Robot Localization with Radio Frequency Identification and Time Difference of Arrival

(1)

IN

DEGREE PROJECT ELECTRICAL ENGINEERING, SECOND CYCLE, 30 CREDITS

STOCKHOLM SWEDEN 2020,

Robot Localization with Radio

Frequency Identification and Time Difference of Arrival

HÅKAN GÖRANSSON

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCE

(2)

Robot Localization with Radio Frequency Identification and Time Difference of Arrival

HÅKAN GÖRANSSON

Master in Systems, Control and Robotics Date: January 19, 2020

Supervisor: Johan Bolin Examiner: Magnus Jansson

School of Electrical Engineering and Computer Science Host company: Envirologic AB

(3)

(4)

iii

Abstract

In today’s robot technology it is important to have a system that enables for a robot to find its own position. This can be achieved with different kinds of sensors, such as LIDAR, GPS and cameras, and sensor fusion.

In this master thesis, we develop a system for a robot that follows a straight wall such that it can find its own position along this wall. This is solved by measuring the Time Difference of Arrival (TDoA) between one RFID-tag and two radio receivers.

We have developed prototypes for the radio receivers that receive the radio signal from the RFID-tag. To improve the TDoA measurement we have developed a state space model for the TDoA measurements along the wall, that is used in a Kalman filter to estimate the TDoA. The radio receivers can at least receive signals down to −52dBm, which is enough to receive radio signals from an RFID-tag.

The simulations of the Kalman filter show satisfactory results if it is not too much noise in the TDoA measurements.

(5)

iv

Sammanfattning

I dagens robotteknik är det viktigt att ha ett system som möjliggör för en robot att hitta dess egna position. Detta kan åstadkommas med olika sorters senso- rer, som LIDAR, GPS och kameror, och sensorfusion.

I denna rapport utvecklar vi ett system som gör att en robot som följer en rak vägg kan hitta dess egen position längs väggen med hjälp av RFID-teknik.

Det är löst genom att beräkna Time Difference of Arrival (TDoA) mellan en RFID-tag och två radiomottagare.

Vi har utvecklat en prototyp av radiomottagarna som tar emot radiosignalen ifrån RFID-taggen. För att förbättra TDoA mätningarna så har vi utvecklat en tillståndsmodell för TDoA mätningarna, vilket kan användas i ett Kalman- filter för att estimera TDoA. Insignalen till tillståndsmodellen är hastigheten på roboten.

Radiomottagaren kan ta emot en radiosignal åtminstone ner till −52dBm, vilket är tillräckligt för att ta emot radiosignaler från en RFID-tagg.

Simuleringarna av Kalman-filtret ger tillfredställande resultat om det inte är för mycket brus i TDoA mätningarna.

(6)

Chapter 1 Introduction

Localization of robots has been a hot research topic the last years. Localiza- tion can be performed with different methods like beacons [1], Angle of Ar- rival (AoA) [2], Time of Flight (ToF) [3], snapshots [4]. It is not only robots we want to localize. Humans can be another thing we want to localize. This is not a problem outdoors, since we have the localization standard Global Posi- tion System (GPS). GPS is not useful indoors since the radio signal from the satellites is not strong enough indoors. There is not yet a standard for indoors positioning systems. Microsoft has had a competition [5] among universities and other companies to aim for some sort of direction to find a standardization for indoors localization systems.

In this master thesis we investigated the possibility to use RFID-technology to enable for a robot to find its own position along a straight wall indoor. This is investigated by measuring the TDoA between one RFID-tag and two receivers.

The antennas for the receivers will be mounted on the robot approximately 1 meter apart.

To improve the measurements of TDoA we have developed a state space model for the TDoA measurements, which can be used in a Kalman filter to estimate and improve the measurements. The goal is to have a system for the robot such that it can find its own position within 0.02 meters along a straight wall.

1

(9)

2 CHAPTER 1. INTRODUCTION

1.1 Background

This master thesis is performed at the company Envirologic AB. Envirologic AB develops and sells cleaning robots that are used in pig barns and a few other environments.

The robot is moving in a straight path and is cleaning boxes along this path, so the robot only moves in one dimension. We only need to find one specific starting reference point for each box, so there is no requirement to have any continuous localization of the robot.

At the moment there are magnets mounted at each starting reference point.

The robot localizes the magnets with an inductive sensor that is mounted on the robot. The drawback of this solution is that the magnets must be mounted before the cleaning and removed after each cleaning, since the pigs are chewing at the magnets and they are being abstracted. The magnets can also be mounted incorrectly such that the starting reference point will be incorrect. The sensing distance between the inductive sensor and the magnets is below 10 cm which can cause problems.

1.2 Thesis Aim

The goal of this thesis is to investigate the possibilities to use passive RFID- tags [6] to find one specific position for each of the boxes that should be cleaned. This will make it easier and more reliable to clean, since the RFID- tags can be mounted such that the pigs cannot reach them and hence they are not being abstracted. With this solution, the RFID-tags do not need to be mounted and removed between each cleaning process.

We will investigate the possibilities to calculate the Time Difference of Ar- rival (TDoA) [7] between one RFID-tag and two antennas which are mounted on the robot, such that the robot can find its own position.

The final requirements are:

• The reading distance between the robot and the RFID-tag should be at least 3 meters.

• The robot should find its own position within 2 cm from the starting reference point.

(10)

CHAPTER 1. INTRODUCTION 3

1.3 Previous Work

There has been plenty of research about this topic and many different ap- proaches to localize objects with RFID technology. Most of them have one thing in common, they are using a software defined radio (SDR). This is a unit that only has the front end of a radio and sample the radio waves such that it is possible to demodulate the signal in software instead of in the hardware. Since the SDR is sampling the radio wave, it is possible to perform signal process- ing with the received radio signal. That is suitable for almost any project that involves radio waves.

One common method is to use Angle of Arrival (AoV). This method uses two or more antennas and compare the phase difference of received radio waves between these antennas. One of these methods uses three different frequencies to be able to calculate both the angle and the distance to the tag, which is explained in [8]. Another method that uses phase difference to calculate the AoA is explained in [9], where they are using an active RFID-tag, which is sending bursts of pure sine waves. The authors are using both a straightfor- ward technique and the maximum likelihood estimation to calculate the phase difference between the two antennas. The results in this paper are satisfying, within a few degrees of error.

There are other algorithms to improve the AoA measurement, the MUSIC- algorithm that is explained in [10] and Root-MUSIC explained in [11]. These methods have one thing in common, which is that there is only one RFID-tag.

This will be a problem in this master thesis, since we cannot know how many RFID-tags that are present in the region of the reception for the RFID-reader.

Another common method to measure distances with radio waves is to use the radio signal strength (RSS). It is strongly affected by disturbances and un- wanted reflections as described in [12]. Another issue that might occur in the application of this master thesis is that one of the antennas can collect a lot of dirt that might affect the RSS between measurements.

Snapshot is another method that can be used in localization problems. It is described in [4]. It is implemented by mounting RFID-tags in the area where we want to localize the robot. The robot must be trained to be able to locate its current position. The training process is achieved by letting the robot drive around in the area and map the RFID-tags that it can read at the current posi-

(11)

4 CHAPTER 1. INTRODUCTION

tion of the robot. This method or estimation can be combined with a particle filter to improve the estimation, as in [4]. The estimation of the robot position has a mean error of 0.2 meters, which is not good enough in our application.

In [13] and [14], the authors describe two different techniques to calculate the Time of Flight (ToF). In [13], the authors describe a method where they are transmitting a signal that begins with frequency f and increase with a bandwidth B under a fixed period of time and then begins from the frequency f again. This creates a sawtooth frequency signal, and by comparing the frequency of the transmitted and the received signal it is possible to calculate the ToF. This method is used since it is much easier to compare a small change in frequency than to calculate time in nanoseconds.

It is necessary to have as wide bandwidth as possible to have a satisfactory measurement, around 300M Hz. But since we are limited between 865M Hz up to 868M Hz it is not possible to use this method. There are also problems with reflections and multipaths that influence the measurements. In [14] the authors describe the same method, but they have improved the accuracy by reducing the effect of reflection and multipath signals. This is accomplished by letting the RFID-tag transmit a custom message which the receiver knows when it should compare the frequencies. The drawback of this method is that they need custom made RFID-tags that are using a specific frequency at 2.45GHz in the ToF measurement.

1.4 Outline

In chapter 2 we introduce the technology of RFID. In chapter 3 we describe localization with TDoA and develop our own model of TDoA for the robot that follows a straight wall. In chapter 4 we develop our radio receiver and the theory of how we calculate the TDoA. In chapter 5 we describe how we estimate the measurements with a Kalman filter to increase the accuracy of the measurements and in chapter 6 we present our results.

(12)

Chapter 2 RFID

2.1 Introduction

RFID technology is used in many implementations today due to its cheap and easy implementations. The RFID-tags can cost from a few cents up to tens of dollar, which depends on the usage. It can be used in cargo tracking, access control, in doorways and different localization methods and is an attractive technology for stores to replace the bar-code in the payment system.

2.2 RFID Technology

In this section of the report, we go through the most essential about RFID technology that we use in the rest of the report.

RFID systems consist of one RFID-reader with one or more antennas. The reader can read the ID of the RFID-tags. The RFID-tags can have a memory that the user of the system can program and read. There exist different kinds of RFID technology, which affect the reading distance. The most common tags are so called passive tags. These tags get their energy from the Continuous Wave (CW) that the RFID-reader transmits. It is the reflection of the CW that the RFID-tag modulates when it is talking to the RFID-reader. There also exist active tags, which get their energy from their own battery. This gives them much longer reading distance, >100 meters [15]. The passive tags are used with different frequencies depending on the reading distance; Low Frequency (LF), High-Frequency (HF) and Ultra-high frequency (UHF) tags. They cover the frequencies from 125 kHz to 134 kHz, 3 MHz to 30 MHz and 860 MHz to 960 MHz, respectively [16].

5

(13)

6 CHAPTER 2. RFID

Figure 2.1: PIE encoding.

There is no frequency standard in the world, so the frequencies are different in different parts of the world.

The UHF reader can communicate with the tag with different modulations, Amplitude Shift Keying (ASK) [17] or Phase Reverse Amplitude Shift Keying (PR-ASK) [17]. The RFID-tag modulates the backscattered [18] radio signal with ASK. The backscattered symbol time is called Tpri, so the data rate is the inverse of Tpri. The data rate of the signal is called the Backscatter Link Frequency (BLF).

As already mentioned, the RFID-reader is using ASK when it communi- cates with the RFID-tag. In the early years of RFID, the RFID-reader modulated the transmitting signal with on-off keying (OOK), where the signal was kept high to indicate a 1-bit and the signal was kept low to indicate a 0-bit, with equal length. This could cause problems in the communication between the RFID-reader and the RFID-tag, since the RFID-tags get their energy from the CW and if the RFID-tag receives a long stream of zeroes then the CW is kept low for a long time so the RFID-tag could lose its power.

To solve this problem the encoding had to be changed, such that the RFID- tag could keep its energy throughout the communication. They started to use pulse-interval encoding (PIE) instead of OOK, which is shown in figure 2.1.

As can be seen, if the RFID-tag receives a long stream of zeroes it still has at least 50 % of the power that is delivered to the RFID-tag.

The backscattered signal from the RFID-tags can be encoded with FM0 or MMS. FM0 switches its state at each bit, so if we have a low 1-bit and the next

(14)

CHAPTER 2. RFID 7

Figure 2.2: FM0 encoding.

Figure 2.3: MMS-2 encoding.

bit is a 1-bit then the signal switches its state such we have a high 1-bit. The 0-bit switches its state as the 1-bit but it also switches from high to low or low to high after ^{T pri}₂ , as can be seen in figure 2.2.

MMS, (miller-modulated subcarrier) modulation consists of three different encodings, MMS-2, MMS-4 and MMS-8, where 2, 4, and 8 stand for the number of Tpri for one bit in the MMS encoding. The signal changes its state with the same frequency as in FM0 encoding. For example, MMS-2 uses two Tpri for one bit in MMS-2 encoding, so MMS-2 has the length of 2 Tpri for one bit, MMS-4 has the length of 4 Tpri for one bit and MMS-8 has the length of 8 Tpri for one bit. Figure 2.3 shows an example of MMS-2 encoding.

(15)

8 CHAPTER 2. RFID

FM0 is more sensitive to noise than MMS encoding and MMS can provide more flexibility between data rate vs noise tradeoff. Hence, if we need high data rate in a noise free environment, one would prefer FM0, otherwise MMS encoding would be the best choice.

The EPC Class 1 Generation 2 tags that we use in this master thesis, is so called reader talk first tags. Where the reader initiates the communication between the reader and the tag. When the RFID-tag antenna receives the CW from the RFID-reader, it starts to radiate the CW. The RFID-tag ASK- modulates the backscattered signal by changing the impedance of the antenna.

The air protocol between the reader and tag has a strict timing when the reader stops transmitting information and when the RFID-reader receives the backscattered signal from the tag.

The time between the RFID-reader stops transmitting and the RFID-tag re- sponds is due to the time it takes for the CW to power up the IC on the RFID- tag. It is this window when there is no transmitting between the RFID-reader and the RFID-tag we use to measure the TDoA.

(16)

Chapter 3 Localization with Radio Signals

3.1 Introduction

TDoA is the time difference for a signal to travel from one transmitter to two or more receivers. When we calculate TDoA in a localization model we usually need at least one more receiver than the number of dimensions in which we want to locate the transmitter. The cleaning robot moves in one dimension and the direction of the robot is known. With this knowledge about the robot, we can use a simplified localization model to find the predetermined positions.

3.2 Localization with TDoA

When we calculate the TDoA between one transmitter and two receivers we calculate the time difference of the arriving signal between the receivers. As can be seen in figure 3.1, if we have one transmitter and two receivers and the time difference is positive we get all possible positions on the red hyperbola, if the time difference is negative we get all possible positions on the blue hyperbola and if the time difference is zero we get all possible positions on the yellow line. So, this is not a sufficient localization model in two dimensions, since this gives us infinitely many possible positions along these lines.

To solve this problem, we can add another receiver, as in figure 3.2, and calculate the time difference between receiver 1 and 2, receiver 1 and 3 and receiver 2 and 3. If we combine all possible positions, we get a unique point where these three lines cross each other.

9

(17)

10 CHAPTER 3. LOCALIZATION WITH RADIO SIGNALS

TDoA with two Receivers

TDoA < 0 TDoA > 0 TDoA = 0

Receiver 2 Receiver 1

Figure 3.1: TDoA with two receivers, resulting in infinitely many possible transmitter positions.

TDoA with Three Receivers

Receiver 1

Hyperbola for Receiver 2 and 3

Receiver 3

Receiver 2 Unique

transmitter location

Figure 3.2: TDoA with three receivers, resulting in a unique transmitter position.

(18)

CHAPTER 3. LOCALIZATION WITH RADIO SIGNALS 11

3.3 State Space Model Along a Straight Wall

In our TDoA model we do not calculate the ToA from the RFID-tag and the antennas. Since we start our timers when the first antenna on the robot has received the last bit from the RFID-reader and stop the timer when our receiver receives the signal from the tag. So, the time we measure is the time for the radio wave to travel back and forth and the time to power up the RFID-tag. We then calculate the time difference, which give us the TDoA.

Since we use distances in our state space model and measure TDoA in time, we have converted our TDoA into distance from now on in the report, if nothing else is stated, by multiplying the TDoA with 2.99792458 · 10⁸.

We are not interested in a continuous position of the robot relative to the RFID-tag in this implementation of TDoA. We are only interested in finding a known predetermined position along the wall. This can be achieved if we make a calibration run, which means that we put the robot at each starting referent point and measure the TDoA between the RFID-tags and the antennas.

Since the RFID-tag is fixed and the robot is following a straight wall, see figure 3.3, by using figure 3.4, we can calculate the TDoA of the robot along the wall according to

z = r

(x − s

2)²+ h²− r

(x + s

2)²+ h² (3.1) where:

s[m] is the distance between the antennas, and the notation [m] indicates the unit meters,

h[m] is the distance from the RFID-tag to the robot when T DoA = 0, x[m] is the position of the robot relative the position when T DoA = 0,

where:

x < 0 before the robot has passed the RFID-tag and x > 0 when the robot has passed the RFID-tag and

z[m] is the T DoA measurement.

The TDoA relative the position along the wall can be seen in figure 3.5.

As can be seen in figure 3.5 we do not have infinitely many positions for each TDoA calculations and since we are calculating the time difference between

(19)

Figure 3.3: The robot along the wall

Figure 3.4: Calculation of TDoA

(20)

CHAPTER 3. LOCALIZATION WITH RADIO SIGNALS 13

the second antenna and the first antenna we get a positive TDoA before the robot has passed the RFID-tag and a negative TDoA when the robot has passed the RFID-tag, so we only get one unique TDoA measurement for each position of the robot along the wall. With this model we do not have infinitely many positions for each TDoA measurement and if we know the distance between the antennas and the distance between the RFID-tag and the antennas on the robot, when T DoA = 0, we can calculate a continuous position along the wall relative the RFID-tag.

Equation 3.2 and 3.3 below show our state space model of the robot along the wall. Equation 3.2 shows the state equations where the states are the position of the robot relative the RFID-tag and the velocity of the robot. Equation 3.3 shows the measurement equations where, zk^mis the measurement model of TDoA and vk^mis the measurement model of the velocity of the robot. Since the robot is following a straight wall, the state equations are linear and the measurement equation become nonlinear.

x_k+1 = x_k+ v_kdt

v_k+1 = v_k+ w_k (3.2)

z^m_k = r

(xk− s

2)²+ h² − r

(xk+ s

2)²+ h²+ e¹_k v^m_k = vk+ e²_k

(3.3)

where:

x[m] is the position of the robot relative to the RFID-tag, and the notation [m]

indicates the unit meters, v[^m_s] is the velocity of the robot, dt[s] is the sampling time,

s[m] is the distance between the antennas on the robot,

h[m] is the distance between the RFID-tag and the antennas on the robot when T DoA = 0,

z[m] is TDoA and

w, e¹ and e²are white gaussian noise.

(21)

-20 -15 -10 -5 0 5 10 15 20

Position [m]

-4 -3 -2 -1 0 1 2 3 4

TDoA [s]

10^-9 Calculated TDoA

Calculated TDoA

Figure 3.5: TDoA [s] along the wall where the position is negative before the robot has passed the RFID-tag and the RFID-tag is placed 3 meters from the wall.

(22)

Chapter 4 Radio Receiver with TDoA

4.1 Receiver Architecture

Radio receivers usually consist of different blocks. The received radio signal is filtered in a band pass filter (BPF) to choose a specific frequency band. The signal is then amplified in a Low Noise Amplifier (LNA) before the signal goes through a mixer that mixes the radio signal with a local oscillator (LO), this gives one lower and one higher intermediate frequency according to

f_IF = f_LO± f_RF (4.1)

where f^IF is the intermediate frequency, f^LO is the local oscillator frequency and f^RF is the received radio frequency. There is usually a low pass filter (LPF) after the mixer since the lower intermediate frequency is easier to demodulate.

This receiver design is not suitable for an RFID-reader. Hence, RFID- readers are usually designed slightly different than a usual radio receiver. When the RFID-reader receives the backscattered radio signal from the RFID-tag it also receives the transmitted CW that powers up the RFID-tag. The CW gives us some problems since it has the exact frequency as the signal we want to receive, and it has higher transmitting power. Since the received CW has high power the LNA could make more harm than benefit because the amplifier could operate in the nonlinear region or even be saturated.

Since we know the exact backscatter frequency of the RFID-tag, we can use this frequency as a LO, by using one of the external antenna outputs from the RFID-reader as a LO to the mixer when the RFID-reader transmits. This

15

(23)

16 CHAPTER 4. RADIO RECEIVER WITH TDOA

gives us a pure DC voltage as an intermediate frequency according to equation 4.1. It is easy to filter this signal with a LPF so the BPF before the mixer is unnecessary. The pure DC voltage after the mixer could have a DC voltage offset, caused by the CW, which we must remove.

Since the phase of the backscattered signal varies with 360^◦ when the distance between the reader and the tag changes with half the wavelength,

≈ 16cm for our UHF tags, it is necessary to use a mixer with an I- and Q- branch as output. The Q-branch is mixed with the LO and the I-branch is mixed with the LO signal that is shifted 90^◦. If the received signal from the tag is in quadrature with the LO, the I-branch has zero output voltage. How- ever, the received signal is then in phase with the LO in the Q-branch and produces maximal voltage output. If the signal phase is in between the I- and Q-branches LO, both branches will have an output voltage. Hence, we can expect that the receiving signal strength will depend on the distance between the reader and tag.

4.2 Designed RFID-receiver

Regular RFID-readers sample the received signal with a fixed frequency, which makes it unsuitable to use in this application. Because we want to stop the timer on our ToF module with the received radio signal we need to use the analog signal from the receiver.

Because of this, we decided to use an off the shelf RFID-reader, NORDIC ID SAMPO S2, to communicate with the RFID-tag and design our own receivers for the TDoA measurement. As antennas for the receiver we used a far field antenna from Impinj. As a front end of the receiver we decided to use a combined mixer with an I/Q demodulator, DC1670A, LTC5584IUF from Linear Technology. The mixer is rated from 30MHz up to 1.4GHz but it can be used with lower frequencies with decreased performance.

(24)

CHAPTER 4. RADIO RECEIVER WITH TDOA 17

We designed the receiver circuit in orcad capture and used PCB editor to design our printed circuit board (PCB). We created our own PCB with a method called tone transfer.

The radio architecture of our demodulator is shown in figure 4.1 and the final PCB of the signal demodulator is showed in figure 4.2.

Figure 4.1: Block diagram of the demodulator.

Figure 4.2: Printed circuit board of the signal demodulator.

The signal from the RFID-tag has a transmitting power between −75dBm and −35dBm [19] and we use 1.5 Volt as a reference voltage in the comparator.

If we want to be able to read the signal at −75dBm we need at least 1.5 voltage signal after the last amplifier, according to equation 4.2 and 4.3

(25)

PdBm =10log( P 1m)

→ P =1m10⁽^PdBm¹⁰ ⁾

(4.2)

With equation 4.2 and ohms law and since we are using a 50 ohms system we get

U =√

P · 50 (4.3)

we need at least a total gain of 36500 from each I/Q-input to the comparator.

We use a BLF of 160kHz in the communication between the RFID-tag and RFID-reader. Hence, we need a filter that removes the leaking LO frequency and the image frequency, according to 4.1. The LPF also removes high frequency disturbances.

Before we can amplify the receiving signal, we have to remove the DC- offset, which is caused by the reflections from the CW into our receiver. This was solved by having a capacitor in series on the inputs to our demodulation board.

We use the op-amp LM7171BIMX/NOPB from Texas Instruments in our circuit, since it has a very high slew rate. The slew rate is 950 _µs^V when it is used with ± 5 DC voltage. The high slew rate is important since we want fast response time in the circuitry.

In the I- and Q-branch input to our PCB we use two op-amps as amplifiers with a combined first order LPF, we have a gain of 200 and a cutoff frequency f = 695kHz in the LPF. After the amplifiers in the I- and Q-branch we have an amplifier where we sum the I- and Q-branches and amplify the signal before we compare it with our reference voltage in the comparator. If the amplified signal is above the reference voltage, we saturate the op-amp to +5V and if the signal is below the reference voltage we saturate the op-amp to -5V. After the comparator, we use a diode to remove the negative voltage in the signal such that we only have a positive voltage in the output from the demodulation board.

(26)

We use a diode, 1N4148WX-TP from Micro Commercial Co. This diode has a maximum forward voltage drop of 1V and, hence, the output from our demodulation board is 4V.

As a timer module we use an evaluation board, TDC7201-ZAX-EVM from Texas Instruments. The evaluation board has two timers that can be used sep- arately or with a common start of both timers. The timer module can measure time from 0.25ns up to 8ms in our configuration of the module, it has a resolution of 55ps with a standard deviation of 35ps. This gives us a resolution of 0.0165 meter and a standard deviation of 0.0105 meters in our ToF measurements. Figure 4.3 shows how the resolution of the time measurement affects the resolution of the position for one antenna when the RFID-tag is 3 meters from the wall, where origo is when T DoA = 0.

0 2 4 6 8 10 12 14 16 18

Position [m]

0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

Resolution [m]

Resolution of the Position as a Function of the Position

Resolution

Figure 4.3: Resolution of the position as a function of the position.

(27)

Figure 4.4: Block diagram of the system.

Since we want to monitor the signal from receiver 1 and stop the timers with our signal from the RFID-tag, the signal from receiver 1 is connected both to the timer module and the raspberry pi 3 and receiver 2 is connected to the timer module, as can be seen in 4.4.

Since the raspberry pi 3 and the evaluation board have an input voltage of 3.3V, we have a voltage divider between the demodulation boards and the inputs on the raspberry pi 3 and the timer module.

We used the raspberry pi 3 to monitor the receiving signal from the RFID- tag, such that we know when we should start the common start on our timer module. When we have started the timer, we use the signal from receiver 1 to stop timer 1 and the signal from receiver 2 to stop timer 2. When the timers have been stopped, we use the raspberry pi 3 to read the registers from the timer module and then calculate the time difference between our receiving signal, such that we get our TDoA measurement.

We had a few problems with the design of the receiver which made it impossible to use. We had to solve this problem before we could test the performance of the receiver. The problem was that we got self-oscillation in our receiver when we connected the demodulation board to the timer module. The frequency of the oscillation was approximately 200kHz.

We had a hard time to solve this problem, since we could not figure out why the self oscillation occurred. We could measure a ground current in our circuit, so we supposed that this could be the problem. The first thing we tested was to have a common ground point for the mixer, timer module, the demodulation board and the raspberry pi 3, but it did not solve the self-oscillation problem.

(28)

Since the timer module has a 50Ω resistor, R2 in figure 4.5 from the input to ground. We had to take that into account when we designed the voltage divider. Since the diode on our demodulating board has a maximum forward voltage drop of 1V, we decided to design the voltage divider with an input voltage of 4.4V such that we do not overload the inputs to the Raspberry Pi 3 and the timer module. According to equation

4.4 − R1i − R2i = 0

→ i = 4.4 R₁ + R₂ R₂i = 3.3

→ R₁ = (4.4 − 3.3)R₂ 3.3

(4.4)

R1 should be 16.6Ω. We used a 15Ω resistor, which is a standard value for a resistor. This gave us 65Ω and 68mA to ground.

Figure 4.5: Voltage divider between the demodulating board and the Rasp- berry Pi 3 and the timer module.

(29)

We investigated if this could cause the self-oscillations in our circuit. By desoldering the 50Ω resistor on the timer module and design a new voltage divider with higher values such we get a lower current into the ground in our circuit. This solution solved our problems with the self-oscillations in the receiver.

There is one important thing that is almost impossible to have control over in the design of the receiver and has a huge impact on the accuracy of the TDoA measurements. That is the total length that the signal must travel inside the receiver. In our case, we have coaxial cables between the mixer and our demodulation board and between the demodulation board and the timer module. We also have the coaxial cables between the antennas and the mixer. If the coaxial cables on both receivers do not have the same length, it will affect the accuracy of the measurements.

If the mixer, demodulator and the timer module are at the same PCB it is easier to control the total length since it is easy to have accurate length of the traces on a PCB. Then the only source of error is the coaxial cable between the antenna and the mixer.

This is not an impossible task to solve since the coaxial cables can be constructed as accurately as possible. It is also possible to calibrate the measurements in the software. If we place the RFID-tag were we should measure T DoA = 0 and take a measurement, then we know how much we have to add or subtract from our measurements.

4.3 Measure TDoA

When we measure the TDoA, we start the common start of the timers when the RFID-reader stops transmitting the signal and stop the timers when our receivers receives the signal from the tag. Hence, when antenna 1 has received the last bit from the RFID-reader we start the command start of the timers, and we stop the timers when the receivers have received the first bit from the RFID-tag.

In figure 4.6 we can see an example where the Raspberry Pi starts the timer and where the timers are stopped by the received signal from the RFID-tag.

We then calculate the TDoA according to T DoA = T imerec1 − T ime_rec2. We get a positive measurement before the robot has reached the RFID-tag and

(30)

a negative value when the robot has passed the RFID-tag, which can be seen in figure 3.5.

To identify the signal from the RFID-reader and the signal from the RFID- tag we can monitor the encoding of the received signal. We use PIE encoding when the RFID-reader is talking with the RFID-tag and MMS-8 encoding when the RFID-tag is talking with the RFID-reader. Hence, if we identify the received signal as PIE encoding, we know that it is the RFID-reader that is talking.

As can be seen in figure 4.6, if we identify the signal as PIE encoding and the RFID-reader stops transmitting we know that we can start the timers with the raspberry pi 3 and wait for the timer to be stopped by the receiving signal from the RFID-tag. The pseudocode for the TDoA measurement is shown in algorithm 1.

Figure 4.7 and 4.8 show the encoding of the signal were we start and stop the timer. The RFID-reader’s transmitted signal is shown in figure 4.7 and can be identified as PIE encoding and the RFID-tag’s transmitted signal is shown in figure 4.8, and it can be identified as MMS-8 encoding. In figure 4.8 we can see that the first bit that the receiver receives is not MMS-8 encoded. Hence, the first signal that is received is probably disturbances. This can cause problems when we should stop our timers. If both receivers receive this disturbance it should not cause any problem.

Algorithm 1 Measure TDoA while encoding = PIE do

if transmitting stops then start common timer

while timer not stopped do if timer stoped then

read register calculate TDoA end

end

(31)

2.815 2.82 2.825 2.83 2.835 2.84 2.845 2.85 2.855

Samples 10⁶

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1

Binary amplitude

Received Signal

Received Signal Start

Stop

Figure 4.6: Received signal from the tag and the RFID-reader.

5.213 5.2132 5.2134 5.2136 5.2138 5.214

Sample ₁₀⁶

0.994 0.996 0.998 1 1.002 1.004

Binary amplitude

PIE Encoding

Start

Figure 4.7: Transmitted signal with PIE encoding.

(32)

4.0154 4.0155 4.0156 4.0157 4.0158

Sample ₁₀⁶

0.99999985 0.9999999 0.99999995 1 1.00000005

Binary amplitude

MMS-8 Encoding

Stop

Figure 4.8: Received signal with MMS-8 encoding.

(33)

Chapter 5 Improve the TDoA Measurement

5.1 Kalman Filter

To improve the TDoA measurement we can use a Kalman filter to estimate the TDoA.

The inputs in the state equations, equation 3.2, are the previous position of the robot, the velocity of the robot and the sampling time, where the sampling time is constant.

The inputs to the measurement equations, equation 3.3, are the position of the robot, the distance between the antennas and the distance between the tag and the antennas on the robot when T DoA = 0 and the velocity of the robot, where the distance between antennas, the distance between the tag and the antennas on the robot when T DoA = 0 is constant. The only inputs we need to measure are the TDoA and the velocity of the robot. The velocity can be easily measured with the encoders on the robot’s wheels.

The Kalman filter is unnecessary to use if the measurements are consis- tent and accurate. But, since there is always noise in the radio signal some measurements could be far from the actual TDoA, or we could even lose a few measurements. If so, we must estimate our TDoA. We chose to use a Kalman filter to estimate the position of the robot relative the RFID-tag. The estimation of the position is used in equation 3.1 to calculate the estimate of TDoA.

We need an Extended Kalman Filter (EKF) in our estimate of the position since our measurement equations are nonlinear.

26

(34)

CHAPTER 5. IMPROVE THE TDOA MEASUREMENT 27

Our states in the state equation are the position and the velocity. Hence, we can estimate both the position and velocity. The algorithm for the EKF is shown in algorithm 2, where we estimate both the position and the velocity.

Algorithm 2 Extended Kalman filter Initialization:

Xˆinit =x_init vinit

, P, Q, R, dt

Extended Kalman filter(Xk−1, P_k−1, z_k):

Prediction:

Xˆ^k = ˆx_k ˆ v_k

=x_k−1+ v_kdt v_k−1

Pˆ^k = FkPk−1F_k^T + Q Update:

Kk = ˆP_kH_k^T(H_kPˆ_kH_k^T + R)⁻¹ Xk = ˆX_k+ K_k(z_k− h(ˆx_k, ˆv_k)) P^k = (I − KkHk) ˆPk

return Xk, P_k

where:

X_initis the initialization matrix of the position for the robot relative the RFID-tag and the velocity,

x the position for the robot relative the RFID-tag, v is the velocity of the robot,

P is the predicted estimate of the state error covariance matrix,ˆ P is the updated estimate of the state error covariance matrix, Q is the process noise covariance matrix,

R is the measurement noise covariance matrix, dt is the sampling time,

K is the Kalman gain,

F is the jacobian of the state model,

H is the jacobian of the measurement model, h is the measurement model matrix and

z is the measurement matrix of TDoA and the velocity.

(35)

28 CHAPTER 5. IMPROVE THE TDOA MEASUREMENT

We will simulate four scenarios to investigate the performance of the Kalman filter:

• Investigate the converging time for the Kalman filter.

• Estimate our TDoA measurement when we have noisy measurements.

• Investigate the TDoA estimation when we have one noisy measurement every two seconds.

• Investigate the performance of the Kalman filter when we have few measurements which are far from the actual TDoA.

(36)

Chapter 6 Results

6.1 Simulation

We simulated all scenarios in this section with these parameters:

h = 3m, s = 1m.

We used a slightly different EKF than the EKF shown in algorithm 2, we used a constant velocity and estimate the position of the robot. Hence, the parameters in the EKF are one dimensional.

The parameters are:

x_init = −2[m], P = 0.1, Q = 0.2, R = 1, dt = 0.1[s] and v = 0.2[^m_s].

Since we use a constant velocity we do not have any measurements of the velocity. Hence, the only measurement noise is the noise in the TDoA measurement, which is gaussian noise with µ = 0m and σ = 0.06m.

We used a constant velocity since we focus on the measurements and the estimate of the TDoA.

To verify the state space model we compared the calculated TDoA, equation 3.1, and the simulation along the wall. As can be seen in figure 6.1 we can confirm that our state space model is accurate.

The EKF must converge as fast as possible since the robot can be initiated close to the desired starting reference point of the cleaning.

We simulated the EKF without noise in the measurements, as can be seen in figure 6.2. The estimation of TDoA has converged at -3 meters when the initial value of the position in the EKF is -2 meters. If our starting reference

29

(37)

30 CHAPTER 6. RESULTS

-20 -15 -10 -5 0 5 10 15 20

Position [m]

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

TDoA [m]

Calculation and Simulation of the State Space Model

Calculated TDoA

Simulated TDoA with the state space model

Figure 6.1: Calculated and simulated TDoA

-5 -4 -3 -2 -1 0 1 2 3 4 5

Position [m]

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

TDoA [m]

Simulation of the Kalman Filter

Actual TDoA Estimate Measurement

Figure 6.2: Simulation of the Kalman filter without noisy measurements.

(38)

CHAPTER 6. RESULTS 31

point would be at -4meters, this will cause problems. We can conclude that it is not an optimal solution to have a fixed initial value in the EKF. A solution to this problem could be to have a dynamic initial value of the position in the EKF.

A more optimal initial value can be found by measuring the TDoA where the robot is initiated and solve for x in equation 3.1 to calculate the initial value of the position.

If the measurements of TDoA are noisy we can estimate our TDoA by using the EKF as can be seen in figure 6.3. The converging time is the same as when we simulated without noise in the measurements. As can be seen in figure 6.3 the estimation of the TDoA has improved compared to the noisy measurements. This can, of course, be a tradeoff between the measurements and the model, if the model is more accurate, we can weigh it higher in the EKF and if the measurements are more reliable, we can weigh the measurements higher in the EKF.

-5 -4 -3 -2 -1 0 1 2 3 4 5

Position [m]

-1 -0.5 0 0.5 1 1.5

TDoA [m]

Figure 6.3: Simulation of the Kalman filter with noisy measurements.

(39)

-5 -4 -3 -2 -1 0 1 2 3 4 5

Position [m]

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

TDoA [m]

Figure 6.4: Simulation of the Kalman filter with one noisy measurement every two seconds.

In the simulation shown in figure 6.4 we only have one measurement every two seconds. We use the EKF when we get a new measurement and between the measurements, we predict our state with the state space model, the estimate is still accurate. There are two reasons for this, the first reason is that we have higher reliability on the model than the measurements in the EKF. Hence, the measurements do not affect the estimate of TDoA too much. The second reason is that the estimate of TDoA is close to the actual TDoA when we start to predict the TDoA with the state space model. Hence, we will follow the actual value of the TDoA.

We have investigated the scenario that some measurements are far from the actual TDoA. In figure 6.5 we can see that this affects our estimate, where we have these erroneous measurements our estimate is far from the actual TDoA.

There are two solutions for this problem, either we can change the weights between the model and the measurement such we have higher reliability on our state space model. However, this solution is not what we would prefer, since it will affect the overall performance of the Kalman filter when we do not have these erroneous measurements.

(40)

The other solution would be that we detect the erroneous measurements, by comparing the current measurement with the previous estimate of the measurement. If the absolute value of the difference between the measurement and the previous estimate of TDoA is above a threshold, we do not use the EKF and instead predict our TDoA with our state state space model.

-5 -4 -3 -2 -1 0 1 2 3 4 5

Position [m]

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

TDoA [m]

(41)

-0.3 -0.2 -0.1 0 0.1 0.2 0.3

Position [m]

-0.1 -0.08 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1

TDoA

TDoA at the starting reference point

This solution is equivalent to the simulation where we only have one measurement every two seconds. As shown in figure 6.4, if we ignore the erroneous measurements and only predict our state with the state space model between these measurements and use the EKF when we have accurate measurements, this would give us higher accuracy of the estimate.

Hence, if we have noisy measurements or miss some measurements we can conclude that it is important to use some sort of filter to have a reliable measurement of the TDoA, such that we can find our predetermined starting referent point for the robot within 2 cm.

However, as we can see in figure 6.6, if our predetermined starting reference point would be T DoA = 0, we would not fulfill the requirement without the EKF, since our TDoA measurement is zero at 25cm from our starting reference point. However, if we use the EKF we would fulfill the requirement for the starting reference point since our estimation of TDoA is within (0 ± 2)cm in this simulation.

(42)

6.2 Receiver and TDoA Measurement

To verify the performance of the receiver we simulated the receiving signal from the RFID-tag with a function generator, RIGOL DG1022, and measured the signal after the demodulator with an oscilloscope, Tektronix 453A.

We connected the outputs from the function generator to the LO and the RF inputs on the mixer. We used one channel as a local oscillator with a frequency of 25M Hz, which is the highest frequency with this function generator, and the other channel as the receiving signal from the RFID-tag, with a carrier frequency of 25M Hz and ASK-modulation with a frequency of 160kHz. We could see the signal at the oscilloscope down to −52dBm which is the lowest signal level that this function generator can generate. Hence, we can conclude that the performance of the receiver is good with an ideal voltage source.

We could not measure the signal from the RFID-tag with the oscilloscope, since it was not possible to distinguish the signal from the RFID-tag and the signal from the RFID-reader. However, we could see signals that was ASK- modulated, so we know that we can receive radio signals.

The LO and RF inputs on the mixer have a maximum input power of 20dBm. Hence, if we used the highest power on the RFID-reader, we would overload the LO input on the mixer, since it is connected to an external antenna output from the RFID-reader. However, we could sample the signal from the RFID-tag if we were using a lower output power from the RFID-reader. To be sure we do not overload the mixer we used an output power of 17dBm, resulting in a reading range <0.4m. We used the Raspberry Pi to sample the signal. That sampled signal can be seen in figure 4.6.

Since the antennas for our receivers should be mounted approximately 1 meter apart, we had to increase the reading range from the RFID-tag to our antennas. We investigated the maximum reading range by using the maximum input power to the mixer. 20dBm gave us only 0.4m.

When we used the upper limit for our mixer-board it short-circuited after a few experiments. The cause of the short-circuit can be two things, either the signal from the RF-input or the LO-input to the mixer-board.

Because we had reached the input power limit of the LO-input at 20dBm

(43)

we need to have different signal strength on the external antenna outputs and the internal antenna, since we used the external antenna output from the RFID- reader as the LO input to the mixer, this problem were known before the short- circuit, this can easily be solved by programing the RFID-reader such that it has different output power on internal and external antennas. The other problem could be if the RFID-readers transmitting signal go through our antennas into the RF-input at the mixer-board. If the problem is from the RF-input on the mixer board it can be more complicated to solve especially if we have problems with the output signal power at 20dBm and we want to use the maximum signal power from the RFID-reader which is 27dBm, to have maximum reading range. To solve this problem, we could measure the signal strength from the RFID-reader at the position were the antennas for our receivers are supposed to be mounted. If the signal strength is too high, the solution might be to change antennas with antennas with smaller beam width than the internal antenna of our RFID-reader, which is 80^◦. Another solution could be to use same antennas for our receivers and for the RFID-receiver.

(44)

Chapter 7 Discussion and Conclusions

7.1 Discussion

The state space model we have developed could be invalid in some cases. Since the environment that the robot is cleaning is wet and the wall is not absolutely straight.

The robot could slip such that the measurement of the velocity could be invalid. If the wall is not absolutely straight the TDoA measurement will vary since the distance between the antennas and the RFID-tag will vary.

Another problem with our state space model is that we need to know the distance between the wall and the RFID-tag. That is not an optimal solution when the system should be installed at the customer’s barns. However, we think it is a problem that has to be solved such that we can use the Kalman filter to achieve the required accuracy of the system.

As can be seen in figure 4.3, if our starting reference point is T DoA = 0, the resolution in the position is around 0.09 meter. Since we want to find the starting reference point within 0.02 meter the desired position would be above 2 meters from the position when T DoA = 0. However as can be seen in figure 6.2, if our starting reference point is above 2 meters from the origo the curve of the TDoA measurements decreases.

Hence, we can conclude that if we want as accurate positioning as possible we want to have our starting reference point as close as possible to 2 meters from where T DoA = 0. This makes the Kalman filter even more important since there is not only the noise in the measurements but also the resolution that affects the positioning.

37

(45)

38 CHAPTER 7. DISCUSSION AND CONCLUSIONS

There is another thing to account for in the state space model, that is the length of the coaxial cables between the antenna and the receiver. In the state space, model we use the time between the RFID-tag and the antennas, but the time we get from the timer module includes the time for the signal to travel inside the coaxial cables, If the state space should be valid we must subtract the time for the signal to travel inside the coaxial cables from the time we get from the timer module.

There are a few problems in this project that must be solved before the TDoA measurement can be tested. The main problem is the input power to the mixer. The first thing that must be done is to decrease the output power on the external antenna port on the RFID-reader such that it is below 20dBm.

The second thing that must be done is to measure the signal power from the RFID-reader where the receiver antennas are mounted on the robot. It could be done by measuring the signal strength in steps to know how much signal power that can be used before it reaches the limit of 20dBm for the RF-input port on the mixer board. If the signal power is too low such that it is not possible to have enough reading range, then it must be investigated to find a solution to the problem. One solution could be to use another antenna with smaller beam width than the internal antenna in the RFID-reader.

Another solution and one that we consider as the best solution is to use only two antennas, which are the antennas for our own designed radio receivers.

This is also preferred since the system will be cheaper and easier to mount on the robot. However, with this solution, the receivers must be redesigned.

Since the algorithm for when we want to start the timers is written in python it could be slow sometimes. It can be a good idea to write it in c++ or even have a microprocessor that is only used to monitor the radio signal and start the common start on the timer module instead of the Raspberry Pi and use the Raspberry Pi to communicate with the timer module and to estimate the TDoA measurement with the EKF.

To make this system robust it needs both radio design knowledge and programming skills. Since there is mostly practical work left, we consider the best solution might be to have two candidate theses. One for the radio design and one for the programming part, which could work in parallel to finish the project.

(46)

CHAPTER 7. DISCUSSION AND CONCLUSIONS 39

We have not fulfilled the final requirements in this master thesis, which are specified in section 1.2.

The reading distance between the RFID-tag and our receivers is only 0.4 meter. The cause of this is that we could not use maximum transmitting power with our RFID-reader, since we short-circuit our mixer at one fifth of the transmitting power.

We could not fulfill the second requirement since we needed a final and functional solution to be able to test it. However, since there will be noise in our measurements we think it will be difficult to fulfill this requirement without any filter that can estimate the measurement. As can be seen in our results we could fulfill these requirements in simulation with an arbitrary noise level in the signal. However, there is no guarantee that it will be fulfilled in the final application.

7.2 Conclusions

We have in this master thesis developed a positioning system for a robot that follows a straight wall indoors. The system measures the TDoA between a RFID-tag and two antennas that are mounted on the robot. We have developed the theoretical and some of the practical aspects of the TDoA measurements such that the robot can find a predetermined position along the wall.

In the theoretical part of the master thesis, we have developed a state space model for the TDoF along the wall. Since there is always noise in the radio signal from the RFID-tag, we have developed a Kalman filter such that we can estimate and improve our measurement of the TDoA.

In the practical part of the master thesis, we developed and constructed two receivers for the antennas that we are using for the TDoA measurements.

We ended up with some problems such that we could not test the TDoA measurements. Hence, there is still a lot of developing to do before it is possible to verify the accuracy of the system. We need to increase the reading distance between the RFID-tag and the receivers for the TDoA measurements.

The power of the radio signal input from the RFID-reader into the receivers must be investigated. If necessary, change to antennas with smaller beam width or change the position of the antennas for both the RFID-reader and

(47)

40 CHAPTER 7. DISCUSSION AND CONCLUSIONS

the receivers for the TDoA measurements.

TDoA measurements is not a new method to localize a transmitter. How- ever, this is as far as we know a new method to use RFID-tags as a transmitter to measure TDoA and use it such that a robot can find its own position in one dimension. Hence, it would be interesting to know the accuracy of the method and if it is accurate enough such that it is possible to use in sensor fusion, to make it possible for the robot to find its own position with higher accuracy.

(48)

Bibliography

[1] Wikipedia contributors. Bluetooth low energy beacon — Wikipedia, The Free Encyclopedia. [Online; accessed 29-November-2019]. 2019. url:

https : / / en . wikipedia . org / w / index . php ? title = Bluetooth_low_energy_beacon&oldid=917185256.

[2] Wikipedia contributors. Angle of arrival — Wikipedia, The Free Ency- clopedia. [Online; accessed 29-November-2019]. 2019. url: https:

//en.wikipedia.org/w/index.php?title=Angle_of_

arrival&oldid=926728264.

[3] Wikipedia contributors. Time of arrival — Wikipedia, The Free Ency- clopedia. [Online; accessed 29-November-2019]. 2019. url: https:

//en.wikipedia.org/w/index.php?title=Time_ of_

arrival&oldid=923877422.

[4] Sebastian Schneegans, Philipp Vorst, and Andreas Zell. “Using RFID Snapshots for Mobile Robot Self-Localization.” In: EMCR, Jan. 2007.

[5] D. Lymberopoulos and J. Liu. “The Microsoft Indoor Localization Com- petition: Experiences and Lessons Learned”. In: IEEE Signal Process- ing Magazine 34.5 (Sept. 2017), pp. 125–140.

[6] M M Zaman Tanim. “How does passive RFID works, briefly explained.”

In: (Jan. 2016). doi: 10.13140/RG.2.2.12361.34402.

[7] Anritsu. Time Difference of Arrival (TDOA). 2017. url: https://

www . rcrwireless . com / wireless / time - difference - of-arrival-tdoa (visited on 12/15/2019).

[8] L. Görtschacher et al. “SDR based RFID reader for passive tag local- ization using phase difference of arrival techniques”. In: IEEE MTT-S International Microwave Symposium (2016).

[9] C. Hekimian-Williams et al. “Accurate localization of RFID tags using phase difference”. In: 2010 IEEE International Conference on RFID (IEEE RFID 2010). Apr. 2010, pp. 89–96.

41

(49)

42 BIBLIOGRAPHY

[10] Mahmoud Mohanna et al. “Optimization of MUSIC algorithm for an- gle of arrival estimation in wireless communications”. In: NRIAG Jour- nal of Astronomy and Geophysics 2.1 (2013), pp. 116–124. doi: 10.

1016/j.nrjag.2013.06.014. eprint: https://doi.org/

10 . 1016 / j . nrjag . 2013 . 06 . 014. url: https : / / doi . org/10.1016/j.nrjag.2013.06.014.

[11] H.K. Hwang et al. “Direction of Arrival Estimation using a Root-MUSIC Algorithm”. In: Lecture Notes in Engineering and Computer Science 2169 (Mar. 2008).

[12] Ambili Parameswaran, Mohammad Iftekhar Husain, and Shambhu Upad- hyaya. “Is RSSI a reliable parameter in sensor localization algorithms:

an experimental study”. In: Field Failure Data Analysis Workshop (F2DA’09) (Jan. 2009).

[13] T. Ussmueller et al. “A multistandard HF/ UHF-RFID-tag with inte- grated sensor interface and localization capability”. In: 2012 IEEE In- ternational Conference on RFID (RFID). Apr. 2012, pp. 66–73.

[14] T. Ussmueller et al. “Roundtrip-Time-of-Flight based localization of passive multi-standard RFID-tags”. In: 2012 IEEE International Con- ference on Wireless Information Technology and Systems (ICWITS).

Nov. 2012, pp. 1–4.

[15] Stephen A. Weis. “RFID (Radio Frequency Identification): Principles and Applications”. In: MIT CSAIL (2007).

[16] atlasRFIDstore. A Guide to RFID Types and How They Are Used. 2018.

url: https : / / www . atlasrfidstore . com / a - guide - to - rfid - types - and - how - they - are - used/ (visited on 12/31/2019).

[17] National Instruments. Amplitude-Shift Keying, Frequency-Shift Keying, and Phase-Shift Keying. 2019. url: https : / / www . ni . com / sv - se / innovations / white - papers / 08 / amplitude - shift-keying--frequency-shift-keying--and-phase- shift-.html (visited on 11/29/2019).

[18] Suzanne Smiley. Explaining Backscatter – From Basic to Advanced Principles. 2018. url: https://blog.atlasrfidstore.com/

explaining-backscatter-from-basic-to-advanced- principles (visited on 11/29/2019).

[19] Daniel Dobkin. The RF in RFID. UHF RFID in Practice. Newnes, 2012.

(50)

TRITA-EECS-EX-2020:11 www.kth.se

Robot Localization with Radio Frequency Identification and Time Difference of Arrival

Robot Localization with Radio

Frequency Identification and Time Difference of Arrival

HÅKAN GÖRANSSON

Robot Localization with Radio Frequency Identification and Time Difference of Arrival

HÅKAN GÖRANSSON

Abstract

Sammanfattning

Contents

Chapter 1 Introduction

1.1 Background

1.2 Thesis Aim

1.3 Previous Work

1.4 Outline

Chapter 2 RFID

2.1 Introduction

2.2 RFID Technology

Chapter 3

Localization with Radio Signals

3.1 Introduction

3.2 Localization with TDoA

3.3 State Space Model Along a Straight Wall

Chapter 4

Radio Receiver with TDoA

4.1 Receiver Architecture

4.2 Designed RFID-receiver

4.3 Measure TDoA

Chapter 5

Improve the TDoA Measurement

5.1 Kalman Filter

Chapter 6 Results

6.1 Simulation

6.2 Receiver and TDoA Measurement

Chapter 7

Discussion and Conclusions

7.1 Discussion

7.2 Conclusions

Bibliography