase-stable coaxial cable(model: MegaPhase TM67) to the power amplifier (model:
Quinstar QPW 50662330) through standard 1.85 mm adapters. The power amplifier
has a gain of 31 dBm, which is high enough to compensate for the cable loss as
well as it also boosts the signal fed to the open-waveguide antenna(OWGA). A DC
power supply powers the Power amplifier and its accessories. A substrate-integrated-
waveguide slot antenna (SIW SA) intercepts the signal and sends the received signal
to the signal analyzer at the receiving end. An external mixer converts the signal
to an intermediate frequency; it also helps in avoiding high cable loss at mmWave
frequencies. The GPIB cables connect the two hardware at the transmitting and
receiving sections for synchronization. To extend the Tx and Rx distance, we can
cascade several GPIB cables together.
The whole measurement setup is placed inside a 50-seater Mercedes Benz
Tourismo bus to conduct the measurement experiments. Performance evaluation
of the 60 GHz downlink channel is carried out over 10 GHz (55-65 GHz) wide band-
width. To create a real-time scenario transmitter is placed near the ceiling at the
front end of the bus, and the receiver is placed in the seat’s drop-down tray to
imitate a hand-held device. The receiver location was changed to different seats to
cover the entire interior space of the bus in the presence and absence of passengers.
Deep Learning
Communication channel models are used for designing cum optimizing communi-
cation systems and also evaluating their system performance via simulations. For
radio channel modeling, empirical or deterministic models are used conventionally.
Empirical channel models, such as the COST-231 model , are derived by giving mea-
surement campaign data as input to the simplistic mathematical models. Although
these models can be used practically, they show very significant deviations from the
actual received channel output parameters such as path loss values , making them
unreliable. Commonly used deterministic models include ray tracing, the Finite-
Difference Time-Domain method or the vector parabolic equation method. Among
the various deterministic models, ray tracing has been widely used to calculate the
radio channel characteristics due to its very high accuracy compared to other de-
terministic models [10]. But these site-specific-physics-based models requires large
computational resources and highly skilled experts for real-time scenarios. This cre-
ates huge bar-barrier along with substantial simulation time and memory required
to trace all the ray paths, when the number of scattering objects and ray intersec-
tions within the simulated space increases [11].
Modern channel modelling aim to overcome these limitations by integrating DL
algorithms which are capable of learning and estimating radio propagation chan-
nel parameters [12]. More particularly, artificial neural networks (ANNs) have been
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 4 of 11
widely used in an effort to replace ray tracing simulators. Artificial neural net-
works help us to imitate human brain by allowing simple learning components to
perform certain computational operations and connect it to other learning compo-
nents. The learning components are referred to as nodes,artificial neurons or just
neurons. These neural networks ”learn” or specifically train themselves from the
data from the measurement campaign,after this for a given input channel parame-
ters the approximate channel behaviour can be estimated [13].
ANN Based DL Model
In ANN or Multi-Layer Perceptrons(MLPs) neurons are arranged into layers, where
each neuron has input connections originating from the previous layer and output
connections pointing towards the next layer. Any typical MLP consists of an input
layer, a number of hidden layers, and an output layer. Here in our ANN there
are four hidden layers between the input and output layer.The input layer has six
neurons for accepting and representing the six input parameters x1, x2, x3, x4, x5, x6
and the output layer has two neurons for two outputs y1, y2.Commonly used input
features of DL channel propagation models are the operating frequency and the Tx-
Rx distance. Other than this there are other categorical input features used here
which are Seat Number where the receiver is present, whether the receiver is present
at the window or aisle seat, whether the receiver is present at the left or right seat
and other miscellaneous information. The hidden layers as well as the output layer
consist of neurons that apply transformations to their input data. Specifically, the
output, s(l)
j, of the j-th neuron in the l-th layer is computed by applying a activation
function, ρh(x) or ρo(x) ,to the weighted sum of the previous layer neuron outputs,
plus a bias term, b(l)
jas shown in (1)
s(l)
j=ρ(
n(l1)
X
i
wij s(l1)
i+b(l)
j) (1)
where s(l1)
iis the output of the i-th neuron of the previous layer, wi,j is the weight
associated with s(l1)
iand n(l1) is the number of neurons in the previous layer.
The number of the neurons in the first, second, third and fourth hidden layers
are 20, 50, 100, 200 respectively. the activation function which we are using in the
hidden layer neurons is “ReLU”(Rectified Linear Units) activation function ρh(x)
while the activation function used in the output layer neurons is “Sigmoid” ρo(x)
as shown in (2) and (3)[13].
ρh(x) = (x, if x0
0,if x < 0)(2)
ρo(x) = 1
1 + ex(3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 5 of 11
Learning Process
Here we are using the ”Adam” optimizer.Adam optimizer is the extended version
of stochastic gradient descent which could be implemented in various deep learning
applications.In Adam instead of adapting learning rates based on the average first
moment as in RMSP, Adam makes use of the average of the second moments of
the gradients. Adam. This algorithm calculates the exponential moving average of
gradients and square gradients. And the parameters of β1 and β2 are used to control
the decay rates of these moving averages. Adam is a combination of two gradient
descent methods, Momentum, and RMSP.
The method is really efficient when working with large problem involving a lot of
input data parameters. It requires less memory and is efficient.We run this model on
the train dataset (X-train, Y-train) and validation dataset(X-val,Y-val) to train our
neural network and also for tuning its hyperparameters. The training and validation
datasets are run with 100 epochs[14].During each epoch the weights are updated by
using (4)
wt+1 =wtˆmt(α
vt+ǫ) (4)
where
α=learning rate(103)
ǫ=a small positive constant(108)
ˆmtand ˆvtare the bias-corrected values which are calculated by using (5),(6) and
(7),(8) respectively,
mt=β1mt1+ (1 β1)[ δL
δwt
] (5)
ˆmt=mt
1β1
(6)
and,
vt=β2vt1+ (1 β2)[ δL
δwt
]
2
(7)
ˆvt=vt
1β2
(8)
Here,
mt=aggregate of gradients at time t [current] (initially, mt= 0)
vt=sum of square of past gradients(initially, vt= 0)
β1=Moving average parameter (constant, 0.9)
β2=Moving average parameter (constant, 0.999)
δL=derivative of Loss Function
δwt=derivative of weights at time t
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 6 of 11
Proposed Methodology
For the measurement campaign in [1,8,9], due to distance constraints, instead
of using a virtual network analyzer (VNA), the authors used a signal analyzer.
The resulting output was the channel transfer function (CTF) amplitude with-
out phase information. As the generation of the power delay profile is needed, the
phase information of CTF is necessary; Hilbert Transform (HT) was used to convert
amplitude-only information into complex information H(f) which is represented as
shown in (9) and (10) [15]
loge(|Hc(f)|)Hilber t T r anfor m
6Hc(f) (9)
Hc(f) = |Hc(f)|6Hc(f) (10)
Using inverse fast Fourier transform (IFFT), time domain channel response can be
procured easily by using (11)
Hc(f)IF F T
hc(t)(.)2
|hc(t)|2(11)
In addition to the above methodology, we propose a deep learning-based (Fig.2)
for generation of CTFs. Here, dynamic parameters extracted from the measurement
campaign like frequency, transmitter-receiver distance, receiver position, and pres-
ence of passengers are fed as input to an ANN model. The ANN model constitutes
a decoder-like Multi- Layer Perceptron(MLP) which is trained with the extracted
parameters(Fig.3). The model generates simulated values of amplitude and phase
information of the CTFs from the given input parameter values. In post-processing
stage the simulated and measured CTF values are compared. The measured and
simulated PDPs are calculated from respective CTFs and comparison is drawn be-
tween them.To furthur validate the results an average PDP trend is generated which
is then sampled at different delays to get tap gains for the TDL model.
Model Validation
The measured and simulated PDPs respectively. To further validate our model, we
sample the continuous measured PDPs into discrete PDPs to derive an equivalent
TDL model for the measured data and compare it with the simulated PDP trend.
Results and Discussion
Sources of Errors
Intra-vehicular scenarios are highly clustered and complex environment. Application
of ANN (or any other ML algorithms) in such an environment are prone to high
amount of errors [16] . For example, the presence or absence of line-of-sight(LOS)
component greatly influence the performance of ANN models [17] . Higher number
of LOS components means more accurate are the ANN based models. Similarly,
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 7 of 11
Algorithm 1 Proposed Algorithm
Require: Pre-Trained ANN Model
Input:
Model Parameters
1 Frequency
2 Tx-Rx Distance
3 Seat Number
4 Window or Aisle
5 Left Side or Right Side
6 Miscellaneous/Others
Training Parameters
1 Measured value of Amplitude of H(f)
2 Measured value of Phase of H(f)
Initialization: Trained Model.
for i=1:length(”Merged Dataset”) do
current datatset ith dataset
T rained Model current dataset
Output P arameter s T r ained M odel
end for
Output:
1 Simulated Values of Amplitude of H(f)
2 Simulated Values of Phase of H(f)
performance of ANN model get affected by the distance and frequency for which
the model is being used. It has been found ANN model are prone to errors when the
receiver is in the vicinity of transmitter [18]. Also, for higher frequency small-scale
fading affects the accuracy of ANN model [19].
Simulation Results
We used the proposed ANN Model across 15 sets of different transmitter-receiver
distance. To investigate the accuracy of ANN based model, comparative analysis
between CTFs extracted from measured data after applying the Hilbert transform
and from simulated CTFs calculated from the proposed ANN model is done. As
shown in Fig.4the trend of simulated CTF extracted from ANN model is approxi-
mately following the upwards and downwards trend of measured CTF values across
the whole bandwidth for a given set of transmitter and receiver(Tx-Rx) distance.
and observed that due to limited number of available datasets it was observed that
the curve tends to overfit at some points.
Tapped Delay Line Model
The dense physical environment inside a bus give rise to highly clustered wireless
channels resulting in distinctive multipaths. Tapped Delay Line models allows to
differentiate between these scattered multipath components with each tap repre-
senting a single component. The PDP obtained from such multipath channels can
be assumed as time average of tap gains at corresponding tap delays.
As evident from Fig.6when a time-average trend for measured PDP was simu-
lated, the average trend closely follows the simulated PDP trend. This time-averaged
PDP trend when sampled at different tap delays and compared with sampled simu-
lated PDPs we found that the difference between the tap gains for average measured
PDP and simulated PDP is quite low(Fig.7). Thus, we can advocate that the sim-
ulated PDP can be considered as an average trend of measured PDP. We further
validated our results with goodness of fit parameters (Table.1) like root mean square
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 8 of 11
error (RMSE) and mean absolute error (MAE) by using (12) and (13).
MAE(i) = PNs
i=1|Ppred (:,i) Pmea(:,i)|
Ns
(12)
RM SE (i) = s|Pmea(:,i) Ppred(:,i)|2
Ns
(13)
Understanding the Errors
As discussed earlier in the section generating channel characteristics synthetically
using ML algorithms is prone to errors due to the complex and cluttered environ-
ment. Thus, we analysed synthetic channel measurements with measured channel
measurements. Table 1shows the RMSE and MAE error metrics for varying distance
between transmitter and receiver. Table 2shows the variation in channel parame-
ters when the passengers were present inside the bus as well as when passengers are
absent. Here, we would like to add that due to low availability of datasets where
the passengers are present error is high compared to when passengers are absent. It
might also reflects that with increase in number of passengers the number of multi-
path components increased thus showing high errors. We also analysed the trends
for errors with change in distance (Fig.8). Here we observed that the trend for a
particular tap is consistently following the trend of time-averaged error (or differ-
ence) trend which signifies that the model remains consistent across the distance
and time vectors used in the study.
Conclusions
This article proposed a six-layer ANN based framework for 60 GHz wireless commu-
nication link inside a bus. The comprehensive study of frequency domain channel
sounding for mmWave measurement campaign inside a bus led to development of
simulated PDP using dynamic channel parameters. The highly comparable trend
between measured and simulated PDP supports that PDPs can be synthetically
generated using ANN. Comparing the simulated PDP trend with the time-averaged
tap delay gains of the measured PDP indicates that proposed ANN based model
can synthetically generate PDPs for given channel parameters. We also found that
with more availability of measurement datasets (especially in presence of passen-
gers) can significantly improve the model performance. From our observations we
can further implies that the ANN based model can extrapolate the PDP values for
the cases where the channel measurement data is unavailable. Such ANN models
are also helpful for Bit-error rate simulation using tapped delay models.
Abbrevations
mmWave: Millimeter Wave; CTF : Channel Transfer Function; PDP : Power
Delay Profile IVC : Intra-Vehicular Communication; IC : Integrated Circuits; GPS
: Global Positioning; ANN : Artificial Neural Network; GHz : Giga Hertz; TDL :
Tapped Delay Line; AP : Access Point; V2I : Vehicle-to-Infrastructure; DC : Direct
Current ; PA : Power Amplifier; MHz : Mega Hertz; dBm : Decibel-Milliwatts
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 9 of 11
OWGA : Open-Waveguide Antenna; SIW SA: Substrate-Integrated-Waveguide
Slot Antenna; GPIB : General Purpose Interface Bus; DL : Deep Learning; MLP
: Multi-Layer Perceptron; ReLU : Rectified Linear Units; RMSP : Root Mean
Sqaure Propagation; VNA : Vector Network Analyzer; ML : Machine Learning;
LOS : Line of Sight; MAE : Mean Absolute Error; RMSE : Root Mean Square
Error;
Availability of data and materials
The data that support the findings of this study are available from the corresponding author, upon reasonable
request.
Competing interests
The authors declare that they have no competing interests.
Funding
This work was developed within a framework of the research grants: project no. 17-27068S sponsored by the Czech
Science Foundation, grant no. LO1401 sponsored by the National Sustainability Program, grant no.
GBMON/13-996/2018/WAT sponsored by the Polish Ministry of Defense, grant no. UGB/22-854/2021/WAT
sponsored by the Military University of Technology, and grant no. CRG/2018/000175 sponsored by SERB, DST,
Government of India.
Author’s contributions
RS conceptualized the work, performed simulations and derived comparatives. ANS wrote the ANN based algorithm
and jointly wrote the manuscript with RS. AC supervised the project. JK, CZ, TM, AP are the foreign collaborators
responsible for obtaining measured datasets along with AC.
Acknowledgements
Not Applicable.
Author details
1ECE Department, National Institute of Technology, M. G. Avenue, 713209 Durgapur, India. 2Institute of
Communications Systems, Faculty of Electronics, Military University of Technology , Warsaw, Poland. 3Department
of Radio Electronics, Brno University of Technology, 61600, Brno, Czech Republic.
References
1. Chandra, A., Mikulasek, T., Blumenstein, J., Prokes, A.: 60 GHz mmW channel measurements inside a bus. In:
2016 8th IFIP International Conference on New Technologies, Mobility and Security (NTMS), pp. 1–5 (2016)
2. Rappaport, T.S., Xing, Y., MacCartney, G.R., Molisch, A.F., Mellios, E., Zhang, J.: Overview of millimeter
wave communications for fifth-generation (5g) wireless networks—with a focus on propagation models. IEEE
Transactions on Antennas and Propagation 65(12), 6213–6230 (2017)
3. Hong, W., Jiang, Z.H., Yu, C., Hou, D., Wang, H., Guo, C., Hu, Y., Kuai, L., Yu, Y., Jiang, Z., Chen, Z., Chen,
J., Yu, Z., Zhai, J., Zhang, N., Tian, L., Wu, F., Yang, G., Hao, Z.-C., Zhou, J.Y.: The role of millimeter-wave
technologies in 5g/6g wireless communications. IEEE Journal of Microwaves 1(1), 101–122 (2021)
4. Azpilicueta, L., opez Iturri, P., Aguirre, E., Astrain, J.J., Villadangos, J., Zubiri, C., Falcone, F.:
Characterization of wireless channel impact on wireless sensor network performance in public transportation
buses. IEEE Transactions on Intelligent Transportation Systems 16(6), 3280–3293 (2015)
5. Suzuki, H., Chen, Z., Collings, I.B.: Analysis of practical mimo-ofdm performance inside a bus based on
measured channels at 5.24 ghz. In: The Second European Conference on Antennas and Propagation, EuCAP
2007, pp. 1–8 (2007)
6. Matolak, D.W., Chandrasekaran, A.: 5 ghz intra-vehicle channel characterization. In: 2012 IEEE Vehicular
Technology Conference (VTC Fall), pp. 1–5 (2012)
7. Li, B., Zhao, C., Zhang, H., Sun, X., Zhou, Z.: Characterization on clustered propagations of uwb sensors in
vehicle cabin: Measurement, modeling and evaluation. IEEE Sensors Journal 13(4), 1288–1300 (2013)
8. Chandra, A., Rahman, A.U., Ghosh, U., Garc´ıa-Naya, J.A., Prokeˇs, A., Blumenstein, J., Mecklenbr¨auker, C.F.:
60-ghz millimeter-wave propagation inside bus: Measurement, modeling, simulation, and performance analysis.
IEEE Access 7, 97815–97826 (2019)
9. Rahman, A.U., Ghosh, U., Chandra, A., Prokes, A.: Channel modelling for 60ghz mmwave communication
inside bus. In: 2018 IEEE Vehicular Networking Conference (VNC), pp. 1–6 (2018)
10. Adeogun, R.: Calibration of stochastic radio propagation models using machine learning. IEEE Antennas and
Wireless Propagation Letters 18(12), 2538–2542 (2019)
11. Seretis, A., Sarris, C.D.: Toward physics-based generalizable convolutional neural network models for indoor
propagation. IEEE Transactions on Antennas and Propagation 70(6), 4112–4126 (2022)
12. Bharti, A., Adeogun, R., Pedersen, T.: Auto-generated summaries for stochastic radio channel models. In: 2021
15th European Conference on Antennas and Propagation (EuCAP), pp. 1–5 (2021)
13. Bakirtzis, S., Chen, J., Qiu, K., Zhang, J., Wassell, I.: An expedient, generalizable and realistic data-driven
indoor propagation model. TechRxiv (2021)
14. Lydia, A., Francis, S.: Adagrad—an optimizer for stochastic gradient descent. Int. J. Inf. Comput. Sci 6(5),
566–568 (2019)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 10 of 11
15. Chandra, A., Mikulasek, T., Blumenstein, J., Prokes, A.: 60 ghz mmw channel measurements inside a bus. In:
2016 8th IFIP International Conference on New Technologies, Mobility and Security (NTMS), pp. 1–5 (2016)
16. Seretis, A., Sarris, C.D.: An overview of machine learning techniques for radiowave propagation modeling. IEEE
Transactions on Antennas and Propagation (2021)
17. Seretis, A., Zhang, X., Zeng, K., Sarris, C.D.: Artificial neural network models for radiowave propagation in
tunnels. IET Microwaves, Antennas & Propagation 14(11), 1198–1208 (2020)
18. Ayadi, M., Zineb, A.B., Tabbane, S.: A uhf path loss model using learning machine for heterogeneous networks.
IEEE Transactions on Antennas and Propagation 65(7), 3675–3683 (2017)
19. Popescu, I., Nafomita, I., Constantinou, P., Kanatas, A., Moraitis, N.: Neural networks applications for the
prediction of propagation path loss in urban environments. In: IEEE VTS 53rd Vehicular Technology
Conference, Spring 2001. Proceedings (Cat. No. 01CH37202), vol. 1, pp. 387–391 (2001). IEEE
Figures
Figure 1 Experimental Set-up for Bus Measurement Campaign
Figure 2 Block Diagram for Proposed Methodology
Figure 3 Flow Chart for Pre-processing
Figure 4 Measured and Simulated CTF Comparison at 3.66 meter distance inside bus
Figure 5 Measured and Simulated PDP Comparison at 3.66 meter distance inside bus
Figure 6 Average Trend for Measured PDP vs Simulated PDP Trend
Figure 7 Tap Gain Comparison for Measured POP and Simulated PDP Trend
Figure 8 Comparison between Average Tap Gain Difference and Tap Gain Difference at 30ns
Tables
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
Shukla et al. Page 11 of 11
Table 1 Error Metrics measured and simulated PDPs.
Tx-Rx
Distance
(in meters)
RMSE
(in dBm)
MAE
(in dBm)
1.18 2.737 2.304
1.47 4.083 2.804
1.66 5.094 3.660
2.24 3.617 2.954
2.35 4.264 3.108
3.66 4.060 2.820
3.7 2.800 2.273
5.12 2.313 1.429
5.16 2.757 1.756
6.66 3.213 2.200
6.76 2.522 1.654
8.18 2.263 1.400
8.36 2.664 1.791
9.72 3.821 2.698
9.75 3.439 2.271
Table 2 Effect of presence of passengers
Passenger
Status
MAE
(in dBm)
RMSE
(in dBm)
Empty 2.27 2.80
Half Filled 4.76 5.12
Fully Filled 12.07 12.27
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
GPIB Cable
Coaxial Cable
PA
Adapter
Cooler
Mixer
SIW SA
Analog Signal
Generator
Signal Source
Analyser
OWGA
55-65 GHz
Transmitter Section Receiver Section
WC
Transmitter
Receiver
Figure No. 1
Measured
Data ANN Model Trained Model Simulated CTF
Test Dataset
TDL Model
(Measured)
PDP
(Measured)
CTF
(Measured)
PDP (Simulated)
Raw Data
Training
Parameter
Extraction
Parameters:
1. Frequency
2. Tx-Rx Distance
3. Seat Number
4. Window/Aisle
5. Left Side/Right
Side
6. Miscellaneous /
Others
Amplitude
input:6
hidden:
100
hidden:
50
hidden:
20
hidden:
200
output: 2
Activation: Relu
Activation:
Sigmoid
Phase
Figure No. 2
Tuning
hyperparameters
Reading of
Merged Dataset
Label Encoding
Model evaluation
Feature
Extraction
Splitting pre-
processed data
Standardizing
or Scaling
Input
Feature Pre-
Processing
Training
Insert
Merged
Dataset
Figure No. 3
55 56 57 58 59 60 61 62 63 64 65
Frequency [GHz]
-95
-90
-85
-80
-75
-70
-65
-60
Channel Transfer Function [dB]
Measured |H(f)|
Simulated |H(f)|
Figure No. 4
0 5 10 15 20 25 30 35 40 45 50
Time Delay [ns]
-160
-140
-120
-100
-80
-60
-40
Received Power [dBm]
Measured PDP
Simulated PDP
Figure No. 5
0 5 10 15 20 25 30 35 40 45 50
Time Delay [ns]
-160
-140
-120
-100
-80
-60
-40
Received Power [dBm]
Average Gain
Measured PDP
Simulated PDP
Figure No. 6
5 10 15 20 25 30 35 40 45 50
Time Delay [ns]
-140
-120
-100
-80
-60
-40
-20
0
Mean Tap Gain [dBm]
Measured PDP
Simulated PDP
Figure No. 7
12345678910
Distance (in meters)
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
3
Mean Tap Gain [dBm]
Average Tap Gain Difference
Tap Gain Difference at 30 ns
Figure No. 8
Original Manuscript PDF
Click here to access/download
Supplementary Material
Manuscript.pdf
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Ever since the deployment of the first-generation of mobile telecommunications, wireless communication technology has evolved at a dramatically fast pace over the past four decades. The upcoming fifth-generation (5G) holds a great promise in providing an ultra-fast data rate, a very low latency, and a significantly improved spectral efficiency by exploiting the millimeter-wave spectrum for the first time in mobile communication infrastructures. In the years beyond 2030, newly emerged data-hungry applications and the greatly expanded wireless network will call for the sixth-generation (6G) communication that represents a significant upgrade from the 5G network – covering almost the entire surface of the earth and the near outer space. In both the 5G and future 6G networks, millimeter-wave technologies will play an important role in accomplishing the envisioned network performance and communication tasks. In this paper, the relevant millimeter-wave enabling technologies are reviewed: they include the recent developments on the system architectures of active beamforming arrays, beamforming integrated circuits, antennas for base stations and user terminals, system measurement and calibration, and channel characterization. The requirements of each part for future 6G communications are also briefly discussed.
Article
Full-text available
This letter proposes a machine learning based method for the calibration of stochastic radio propagation models. Model calibration is cast as a regression problem involving mapping of the channel transfer function or impulse response to the model parameters. A multilayer perceptron is trained with summary statistics computed from synthetically generated channel realizations using the model. To calibrate the model, the trained network is used to estimate the model parameters from channel statistics obtained from measurements. The performance of the proposed method is evaluated with propagation graph and Saleh-Valenzuela models using both simulated data and in-room channel measurements. Results show accurate estimation of the parameters of both models.
Article
Full-text available
Millimeter-wave (mmWave) transmission over the unlicensed 60 GHz spectrum is a potential solution to realize high-speed internet access, even inside mass transit vehicles. The solution involves communication between users and a mmWave-band on-board unit which aggregates/ disseminates data streams from/ to commuters, and maintains connection with the nearest terrestrial network infrastructure node. In this paper, we provide a measurement-based channel model for 60 GHz mmWave propagation inside a typical inter-city bus. The model characterizes power delay profile (PDP) of the wireless intra-vehicular channel and is derived from about a thousand data sets measured within the bus. The proposed analytical model is further translated into a simple simulation algorithm which generates in-vehicle channel PDPs. Different goodness-of-fit tests confirm that the simulated PDPs are in good agreement with the measured data. Finally, a tapped-delay-line (TDL) channel model is formulated from the proposed PDP model, and the TDL model is used to study bit error rate (BER) performance of mmWave link inside bus under varying data rates and link lengths.
Conference Paper
Full-text available
The 5G vision acknowledges intravehicular communication as a means to enable passenger connectivity on the move. The capacity demand in a public transport vehicle is multi-fold compared to personal cars as there are more people on-board. In order to meet the demand, 5G standardization bodies prescribe moving the spectrum up to the millimetre wave (mmWave) regime. In this paper, we focus on buses as they are the most pervasive form of public transportation, and provide a wideband wireless channel model for 60GHz mmWave propagation inside bus. The model characterizes power delay profile (PDP) of the wireless intravehicular channel and is derived from about a thousand measured datasets within a bus. The proposed analytical model is further translated to a simple simulation algorithm which generates in-vehicle channel PDPs. The simulated PDPs are in good agreement with the measured data.
Article
Efficient and realistic indoor radio propagation modelling tools are inextricably intertwined with the design and operation of next generation wireless networks. Machine learning (ML)-based radio propagation models can be trained with simulated or real-world data to provide accurate estimates of the wireless channel characteristics in a computationally efficient way. However, most of the existing research on ML-based propagation models focuses on outdoor propagation modelling, while indoor data-driven propagation models remain site-specific with limited scalability. In this paper, we present an efficient and credible ML-based radio propagation modelling framework for indoor environments. Specifically, we demonstrate how a convolutional encoder-decoder can be trained to replicate the results of a ray-tracer, by encoding physics-based information of an indoor environment, such as the permittivity of the walls, and decoding it as the path-loss (PL) heatmap for an environment of interest. Our model is trained over multiple indoor geometries and frequency bands, and it can eventually predict the PL for unknown indoor geometries and frequency bands within a few milliseconds. Additionally, we illustrate how the concept of transfer learning can be leveraged to calibrate our model by adjusting its pre-estimate weights, allowing it to make predictions that are consistent with measurement data.
Article
A fundamental challenge for machine learning models for electromagnetics is their ability to predict output quantities of interest (such as fields and scattering parameters) in geometries that the model has not been trained for. Addressing this challenge is a key to fulfilling one of the most appealing promises of machine learning for computational electromagnetics: the rapid solution of problems of interest just by processing the geometry and the sources involved. The impact of such models that can “generalize" to new geometries is more profound for large-scale computations, such as those encountered in wireless propagation scenarios. We present generalizable models for indoor propagation that can predict received signal strengths within new geometries, beyond those of the training set of the model, for transmitters and receivers of multiple positions, and for new frequencies. We show that a convolutional neural network can “learn" the physics of indoor radiowave propagation from ray-tracing solutions of a small set of training geometries, so that it can eventually deal with substantially different geometries. We emphasize the role of exploiting physical insights in the training of the network, by defining input parameters and cost functions that assist the network to efficiently learn basic and complex propagation mechanisms.
Article
Gradient Descent algorithms though popularly used, it still remains to work as a black-box with much of the tuneable hyper-parameters being unexplored. These hyper-parameters employ proximal functions to control the gradient steps which supports for online and adaptive learning. The earlier algorithms required the manual initialization of the hyper-parameters before the training process begins, and it remains static throughout the training. By incorporating Optimizers for existing algorithms, hyper-parameter initialization and updation is automatically done by the algorithm. This article tends to give an insight about these hyper-parameters, its nature and the purpose of adding these hyper-parameters to enhance the performance of Gradient Descent Algorithms.
Article
This paper provides an overview of the features of fifth generation (5G) wireless communication systems now being developed for use in the millimeter wave (mmWave) frequency bands. Early results and key concepts of 5G networks are presented, and the channel modeling efforts of many international groups for both licensed and unlicensed applications are described here. Propagation parameters and channel models for understanding mmWave propagation, such as line-of-sight (LOS) probabilities, large-scale path loss, and building penetration loss, as modeled by various standardization bodies, are compared over the 0.5-100 GHz range.