TECHNICAL UNIVERSITY OF CRETE

Department of Electronic and Computer Engineering

Software-Defined Radio Impementation

of OFDM

Author: Supervisor:
Stylianos Leonidakis Prof. Athanasios Liavas
August 2014
 Acknowledgements

First of all, i would like to thank my supervisor, Professor Athanasios Liavas for suggesting
me with such an interesting topic and for his overall guidance and suggestions throughout
my thesis. I would also like to thank my family for being so understanding and supportive.
Last but not least, i would like to thank my friends for their invaluable help and support
throughout my student years.
 Abstract

Wireless communications is by any measure the fastest growing segment of the commu-
nications industry. Cellular phones, Wireless Local Area Networks (WLANs), Smart Homes
and telemedicine are some applications that make extensive use of wireless communications.
Orthogonal Frequency Division Multiplexing (OFDM) is a modulation that has been
applied to a wide variety of applications and is adopted in many recently standardized
broadband communication systems. OFDM is an appealing technique for achieving high-bit-
rate data transmision, due to its robustness against the multipath fading and Inter Symbol
Interference. Since the OFDM transmission is vulnerable to time and frequency offsets,
accurate estimation of these parameters is one of the most important tasks of the OFDM
receiver. In this thesis, we implement a time synchronization algorithm based on correlation
of two identical parts of an OFDM pilot symbol. In addition, knowledge of the channel in the
transmitter is very useful for effective transmission and/or interference mitigation (especially
in static environments). Thus, channel feedback is utilized throughout the experiments
conducted for this thesis.
The experiments were conducted on a software-defined-radio (SDR) testbed based on the
Universal Software Radio Peripheral (USRP).
Contents

1 Introduction 1
1.1 Multicarrier Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Software Defined Radio (SDR) . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 OFDM Structure 3
2.1 OFDM Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 OFDM Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Time synchronization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Carrier Frequency Offset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.1 Flat Fading Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.2 Frequency Selective Channel . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Channel Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5.1 Flat Fading Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3 Software Tools and USRP 17

3.1 Universal Software Radio Peripheral (USRP) . . . . . . . . . . . . . . . . . 17
3.2 USRP 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 GNU Radio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.1 GNU Radio Companion . . . . . . . . . . . . . . . . . . . . . . . . . 20

4 Experimental part 21
4.1 Single Antenna Point to Point System with channel feedback . . . . . . . . . 21
4.2 Single to Dual Antenna Point to Point System with channel feedback . . . . 25
4.3 Single Point to multi Point System with channel feedback . . . . . . . . . . . 28
4.4 Interference cancellation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.4.1 Orthogonal Beamforming . . . . . . . . . . . . . . . . . . . . . . . . 31

5 Conclusion 35
List of Figures

2.1 OFDM Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.2 An OFDM frame in the transmitter . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 OFDM Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.4 An OFDM frame in the receiver . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.5 Structure of an OFDM synchronization symbol . . . . . . . . . . . . . . . . 6
2.6 Real and imaginary part of Synchronization Symbol. . . . . . . . . . . . . . 8
2.7 Pilot based method for high SNR (10 dB). . . . . . . . . . . . . . . . . . . . 10
2.8 Pilot based method for low SNR (3 dB). . . . . . . . . . . . . . . . . . . . . 11
2.9 A realization of |F [{zn }] | . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1 The USRP block diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.1 A point to point system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.2 Process carried out in Node B . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.3 Process carried out in Node A . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.4 Channel Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.5 Single to Dual Antenna point to point . . . . . . . . . . . . . . . . . . . . . 25
4.6 Data symbols for both antennas . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.7 Point to Multi point system . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.8 Channel estimation of the receiver that will not receive any future packets . 29
4.9 2x2 system, Rx1 only receiving data signals . . . . . . . . . . . . . . . . . . 32
4.10 Received packet in Rx2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.11 Received packet in Rx1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Chapter 1

Introduction

Wireless Communication

Wireless communication is the transfer of information between two or more points that are
not connected by an electrical conductor. The most common way of wireless communication
is by electromagnetic waves.
In the past decades, wireless communication systems gained enormous popularity over
wired systems, mainly because they are in many applications much cheaper to implement.
Besides, in places where the environment hampers cable deployment, wireless connection
remains the only means of communication.

1.1 Multicarrier Modulation

Digital bandpass modulation techniques can be broadly classified into two categories. The
first is single-carrier modulation, where data is transmitted by using one radio frequency
carrier. The other is multicarrier modulation, where data is transmitted by simultaneously
modulating multiple radio frequency carriers. By using multicarrier modulation, we divide
the stream of data into several bit streams, each of them having a lower bit rate.
During the early stages of multicarrier modulation, the signal bandwidth was divided
into non-overlapping frequency subchannels, with each transmitting a different data stream
from a common source. The fact that there was no overlap between the data streams helped
eliminating interference among different data streams (interchannel interference). On the
other hand, it resulted in a very inefficient use of the available spectrum. These systems
involved a high hardware complexity since parallel data transmission was implemented using
a large number of oscillators, each tuned to a specific subcarrier.

1
 Orthogonal Frequency Division Multiplexing (OFDM) is a multicarrier modulation tech-
nique where data symbols are modulated on a number of regularly spaced subcarriers. These
subcarriers have the minimum frequency separation required to maintain orthogonality (of
their corresponding time domain waveforms) yet the signal spectra of the different subcarri-
ers overlap in the frequency domain. The spectral overlap results in a waveform that utilizes
bandwidth very efficiently. The most important feature of OFDM is that, by choosing the
subcarrier spacing properly in relation to the channel coherence bandwidth, we can convert a
frequency selective channel into a number of parallel frequency flat subchannels. Techniques
that are appropriate for flat fading channels can then be applied [8].
OFDM has a number of advantages compared to single carrier transmission. Firstly,
OFDM has the ability to cope with severe channel conditions such as narrowband inter-
ference and frequency-selective fading due to multipath. In addition, wideband channel
equalization is attained through many narrow band channel equalizations. Also, Inter Sym-
bol Interference (ISI) is easily eliminated by adding a comparatively short Cyclic Prefix at
the start of each packet. OFDM is a signaling technique that is widely adopted in many
recently standardized broadband communication systems, mainly due to its ability to cope
with frequency selective fading which is common in high data rate communications.

1.2 Software Defined Radio (SDR)

SDR is a rapidly evolving technology that receives enormous atttention and generates interest
in the telecommunication industry. SDR is a revolution in radio design due to its ability to
create radios that can change on the fly, creating new choices for users.
An SDR system is a radio communication system where standard components that are
implemented in hardware (e.g. filters, amplifiers, mixers modulators/demodulators, detec-
tors etc.) are instead implemented using software in a personal computer or an embedded
system. Instead of changing the circuitry to handle different types of radio signals, we can
just change the software. This makes SDR very flexible because we can just replace the
software and operate with completely different protocols.
A basic SDR system may consist of a personal computer and the necessary special-purpose
hardware to transmit and receive electromagnetic waves (a device such as a USRP). A
significant amount of signal processing is handed over to the general-purpose processor of
the personal computer rather than being done in the special-purpose hardware.

2
Chapter 2

OFDM Structure

2.1 OFDM Transmitter

Figure 2.1: OFDM Transmitter

As we mentioned earlier, the difference between OFDM and other FDM systems is orthog-
onality. Orthogonality indicates that there is a precise mathematical relationship between
the frequencies of the carriers of the system. While in normal FDM systems carriers are
spaced apart and guard bands are introduced between adjacent carriers, in OFDM, carriers
are arranged in such a way that they are overlapping, but because of the orthogonality, the
received signal is not affected by Inter Carrier Interference.
In Figure 2.1, we depict the operation of an OFDM transmitter. In the first place, a
training sequence is added in the beginning of each OFDM symbol (trainings’ format will
be addressed in detail in Time synchronization chapter) and then data and trainings are
passed through an Inverse Discrete Fourier Transform. Then a Cyclic Prefix is added at the
beginning of the symbol as a guard interval to cope with Inter Symbol Interference. Once
Cyclic Prefix is added, a Pulse Shaping Filter (in our case, Square Root Raised Cosine) is
applied to the transmitted sequence. Afterwards, data is converted to Analog with the use
of the FPGA Digital to Analog converter and then transmitted through the channel.

3
 Figure 2.2: An OFDM frame in the transmitter

2.2 OFDM Receiver

Figure 2.3: OFDM Receiver

Receiver’s operations are roughly the opposite of the transmitter. Once the signal is
received, the FPGA Analog to Digital converter converts data into a digital signal. A
Matched Filter is applied to the signal (usually a Square Root Raised Cosine filter). Then
the Cyclic Prefix is removed and the rest of the received signal is passed through a Discrete
Fourier Transform. Time synchronization, Carrier Frequency Offset and channel estimation
will be adressed later in this chapter.

4
 In order to avoid a large number of modulators and filters at the transmitter and comple-
mentary filters and demodulators at the receiver, it is desirable to use digital signal processing
techniques such as Discrete Fourier Transform (DFT). The sinusoids of the DFT form an
orthogonal basis set which is necessary to eliminate Inter Symbol Interference. The OFDM
transmitter is implemented using the IDFT and the receiver using DFT. Both IDFT and
DFT would usually require N 2 multiplications and additions. However, by calculating the
output using the Fast Fourier Transform (FFT), we reduce the number of computations to
the order of N log N .

Figure 2.4: An OFDM frame in the receiver

5
2.3 Time synchronization
A number of time synchronization algorithms for the OFDM systems have been proposed.
Most of these algorithms are based on the correlation of identical parts of the OFDM symbols.
One way of identifying the start of the packet is to use the correlation between the cyclic
prefix in the beginning and in the end of the packet. Another way is to create two identical
sequences of training symbols in the start of each OFDM frame.

Figure 2.5: Structure of an OFDM synchronization symbol

The transmitter inserts a synchronization symbol in the beginning of each frame of OFDM
data symbols. The synchronization symbol consists of two identical parts. In the frequency
domain, this means that symbols are only transmitted in every even frequency (tone). Before
each synchronization symbol we insert a (CP) Cyclic Prefix.
The length of the cyclic prefix must be as big as that of the channel impulse response in
order to keep the two halves of the training symbols almost1 identical at the output of the
channel.
The synchronization symbol is placed in the start of each frame. The time synchronization
process is carried out in the time domain, searching for identical symbols in the first and
the second half of the packet. In Figure 2.6, we can observe the two identical halves of the
Synchronization symbol in the time domain.

1 The two halves are not exactly identical because of the phase difference caused by carrier frequency offset.

6
 Let S be the training sequence used as an input to the IDFT module. Then, the output
of the IDFT in the transmitter is denoted by S̃ = [S˜0 , S˜1 , ..., S̃N −1 ]T , where

N −1
˜ 1 X 2πkn
Sn = √ Sk ej N . (2.3.1)
N k=0

The training symbol sequence to be transmitted is expressed as

 T
S̃ = S N −L+1 , S N −L+2 , ..., S N −1 , S0 , S1 , ..., S N −1 , S0 , S1 , ..., S N −1 . (2.3.2)
|2 2
{z 2
} | {z 2 } | {z 2 }
CP F irstHalf SecondHalf

The transmitted signal after pulse shaping will be

X
X(t) = S̃ n gT (t − nT ). (2.3.3)
n

At the channel output the received signal will be

X
Y (t) = c(t) ∗ X(t) + W (t) = S̃ n h(t − nT ) + W (t) (2.3.4)
n

where h(t) = c(t) ∗ gT (t).

With Carrier Frequency Offset and phase offset, we have

X
Y (t) = e−j(2π∆F t+φ) S̃ n h(t − nT ) + W (t). (2.3.5)
n

T
If we sample with period Ts = over
, we obtain

X
Y [k] = Y (kTs ) = e−j(2π∆F kTs +φ) S̃ n h(kTs − nT ) + W (kTs )
n
X (2.3.6)
= ejφ e−j2π∆f k S̃ n hk,n + W [k]
n

7
 Figure 2.6: Real and imaginary part of Synchronization Symbol.

where ∆f := ∆F Ts . If we incorporate ejφ into h we get

Y [k] = e−j2π∆f k r[k] + W [k] (2.3.7)

where r[k] = ejφ
P
n S̃ n hk,n or, in a slightly different notation,

yn = e−j2π∆f n rn + wn . (2.3.8)

8
 At the receiver, we need to find the correct start of the packet. A way to do this, using
 T
the training sequence S̃ = S0 , S1 , ..., S N −1 , is to compute the statistic [2]
2

N
2
−1  ∗  
X
P (d) = yd+m S̃m yd+m+ N S̃m (2.3.9)
2
m=0

where d is a time index corresponding to the first sample in a window of N samples and m
is a time index that runs for each d. By incrementing d, we slide the window in order to find
the start of the packet. We define the optimal d as

d∗ = argmax|P (d)|. (2.3.10)

For an analysis of this estimator see [6]

9
 Figure 2.7: Pilot based method for high SNR (10 dB).

In Figure 2.7 we plot a realization of |P (d)|. As we can see, the main peak is approxi-
mately twice the amplitude of the other peaks of |P (d)|.

This method for time synchronization in OFDM systems is robust and is effective in low
SNR as well. In Figure 2.8, we see that even when the SNR is low the peak of the correct
timing synchronization is still observable and once again approximately twice the amplitude
of the other peaks of |P (d)|.

10
Figure 2.8: Pilot based method for low SNR (3 dB).

11
2.4 Carrier Frequency Offset
In wireless communications, baseband signals are often modulated with a high frequency
carrier in the transmitter, making them more resilient in long distance transmissions, and
demodulated with the same frequency in the receiver so as the signal to be proccessed in the
baseband.

A difference in the frequency between the oscillators in the receiver and the transmitter
and/or the Doppler effect causes a shift in frequency domain. This is called Carrier Frequency
Offset (CFO) and needs to be estimated and eliminated because a demodulated signal with
even a small shift in the frequency domain can result in a large Bit Error Rate (BER).
First of all, time synchronization needs to be achieved in the receiver in order to identify
the beginning of the block. As shown earlier, once we have detected the start of the packet
the received signal affected by CFO and channel’s impact can be expressed as

yn = e−j2π∆f n rn + wn . (2.4.1)

2.4.1 Flat Fading Channel

Assuming a flat channel for the duration of the block, then

yn = e−j(2π∆f n) e−jφ h0 S̃n + wn . (2.4.2)

Denoting h00 = e−jφ h0 , then the received samples can be expressed as

yn = e−j(2π∆f n) h00 S̃n + wn . (2.4.3)

12
 If we multiply yn with the complex conjugate of the known training sequence S̃n∗ we denote
zn

zn = yn S̃n∗ = e−j(2π∆f n) h00 S̃n S̃n∗ + wn0 (2.4.4)

= e−j(2π∆f n) h00 |S̃n |2 + wn0
zn is a noisy complex exponential with frequency f 0 = −∆f .

We can estimate the frequency f 0 if we use Fourier Transform on zn and find the maximum
in the digital frequency domain F∗ = argmax|F [{zn }] |.

Figure 2.9: A realization of |F [{zn }] |

13
 Once F∗ is estimated, it can be used to eliminate the effect of the Carrier Frequency Offset
by creating a sequence

mn = e−j2πF∗ n yn (2.4.5)
mn is the training sequence without CFO and affected only by channel’s impact.

2.4.2 Frequency Selective Channel

In order to estimate the carrier frequency offset, the receiver exploits the special structure
of the OFDM synchronization symbol. So we assume that time synchronization has already
been achieved. As shown earlier, the received signal can be expressed as

yn = e−j2π∆f n rn + wn , for n = 1, 2, ...N (2.4.6)

then the received samples in the first half can be expressed as

N
yn = e−j2π∆f n rn + wn , for n = 1, 2, ... (2.4.7)
2
and the second half of the received samples as

N
yn+ N = e−j(2π∆f (n+ 2 )) rn+ N + wn+ N (2.4.8)
2 2 2

N N
= e−j(2π∆f (n+ 2 )) rn + wn+ N , for n = 1, 2, ...
2 2

because of the structure of the Synchronization symbol we know that rn+ N = rn .

2

14
 As a consequence,

  ∗
∗ −j2π∆f n −j2π∆f (n+ N )
yn yn+ N = e rn + wn e 2 r + w
n n+ N
2 2

N
= e−j2π∆f n rn ej2π∆f (n+ 2 ) rn∗ + e−j2π∆f n rn wn+
∗
N+
2
N
+ wn ej2π∆f (n+ 2 ) rn∗ + ∗
wn wn+ N (2.4.9)
2

= ejπ∆f N |rn |2 + w˜n

where

N
w˜n = e−j(2π∆f n) rn wn+
∗
N + wn e
j(2π∆f (n+ 2 ) ∗ ∗
rn + wn wn+ N.
2 2

Ignoring the noise part, we can estimate the difference in phase ∆f between the two halves
of the synchronization symbol. Using all the samples of the training sequence, the carrier
frequency offset can be estimated by [6]

N N
2
X −1   2
−1 
X
∗ jπ∆f N 2
arg yn yn+ N = arg e |rn | = π∆f N. (2.4.10)
2
n=0 n=0

Thus, an estimate of the carrier frequency offset can be derived as [6]

 N −1 
2P
∗
arg yn yn+ N
n=0 2
f∗ = . (2.4.11)
πN

Once f∗ is calculated, it can be used to eliminate the effect of CFO by creating a sequence

mn = ej2πf∗ n yn (2.4.12)
mn is the training sequence without CFO and affected only by channel’s impact.

15
2.5 Channel Estimation

2.5.1 Flat Fading Channel

Once Carrier Frequency Offset has been cancelled the next step is to estimate the channel
using the corrected training sequence
mn = h00 S̃n + wn .

One way to estimate the channel is [3]

N −1
1 X mn
h= . (2.5.1)
N l=1 S̃n

The channel estimation used above works only for flat fading channels. Another process
is followed in case of a frequency selective channel.

This estimate can be used as the actual value of the channel to recover the transmit signal
as
h∗
y˜n 0 = mn ' S̃n + w˜n 0 . (2.5.2)
|h|2

16
Chapter 3

Software Tools and USRP

3.1 Universal Software Radio Peripheral (USRP)

The Universal Software Radio Peripheral (USRP) products are computer hosted software
radios. They are designed and sold by Ettus Research, LLC and its parent company, National
Instruments. The USRP product family is intended to be a comparatively inexpensive
hardware platform for software radio, and is commonly used by research labs and universities
all around the world. USRPs connect to a host computer through a high-speed USB or
Gigabit Ethernet link. Some USRP models integrate the general functionality of a host
computer with an embedded processor that allows the USRP Embedded Series to operate
in a standalone fashion.

3.2 USRP 1

The USRP1 is the original Universal Software Radio Peripheral hardware (USRP) that
provides entry-level RF processing capability. It is intended to provide SDR development
capability for cost sensitive users and applications. The architecture includes an Altera
Cyclone FPGA, on 64 MS/s dual ADC on 128 MS/s dual DAC and USB 2.0 connectivity to
provide data to host computers. A modular design allows the USRP1 to operate from DC to
6 GHz. The USRP1 platform can support two complete RF daughterboards. This feature
makes the USRP ideal for applications requiring high isolation between transmit and receive
chains, or dual-band transmit/receive operation. The USRP1 can stream up to 8 MS/s to
and from host applications, and users can implement custom functions in the FPGA fabric.
[4]

17
 RFX2400 daughterboard

The RFX2400 daughterboard is a full duplex trasceiver designed specifically for operation
in 2.4 GHz band. The daughterboard has a Transmitter/Receiver antenna plug and a receiver
antenna plug. The transmitter inputs the baseband signal which comes from the mainboard
and it is modulated in the central frequency chosen by the software (gnu radio). Pulse
shaping is implemented through software as well. The output of the transmitter goes to a
two-sided switch which is connected to the input of the transmitter from the one side and
to the antenna plug from the other side. This switch is controlled through software and
it depends whether data is trasmitted or received. The receiver consists of an oscillator
whose frequency is controlled by software, and a mixer that mixes the signal coming from
the antenna with a sinusodial signal coming from the oscillator. This is how the receiver
demodulates the signal and it is sent to the mainboard for sampling.
We must point out that RFX2400 daughterboards do not have an embedded clock and
FPGA’s clock is used instead. So their use is required when we need perfect synchronization
(e.g. when transmitting from both antennas simultaneously).

Mainboard

The USRP has a 4 high-speed analog to digital converters (ADCs) each with 12 bits per
sample with 64 MS/s and 4 high-speed digital to analog converters (DACs) each with 14
bits per sample and 128 MS/s. These 4 buses are connected to an Altera cyclone FPGA and
in turn the FPGA is connected through a USB2 port to the computer. So we have 4 input
channels and 4 output channels available for use.

18
 Figure 3.1: The USRP block diagram

FPGA

According to the above, the maximum rate that a USRP sends and receives at the same
time is
(64M S/s ∗ 12bit/Sample + 128M S/s ∗ 14bit/Sample) ∗ 2 = 640M byte/s

But the data rate that USB2 supports is 32Mbyte/s. On the FPGA there are two digital
filters for decimation and interpolation. The decimation filter’s input is the flow of the
samples from the ADC which comes with a rate of 64MS/s. The decimation filter subsamples
its input flow by a decimation rate factor, which is chosen through the software. Accordingly,
the interpolation filter oversamples the data flow by an interpolation rate factor, which is
also chosen through the software, so that the final sampling rate is 128MS/s (sampling rate
of the DAC). Finally, as the data rate of the USB is not stable, we need to use a buffer before
the interpolation filter in which the samples coming from the PC are stored and then they
are read at a rate of (128MS/s)/interpolation so that in the end we have a 128MS/s. [4]

19
3.3 GNU Radio
USRP is specifically designed to be used with GNU radio. GNU radio is a free and open
source signal processing package for building SDRs. It is widely used in academic and com-
mercial environments to support wireless communication research. All the signal processing
blocks are written in C++ and Python is used to create the network or graph and glue
these blocks together. Every block has a predefined number of input/output interfaces and
performs one or more communication functions in the software domain. Each block can be
edited, upgraded or even implemented independently, without interfering with the whole
communication chain.

3.3.1 GNU Radio Companion

The GNU Radio Companion (GRC) is a graphical programming interface that allows to
generate GNU Radio flow graphs without writing the Python code. All the GNU Radio
blocks can be accessed in the GRC. This graphical interface has been implemented in XML
language. When a new block is generated and if this is required in the GRC, then definition
of the block has to be implemented in the XML language. The definition includes: which
Python function or class should be called; what are the arguments for the function or class;
what is the type of the input/output data streams. The GRC facilitates the user to make
the flow graphs by dragging and dropping the blocks from the library. When the GRC runs
a model, it generates a Python file in the same directory. This Python file contains the flow
graph which the Python interpreter runs. [7]

20
Chapter 4

Experimental part

4.1 Single Antenna Point to Point System with channel feedback

This is the most basic system we implemented. We have one transmitter and one receiver,
both with one antenna. The basic aspects of our protocol are the following:

• We have 2 distinct USRPs so we denote them as Node A and Node B. Each of these
nodes is a transceiver, that is, it alternates between a Transmitter (Tx) and a Receiver
(Rx) mode.
• When one of these nodes is in Tx mode, it sends N(' 50) packets and then turns to Rx
mode.
• When a node is in Rx mode, it waits until it detects a packet. Once a packet is detected,
after processing, the training sequence is checked and if the Bit Error Rate is zero we
assume that a correct packet has been detected.
• After the correct packet has been detected, if it is node B, the flat fading channel hest
is estimated and coded into data, if it is node A, hest the flat fading channel is decoded
from the received packet and it turns to Rx mode.

21
 Figure 4.1: A point to point system

We assume that CFO correction and time synchronization was carried out at the receiver
so using 2.5.1, we get a complex number h = hr + jhi for the actual value of the channel.
Figure 4.2 shows the steps after channel estimation.

Figure 4.2: Process carried out in Node B

The real and the imaginary part of the channel, hr and hi , are converted into binary
numbers. Then each of the binary numbers is converted to a 4-QAM sequence using Gray
Code. Then both symbol sequences are concatenated and used as a data packet ready to be
transmitted.

Now both nodes’ function has switched from Transmitter to Receiver and vice versa.
Node B now sends the same number of packets as Node A now adding the channel feedback
in the end of the packet.

The process that Node A carries out, once the channel is estimated and eliminated, is the
inverse procedure as shown in Figure 4.3

22
 Figure 4.3: Process carried out in Node A

In order to see how the channels look like in a static environment, we estimated the
channels for 100 consecutive packets. In Figure 4.4, we plot the channel estimates. We
observe that the channel estimates are uniformly distributed on a circle of the complex
plane. This happens because their magnitude does not change due to the static environment.
However, each channel has a random phase, which seems to be uniformly distributed in [0,
2π).

23
Figure 4.4: Channel Estimates

24
4.2 Single to Dual Antenna Point to Point System with channel
feedback

This system is quite similar to the previous one with one difference. Our Receiver, Node B,
has two antennas instead of one. Node A sends again a number of packets and both antennas
in Node B are receiving until they detect a packet. Once packets are sent, Node A turns to
a receiver.

Figure 4.5: Single to Dual Antenna point to point

25
 The received signal after CFO-cancellation and assuming perfect time synchronization for
each node is
     
y1 h1 w
= s+ 1 (4.2.1)
y2 h2 w2
or, in vector notation:
y = hs + w.

Both antennas of Node B are looking for a packet. Once one of them detects a packet
both inputs are being processed. With a similar way as shown previously, Node B estimates
both channels (h1 and h2 ) and are coded as data in order to be sent back to Node A.
Channel feedback can be sent back to Node A from either or both of the antennas of node
B.
Once Node A has the channel estimates (h1 and h2 ) we calculate the ratio of the magnitude
of these channels
max(|h1 |, |h2 |)
c= . (4.2.2)
min(|h1 |, |h2 |)

This ratio shows us how stronger is one channel compared to the other. This ratio can
be quite big, if for instance there is no line of sight between Node A and one of the antennas
on the receiver.
We suppose that we need to transmit different data to each antenna in Node B. So all
future packets will have the data sequence split into two halves, each half for a different
antenna. So we will use this ratio c to compensate for channel’s attenuation in the weak
channel antenna.
We assume |h1 | > |h2 | so we need to amplify the symbols sent to the bottom antenna as
shown in figure 4.5. So the data packet will have the following form

26
 Figure 4.6: Data symbols for both antennas

This way, we manage to make both antennas receive data symbols with the same energy
regardless of the magnitude of each channel.

27
4.3 Single Point to multi Point System with channel feedback

This system assumes one antenna transmitter that needs to receive the channel estimates
from a number of independent nodes.

Figure 4.7: Point to Multi point system

This system can be generalized for a bigger number of nodes. The difference of this
experiment is that the receivers have to send back their feedback seperately. So now for this
system we have to add a header. When Node A receives a packet this header will work as an
identification of the Node that sent its channel feedback. Thus Node A will have to receive
a number of packets (depending on the nodes) in order to receive the channel estimates of
each node.
Once all channel estimates have been received our system follows the same procedure as
in the previous system amplifying appropriately the data packets that correspond to nodes
with smaller channel absolute value.

28
4.4 Interference cancellation

In this experiment we study the interference introduced from another transmitter through
the same frequency band in a simple Point to Point system. Let Tx1 with Rx1 be a commu-
nication pair which needs to communicate without causing interference to Rx2 node.

Figure 4.8: Channel estimation of the receiver that will not receive any future packets

29
 As shown in Figure 4.8 the received signal y2 for Rx2 will be

s11 s21
     
y2,1 w2,1
 y   s12 s22  " #  w 
 2,2    h1  2,2 
y2 =  = .
 ..   .. .
.

 h2 +  .. 
  (4.4.1)
 .   .   . 
y2,n s1n s2n w2,n
or, in vector notation

y2 = Sh + w (4.4.2)
where w is Additive White Gaussian Noise

Using the above equation we need to estimate both channels and then send their estimates
back to Tx1 as feedback. The Least Squares channel estimates can be found by minimizing
the following square error quantity

hLS = argmin||y2 − Sh||22 . (4.4.3)

h

The solution is given by [1]

−1
hLS = S H S S H y2 . (4.4.4)

30
4.4.1 Orthogonal Beamforming

Beamforming is a signal processing technique used for directional signal transmission. In

our case we use the channel feedback in order to cancel the interference of our transmitter
at the specific node. So between Tx1 and Rx2 if we estimate the channel, as showed earlier,
all future packets will pass through the same channel just with a different random phase.
Assuming a static environment, for every future packet we will have

   
h1∗ jφ h1
=e , (4.4.5)
h2∗ h2

where h1 and h2 are the initial channels as shown in Figure 4.8 and h1∗ , h2∗ all future
channels. We need to beamform the transmit signal appropriately in order to arrive at Rx2
very weak. Tx1 will transmit w1 s and Tx2 will transmit w2 s. So we beamform the signal
sent from Tx1 with w1 = h2 and signal sent from Tx2 with w2 = (−h1 ) .
Thus after beamforming Rx2 will receive

y2,n = (h2 )h1∗ sn + (−h1 )h2∗ sn + w2,n (4.4.6)

= h2 h1 ejφ sn + (−h1 )h2 ejφ sn + w2,n
= 0 + w2,n
= w2,n ,

where w2,n is Additive White Gaussian Noise

31
For Rx1 the received signal will be

y1,n = (h2 )h3 sn + (−h1 )h4 sn + w1,n (4.4.7)

= h2 h3 sn − h1 h2 sn + w1,n
= (h2 h3 − h1 h2 )sn + w1,n
= h0 sn + w1,n ,

where w1,n is Additive White Gaussian Noise

Figure 4.9: 2x2 system, Rx1 only receiving data signals

32
 In Figures 4.10 and 4.11 we can see two beamformed packets received from Rx2 and Rx1
respectively.

Figure 4.10: Received packet in Rx2

Figure 4.11: Received packet in Rx1

33
 In this experiment, we observe that indeed the packet received at Rx2 is utterly attenuated
compared with the packet received from Rx1 .
This way we have managed to establish a communication between Tx1 and Rx1 without
Rx2 receiving any data. Exploiting the channel’s feedback sent, even though both receivers
are tuned in the same frequency just one of them receives data. This can be generalized for
a greater number of receivers and transmitters that communicate over the same frequency
bandwidth. More specifically, we can achieve interference cancellation at n − 1 receiving
antennas with n transmit antennas. [5]

34
Chapter 5

Conclusion

The importance of multi-carrier communication systems has been established in the present
communication era. Multi-carrier communication systems, especially OFDM, have evolved
as one of such potential candidates which are bandwidth efficient and robust to multipath
channel conditions. In this thesis, a number of experiments were conducted in an SDR
testbed using OFDM. In many cases, knowledge of the channel at the transmitter can
be useful thus channel feedback was exploited throughout the experiments. A number of
transceivers were used in order to make different communication systems using the same
hardware and loading different programmes.

35
Bibliography

[1] S. M. Kay, ”Fundamentals of Statistical Signal Processing: Estimation Theory”, Prentice-

Hall, 1998, Chapter 8.
[2] Ch. N. Kishore and V. U. Reddy,”A Frame Synchronization and Frequency Offset Esti-
mation Algorithm for OFDM System and its Analysis,” EURASIP journal on Wireless
Communications and Networking.
[3] Athanasios P.Liavas, Lectures of Telecommunication Systems II 2010.
[4] USRP-Ettus Research, a National Instruments Company.
[5] Nikolaos Sapountzis, Bachelor Thesis, ”Interference Cancellation in Wireless Communi-
cations,” 2013.
[6] Ioannis Kardaras, Master thesis ”Software-Defined Radio Implementation of an OFDM
Link,” 2010.
[7] Omair Sarwar, Master Thesis ”Software Defined Radio (SDR) for Deep Space Commu-
nication,” 2013.
[8] Li, Ye, Gordon L. (Eds.), Orthogonal Frequency Division Multiplexing for Wireless Com-
munications, Springer, 2006.

36