Download as pdf or txt
Download as pdf or txt
You are on page 1of 60

EXAMINING THE EFFICACY OF FHSS AND OFDM

PROTOCOLS FOR DRONE COMMUNICATIONS IN A


MULTIPATH ENVIRONMENT

A THESIS PRESENTED BY

JOSHUA LABRANCHE

SUBMITTED TO THE FRANCIS COLLEGE OF ENGINEERING


UNIVERSITY OF MASSACHUSETTS LOWELL,
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE
DEGREE OF

MASTER OF SCIENCE IN ENGINEERING


FEBRUARY 2022

DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING


©2022 by Joshua LaBranche
All rights reserved
EXAMINING THE EFFICACY OF FHSS AND OFDM
PROTOCOLS FOR DRONE COMMUNICATIONS IN A
MULTIPATH ENVIRONMENT

BY

JOSHUA LABRANCHE
B.S. UNIVERSITY OF MASSACHUSETTS LOWELL (2014)
M.S. UNIVERSITY OF MASSACHUSETTS LOWELL (2017)

SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS


FOR THE DEGREE OF MASTER OF SCIENCE
DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING
UNIVERSITY OF MASSACHUSETTS LOWELL

Signature of
Author: Date: 08 December, 2021

Signature of Thesis^
< )
.
/y y
Supervisor:______ &
Name Typed: Jay A. Weitzen

Signatures of Other Thesis Committee Members:

Committee Member Signature:


Name Typed: Paul M. Robinette

Committee Member Signature:


Name Tÿped: Thanuka L. Wickramaranthne
EXAMINING THE EFFICACY OF FHSS AND OFDM

PROTOCOLS FOR DRONE COMMUNICATIONS IN A

MULTIPATH ENVIRONMENT

BY

JOSHUA LABRANCHE

ABSTRACT OF A THESIS SUBMITTED TO THE FACULTY OF THE


DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF


MASTER OF SCIENCE
UNIVERSITY OF MASSACHUSETTS LOWELL
FEBRUARY 2022

Thesis Supervisor: Jay A. Weitzen, Ph.D.


Professor, Department of Electrical and Computer Engineering
Abstract
In this thesis an end-to-end simulation model, with an emphasis on high multipath
indoor channel models, was constructed and executed to test the viability of reliable com­
munications between a user operated drone transmitter and the corresponding vehicle’s
receiver. The models constructed represent frequency hopping spread spectrum (FHSS)
and orthogonal frequency division multiplexing (OFDM) waveforms used by commercial-
off-the-shelf (COTS) drones for wireless communication. A high multipath channel model
was constructed based upon previous research done by the 802.1 in task group (TGn). Per­
formance results were assessed by measuring packet-error-rates (PERs). Communication
is deemed to be viable if these PERs allow for a sufficient amount of effective throughput
that would enable reliable communications of both video and control signals.

ii
Acknowledgments

First and foremost I would like to thank Dr. Rachel Learned (MITLL) for giving me
the opportunity to pursue my career and ultimately this thesis, the guidance to continue to
evolve my skill-set, and always assisting me in becoming the professional I strive to be. A
huge thanks to Dr. Jay Weitzen (University of Massachusetts Lowell) for being an excellent
professor during my graduate studies and continuing to be involved with my academic
pursuits through advising me throughout the entire thesis process. I’d also like to thank
David Romero (MITLL) for his continued endeavor of mentoring and guiding my career
development and my advancement of the understanding of the applicable concepts of this
thesis and beyond. A special thanks to Dr. Trey Shenk (MITLL) and Howard Goldowsky
(MITLL) for their outstanding companionship and encouragement as co-workers, as well
as helping me talk myself through concepts discussed in this thesis. I would also like to
acknowledge the good professors at the University of Massachusetts Lowell, especially
Kiwi, who have taught me so much throughout the entirety of my academic career. Lastly,
I want to thank and congratulate my parents for all the hard work and sacrifices they have
made to help make any of this possible.

iii
Contents
1 Introduction 1
1.1 Problem Statement............................................................................................... 1
1.2 COTS Drone Communications Comparison.................................................... 2
1.3 Channel Model: Multipath and Fading.............................................................. 4
1.4 FHSS Communications........................................................................................ 7
1.5 Orthogonal Frequency Division Multiplexing Communications.................... 12
1.6 Related Works........................................................................................................ 17

2 Methodology 20
2.1 FHSS for Telemetry................................................................................................... 20
2.2 OFDM for Video/Telemetry ................................................................................... 24

3 Results 29
3.1 FHSS Results............................................................................................................. 29
3.2 OFDM Results.......................................................................................................... 29

4 Discussion 33
4.1 FHSS Results Discussion..................................................................................... 33
4.2 OFDM Results Discussion.................................................................................. 33

5 Conclusions 36

6 Recommendations 37

7 Literature Cited 38

A Appendix 41
A.l MIMO Overview for 802.1 In.................................................................................. 41
A.2 802.11 TGn Channel Models.................................................................................. 44

iv
List of Figures
1 Comparing spectrum captures of DJI’s proprietary communication systems
Lightbridge (top) and OcuSync (bottom) [5]....................................................... 4
2 Example of multipath. As the signal propagates from the transmitter (Tx)

to the receiver (Rx), if no obstructions exist the signal can travel along the
LOS path p\. However, if there exists obstacles such as reflective surfaces
the signal may travel along either paths or p^. The signal can traverse
any of the three paths, and the receiver observes the superposition of these
signals with their respective time delay tn........................................................... 5
3 Finite shift register representation of the 1/2 rate K = 3 convolutional code
used............................................................................................................................ 8
4 State diagram example of a particular convolutional encoder sequence. The
initial configuration of the state machine begins in state 0, and as the bits are
passed through the encoder the state transitions based upon the input bits
and the current bits in the registers. As the encoder transitions states only a
certain number of next states are possible from the current state, and so the
process of encoding can be viewed as a traversal from state to state along
this diagram where the encoded sequence consists of the outputs from each
state............................................................................................................................ 9
5 FHSS transmitter modulator.................................................................................. 11
6 FHSS waveform spectrum example........................................................................ 12
7 OFDM waveform spectral density example........................................................... 12
8 16-QAM Constellation Diagram in a high signal-to-noise ratio (SNR) envi­
ronment. The x abscissa is known as the In-Phase portion of the waveform
and represents the cosine elements. The y abscissa is known as the Quadra­
ture portion of the waveform and represents the sine elements............................. 14
9 Constellation diagram examples of waveforms of particular interest to this
study...................................................................................................................... 15
10 OFDM waveform spectrogram example.............................................................. 16

v
11 Comparison of an OFDM waveform to a traditional single carrier modu­
lated waveform............................................................................................... 17

12 Simulink Overview of the FHSS Model................................................................... 20


13 Finite shift register representation of the 1/2 rate convolutional code used. . 21
14 High level overview of interleaving and deinterleaving......................................... 21
15 Simulink block diagram of the FHSS modulator................................................... 22
16 Simulink block diagram of the FHSS demodulator............................................. 23
17 Example of the parameters defining the TGn channel model D........................... 26
18 Channel response representations of channel model D.......................................... 27

19 Simulation results comparing the FHSS BER for the case of a strictly
AWGN channel and a high multipath channel............................................... 29
20 Comparison of PERs for various OFDM modulation and coding schemes.

Channel model D used...................................................................................... 30


21 Comparison of PERs for various OFDM modulation and coding schemes.
Channel model D used...................................................................................... 30
22 Comparison of PERs for various OFDM modulation and coding schemes.

Channel model D used...................................................................................... 31


23 Channel Model D:PERs and Effective Throughput measurements............... 31
24 Channel Model B:PERs and Effective Throughput measurements............... 32
25 Channel Model C:PERs and Effective Throughput measurements............... 32
26 Tables containing the parameters used in channel models B and C................. 45

vi
List of Tables
1 Comparison of UAV RC and Video transmission protocols.............................. 3
2 Overview of Wi-Fi used in the UAV communications........................................ 3
3 Overview of the FHSS simulation parameters utilized........................................... 24
4 2x2 MIMO 802.1 In MCS modulation and coding schemes. Note that
MCS values of 8 — 15 correspond to the case of two spatial streams (MCS
of 1 — 7 is one spatial stream, etc.)............................................................................ 25
5 Overview of the OFDM simulation parameters utilized......................................... 28
6 K-factors of each of the 6 models for the LOS and NLOS configurations.. . 44
7 Breakpoint distances and shadow loss standard deviations for each of the 6
channel models.................................................................................................. 48

vii
1 Introduction

Unmanned Aerial Vehicles (UAVs), often known as drones, can be utilized for a multi­
tude of tasks including, but not limited to, mobile sensor networking, small scale logistics,
surveillance and intelligence. These UAVs are controlled by users through a wireless radio
frequency (RF) communication protocol. It is important to classify and assess the effi­
ciency and reliability of such protocols for many different RF environments. In particular,
a high multipath environment is of interest due to the operation of the UAVs designed to
operate out of doors but are instead used in indoor and subterranean environments.
This thesis begins by presenting findings on the communication protocols and sub­
sequent waveforms used by the current state-of-the-art UAVs as well as a background
overview of these models. This is required foremost because it is important to under­
stand specifically which protocols are used. Next, the channel model and effects used are
discussed since it is important to model the environments of interest with a high degree
of accuracy. From there the detailed methodology of the simulation components are dis­
cussed with an emphasis on reconstruction. Lastly, the results of the simulations and their
respective analysis are discussed followed by the conclusions of the research and any future
recommendations.

1.1 Problem Statement

UAV communication protocols are designed to allow for a robust relay of information
between the vehicle and the respective controller. Starting with radio-control (RC) signals
from the controller to the vehicle, telemetry signals from the vehicle back to the controller,
and possibly video captured on the vehicle transmitted back to the controller to be saved or
streamed.
Additional consideration needs to be given towards the operating environment in ques­
tion. UAV communication protocols are typically designed for an ideal environment con­
sisting of a clear line-of-sight (LOS) channel. This is due to the fact that most UAVs are
operated in an outdoor environment where a LOS channel is readily available. If instead,
the UAV is operated in an indoor environment, and in particular, a corridor environment,

1
then the existence of a LOS channel is less probable. Additionally, the corridor environ­
ment presents many places for scatters and reflections to occur and thus a high amount of
multipath to be realized at the receiver.
MATLAB was used to construct representative models of the protocols in question,
models of the channels, and a method to simulate packet or bit error losses and rates
(PERs, BERs) for analysis of the system. Since the specific protocols vary for each make
and model combination of commercial off-the-shelf (COTS) UAVs, a general protocol was
developed for the two main waveform types utilized: frequency hopping spread spectrum
(FHSS) and orthogonal frequency division multiplexing (OFDM). With the protocols de­
veloped, the channels are tested and the results are analyzed by means of the amount of
packets or bits in error. Also, only the physical layer was developed and tested here, and
each packet was constructed by a random sequence of information.

1.2 COTS Drone Communications Comparison

It was found that commercial-off-the-shelf (COTS) UAVs use two main classes of
waveforms as their means of communication. The first type of waveform is known as
Frequency Hopped Spread Spectrum (FHSS) which is typically used in Bluetooth devices.
The second is Orthogonal Frequency Division Multiplexing (OFDM) which is traditionally
used in Wi-Fi and LTE devices. Not all UAV’s devices are capable of video transmissions,
but the ones that are generally 1 use some form of OFDM to communicate the video in­

formation. For radio control (RC) and telemetry communications UAVs use some form of
either FHSS or OFDM.
For example, in Table 1 it can be seen that the Parrot ANAFI, Yuneec Mantis Q, and the
DJI Mavic Air 2 all use OFDM to transmit both the control and video signals. In contrast,
the DJI Mavic Pro 1, DJI Mavic Pro 2, and the DJI Phantom 4 all use an FHSS waveform
for control communications and an OFDM waveform for the video communications. It
should be noted that the DJI products listed in the table use a purposefully designed pro­
prietary software by DJI for their communication systems, whereas the Yuneec Mantis Q
and the Parrot ANAFI both leverage 802.1 lx waveforms for their communication systems.
'it was found that some UAVs used analog Frequency Modulation (FM) for video communications.

2
Additionally, the operating frequency band for the systems can vary from model to model
as can be seen by the DJI Mavic Pro 2 being able to operate in both the 2.4 GHz and 5.8
GHz frequency bands, but the DJI Mavic Pro 1 only utilizes the 2.4 GHz band.
Remote Control Video
Brand - Model Protocol (Waveform) 2.4 GHz 5.8 GHz Protocol (Waveform) 2.4 GHz 5.8 GHz
DJI - Mavic Pro 2 OcuSync 2.0 (FHSS) X X OcuSync 2.0 (OFDM) X X
DJI - Mavic Pro 1 OcuSync (FHSS) X OcuSync (OFDM) X
DJI - Phantom 4 Lightbridge (FHSS) X Lightbridge (OFDM) X
DJI - Mavic Air 2 DJI Enhanced Wi-Fi (OFDM) X DJI Enhanced Wi-Fi (OFDM) X X
Yuneec - Mantis Q Wi-Fi 802.11 a/n (OFDM) X Wi-Fi 802.11 a/n (OFDM) X
Parrot - ANAFI Wi-Fi 802.11 a/b/g/n (OFDM) X X Wi-Fi 802.11 a/n (OFDM) X X

Table 1: Comparison of UAV RC and Video transmission protocols.

Wi-Fi for UAV Communications

In the absence of proprietary software/hardware used for communications, a reliable


alternative is to leverage predefined Wi-Fi 802.11 protocols and standards. The advantage
of this is the ability to create cost effective vehicles since the availability of transceiver
chip sets for 802.11 communications is high. In Table 2 a comparison of how the DJI
Mavic Air 2, Yuneec Manits Q, and the Parrot ANAFI use Wi-Fi and 802.11 protocols is
given. Interesting to note is that the Yuneec Mantis Q only uses the 5 GHz channels for
communications, which would decrease the reliable communication distance compared to
the 2.4 GHz signals, at the benefit of a higher data throughput, in the same channels.

DJI Mavic Air | Yuneec Mantis Q | Parrot ANAFT


Protocol DJI Enhanced Wi-Fi 802.11 b/g/n 802.11 a/n 802.11 a/b/g/n
Bandwidth 10 MHz 20 MHz 20 MHz 10 MHz/20 MHz
2.4 GHz Frequencies 2.412 - 2.462 GHz - 2.412 - 2.462 GHz
2.4 GHz Channels 1-13 - 1-13
5.18-5.24 GHz 5.18-5.24 GHz
5 GHz Frequency 5.745 - 5.825 GHz
5.745 - 5.825 GHz 5.745 - 5.825 GHz
5 GHz Channels 149,152,157,161,165 34,44,48,149,157,165 34,44,48,149,157,165

Table 2: Overview of Wi-Fi used in the UAV communications.

Frequency Hopping Spread Spectrum Considerations

The remaining drones in Table 1 use a form of FHSS for Radio Control (RC) commu­
nications and OFDM for the video communications. The benefit of this design is a more
robust and reliable aerial vehicle since a loss of video transmission does not equate to a

3
loss of RC transmission, unless the pilot is steering by utilizing the camera. This would ef­
fectively enable the possibility of a user being able to continue to control a UAV (visually)
in the absence of reliable OFDM signaling. Figure 1 shows a spectrum capture for the DJI
Phantom 4’s Lightbridge communication system in the top, and the DJI Pro l’s OcuSync
communication system on the bottom. The spikes appearing in both spectrum plots are the
underlying FHSS RC signal, and the wider bands of stronger (brighter) energy correspond

to the OFDM video signal.

Figure 1: Comparing spectrum captures of DJI’s proprietary communication systems


Lightbridge (top) and OcuSync (bottom) [5],

The difference between the two systems is the hardware and software used for imple­
mentation. The Lightbridge system was developed first by DJI and includes both custom
designed hardware and software for implementation. On the other hand, the OcuSync sys­
tem was developed later and takes a more software-defined radio approach which would
allow for future developments to be easily implemented without swapping out hardware
components. The differences from an RF point-of-view are minimal in comparison.

1.3 Channel Model: Multipath and Fading

Ideal situations for wireless communications exists in channels where the transmitter
and receiver are within a visible line-of-sight (LOS) with no physical obstructions present
to allow for reflections or scatters of the signal-of-interest (SOI) to be observed at the
receiver. However, some channels do contain physical obstacles that allow for the signal to

4
P2

Rx

P3

lpt1 < IP2I < IP3I => Tl < T2 < t3

Figure 2: Example of multipath. As the signal propagates from the transmitter (Tx) to the
receiver (Rx), if no obstructions exist the signal can travel along the LOS path p\. However,
if there exists obstacles such as reflective surfaces the signal may travel along either paths
P2 or p3- The signal can traverse any of the three paths, and the receiver observes the
superposition of these signals with their respective time delay Tn.

be reflected about a surface’s face or scattered about a surface’s vertex to be received at the
receiver with a seemingly probabilistic amplitude scaling and time delay with respect to the
LOS signal path. When this occurs the communication operates in a multipath channel, and
the various amplitude variations are known as fading. The receiver then observes the SOI
as a superposition of these signal copies that traverse multiple paths to the receiver. If the
transmitted signal is denoted as s(t\ then the received signal in a multipath environment
can be modeled as,

=£«»(') exp(-27tjfcTn(t)Xt ~ (1)


n

where % represents the nth signal copy’s respective scaling parameter that may vary

with time t, Tn(t) represents the time delay for the received signal for time t, and fc is the
carrier frequency. Also, if the channel is modeled as an impulse of the respective signal
delays and amplitudes it can be defined similarly as,

= ^aw(t)exp(-2^j/cTn(t))5(t - "%(;)) (2)


n

For the case when the impulse response of the channel model c(r;t) is modeled as
a zero-mean complex Gaussian random process, the envelope of the channel |c(f;t)| is

5
Rayleigh distributed. When the Gaussian process can no longer be modeled as zero-mean
and there is a dominant LOS path between the transmitter and the receiver, the envelope
takes on a Rice distribution. The channel overall is modeled as a linear time variant fading
multipath channel.

Rician and Rayleigh Fading

There are two probability distributions of interest used to model the multipath com­
ponent’s received amplitude. They are known as a Rician and Rayleigh distribution. The
Rayleigh distribution is used when no LOS is present between the transmitter and receiver.
Conversely, the Rician distribution is used for fading channel models where a LOS path is
present. The two parameters used to define a Rician distribution are

K=w

fl = A2 + 2a2

where A2 is the power in the LOS path at the receiver, and 2a2 is the cumulative power

in the other non-LOS paths. This results in the following distribution

(^+1)^\ y l
pM =

which represents the probabilistic nature of the distribution of the received signal am­
plitudes. If K —> oo then the multipath components become insignificant and the channel
is representative of a LOS channel with a single dominant signal path. Alternatively, in
the case where no LOS is present and K —> 0 the Rician distribution approaches that of a
Rayleigh one. This can be shown by setting K = 0 and realizing the summation is equal to
the modified Bessel function of the first kind, 7q(*) which can be shown to equal 1 when
the argument is 0,

6
1 y*
7q(z) = exp(zcos(z))A

Zq(O) = — [ Idt
TC Jo
fo(O) = ^-O]

Zo(O) = 1

Thus, the distribution becomes Rayleigh and is given by

2x f—x^X
PM = —exp 1 — 1 x>0 (4)

1.4 FHSS Communications

To start, a mathematical representation for a Frequency Hopping Spread Spectrum


(FHSS) waveform is given. This overview gives a general model of an FHSS waveform,
and includes some particulars used for this body of work. It should be noted that in essence
there are two fundamental forms of FHSS waveforms. The first being fast hopping and the
other being slow hopping. For fast hopping, the hop duration is less than that of the symbol
duration2, whereas for slow hopping the hop duration is much longer than the symbol dura­

tion. In this body of work there is a focus on a slow hopping system. Fast hopping systems
are more agile and therefore more resilient to interference or intentional signal jamming,
but at the cost of increased complexity in terms of equivalent performance to that of a slow
hopping system.

Convolutional Coder and Viterbi

2Hop duration measures the amount of time for which the signal exists on a single frequency channel
determined by the carrier waveform and is a predefined parameter. The symbol duration measures the amount
of time for which the receiver processes a single symbol and is determined by the sampling frequency as well
as the oversampling factor. For example, if our receiver samples at a rate of 10 nanoseconds, or 100 MHz,
and is designed to sample each symbol 100 times, then the symbol duration is equal to 1000 nanoseconds, or
I microsecond.

7
In the interest of making the system more robust to long periods of consecutive errors
known as burst errors, a convolutional encoder and interleaver are employed prior to the
modulating of the data. The interleaver is a defined process to allow for the system to
rearrange the bit sequence prior to transmission and then perform the inverse function again
on the received data to arrange back to the original sequence. The particular interleaver
used is discussed in more detail in section 2.1. When operating in a fading environment,
if a packetized hop is subjected to a deep fade then it is reasonable to assume that the
receiver will be unable to decode this packetized hop and thus produce consecutive bit
errors known as burst errors. Interleaving is recommended when using convolutional codes
in a fading environment because the decoding performance of these codes are sensitive to
burst errors. From [14], theorem 7.2.2 defines the length of at most a single burst error
within a codeword that can be corrected by interleaving a given burst error correcting code.

In practice, the convolutional encoder follows the bit operations shown in Figure 3. To
begin, an input bit is fed into the encoder and at each state s, a bit-wise xor operation is
performed. The output bit gi is constructed as the binary sum of the input data, output of
51, and the output of $2. Similarly, the output bit g2 is constructed as the binary sum of
the input data, output of 5q, output of 5i, and the output of 52. These two output bits are
concatenated to produce the encoded data sequence. Since I bit was input and 2 bits were
output, this code is known as a 1/2 rate code and the code rate gives the ratio of the number
of information bits to the number of transmitted bits.

9, output

Input
data i
z-1 Z-1 Z-1

output

Figure 3: Finite shift register representation of the 1/2 rate K = 3 convolutional code used.

Shown in Figure 4 is an example of the sequence output from this particular encoder
as well as the states traversed to obtain the sequence. This representation is especially use­
ful when considering how to decode the sequence. Since the encoder can be graphically

8
represented with the trellis shown in Figure 4. One widely known method of decoding is
to find the most likely path of traversal given the output sequence and the trellis diagram.
The accuracy of this approach is increased as the length of the trellis that is used is also
increased. This approach is known as the Viterbi algorithm [27, 20].

Output: 10 01 10 11

Figure 4: State diagram example of a particular convolutional encoder sequence. The initial
configuration of the state machine begins in state 0, and as the bits are passed through the
encoder the state transitions based upon the input bits and the current bits in the registers.
As the encoder transitions states only a certain number of next states are possible from the
current state, and so the process of encoding can be viewed as a traversal from state to state
along this diagram where the encoded sequence consists of the outputs from each state.

Mathematical Representation

As previously mentioned, some of the systems use FHSS waveforms to communicate.


These FHSS waveforms are generated by using a pseudo-random sequence of carrier fre­
quencies for data to be modulated upon for transmission. The data is digitally modulated
at baseband to impose the data on the baseband carriers which are then modulated to the
desired frequency, and for the desired time duration, as determined by the pseudo-random
generator. To start, the data signal is a special case of an orthogonal binary frequency shift
keying (BFSK) waveform that is shaped with a low pass Gaussian filter to preserve spectral
efficiency. Next, the rectangular 0 and 1 binary data is shaped with a low pass Gaussian
filter given by,

. . 1 . t2 .
(5)

9
where a is related to the filter’s 3-dB bandwidth B by:

vN2)
(6)
2 xB

The 3-dB bandwidth B multiplied by the bit duration % gives what is known as the
BT product, and is a defining parameter of the GFSK waveform. The bandwidth B is
determined from the bit duration and the defined BT product. In particular, the bit duration
used was 2 fis and the BT was 1, giving a 3-dB bandwidth of B = 500 kHz. For a BFSK

waveform, the data is transmitted by changing the waveform’s frequency with respect to a
desired 0 or 1.

y^cos(2 0
— (7)
0 otherwise

where z = [1,2], Eb is the energy per bit, and Tb is the bit duration. If the number of
frequencies is increased from 2 to M, the signal becomes an M-ary FSK signal. Using
a baseband GFSK signal to modulate the data, the signal is then spread across the de­
sired bandwidth by mixing the signal with the pseudo-random FSK carriers to generate
a Frequency Hopping Spread Spectrum waveform. Given in eq. 8 is the mathematical
representation for a FHSS waveform,

Sj(t) = Ajdj(t) cos ([Wq + hj(t)Aw\t + 0j) 0 <t <T (8)

where Aj,dj(t), Wq,Aw, Qj are the signal amplitude, data signal (GFSK), hopping
waveform center frequency, hopping pattern function, hopping frequency spacing, and the
signal phase respectively. The hopping pattern is used in conjunction with the center
frequency Wb and frequency spacing Aw to generate the respective hop’s carrier frequency.
In particular, using a Wq in the 2.4 GHz band and a Aw of 1 MHz, an hj(f) can be con­
structed such that for a desired hop duration time, Th seconds per hop, produces a pseudo­
random integer value from the set [—24,24] 3 Z representative of the 1 MHz offset from
Wq. Figure 5 shows the block diagram used for generating the FHSS waveform.
Notice that an FSK block is used in the place of the frequency synthesizer. Effectively,

10
"TJTlWXM
CPM

Out
GFSK
Modulation

ulww
To 50-FSK
Frame

Generate
49 possible carriers
-24MHz to 24MHz

Figure 5: FHSS transmitter modulator.

the FSK block provides a means to synthesize pseudo-random carrier frequencies of a pre­
defined number of hopping frequencies to be mixed with the GFSK data signal.

FHSS Example

For example, in Figure 6 a baseband GFSK signal was generated and then modulated
with one of a possible 49 frequencies to be transmitted. In practice, a separate FSK mod­
ulator was used to construct the frequency hops that are represented by the spikes in the
top plot of the figure. The time duration of the pulse is shown in the bottom, and the hop
duration is shorter than the time window so the current hop goes from -1 MHz to 4 MHz.
The advantage of this particular waveform is the agile nature of the short hop duration
and the ability to spread the signal across the spectrum. The agile nature is advantageous
to any deficiencies within a given hop band, or a particular sub-band within the frequency
band of interest, since the waveform does not dwell in this region for long if at all. Impair­
ments such as interference and frequency fading can be minimized with proper construction
of the FHSS waveform.

11
Wl

Frequency (MHz)

Figure 6: FHSS waveform spectrum example.

1.5 Orthogonal Frequency Division Multiplexing Communications

The other main waveform utilized for UAV communications is OFDM. Orthogonal
frequency division multiplexing is a form of frequency division multiplexing where signals
are transmitted simultaneously on different orthogonal subcarriers. Orthogonal here means
that the subcarriers are separated in frequency by the inverse of the signal’s symbol period
such that the nulls of adjacent subcarriers coincide with the peak of the main lobe of the
subcarriers in the frequency domain.

-20

-25 -

-30 -
PSD (dB)

CO

Q
CD
-35 -

-40 -

-45 L
-10 -5 o 5 10
Frequency (MHz)

Figure 7: OFDM waveform spectral density example

12
M-PSK and M-QAM

The underlying signals that are transmitted over orthogonal subcarriers are either Binary
Phase Shift Keying (BPSK), Quadrature Phase Shift Keying (QPSK), and either 16 or 64
M-ary Quadrature-Amplitude Modulation (M-QAM).
To begin, QAM signals can be represented as:

$m(r) =ABKg(r)cos(2rc/^)+Amjg(r)sm(2^ 0 <t<T (9)

where m = 1,2Amc and Ams are the information-bearing signal amplitudes, and
g(t) is the signal pulse. In particular the raised cosine spectrum is utilized, and the square
root is applied to both the transmit pulse and the receiver’s matched filter. Therefore,

sin(m/T) cos
Srrcft) — m/T l-(4f2;2/T2) 1 J

Alternatively, eq. 9 can be rewritten as,

(y/Amc+A%Mt) cos(2 afct + tan-1 (Amî/Amc)) 0 <t<T (11)

and it can be seen that both the resulting waveform’s amplitude and phase are a function
of the information-bearing signal amplitude levels. It should be noted that these amplitude
levels Amc and A^ are independent sets. For example, if the signal amplitudes Amc and Ams
can be represented by a discrete set of amplitudes then the signals are known as Rectangular
QAM. In particular, for the set of amplitudes {(2m— Y—M^a, m = 1,2, ...4), where a
is the distance between adjacent points, the resulting QAM signal is a 16-QAM signal as
shown in Figure 8
Next, PSK signals are ones that maintain a constant amplitude and utilize only changes
in phase to modulate the data. These signals can be represented as follows,

=g(t)[cos(^(m- l))cos(2%/^) -sin(^(m- l))sm(2n^ 0 < t < T (12)

13
16 OAM
Figure 8: 16-QAM Constellation Diagram in a high signal-to-noise ratio (SNR) environ­
ment. The x abscissa is known as the In-Phase portion of the waveform and represents the
cosine elements. The y abscissa is known as the Quadrature portion of the waveform and
represents the sine elements.

where g(t) is again the signal pulse shape grrc(t), and the ®m = ’ m = 1,2, ...M
are the possible phases used to represent the data.
Notice that eq. 11 and eq. 12 are equivalent when M = 4 and Ams = Amc = ±V2/2.

This is the special case of Rectangular 4-QAM and QPSK which can be used to represent
the same waveform. Also, when M = 2 for eq. 12, the waveform is known as BPSK and
contains two IQ points each representing a bit of data. Shown in Figure 9 are examples of
the four waveforms of particular interest to this study. Two of the signals are PSK, and the
other two signals are M-QAM signals.
The benefit of increasing the number of points on the constellation diagram is to allow
for more data to be modulated per symbol and increase the effective data rate. For example,
a user would be able to transmit twice as much data per symbol for the case of a QPSK
signal in comparison to that of a BPSK signal since the number bits per symbol could be
doubled. The cost of this increase in data rates is an increase in the power required for
equivalent probabilistic error performance. This is due to the minimum distance between
constellation points decreasing as the number of constellation points increases for equiva­
lent energy per symbol levels. In short, the higher the data rate the more power needed for
comparable error-free transmission.

OFDM and the Subcarriers

14
BPSK QPSK

16 QAM 64 QAM

Figure 9: Constellation diagram examples of waveforms of particular interest to this study.

Orthogonal Frequency Division Multiplexing (OFDM) is a communication method that


transmits multiple data signals simultaneously by separating the signals in frequency Af
such that the nulls of adjacent subcarriers coincide with the peaks of the main lobes of the
subcarrier. This is the orthogonal portion of OFDM. The frequency part of OFDM comes
from subcarriers being divided into frequency bins across the band. Multiplexing is a nice
way to say simultaneous communication of multiple signals on the same communication
channel. Therefore OFDM signals are waveforms that communicate multiple signals si­
multaneously by means of dividing up the frequency spectrum for the signals such that
they are orthogonal. An example of an OFDM signal is given in Figure 10.
In particular, the signals that modulate these subcarriers are PSK and QAM signals. If
these signals are denoted as X^, then the mathematical representation for the inverse Fourier
transform of the received signal is,

= J^exp^TT/kAyt) 0 < t < Tift (13)


N k

where AF = 1/Tift is the subcarrier frequency spacing, Tift is the inverse Fourier trans-

15
160 -
10

Time (ps)
Frequency (MHz) -1°

Figure 10: OFDM waveform spectrogram example.

form (IFT) symbol period, and N is the number of samples in the IFT. The represent
the yh subcarrier’s modulated symbols signal. These are the previously discussed PSK or

QAM signals.
On the receive side, the data can be extracted from the received signal by taking a
Fourier transform such that

(14)

A comparison of the OFDM signals to traditional single carrier modulation is given in


Figure 11. It should be noted that the spectrum allocation and effective data rates are of
particular interest. OFDM waveforms provide the benefit of being well suited for wide­
band systems in environments with frequency selective fading or narrowband interference
since not all subcarriers may be subjected to a fade or interference. Additionally, OFDM
waveforms are tolerant of time synchronization errors, but are sensitive to frequency offsets
and phase noise since they will introduce intersymbol interference between time adjacent
symbols that will degrade performance.

16
Single Carrier Modulation

■8 *

Orthogonal Frequency Division Multiplexing

»»
•*
*•
•*

I *•
»•
•a

•e

WWWW1
Figure 11: Comparison of an OFDM waveform to a traditional single carrier modulated
waveform.

1.6 Related Works

A detailed overview of the wireless communication technology employed by UAVs is


discussed in [36] which covers a wide spectrum of signal processing and communication
topics directly applicable to those found in UAV communications.
An important factor in reliable communication of UAVs, especially swarms of UAVs
operating in close proximity, is the need for proper detection and classification of the users.
Various UAV protocols addressing both security and swarm applications are discussed in
[16,33]. Given in [1] is a toolbox that can be utilized to detect and classify UAVs based on
flowcharts produced in GNU radio3. Drone detection using a short time Fourier transform

(STFT) was shown in [17]. Machine learning has also been utilized to detect and classify
UAVs in [7].
For the case when there is a desire the remove the ability for the drone to be piloted by
means of RF communications, machine learning techniques have been researched for the
3GNU Radio is a signal-processing software package that is distributed under the GNU general public
license (GPL) which allows for the distribution to be free. GNU Radio is analogous to MATLAB’s simulink
blocks, but with a particular focus on signal processing.

17
applicability to automate UAV flight, and autonomous UAV navigation is examined in [21]
using deep neural networks.
UAVs are also susceptible to interference and jamming threats. Reference [18] gives an
overview of SDR (Software Defined Radio) based attacks on UAVs. Also, [26] discusses
the ability for a UAV to be leveraged to jam other UAVs. Recent work has been done in
[29] over viewing a multitude of techniques used for the jamming of UAV communica­
tions. More specifically, in [25] the jamming of spread spectrum communications such as
FHSS was analyzed. Detection for unauthorized UAVs was discussed in [12], The ability
to detect and subsequently hi-jack a UAV was shown in [13], Anti-jamming wireless sensor
network protocols was discussed in [19]. Ultimately, in 2020, a comprehensive overview of
the different UAVs and the common approaches for attacking these jammers is given in [3].

Orthogonal Frequency Division Multiplexing and 802.1in Wi-Fi

A majority of the work in this report in reference to 802.1 in and MIMO systems comes
from [22]. The text references many other related works that have been researched prior to
the publication. Channel models for indoor multipath environments have been discussed
in [4] for SISO (Single-Input Single-Output) systems and [37] for MIMO (Multiple-Input
Multiple-Output) systems. UAV performance was assessed using the 802.11a standards
in [38]. For the deployment of 802.1 In, a task group was assembled in order to charac­
terize the relevant channel models that the chip sets designed to implement the standards
of 802.1 in would be subjected to [6]. This work built upon previous work done by [30]
and [31] includes specific applications to indoor environments with a particular focus on
multipath. From here, the applicability of these channels models and any recommended
changes is given in [34], where the performance of the empirical model is compared to the
real-world data analysis.

Frequency Hopped Spread Spectrum Model Inspirations

In order to model the frequency hopped spread spectrum waveforms used by some

18
UAVs, a definitive characterization is required. It is shown in [2] that some of the UAV
transceivers such as HiTec, Corona, FrSky, Tactic, Futaba S-FHSS, and Wi-Spy’s USB
spectrum analyzer utilize the TI-CC2500 [15] chip set. It was decided to model the FUSS
signal after the capabilities defined in the data sheet provided for the TI-CC2500 transceiver.
Also, the hopping duration of the FUSS system used in this report was derived from the
hopping duration found in [32] where the security of an FUSS signal is discussed based on
the ability to detect and reverse engineer the signal. A MATLAB model using Simulink
was developed. MATLAB models for FUSS systems are given in [9, 23]. Also, an early
mathematical description of FUSS systems is given in [28].
Several approaches for adaptive frequency hopping is discussed in [24] which could be
leveraged to make the UAV more resilient to multipath by avoiding portions of the spectrum
that experience known deep fades and to jamming by avoiding portions of the spectrum that
are occupied by interfering signals.

19
2 Methodology

2.1 FHSS for Telemetry

As shown in Table 1, there are several UAV models that utilize FHSS for remote con­
trol communications. This includes the ability for the controller to transmit information
about the user requested response with respect to the UAV’s movement, flight speed, and
positioning. In response, the UAV receives the control information and transmits back to
the user its respective data such as position, speed, and power levels. In this section, an
overview of the Simulink model constructed to produce representative FHSS signal and
measure the effects of indoor multipath on system performance is given.

Simulink Model

The model was constructed in Simulink in order to leverage preexisting library func­
tionality. The simulation creates, transmits, receives, and analyzes information on a per hop
basis. Given in Figure 12 is the high-level overview of the model. The basic components
are a transmitter, a channel, and a receiver with some performance metrics for analysis.
Transmitter

Channel

Bernoulli
Binary
Convolutional
Encoder FH-CPM [58M0^| r-----
Rician
ï
Modulator
Binary Assemble
Data Source Packets Frequency hopping
al 1 Mbps GFSK Modulator

Receiver

Hop Frequency in MHz

I5324I 0x1] '


I Channel Ix.Symbols^ |8xli 0.029711
ist Môoicn Matrix
Demodulator De interleaver pH} Viterbi Decoder
lop ..Index
339e*05|
Frequency hopping Dis-assemble
FM Demodulator

Figure 12: Simulink Overview of the FHSS Model.

Transmitter

The transmitter block begins with a random stream of binary data at a rate of 2 Mbps.
The data is then encoded using a 1/2 rate code with constraint length 4 as shown in Figure

20
13. The convolutional encoder is implemented in Simulink using the preexisting Convolu­
tional Encoder block.

Input
data i

Figure 13: Finite shift register representation of the 1/2 rate convolutional code used.

Next, the data is interleaved by groups of 16 bits through the use of a 4 x 4 matrix
interleaver. This interleaver works by sequentially reading the data into the matrix row by
row, beginning with the first row and first column. The matrix is then read out starting from

the last row and last column, on a column-by-column basis. Figure 14 shows the general
principle behind how an interleaver works.

Interleaver Interleaver
Write buffer Read buffer
Innn
Packet FEC
Engine Encoder

Interleaver Interleaver
Write buffer Read buffer

FEC Packet
Decoder Engine

Figure 14: High level overview of interleaving and deinterleaving.

Afterwards, the encoded and interleaved data is then packetized into lengths of 5824
bits each. Since the data stream is generated at 2 Mbps, and the code rate is 1/2, then the
effective data rate of the packets is 2* (1/2) = 1 Mbps. Thus, 5284 bits at 1 Mbps equates
to a hop duration of approximately .0058 seconds as desired. These packets are sent to the
modulator block to be prepared for transmission.
Next, the data is first modulated onto a baseband Gaussian Frequency Shift Keying

21
"T_rfw\M
CPM
M 1

Out
GFSK
Modulation

To
nudvwA
50-FSK
Frame

Generate
49 possible carriers
-24MHz to 24MHz

Figure 15: Simulink block diagram of the FHSS modulator.

(GFSK) waveform. The GFSK waveform is constructed by shaping the data with a Gaus­
sian filter with a modulation index of 0.3 and a bandwidth-time product (BT) of 1. The
modulating waveform is constructed from the input hopping frequency and then multiplied
(mixed) with the GFSK data to produce the hopped data waveform. This process is re­
peated for each packet/hop to be transmitted.

Channel

The channel model consists of two blocks. The first block is a single-input single­
output (SISO) fading channel (Rician or Rayleigh) model. The channel is parameterized to
match the properties of the channel used in the simulation of the OFDM waveforms, and
the channel modeling indoor multipath comes from research done in [6]. It should be noted
that the channel varies over time as a result of the maximum Doppler parameter, which a
value of 15 Hz was used for this scenario.
The output of the fading SISO channel is input into a simple additive white Gaus­
sian noise (AWGN) block. A noise vector is constructed and added to the signal with the
appropriate variance as determined by the input signal’s power level and the desired signal-
to-noise ratio (SNR) power level in dB.

Receiver

22
The receiver’s demodulator block is shown in detail in Figure 16. To begin, the hopped
signal is demodulated back to baseband by the reverse process completed in the transmit
side. The hopping frequency is used to construct the conjugate of the original modulating
waveform such that when multiplied with the received waveform brings it back to base­
band. Since the 2-ary GFSK waveforms are a specific form of the more general 2-ary FSK
waveforms, a 2-FSK demodulator was used to demodulate the baseband GMSK received
waveform.

2-FSK

Out
To
50-FSK
Frame

Generate
49 possible carriers
-24MHz to 24MHz

Figure 16: Simulink block diagram of the FHSS demodulator.

Next, the output from the demodulator block is passed to a Viterbi decoder to reverse
the convolutional encoder process and then to the deinterleaver. The output of the dein­
terleaver is the received data stream, which is compared to the transmitted data stream for
analysis.

Simulation Parameters

Working from the information given in [15] and the work done in [32], the relevant
parameters of the signal used in the simulation are populated. The data rate of 500 kBaud
is the highest data rate allowed by the Texas Instruments CC2500 transceiver. This rate is
a result of a bit duration of 2 gs, the data being sampled at rate of 2 samples per symbol,
and the coding rate being 1/2 which results in the effective rate of 500 kBaud/sec, i.e.

2xl06^) 1 syms 1 = 500x 10^


sec / 2 samps 2 sec

Also, the multipath model consists of a single delay tap in addition to the original LOS

23
FHSS Simulation Parameters
Hop Bandwidth 500 kHz
Hop Duration .0058 secs
Frequency Channels/Separation 50 Channels with 1 MHz Spacing
Convolutional Code 1/2 Rate Code, Depth = 4
Data Rate 500 kBaud
Processing Bandwidth 100 MHz
multipath Taps To = 0, attno = 0; Ti = 30 ns, attni = -0.9 dB
Signal to Noise Ratios [6,22] dB
Table 3: Overview of the FHSS simulation parameters utilized.

channel, where the delay occurs 30 ns after the LOS signal with a decrease in the mean
power with respect to the LOS signal of 0.9 dB. The actual amplitude of this delay follows
a Rician distribution with a K value equal to 3. Also, the Doppler used has a maximum
shift of 15 Hz which follows a Jake’s model [31]

l/l < fd (15)


1-#//.%) '

where fa is the maximum Doppler shift and /3 is a parameter set such that S(f) = .1.
More details of this particular channel model and how it was applied can be found in the
Appendices of this report.

2.2 OFDM for Video/Telemetry

From Table 1, there are several other UAV models that utilize only OFDM for all com­
munications. This includes the usage of FHSS for systems that use an OFDM, but addi­
tionally OFDM provides a much higher data throughput which enables the real-time trans­
mission of streaming video feeds. Since some systems use OFDM for all communications,
and other systems use OFDM strictly for video transmission, there is no focus on which
information is being transmitted (RC, telemetry, or video), but only the PER achieved by
the protocol.

MATLAB Model

24
MCS Coding Scheme Encoding Rate Data Rate (Mbps)
8 BPSK 1/2 27.0
9 QPSK 1/2 54.0
10 QPSK 3/4 81.0
11 16-QAM 1/2 108.0
12 16-QAM 3/4 162.0
13 64-QAM 2/3 216.0
14 64-QAM 3/4 243.0
15 64-QAM 5/6 270.0

Table 4: 2 x 2 MIMO 802.1 In MCS modulation and coding schemes. Note that MCS
values of 8 — 15 correspond to the case of two spatial streams (MCS of 1 — 7 is one spatial
stream, etc.).

Preexisting libraries were leveraged to build the model, however Simulink was not used
for the OFDM waveform and instead a scripted simulation was constructed. This offered
the benefit of a larger library of functions, specifically those aimed towards 802.1 in proto­
cols.
The script begins by defining the parameters for the high-throughput (HT) wlan wave­
form. The waveform is constructed for the 40 MHz bandwidth option, a 2 x 2 multiple input
multiple output (MIMO) transmit/receive configuration, a physical layer convergence pro­
cedure (PLCP) service data unit (PSDU) length of 1024, and various LDPC encoding rates
and modulation scheme combinations as shown in Table 4.

Channel

UAVs are not typically designed for use in indoor environments where corridors and
walls produce high indoor multipath to be realized at the receiver. In order to examine
the reliability of the communication link of the UAVs, a representative channel model of
the desired environment is required. Also, standardized models are used that are based on
previous research, since there has been sufficient work done for 802.11 in characterizing
various environments to be realized when deploying 802.11 protocols. In particular, there
was a focus on the 802.1 in channel models examined by the 802.1 In task group known as

25
TGn.
Table 3.7 Channel model 0 (Erceg et al., 2004}

Cluster 1 Cluster 2 Cluster 3


Tap Excess Power AoA AS Rx AoD AS Tx Power AoA AS Rx AoD ASTx Power AoA AS Rx AoD ASTx
index delay [ns] [dB] Fl n n [ 1 [dB] [°] [ 1 [ ] [] [dB] n n n [°]

1 0 0 158.9 27.7 332.1 27.4


2 10 -0.9 158.9 27.7 332.1 27.4
3 20 -1.7 158.9 27.7 332.1 27.4
4 30 -2.6 158.9 27.7 332.1 27.4
5 40 -3.5 158.9 27.7 332.1 27.4
6 50 -4 3 158.9 27.7 332.1 27.4
7 60 -5.2 158.9 27.7 332.1 27.4
8 70 -6.1 158.9 27.7 332.1 27.4
9 80 -6.9 158.9 27.7 332.1 27.4
10 90 -7.8 158.9 27.7 332.1 27.4
11 110 -9.0 158.9 27.7 332.1 27.4 —66 320.2 31.4 49.3 32.1
12 140 -11.1 158.9 27.7 332.1 27.4 -9.5 320.2 31.4 49.3 32.1
13 170 -13 7 158.9 27.7 332.1 27.4 -12.1 320.2 31.4 49.3 32.1
14 200 -16.3 158.9 27.7 332.1 27.4 -14.7 320.2 31.4 49.3 32.1
15 240 -19 3 158.9 27.7 332.1 27.4 -17.4 320.2 31.4 49.3 32.1 -18.8 276.1 37.4 275.9 36.8
16 290 -23.2 158.9 27.7 332.1 27.4 -21.9 320.2 31.4 49.3 32.1 -23.2 276.1 37.4 275.9 36.8
17 340 -25.5 320.2 31.4 49.3 32.1 -25.2 276.1 37.4 275.9 36.8
18 390 -26.7 276.1 37.4 275.9 36.8

Figure 17: Example of the parameters defining the TGn channel model D.

In [6] several channel models are introduced that are used to construct Multiple-Input
Multiple-Output (MIMO) correlation matrices representative of the channel being modeled.
The models begin with predetermined power delay profiles (PDP)s. PDFs are clustered
and extended such that they overlap, and corresponding tap powers are calculated by the
method presented in Appendix A of [6]. A Laplacian distribution [31] is used to model the
assumed power angular spectrum shape of each cluster and tap. The angular spread and
angle of arrival is then assigned to each cluster and corresponding tap. Next, the antenna
configuration is assumed to be uniform-linear and the resulting correlation matrices are
computed. Lastly, Doppler and shadowing effects are applied to the signal. Instead of
modeling the path loss, results were plotted as a function of SNR which takes into account
path loss, transmit power, and receiver processing gain. Parameters of channel model D
are given in Figure 17, and a more detailed overview of each channel model used for this
thesis is given in the appendix.
Shown on figure 18a, is the output of the channel when a single impulse is passed
through. There is an obvious delay and multipath effect occurring as seen by the shifted
pulse’s shape. In Figure 18b, the PSD of the output of the channel when an OFDM signal is
passed through can be seen. The signal is in a noticeable fade around the -8 MHz frequency

26
offset from the center.

-20
Impulse Response Transmitted Signal
Channel Output Channel Output
-25
0.8

-30

1 0.6

I 0/

I 0.2

-50

-0.2 L -55 L
0 5 10 15 20 25 30 35 -20 ■15 20
Sample Index Frequency (MHz)

(a) Impulse response (b) Channel response

Figure 18: Channel response representations of channel model D.

With these defined, the properties of the multipath channel are defined. A TGn channel
model is constructed for the desired delay profile defined by the various alpha-channels (B,
C, and D), the number of transmit and receive antennas, the distance between the trans­
mitter and receiver, the effect of any florescent lighting, and the relative Doppler or en­
vironmental speed. Note that no large-scale fading effects were simulated because the
environment under test is an indoor corridor where the effects of large scale fading are neg­
ligible, if existent at all.

Receiver

The transmitter and channel are well defined by the appropriate parameterization and
calling of preexisting library functions. Once a waveform is constructed and passed through
both a TGn channel and an AWGN channel, it is received. To start, both a coarse estimate
of timing and then frequency are found and corrected for. From here, fine timing and fre­
quency estimates are calculated and corrected for in the received signal. Next, a channel
estimate is computed from the demodulated time-aligned and frequency-corrected received
signal. The channel estimate and received signal are then passed to the data extractor where
the channel estimate is used to correct for any channel effects on the received signal, phase
offset estimation is computed and applied, and the LDPC decoder is applied which makes

27
use of log-likelihood ratios requiring noise estimates acquire from the received signal. The
channel estimate provides a means of attempting to reverse the effects the multipath has on
the received signal.

Simulation Parameters

OFDM Simulation Parameters


Channel Bandwidth 40 MHz
MIMO Configuration 2 Tx Antennas, 2 Rx Antennas, 2 Spatial Streams
Subcarriers and Spacing 128 subcarriers (108 for data) spaced 312.5 kHz apart
Coding Used Low Density Parity Check (LDPC) codes of various rates
802.1 in Channels Used Channel Models B, C, and D
Signal to Noise Ratios [0,40] dB

Table 5: Overview of the OFDM simulation parameters utilized.

It is important to note that the modulation and coding schemes specifically used are not
given in Table 5 since they are subject to change. For example, one of the following four
modulation schemes are employed: BPSK, QPSK, 16-QAM, and 64-QAM. Additionally,
the LDPC code rate applied to each respective modulation scheme varies as given in Table
4. An overview of the channel models used and their respective configurations is given in
Appendix A.2.

28
3 Results

3.1 FHSS Results

Designing of the simulations started by confirming the simulated BER aligns with that
of a theoretical BFSK curve [27]. This is shown in Figure 19 by the theoretical curve in
the dashed line and the overlapping simulated BERs marked with crosses along the solid
line. Also included is the theoretical AWGN curve when the convolutional encoder is used,
shown by the dotted curve. From here, the signal model was subjected to a Rician channel
model as described by the parameters in Table 3. The simulation was again compared to
that of a theoretical BFSK signal in a Rician SISO channel as shown in Figure 19 by the
dashed curve with crosses.

FHSS BER Performance Results


10l
*

10'2

I 10"

I
DO 10'8

-------- - FSK AWGN Theory


10'8
.... *..... FSK AWGN Theory w/ Encoder
-----1----- FSK AWGN Simulation
--%- FSK Rician K = 3 Theory
—x— FSK Rician K = 3 Simulation w/ Encoder
10-1°
0 5 10 15 20 25
Eb/No (dB)

Figure 19: Simulation results comparing the FHSS BER for the case of a strictly AWGN
channel and a high multipath channel.

3.2 OFDM Results

In Figures 20, 21, and 22 the PER is plotted as a function of received SNR. Each figure
consists of three curves: one for the AWGN only channel model shown by the dashed fine
with crosses, one for the TGn channel model D with a LOS between the transmitter and
receiver shown by the solid line with plus symbols, and one for the TGn channel without

29
a LOS shown by the solid line with asterisks. Additionally, any curves that do not appear
to have data points for all SNR simply achieved a PER less than that of the minimum
detectable value for those SNR values.

H—BPSK 1/2 TGn I H--- QPSK 1/2 TGn


BPSK 1/2 TGn —* QPSK 1/2 TGn
-x--BPSK 1/2 AWGN | -x- QPSK1/2AWGN|

SNR (dB) SNR (dB)

(a) BPSK signal with 1/2 rate LDPC code (b) QPSK signal with 1/2 rate LDPC code

Figure 20: Comparison ofPERs for various OFDM modulation and coding schemes. Chan­
nel model D used.

—I--- QPSK 3/4 TGn I ■ 16-QAM 1/2 TGn I


-*— QPSK 3/4 TGn ■16-QAM 1/2 TGn \
- QPSK 3/4 AWGN | ■ 16-QAM 1/2 AWGN |

SNR(dB) SNR (dB)

(a) QPSK signal with 3/4 rate LDPC code (b) 16-QAM signal with 1/2 rate LDPC code

Figure 21 : Comparison of PERs for various OFDM modulation and coding schemes. Chan­
nel model D used.

30
■ 16-QAM 3/4 TGn I
■ 16-QAM 3/4 TGn
• 16-QAM 3/4 AWGN |

-4--- 64-QAM3/4TGn I
-*—64-QAM 3/4 TGn
- x - • 64-QAM 3/4 AWGN |

SNR (dB) SNR (dB)

(a) 16-QAM signal with 3/4 rate LDPC code (b) 64-QAM signal with 3/4 rate LDPC code

Figure 22: Comparison ofPERs for various OFDM modulation and coding schemes. Chan­
nel model D used.

Comparing TGn Channel Models

The previous results focused solely on the TGn channel model D. In addition to model
D, models B and C were also simulated for the case of a non-LOS transmission. Also
included is the measured effective throughput of the simulation. Effective throughput is
computed by measuring the PER of the simulations, equating this PER to a BER by assum­
ing a packet in error equates to errors in all bits of the packet (i.e., the packet is dropped),
and then multiplying by the data transmission rate.

Throuahout for Channel Model-D: NLOS


300
5/6 Rate 64-QAM
3/4 Rate 64-QAM
2/3 Rate 64-QAM
250
mel Model-D: NLOS
3/4 Rate 16-QAM
10° 1/2 Rate 16-QAM
3/4 Rate QPSK
1/2 Rate QPSK
1/2 Rate BPSK
5 Mbps Threshold

100

50

SNR (dB)

0 5 10 15 20 25 30 35 40
(a) PERs results for channel model D SNR(dB)

(b) Effective throughput results for channel model


D

Figure 23: Channel Model D: PERs and Effective Throughput measurements

31
Throughput for Channel Model-B: NLOS
300
—x— 5/6 Rate 64-QAM
----- 1----- 3/4 Rate 64-QAM
250 2/3 Rate 64-QAM
3/4 Rate 16-QAM
Ml"! 1/2 Rate 16-QAM
3/4 Rate QPSK
BPSK1/2TGn Io 200 1/2 Rate QPSK
-A—QPSK 1/2 TGn
-e—QPSK3/4TGn 1/2 Rate BPSK
s— 16-QAM 1/2 TGn 5 Mbps Threshold
-*— 16-QAM 3/4 TGn
64-QAM 2/3 TGn
H--- 64-QAM 3/4 TGn cl 150
—64-QAM 5/6 TGn t
BPSK1/2AWGN
QPSK 1/2 AWGN
-0—QPSK 3/4 AWGN
-B -16-QAM 1/2 AWGN 100
16-QAM 3/4 AWGN
-0—64-QAM 2/3 AWGN
64-QAM 3/4 AWGN
-X - -64-QAM 5/6 AWGN

SNR (dB)

0 5 10 15 20 25 30 35 40
(a) PERs results for channel model B SNR(dB)

(b) Effective throughput results for channel model


B

Figure 24: Channel Model B: PERs and Effective Throughput measurements

Throughput for Channel Model-C: NLOS

—x— 5/6 Rate 64-QAM


----- 1----- 3/4 Rate 64-QAM
250 2/3 Rate 64-QAM
3/4 Rate 16-QAM
1/2 Rate 16-QAM
3/4 Rate QPSK
BPSK 1/2 TGn 1/2 Rate QPSK
— QPSK 1/2 TGn
-0— QPSK 3/4 TGn 1/2 Rate BPSK
— 16-QAM 1/2 TGn 5 Mbps Threshold
-*— 16-QAM 3/4 TGn
- 9—64-QAM 2/3 TGn
- 4--- 64-QAM 3/4 TGn
- M— 64-QAM 5/6 TGn
-0 - BPSK 1/2 AWGN
-A—QPSK 1/2 AWGN
O -QPSK 3/4 AWGN
B -16-QAM 1/2 AWGN 100
-16-QAM 3/4 AWGN
-O--64-QAM 2/3 AWGN
-4— 64-QAM 3/4 AWGN
x 64-QAM 5/6 AWGN
50
15 20 25
SNR (dB)

0 5 10 15 20 25 30 35 40
(a) PERs results for channel model C SNR (dB)

(b) Effective throughput results for channel model


C

Figure 25: Channel Model C: PERs and Effective Throughput measurements

32
4 Discussion

4.1 FHSS Results Discussion

Adding additional taps only reduces performance and is especially noticeable at higher
SNR values, as seen by the solid line with crosses. This reduction in performance is a result
of destructive combining of the signal from the various paths that current signal processing
techniques used in this simulation are unable to account or correct for. Important to note
is that the FHSS simulation, shown in Figure 19, was able to achieve a BER of 1/100 at
approximately 12 dB SNR and 1/1000 at approximately 18 dB.
For example, given the 1 Mbps transmission rate and a 50% BER, the effective through­
put would be an unreliable 500 Kbps, which is 5 times faster than 2G cellular services used
for the transmission of text messages. The minimum measured effective throughput for the
FHSS protocol was 922.6 Kbps, or a BER of 0.0774058, at 6 dB receive SNR. This would
mean that at 6 dB receive SNR the FHSS protocol simulated would be able to operate at a
data rate 9 times greater than that realized by 2G cellular devices used to transmit text. At
the highest SNR simulated of 12 dB, the effective throughput results in 999.56 kbps. Again,
it is important to note that the SNR realized at the receiver is dependent upon the transmit
power, the path loss due to transmit distance, receiver processing gain, and receiver noise
sensitivity.

4.2 OFDM Results Discussion

For the case of BPSK modulation and a 1/2 rate LDPC encoding scheme shown in
Figure 20a, the loss in performance is about 6 dB at 1 /100 packet error rate for the LOS
case versus the AWGN case. The NLOS case is even worse than the LOS case, as expected,
and the loss there is closer to 11 dB. Observing the results in Figure 20b shows a similar
story, but since the modulation scheme requires more energy for the same probability of
error, the 1/100 PER is achieved for higher SNR values. In comparison, the difference is
approximately 7 and 10 dB between the AWGN TGn channel results.
As the code rate is increased from 1/2 to 3/4, the amount of power required for the

33
same probability of error again increases. This can be seen by the 5 dB loss in performance
between Figure 20b and 21a. Also, as the modulation scheme moves from being constant
modulus to one that is not constant modulus, and the amount of power to maintain the same
probability of error is again increased by a minimum distance argument [27]. In Figure 21b
the PER performance achieves 1 /100 at approximately the 21.5 dB SNR level for LOS and
at approximately 25 dB SNR level for NLOS.
Lastly, in Figures 22a and 22b the simulations show that using a 64-QAM modula­
tion scheme requires more transmit power for equivalent performance with constant noise
power realized at the receiver in the high indoor multipath models. This is realized by the
additional 5 dB required for equivalent performance. The main reason for this dramatic
decline is due to the high level of intersymbol interference introduced from the fades for
such a sensitive constellation with respect to the minimum distance argument.
The throughput is denoted as effective because it only measures the successful bits
received per unit of time (seconds). The reason for this is that from [8, 11, 35], a live
stream of resolution 1920x1080 requires a data rate of 3 — 6 Mbps (megabits-per-second)
for Facebook and Google, and 4.5 Mbps for Twitch. Reducing the resolution to 1280x720

would require approximately 1.5 — 2 Mbps less for successful transmission than for the
higher resolution setting. With this in mind, a threshold of 5 Mbps was selected indicating
that successful live video transmission can be communicated with the use of Wi-Fi network
protocols. If the performance is able to achieve above this threshold then a user operator
would be able to reliably communicate and steer the UAV via video. In Figure 23a the
results for all the MCS values used are compared to the same signals in an AWGN channel.
The next channel model examined is channel model B shown in Figure 24a. Important
to note is that the performance, when comparing model B and D, varies in terms of which
modulation and coding combination is superior. In particular, the 1 /2 rate QPSK signal
achieves a PER of 1/100 at approximately 22 dB in channel model D and at approximately
23 dB in channel model B, a difference of about 1 dB. Alternatively, for a 3/4 rate 64-QAM
signal, a PER of 1/100 is never realized in channel model B. This same PER is achieved
for the same 3/4 rate 64-QAM signal in channel model D at approximately 40 dB.
Lastly, channel model C is examined and the results are given in Figure 25a. For the

34
case of the 1/2 rate QPSK signal, a PER of 1/100 is achieved at approximately 24 dB SNR.
As for the higher order 64-QAM signals, they are never able to achieve a PER of 1/100.
For the sake of comparison, examining the 3/4 rate 16-QAM signals in channel model B
and C shows a PER of 1/100 achieved at approximately 37 dB for both.
For all channel models, the effective throughput was able to achieve at least 5 Mbps
for all coding and modulation schemes for a received SNR of greater than or equal to 20
dB. Thus, as long as the link is able to maintain an SNR at the receiver of at least 20 dB,
the protocols are effective enough to ensure successful video communications. If a 20 dB
received SNR is not possible, 5 mbps of throughput can be achieved by using a 1/2 rate
BPSK modulation and coding scheme with a received SNR of 4 dB for channels B and C,
and 5 dB for channel D.

35
5 Conclusions

In this report the BER of a FHSS system, which was developed in MATLAB by means
of a Simulink flow-graph, was analyzed. This FHSS system was developed with the in­
tention of being representative of a UAVs FHSS communication system. Additionally, the
PER of various OFDM waveforms that adhere to 802.1 in Wi-Fi was analyzed. In both
cases the signals were tested in both an AWGN and a multipath channel model.
Beginning with the idea that successful piloting of the UAV when no LOS exists re­
quires reliable video communications. It was found that an effective throughput of 5 Mbps
is sufficient to provide a reliable video communications link. For all 802.1 in signals tested
when subjected to the three channel models utilized, the protocols are able to successfully
achieve 5 Mbps of effective throughput if the SNR realized at the receiver is at least 3 or
4 dB. The SNR realized at the receiver is a function of transmit power, path loss, channel
impairments, receiver signal processing for gain, and receiver noise sensitivity.
This implies that with a received signal power that is at least 7 dB greater than the
receiver’s noise sensitivity, or noise floor, the system is able to maintain a reliable and
efficient video communications link and allow for the pilot to continue to safely control
the UAV. Also, it is reasonable to assume that if the video transmission is successful, the
remaining radio communications will also be successful. This is because the effective
throughput required for successful message transmission is much less than that for video
transmission.
It can be concluded that the protocols examined in this report are sufficient enough to
provide reliable and effective communications between the RC controller and the UAV plat­
form when being subjected to the various environment models discussed. This is provided
that the transmit power, path loss due to distance, receiver processing gain, and receiver
noise sensitivity is able to maintain one of the aforementioned SNR levels.

36
6 Recommendations

One possible avenue for future research and development is to analyze the performance
of a dynamic FHSS system that is able to measure the characteristics of the channel of
operation and adapt to only hop into portions of the allocated spectrum where efficient and
reliable communication can be achieved. Machine learning techniques could potentially be
leveraged to classify the spectrum of interest and make decisions for the transceivers with
respect to spectrum allocation. This could also be leveraged for the OFDM waveforms and
different channels could be occupied by the OFDM waveform of interest.
Another avenue for future research is to attempt to increase the throughput of these
systems with increased hardware by means of increasing the number of transmit antennas,
receive antennas, and spatial transmission streams. This is already being deployed in 5th
and 6th generation Wi-Fi devices such as 802.1 lac. This has the drawback of increased cost

of the additional hardware, whereas dynamic spectrum allocation could be implemented in


software, and UAVs that utilize SDR based platforms would be able to add this capability
at a relatively low cost. Alternatively, the indoor corridor environment of interest could be
fitted with repeater devices that would enable a longer reliable communication channel, but
this again would come at the cost of hardware and proper design implementation.
Lastly, the reliability of these improved designs should be subjected to interference
sources that may or may not be intentionally placed in the spectrum of operation. Also,
since the multipath of the signal at the receiver is analogous to interference observed by
the receiver a consideration should be given to anti-interference techniques as a means

of digitally removing the multipath components, for example by means of a Successive


Interference Cancellation (SIC) algorithm.

37
7 Literature Cited
[1] A. Bello. Radio frequency toolbox for drone detection and classification, Spring 2019.

[2] D. E. Buxton. Rc spread spectrum demystified, https://www.rchelicopterfun.


com/RC-Spread-Spectrum.html. [Online; accessed 15-October-2008].

[3] V. Chamola, P. Kotesh, A. Agarwal, Naren, N. Gupta, and M. Guizani. A comprehen­


sive review of unmanned aerial vehicle attacks and neutralization techniques. Ad Hoc
Networks, 111:102324, 2020.

[4] Y. S. Cho, J. Kim, W. Y. Yang, and C. G. Kang. SISO Channel Models, pages 25-70.
2010.

[5] D. B. Drones. Dji ocusync 2.0: What you need to know about this fpv transmission
system, http://djibestdrones.com/dji-ocusync-2-0/ . [Online; accessed 15-
October-2008],

[6] V. Erceg, L. Schumacher, and P. Kyritsi. Tgn channel models. May 2004.

[7] M. Ezuma, F. Erden, C. K. Anjinappa, O. Ozdemir, and I. Guvenc. Micro-uav de­


tection and classification from rf fingerprints using machine learning techniques. In
2019 IEEE Aerospace Conference, pages 1-13, 2019.

[8] Facebook. What are the video format guidelines for live streaming on facebook?
https://www.facebook.com/help/1534561009906955. [Online; accessed 15-
October-2008],

[9] X. Fan and Z. Tan. Simulink implementation of frequency-hopping communication


system and follower jamming. In 2018 IEEE International Conference on Automa­
tion, Electronics and Electrical Engineering (AUTEEE), pages 192-195, 2018.

[10] G. Foschini and M. Gans. On limits of wireless communications in a fading environ­


ment when using multiple antennas. Wireless Personal Communications, 6:311-335,
1998.

[11] Google. Choose live encoder settings, bitrates, and resolutions, https : //support.
google. com/youtube/answer/2853702. [Online; accessed 15-October-2008].

[12] s. Güvenç, O. Ozdemir, Y. Yapici, H. Mehrpouyan, and D. Matolak. Detection, lo­


calization, and tracking of unauthorized uas and jammers. In 2017IEEE/AIAA 36th
Digital Avionics Systems Conference (DASC), pages 1-10, 2017.

[13] M. Haluza and J. Cechak. Analysis and decoding of radio signals for remote control
of drones. In 2016 New Trends in Signal Processing (NTSP), pages 1-5, 2016.

[14] D. G. Hoffman. Coding theory : the essentials. M. Dekker, New York, 1991.

[15] T. Instruments. Cc2500 low-cost low-power 2.4 ghz rf transceiver, https : //www.
ti. com/lit/ds/symlink/cc2500.pdf. [Online; accessed 15-October-2008].

38
[16] J. Jiang and G. Han. Routing protocols for unmanned aerial vehicles. IEEE Commu­
nications Magazine, 56(l):58-63, 2018.

[17] B. Kaplan, Ibrahim Kahraman, A. Gôrçin, H. A. Çirpan, and A. R. Ekti. Measurement


based fhss-type drone controller detection at 2.4ghz: An stft approach. 2020.

[18] F. Le Roy, C. Roland, D. Le Jeune, and J.-P. Diguet. Risk assessment of sdr-based
attacks with uavs. In 201916th International Symposium on Wireless Communication
Systems (ISWCS), pages 222-226, 2019.

[19] L. Liu, G. Han, S. Chan, and M. Guizani. An snr-assured anti-jamming routing pro­
tocol for reliable communication in industrial wireless sensor networks. IEEE Com­
munications Magazine, 56(2):23-29, 2018.

[20] D. Nojima, Y. Nagao, M. Kurosaki, H. Ochi, A. Ishikawa, S. Fukagawa, and


A. Tahira. Soft decision viterbi decoder for fsk demodulation under fast fading chan­
nel. In International Conference on Communications and Electronics 2010, pages
222-227, 2010.

[21] R. P. Padhy, S. Verma, S. Ahmad, S. K. Choudhury, and P. K. Sa. Deep neural network
for autonomous uav navigation in indoor corridor environments. Procedia Computer
Science, 133:643-650, 2018. International Conference on Robotics and Smart Man­
ufacturing (RoSMa2018).

[22] E. Perahia. Next generation wireless LANs : 802.1In and 802.1lac. Cambridge
University Press, Cambridge, UK, 2013.

[23] U. N. Pintukumar Yadav. Performance analysis of fhss transceiver model in matlab.


International Research Journal of Engineering and Technology(IRJET), 2(2), 2015.

[24] P. Popovski, H. Yomo, and R. Prasad. Strategies for adaptive frequency hopping in
the unlicensed bands. IEEE Wireless Communications, 13(6):60-67, 2006.

[25] K. Parlin. Jamming of spread spectrum communications used in uav remote control
systems, Spring 2017.

[26] K. Parlin, M. M. Alam, and Y Le Moullec. Jamming of uav remote control systems
using software defined radio. In 2018 International Conference on Military Commu­
nications and Information Systems (ICMCIS), pages 1-6, 2018.

[27] J. Proakis. Digital communications. McGraw-Hill, Boston, 2008.

[28] G. H. Raz. A mathematical model and simulation of frequency hopping interferences


to fm systems. Mathematical and Computer Modelling, 11:988-993, 1988.

[29] D. J. Rozenbeek. Evaluation of drone neutralization methods using radio jamming


and spoofing techniques, 2020.

[30] A. Saleh and R. Valenzuela. A statistical model for indoor multipath propagation.
IEEE Journal on Selected Areas in Communications, 5(2): 128—137, 1987.

39
[31] L. Schumacher, K. Pedersen, and P. Mogensen. From antenna spacings to theoretical
capacities - guidelines for simulating mimo systems. In The 13th IEEE International
Symposium on Personal, Indoor and Mobile Radio Communications, volume 2, pages
587-592 vol.2, 2002.

[32] H. Shin, K. Choi, Y. Park, J. Choi, and Y. Kim. Security analysis of fhss-type drone
controller. In H.-w. Kim and D. Choi, editors, Information Security Applications,
pages 240-253. Springer International Publishing, 2016.

[33] P. J. Singh, R. de Silva, and I. Seher. Comparison of communication protocols for


uavs and vanets. In 2016 International Conference on Computing, Communication
and Automation (ICCCA), pages 616-619, 2016.

[34] K. Staniec and M. Kowal. Measurement evaluation of the tgn radio channel models
usefulness in predicting wlan performance. Progress In Electromagnetics Research,
137:311-333,01 2013.

[35] Twitch. Broadcasting guidelines, https://stream.twitch.tv/encoding/. [On­


line; accessed 15-October-2008].

[36] E. Vinogradov, H. Sallouha, S. D. Bast, M. M. Azari, and S. Pollin. Tutorial on UAV:


A blue sky view on wireless communication. CoRR, abs/1901.02306, 2019.

[37] W. Xu and S. A. Zekavat. A high-performance measure for non-line-of-sight identi­


fication in mimo-ofdm-based sensor networks. IEEE Systems Journal, 8(1): 125-130,
2014.

[38] E. Yanmaz, R. Kuschnig, and C. Bettstetter. Channel measurements over 802.1 la-
based uav-to-ground links. In 2011 IEEE GLOBECOM Workshops (GC Wkshps),
pages 1280-1284, 2011.

40
A Appendix

A.l MIMO Overview for 802.1in

This appendix gives some background on a channel model consisting of one transmit
and one receive antenna, also known as a Single-Input Single-Output (SISO) model. Next,
a Multiple-Input Multiple-Output (MIMO) model is introduced with the inclusion of Spa­
tial Division Multiplexing (SDM). The parameters of the 802.1 in model used in this thesis
consisted of two transmit and receive antennas with two spatial streams. This overview is
also given in [22] in chapter 3. The information is repeated here for convenience.

Single-Input Single-Output (SISO) Basics

A simplified communications system can be described by the following,

y = p2-h-x+z (16)

where y is the received signal, p is the average SNR, h is the channel fading coefficient,
x is the transmitted signal with unit expected power, and z is a zero-mean unit variance
Gaussian random variable. The channel coefficient can be described in a probabilistic
manner by the proper modeling of the channel fades by use of, but not limited to, the
previously mentioned Rayleigh and Rician distributions. From [10] the Shannon channel
capacity measured in the bits per second divided by the signal bandwidth, is given as

C = log2(l+p#) (17)

In order to extract the original transmitted data, the received signal is equalized by
applying the inverse of the fading coefficient multiplied by the SNR to the received signal.
This results in a noisy data estimate x as given as

41
x=(p2-h)~1-y (18)

= ■ (p2-h-x+z) (19)
= x+(p2-h)~1 -z (20)

Multiple-Input Multiple-Output (MIMO) Basics

Now, adding another transmit and receive antenna to the system will allow for the
received signal to represented by a system of equations. Given in matrix notation, the
model descriptions is given by

yi zi
(21)
% Z2

Here it is assumed that both transmitters transmit the same spatial data stream. This
means that the data rate would not effectively be increased, but signal processing techniques
that make use of the receive signal diversity can be applied to improve reliable and efficient
data throughput.
Suppose that the number of transmit antennas is M, and the number of receive antennas
is N. This would limit the number of spatial data streams to the number of transmit antennas
M. This system can be described as follows,

yi An A12 • • Aim Xi Zi

72 &21 &22 ’ • A%w *2 Z2


_ P
Vm
+ (22)

yN hm Av2 ■ • h-NM Xm ZN

and can also be written in matrix notation as

42
Yn — • HnxmXm+Z)v (23)
VAf
In the OFDM system, each subcarrier is described by eq. 23. Given in [10], the Shannon
capacity for this system is

( p2
C — log2 Jet + (24)
\ M

where det(S) is the determinant of the matrix 5, and H* is the conjugate transpose of
the matrix H. Applying the same equalizer strategy for the SISO system, the noisy data
estimate is given as

H*
(25)

For the case where M = N, as is in this report, the equation can be simplified to

^= h\ Z (26)
[M /

where it should that the reliance of an existing inverse to the matrix H would render the
solution indeterminate if the matrix H is singular and therefore does not have an existing
inverse. The focus of the analytic model of the system requires proper modeling of the
coefficients h^M in the matrix H, as well as the correct assignment to the scalar factor p
determined by the desired SNR.

43
A.2 802.11 TGn Channel Models

The channel models used to construct the matrix H is given in [6] with inspiration
derived from [31] and [30], among others. There are a total of 6 models corresponding to
different indoor environments. To begin, a clustering approach to the delay taps is applied
based on the work from [30]. In this model, taps are grouped in log-decaying clusters
that may overlap. To compute the power for each tap, the powers of overlapping taps is
summed. Each tap is Rician distributed with a fixed and random component. The fixed
component is determined from a LOS path, and the remaining component is random and
Rayleigh distributed. The taps h can be described as

(27)

where X is a zero-mean and unit variance Gaussian random variable, Q is derived from
the angle of arrival of the LOS path, K is the Rician K-factor, and P is the power of the
tap. The impulse response only takes on a nonzero value for K for the first tap in the model
and all subsequent taps have a K-factor of 0. The LOS and NLOS parameter is determined
by a fixed distance between the transmitter and the receiver. An overview of each model’s
K-factor is given in Table 6.

LOS NLOS
Channel K-factor (tap 1) K-factor (taps > 1) K-factor (tap 1) K-factor (taps > 1)
A 1 0 0 0
B 1 0 0 0
C 1 0 0 0
D 2 0 0 0
E 4 0 0 0
F 4 0 0 0

Table 6: K-factors of each of the 6 models for the LOS and NLOS configurations.

The parameters for channel model B and C are given in Figure 26.
Antenna Correlation
Noting that each tap in eq. 22 can be represented by eq. 27, the matrix H can thus be
expressed as

44
Table 3.6 Channel model C (Erceg et al., 2004)

Cluster 1 Cluster 2
Table 3.5 Channel model B (Erceg et al., 2004)
Tap AS Rx AS Tx AS Rx AS Tx
Cluster 1 Cluster 2 delay [ns] [dB] [°] M [°] [ ] [dB] [°] Fl [°] [°]

Tap AS Rx AoD AS Tx AS Rx AoD AS Tx 1 0 0 290.3 24.6 13.5 24.7


delay [ns] [dB] [°] [°] h [°] [dB] [°] M [°] n 2 10 -2.1 290.3 24.6 13.5 24.7
3 20 -4.3 290.3 24.6 13.5 24.7
1 0 0 4.3 14.4 225.1 14.4
4 30 -6.5 290.3 24.6 13.5 24.7
2 10 -5.4 4.3 14.4 225.1 14.4
40 -8.6 290.3 24.6 13.5
3 20 -10.8 14.4 225.1 14.4 118.4 25.2 106.5 25.4
6 50 -10.8 290.3 24.6 13.5 24.7
4 30 -16.2 14.4 225.1 14.4 —6 3 118.4 252 106.5 25.4
7 60 -13.0 290.3 24.6 13.5 24.7 -5.0 332.3 22.4 56.4 22.5
5 40 -21.7 4.3 14.4 225.1 14.4 -9.4 118.4 25.2 106.5 25.4
8 70 -15.2 290.3 24.6 13.5 24.7 332.3 22.4 56.4 22.5
6 50 -12.5 118.4 106.5 25.4
9 80 -17.3 290.3 24.6 13.5 24.7 -93 332.3 22.4 56.4 22.5
7 60 -15.6 118.4 106.5 25.4
10 90 -19.5 290.3 24.6 13.5 24.7 -11.5 332.3 22.4 56.4 22.5
8 70 -18.7 118.4 25.2 106.5 25.4
11 110 -13.7 332.3 22.4 56.4 22.5
9 80 -21.8 118.4 25.2 106.5 25.4
12 140 -15.8 332.3 22.4 56.4 22.5
13 170 -18.0 332.3 22.4 56.4 22.5
14 200 -20.2 22.4 56.4 22.5
(a) Channel Model B Parameters
(b) Channel Model C Parameters

Figure 26: Tables containing the parameters used in channel models B and C.

~eFii . . . ej01M
Xu X12 ••- Xl M

. . . ej^M
i / 1 *21 X22 ’•* X2M
H = VP + a / K+1

ej9m ej^N2 ... ejÔNM


Xm Xn2 • • • XNM
(28)
where the random elements Xnm are subjected to the correlation present between trans­
mit and receive antennas.
To start, the correlation X is computed by

x= [x] ([W/zy (29)

where X represents another zero-mean and unit variance Gaussian random variable.
Furthermore, the matrices Rrx and Rrx denote the correlation matrices with entries
and pRXjj denoting the complex correlation coefficients between the ith and jth transmit and

receive antennas respectively.


Specifically for the 802.1 in channel models, these complex correlation coefficients are
a byproduct of the power angular spectrum (PAS), which takes into account the angular
spread and angle of arrival/departure for each impulse tap. The PAS distribution used for
802.1 in is given as the truncated Lapacian [31] given by,

45
1 Ne n
—Vz\ 0 — <%l (30)
Ok

where Nc is the number of clusters in the model, Pk is the tap power for cluster k, (%
is the angular spectrum of the tap, and % is the tap angle of incidence. Assuming a linear
antenna array, the correlation coefficients p are computed as

P = Rxx (D) + jRxy (D) (31)

where the correlation functions Rxx(D) and Rxy(D) are given by

It / Ojt!) \

/ cos (
\ A
/ 2^n
sin ) PA5(0) d0
/
\
(32)

sin I sin^ ) PASt^d^ (33)


Æ \ A /

Doppler Model

The Doppler spectrum used in the 802.1 In model is bell shaped and described as

|/| — fmax (34)

where fd is the Doppler spread, fmax is the maximum frequency component of the
Doppler power spectrum, and P is a constant used to set the value of S(ff in the 802.1 in
model, a value of 5(/) = .1 was used which gives the value for p. The value of fd = vq/A
is used to determine the Doppler spread, and this can be configured by setting either the
environmental speed vq or the Doppler spread itself, measured in Hz. The coherence time
of this spectrum is determined by Tc = {y/P/2nfd) • ln(2).4

^Channel model F uses an alternate Doppler spectrum to account for the inclusion of a moving object
within the channel that causes a second Doppler spike in the spectrum. However, since channel model F was
not simulated in this report the definition of channel model F’s respective Doppler spectrum has been omitted.

46
Path Loss

To compute the path loss5 from the transmitter to receiver, a free space (FS) loss, with

a slope of 2, was used up to a breakpoint distance. After this break point distance, the slope
is increased to 3.5. The expression for the losses are given as

L(d) = Lps(d) + SF d < dpp (35)

Ud) = Lps^d) + 351og10 | -—) +SF d > dpp (36)


\dBP J

where d is the distance of separation in meters, dpp is the breakpoint distance in meters.
The free space loss Lps, with the frequency denoted as f, is given by,

LpS(d) = 2Olog^) + 201og10(/) - 147.5 (37)

and the log-normal distribution used to model the shadow fading random variable SF is
given by,

<38)

The parameters of each channel for the break point distances and the shadow fading
standard deviation asF is given in Table 7. It should be noted that for each channel model
if the distance of separation d is less than or equal to the breakpoint distance dgp, the
channel is described as a LOS channel. For the case where the separation distance d is
greater than that of breakpoint distance dpp then the channel is described as NLOS.

5Note that the path loss was omitted from this body of work and instead the simulation computed error
rates as a function of SNR which takes into account the path loss due to transmission distance as well as the
transmit power, receiver processing gain, and the receiver noise sensitivity.

47
Shadow Fading standard deviation
Channel Model Breakpoint Distance dpp LOS NLOS
A 5 3 4
B 5 3 4
C 5 3 5
D 10 3 5
E 20 3 6
F 30 3 6
Table 7 : Breakpoint distances and shadow loss standard deviations for each of the 6 channel
models.

48
Biographical Sketch of Author

A graduate of the University of Massachusetts Lowell in 2014 as a Bachelor of Sci­


ence in Mathematics and in 2017 as a Master of Science in Mathematics with a concen­
tration on Applied and Computational Mathematics, Joshua LaBranche has been working
at MIT Lincoln Laboratory in Lexington, Massachusetts since February 2018 as a Signal
Processing and Communications engineer. The focus of the work is on theoretical and
prototypical communication system development and analysis. His current research inter­
ests include interference mitigation, machine learning applications for communications and
signal processing, advanced communication systems prototype development and analysis,
and advanced 5G modeling and deployment.

49
ProQuest Number: 28867279

INFORMATION TO ALL USERS


The quality and completeness of this reproduction is dependent on the quality
and completeness of the copy made available to ProQuest.

ProQuest.

Distributed by ProQuest LLC ( 2022 ).


Copyright of the Dissertation is held by the Author unless otherwise noted.

This work may be used in accordance with the terms of the Creative Commons license
or other rights statement, as indicated in the copyright statement or in the metadata
associated with this work. Unless otherwise specified in the copyright statement
or the metadata, all rights are reserved by the copyright holder.

This work is protected against unauthorized copying under Title 17,


United States Code and other applicable copyright laws.

Microform Edition where available © ProQuest LLC. No reproduction or digitization


of the Microform Edition is authorized without permission of ProQuest LLC.

ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346 USA

You might also like