OFDM Underwater Acoustic Communication System Implementation

on FPGA

A Thesis Presented
by

Diego Andres Cuji Dutan

to

The Department of Electrical and Computer Engineering

in partial fulfillment of the requirements

for the degree of

Master of Science

in

Electrical and Computer Engineering

Northeastern University
Boston, Massachusetts

June 2019
To my family.

i
Contents

List of Figures iv

List of Tables vi

List of Acronyms vii

Acknowledgments ix

Abstract of the Thesis x

1 Introduction 1
1.1 Motivation and contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 State of the Art 4

2.1 Underwater Acoustic Communication Channel Models . . . . . . . . . . . . . . . 5
2.2 Range, bandwidth, and system requirements . . . . . . . . . . . . . . . . . . . . . 6
2.3 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3.1 Non-Coherent Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3.2 Coherent Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.3 Multi-carrier Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.4 Spatial Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 Underwater Acoustic Channel Modeling 11

3.1 Path loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 Signal-to-Noise ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Multipath . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.5 The Doppler Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Orthogonal Frequency Division Multiplexing and System Design 18

4.1 Orthogonal Frequency Division Multiplexing (OFDM) Data Transmission and Re-
ception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

ii
 4.2 Differentially coherent detection . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2.1 Method based on multiple fast Fourier transform (FFT) demodulation . . . 23
4.3 Doppler compensation for mobile communications . . . . . . . . . . . . . . . . . 24
4.3.1 Beta (β) algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.4 Transmitter design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.5 Receiver design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Underwater Platform Overview 29

5.1 Hardware Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.1.1 Microzed Board . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Software Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3 Hardware Description Language (HDL) Design . . . . . . . . . . . . . . . . . . . 34
5.3.1 Simulink Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.3.2 Block Diagram Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6 Results 41
6.1 Underwater Acoustic (UWA) Modem . . . . . . . . . . . . . . . . . . . . . . . . 41
6.1.1 Hardware components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.1.2 Prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2.1 System parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.2.2 Experimental Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

7 Conclusion 51

Bibliography 53

iii
List of Figures

3.1 Absorption Coefficient. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3.2 Empirical model of the noise power spectral density for UWA channel [40]. . . . . 13
3.3 Frequency-distance dependence part of the Signal-to-Noise Ratio (SNR). The spread-
ing factor is k = 1.5 for the path-loss. The noise Power Spectral Density (PSD) is
calculated for moderate shipping activity (s = 0.5) and no wind (vω = 0 m/s) [36]. 14
3.4 Geometry of the channel. [36]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.5 Channel response functions [36]. . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.6 Motion-induced Doppler effects in time and frequency domains [36]. . . . . . . . . 17

4.1 OFDM transmitter scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.2 OFDM receiver scheme. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.3 Partial Fast Fourier Transform (FFT) demodulation scheme [3]. . . . . . . . . . . . 24
4.4 Motion-induced Doppler shift is not uniform in a wideband system [15]. . . . . . . 25
4.5 Illustration of the structure of a sample Cyclic Prefix (CP)-OFDM frame. . . . . . 27
4.6 Block diagram of the CP-OFDM transmitter. . . . . . . . . . . . . . . . . . . . . . 27
4.7 Block diagram of the CP-OFDM receiver. . . . . . . . . . . . . . . . . . . . . . . 28

5.1 Microzed board [48]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.2 AXI4-Stream Signal Diagram [51]. . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3 High-Level SW Tool Workflow for Zynq PL and PS. . . . . . . . . . . . . . . . . 34
5.4 OFDM digital implementation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.5 OFDM transmitter and receiver models. . . . . . . . . . . . . . . . . . . . . . . . 37
5.6 System Parameters. A Matlab function, that includes the underwater system pa-
rameters, such as number of carriers, number of blocks, sampling frequency, etc,
generates a .mat file called ”ofdm SystemParameters” and this file is further called
inside the Simulink models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.7 OFDM Transmitter and Receiver Platform. . . . . . . . . . . . . . . . . . . . . . 39

6.1 Hydrophone TC4013 frequency response [53]. . . . . . . . . . . . . . . . . . . . . 43

6.2 Hardware block diagrams. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.3 Tank and pool dimensions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.4 Channel profile based on preamble correlation measured in (a) tank and (b) pool. . 46
6.5 Geometry of the experiment in (a) tank and (b) pool. . . . . . . . . . . . . . . . . 47

iv
6.6 The estimated Cumulative Distribution Function (CDF)s of the Mean Squared Error
(MSE) per block and MSE per frame measured in tank (top) and pool (bottom)
setups. The experiments, both in tank and pool, consist of 50 frame transmissions
and each frame is composed of 8 OFDM blocks. The CDFs in (a) and (c) reflect
the recorded MSE on each OFDM block, while the CDFs in (b) and (d) reflect the
average MSE per frame on each receiving point during the experiment. . . . . . . . 49
6.7 Average block error rate (Block Error Rate (BLER)) and bit error rate (Bit Error
Rate (BER)) measured in tank (top) and pool (bottom) setups. The results in (a) and
(c) reflect the average BLER for 50 frame transmissions recorded on each position,
in tank (A, B, C, and D) and in pool (A, B, and C). The plots (b) and (d) indicate the
BER calculated on each OFDM frame transmission. . . . . . . . . . . . . . . . . . 50

v
List of Tables

2.1 Available bandwidth for UWA communication ranges [7]. . . . . . . . . . . . . . . 6

2.2 Underwater communications applications and corresponding required BER [8]. . . 7
2.3 Evolution of non-coherent modulation techniques [8]. . . . . . . . . . . . . . . . . 8
2.4 Evolution of coherent modulation techniques [8]. . . . . . . . . . . . . . . . . . . 8
2.5 Evolution of multi-carrier modulation techniques [8]. . . . . . . . . . . . . . . . . 9

5.1 Software tools and corresponding versions. . . . . . . . . . . . . . . . . . . . . . 34

6.1 Hardware components of UWA prototype. . . . . . . . . . . . . . . . . . . . . . . 41

6.2 Experiment Signal Parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.3 MSE performance for 90% of the OFDM blocks and frames recorded in tank. . . . 48
6.4 MSE performance for 90% of the OFDM blocks and frames recorded in pool. . . . 48

vi
List of Acronyms

ADC Analog-to-Digital Converter.

ARM Advanced RISC Machine.

AXI Advanced eXtensible Interface.

BER Bit Error Rate.

BLER Block Error Rate.

BPSK Binary Phase Shift Keying.

CDF Cumulative Distribution Function.

CP Cyclic Prefix.

DAC Digital-to-Analog Converter.

DFT Discrete Fourier Transform.

DMA Direct Memory Access.

DSP Digital Signal Processing.

FFT Fast Fourier Transform.

FIFO First In First Out.

FPGA Field Programmable Gate Array.

FSK Frequency Shift Keying.

HDL Hardware Description Language.

ICI Inter-carrier Interference.

IDFT Inverse Discrete Fourier Transform.

IFFT Inverse Fast Fourier Transform.

IP Intellectual Property.

vii
ISI Inter-symbol Interference.

MSE Mean Squared Error.

MIMO Multiple-Input Multiple-Output.

OFDM Orthogonal Frequency Division Multiplexing.

PL Programmable Logic.

PS Processing System.

PSD Power Spectral Density.

PSK Phase Shift Keying.

QAM Quadrature Amplitude Modulation.

QPSK Quadrature Phase Shift Keying.

SDK Software Development Kit.

SISO Single Input Single Output.

SNR Signal-to-Noise Ratio.

SoC System on a Chip.

UWA Underwater Acoustic.

viii
Acknowledgments

My sincere gratitude to my advisor Prof. Milica Stojanovic for her support, encouragement,
and patience throughout this thesis. She has been crucial during my Master’s study program to define
the direction of my research. I feel grateful with Prof. Miriam Leeser who has guided me to improve
my research work.
Also, I would like to thank my colleagues in the lab. In special to Amir Tadayon for sharing
his vast knowledge in the field of signal processing. My humble recognition to Emrecan Demirors
and Jiacheng Shi who were always ready to support and help.
My eternal gratitude to my beloved parents and siblings. They have been part of my growth
as a human and as a professional. Mayte has been an important person during this study experience
abroad. I will always remember her with love.

ix
 Abstract of the Thesis

OFDM Underwater Acoustic Communication System Implementation on

FPGA
by
Diego Andres Cuji Dutan
Master of Science in Electrical and Computer Engineering
Northeastern University, June 2019
Dr. Milica Stojanovic, Advisor

This thesis focuses on the design and implementation of a highly flexible underwater acous-
tic (UWA) communications prototype that can be used for rapid testing and research purposes. The
prototype contains software and hardware targeted to the Programmable Logic (PL) and Processing
System (PS) sections of the Xilinx Zynq System on a Chip (SoC) device, respectively. The system
model is developed by exploiting the Simulink software tool, which offers both model simulation
and hardware description language (HDL) code generation. The HDL code is deployed onto the
embedded Field Programmable Gate Array (FPGA) of the Zynq SoC.
We propose a system model based on orthogonal frequency division multiplexing (OFDM)
with cyclic prefix and differentially coherent detection. OFDM is a robust multi-carrier modulation
technique used to combat the frequency-selectivity and multipath effects of the UWA channels. The
innate OFDM virtue of having narrowband signal on each carrier, motivates the straightforward
implementation of differentially coherent detection. This technique avoids estimating the channel
coefficients by taking advantage of the existing coherence between consecutive OFDM carriers, and
thus, the overall system complexity is reduced significantly.
Finally, the prototype is evaluated for short range point-to-point communication links
established in tank and pool setups. We report on the system performance in terms of data detection
mean squared error and bit error rate from a system that uses the 120 − 130 kHz band.

x
Chapter 1

Introduction

There is a growing interest and emerging need for UWA communication platforms which
allow easy flexibility to adapt to different scenarios according to the changing environmental condi-
tions, and high data rate transmission. Moreover, in existing commercial modems, its proprietary
software-hardware feature restricts to make changes on the signal processing algorithms running
on the physical layer. Thus, testing novel modulation techniques or adding new functionality to the
existing ones are not feasible.
In this thesis we develop a highly flexible underwater acoustic communications platform
that can be used for rapid testing and research purposes. The UWA platform contains software and
hardware targeted to the PL and PS sections of the Zynq SoC device, respectively. The system
model provides a unified simulation and rapid prototyping environment by exploiting Simulink tools.
Simulation code developed in Simulink is converted to HDL code that further runs on the PL part of
the Zynq SoC.
The system is designed to use OFDM as a robust multi-carrier modulation technique
to combat the frequency-selectivity of the UWA channel. The appeal of OFDM relies on the
computational efficiency of FFT-based processing and the fact that it easily scales to different system
bandwidths. Similarly, the OFDM advantage of having narrowband signal on each carrier, motivates
an straightforward implementation of differentially coherent detection.
Differentially coherent detection is a low complexity alternative that avoids estimating
the channel coefficients. When there is perfect channel knowledge, differential detection has a
theoretical 3dB penalty in performance compared to coherent detection techniques over Rayleigh
fading channels. However, inaccurate channel estimation can lead to degrade the performance of
systems based on coherent detection, and thus differentially coherent detection reports superior

1
CHAPTER 1. INTRODUCTION

performance.
The developed UWA platform prototype has been tested and evaluated in both indoor
(water tank) and outdoor (sea water pool) environments. We present the system performance results
in terms of data detection mean squared error from a system designed to use the 120-130 kHz band.

1.1 Motivation and contribution

OFDM has been extensively adopted in wireless radio communications and similarly in
UWA communications for several years. There is evidence that all the real-time OFDM modems
have implemented coherent detection techniques [1]. However, we capitalize on the design and
implementation of UWA prototype based on both OFDM and differentially coherent detection.
The allure of differentially coherent detection in combination with OFDM was first in-
troduced in [2], and their performance was demonstrated using synthetic data. Furthermore, an
expanded study developed in [3] reports the system performance applied on recorded data from
the MACE’10 experiment. This technique relies on partial-FFT demodulation to mitigate the Inter-
carrier Interference (ICI) distortions along with differentially coherent detection to maintain the
system complexity at minimum. Moreover, the authors in [4] present a novel Doppler compensation
method for OFDM systems. This approach is built upon the FFT demodulation and differentially
coherent detection. Thus, all these facts motivate us to develop an FPGA-based UWA platform which
incorporates differentially coherent detection as a low-computational technique.
The contribution of this thesis are:

• Design and implementation of an OFDM scheme that integrates differential encoding and
differentially coherent detection in frequency domain.

• Development of the system Simulink model which is easy scalable for different system
requirements.

• Design, implementation, and testing of UWA platform prototype. This platform contains:
1) Microzed board which is built around a Zynq 7000 SoC. The HDL code generated from
Simulink is targeted to the PL section and the software part designed in Xilinx Software
Development Kit (SDK) is targeted to the PS section of the Zynq SoC device. 2) The Digital-to-
Analog Converter (DAC) and Analog-to-Digital Converter (ADC) boards are used to generate
the OFDM signal and sample the received signal, respectively. These boards are directly

2
CHAPTER 1. INTRODUCTION

connected to the Microzed board through a parallel bus. 3) Teledyne 4013 omnidirectional
transducers. 4) Amplifier and pre-amplifier.

1.2 Thesis Organization

This thesis is organized as the following. Chapter 2 summarizes the state-of-the-art

work in UWA communications. In Chapter 3, we consider the UWA channel modeling and the
intrinsic effects on the signal propagation, such as path loss, noise, multipath, and the Doppler effect.
Chapter 4 describes the OFDM principle, the system parameters, OFDM system design, and block
diagrams of the transmitter and receiver based on cyclic prefix (CP). In this chapter, we also propose
the use of differentially encoding and differentially coherent detection. Chapter 5 introduces the
system implementation approach on FPGA, design work-flow using MathWorks Simulink, HDL
code generation, block diagrams built in Vivado Design Suite, and software developed in Xilinx
SDK. Additionally, Chapter 5 discusses the internal architecture of the Zynq SoC device, the process
of targeting the hardware and software parts into the PL and the PS, respectively, and the Advanced
eXtensible Interface (AXI)4-Stream protocol used to communicate the PL with the PS sections. In
Chapter 6, we present and analyze the system performance results from testings in tank and pool.
Finally, we summarize our conclusions and future work in Chapter 7.

3
Chapter 2

State of the Art

Underwater wireless communications is of great interest to the military, industry, and the
scientific community, as it plays an important role in tactical surveillance, pollution control, oil control
and maintenance, offshore explorations, climate change monitoring, and oceanography research [5].
The transmission of information through underwater channels is performed using electromagnetic
waves, optical waves, and acoustic waves. Even tough electromagnetic waves propagate at extra
low frequency ranges (30 Hz to 300 Hz), they require large antennas, high transmitting power, and
are suited for very short distances. Optical waves are mainly affected by scattering, temperature
fluctuations, dispersion and beam steering. Consequently, optical underwater communications are
also limited to short distances due to severe water absorption at optical-frequency bands [6]. However,
acoustic waves can establish relative long range links (in the order of km), and currently is the most
common technique used for underwater communications.
Underwater acoustic signals occupy bandwidths in the order of few kHz and are placed on
center frequencies of similar order. This is a feature of wideband systems. The acoustic signal that
propagates through the underwater channel is mainly affected by absorption, which depends on the
signal frequency and distance, multipath propagation, and the relative low speed of sound (1500 m/s)
that creates severe Doppler distortions. Thus, all these combined effects establishes one of the most
challenging communications mediums [7].
Although, the first underwater acoustic communication systems were based on analog
amplitude modulation, the advent of digital communication techniques and the availability of
Digital Signal Processing (DSP) boards allowed the development of more sophisticated underwater
communication devices. Initially, non-coherent systems, such as Frequency Shift Keying (FSK),
reported high robustness to the severe UWA channel impairments, low bandwidth efficiency, and

4
CHAPTER 2. STATE OF THE ART

range in the order of few kilometers. Then, coherent detection techniques were introduced to increase
the bandwidth efficiency, but the receiver complexity also increased considerably [7].
Research efforts have been made during the last two decades in terms of the communication
range and data throughput. Multi-carrier modulation techniques emerged to provide high data rate
transmission, computational efficiency, and high performance against the frequency-selectivity of
the channel. Also, spatial modulation methods, which were widely used in radio systems, presented
even more improvement on the overall system performance. Nowadays, the investigation is focused
on the development of multi-agent underwater networks [8].
In the next sections, we present a detailed review of the state of the art work on the
characterization of the underwater acoustic channel, system requirements, system design, modulation
techniques, and underwater networks.

2.1 Underwater Acoustic Communication Channel Models

UWA channels are essential part of system design and prospective system deployment. An
appropriate channel modeling based on the physical wave propagation features impacts on the overall
system performance [9]. In a communication system, the transmitter sends information through a
signal that propagates through the medium, where the transmitted signal experiences several effects
due to the channel impairments. The signal distortions are caused by the frequency-dependent path
loss, multipath propagation, and the low speed of sound in water, 1500 m/s. Also, the relative
transmitter-receiver motion introduces extreme Doppler distortions, namely frequency shifting and
spreading [10].
In addition to the deterministic properties of wave propagation on the UWA channel, other
random signal fluctuations occur due to surface waves, internal turbulence, variations in the sound
speed, and other small-scale phenomena. Thus, a complete statistical channel model must encompass
the deterministic physical characterization along with the random channel variations [11].
Even though UWA communications lacks standardization, there are several studies in the
literature regarding channel modeling. A deterministic approach grounded on ray tracing theory is
shown in [12]. Also, several channel modeling studies using stochastic theory were performed by
processing data collected from real experiments. These approaches are based in fitting Ricean fading
and Rayleigh fading to experimental measurements [13].
The statistical characterization of the underwater channel is described by the large-scale
fluctuations, which are caused by random displacements in the order of many wavelengths; and

5
CHAPTER 2. STATE OF THE ART

small-scale effects caused by displacements in the order of few wavelengths. An example of large-
scale analysis is shown in [14]. And a rigorous model that studies not only large-scale fluctuations,
but also small-scale variations is presented in [9]. However, small-scale propagation analysis is still
currently an on-going research topic.

2.2 Range, bandwidth, and system requirements

In terms of range, UWA communication systems are classified in: very long, long, medium,
short, and very short distances [7]. However, there is evidence that the available bandwidth highly
depends on the operational frequency and link distance (Figure 1 in [15]). This fact creates the
necessity of adequate modulation schemes which exploits the available limited bandwidth of the
channel to its maximum. Additionally, the bandwidth-distance dependence has further implications
in underwater network design. Table 2.1 summarizes the classification of UWA communication
systems according to the range and feasible bandwidth.

Type Range [km] Bandwidth [kHz]

very long > 100 <1
long 10 − 100 2−5
medium 1 − 10 ≈ 10
short 0.1 − 1 20 − 50
very short < 0.1 > 100

Table 2.1: Available bandwidth for UWA communication ranges [7].

Regarding the system requirements, the most common applications in underwater channels
are: remote control of objects, telemetry, digital voice communication, and video transmission.
Table 2.2 outlines these applications along with the appropriate transmission data rates and BER [7].
The performance of underwater acoustic communication systems is measured through
the BER and the bandwidth efficiency (bits per seconds per Hertz), which have been subject of
improvement for more than two decades of research work. The researchers have focused on different
modulation techniques, starting from single carrier non-coherent modulation, coherent modulation,
and nowadays, most of the scientific production is related with multi-carrier modulation techniques.

6
CHAPTER 2. STATE OF THE ART

Application Rate [kbit/s] BER

remote control 1 < 10−4
telemetry ∼ 10 10−3 − 10−4
digital voice communication ∼1 10−2
video transmission < 64 10 − 10−4
−3

Table 2.2: Underwater communications applications and corresponding required BER [8].

2.3 System Design

In order to combat against the severe fluctuations of the underwater acoustic channel, the
early communication systems were designed based on non-coherent modulation techniques. Even
though non-coherent systems in the form of FSK rely on signal energy detection and offer high
robustness to channel impairments, they are not bandwidth efficient. An example of non-coherent
system was known as DATS (Digital Acoustic Telemetry System) in the 80s [16].
Coherent systems were introduced in the 90s to increase bandwidth efficiency. Phase Shift
Keying (PSK) and Quadrature Amplitude Modulation (QAM) are examples of coherent techniques,
which increase the rate per bandwidth use. However these methods require a receiver able to track
the time-varying channel and perform coherent detection.
Despite of the noticeable improvement provided by coherent systems, their receiver struc-
ture is subject to use long equalizers to correct the extreme channel impairments. Therefore, research
focused on the development of reduced complexity systems that effectively use the limited bandwidth
channel such as multi-carrier modulation schemes [7].
Apart from single transmitter and single receiver systems, usually referred to Single Input
Single Output (SISO) systems, Multiple-Input Multiple-Output (MIMO) structures were introduced
to increase the bandwidth efficiency and improve the system performance (BER) [17].

2.3.1 Non-Coherent Modulation

Non-coherent systems in the form of FSK require low complexity receivers based on energy
detection. These techniques have worked on rapid phase variation channels in shallow-water in both,
long and medium links. The multipath effect of the UWA channel, which introduces Inter-symbol
Interference (ISI), is compensated by inserting guard times between consecutive symbols in the
transmitted signal. In addition, frequency channels that are separated by more than the coherence
bandwidth Bc can be used at the same time [8]. Table 2.3 recaps the research work observed on

7
CHAPTER 2. STATE OF THE ART

non-coherent systems, particularly the bandwidth efficiency does not exceed 0.47.

Publication Rate Bandwidth Bandwidth Range BER

[kbit/s] [kHz] Efficiency [km]
Catipovic (1984) [18] 1.2 5 0.24 3 (shallow) ∼ 10−2
Freitag (1990) [19] 2.5 20 0.13 3.7 (deep) 10−4
Freitag (1991) [20] 0.6 5 0.12 2.9 (deep) 10−3
Mackelburg (1991) [21] 1.25 10 0.13 2 (deep) N/A
Scussel (1997) [22] 0.6 − 2.4 5 0.47 10 (deep) - N/A
5 (shallow)

Table 2.3: Evolution of non-coherent modulation techniques [8].

2.3.2 Coherent Modulation

As the increase of bandwidth efficiency was required, then phase-coherent modulation

schemes were introduced, namely PSK and QAM. In the early developments, the rapid phase
variations of the UWA channel stopped the utilization of coherent techniques; however, the arrival
of sophisticated phase-tracking algorithms allowed the implementation of coherent receivers and
consequently data rates were considerable boosted. Table 2.4 comprises the evolution of coherent
modulation techniques.

Publication Rate Bandwidth Bandwidth Range [km] BER

[kbit/s] [kHz] Efficiency
Mackelburg (1981) [23] 4.8 8 − 14 0.6 4.8 (deep) 10−6
Olson (1985) [24] 2 2 − 10 1.0 6 (deep) < 10−3
Howe (1992) [25] 1.6 10 − 50 0.16 0.1 (shallow) < 10−3
Suzuki (1992) [26] 16 8 − 20 2.0 6.5 (deep) 10−4
Jones (1997) [27] 20 10 − 50 2.0 1.0 (deep) 10−2

Table 2.4: Evolution of coherent modulation techniques [8].

Phase-coherent systems are classified into two categories: differentially coherent and
purely phase-coherent. An advantage of using differentially encoded PSK (DPSK) with differentially
coherent detection is that it allows simple carrier recovery [7]. DPSK serves as an intermediate
solution between non-coherent and purely coherent systems in terms of bandwidth efficiency [28].

2.3.3 Multi-carrier Modulation

Multi-carrier modulation techniques are used to overcome the frequency-selective effects

of the UWA channel. Multi-carrier techniques in the form of OFDM have been widely used in

8
CHAPTER 2. STATE OF THE ART

radio communication systems and is currently adopted in UWA systems. OFDM offers remarkable
robustness against the frequency-selective channel as it translates the frequency-selective channel
into several narrowband channels, each one experiencing flat fading, and thus, eliminating the need
of long equalizers in the time domain. In addition, the allure of OFDM relies on its computationally
efficient FFT processing and that it easily scales to different bandwidths.
OFDM is also favorable for MIMO processing, adaptive modulation, differentially coherent
detection, and partial-band interference suppression. Major issues of OFDM include: high sensitivity
to frequency offset and time-variation of the channel, sensitivity to non-lineal distortions, and power
efficiency.
Additionally, OFDM incorporates two stages: pre-FFT and post-FFT processing. The
front-end resampling is required when the motion-induced Doppler frequency shifting exceeds the
sub-carrier spacing. Even though initial resampling is performed, some Doppler shifting can still
remain because of the slow velocity of propagation in the acoustic channel. A collection of achievable
data rates for multi-carrier modulation techniques is illustrated in Table 2.5.

Publication Modulation Rate [kbit/s] Bandwidth Range [km] BER

[kHz]
Stojanovic (2006) [29] QPSK 30 24 2.5 (shallow) ∼0
Li (2008) [30] QPSK 1.5 − 25 3 − 50 0.5 (deep) 10−5
Li (2008) [30] 16-QAM 12, 25, 50 12, 25, 50 0.5 (deep) 10−5
J. Huang (2011) [31] QPSK, 5.2, 10.4 9.77 1.0 (shallow) 10−3
16-QAM

Table 2.5: Evolution of multi-carrier modulation techniques [8].

2.3.4 Spatial Modulation

Limited bandwidth and low spectral efficiency are the main shortcomings in UWA commu-
nications. Certainly, coherent modulation schemes demonstrated an exceptional boost in the field as
it enabled to increase the bandwidth efficiency. However, the spatial structure of the UWA channel
can introduce additional bandwidth efficiency gain [32].
Spatial modulation is a technique that controls the spatial distribution of signal energy, such
that the UWA channel supports multiple parallel independent communication channels [32]. The
exploitation of the several parallel channels, best known as diversity gain, leads to an improvement in
the performance of the whole system and thus increasing the data rate transmission [33]. A system

9
CHAPTER 2. STATE OF THE ART

that uses multiple transmitters and multiple receivers is referred to as MIMO system. MIMO systems
have been developed over UWA channels using single-carrier and multi-carrier modulation schemes.

2.4 Networks

Even though the bandwidth-distance dependence in UWA communications is a consider-

able obstacle, it allows the idea of implementing underwater multi-hopping and further networks.
In addition, the transmission power and the low speed of sound are major issues that influence the
design of network protocols. Also, large delays between transactions must be taken in account so
that it can reduce the throughput of the system considerably [15].
Additionally, power efficiency is a desirable characteristic for underwater networks since
underwater instruments are battery-powered. Optimization of energy consumption can be accom-
plished by transmitting at higher rate or reducing the number of re-transmissions. However, in a
random access network, where several users are enabled to continuously transmit information, this
task is limited by the channel latency [34].
In summary, the intrinsic propagation characteristics of UWA channels, power, and data
rate must be taken into consideration in network design.

10
Chapter 3

Underwater Acoustic Channel Modeling

The underwater acoustic channel is mainly described by a path loss which depends on the
transmission distance and signal frequency ([35], [36]). The UWA channel offers a limited bandwidth
that is also conditioned on the transmission distance. For instance, the available bandwidth is reduced
to some kHz when the distance is in the order of few tens of kilometers [37].
The acoustic signal that propagates through the UWA channel has the following limitations:
the time-varying multipath propagation and the low speed of sound (1500 m/s) [11]. Moreover,
the inherently wideband feature of UWA communication systems, in which the bandwidth and the
center frequency are in the same order, can cause severe frequency-selective signal distortions and
motion-induced Doppler spreading and shifting [15]. These effects together are contrasted with the
worst properties of radio channels: poor link quality of a terrestrial system, and high latency of a
satellite system [11].

3.1 Path loss

The path loss of the acoustic channel is characterized by its dependency of the signal
frequency, and it is expressed as

 l k
A(l, f ) = a(f )l−lr (3.1)
lr

where l is the distance between transmitter and receiver with a reference lr , and f is the signal
frequency. The path loss exponent k usually takes values between 1 and 2 for cylindrical and

11
CHAPTER 3. UNDERWATER ACOUSTIC CHANNEL MODELING

spherical spreading, respectively. The absorption coefficient a(f ), measured in dB/km, is evaluated
according to the Thorp’s empirical formula ([38]). Figure 3.1 shows its rapid growth with frequency.
350

300

250

200

150

100

50

0
0 100 200 300 400 500 600 700 800 900 1000

Figure 3.1: Absorption Coefficient.

3.2 Noise

Noise in an acoustic channel is generated from four sources: turbulence Nt (f ), shipping

Ns (f ), waves Nω (f ), and thermal noise Nth (f ) [11]. The empirical approximations of the noise
PSD as function of frequency f [kHz] are defined in ([35], [39]) and measured in [dB re µ Pa per
Hz].

10 log10 Nt (f ) = 17 − 30 log10 (f ) (3.2)

10 log10 Ns (f ) = 40 + 20(s − 0.5) + 26 log10 (f ) − 60 log10 (f + 0.03) (3.3)
√
10 log10 Nω (f ) = 50 + 7.5 vω + 20 log10 (f ) − 40 log10 (f + 0.4) (3.4)
10 log10 Nth (f ) = −15 + 20 log10 (f ) (3.5)

In equation (3.3), the shipping activity s ranges from 0 to 1, for low and high activity
respectively, and the wind speed vω is measured in m/s (equation (3.4)). The total noise PSD N (f )

12
CHAPTER 3. UNDERWATER ACOUSTIC CHANNEL MODELING

is given by

N (f ) = Nt (f ) + Ns (f ) + Nω (f ) + Nth (f ) (3.6)

Figure 3.2 depicts the noise power spectral density for different levels of shipping traffic
and wind speed. The the total noise PSD N (f ) decays at a rate of approximately 18 dB/decade.
110
turbulence
100
low activity(s=0)
medium activity(s=0.5)
90
high activity(s=1)
80 wind at 10m/s
wind at 0m/s
70 thermal
overall
60

50

40

30

20

10

0
100 101 102 103 104 105 106

Figure 3.2: Empirical model of the noise power spectral density for UWA channel [40].

3.3 Signal-to-Noise ratio

The signal-to-noise ratio (SNR) depends not only on the frequency, but also on the trans-
mitting distance. To illustrate this distance-frequency dependence, using the path loss A(l, f ), noise
PSD N (f ), and assuming that the transmitted signal is a tone with frequency f and power P , then
the SNR at the receiver can be evaluated as below.

P
SN R(l, f ) = (3.7)
A(l, f )N (f )∆f

where ∆f is the narrowband received bandwidth around the frequency f .

13
CHAPTER 3. UNDERWATER ACOUSTIC CHANNEL MODELING

Figure 3.3: Frequency-distance dependence part of the SNR. The spreading factor is k = 1.5 for
the path-loss. The noise PSD is calculated for moderate shipping activity (s = 0.5) and no wind
(vω = 0 m/s) [36].

The frequency-distance dependence of the SNR is determined by the factor A(l, f )N (f )

(equation (3.7)). For each transmitting distance there exists a narrow range of optimal frequencies
(bandwidth) B(l) where the received SNR level is maximum [36]. Figure 3.3 depicts plots of
the function 1/A(l, f )N (f ) for different distances. The bandwidth is extremely limited by the
transmitting distance. For instance, at 100km the available bandwidth is approximately 1kHz, while
for shorter distances the bandwidth increases.

3.4 Multipath

In UWA channels, multipath is generated by: sound reflection at the surface, bottom and
any objects, and sound refraction in the water [15]. The geometry of the channel and its reflection
properties define the impulse response of an acoustic channel, including the important propagation
paths and delays. The frequency response of the pth path is denoted by

Γp
Hp (f ) = p (3.8)
A(lp , f )

where Γp is the total reflection coefficient along the p-th propagation path and A(lp , f ) is the

14
CHAPTER 3. UNDERWATER ACOUSTIC CHANNEL MODELING

propagation loss that belongs to this path and acts as a low-pass filter. Moreover, the overall channel
frequency response is

P
X −1
H(f ) = Hp (f )e−j2πf τp (3.9)
p=0

and its corresponding channel impulse response

P
X −1
h(t) = hp (t − τp ) (3.10)
p

With the aim of illustrating the multipath effect as in [36], let us define a system with
transmitter and receiver, both located at 75m depth from the bottom and separated by 3 km. The
spreading factor is k = 1.5 and 3dB loss associated with each bottom reflection.

p5

3 km p1

TX RX
p0

p2

p3 p4

75 m

Bottom

Figure 3.4: Geometry of the channel. [36].

The transmitted signal propagates through different paths p with length lp . Assuming that
the speed of propagation c is constant then the first arrival p0 occurs at t0 = l0 /c, which is the
reference time used to calculate the other relative path delays as τp = lp /c − t0 . Additionally, ideal
surface reflection coefficient amount to -1, while the bottom reflection coefficients depend on the
material type and grazing angle.

15
CHAPTER 3. UNDERWATER ACOUSTIC CHANNEL MODELING

10-4
0.1 5

0.08 4

0.06 3

0.04 2

0.02 1

0 0
0 0.02 0.04 0.06 0.08 -200 -100 0 100 200

Figure 3.5: Channel response functions [36].

Figure 3.5 illustrates the overall transfer function and response magnitudes |H(f )| and
|h(t)|. The total multipath spread Tmp is specified by the longest path delay, and typically ranges on
the order of tens of milliseconds [15].

3.5 The Doppler Effect

The relative motion between the transmitter and receiver contributes to the variability of
the channel impulse response. The Doppler effect causes frequency shifting and frequency spreading
in the acoustic signal. This phenomena is proportional to the ratio a = v/c of the relative velocity
between transmitter and receiver, and the speed of sound. The speed of sound in underwater channels
is slow (around 1500 m/s), thus the motion-induced Doppler distortion in underwater channels is far
more severe than wireless radio channels.
To gain deep understanding of motion-induced Doppler effects, let us suppose that the
transmitter and receiver move relative to each other, and the transmitted signal s(t) = <{g(t)ej2πfc t }
contains a rectangular pulse g(t) which is modulated onto a carrier of frequency fc . Neglecting the
multipath effect and path dispersion of the channel, the single-path received signal is modeled as

 l(t0 ) − vt 
r(t) = s t0 + t − (3.11)
c

16
CHAPTER 3. UNDERWATER ACOUSTIC CHANNEL MODELING

where the relative velocity v is constant and l(t0 ) is the distance traveled by the signal arriving at t0 .
Reordering the variables of equation (3.11) we have that

r(t) = s (1 + a)t − τ = <{g((1 + a)t − τ )ej2πfc [(1+a)t−τ ] } (3.12)

where τ = l(t0 )/c − t0 is the signal delay. Hence, the equivalent baseband received signal is defined
as

v(t) = e−j2πaτ g((1 + a)t − τ )ej2πafc (3.13)

There are two effects on the signal causing distortion. First, the motion-induced Doppler
spreading where the pulse g(t) is scaled in time by (1 + a) and delayed by τ , so that the original
pulse duration T is observed at the receiver as T /(1 + a). Similar to the transmitted bandwidth B
which is observed as B(1 + a). Second, the motion-induced Doppler shifting where a frequency
offset afc is introduced. The effects of frequency shifting and spreading are illustrated in Figure 3.6.

Figure 3.6: Motion-induced Doppler effects in time and frequency domains [36].

17
Chapter 4

Orthogonal Frequency Division

Multiplexing and System Design

OFDM is a multi-carrier modulation technique widely used in wireless radio systems.

It has also been implemented in acoustic communications due to its high bandwidth efficiency,
scalable system design, computational efficiency of FFT-based processing, and its robustness against
frequency-selective channels.
Unlike the single carrier systems, OFDM splits the available signal bandwidth into a
collection of equally-spaced narrowband chunks, which are referred to as inter-carrier frequency
separation. Thus, each narrowband carrier experiences flat fading, and therefore, eliminates the use
of long equalizers. In addition, OFDM is suitable for adaptive modulation and differentially coherent
detection techniques.
OFDM is highly sensitive to frequency offset and time-variation of the channel. Frequency
Doppler shifts are easily introduced in UWA channels due to its wideband nature, where the ratio
between the carrier frequency and the bandwidth is small.

4.1 OFDM Data Transmission and Reception

Let us consider a signal defined in a bandwidth B. The signal is transmitted through a

channel with coherence bandwidth Bc < B, hence the signal experiences frequency-selective fading.
The OFDM technique splits the bandwidth B into a collection of K narrowband sub-channels, each
one defined over ∆f = B/K. The value of K is selected such that ∆f << Bc , this ensures that
each sub-channel only faces flat fading.

18
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

Symbol d0 ×
Mapping

Serial-to- P
Data u(t) s(t)
Parallel ej2π0∆f t × × <{·}
source (S-P)

Symbol dK−1 g(t)

× ej2πf0 t
Mapping

ej2π(K−1)∆f t

Figure 4.1: OFDM transmitter scheme.

Figure 4.1 illustrates the OFDM scheme. The incoming stream of data is allocated in K
segments and converted into a K parallel stream of symbols. Each symbol dk belongs to a pre-defined
alphabet, e.g. PSK, and is modulated by a carrier frequency k∆f and defined over a bandwidth ∆f .
The equivalent baseband signal u(t) results from the sum of the K modulated segments and the
multiplication by a rectangular pulse g(t). The passband signal is the consequence of taking the real
part of the up-shifted version of u(t) by the initial carrier frequency f0 ,
( K−1 )
X
j2πfk t
s(t) = < dk e g(t) (4.1)
k=0

( K−1 )
X
j2πk∆f t j2πf0 t
s(t) = < dk e g(t)e (4.2)
k=0

n o
s(t) = < u(t)ej2πf0 t (4.3)

K−1
X
u(t) = dk ej2πk∆f t g(t) (4.4)
k=0

where g(t) is a rectangular pulse with duration T = 1/∆f and defined in T 0 = T + Tg , where Tg is
the guard interval and must be greater than the multipath delay spread of the channel T g > Tmp . The
signal s(t) passes through a channel h(t, τ ) and Gaussian noise n(t) is added, the noise variance is
σ 2 /2. The impulse response of the channel is defined as in [41], where τ is the delay, t is the channel
observation time and P the number of propagation paths. Additionally, hp (t) indicates the p-th path

19
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

real-valued gain and τp (t) is the propagation delay per path.

P
X −1
h(t, τ ) = hp (t)δ(t − τp (t)) (4.5)
p=0

The received passband r(t) and baseband v(t) signals (see Figure 4.2 ) are given as

P
X −1
r(t) = hp (t)s(t − τp (t)) + n(t) (4.6)
p=0
P
X −1 n o
= hp (t)< u(t − τp (t))ej2πf0 (t−τp (t)) + n(t)
p=0
(" P −1 # )
X
−j2πf0 τp (t) j2πf0 t
=< hp (t)u(t − τp (t))e + ω(t) e
p=0
n o
j2πf0 t
= < v(t)e

where v(t) is inferred as the complex baseband equivalent received signal and ω(t) is zero-mean
additive Gaussian noise with variance σ 2 .

P
X −1
v(t) = hp (t)u(t − τp (t))e−j2πf0 τp (t) + ω(t) (4.7)
p=0
P −1
" K−1 #
X X
= hp (t) dk e j2πk∆f (t−τp (t))
g(t − τp (t)) e−j2πf0 τp (t) + ω(t)
p=0 k=0
K−1
" P −1 #
X X
= dk hp (t)e−j2πfk τp (t) ej2πk∆f t g(t − τp (t)) + ω(t)
k=0 p=0
K−1
X
= dk Hk (t)ej2πk∆f t g(t − τp (t)) + ω(t)
k=0

The channel transfer function at frequency fk is defined as

X
Hk (t) = hp (t)e−j2πfk τp (t) (4.8)
p

20
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

R y0
× T

s(t) Channel r(t) v(t) e−j2π0∆f t

+ ×
h(t, τ )

n(t) R yK−1
e−j2πf0 t × T

e−j2π(K−1)∆f t

Figure 4.2: OFDM receiver scheme.

Z T0
1
(a)
yk = v(t)e−j2πk∆f t g(t)dt (4.9)
T 0
Z " K−1 #
1
(b)
T X
= dl Hl (t)ej2πl∆f t + ω(t) e−j2πk∆f t dt
T 0 l=0
K−1
" Z #
(c) X 1 T
j2π(l−k)∆f t
= dl Hl e dt + zk
T 0
l=0
K−1
X
(d)
= dl Hl δkl + zk
l=0
(e)
= dk Hk + zk

In equation (4.9), the equality (a) comes from the fact that g(t) is defined over the interval
t ∈ [0, T ]. Moreover, we assume that the channel is time-invariant over the interval of an OFDM
block, Hl (t) = Hl for t ∈ [0, T ] as stated in (b). The orthogonality property between sub-carriers
is expressed in (c), so the integral results in a delta-function dkl which is one when k = l and
zero otherwise (d). The last expression (e) describes the effect of a flat channel Hk on each carrier
frequency fk , meaning that OFDM decomposes the wideband channel into a set of narrowband
orthogonal sub-channels.
Finally, the multi-carrier modulation and demodulation schemes, illustrated on Figure 4.1
and Figure 4.2, are replaced by the Inverse Discrete Fourier Transform (IDFT) and Discrete Fourier
Transform (DFT) respectively. The IDFT and DFT are implemented through the Inverse Fast Fourier

21
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

Transform (IFFT) and FFT algorithms.

4.2 Differentially coherent detection

Channel estimation is an indispensable part of a coherent OFDM system, and its accuracy
greatly influences the overall system performance. For a time-invariant channel, estimates can be
obtained in the initial OFDM block, which are then applied to detect the data throughout a frame
of OFDM blocks. While for a time-varying channel, channel estimates have to be computed and
updated in every OFDM block. The methods for estimating the channel coefficients are frequently
based on pilot sub-carriers or decision-feedback, and the algorithms can range from the traditional
least squares to more complex ones, such as basis and matching pursuit which exploit the sparsity of
the acoustic channel [42]. However, the long delay spread, time-varying channels, and the wideband
nature of the UWA signals are factors that contribute to an imperfect channel estimation in coherent
systems. And thus, this motivates the use of differentially coherent detection as a low complexity
method which eliminates the need for channel estimation [3].
Differentially coherent detection is based on the assumption that the channel response does
not change significantly either between adjacent carriers or between blocks. The first one implies
differential encoding and detection in frequency, while the latter involves differential encoding
and detection in time ([2],[43]). We consider encoding/detection in frequency domain which has
the following advantages: 1) reinforces the carrier coherence assumption, and 2) increases the
transmission rate [44]. On the transmitter side, the data symbols dk are obtained by differential
encoding the original i.i.d. symbols bk , i.e. dk = dk−1 bk , where bk and dk belong to the same
PSK alphabet, with d0 = 1. While on the receiver, the estimates of the differentially-encoded data
symbols bk are obtained as

∗ y
yk−1 k
b̂k = , k = 0, . . . , K − 1 (4.10)
|yk−1 |2

In equation (4.10), we can observe two facts: 1) consecutive channel coefficients can be
approximated as Hk ≈ Hk−1 , and 2) the received signal yk−1 acts as the channel estimate, i.e.
∗ /|y 2
yk−1 k−1 | is the normalized channel coefficient. To better understand the concept of differentially

coherent detection, consider the received signals on consecutive carriers yk−1 = dk−1 Hk−1 and
yk = dk Hk where zk−1 = zk = 0. Hence, the right side of the equality (4.10) becomes,

22
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

∗ y
yk−1 ∗ ∗
k (a) dk−1 Hk−1 dk Hk
= (4.11)
|yk−1 |2 |dk−1 Hk−1 |2
∗ ∗
(b) dk−1 Hk−1 dk−1 Hk−1 bk
= ∗ ∗ d
dk−1 Hk−1 k−1 Hk−1
(c)
= bk

where equality (a) is obtained by replacing yk and yk−1 . The assumptions Hk = Hk−1 and
dk = dk−1 bk are observed in equality (b). Finally, equality (c) capitalizes the concept of differentially-
coherent detection.
In addition, differentially coherent detection is closely associated with the efficient use
of system bandwidth, which is given by the ratio of the symbol rate to bandwidth, R/B ∼ 1/(1 +
Tg B/K), where Tg is the guard interval. As K increases for a given bandwidth B, both the
bandwidth efficiency and the coherence between adjacent carrier increase, this causes the carrier
spacing ∆f = B/K to become smaller. Thus, favoring the use of differential encoding in frequency
[2]. However, K is limited by the time-coherence of an OFDM system. Since the block duration
T = 1/∆f increases along with K, so does the motion-induced Doppler as well as the channel
response fluctuates within one block. In consequence, the orthogonality between carriers is lost and
ICI must be compensated.
In the literature, two methods have been introduced to overcome the motion-induced
Doppler and the ICI in UWA OFDM systems. The authors in [3] propose four methods for ICI
mitigation. This study suggests the implementation of differentially-coherent detection to reduce the
overall system complexity. While in [4], a method for compensating the motion-induced Doppler
is establish by applying differentially-coherent detection. We briefly overview these methods in
Section 4.2.1 and Section 4.3.1, respectively.

4.2.1 Method based on multiple fast Fourier transform (FFT) demodulation

This study assumes the notion that the channel variations may be decomposed based on a
set of pre-defined functions. The received signal is projected onto these functions, and the projections
are passed on to FFT demodulation and subsequent to a combiner block along with differentially
coherent detection. The equivalent baseband received signal is expressed as in (4.7), where the
channel coefficients Hk (t) permit a decomposition onto a set of known functions Φi (t), such that

23
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

PI−1 PI−1
Hk (t) ≈ i=0 Hk,i Φi (t). If this decomposition can be performed then vk ≈ i=0 Hk,i yk,i , where
yk,i is obtained by taking the FFT of the product Φ∗i (t)v(t) as

Z T
yk,i = Φ∗i (t)v(t)e−j2πk∆f t dt (4.12)
0

T = K/B

T/I

FFT FFT FFT

y k1 y k2 y kI

Combiner
xk

Differentially Coherent
Detection

bk

Figure 4.3: Partial FFT demodulation scheme [3].

Figure 4.3 illustrates a total of I FFT operations to be performed. Since the channel
Hk (t) fluctuates slowly, then I can be small. The outputs yk,i are fed into a combiner and further
differentially decoded.

4.3 Doppler compensation for mobile communications

The main obstacle in OFDM systems applied to UWA acoustic channels are the severe
Doppler distortions induced by the relative motion between the transmitter and receiver. The
wideband nature of UWA communication systems causes non-uniform shifting on each frequency fk ,
so that the shifted amount is not equal for all sub-carriers. Figure 4.4 illustrates the Doppler effect in
a wideband acoustic system creating non-uniform frequency shifting [15].
The motion-induced Doppler shifting can be mitigated by resampling the received signal by
a factor a. The Doppler scaling factor is estimated as a = Trx /Ttx − 1, where Ttx is the transmitted
frame duration and Trx is the estimated received frame duration [15].

24
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

B= K ∆f

∆f

f0 f1 f2 fk f K −1 f

f
f k (1 + a )

Figure 4.4: Motion-induced Doppler shift is not uniform in a wideband system [15].

The resampling process of the signal at a constant rate a only removes the Doppler
distortions created by sources moving at a constant speed. However, the Doppler distortions are
time-varying due to the random motion of the transmitter-receiver and channel time variations. Thus,
initial resampling performs coarse Doppler compensation of the entire received signal (frame of
OFDM blocks) and leaves a residual Doppler factor that must be corrected on each OFDM block for
best ICI reduction ([3],[45]).
The received signal is bandpass filtered, synchronized, resampled, and downshifted by the
lowest carrier frequency f0 . The passband resampled signal is obtained as r(t) = r̃(t/(1 + a)) and
the equivalent received baseband signal is expressed as

K−1
X
jβt
v(t) = e dk Hk (t)ej2πk∆f t + ω(t) (4.13)
k=0

where β is the unknown residual frequency offset. In summary, two steps must be taken for
compensating Doppler distortions: 1) front-end resampling of the entire frame of OFDM blocks and
2) high-resolution Doppler compensation on each block to mitigate the ICI.

25
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

4.3.1 Beta (β) algorithm

A low complexity method based on hypothesis testing and differentially coherent detection
[4]. Several hypothesized frequency offset values are used in steps of a fraction of ∆f , the length of
the fraction depends on the accuracy of the final estimated β̂ and the computational complexity. The
hypothesized value β̂ is used to compensate and demodulate the signal as

Z
yk = v(t)e−j β̂t e−j2πk∆f t dt (4.14)
T

where k = 0, . . . , K − 1, and the values yk are used to find the estimates of the differentially-encoded
data symbols by using equation (4.10).
By using equally-spaced pilot data symbols bk , k ∈ κp , the composite error is defined as

X
E(β̂) = |bk − b̂k |2 (4.15)
k∈κp

and the estimate of β̂ ∗ is obtained as

β̂ ∗ = arg min E(β̂) (4.16)

β̂

4.4 Transmitter design

The OFDM transmitted signal consists of K carriers occupying a bandwidth B = K∆f ,

which is expressed as

 K−1
X 
j2πfk t
s(t) = Re dk e , t ∈ [0, T + Tg ] (4.17)
k=0

where fk = f0 +k∆f is the k-th sub-carrier frequency, f0 is the first carrier frequency, and the carrier
spacing ∆f is assumed to be small enough so that Hp (f ) ≈ Hp (fk ) for f ∈ [fk −∆f /2, fk +∆f /2].
The OFDM block duration T = 1/∆f must be greater than the multipath spread of the channel Tmp ;
and if an efficient use of system bandwidth is required then ∆f should be on the order of 100 Hz or
less [15].
This system assumes the use of CP to preserve the orthogonality of the carriers and to
eliminate inter-block interference between consecutive OFDM blocks [46]. The cyclic prefix duration

26
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

Tg , also referred to as guard interval, must be chosen such that its value is longer than the multipath
spread of the channel, Tg ≥ Tmp .
An illustration of the structure of the CP-OFDM frame is depicted in Figure 4.5. Each
frame includes a preamble, Nb OFDM blocks, and a postamble. The synchronization preamble
and postamble are short signals formed from a pseudo-noise m-sequence with m = 5 mapped to a
unit-amplitude Binary Phase Shift Keying (BPSK) alphabet, i.e., {−1, +1}. The carrier frequency
fc of the preamble-postamble is fc = f0 + B/2 where f0 is the first carrier frequency of the OFDM
signal and B is the acoustic bandwidth. The pause interval between the preamble and the OFDM
signal is Tpse .

Preamble Tg >> Tmp Postamble

Tpse Tpse
CP #1 CP #2 CP #Nb
t
T = K/B

Figure 4.5: Illustration of the structure of a sample CP-OFDM frame.

A complete block diagram of the transmitter is showed in Figure 4.6, where the codeword
bits c are mapped into PSK data symbols bk . The data symbols dk are obtained by differential
encoding the original i.i.d. symbols bk , i.e. dk = dk−1 bk , where bk and dk belong to the same PSK
alphabet, with d0 = 1. Taking IFFT on the differentially-encoded data symbols and inserting CP
yields the baseband signal u(t). Finally, u(t) is upshifted by the first carrier frequency f0 .

I
c Symbol bk Differential dk F Cyclic s(t)
× <{·}
Mapping Encoding F Prefix
T
ej2πf0 t

Figure 4.6: Block diagram of the CP-OFDM transmitter.

27
CHAPTER 4. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING AND SYSTEM DESIGN

r̃(t) Lowpass Frame Doppler scale Bandpass

× ×
filtering synchronization coarse estimation resampling

e−j2πfc t ej2πfc t

Cyclic prefix
F yk Differential
Frequency v(t) b̂k
× F
offset correction removing Decoding
T

e−j2πfo t

Figure 4.7: Block diagram of the CP-OFDM receiver.

4.5 Receiver design

The passband received signal, modeled as in equation (4.6), is downshifted to baseband

by demodulating with a carrier at frequency fc = f0 + B/2. Then the signal is filtered using a
lowpass filter (LPF) with cutoff frequency fcutof f = B/2. At this point, the signal is correlated
with the transmitted preamble to find the beginning of the OFDM frame. After synchronization,
the signal is up-shifted again so that it lives in the frequency band [0, B]. The cyclic prefix (CP) is
removed, as in equation (4.7), and using the fact that carriers are orthogonal over t ∈ [0, T ] then the
demodulation is performed through the FFT to obtain yk = dk Hk + zk k = 0, . . . , K − 1, where
Hk is the channel frequency response and zk contains the equivalent complex baseband noise. The
differentially coherent detection is applied on yk to obtain estimates of the data symbols b̂k .

28
Chapter 5

Underwater Platform Overview

In this chapter we introduce the hardware development of underwater communication

devices using the FPGA technology. This device implements high performance signal processing in
hardware level. We also outline the hardware, HDL and software design of this platform.
The main part of this development is the Microzed board, which includes the Xilinx Zynq-
7000 All Programmable SoC (Zynq SoC). The Zynq SoC is made of two components, the PS and
the PL parts. The PS is formed around a dual-core Advanced RISC Machine (ARM) processor and
can support software routines and/or operating systems, while the PL section comprises an FPGA,
ideal to implement high-speed logic. The PL-PS communication is established by using industrial
standard AXI connections [47].
A complete system platform is illustrated from its design in MathWorks Simulink, HDL
code generation, Xilinx Vivado Design Suite system integration, and application development in
Xilinx SDK.

5.1 Hardware Overview

5.1.1 Microzed Board

Microzed board (Figure 5.1) is a low-cost development kit that enables hardware and
software developers to create and evaluate Zynq-7000 All Programmable SoC designs [48]. The
most important features that Microzed support are:

• Processor: Zynq-7000 All Programmable SoC XC7Z020-CLG400-1

• Memory: 1 GB DDR3, 128 Mb Quad-SPI Flash, 4GB mcroSD card

29
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

• Communication: 10/100/1000 Ethernet, USB 2.0 and USB-UART

• General purpose I/O and PMOD capability

• Clocking: 33.33 MHz clock source for PS

• Xilinx Platform Cable JTAG connector

Figure 5.1: Microzed board [48].

The Zynq 7020 SoC includes a dual ARM Cortex-A9 CPU based PS and Xilinx hardware
PL. The ARM cores are the Cortex A9 MPcore which are able to run up to 1 GHz. The PS includes
L2 cache and enables its embedded computing capability using DDR2 and DDR3 SDRAM memory,
Flash memory, Gigabit Ethernet, general purpose I/O, and UART technologies. Also, the Xilinx AXI
Interface provides high-speed memory-access between the PS and the PL [47].

5.1.1.1 Processing System (PS)

The processing system of the Zynq SoC includes a dual core embedded ARM Cortex-A9
processor operating at up to 1 GHz. It can support memory controllers and I/O peripherals.
The processing system core acts as a logical connection between the PS and the PL and
allows to integrate custom and embedded Intellectual Property (IP) cores with the processing system
unit inside Vivado Design Suite [49].

30
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

5.1.1.2 Programmable Logic (PL)

The PL section is composed of general purpose FPGA logic fabric, slices and Configurable
Logic Blocks (CLB), and Input/Output Blocks (IOBs) for interfacing. Some of these terms are
explained below [47].

• Configurable Logic Block(CLB) are small groupings of logic elements that are laid out in a
two-dimensional array on the PL, and connected to other similar resources via programmable
interconnects. Each CLB is positioned next to a switch matrix and contains two logic slices.

• Slice a sub-unit within the CLB, which contains resources for implementing combinatorial
and sequential logic circuits. Zynq slices are composed of 4 Lookup Tables, 8 Flip-Flops, and
other logic.

• Lookup Table (LUT) is a flexible resource that can implement a logic function of up to six
inputs, Read Only Memory (ROM), Random Access Memory (RAM), or a shift register.

• Flip-Flop (FF) a 1-bit register with reset functionality.

• Switch Matrix each one sits next to one CLB and provides a flexible routing facility for
making connections.

• Input/Output Blocks (IOBs) are resources that provide interfacing between the PL logic
resources, and external circuitry. Each IOB can handle a 1-bit input or output signal.

There are also two special purpose components at the PL level: Block RAMs and DSP48E1
slices for high-speed arithmetic.

• Block RAM (BRAM) is a configurable memory module used to store data and implements
Random Access Memory (RAM), Read Only Memory (ROM), and First In First Out (FIFO)
buffers. Each BRAM can store up to 36 Kb of information.

• DSP48E1 slices are dedicated hardware implementing high-speed arithmetic on signals with
medium to long arithmetic word lengths. Typically, they comprise a pre-adder/subtractor,
multiplier, and post-adder/subtractor with logic unit.

31
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

5.1.1.3 Interfacing the Processing System with the Programmable Logic

The ability to connect the PL with the PS is an appealing feature that Zynq SoC implements
to form a completely integrated system. AXI interconnects and interfaces are used to connect the PS
and the PL, and there are three different types of AXI communication, which are presented below:

• AXI4 is suited for memory-mapped links, provides a data burst transfer of up to 256 data
words.

• AXI4-Lite supports only one data transfer per connection, it is also a memory-mapped
communication.

• AXI4-Stream is designed for high-speed streaming data, supports burst transfers of size in
the order of few Mega-bytes and is not a memory-mapped based communication.

For this thesis the AXI4-Stream communication protocol is used, which is described next.

5.1.1.4 AXI4-Stream Interface

In general the AXI protocol is implemented to communicate the Programmable Logic

with the Processing System, depending on the application the designer can use either AXI4-Lite
or AXI4-Stream as described before. AXI4-Stream best fits when the target application requires
high-data throughput. More often a bus of 32-bit width is used, which implies 4 byte per clock cycle
in hardware. For instance, if the PL is running a clock of 100 MHz, the data is transferred at a rate of
400 Mega-bytes per second [50].
The AXI4-Stream protocol consists of a bus of signals specifically designed such that
the two interfaces, usually known as master and slave, can understand each other and make a data
transfer. These signals include: TVALID, TDATA, TREADY, TLAST and TUSER.
Figure 5.2 illustrates two packet transfers using the AXI4-Stream protocol (reference
Xilinx AXI reference guide). The first transaction consists of four data cycles D0 to D3. The last
data clock is marked by TLAST asserted signal. It is clear that during the transaction both TVALID
and TREADY are always high. Similarly, a second transaction of n cycles is shown. The signals
TVALID and TREADY are used for flow control. The transmitter (master) controls how much data
is transferred by asserting the TVALID signal and the receiver (slave) de-asserts TREADY signal
when it is not ready to receive more data.

32
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

CLK

TVALID

TREADY

TLAST

TDATA D0 D1 D2 D4 D0 D1 Dn-1 Dn

First Transfer Second Transfer

Figure 5.2: AXI4-Stream Signal Diagram [51].

5.2 Software Tools

The process of targeting custom designs on the Microzed board is carried out by using
commercially available tools from MathWorks and Xilinx. MathWorks Simulink is used to create
and simulate the models; it also provides the HDL Coder tool which allow us to target the PL section
of the Zynq [52]. Through the HDL Coder we are able to generate IP cores from the Simulink
models [47]. The IP core generated is integrated along with other components such as the Processing
System unit, AXI Direct Memory Access (DMA)s, and AXI Interconnects inside Xilinx Vivado
Design Suite software. The Xilinx Vivado is a tool that allows designers to define the PL section in a
complete block diagram, where each element is interconnected with others using wires in a graphical
fashion. Xilinx Vivado performs two operations, synthesis and implementation, to finally generate
the bitstream file that is further programmed on the target board.
Additionally, the PS part is performed by Xilinx SDK, that makes use of the generated
hardware file (bitstream). Xilinx SDK is based on Eclipse and provides a friendly graphical user
interface to develop C/C++ code that can run on the dual-core ARM processor of the Zynq.
Figure 5.3 depicts the workflow used to develop this project. There are two paths clearly
defined; the first one targets the PL section, while the second path targets the PS part of the Zynq
device. Following the PL targeting procedure, the Simulink models are compiled and converted to
HDL code, namely Verilog or VHDL code. The IP cores are created in Vivado Design Suite using
the previously generated HDL code, and the final step encompasses the creation of a complete block

33
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

Software Tool Version

Matlab Simulink R2017b
HDL Coder Compatible with R2013b to R2019a
Vivado Design Suite 2017.4
Xilinx SDK 2017.4

Table 5.1: Software tools and corresponding versions.

design in Vivado Design Suite which further generates the bitstream used to target the PL.
Furthermore, the PS targeting process is carried out by using Xilinx SDK where the
application is developed using C language. The software tool versions are described in Table 5.1.
Host PC running SW tools

MathWorks Simulink™
Model

HDL Coder™

HDL Code

IP Core Application

Xilinx Vivado® Xilinx SDK®

FPGA Bitstream ARM Executable

PL PS

Zynq® SoC

Figure 5.3: High-Level SW Tool Workflow for Zynq PL and PS.

5.3 HDL Design

Let us consider an OFDM baseband signal defined by equation (4.4) and bandwidth B,
which is sampled at frequency fs > 2(f0 + B), as illustrated in Figure 5.4; f0 is the first carrier
frequency of the passband signal and fs = 1/Ts .

34
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

Nyquist Limit
...
B f0 fc f0 + B fs f

Figure 5.4: OFDM digital implementation.

The resulting expressions (5.1) and (5.2) are obtained by replacing ∆f = 1/T and
T = Ns Ts in equation (4.4).

K−1
X
u(nTs ) = dk ej2πk∆f nTs (5.1)
k=0

K−1
X j2πkn
u(n) = dk e Ns , n = 0, . . . , Ns (5.2)
k=0

For convenience, fs = LB > 2(f0 + B) so that L is a factor multiple of two with Ns =

LK; equation (5.3) is interpreted as the IDFT of length Ns by zero-padding the sequence {dk }K−1
0
as stated in equation (5.4). The IDFT and DFT operations are performed by the corresponding
efficient algorithms IFFT and FFT respectively.

s −1
NX
j2πkn
u(n) = dk e Ns , n = 0, . . . , Ns (5.3)
k=0

 
T
u(n) = IFFTNs [d0 . . . dK−1 0| 0{z. . . 0} ] (5.4)
| {z }
K elements (L-1)K elements

The OFDM passband sampled signal is defined by,

s(n) = <{u(n)ej2πf0 nTs } (5.5)

which is converted to the analog domain through the DAC and further transmitted through the
channel. On the receiver side, the passband signal is sampled by the ADC at frequency fs such that
the Nyquist sampling theorem is satisfied. Let v(n) = v(nTs ), n = 0, . . . , Ns − 1 be samples of the

35
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

received signal after performing several operations over the signal (described in 4.5) in the digital
domain. The demodulation process is implemented through the FFT (equation (??)).
 
1
yk = FFTNs [v0 . . . vK−1 vK . . . vNs −1 ]T (5.6)
Ns | {z } | {z }
K elements (L-1)K elements

the resulting vector yk from (5.6) contains Ns elements, so the differentially coherent detection is
performed only over the first K samples.
The Simulink models for the transmitter and receiver are built using equations (5.4),
(5.5), and (5.6). An illustration of the models is described in Figure 5.5, the transmitter includes
the following sub-systems: Symbol mapping, differential encoder, IFFT, cyclic prefix insertion,
preamble insertion, and baseband to passband. The receiver encompasses the following blocks:
Passband to baseband, frame synchronization, cyclic prefix removal, FFT, differential decoding, and
detection. These sub-systems are explained in the next section.

5.3.1 Simulink Models

The OFDM models are built in Simulink which provides tools and libraries that ease the
simulation and prototyping of the UWA platform. We develop two independent models, transmitter
and receiver, each one is split into several sub-systems. The transmitter model is observed in
Figure 5.5(a) where each small system performs some defined operations that are explained below:

• Symbol Mapping
This block performs a mapping operation between a sequence of information bits c and a
sequence of symbols bk , according to a predefined linear modulation scheme, i.e. Quadrature
Phase Shift Keying (QPSK).

• Differential Encoder
The encoder works at symbol level, so that differentially encoded data symbols dk are obtained
from the original symbols bk , i.e. dk = dk−1 bk , with d0 = 1.

• IFFT
This sub-system implements the operation described by the equation (5.3). A sequence of K
differentially encoded symbols {dk }K−1
0 is zero-padded by a sequence of (L − 1)K elements
(equation (5.4)) and then the IFFT operation is applied. The result is referred to as OFDM
block or symbol.

36
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

OFDM TX IP Core

Cyclic
From PS AXI4- Baseband To DAC
Symbol Differential OFDM Prefix and
Stream IFFT to Pass-
Mapping Encoding Mapper Preamble
Interface band
Insertion

(a)

OFDM RX IP Core

From ADC Passband AXI4- To PS

Frame OFDM Differential
to Base- FFT Detection Stream
Synchro Demapper Decoding
band Interface

(b)

Figure 5.5: OFDM transmitter and receiver models.

• Cyclic Prefix Insertion

The first Ng samples of each OFDM block are inserted between consecutive OFDM blocks.
The number Ng is chosen according to Tg = Ts Ng where Tg is the guard interval or cyclic
prefix duration (described in Chapter 4).

• Preamble Insertion
A pseudo-noise m-sequence, where m = 5, is used to perform synchronization on the receiver
side. In addition, a pause time interval Tpse is inserted between the preamble-first OFDM
block, and the last OFDM block-postamble (preamble of the next frame).

• Baseband to Passband
The stream of OFDM blocks are up-shifted by the first carrier frequency f0 as defined by
equation (5.5) and the preamble-postamble sequences are placed at fc .

The sub-systems of the receiver model (Figure 5.5(b)) are defined:

• Passband to Baseband
The samples of the received signal from the ADC are downshifted and filtered to obtain the
complex baseband received signal.

37
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

• Frame Synchronization
The complex baseband signal is correlated with the transmitted preamble sequence. The
OFDM blocks are synchronized by finding the maximum of the correlation function.

• Cyclic Prefix Removal

The last Ng samples of each OFDM block, which correspond to the cyclic prefix, are removed.

• FFT
This system describes the OFDM demodulation stated in equation (5.6). The length of the
FFT and the output sequence yk is Ns , so the first K elements of yk are kept and the others
are discarded.

• Differential Decoding
The samples {yk }K−1
k=0 are differentially decoded by using equation (??) to obtain the decoded
symbols b̂k .

• Detection
The differentially decoded symbols b̂k are converted to a sequence of information bits ĉ.

In addition to the OFDM sub-systems, there are two blocks designed to interface the
PS with the PL and vice versa, and use the AXI4-Stream protocol. On the transmitter model, a
user-defined bit sequence c is transferred from the PS to the PL. Similarly, on the receiver side the
detected bit-sequence ĉ is transferred from the PL to the PS.

Custom Parameters
- Number of Carriers
- Number of Blocks
- Sampling Frequency MATLAB function Simulink Model
.
.
.
ofdmSystemParameters.mat

Figure 5.6: System Parameters. A Matlab function, that includes the underwater system parameters,
such as number of carriers, number of blocks, sampling frequency, etc, generates a .mat file called
”ofdm SystemParameters” and this file is further called inside the Simulink models.

The model presented in Figure 5.5 has the ability to work with different system parameters,
such as bandwidth, number of carriers, number of OFDM blocks, and sampling frequency. The
system parameters are defined by a Matlab function that generates a structured file to be called inside

38
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

the Simulink model (see Figure 5.6). Even though changing parameters is possible, we have to
consider that they will further define the resources used in the PL, which are limited.

Controller
Memory
DRAM

DRAM
AXI4-Stream OFDM TX To
AXI DMA
IP Core DAC

control
ARM

ARM

ZYNQ Processing System ZYNQ Programmable Logic

Microzed Board

Controller

Memory
DRAM
DRAM

From OFDM RX AXI4-Stream

AXI DMA
ADC IP Core

control
ARM

ARM

ZYNQ Programmable Logic ZYNQ Processing System

Microzed Board

Figure 5.7: OFDM Transmitter and Receiver Platform.

5.3.2 Block Diagram Design

The block diagrams inside the Zynq SoC are built such that the PS and PL can interact
with each other. The so-called OFDM IP cores generated from the Simulink models through the

39
CHAPTER 5. UNDERWATER PLATFORM OVERVIEW

HDL Coder tool, are targeted to the PL and designed to communicate with the platform end-user.
Thus, the end-user send or receive information from the OFDM IP cores through an user application
running on the PS.
The process of communicating PS and PL is carried out by the AXI DMA core and is
defined in the Xilinx Vivado Design Suite (see Figure 5.7). AXI DMA provides high-bandwidth
direct memory access between memory and AXI4-Stream target peripherals. For instance, user data
is placed on the on-chip memory DRAM, then the AXI DMA bridges such data from memory to the
AXI4-Stream channel that connects the OFDM IP core, all sequentially controlled by the PS. On the
other side, the AXI DMA receives data from the AXI4-Stream channel and places on DRAM that
can be further accessed by an application running on the PS.

40
Chapter 6

Results

This chapter describes the architecture of the UWA prototype which includes hardware
components such as DAC and ADC boards, power amplifiers, and acoustic transducers. We briefly
point out the most relevant technical characteristics of each hardware element, and a complete block
diagram of the prototype is presented.
In addition, we report experimental results from a system deployed in both indoor en-
vironment (water tank) and outdoor environment (sea water pool). The performance of the UWA
prototype is evaluated in terms of data detection mean squared error (MSE) and bit error rate (BER)
from a system designed to use the 120 − 130kHz band.

6.1 UWA Modem

The UWA platform prototype is built around the Microzed board that incorporates the
Zynq SoC device. The prototype also contains other components, namely analog-to-digital and
digital-to-analog converters (ADC and DAC respectively), power amplifiers, and acoustic transducers.
Table 6.1 summarizes the hardware components with their corresponding models.

Hardware Device Model

DAC board DC2459A Evaluation Board
ADC board DC528A Demonstration Circuit
Power Amplifier ZHL-6A+
Pre-amplifier Teledyne VP2000
Hydrophone Teledyne Reson TC4013

Table 6.1: Hardware components of UWA prototype.

41
CHAPTER 6. RESULTS

6.1.1 Hardware components

• DC2459A Evaluation Board

This demonstration circuit features the LTC1668 of 16-bit input and sampling frequency of
50Msps. The Microzed board generates a digital signal of 16-bit resolution that drives the
input of the DAC evaluation circuit. The DAC analog output is amplified by the ZHL-6A+
device.

• DC528A Demonstration circuit

The DC528A evaluation board based on the LTC1740 which has 14-bit resolution and sampling
frequency up to 6Msps. The ADC board converts the analog signal received from the transducer
to a 14-bit digital pattern that is fed to the Microzed board.

• Mini-Circuits ZHL-6A+ Power Amplifier

The Mini-Circuits ZHL-6A+ is an amplification device that offers a gain up to 10dB. The
operational band ranges from 2.5kHz to 50MHz.

• Teledyne VP2000 Pre-amplifier

The Teledyne VP2000 serves both in amplifying and filtering the received signal. It also
provides variable gain selection in six levels from 0 to 50dB. The filter offers a range of 12
high-pass and 12 low-pass filters with variable cutoff frequencies, so that it allows ideal band
pass filtering.

• Hydrophone TC4013
The Teledyne Reson TC4013 is used for both transmission and reception as it offers a relative
wideband usable frequency range of 1Hz to 170kHz (see Figure 6.1). The TC4013 also
provides uniform omnidirectional sensitivities in both horizontal and vertical planes up to high
frequencies.

6.1.2 Prototype

The development of the UWA prototype consists in two parts: 1) the FPGA modeling
explained in Chapter 5 and 2) the hardware components. The workflow used to design the FPGA
model (illustrated in Figure 5.3) produces the bitstream file that is targeted to the Zynq 7000 SoC
enclosed in the Microzed board. The FPGA DSP capabilities are integrated with real-world analog

42
CHAPTER 6. RESULTS

Figure 6.1: Hydrophone TC4013 frequency response [53].

signals through the use of hardware components, namely ADC and DAC devices. Figure 6.2 depicts
hardware block diagram, a power amplifier is used in both the transmitter and receiver sides.

• Transmitter
The output of the Microzed board is a bus of 16-bit digital outputs that are connected to the
digital input ports of the DAC DC2459A Evaluation Board. The analog output signal of the
DAC is amplified by the Mini-Circuits ZHL-6A+ power amplifier, and is further connected to
the Teledyne Reson TC4013 transducer.

• Receiver
The signal received by the Teledyne Reson TC4013 is pre-amplified through the Teledyne
VP2000. The pre-amplified signal is sampled by the ADC DC528A board, where each sample
is represented by 14-bit length words, and thus, a digital bus of 14-bits is connected between
the ADC and the Microzed board.

43
CHAPTER 6. RESULTS

Transmitter
Power Amplifier

FPGA 16 Digital
Outputs
Analog Transducer
DAC
Signal
K
Board

Microzed Board – DAC Transducer

Xilinx-7000 SoC DC2459A-A TC-4013
ZHL-6A+

Receiver
Pre Amplifier

FPGA
Analog 14 Digital
Transducer Signal Outputs
K ADC
Board

Transducer Teledyne ADC Microzed Board –

TC-4013 VP2000 DC528A Xilinx-7000 SoC

Figure 6.2: Hardware block diagrams.

44
CHAPTER 6. RESULTS

6.2 Performance Evaluation

To asses the system performance, we performed several experimental testings in tank and
pool, whose dimensions are showed in Figure 6.3. The experimental OFDM signal was transmitted
in the band of 120kHz to 130kHz where the transducer transmitting sensitivity response is relatively
flat as depicted in Figure 6.1.

150 cm

TX d RX

95 cm
295 cm

100cm
TX d RX

600cm

Figure 6.3: Tank and pool dimensions.

6.2.1 System parameters

The signal bandwidth is B = 10kHz placed at the central frequency fc = 125kHz. The
CP-OFDM system has a guard interval of Tg = 25ms and total number of carriers K = 1024. The
carrier spacing ∆f = 9.7656Hz and the OFDM block duration is T = 102.4ms. Each OFDM frame
contains Nb = 8 OFDM blocks and a total of Nd = 8192 data symbols. Table 6.2 summarizes the
system parameters.

6.2.2 Experimental Result

The experiment consists of two parts: 1) measurement of the channel impulse response
based on the preamble signal, and 2) measurement of the system performance.

45
CHAPTER 6. RESULTS

first carrier frequency f0 [KHz] 120

bandwidth B [KHz] 10
guard interval Tg [ms] 25
pause interval Tpse [ms] 25
preamble carrier frequency fc [KHz] f0 + B/2
preamble generating polynomial x5 + x4 + 1
number of carriers K 1024
number of OFDM blocks Nb 8

Table 6.2: Experiment Signal Parameters.

0.6 0.6

0.4 0.4

0.2 0.2

0 0
-0.5 0 0.5 1 1.5 -0.5 0 0.5 1 1.5

0.6 0.6

0.4 0.4

0.2 0.2

0 0
-0.5 0 0.5 1 1.5 -0.5 0 0.5 1 1.5

(a) (b)

Figure 6.4: Channel profile based on preamble correlation measured in (a) tank and (b) pool.

6.2.2.1 Channel impulse response

The approximate channel impulse response is obtained by correlating the preamble signal,
which is a pseudo-noise sequence with m = 5. The normalized correlation of the received preambles
with the transmitted one is illustrated in Figure 6.4. The first-row plot shows the magnitude square of
the correlation of the preambles overlapped as a function of the delay time. While the second-row
plot depicts the average correlation. In both figures, the direct path signal occurs at τ = 0ms where
the correlation function peaks. The plots also show some signal replicas that occur in different path
delays. The total multipath spread Tmp of the channel is governed by the longest path delay. The tank
multipath spread (Figure 6.4(a)) reports an approximate of Tmp ≈ 1.1ms, while the pool multipath
spread (Figure 6.4(b)) accounts for Tmp ≈ 0.75ms.

46
CHAPTER 6. RESULTS

Tx
Rx
≈ 380 cm

B
A Tx ≈ 550 cm
Rx
D

B D
E 95 cm

A(120,80)
≈ 380 cm
C B(100,50)
C(150,50)
D(120,20)
E(200,50)
(0,0) C

295 cm

(a) (b)

Figure 6.5: Geometry of the experiment in (a) tank and (b) pool.

6.2.2.2 System performance

The experiment involves multiple transmissions, each containing the OFDM signal whose
parameters are listed in Table 6.2. There is a total of 50 transmissions taken from each four fixed
positions in the tank and three locations in the pool. Figure 6.5 illustrates the geometry of the
experiment in tank and pool setups. In the tank, the transmitter is placed on point E, and the receiver
is set on points A, B, C, and D. In the pool, the transmitter is specified by point D, while the receiver
is located on A, B, and C.
The performance of the system is presented in terms of data detection mean squared error
(MSE) and bit error rate (BER). We report on the estimated cumulative density function (CDF) of
the MSE measured in each signal frame. Furthermore, we show both the bit error rate (BER) and
block error rate (BLER) of the system without channel coding.
The MSE corresponding to an OFDM signal with K carriers is measured in the n-th block
of the i-th frame as

K−1
1 X i
M SE i (n, K) = |bk (n) − b̂ik (n)|2 (6.1)
K −1
k=1

and the MSE per frame is obtained as

47
CHAPTER 6. RESULTS

Nb
i 1 X
M SE (K) = M SE i (n, K) (6.2)
Nb
n=1

The estimated cumulative distribution function (CDF) of the MSE per block, measured
in the tank and the pool, is illustrated in Figure 6.6(a) and Figure 6.6(c), respectively. Each figure
reports measurements from four points in tank and three locations in pool (Figure 6.5). Moreover,
the average MSE per frame received is showed in Figure 6.6(b) and Figure 6.6(d) for tank and pool
setups, accordingly.
The MSE measured in 90% of the total OFDM blocks and frames in the tank is summarized
in Table 6.3. The best performance is observed on position D, whose MSE per block is less than
−20.66dB and MSE per frame is below −21.14dB.

Position MSE [dB] per block MSE [dB] per frame

A < −16.86 < −17.67
B < −15.01 < −15.86
C < −16.46 < −17.2
D < −20.66 < −21.14

Table 6.3: MSE performance for 90% of the OFDM blocks and frames recorded in tank.

Similarly, the experiment in pool for 90% of the total OFDM blocks and frames is outlined
in Table 6.4. The best MSE result is reported on position C, whose MSE per block is less than
−9.27dB and MSE per frame is less than −10.09dB.

Position MSE [dB] per block MSE [dB] per frame

A < −9.09 < −9.92
B < −7.85 < −8.35
C < −9.27 < −10.09

Table 6.4: MSE performance for 90% of the OFDM blocks and frames recorded in pool.

In Figure 6.7, we demonstrate the performance of the system in terms of bit error rate BER
versus frame index and average block error rate (BLER) versus block index.
The results from tank experiments (Figure 6.7(a) and (b)) describe an average BLER
uniformly distributed on each OFDM block. The best average BLER is observed on position D,
which is in the order of 3 × 10−4 . While the worst average BLER results from position B, which

48
CHAPTER 6. RESULTS

accounts for 1.7 × 10−3 . The BER measurements on each frame transmission present similar error
rates on average.
Equivalently, the experiments in pool (Figure 6.7(c) and (d)) present identical average
BLER values across the OFDM block index. The best average BLER performance reaches 2.6×10−3
on position C. The worst average BLER is seen on position B which accounts for 3.6 × 10−3 . The
BER measurements on each frame transmission present similar error rates on average.

1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
-25 -20 -15 -10 -25 -20 -15 -10

(a) (b)
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
-15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5 -15 -14 -13 -12 -11 -10 -9 -8 -7 -6 -5

(c) (d)

Figure 6.6: The estimated CDFs of the MSE per block and MSE per frame measured in tank (top)
and pool (bottom) setups. The experiments, both in tank and pool, consist of 50 frame transmissions
and each frame is composed of 8 OFDM blocks. The CDFs in (a) and (c) reflect the recorded MSE
on each OFDM block, while the CDFs in (b) and (d) reflect the average MSE per frame on each
receiving point during the experiment.

49
CHAPTER 6. RESULTS

10-2 10-2

10-3 10-3

10-4 10-4
1 2 3 4 5 6 7 8 0 5 10 15 20 25 30 35 40 45 50

(a) (b)
10-2 10-2

10-3 10-3
1 2 3 4 5 6 7 8 0 5 10 15 20 25 30 35 40 45 50

(c) (d)

Figure 6.7: Average block error rate (BLER) and bit error rate (BER) measured in tank (top) and
pool (bottom) setups. The results in (a) and (c) reflect the average BLER for 50 frame transmissions
recorded on each position, in tank (A, B, C, and D) and in pool (A, B, and C). The plots (b) and (d)
indicate the BER calculated on each OFDM frame transmission.

50
Chapter 7

Conclusion

In this research work, we designed and implemented an UWA platform based on CP-
OFDM which incorporates differentially coherent detection in frequency domain. Particularly, this
detection technique eliminates the need for channel estimation, and thus, increases the system
robustness on rapidly varying UWA channels. Differentially coherent detection has reported excep-
tional performance results in terms of MSE of the data detected and its appealing obeys to the low
computational resources used.
Furthermore, we developed hardware and software for the UWA platform. By exploiting
the Simulink capabilities, the CP-OFDM system is modeled and simulated, which also allows us
to conveniently change the system parameters, such as bandwidth, number of carriers, number of
OFDM blocks per frame, and the sampling frequency. The HDL code is then generated from the
Simulink model and further the OFDM IP cores are created for the transmitter and the receiver.
The OFDM IP cores are interconnected with other IPs, namely AXI DMA and ZYNQ Processing
System, in a block design using the Vivado Design Suite. This software tool synthesizes, implements,
and generates the bitstream file that is targeted on the Microzed. The software part is used to
interface the end user with the UWA platform and it is generated using the Xilinx SDK using C
programming language. The Microzed board is the main component of the final UWA prototype,
which encompasses several hardware components, namely ADC and DAC boards to perform the
conversion between the digital and analog domain or vice versa, power amplifiers, and acoustic
transducers.
Finally, we evaluated the performance of the UWA prototypes in both indoor (water tank)
and outdoor (sea water pool) environments. The experiment was split in two parts. First, we
conducted several transmissions of the preamble signal, which allowed us to obtain the channel

51
CHAPTER 7. CONCLUSION

impulse response and measure the multipath delay spread of the channel. The second part consisted
in assessing the system performance. The results are obtained from about 50 real time transmissions
during the experiment. The experiments in tank reported an average MSE below −20.66dB for
90% of OFDM blocks and achieves BLER as low as 3 × 10−4 without channel coding. Also, the
experiments in pool showed an average MSE below −9.27dB for 90% of OFDM blocks and achieves
BLER as low as 2.6 × 10−3 without channel coding.

Future work

The current FPGA-based UWA prototypes can establish a fixed point-to-point real-time
communication scenario. However, we plan to implement the Doppler compensation algorithm,
described in Section 4.3.1, as a future work. This high-resolution Doppler compensation method will
allow us to deploy mobile underwater acoustic communication and further multichannel implementa-
tion.

52
Bibliography

[1] H. S. Dol, P. Casari, T. van der Zwan, and R. Otnes, “Software-defined underwater acoustic
modems: Historical review and the nilus approach,” IEEE Journal of Oceanic Engineering,
vol. 42, no. 3, pp. 722–737, July 2017.

[2] M. Stojanovic, “A method for differentially coherent detection of ofdm signals on doppler-
distorted channels,” in 2010 IEEE Sensor Array and Multichannel Signal Processing Workshop,
Oct 2010, pp. 85–88.

[3] Y. M. Aval and M. Stojanovic, “Differentially coherent multichannel detection of acoustic ofdm
signals,” IEEE Journal of Oceanic Engineering, vol. 40, no. 2, pp. 251–268, April 2015.

[4] A. Tadayon and M. Stojanovic, “Frequency offset compensation for acoustic ofdm systems,” in
OCEANS 2017 - Anchorage, Sep. 2017, pp. 1–5.

[5] J. Heidemann, M. Stojanovic, and M. Zorzi, “Underwater sensor networks: applications,

advances and challenges,” Philosophical Transactions of the Royal Society A: Mathematical,
Physical and Engineering Sciences, vol. 370, no. 1958, pp. 158–175, 2012.

[6] H. Kaushal and G. Kaddoum, “Underwater optical wireless communication,” IEEE Access,
vol. 4, pp. 1518–1547, 2016.

[7] M. Stojanovic, “Acoustic (Underwater) Communications,” in Encyclopedia of Telecommunica-

tions, J. G. Proakis, Ed. John Wiley and Sons, 2003.

[8] T. Melodia, H. Kulhandjian, L. Kuo, and E. Demirors, “Advances in Underwater Acoustic

Networking,” in Mobile Ad Hoc Networking: Cutting Edge Directions, second edition ed.,
S. Basagni, M. Conti, S. Giordano, and I. Stojmenovic, Eds. Inc., Hoboken, NJ: John Wiley
and Sons, 2013, pp. 804–852.

53
BIBLIOGRAPHY

[9] P. Qarabaqi and M. Stojanovic, “Statistical characterization and computationally efficient

modeling of a class of underwater acoustic communication channels,” IEEE Journal of Oceanic
Engineering, vol. 38, no. 4, pp. 701–717, Oct 2013.

[10] A. Radosevic, J. G. Proakis, and M. Stojanovic, “Statistical characterization and capacity of

shallow water acoustic channels,” in OCEANS 2009-EUROPE, May 2009, pp. 1–8.

[11] M. Stojanovic and J. Preisig, “Underwater acoustic communication channels: Propagation

models and statistical characterization,” IEEE Communications Magazine, vol. 47, no. 1, pp.
84–89, January 2009.

[12] C. T. Tindle, “Wavefronts and waveforms in deep-water sound propagation,” The Journal of
the Acoustical Society of America, vol. 112, no. 2, pp. 464–475, 2002. [Online]. Available:
https://doi.org/10.1121/1.1489437

[13] F.-X. Socheleau, J.-M. Passerieux, and C. Laot, “Characterisation of time-varying

underwater acoustic communication channel with application to channel capacity,” in
Underwater Acoustic Measurements, Nafplion, Greece, Jun. 2009. [Online]. Available:
https://hal.archives-ouvertes.fr/hal-00473702

[14] B. Tomasi, G. Zappa, K. McCoy, P. Casari, and M. Zorzi, “Experimental study of the space-
time properties of acoustic channels for underwater communications,” in OCEANS’10 IEEE
SYDNEY, May 2010, pp. 1–9.

[15] M. Stojanovic, “Underwater acoustic communications: Design considerations on the physical

layer,” in 2008 Fifth Annual Conference on Wireless on Demand Network Systems and Services,
Jan 2008, pp. 1–10.

[16] J. Catipovic, A. Baggeroer, K. Von Der Heydt, and D. Koelsch, “Design and performance
analysis of a digital acoustic telemetry system for the short range underwater channel,” IEEE
Journal of Oceanic Engineering, vol. 9, no. 4, pp. 242–252, October 1984.

[17] G. Qiao, Z. Babar, L. Ma, S. Liu, and J. Wu, “Mimo-ofdm underwater acoustic communication
systems—a review,” Physical Communication, vol. 23, pp. 56 – 64, 2017. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1874490716301550

54
BIBLIOGRAPHY

[18] J. Catipovic, A. Baggeroer, K. Von Der Heydt, and D. Koelsch, “Design and performance
analysis of a digital acoustic telemetry system for the short range underwater channel,” IEEE
Journal of Oceanic Engineering, vol. 9, no. 4, pp. 242–252, October 1984.

[19] L. E. Freitag and J. S. Merriam, “Robust 5000 bit per second underwater communicationsystem
for remote applications,” in Marine Instrumentation, Marine TechnologySociety Proceedings,
vol. 1, Feb 1990, p. 201–207.

[20] L. E. Freitag, J. S. Merriam, D. E. Frye, and J. A. Catipovic, “A long-term deep-water acoustic

telemetry experiment,” in OCEANS 91 Proceedings, vol. 1, Oct 1991, pp. 254–260.

[21] G. R. Mackelburg, “Acoustic data links for uuvs,” in OCEANS 91 Proceedings, Oct 1991, pp.
1400–1406.

[22] K. F. Scussel, J. A. Rice, and S. Merriam, “A new mfsk acoustic modem for operation in adverse
underwater channels,” in Oceans ’97. MTS/IEEE Conference Proceedings, vol. 1, Oct 1997, pp.
247–254 vol.1.

[23] G. Mackelburg, S. Watson, and A. Gordon, “Benthic 4800 bits/s acoustic telemetry,” in OCEANS
81, Sep. 1981, pp. 72–72.

[24] L. Olson, J. Backes, and J. Miller, “Communication, control, and data acquisition systems on
the ishte lander,” IEEE Journal of Oceanic Engineering, vol. 10, no. 1, pp. 5–16, January 1985.

[25] G. S. Howe, O. R. Hinton, A. E. Adams, and A. G. J. Holt, “Acoustic burst transmission of

high rate data through shallow underwater channels,” Electronics Letters, vol. 28, no. 5, pp.
449–451, Feb 1992.

[26] M. Suzuki, T. Sasaki, and T. Tsuchiya, “Digital acoustic image transmission system for deep-sea
research submersible,” in OCEANS 92 Proceedings Mastering the Oceans Through Technology,
vol. 2, Oct 1992, pp. 567–570.

[27] J. C. Jones, A. Di Meglio, L. S. Wang, R. F. W. Coates, A. Tedeschi, and R. J. Stoner, “The

design and testing of a dsp, half-duplex, vertical, dpsk communication link,” in Oceans ’97.
MTS/IEEE Conference Proceedings, vol. 1, Oct 1997, pp. 259–266 vol.1.

[28] I. F. Akyildiz, D. Pompili, and T. Melodia, “Underwater acoustic sensor networks: research
challenges,” Ad Hoc Networks, vol. 3, no. 3, pp. 257 – 279, 2005. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1570870505000168

55
BIBLIOGRAPHY

[29] M. Stojanovic, “Low complexity ofdm detector for underwater acoustic channels,” in OCEANS
2006, Sep. 2006, pp. 1–6.

[30] Baosheng Li, Shengli Zhou, Jie Huang, and P. Willett, “Scalable ofdm design for underwater
acoustic communications,” in 2008 IEEE International Conference on Acoustics, Speech and
Signal Processing, March 2008, pp. 5304–5307.

[31] J. Huang, S. Zhou, J. Huang, C. R. Berger, and P. Willett, “Progressive inter-carrier interference
equalization for ofdm transmission over time-varying underwater acoustic channels,” IEEE
Journal of Selected Topics in Signal Processing, vol. 5, no. 8, pp. 1524–1536, Dec 2011.

[32] D. B. Kilfoyle, J. C. Preisig, and A. B. Baggeroer, “Spatial modulation experiments in the

underwater acoustic channel,” IEEE Journal of Oceanic Engineering, vol. 30, no. 2, pp. 406–
415, April 2005.

[33] Lizhong Zheng and D. N. C. Tse, “Diversity and multiplexing: a fundamental tradeoff in
multiple-antenna channels,” IEEE Transactions on Information Theory, vol. 49, no. 5, pp.
1073–1096, May 2003.

[34] E. M. Sozer, M. Stojanovic, and J. G. Proakis, “Underwater acoustic networks,” IEEE Journal
of Oceanic Engineering, vol. 25, no. 1, pp. 72–83, Jan 2000.

[35] D. E. Lucani, M. Stojanovic, and M. Medard, “On the relationship between transmission power
and capacity of an underwater acoustic communication channel,” in OCEANS 2008 - MTS/IEEE
Kobe Techno-Ocean, April 2008, pp. 1–6.

[36] M. Stojanovic, “On the relationship between capacity and distance in an underwater
acoustic communication channel,” in Proceedings of the 1st ACM International Workshop on
Underwater Networks, ser. WUWNet ’06. New York, NY, USA: ACM, 2006, pp. 41–47.
[Online]. Available: http://doi.acm.org/10.1145/1161039.1161049

[37] D. E. Lucani, M. Medard, and M. Stojanovic, “Underwater acoustic networks: Channel models
and network coding based lower bound to transmission power for multicast,” IEEE Journal on
Selected Areas in Communications, vol. 26, no. 9, pp. 1708–1719, December 2008.

[38] L. M. Brekhovskikh, Yu, L. M. Brekhovskikh, and Y. Lysanov, Fundamentals of Ocean

Acoustics. Springer, 2003.

56
BIBLIOGRAPHY

[39] R. Coates, Underwater Acoustic Systems, ser. New Electronics. Macmillan Education, Limited,
1990.

[40] L. M. Wolff, E. Szczepanski, and S. Badri-Hoeher, “Acoustic underwater channel and network
simulator,” in 2012 Oceans - Yeosu, May 2012, pp. 1–6.

[41] A. Radosevic, R. Ahmed, T. Duman, J. Proakis, and M. Stojanovic, “Adaptive ofdm modulation
for underwater acoustic communications: Design considerations and experimental results,”
IEEE Journal of Oceanic Engineering, vol. 39, no. 2, pp. 357–370, 2014.

[42] M. Stojanovic, “Mimo ofdm over underwater acoustic channels,” in 2009 Conference Record
of the Forty-Third Asilomar Conference on Signals, Systems and Computers, Nov 2009, pp.
605–609.

[43] Y. M. Aval and M. Stojanovic, “A method for differentially coherent multichannel processing
of acoustic ofdm signals,” in 2012 IEEE 7th Sensor Array and Multichannel Signal Processing
Workshop (SAM), June 2012, pp. 73–76.

[44] Y. Aval, S. K. Wilson, and M. Stojanovic, “Differentially coherent detection: Lower complexity,
higher capacity?” in 2016 Information Theory and Applications Workshop (ITA), Jan 2016, pp.
1–5.

[45] B. Li, S. Zhou, M. Stojanovic, L. Freitag, and P. Willett, “Multicarrier communication over
underwater acoustic channels with nonuniform doppler shifts,” IEEE Journal of Oceanic
Engineering, vol. 33, no. 2, pp. 198–209, April 2008.

[46] M. Stojanovic and S. Tadayon, “Estimation and tracking of time-varying channels in ofdm sys-
tems,” in 2014 52nd Annual Allerton Conference on Communication, Control, and Computing
(Allerton), Sep. 2014, pp. 116–122.

[47] L. H. Crockett, R. A. Elliot, M. A. Enderwitz, and R. W. Stewart, The Zynq Book: Embedded
Processing with the Arm Cortex-A9 on the Xilinx Zynq-7000 All Programmable Soc. UK:
Strathclyde Academic Media, 2014.

[48] AVNET. Microzed. [Online]. Available: http://zedboard.org/product/microzed

[49] Xilinx. Processing system 7. [Online]. Available: https://www.xilinx.com/support/

documentation/ip documentation/processing system7/v5 5/pg082-processing-system7.pdf

57
BIBLIOGRAPHY

[50] ——. Axi reference guide. [Online]. Available: https://www.xilinx.com/support/documentation/

ip documentation/ug761 axi reference guide.pdf

[51] ——. Axi dma ip core guide. [Online]. Available: https://www.xilinx.com/support/

documentation/ip documentation/axi dma/v7 1/pg021 axi dma.pdf

[52] MathWorks. Hdl coder. [Online]. Available: https://www.mathworks.com/products/hdl-coder.

html

[53] T. Marine. Reson tc 4013 - hydrophone. [Online]. Available: http://www.teledynemarine.com/

reson-tc4013?ProductLineID=48

58