Software Defined Radio for Medical

Applications
Master’s thesis in Master Biomedical Engineering

LAURA GUERRERO

Department of Electrical Engineering

C HALMERS U NIVERSITY OF T ECHNOLOGY
Gothenburg, Sweden 2019
 Master’s thesis 2019:EENX30

Software Defined Radio for Medical Applications

LAURA GUERRERO

Department Electrical Engineering

Division of Signal Processing and Biomedical Engineering
Biomedical Electromagnetics
Chalmers University of Technology
Gothenburg, Sweden 2019
Software Defined Radio for Medical Applications

Laura Guerrero

© LAURA GUERRERO, 2019.

Supervisor: Xuezhi Zeng, Signals and systems

Examiner: Andreas Fhagel, Signal and systems

Master’s Thesis 2019:EENX30

Department of Electrical Engineering
Division of Signal Processing and Biomedical Engineering
Biomedical electromagnetics group
Chalmers University of Technology
SE-412 96 Gothenburg
Telephone +46 31 772 1000

Cover: USRP2901 Software Defined Radio Device

Gothenburg, Sweden 2019

iv
Software Defined Radio for Medical Applications
Laura Guerrero
Department of Electrical Engineering
Chalmers University of Technology

Abstract
A portable, cheap and accurate system is necessary for pre-hospital diagnosis, and
recently software defined radio technology (SDR), has attracted the interest of re-
searchers for these same reasons. The objective of this master thesis is to evaluate
the feasibility of using SDR technologies for Medical applications. In this project,
the parameters that affect the measurement quality were studied. Furthermore, cal-
ibration strategies for random phase offset between the transmitter and the receiver
were successfully developed and evaluated. Then the measurement repeatability
and accuracy of the system were evaluated, the results show that the calibration
system made the measurements repeatable, making the system practical for medical
monitoring. On the other hand, in order to achieve higher accuracy, the calibration
methods need to be further developed. Additionally, the possibility of using the
SDR for other medical application was explored, the results show that it can be
effectively used in non-contact vital sign monitoring.

Keywords: SDR,medical applications, vital sign detection,measurement repeatabil-

ity, measurement accuracy,medical diagnosis.

v
Acknowledgements
I would like to express my deepest gratitude to my supervisor, Xuezhi Zeng, thank
you for so patiently answering my questions and for guiding me in right direction.
It would have been impossible to accomplish this work without your warm and kind
help. You are a role model.
I would like to thank my parents and my brother for their support and kindness,
I would have not been able to accomplish my goals without the constant motivation
you gave me.
My special thanks to my friends from the Study Room: Asta, Mauricio, Isabel,
Berglind, Tryggvi, Alberto, Ryan, Abdul and Alex. You created an unforgettable
time in my life and I will be forever thankful for the support I received from you.

Laura Guerrero, Gothenburg, June 2019

vii
 Contents

List of Figures xi

1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Microwave Imaging Approaches . . . . . . . . . . . . . . . . . . . . . 1
1.3 Current Imaging System Setups . . . . . . . . . . . . . . . . . . . . . 2
1.4 Motivation and Project aim . . . . . . . . . . . . . . . . . . . . . . . 2
1.5 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Microwave Measurement for Medical Diagnostics 5

2.1 Microwave Measurement for Medical Diagnostics . . . . . . . . . . . . 5
2.2 Measurement Technology . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Time Domain Measurements . . . . . . . . . . . . . . . . . . . 6
2.2.1.1 Pulse Generator . . . . . . . . . . . . . . . . . . . . 6
2.2.1.2 Real Time/Sampling Oscilloscope . . . . . . . . . . . 7
2.2.2 Frequency Domain Measurements . . . . . . . . . . . . . . . . 8
2.2.2.1 Frequency Modulated Continuous Wave (FMCW) . . 8
2.2.2.2 Stepped Frequency Continuous Wave (SFCW) . . . . 9

3 Software Defined RF System 11

3.1 Advantages of SDR Systems . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 USRP 2901 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2.1 Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2.1.1 Receiver and Transmitter Path . . . . . . . . . . . . 12
3.2.1.2 Internal RF Switch . . . . . . . . . . . . . . . . . . . 14
3.2.1.3 RFPLL Synthesizers . . . . . . . . . . . . . . . . . . 14

4 Fundamental Test with the SDR 15

4.1 Power and Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 IQ Rate and Noise Level . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3 Measurement Repeatability . . . . . . . . . . . . . . . . . . . . . . . 19
4.4 Calibration Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.4.1 Strategy 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.4.2 Strategy 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.4.3 Strategy 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

ix
Contents

5 Real Measurements 29
5.1 Initial accuracy test . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Antenna measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

6 Other Applications 37
6.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

7 Conclusion 41
7.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

Bibliography 43

A Appendix 1 I
A.1 LO functionality block . . . . . . . . . . . . . . . . . . . . . . . . . . I

x
 List of Figures

2.1 Measurement configuration for medical diagnostics. . . . . . . . . . . 5

2.2 Time domain measurement system. . . . . . . . . . . . . . . . . . . . 6
2.3 Sequential sampling oscilloscope. . . . . . . . . . . . . . . . . . . . . 8
2.4 Frequency modulated continuous wave radar. . . . . . . . . . . . . . . 9
2.5 Frequency vs time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.1 Block diagram of USRP [34]. . . . . . . . . . . . . . . . . . . . . . . . 12

3.2 Block diagram of the USRP 2901 [37]. . . . . . . . . . . . . . . . . . 13

4.1 Basic measurements setting. . . . . . . . . . . . . . . . . . . . . . . . 15

4.2 Magnitude of signal received from 10dB RF attenuator. . . . . . . . . 16
4.3 Comparison of the gain behavior depending on the frequency. . . . . 16
4.4 Power of received signal when using 50 dB gain. . . . . . . . . . . . . 16
4.5 Noise level depending on the gain. . . . . . . . . . . . . . . . . . . . . 17
4.6 Magnitude of signal and noise level depending on IQ rate. . . . . . . 18
4.7 IQ rate and power spectrum. . . . . . . . . . . . . . . . . . . . . . . . 18
4.8 Received magnitude and phase in a 100 repetitions . . . . . . . . . . 19
4.9 Measuremets setup used in strategy 1. . . . . . . . . . . . . . . . . . 20
4.10 Magnitude and phases obtained with strategy 1 (not averaged data). 22
4.11 Magnitude and phases obtained with strategy 1 averaged over 5000
samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.12 Mean and standard deviation of magnitude in strategy 1 of averaged
and not averaged data. . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.13 Mean and standard deviation of phase in strategy 1 of averaged and
not averaged data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
4.14 Measurement setup used in strategy 2. . . . . . . . . . . . . . . . . . 24
4.15 Magnitude and phases obtained with strategy 2 (not averaged data). 26
4.16 Magnitude and phases obtained with strategy 2 averaged over 1200
samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.17 Measurement setup used in strategy 3 . . . . . . . . . . . . . . . . . . 27
4.18 Magnitude and phases obtained with strategy 3 (not averaged data). 28
4.19 Magnitude and phases obtained with strategy 3 averaged over 1200
samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5.1 Measurement setup using the 6 dB attenuator as object of interest . . 29

5.2 Measurement setup using the 10 dB attenuator as object of interest . 29

xi
List of Figures

5.3 Magnitude and phase of difference between the transmission coeffi-

cients using 6 dB and 10 dB attenuators as objects . . . . . . . . . . 30
5.4 Measurement setup using no object (cables) . . . . . . . . . . . . . . 30
5.5 Measurement using 6 dB as object with three different calibration
coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.6 Measurement using 10 dB as object with three different calibration
coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.7 Picture of imaging tank used for measurements . . . . . . . . . . . . 33
5.8 Adjacent antenna pair response obtained with SDR and VNA . . . . 33
5.9 Phase of adjacent antenna pair response obtained with SDR and VNA 34
5.10 Most distant antenna pair response obtained with SDR and VNA . . 34
5.11 Phase of most distant antenna pair response obtained with SDR and
VNA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.12 Measured reflection coefficient at the RF ports of the SDR board . . 36

6.1 Basic test scenario for vital sign detection. . . . . . . . . . . . . . . . 37

6.2 Received signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.3 Signal After respiration filer in Time domain and Frequency domain. 39
6.4 Auto-correlation of respiration. . . . . . . . . . . . . . . . . . . . . . 39
6.5 Signal After Heartbeat filer in Time domain and Frequency domain . 40

A.1 Sinusoid signal with 0Hz and amplitude 0,7 . . . . . . . . . . . . . . II

A.2 Received signal while not configuring the LO frequency . . . . . . . . II
A.3 Received signal while not configuring the LO frequency . . . . . . . . II

xii
 1
Introduction

In this section, Ultra Wide Band (UWB) microwave imaging is introduced including
a background and current imaging methods.

1.1 Background
Traditional imaging systems used for diagnostic, like Magnetic Resonance Imaging
(MRI) and X-ray Computed Tomography (CT), are powerful but normally only
available in hospitals. Conditions affecting circulation and pressure in the brain,
like stroke, are life threatening, and in these cases the sooner treatment is applied,
to restore the circulation and pressure, the greater the likelihood of recovery. This
is the reason why a pre-hospital diagnostic system is necessary.
The rapid development of microwave integrated circuit technology has allowed
the size and cost of microwave electronics to be dramatically reduced, so microwave-
based devices are particularly suitable for pre-hospital diagnostic systems [1].
Microwave for biomedical imaging has been broadly studied in the past years
for stroke, trauma and breast cancer detection [1–4]. Microwave diagnostics relies
on the existence of dielectric contrast between healthy tissues and lesions. Often we
consider only permittivity and conductivity and ignore permeability, as biological
tissue is in general non-magnetic.
Several studies have found that the permittivity of cancerous breast tissue is
higher than normal breast tissue and also higher than non-malignant breast tumors
[5, 6]. Other studies have analyzed the dielectric properties of the brain and found
a difference in permittivity and conductivity between white and gray matter [7].
Moreover, it has been found that the dielectric properties of brain tissue declines
with age [8, 9].
Normally the tissues studied are healthy tissues but diseases and injuries might
lead to changes in dielectric properties [1]. An example of such change is the co-
agulation of blood while bleeding. Another example is the case during an ischemic
stroke where the dielectric properties will change due to loss of circulation [10].

1.2 Microwave Imaging Approaches

In microwave imaging techniques, an electromagnetic wave is used to illuminate the
object that is being tested. Scattering will happen because of natural dielectric
variations of the tissue. Two approaches can be taken: Tomographic methods and

1
1. Introduction

radar-based techniques [3]. In tomographic imaging methods the measured scatter-

ing data is used to quantitatively reconstruct spatial representation of the dielectric
profile of the tissue.
A nonlinear Reconstruction algorithm, such as the conjugate gradient algorithm
is used [4, 11]. The algorithm’s objective is to minimize a penalty function of the
difference between the measured signal and a simulated signal, obtained by using the
finite-difference time domain method or finite element method [12]. This minimiza-
tion is accomplished by iteractively updating the dielectric parameters in the model.
Since these algorithms can be computationally demanding, reconstruction times can
become unfeasible. For 2-D imaging the reconstruction times are reasonable but for
3-D reconstruction reconstruction times can be several hours in supercomputers.
In radar-based imaging method, an UWB signal illuminates the object from
different directions. Unlike Tomography, where the objective is to get a quantita-
tive reconstruct of a spatial representation of the dielectric profile of the scattering
targets; the objective of radar imaging is to use the amplitude and echo time to
detect and position strong scatterers, such as tumors by using simple delay-and sum
algorithms [13].
This reconstruction strategy is similar to a global positioning receiver, where
the position is determined by triangulating using propagation times from different
directions [1].

1.3 Current Imaging System Setups

The measurements of UWB data for microwave imaging, can be carried out in the
frequency domain or the time domain. Microwave imaging system are normally con-
structed using lab intruments. Many experimental systems for biomedical imaging
use frequency domain measurement systems, and usually they are based on Vector
Network Analyzers (VNA) [4, 11, 12, 14]. The VNA is a powerful measurement plat-
form that provides a high accurate measurement solution in experimental uses. The
VNA uses the stepped frequency modulations technique to perform measurements in
wide frequency bands, in each frequency step, a sinusoidal signal will illuminate the
object, the scattered signals are received by the VNA and the scattering parameters
will be calculated.
There are also experimental setups working in time domain consisting of a pulse
generator and an real-time or sampling oscilloscope [15–18].The pulse generator
is used to illuminate the object and the oscilloscope measures the time domain
response.

1.4 Motivation and Project aim

Among all the microwave systems that have been investigated for medical imag-
ing, most of them are based on lab-instrument, which are bulky and expensive. A
portable, cheap and accurate system is crucial for promoting the development of
microwave technology, especially for pre-hospital applications.

2
 1. Introduction

Most recently, a new type of system that has been mainly employed for radio
communications has attracted interest from researchers in this field [19, 20]. This
system is based on software defined radio technology (SDR) and the whole system is
built on a single circuit board which consists of a radio frequency integrated circuit
(RFIC) transceiver and a digital signal processing (DSP) unit.
The aim of this master thesis project is to investigate a compact and cheap
SDR board for medical applications. The main objectives include:
• Configure the board for required measurement.
• Investigate parameters that affect the measurement quality.
• Develop and evaluate calibration strategies for random phase offset between
the transmitter and the receiver.
• Investigate measurement repeatability and accuracy for medical diagnosis.
• Study the possibility of using the SDR for other medical applications.

1.5 Thesis Outline

The main content of the thesis is summarized here: Chapter 1 gives an overall
introduction to the topic and the project motivation and aim. Chapter 2, gives
an overview of microwave imaging for medical applications. Chapter 3, introduces
conventional UWB measurement technologies. In Chapter 4, the parameters that
affect the measurement quality are studied, calibration strategies are developed and
evaluated, and the repeatability of the system is evaluated. In Chapter 5, the
accuracy of the measurements is evaluated. In Chapter 6, the possibility of using
the SDR for other medical applications is explored.

3
1. Introduction

4
 2
Microwave Measurement for
Medical Diagnostics

In this chapter an introduction to microwave measurements for medical diagnosis is

presented. Additionally, two conventional technologies for utra wide band (UWB)
measurements are introduced: time domain and frequency domain technologies. The
architecture of each system are presented and the measurement strategies described
in detail.

2.1 Microwave Measurement for Medical Diag-

nostics
A common antenna configuration scenario is shown in Figure 2.1. The antennas
are placed around the object that is going to be tested and each antenna transmits
a waveform that will help illuminate the object. The scattered waveform will be
received by antennas and subsequently measured.

Figure 2.1: Measurement configuration for medical diagnostics.

The choice of frequency band is generally a trade-off between penetration depth

and imaging resolution. A higher frequency can result in a better resolution, but
experience greater attenuation when penetrating to human tissue, which are very
lossy. In order to achieve a good resolution and a sufficient penetration depth, a
wide frequency spectrum is needed. This is the reason why there is interest in using

5
2. Microwave Measurement for Medical Diagnostics

UWB technology in medical applications. Specifically, a system may be classified

as an UWB system when its operating frequency range is wider than 500 MHz
or 20 fractional bandwidth [21]. With this technique the higher frequencies are
used to increase the resolution in the reconstructed image, allowing to resolve finer
structures, and the lower frequencies will allow the imaging of larger structures
and high contrast objects [3]. In microwave tomography, the lower frequencies are
necessary as the initial guess for proceeding iteration, since scattered data is non-
linearly related to the scattering target and the higher the frequency, the higher is
the non-linearity [4].

2.2 Measurement Technology

The UWB microwave measurements systems for medical diagnostics are usually built
from instruments found in laboratories. The measurements can be divided in two
strategies: time domain measurements, using a pulse generator and an oscilloscope
[15] and frequency domain measurements using a VNA.

2.2.1 Time Domain Measurements

Time domain measurements system consist of a pulse generator (a step source) and a
real-time or sampling oscilloscope to measure the scattered signal, Figure 2.2 shows
an example of a time domain measurement system used for experiments with breast
phantom imaging.

Figure 2.2: Time domain measurement system.

2.2.1.1 Pulse Generator

The pulse generator is fundamental in a time domain UWB imaging system. Pulse
generators provide a number of pulse shapes, the pulse shapes can be classified into
the following categories: rectangular, gaussian impulse, gaussian monocycle pulse
(first derivative of the gaussian impulse signal [21]), and step [16]. The spectral
energy will be defined by the pulse shape.

6
 2. Microwave Measurement for Medical Diagnostics

The classical pulse generation techniques are discrete RF transistor, avalanche

transistor, step-recovery diode (SRD) and tunnel diode [17]. Nonlinear transmission
lines (NLTLs) and analog/digital monolithic microwave integrated Circuits (MMIC)
can be used to build ultra-fast risetime and high amplitude pulse generators [22].
According to time-frequency relation, the bandwidth required for microwave medical
imaging corresponds to a pulse with duration less than hundred picoseconds (ps) [17].
To generate picosecond pulses, SRD and NLTLs are frequently used [23, 24].
SRD is used in many pulse generator designs for UWB applications, since it presents
a simple, compact and low-cost solution [18, 25].

2.2.1.2 Real Time/Sampling Oscilloscope

Modern digital oscilloscopes can be divided in two categories depending on how they
operate: real-time oscilloscopes and sampling (or equivalent time) oscilloscopes [26].
A real-time sampling oscilloscope captures an entire waveform on each trigger
event, for this reason a large number of data points must be captured in one con-
tinuous record. Generally this type of system uses interleaving of multiple analog-
to-digital converters (ADCs) in parallel, in order to increase the converters effective
sample rate [27]. Normally real-time oscilloscopes have high input impedance and
are devised to measure voltages inside electrical circuits operating in real time. The
bandwidth of the oscilloscopes is limited to 500 MHz [26].
In the equivalent time sampling oscilloscope, the effective sample rate of the
ADC can be extended beyond the nominal sampling rate of the ADC [17]. Instead of
dealing with the whole signal in real time, the equivalent time sampling oscilloscope
collects samples of the instantaneous voltage of the input signal on successive samples
and will construct the samples to form a picture of the waveform [28]. There are
two types of equivalent sampling: sequential sampling and random sampling.
A detailed description of random sampling can be found in [29]. In sequential
sampling oscilloscopes, the stroboscopic technique is used. The stroboscopic tech-
nique is the one used to slow down the motion (or frequency) of events which are
too fast to observe by conventional means [29]. In other words a slow replica of the
signal will be assembled from individual measurements of many repetitions of the
signal (as shown in Figure 2.3).

7
2. Microwave Measurement for Medical Diagnostics

Figure 2.3: Sequential sampling oscilloscope.

In the first measurements cycle the signal is measured at the starting point
T = T0 . Then a sequential delay (∆t) is added at each cycle. In the second
measurement cycle, the amplitude is measured at time T = T0 + ∆t, the third
measurement cycle is measured at time T = T0 + 2∆t. This process will be repeated
until the entire pulse length is sampled.
Sampling oscilloscopes have a higher accuracy than real time oscilloscopes, since
they have a higher effective sampling than the real time oscilloscope, meaning that
they have smaller aliasing errors. However, the measurement time is determined by
the sample rate and the number of sample points, and sampling oscilloscopes recon-
struct the waveform from many repetitive measurements. Consequently, sampling
oscilloscopes have higher measurement accuracy at the expense of measurement
speed [17].

2.2.2 Frequency Domain Measurements

The frequency domain measurements can be divided in two strategies: frequency
modulated continuous wave (FMCW) [30], and stepped frequency continuous wave
(SFCW) [19].

2.2.2.1 Frequency Modulated Continuous Wave (FMCW)

In a FMCW radar a chirp signal is used, a chirp signal is a signal in which the
frequency increases or decreases with time (frequency ramp). The chirp signal is
generated with a constant amplitude sinusoidal signal whose frequency is linearly
swept in a band over a specific time, the signal generated can be expressed with the
Equation 2.1 [31]
!
f1 − f0 2
s(t) = cos 2π(f0 t + t) , (2.1)
2T
Where f0 is the starting frequency, f1 is the final frequency and T is the time
it takes to sweep from f0 to f1 .
In a traditional FMCW radar, the chirp is transmitted, it propagates at the
speed of light and will hit the target after a propagation delay, the signal then

8
 2. Microwave Measurement for Medical Diagnostics

bounces back and is received by the receiving antenna after another time delay. It
is then mixed with the instantaneous chirp signal, and down-converted, the down-
converted signal will have a single tone fd which is the beat frequency resulting from
the time delay between the transmitted and returned signals.

Figure 2.4: Frequency modulated continuous wave radar.

The range of a target can be determined from Equation 2.2:

vfd
R= , (2.2)
2m
where m is the rate of the frequency sweeping.
The drawback of this technique is that it is difficult to achieve an accurate and
constant frequency’s sweeping rate over a wide band given the non-linearity of the
synthesizer, specially when a voltage-controlled oscillator (VCO) is used instead [32].

2.2.2.2 Stepped Frequency Continuous Wave (SFCW)

SFCW radars transmit successive trains of sinusoidal signals at different frequencies,
as shown in Figure 2.5. The transmitted frequencies are separated by a constant
frequency step f and the total bandwidth B is N*f where N is the number of steps.
PRI is the transmission duration of each sinusoidal signal and the total time to
complete the frequency sweep is N PRI [34]. The received signal at each frequency is
then down-converted to a intermediate-frequency (IF) signal, which is then further
converted into the base-band in-phase (I) and in quadrature (Q) signals, which
contain the amplitude and phase information of the targets. Then the I/Q signals
are converted into digital signals with an analog-to-digital converter (ADC). The
acquired digital data over a wideband is then transformed into a "synthetic pulse"
in time domain by using the Inverse Discrete Fourier Transform (IDFT), this pulse
depicts the target and can be processed to reveal the target’s features [32].

9
2. Microwave Measurement for Medical Diagnostics

Figure 2.5: Frequency vs time.

This is the measurement strategy employed in VNAs. With VNA measurement,

the object response at each frequency is expressed as scattering parameters, which
are the ratio of received wave to transmitted wave.

10
 3
Software Defined RF System

As mentioned in Section 2, most of the microwave based medical diagnostic systems

are constructed with bulky and expensive laboratory instruments. This is useful for
experimental use, however, for pre-hospital diagnosis a small, cheap and accurate
system is necessary. For this reason, SDR technology, mainly employed for radio
communications, has attracted attention in this field. This chapter gives a intro-
duction of the explored SDR board, USRP2901. The block diagram and functions
are described in detail.

3.1 Advantages of SDR Systems

The traditional hardware based radio devices can become limiting for cross function-
ality and encounter the problem that they can only be modified through physical
intervention. This can result in an increased production cost and reduced flexibility
in support of multiple waveform standards [33]. The advantage of using SDR is
that it presents an efficient and inexpensive solution to all the problems of the tra-
ditional hardware based radios. In the SDR the physical layer processing is carried
out through adaptable software operating on programmable processing technologies,
this includes the field programmable gate arrays (FPGA), digital signal processor
(DSP), general purpose processors (GPP), programmable system in chip (SoC). This
allows new features to be added to existing radio systems without requiring adding
new hardware.

3.2 USRP 2901

The National Instruments universal software radio peripheral (USRP) devices are
wideband SDRs which can be used for different RF applications. As shown in Figure
3.1, this board mainly includes a wide band RF transceiver and a FPGA, which are
respectively an analog measurement platform and a digital processing unit. The
investigated board has a frequency range of 70 MHz to 6 GHz, gain range of 76 dB,
and maximum instantaneous real-time bandwidth is of 56 MHz. The transmitter
has a maximum output power of 20 dBm and the receiver has a maximum input
power of -15 dBm.

11
3. Software Defined RF System

Figure 3.1: Block diagram of USRP [34].

3.2.1 Block Diagram

The block diagram of the board is presented in Figure 3.2. The board is composed
of two channels and each channel has two ports, one of the ports can act as either
receiver or transmitter and is labeled Tx/Rx1 and the other port labeled Rx2 is a
receiving only port.

3.2.1.1 Receiver and Transmitter Path

The transmit path starts with the host computer, it synthesizes base-band IQ signals
and transmits the signals to the device over a USB 3.0 or USB 2.0 connection.
Then the digital up-converter (DUC) mixes, filters, and interpolates the I and
Q signals to 61.44 MS/s. The Spartan 6 FPGA provides digital up-conversion
functionality in the board, which includes fine-frequency tuning and several filters
for decimation.
Next the digital-to-analog converter (DAC) converts the data from the I and
Q channels. The AD9631 transceiver includes two fully programmable 128-tap FIR
filter with interpolation options. The FIR output is sent to a series of interpolation
filters that provide additional filtering and data rate interpolation prior to reaching
the DAC.
The baseband I and Q signals can be represented with equations 3.1 and 3.2 [35]

I = A cos(θ) (3.1)
Q = A sin(θ) (3.2)
The mixer up-converts the signals to an RF frequency. The PLL controls the
VCO to make sure that the device clock and LO can be frequency-locked to a
reference signal.
After mixing, the signal becomes [32, 36]:

x = A cos(wt + θ) = I cos(wt) + Q sin(wt), (3.3)

12
 3. Software Defined RF System

where w = 2πf and w, θ and A are the RF frequency, phase and amplitude of the
transmitted signal.
The received signal can be expressed with Equation 3.4

r = B cos(w(t − τ ) + θ), (3.4)

where B is the amplitude of the received signal and τ the elapsed time between
transmission and reception.
The receiver chain begins with a low-noise and drive amplifier to amplify the
incoming signal from the port. Next the signal is down-converted by the mixer to
the baseband or IF in-phase (I) and quadrature (Q) components. Following this,
the I and Q data is digitized by a 12 bit analog-to-digital converter (ADC). Then
the FPGA digital down-converter (DDC) mixes, filters, and decimates the signal to
a user-specified rate. After decimation, the data is streamed to a host computer
over a USB 3.0 or USB 2.0 connection.

Figure 3.2: Block diagram of the USRP 2901 [37].

13
3. Software Defined RF System

3.2.1.2 Internal RF Switch

As it can be seen in Figure 3.2, the RF switches are located after each port. They
have two different positions allowing for transmit and receive operations to occur on
the same shared port. Having the RF switch presents the advantage of increasing
the number of receivers available. It has the disadvantage that the transmitting and
receiving port are connected, meaning that there will be greater leakage between
ports.

3.2.1.3 RFPLL Synthesizers

The function of the synthesizer is to generate a range of frequencies from a single
reference frequency, a carrier frequency to move up the base band signal to a desired
RF signal. The transceiver contains two independent RFPLL synthesizers. The PLL
synthesizers are of a fractional-N architecture with completely integrated VCO and
loop filters.

14
 4
Fundamental Test with the SDR

In this chapter, several fundamental test were performed in order to gain more
understanding of the system. Particularly, the measurement repeatability was in-
vestigated.

4.1 Power and Gain

As mentioned in Section 3, the receiver has a maximum input power of -15 dBm, it
is important that the received power does not go over this limit, to avoid damaging
the device during tests. In addition, the good knowledge of power level at different
frequencies is the prerequisite for achieving the best measurement quality.
The setup used, for investigating the power, is shown in Figure 4.1. The trans-
mitter/receiver in channel zero (Tx/Rx1) is connected to the transmitter/receiver
in channel one (also labeled Tx/Rx1), and a 10 dB RF attenuator is used to protect
the receiver from a too high input power.

Figure 4.1: Basic measurements setting.

Figure 4.2 displays the magnitude of the received signals at each frequency in
the range 500 MHz-6 GHz. It can be seen that the power of the received signal
decreases with the frequency, and in this particular image the power drops from
-34dB to -50dB in the frequency range 500 MHz-3 GHz, having a -15 dB drop. This
is an expected behavior in the device, given the received signal gain changes shown
in the specifications of the AD9361 [38].
To improve the measurement quality of weak signals, the gain in the amplifiers
must be increased accordingly. However, this may result in the saturation of stronger
signals. This is the reason why, a detailed description of the saturation point of the
received signals is needed. It was decided that the gain variations would be done on
the receiver’s amplifier, to avoid damaging the device on the test by exceeding the
maximum input power in the receiver.

15
4. Fundamental Test with the SDR

Figure 4.2: Magnitude of signal received from 10dB RF attenuator.

A simple evaluation is done to compare the point of saturation of each frequency

band, the test was done with 8 different frequencies (500 MHz, 1 GHz, 1.5 GHz, 2
GHz, 2.5 GHz, 3 GHz, 3.5 GHz and 4 GHz), the received power level for different
gain settings were shown in Figure 4.3.

Figure 4.3: Comparison of the gain behavior depending on the frequency.

It can be seen that the saturation occurs at the later position (a higher gain)
for a higher frequency, which can be expected from the data shown in Figure 4.2.
Figure 4.4 shows an example of a saturated signal. In this example, the gain in the
receiver was set to 50 dB and the gain in the transmitter was set to 0 dB.

Figure 4.4: Power of received signal when using 50 dB gain.

It is important to point out the noise floor also increases with the gain, as
shown in Figure 4.5. However, the increase of the noise floor with the gain is less
than that of the power, which consequently resulting in a high signal to noise ratio

16
 4. Fundamental Test with the SDR

(SNR). This conclusion however is only valid when the system operates in the linear
region.

Figure 4.5: Noise level depending on the gain.

4.2 IQ Rate and Noise Level

UWB measurements performed in frequency domain systems require doing a fre-
quency sweeping of a large number of discrete frequency points, which can end up
being time consuming. Since there is an increased demand on measurement speed,
it is important to study the parameters that will have an effect on the measurement
speed. The IQ rate will define how fast samples are taken, so it is important to
measure the response from the received signal when this parameter is changed.
In the test, a CW signal at 915 MHz was studied. Both the transmitter and
receiver gain were set to 0 dB. Figure 4.6 show the magnitude of the signal and
the magnitude of the floor noise level for different IQ settings.The magnitude of the
signal will stay stable before reaching an IQ rate higher than 50MHz,where there is
a small magnitude drop from -35dB to -38dB. On the contrary, the magnitude of the
noise floor will increase with the IQ rate. Therefore, the SNR of the signal decreases
as the IQ rate increases. When the IQ rate increases, the duration of acquisition
for the measurement will decrease in time. This means that the frequency bin will
increase in width, so the noise displayed increases. An example of the difference the
IQ rate generates on the signals is shown in Figure 4.7. The Figure displays the IQ
data and power spectrum of two different signals with IQ rate 3.5MHz and IQ rate
60MHz. It can be observed that the noise level on the signal will change according
to Figure 4.6.

17
4. Fundamental Test with the SDR

(a) Magnitude of received signal.

(b) Noise level of received signal.

Figure 4.6: Magnitude of signal and noise level depending on IQ rate.

(a) IQ data (3.5MHz). (b) Power spectrum (3.5MHz).

(c) IQ data (60MHz). (d) Power spectrum (60MHz).

Figure 4.7: IQ rate and power spectrum.

18
 4. Fundamental Test with the SDR

4.3 Measurement Repeatability

The repeatibility is how much the results of successive measurements will agree, if
the same test is done in the same measurement setup. The generation of repeatable
experiments is essential in the microwave diagnosis and microwave imaging field. An
example is when monitoring is implemented, the repeatability is important, since it
is crucial to differentiate between the differences occurring due to object changes,
and differences occurring due to the measurement setup.
A fundamental test was performed to investigate the measurement repeatability,
and the measurement setup is the same as in Figure 4.1. In the test, a CW signal at
915 MHz was studied. Both the transmitter and receiver gain were set to 0 dB. The
I/Q rate in both transmitter and receiver is set to 500kHz. The measurement was
done 100 times with the same setup, and the magnitude and phase of the received
signal for these 100 measurements were given in Figure 4.8

(a) Magnitude

(b) Phase

Figure 4.8: Received magnitude and phase in a 100 repetitions

It can be seen that both the magnitude and phase varies from run to run. While
the magnitude changes in a small range, the phase variation is large and completely
random. This is due to the internal architecture of the device, as shown in Section
3. In the device, the transmitter and receiver have different PLLs and the phase of
the generated clocks are therefore not aligned, resulting a random phase offset from
run to run. Therefore, an effective strategy to calibrate the random phase offset is
needed.

19
4. Fundamental Test with the SDR

4.4 Calibration Strategies

To address this repeatibility problem, three calibration strategies are proposed and
evaluated in this section. In each calibration strategy, two attenuators (10dB and
6dB) were used, one of them acting as the object and the other acting as the reference
line. Since the goal of the calibration is to make the measurement repeatable,
the experiment will be repeated many times with the same setup to evaluate the
performance of each calibration strategy.

4.4.1 Strategy 1
In this strategy, the idea is to measure the object and the reference with the same
physical transmitter and the same physical receiver. In section 3, it was explained
that there is one physical receiver in each channel and the internal RF switch can
be used to switch between port Tx/Rx1 (that can act as both transmitting and
receiving port), and port Rx2 (that is a receiving only port).
The strategy is shown in Figure 4.9. The transmitter in channel 0 is connected
to a 3 db power divider, and each of the output ports of the power divider is
connected to an attenuator which is further connected to a port in channel 1. With
this strategy, the receiver will first be connected to the port Tx/Rx1, measuring
signal attenuated by the 10 dB attenuator, then the receiver will switch to the Rx2
port, receiving signal coming from the 6 dB RF attenuator.

Figure 4.9: Measuremets setup used in strategy 1.

In each test, a frequency sweep was done in the range 500 MHz - 6 GHz,using
steps of 110 MHz (50 frequency points). The transmitter sent 16384 sample points,
and the two receiving ports (connected to object and reference line), collected 8192
samples each. Both TxGain and RxGain are set 0 dB and the IQ rate is 500 kHz
for the transmitter and the receiver.
The test was repeated 20 times, Figure 4.10 shows the magnitude and phase
of the relative measurement, which is the ratio of the object data (10 dB) to the
reference data (6 dB). No averaging was applied to the measurement data. Figure
4.11 displays the same 20 tests with the data averaged over 5000 samples.
It is shown that both the magnitudes and phases of the relative measurement
are stable from run to run, indicating a good repeatability. It can also be seen that

20
 4. Fundamental Test with the SDR

the measurement uncertainty is bigger in the non-averaged case than in the averaged
case, and the measurement uncertainty increases in higher frequencies. This can be
explained by the power drop experienced in the received signals (Figure 4.2) in
higher frequencies, since the signals become much smaller compared to the size of
the noise.
The mean value and the standard deviation of the magnitude of the received
signals in each frequency is show in Figure 4.12. It can be seen that the SNR of the
signal increased 20 dB when averaging was done on the signals. Likewise the mean
value and variance of the phases are shown in Figure 4.13. The standard deviation
of the phase of the not averaged data can get values as high as 15 degrees in higher
frequencies. On the other hand the standard deviation of the averaged data is under
0.2 deg for all frequency points.
The deviation is much smaller in the averaged samples than the non-averaged
samples; furthermore, the size of the random noise is bigger in higher frequencies
compared to the magnitude of the signal. The averaging of the measurements helped
reduce its effect on the data. Since the averaging of the samples improved the mea-
surement uncertainty at all frequencies, the conclusion reached is that this strategy
is limited by the size of the random error.
Although this strategy succeeded in stabilizing the magnitude and phase in
each frequency point, there are some disadvantages. As the receiver needs to be
sequentially switched to two receiving ports during the measurement, the received
data includes a transition part which needs to be excluded before the data processing.
Furthermore, the received sample points need to be splitted between each receiving
session.

21
4. Fundamental Test with the SDR

(a) Magnitude of 20 measurements (no averaging).

(b) Phase of 20 measurements (no averaging).

Figure 4.10: Magnitude and phases obtained with strategy 1 (not averaged data).

(a) Magnitude of 20 measurements (averaged over 5000 samples).

(b) Phase of 20 measurements (averaged over 5000 samples).

Figure 4.11: Magnitude and phases obtained with strategy 1 averaged over 5000
samples.

22
 4. Fundamental Test with the SDR

(a) Mean value of magnitude of 20 measurements.

(b) Standard deviation of magnitude of 20 measurements.

Figure 4.12: Mean and standard deviation of magnitude in strategy 1 of averaged

and not averaged data.

(a) Mean value of the phase of 20 measurements.

(b) Standard deviation of the phase of 20 measurements.

Figure 4.13: Mean and standard deviation of phase in strategy 1 of averaged and
not averaged data.

23
4. Fundamental Test with the SDR

4.4.2 Strategy 2
To attempt to overcome the problems found in strategy 1, another strategy is im-
plemented. In this strategy, the object signal and the reference signal use different
receivers in different channels, this allows to receive in parallel. The strategy works
on the assumption that both receivers are identical. Additionally, since the receivers
will be located farther apart, then there should be a smaller amount of leakage be-
tween them.
The architecture of the RF transceiver in the board does not permit to send a
signal from one channel and receive in two different channels. Therefore, the two
transmitting channels will be activated, even if only one transmitter is sending data.
The unused transmitting port will not send any data and the port will be terminated
with a 50 Ω coaxial matched load. This is done to ensure that the transmitter will
not interfere with the other signals.
The strategy is shown in Figure 4.14, the transmitter/receiver in channel 0 is
connected to a 3 dB power divider, and each of the output ports of the power divider
is connected to an attenuator, which is further connected to the receiving only port
in each channel. Again, a frequency sweep was done in the range 500 MHz - 6 GHz,
using the same settings as in Strategy 1.

Figure 4.14: Measurement setup used in strategy 2.

The measurements were only repeated 5 times for this strategy, since the re-
peatability of the measurements was not good. Figure 4.15 shows the magnitude
and phase of the relative measurement when no averaging was applied to the mea-
surement data. Likewise Figure 4.16 shows the magnitude and phase of the relative
measurement, with data averaged over 1200 samples.
It can be seen that the repeatability of the measurements is not as good as in
Strategy 1, since only 5 measurements were done and the value of the magnitude
and phase of the relative measurements are stable, but fluctuating in a big margin.
The uncertainty increases with the frequency, just like in Strategy 1. This is
reduced by the averaging. It can be seen that the averaging did not help reduce the
uncertainty in frequency points under 3 GHz. This happens because this strategy
is limited by deterministic noise, therefore, the averaging does not help reduce it.
When tested, the receivers did not behave in the same way. This deterministic noise

24
 4. Fundamental Test with the SDR

could be explained by differences in the hardware in both receivers. This difference

in hardware will increase the number of error sources, because the end results will
have the added error of the difference between the receivers.

25
4. Fundamental Test with the SDR

(a) Magnitude of 5 measurements (no averaging).

(b) Phase of 5 measurements (no averaging).

Figure 4.15: Magnitude and phases obtained with strategy 2 (not averaged data).

(a) Magnitude of 5 measurements (averaged over 1200 samples).

(b) Phase of 5 measurements (averaged over 1200 samples).

Figure 4.16: Magnitude and phases obtained with strategy 2 averaged over 1200
samples.

26
 4. Fundamental Test with the SDR

4.4.3 Strategy 3
Considering the architecture of the board, another strategy similar to strategy 2 was
implemented. In this strategy each RF attenuator will be connected to a different
channel, this can be seen in Figure 4.17.The setup of the cables change but the
program running is the same as in strategy 2. Instead of sending no data in the
second channel, The same signal is sent in both transmitters. A frequency sweep
with the same settings as the other strategies is done.

Figure 4.17: Measurement setup used in strategy 3

The measurements were only repeated also 5 times for this strategy, since the
results were not as promising as Strategy 1. Figure 4.18shows the magnitude and
phase of the relative measurement when no averaging was applied to the measure-
ment data. The magnitude and the phase are better than strategy 2, but still the
results are not as good as the results in strategy 1. Figure 4.19 shows the magnitude
and phase of the relative measurement, with data averaged over 1200 samples. The
averaging helped with the uncertainty increase with the frequency, like in the other
strategies, but did not decrease the uncertainty in lower frequencies. The conclusion
is that this strategy is limited by deterministic noise and comes from hardware dif-
ferences. This strategy uses two different transmitters and two different receivers, so
there should be more error sources than in strategy 2. Yet, the repeatability of the
measurements is better in this strategy than in strategy 2 for an unknown reason
to the author.

27
4. Fundamental Test with the SDR

(a) Magnitude of 5 measurements (no averaging).

(b) Phase of 5 measurements (no averaging).

Figure 4.18: Magnitude and phases obtained with strategy 3 (not averaged data).

(a) Magnitude of 5 measurements (averaged over 1200 samples).

(b) Magnitude of 5 measurements (averaged over 1200 samples).

Figure 4.19: Magnitude and phases obtained with strategy 3 averaged over 1200
samples.

28
 5
Real Measurements

In this chapter, real measurement will be performed in order to evaluate the mea-
surement accuracy.

5.1 Initial accuracy test

The accuracy is how much the values of the measurements are close to the true val-
ues. The accuracy of the measurements is important for the image reconstruction
in microwave tomography systems, because it is the prerequisite for the successful
reconstruction of the internal composition of the object. Figure 5.1 and 5.2 are sim-
ple measurement setups to resemble a measurement example for microwave imaging.
A power splitter was used to split the power to two RF paths. One is the refer-
ence measurement path and other is the object measurement path. In Figure 5.1,
the 6 dB attenuator is the measurement object of interest and In Figure 5.2, the
measurement object is the 10 dB attenuator.

Figure 5.1: Measurement setup using the 6 dB attenuator as object of interest

Figure 5.2: Measurement setup using the 10 dB attenuator as object of interest

29
5. Real Measurements

Figure 5.3 the difference between the transmission coefficient obtained from the
SDR measurement and the one measured with a VNA.

(a) Magnitude of difference between the transmission coefficients

(b) Phase of difference between the transmission coefficients

Figure 5.3: Magnitude and phase of difference between the transmission coefficients
using 6 dB and 10 dB attenuators as objects

It can be seen that the difference, between the magnitude obtained with the
SDR measurement, and the magnitude obtained with the VNA is less than 1 dB
difference for frequencies under 3 GHz. On the contrary, the phase estimated with
the SDR is different from the phase estimated with the VNA.
To further investigate this phase error source, some calibration strategies can
be implemented. The first setup that will be used is displayed in Figure 5.4. The
transmission coefficients obtained with only cables will be used as calibration coef-
ficients. Additionally, measurements using 6 dB attenuator as object and 10dB as
object will also be used as calibration coefficients.

Figure 5.4: Measurement setup using no object (cables)

The calibration coefficients will be subtracted from the measurements obtained

30
 5. Real Measurements

while using the 6 dB attenuator as object, the results of this operation are showed in
Figure 5.5. It can be seen that the results using the calibration coefficients obtained
with the cables, have a really oscillatory behaviour. The results using the calibration
coefficients obtained with the attenuators are very similar to the measurements of
the VNA.
Figure 5.6 displays the results of subtracting the calibration coefficients from
measurements done with the 10 db attenuator as object. Similar results can be seen
in these measurements. The calibration done with the 6 db and 10 db measurements
give a better result that the calibration done with only cables.

31
5. Real Measurements

(a) Magnitude of measurement using 6 dB as object

(b) Phase of measurement using 6 dB as object

Figure 5.5: Measurement using 6 dB as object with three different calibration

coefficients

(a) Magnitude of measurement using 10 dB as object

(b) Phase of measurement using 10 dB as object

Figure 5.6: Measurement using 10 dB as object with three different calibration

coefficients

32
 5. Real Measurements

5.2 Antenna measurements

We investigated the measurement accuracy of a real microwave imaging antenna
system, as shown in Figure 5.7. This imaging tank is composed of 20 monopole
antennas that evenly distributed in a circle of radius 20 cm.

(a) Side view of imaging tank (b) Antenna array inside imaging tank

Figure 5.7: Picture of imaging tank used for measurements

The transmission coefficient between different antennas pairs were estimated

from SDR measurement and compared with the VNA data. Figure 5.8 shows the
magnitude results obtained with an adjacent antenna pair. Figure 5.10 show the
magnitude results obtained with the most distant antenna pair. It can be seen
that the magnitude response from the adjacent antenna pair is similar to the object
magnitude response obtained with the VNA. Particularly in frequencies under 3GHz.
However, the phase is different (as seen in Figure 5.9). This is the same behaviour
seen in the measurements using attenuators as objects. In the most distant antenna
pair the quality of the estimation decrease. There is also a phase difference between
the estimation and the VNA data (Figure 5.11).

Figure 5.8: Adjacent antenna pair response obtained with SDR and VNA

33
5. Real Measurements

Figure 5.9: Phase of adjacent antenna pair response obtained with SDR and VNA

Figure 5.10: Most distant antenna pair response obtained with SDR and VNA

Figure 5.11: Phase of most distant antenna pair response obtained with SDR and
VNA

34
 5. Real Measurements

5.3 Discussion
After studying the possible reasons for these results, it was found that the reflection
coefficient is very high in the ports of the SDR board. Furthermore, the reflection
coefficient changes when the port is in active state and when it is in inactive state
as seen in Figure 5.12.
The decrease in magnitude in the received signals (Section 4.1) can also be
explained by the increase of the reflection coefficient in higher frequencies. This
means there will always be a decrease in quality of the results in higher frequencies.
This makes the calibration more difficult, because this means that a great part
of the signal will be reflected back and leaked between the ports. The current power
splitter, presents a 20dB isolation, which is not enough to attenuate the effect of
the signals. This explains the difference between the phase estimated with SDR
measurements and the phase obtained from VNA data, as seen in Figure 5.3.
It can be seen in Figure 5.5 that the calibration of the phase difference is more
accurate if the object used for the calibration coefficients is similar to the object
being measured. The same can be said for the results shown in Figure 5.6. The
calibration coefficients obtained using the setup shown in Figure 5.4, display an
oscillatory behaviour. This happens because the leaked signal is stronger.
In Figure 5.8 and Figure 5.10, it can be seen that the magnitude of the received
signals decrease the farther away the antennas are. This makes the results more
vulnerable to noise in the most distant pair.
One way to resolve this issue would be to take into account the effect that the
leakage is creating into the calculations from the calibration, and another one would
be to have more isolation in the connecting devices.

35
5. Real Measurements

(a) Reflection coefficient of port Tx/Rx1 at inactive state

(b) Reflection coefficient port Tx/Rx1 at active state

(c) Reflection coefficient port Rx2 at inactive state

(d) Reflection coefficient port Rx2 at active state

Figure 5.12: Measured reflection coefficient at the RF ports of the SDR board

36
 6
Other Applications

Given the flexibility of the SDR, it can also be used for other medical applications,
like non contact measurement of respiration and heartbeat rate. Respiration rate
and heartbeat rate are important physiological variables for the human being. The
monitoring of this parameters becomes crucial in medical applications and rescue
missions, where doppler radars can be used for the through-obstacle detection of
respiration and heartbeat rate, achieving easier and fast localization of survivors in
earthquake rubble. Usually a doppler radar senses motion by transmitting a radio
wave signal, and receiving the motion-modulated signal that was reflected from the
target. The reflected wave is frequency modulated by the periodic movements on
the surface of the chest generated by respiration and heart movements. If the radar
is illuminating the whole chest at once, the detected motion will be an average of all
displacement in the area [39]. The amplitude of the chest displacement is expected
to be around 10mm for respiration and 0.1mm for heart activity [40].

6.1 Method
The experiment was conducted with only one person sitting or standing (Figure
6.1), with the radar located in the same height as the chest. The distance between
the radar and the person could be adjusted during experiments [39, 41].

Figure 6.1: Basic test scenario for vital sign detection.

A metronome was used to have a reference measurement of the respiration rate.

The reference respiration rate is 16 BPM. No reference data was obtained for the
heartbeat rate.

37
6. Other Applications

The measurements will have a duration of 30s to make sure to get a whole
period of the respiration and heartbeat signals. Since the healthy adult breathes
from 12-20 times per minute, the signal will be filtered with a butterworth band
pass filter of 6th order with the range between 0.15 Hz and 0.5 Hz (9-30 breaths per
minute), to obtain the respiration signal. The resting heartbeat of a healthy adult
is in the range 60-100 beats per minute, so the received signal will also be filtered
with a filter with range between 0.9Hz and 2.5 Hz (54-150 beats per minute).
Two methods will be used to find the respiration and heartbeat rate of the
filtered and normalized signals, the FFT, and the Auto-correlation. The FFT of the
signal will be taken and the maximum value (peak) in the frequency domain will
represent the frequency of respiration or heartbeat rate. The Auto-correlation can
also be used to detect the presence of the period and determine their duration, since
the auto-correlation sequence of a periodic signal has the same periodic character-
istics as the signal [42]. In the auto-correlation the first peak (p1), after the main
peak in located in 0, will indicate the periodicity of the signal. The respiration’s rate
can be found with the following formula (a detailed explanation of this principle is
discussed in [42]):
60
fp 1 = , (6.1)
t(p1 )
where fp1 is the respiration rate and t(p1 ) is the time in seconds it takes for p1 to
occur in the auto-correlation signal.

6.2 Results
Figure 6.2 displays the reflected signal received by the board after being modulated
by the chest displacement of the subject.

Figure 6.2: Received signal.

It is noticeable that the signal has a periodical pattern. It must be noted that
the information from the displacement generated from respiration and displacement
generated by the heartbeat are both included in this signal, in other words, it is the
sum-up of the total displacement on the chest area. This signal is filtered with the
respiration filter, and the resulting signal is shown in Figure 6.3a. Looking at the
signal, 8 periods in 30 s can be counted, meaning that the respiration rate is of 16
breaths per minute.

38
 6. Other Applications

Figure 6.3b shows the signal in time domain and the FFT of the signal. It
can be seen that the peak is located in 16 breaths per minute, confirming that the
estimation of the respiration rate with the SDR measurements is correct.

(a) Time domain signal.

(b) Frequency domain signal.

Figure 6.3: Signal After respiration filer in Time domain and Frequency domain.

The auto-correlation is also used to determine the respiration rate, the result
from auto-correlating the filtered and normalized signal is shown in Figure 6.4.

Figure 6.4: Auto-correlation of respiration.

The first peak is in 3.84s, using Equation 6.1 the respiration rate can be found.
The resulting respiration rate is 16.3 breaths per minute which is a number close
to the appropriate respiration rate. The signal filtered with the heart beat filter
is displayed in Figure 6.5 in time domain and frequency domain. In the frequency
Domain it can be seen that a peak is found in 82 beats per minute indicating this
is a possible value of the heartbeat rate. This is an expected value since the normal
rate for adults is from 60 to 100 beats per minute.

39
6. Other Applications

(a) Time domain signal.

(b) Frequency Domain signal.

Figure 6.5: Signal After Heartbeat filer in Time domain and Frequency domain

6.3 Discussion
The respiration rate and heartbeat rate were correctly estimated using the SDR
measurements. The testing environment will have an effect of the data received, if
there is more than one person in the way of the radar, the signals get mixed and it
is difficult to separate them.
The advantage of the auto-correlation method is that even if the signal is shorter
in time, it will be able to find the respiration rate, since it only needs one period
of the signal to find the periodicity. It is also more robust against noise than the
FFT method. This is useful if rapid measurements are needed, such as real time
monitoring of the vital signs of elderly people.

40
 7
Conclusion

This study set out to investigate the feasibility of using SDR for microwave-based
medical applications. As part of this research, three different calibration strategies
were implemented and tested to attempt to fix the random phase problem encoun-
tered due to the use of different PLLs in the transmitter and the receiver in the
SDR technology. Among the calibration strategies tested, it was found that strat-
egy 1 performed better than strategy 2 and strategy 3. With this strategy, a good
measurement repeatability was obtained, which could allow the device to be used
for medical monitoring.
The measurement accuracy, however, is affected by the strong reflection of
the RF ports of the SDR board. A more effective calibration approach needs to
be developed in order to improve the measurement accuracy. Furthermore, it was
shown that the SDR technology has a great potential to be used for non-contact
vital sign monitoring at a distance.

7.1 Future Work

Due to the limitations set by the reflection coefficients in the ports of the SDR
it would be interesting to further develop the calibration methods to improve the
measurement accuracy. One approach is to take the effect of leakage into account
in the calibration and the other approach is to increase the isolation in the devices
used for the calibration. This would decrease the leaked signal from port to port,
allowing for the simple calibration method to estimate the absolute phase. Imaging
test was missing in this report and it is interesting to perform the test after the
accuracy of the measurements is improved.

41
7. Conclusion

42
 Bibliography

[1] Andreas Fhager, Stefan Candefjord, Mikael Elam, and Mikael Persson. Mi-
crowave diagnostics ahead: Saving time and the lives of trauma and stroke
patients. IEEE Microwave Magazine, 19(3):78–90, 2018.
[2] Conceicao Raquel Cruz, Johan Jacob Mohr, and Martin OHalloran. An Intro-
duction to Microwave Imaging for Breast Cancer Detection. Springer Interna-
tional Publishing, 2018.
[3] A. Fhager, P. Hashemzadeh, and M. Persson. Reconstruction quality and
spectral content of an electromagnetic time-domain inversion algorithm. IEEE
Transactions on Biomedical Engineering, 53(8):1594–1604, 2006.
[4] Maryam Naseri. Microwave tomography for breast cancer detection. Master’s
thesis, Chalmers University f technology, 2015.
[5] T. S. England. Dielectric properties of the human body for wave-lengths in the
1–10 cm. range. Nature, 166(4220):480–481, 1950.
[6] A.j. Surowiec, S.s. Stuchly, J.r. Barr, and A. Swarup. Dielectric properties of
breast carcinoma and the surrounding tissues. IEEE Transactions on Biomed-
ical Engineering, 35(4):257–263, 1988.
[7] K R Foster, J L Schepps, R D Stoy, and H P Schwan. Dielectric properties
of brain tissue between 0.01 and 10 ghz. Physics in Medicine and Biology,
24(6):1177–1187, 1979.
[8] A Peyman, S J Holden, S Watts, R Perrott, and C Gabriel. Dielectric properties
of porcine cerebrospinal tissues at microwave frequencies:in vivo,in vitroand sys-
tematic variation with age. Physics in Medicine and Biology, 52(8):2229–2245,
2007.
[9] Rachael I. Scahill, Chris Frost, Rhian Jenkins, Jennifer L. Whitwell, Martin N.
Rossor, and Nick C. Fox. A longitudinal study of brain volume changes in
normal aging using serial registered magnetic resonance imaging. Archives of
Neurology, 60(7):989, 2003.
[10] Serguei Semenov, Toan Huynh, Thomas Williams, Brian Nicholson, and Anna
Vasilenko. Dielectric properties of brain tissue at 1ghz in acute ischemic stroke:
Experimental study on swine. Bioelectromagnetics, 38(2):158–163, 2016.
[11] Puyan Mojabi, Majid Ostadrahimi, Lotfollah Shafai, and Joe Lovetri. Mi-
crowave tomography techniques and algorithms: A review. 2012 15 Interna-
tional Symposium on Antenna Technology and Applied Electromagnetics, 2012.
[12] Johan Köster. Image reconstruction and fdtd modelling of an antenna array for
3d microwave tomography. Master’s thesis, Chalmers University of technology,
2011.

43
Bibliography

[13] Hooi Been Lim, Nguyen Thi Tuyet Nhung, Er-Ping Li, and Nguyen Duc Thang.
Confocal microwave imaging for breast cancer detection: Delay-multiply-and-
sum image reconstruction algorithm. IEEE Transactions on Biomedical Engi-
neering, 55(6):1697–1704, 2008.
[14] Beadaa Jasem Mohammed. Design and implementation of microwave imaging
systems for medical applications. PhD thesis, The University of Queensland,
Australia, 2014.
[15] Joshua Chong Yue Lai, Cheong Boon Soh, Erry Gunawan, and Kay Soon Low.
Uwb microwave imaging for breast cancer detection — experiments with hetero-
geneous breast phantoms. Progress In Electromagnetics Research M, 16:19–29,
2011.
[16] James.R.Andrews. Picosecond Pulse Generator for Uwb Radars. Picosecond
Pulse Lab, application note an-9 edition, 2000.
[17] Xuezhi Zeng. Time domain systems for microwave imaging: accuracy evalua-
tions and prototype design. PhD thesis, Chalmers, 2013.
[18] Cemin Zhang and Aly Fathy. Reconfigurable pico-pulse generator for uwb
applications. 2006 IEEE MTT-S International Microwave Symposium Digest,
2006.
[19] J. Marimuthu, K.s. Bialkowski, and A.m. Abbosh. Reconfigurable software
defined radar for medical imaging. 2014 1st Australian Microwave Symposium
(AMS), 2014.
[20] Anthony E. Stancombe and Konstanty S. Bialkowski. Portable biomedical mi-
crowave imaging using software- defined radio. 2018 Asia-Pacific Microwave
Conference (APMC), 2018.
[21] Cam Nguyen and Jeongwoo Han. Time-domain ultra-wideband radar, sensor
and components. SpringerBriefs in Electrical and Computer Engineering, 2014.
[22] James.R.Andrews. Picosecond Pulse Generator techniques and pulser capabili-
ties. Picosecond Pulse Lab, application note an-19 edition, 2008.
[23] Lianfeng Zou, Shulabh Gupta, and Christophe Caloz. A simple picosecond
pulse generator based on a pair of step recovery diodes. IEEE Microwave and
Wireless Components Letters, 27(5):467–469, 2017.
[24] Jianming Zhou, Qiuyuan Lu, Fan Liu, and Yinqiao Li. A novel picosecond
pulse generation circuit based on srd and nltl. Plos One, 11(2), 2016.
[25] J.s. Lee and C. Nguyen. Uniplanar picosecond pulse generator using step-
recovery diode. Electronics Letters, 37(8):504, 2001.
[26] D. Williams, P. Hale, and K.a. Remley. The sampling oscilloscope as a mi-
crowave instrument. IEEE Microwave Magazine, 8(4):59–68, 2007.
[27] Maxim Integrated. Multiply Your Sampling Rate with Time-Interleaved Data
Converters, tutorial 989 edition, 2001.
[28] Keysight Technologies. What is the difference between an equivalent time sam-
pling oscilloscope and a real-time oscilloscope?, application note1608 edition.
[29] Ian Hickman. Oscilloscopes: how to use them, how they work. Elsevier, 2001.
[30] Changzhi Li Changzhan Gu. From tumor targeting to speech monitoring: Accu-
rate respiratory monitoring using medical continuous-wave radar sensors. IEEE
Microwave Magazine, 15(4):66–76, 2014.

44
 Bibliography

[31] Adrian Figueroa, Belal Al-Qudsi, Niko Joram, and Frank Ellinger. Comparison
of two-way ranging with fmcw and uwb radar systems. 2016 13th Workshop on
Positioning, Navigation and Communications (WPNC), 2016.
[32] Cam Nguyen and Joongsuk Park. Stepped-frequency radar sensor analysis.
Stepped-Frequency Radar Sensors SpringerBriefs in Electrical and Computer
Engineering, page 39–64, 2016.
[33] Stephanie Hamill. Introduction to sdr - wireless innovation forum.
[34] What is ni usrp hardware?
[35] Donald R. Wehner, Bruce Barnes, Benjamin C. Flores, and Donald R. Wehner.
High-resolution radar. Artech House, 1996.
[36] David Andrew Noon. Stepped-Frequency Radar Design and Signal Processing
Enhances Ground Penetrating Radar Performance. PhD thesis, The University
of Queensland, Queenslan, 1996.
[37] National Instruments. AN INTRODUCTION TO SOFTWARE DEFINED
RADIO With NI LabVIEW and NI USRP. 2013.
[38] Analog Devices. Rf agile transceiver ad9361, 2013.
[39] Olga Boric-Lubecke, Wansuree Massagram, Victor M. Lubecke, Anders Host-
Madsen, and Branka Jokanovic. Heart rate variability assessment using doppler
radar with linear demodulation. 2008 38th European Microwave Conference,
2008.
[40] T. Kondo, T. Uhlig, P. Pemberton, and P. D. Sly. Laser monitoring of chest
wall displacement. European Respiratory Journal, 10(8):1865–1869, 1997.
[41] Hongming Shen, Chen Xu, Yongjie Yang, Ling Sun, Zhitian Cai, Lin Bai,
Edward Clancy, and Xinming Huang. Respiration and heartbeat rates mea-
surement based on autocorrelation using ir-uwb radar. IEEE Transactions on
Circuits and Systems II: Express Briefs, 65(10):1470–1474, 2018.
[42] Guanghao Sun and Takemi Matsui. Rapid and stable measurement of res-
piratory rate from doppler radar signals using time domain autocorrelation
model. 2015 37th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society (EMBC), 2015.
[43] Configuration:advanced:lo frequency property.

45
Bibliography

46
 A
Appendix 1

A.1 LO functionality block

The USRP includes DC offset correction in the RFIC AD9361, which will use a
filter to eliminate any DC components. This is a desirable trait in communications
systems to prevent bit errors from happening, but in order to send a sinusoid of
0Hz, the LO frequency needs to be configured.
The carrier frequency is obtained in hardware by combining the frequency trans-
lation on the local oscillator (LO) and the additional frequency translation in the
digital signal processing on the FPGA [43].
If only the value of the carrier frequency is specified, the NI-USRP will set the
LO frequency as close as possible to the carrier frequency and then carry out the rest
of the frequency translation using the Digital Signal Processor(DSP) in the FPGA.
The difference between the LO frequency and the carrier frequency cannot be
greater than half the device bandwidth, since the DSP of the FPGA will not be able
to process the rest of the signal due to a lack of sufficient bandwidth.
Taking everything into account, the LO frequency needs to be configured to
a value far enough from the carrier frequency that the DC offset correction in the
AD9361 will not filter it out, but still close enough that the bandwidth of the Digital
signal processor will be able to process the remaining signal.
To prove how the filtering of the DC signal works, a simple experiment can be
conducted using the configuration shown in Figure 4.1.
A signal of frequency 0Hz and amplitude 0,7 will be sent (shown in Figure
A.1), the carrier frequency (fc ) will then be configured to be 915MHz and the LO
frequency (flo ) will not be configured, thus, allowing the NI-USRP to choose an LO
frequency as close to 915M as possible.
Figure A.2a shows the received signal, both I and Q signals are filtered out by
the DC correction filter in the AD9361 since the signal just goes to amplitude 0.
Likewise, Figure A.2b shows the power spectrum of the signal, since nothing was
received, it can be seen that the power is at a value of -80dB.
By experimenting with different values, it was found that the value fc −flo =5MHz
was a good difference between the carrier frequency and LO frequency. This con-
clusion was reached, since the 5MHz signal is different enough from being a DC
signal,that it will not be filtered out, but small enough that the Digital signal pro-
cessor from the FPGA will be able to process it.

I
A. Appendix 1

Figure A.1: Sinusoid signal with 0Hz and amplitude 0,7

(a) Received signal (b) Power Spectrum

Figure A.2: Received signal while not configuring the LO frequency

the previous experiment is repeated, but this time using fc − flo =5MHz, mean-
ing that the LO frequency is configured to be flo =910MHz, since the carrier fre-
quency is fc =915MHz.The results are displayed in Figure A.3 .

(a) Received signal (b) Power Spectrum

Figure A.3: Received signal while not configuring the LO frequency

The Received signal can be seen in Figure A.3a,the DC signal is now received
and down-converted without problems.The power spectrum of the received signal
when configuring the LO frequency is shown in Figure A.3b.The power is at a value
of -20dB.

II