UPTEC F 19002
Examensarbete 30 hp April 2019
Battery-free Visible Light Sensing
Andreas Soleiman
Teknisk- naturvetenskaplig fakultet UTH-enheten
Besöksadress:
Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0
Postadress:
Box 536 751 21 Uppsala
Telefon:
018 – 471 30 03
Telefax:
018 – 471 30 00
Hemsida:
http://www.teknat.uu.se/student
Abstract
Battery-free Visible Light Sensing
Andreas Soleiman
In this thesis, we show that it is possible to design a battery- free light sensing system that can sense and communicate hand gestures while operating fully on harvested power from indoor light. We present two main innovations that push our system to tens of microwatts of power to enable battery-free operation.
First, we introduce a novel visible light sensing system that can track variations in light intensity by using a solar cell as a sensor. Solar cells are unlike photodiodes optimized for energy yield in the visible light region and hence do not require any power hungry active components such as an operational amplifier.
Furthermore, solar cells can operate under more diverse light conditions as they are not susceptible to saturation under bright light. Second, we devise two ultra-low power communication mechanisms based on radio frequency backscatter to transmit sensor readings at various resolutions without the need of any energy- expensive computational blocks. We design two battery-free and self-powered hardware prototypes that are based on these two innovations. Our first design utilizes an on-board comparator based circuit to perform a 1-bit digitization of changes in light readings, consuming only sub-microwatt of power for digitization.
For our second prototype, we design an analog backscatter
mechanism that can map raw sensor readings directly to backscatter transmissions. We demonstrate the feasibility of our designs when sensing significant changes in light intensity caused by shadows from hand gestures, and reconstruct these at a receiving device.
Our results demonstrate the ability to sense and communicate various hand gestures at a peak power of 20 microwatts when performing 1-bit digitization, and a mean power of 60 microwatts when performing analog backscatter. Both designs represent orders of magnitude improvement in terms of power consumption over state- of-the-art visible light sensing systems.
Handledare: Ambuj Varshney
Popul¨ arvetenskaplig Sammanfattning
I dagens moderna samh¨alle existerar det ett flertal miljarder vardagliga f¨orem˚ al som ¨ar uppkopplade till internet och som har f¨orm˚ agan att kommunicera tr˚ adl¨ost med varandra. Dessa typer av f¨orem˚ al h¨or till det som kallas f¨or Internet of Things (IoT), eller Sakernas Internet. De kan vara allt fr˚ an smarta termostat som m¨ater temperatur i hemmet och kommunicerar detta tr˚ adl¨ost, k¨oksredskap som smarta kaffebryggare eller mikrov˚ agsugnar med mera, och smart belysning som kan kontrolleras av mobiltelefoner. Dock existerar det m˚ anga problem som beh¨over hanteras d˚ a det finns s˚ a m˚ anga uppkopplade f¨orem˚ al som ¨ar tillg¨angliga: alla konkurerrar om en plats i n¨atverket f¨or att kommunicera eller ta emot sensordata, vilket s¨atter st¨orre begr¨ansningar f¨or aktiv kommunikationstid och ¨okar risken f¨or interferens mellan tr˚ adl¨osa signaler. Ut¨over detta uppst˚ ar s¨akerhetsrisker d˚ a fler f¨orem˚ al kommunicerar k¨anslig information ¨over internet, eftersom det skapar m¨ojlighter f¨or obeh¨origa att f˚ a tillg˚ ang till k¨anslig data eller ta kontroll ¨over olika apparater i hemmet.
Forts¨attningsvis inneb¨ar en numer¨ar ¨okning av antalet uppkopplade f¨orem˚ al ocks˚ a en drastisk ¨okning av sammantagen energif¨orbrukning. Detta strider mot en h˚ allbar utveckling eftersom det s¨atter st¨orre krav p˚ a utvinning av resurser; mer energi g˚ ar ˚ at till att driva en ¨okning av antalet IoT enheter samtidigt som nya m˚ aste tillverkas f¨or att ers¨atta de som slutar att fungera. Dessa faktum resulterar i att uppkopplade f¨orem˚ al tvingas att antingen drivas ur eluttag, vilket begr¨ansar
˚ atkomligheten av f¨orem˚ alen till en dedikerad plats d¨ar uttagen finns, eller genom batterier som s¨atter krav p˚ a materialutvinning och eventuella kostnader f¨or un- derh˚ all eftersom de frekvent beh¨over ers¨attas. Dessa faktorer skapar sin tur stora utmaningar f¨or att integrera sensorer i sv˚ ar˚ atkomliga platser, exempelvis i mark, v¨aggar och tak.
1
Detta projekt har till ¨andam˚ al att tackla kritiska milj¨oproblem som uppst˚ ar i samband med en drastisk ¨okning av IoT enheter i syfte att m¨ojligg¨ora en h˚ allbar utveckling av IoT. Detta kr¨aver en innovativ utveckling av nya former av sensorer och mekanismer som ¨ar energieffektiva i praktiken och som kan h˚ alla i l˚ ang tid utan att beh¨ova underh˚ allas. En viktig l¨osning som detta projekt tar fram ¨ar att reducera komplexiteten i sensorer genom att eliminera energikr¨avande kompo- nenter som de processorer som bearbetar sensordata lokalt. Genom att ta bort dessa processorer och ist¨allet l˚ ata uppkopplade enheter kommunicera sin data di- rekt till en mottagare, utan att bearbeta det lokalt, och utan att kompromissa p˚ a funktionalitet, s˚ a skapar det m¨ojlighet f¨or enheterna att f¨orbruka minimalt med energi. Denna energi kan enkelt samlas in fr˚ an omgivningen och d¨arefter driva komponenterna tillr¨ackligt l¨ange f¨or att uppn˚ a ¨onskad funktionalitet. Potentialen i den funktionen ¨ar att det kan eliminera batterier och d¨armed l¨osa ett av de allra mest kritiska problemen n¨ar det kommer till energif¨orbrukning. Detta ¨ar vad vi kallar f¨or batterifria system, och de har potentialen att kunna existera v¨al f¨orbi en m¨anniskas livsl¨angd.
D¨armed g¨ors ett f¨ors¨ok till att svara p˚ a f¨oljande fr˚ aga: ¨ Ar det m¨ojligt att konstruera ett batterifritt tr˚ adl¨ost igenk¨anningssystem som kan utplaceras i stor skala och p˚ a ett h˚ allbart s¨att?
Projektetet bidrar med ett svar p˚ a denna fr˚ aga genom att modellera och designa
ett tr˚ adl¨ost och batterifritt elektroniskt system som kan utnyttja synligt ljus fr˚ an
inomhusbelysning, b˚ ade som k¨alla till energi f¨or att g¨ora systemet sj¨alvdrivande,
men ocks˚ a f¨or igenk¨anning av skuggor. En originell applikation som tas fram i detta
projekt ¨ar att detektera handgester ur deras unika skuggm¨onster n¨ar dessa skuggor
faller ¨over en liten tunnfilmssolcell. D¨arefter kommuniceras denna information
tr˚ adl¨ost till en mottagare som kan identifiera typen av handgest. Detta kan g¨ora
det m¨ojligt f¨or applikationsscenarion som exempelvis att styra olika funktioner
i smarta hem med hj¨alp av handgester som utf¨ors ovanf¨or solceller. Solcellerna anv¨ands med andra ord b˚ ade som enheter f¨or att samla in energi, men ocks˚ a som sensorer d˚ a de kan m¨ata avvikelser i ljusintensitet p˚ a grund av skuggor.
H¨ar utnyttjas ocks˚ a en ny form av tr˚ adl¨os kommunikationsmekanism som kallas f¨or backscatter. Backscatter-antenner absorberar eller reflekterar signaler i om- givningen f¨or att kommunicera ny information, ist¨allet f¨or att generera dessa sig- naler i aktiva radiochip som konsumerar relativt mycket energi. D¨armed sparar backscatter-antenner tiotals ordningar mer i energim¨angd j¨amf¨ort med vanliga radioantenner. De tr˚ adl¨osa signalerna som utnyttjas f¨or backscattering kan ex- empelvis h¨arstamma fr˚ an enheter som ¨ar uppkopplade via WiFi eller Bluetooth, eller FM signaler som genereras i radiotorn. Potentialen f¨or denna typ av kommu- nikation ¨ar enorm, eftersom en ¨okning av antalet uppkopplade signaler ocks˚ a leder till en ¨okning av m¨angden elektromagnetiska signaler som befinner sig i rymden.
D¨arf¨or r˚ ader det inga tvivel om att just backscattering ¨ar en framtida mekanism som ¨ar p˚ a v¨ag att etablera sig.
3
Acknowledgement
The research presented in this thesis project was performed under the supervision of Dr. Ambuj Varshney and Professor Thiemo Voigt. The outcome of this research was published at the ACM Workshop on Visible Light Communication Systems (ACM VLCS), which was held in conjunction with ACM MobiCom 2017, and where it won the best paper award. Moreover, this research project allowed me to participate in the graduate category of the ACM Student Research Competition (ACM SRC) at ACM MobiCom 2017, after submitting a poster of our work. I was very happy to win the gold medal in this competition, and it further allowed me to participate in the ACM SRC Grand Finals where I could compete with highly skillful and motivated SRC winners from other prestigious computer science venues. In addition, a follow-up of the research project which used our concepts of analog backscatter was also published as a demonstration at ACM WiSec 2018, and won the best demo award.
During the course of this thesis project, I have been very fortunate to have the opportunity and freedom to spend qualitative hours in a research lab at Uppsala University, in close proximity to my research advisers. I would first like to thank Professor Thiemo Voigt for entrusting me with this fantastic opportunity to work full time as a research assistant in a highly competitive research environment.
Thanks to his confidence in the quality of this project and my abilities, I have been able to pursue research as a profession while working on my thesis project.
Next, I would like to send a very special and warm thanks to my supervisor Dr.
Ambuj Varshney, who have spent countless hours working closely with me, and
teaching me the ways of systems research. He has through his effort helped wire me
to think analytically and critically about groundbreaking research problems. But
more importantly, he has always maintained faith in my progress, while motivating
me during my most difficult times. This is more than I could ever ask from a supervisor. Further, through my many interactions with Ambuj, I have developed the confidence to believe that I can achieve success in everything I decide to do with pure dedication and hard work, and by having a high fault tolerance. For that, I am deeply thankful. I am certain that this mentality will benefit me greatly in all my future endeavors. Finally, I would like to thank all my colleagues at the department; namely Kasun for his guidance in hardware design and patience with my incompetence, Carlos for sharing his deep theoretical knowledge in radio concepts, and Sam, Lorenzo, George, and Haris for being my close friends to share general knowledge and beers with.
5
Contents
1 Introduction 21
1.1 Motivation . . . 22
1.2 Problem Statement . . . 24
2 Background 27 2.1 Visible Light Sensing . . . 27
2.2 Radio Frequency Backscatter . . . 28
2.3 Mitigating Computational Overhead . . . 35
3 Related Work 37 3.1 Visible Light Sensing . . . 37
3.2 Ultra-low Power Gesture Recognition . . . 40
3.3 Radio-Frequency Backscatter . . . 41
3.4 Ambient Backscatter . . . 42
3.5 Eliminating Computational Blocks . . . 43
3.6 Analog Backscatter . . . 44
4 System Design 45 4.1 Overview . . . 45
4.2 RF-Emitter . . . 50
4.3 Backscatter Frontend . . . 53
7
4.4 Thresholding Visible Light Marker . . . 55
4.5 Analog Visible Light Marker . . . 58
4.6 Signal Reception and Processing . . . 63
4.7 FM Backscatter Peak Detection . . . 67
5 Experimental Evaluation 69 5.1 Light Sensor Performance . . . 70
5.2 Light Sensor Responsiveness at Changing Light Conditions . . . 71
5.3 Power Consumption of the VLMs . . . 73
5.4 Energy Harvesting Performance . . . 73
5.5 Analog Backscatter . . . 76
5.6 Ambient FM Backscatter . . . 81
5.7 Application Scenario: Hand Gesture Sensing . . . 84
6 Discussion 87 6.1 Discussion . . . 87
7 Conclusion and Future Work 91 7.1 Conclusion . . . 91
7.2 Research Challenges and Future Work . . . 92
List of Figures
2-1 The visible spectrum. The figure shows the region of the electro- magnetic spectrum at which radiation, with range of wavelengths between approximately 400 to 700 nm, is visible to the human eye. . 28 2-2 Demonstration of the basic principle of visible light sensing.
In this scenario, a moving hand obstructs a light sensor by casting a shadow over it, causing a unique pattern in the perceived light readings. The right image shows a signal representation of the event. 29 2-3 Backscatter transmitter frontend. By switching the impedance
states between Z
1and Z
2, corresponding to bit ’0’ and bit ’1’, re- spectively, the backscatter transmitter modulates the RF signal. In this illustration, the incident RF signal is modulated in amplitude and in frequency. . . 31 2-4 Frequency-shifted backscatter. The captured image shows the
frequency spectrum with a carrier signal centered at 2.48 GHz and the backscattered signal with its two mirrored frequencies located 500 kHz away from the carrier, i.e. at 2.4795 GHz and 2.4805 GHz. . 32
9
2-5 Monostatic vs. bistatic setup. As a backscatter tag moves away from the reader in a monostatic setup, the received signal power decreases in accordance with the inverse square law. In a bistatic setup, the received signal power increases as the tag moves closer to the receiver or the carrier source. . . 33 2-6 Ambient backscatter and frequency shifting. An RF source
such as a radio tower generates a carrier signal at frequency f
c, which in turn is reflected by a backscatter tag at the frequency f
c+ ∆f and received by a commodity receiver located at some distance away from the tag. . . 34 2-7 Enabling ultra-low power wireless sensing by avoiding energy-
expensive computational blocks. The block diagram shows the key design. The idea is that the light sensor is directly coupled to the communication mechanism. . . 35 3-1 A typical visible light sensing sensing system. These systems
use photodiodes coupled with transimpedance amplifiers which are susceptible to saturation under bright light. Furthermore, they dig- itize using energy-expensive microcontrollers or field-programmable gate arrays. . . 39 3-2 The Battery-free Cellphone. Their prototype design avoids
computational blocks by a mechanism to directly couple the sen-
sor to the RF backscatter communication module. We build on a
similar principle in our design. . . 44
4-1 Visible light sensing and communication at µWs of power.
In this scenario, a commodity radio generates a carrier signal, and the VLM senses a shadow under sunlight. In another scenario, the VLM can backscatter RF signals from radio towers and sense shadows under artificial light. . . 46 4-2 Battery-free visible light marker prototype platform. The
enumerated markers in the figure illustrate the main components of the system: (1) The energy-harvesting architecture consisting of a large thinfilm solar cell, and an energy harvesting management module with an external storage capacitor. (2) The light sensor consisting of a small thinfilm solar cell. (3) Our custom circuit design (aVLM/tVLM) consisting of components that convert sensor readings to frequency shifted backscatter signals. (4) The antenna used for communication within the 2.4 GHz band. . . 47 4-3 The thinfilm solar cell used for sensing. This solar cell has a
thickness of 0.2 mm and a size of 12.7 x 64 mm, which makes it an attractive device for wearable sensing. It has an open circuit voltage of 4.5 V and can also be used to sense variations in the visible light.
For our sensing purposes, we choose to use this particular solar cell. 49 4-4 The BQ25570 harvester. This particular board has the BQ25570
chip (center) along with an on-board 100 µF capacitor and place- holders for header pins. . . 50 4-5 The thinfilm solar cell we choose for energy harvesting. This
particular cell has a larger surface area (74 x 114 mm), as compared to the previous solar cell we chose for sensing. Furthermore, it has more cell connections in parallel, which improves energy yield and therefore also makes it better suited for energy harvesting purposes. 50
11
4-6 USRP-B200 Software Defined Radio. We use this radio to generate constant tones for backscatter experiments, and to perform complex signal processing tasks at reception. . . 52 4-7 The CC1310 Launchpad. We use the low-cost CC1310 wireless
MCU to sample the RSSI at sub-1 GHz. . . 52 4-8 RTL SDR USB Dongle and an FM antenna. The RTL-SDR is
an inexpensive receiver (25 USD) and can be used to receive signals at frequencies between 52 - 2200 MHz. We will use this to receive audio signals from a nearby radio station. . . 53 4-9 Antennas used. The top antenna is a telescopic FM antenna with
a soldered SMA connector. We use the middle and the bottom antennas are for experiments at 2.4 GHz, and 868 MHz, respectively. 53 4-10 Waterfall illustration of an FM signal centered at 106.5
MHz. The ambient FM signal is received at a campus area in Uppsala. The red, orange and yellow colors indicate a strong signal, and the blue colors indicate weaker signal strength. Blue vertical lines near the signal typically indicate weak harmonics or concurrent transmissions. . . . 53 4-11 Schematic diagram of the principal backscatter operation.
Here, the value of RSET determines the output frequency of the oscillator, i.e. ∆f . When connecting a voltage source over RSET, the oscillator acts as a VCO. . . 54 4-12 Schematic of the thresholding visible light marker. The
tVLM can detect and communicate shadow events without the use of any power-hungry components. . . 57 4-13 Hardware design of the thresholding visible light marker (tVLM).
The dimension of the tVLM is 25x25 mm. . . 57
4-14 Realization of analog backscatter using a VCO. The illustra- tion shows the main design blocks of the analog backscatter tag.
Here, V (t) denotes voltage as a function of time and the input to the VCO. The output frequency of the VCO, ∆f (V ), is a function of the input analog signal and used to control the frequency of the switching circuit. This gives the frequency shift of the backscatter signal. As a result, the backscattered signal appears at a frequency f
c+ ∆f (V ), where f
crepresents the frequency of the incident RF signal, i.e. the carrier signal. . . 60 4-15 Schematic of the analog visible light marker. The aVLM
transforms analog signals to corresponding changes in the frequency of the backscatter signal. . . 61 4-16 Hardware prototype of the analog visible light marker. This
marker has the ability to map analog sensor readings to frequencies of the backscatter signal. . . 61 4-17 Signal processing flowgraph implemented using GNU Ra-
dio Companion graphical development tool. The signal is streamed from the receiver using the USRP Source block and the signal is stored using the File Sink block after several signal pro- cessing operations. In this flowgraph, the band-pass filtering block is bypassed, as highlighted by the yellow color. . . 66
13
5-1 Light sensing responsiveness at 500 lx of various solar cells.
For the VLM, we choose the thinfilm solar cell, as highlighted by the green color. This is because it has a relatively high SNR and a flexible design which allows for more versatile deployment usage, such as on wearable sensing devices. Furthermore, we observed that the thinfilm solar cell has the highest short circuit current, hence making it the best choice for energy-harvesting. . . 71 5-2 Comparison of unamplified photodiodes to the thinfilm so-
lar cell. The result shows that the thinfilm solar cell is clearly unmatched in sensing responsiveness compared to the photodiodes.
We note that that the responsiveness of photodiodes could be en- hanced by amplification using a TIA. However, that would increase the power consumption significantly which makes it less feasible for battery-free operation. . . 71 5-3 Lux-to-voltage ratio of a thin film solar cell. The output
voltage of the thin film solar cell varies in close resemblance to a logarithmic curve. . . 72 5-4 The cold start mode of the BQ25570 harvester. We per-
formed an experiment to evaluate the voltage boosting functionality
of the harvester when connecting a harvesting solar cell to it. Then
we perform a shadow over the solar cell and measure how long it
takes for the harvester to exit this mode. When the input is above
1.5 V, the harvester starts to boost the voltage up to 2 V. . . 75
5-5 Setup for the 2.48 GHz analog backscatter experiment. . . 77
5-6 Identified backscattered signal frequencies when the input voltage to the aVLM is 30 mV. The processing algorithm iden- tifies the backscattered signal frequency shifts by finding the corre- sponding signal amplitudes. The algorithm is able to identify these peaks because the backscattered signals are stronger than the back- ground transmissions. . . 78 5-7 Identified backscattered signal frequencies when the input
voltage to the aVLM is 300 mV. . . 79 5-8 Identified backscattered signal frequencies when the input
voltage to the aVLM is 510 mV. . . 79 5-9 Frequency output of the VCO of the aVLM and the re-
ceived backscattered signal at varying input voltages. The experiment shows that there is an almost linear relationship be- tween the input voltage and the received backscattered frequency.
However, we observe that a nonlinear behaviour occurs at higher input voltages. We also observe that there is a slight difference in frequency between the received signals and the VCO output; the maximum difference is around 14 kHz. . . . 80 5-10 Relative error between the actual input voltages to the
VCO and the approximated voltages at a separation dis- tance of 1 m between the tag and the carrier source. We observe that at varying separation distances between the tag and the receiver, the relative errors between the input voltages and the approximated ones stay almost constant (as expected) and below 5 %. Thus, the accuracy of the approximating function is at mini- mum close to 95 %. . . 81
15
5-11 Relative error between the actual input voltages to the VCO and the approximated voltages at a separation dis- tance of 2 m between the tag and the carrier source. As expected, the relative error still stays below 5 % and barely changes when moving the tag away from the carrier generator. The slight offsets between the relative errors here and in Figure 5-10 is likely an effect of the unpredictable nature of RF, which gives rise to slight variations in the received backscattered signals. . . 82 5-12 Ambient FM Backscatter at 1 m separation between the
tag and the receiver. The figure shows the signal power as a function of the frequency offset of the backscattered signal, i.e. ∆f . This particular backscattered signal is received at a frequency offset of 200 kHz. . . 82 5-13 Ambient FM backscatter at 2 m separation between the
tag and the receiver. . . . 83 5-14 Peak detection of FM signals backscattered at an offset
of 100 kHz at 1 m separation between the tag and the re- ceiver. The red dashed line is drawn at the center frequency, which is 106.5 MHz. The moving average filter extracts a very smooth version of the FM signal but distorts the relative amplitude values.
Therefore, we normalize the filtered signal amplitudes as a method to demonstrate an indicator of the relative signal strengths. . . 83 5-15 Examples of hand gestures that can be performed using
the tVLM. Swipe is represented by a brief hand movement over
the tVLM. Two, and Four taps are represented by two or four slower
palm movements. . . . 84
5-16 Sensing hand gestures. We detect three hand gestures (Swipe, Two taps, Four taps) at 20 µWs of power. The top two rows show the output at the VLM, the bottom row shows the received signal at the end-device. . . 85 5-17 Sensing hand gestures using aVLM . Analog backscatter en-
ables us to detect four hand gestures: Push, Pull, Block, and Punch
that requires higher sensing resolution. The figures illustrate a 6-bit
representation these gestures which are obtained and reconstructed
at the receiver. . . 86
A-1 Output power of the USRP-B200 SDR at 2.48 GHz. . . 106
Acronyms
ADC analog-to-digital converter.
aVLM analog visible light marker.
BLE Bluetooth Low Energy.
COTS commercial off-the-shelf.
DC direct current.
FFT Fast Fourier Transform.
FPGA field-programmable gate array.
IoT Internet of Things.
ISM industrial, scientific and medical.
LED light-emitting diode.
LPWAN low-power wide-area.
MCU micro-controller.
NES networked embedded systems.
19
RCS radar cross section.
RF radio frequency.
RFID radio-frequency identification.
RSSI received signal strength indicator.
SDR software-defined radio.
SNR signal-to-noise ratio.
SPDT single pole, double throw.
TIA transimpedance amplifier.
tVLM thresholding visible light marker.
VCO voltage controlled oscillator.
VLC visible light communication.
VLM Visible Light Marker.
VLS visible light sensing.
Chapter 1 Introduction
The vision of enabling ubiquitous wireless connectivity and interoperability be- tween all types of computational and sensing systems has within the past few decades been embraced by many institutions and enterprises worldwide. As a re- sult, a vast quantity of networked embedded systems (NES) have emerged which either simplify or augment our everyday lives. Many of the state-of-the-art sys- tems can seamlessly interact with smartphones or computers to, for instance, con- trol common physical objects through hand gestures in smart homes [1, 2] or to perform biomedical sensing [3, 4]. These could be anything from smart wear- able systems that can record and communicate human vital signs such as heart rate and blood pressure [5], smart thermostats that communicate temperature variations in heating systems [6], UV sensing systems that can measure and com- municate the intensity of incident UV-light to smartphone apps [7], and indoor localization systems that leverage smart beaconing techniques to reveal position information [8, 9, 10, 11, 12]. When NES includes sensors that are coupled with processing devices in the form of computers or smartphones, they allow for infer- ence of additional information caused by abnormal variations in sensor readings;
for example, the effects of an increased exposure to UV light due to high inten-
sity levels, or the potential sources of malign changes to human vitals caused by significant changes in the vital signs over a period of time.
However, the quantity of NES has experienced a rapid increase over recent years. The number of connected embedded systems by the year 2015 was 15 bil- lion, and it is expected that this number will almost double by 2022, reaching 29 billion [13]. The rapid increase of such systems gives rise to new and alarm- ing problems that has to be addressed to allow for future interoperability and for environmental sustainability, particularly when the aim is to deploy them at a pervasive scale. There are several problems that needs to be addressed; first, many NES share the same wireless spectrum and a higher density of devices tends to lead to more systems interfering with each other [14, 15]. Second, most sys- tems (including the state-of-the art) are either powered by batteries or by mains, which severely restricts the possibilities for pervasive deployment. Batteries in- duce maintenance costs and are not always easy to replace [16, 17]. Furthermore, drawing power from mains to power these systems is severely limited by the place- ment of the sources from which the power can be drawn. Thus, these systems are most often infeasible to be deployed in hard to reach places such as within the concrete of roads or on top of buildings. Moreover, the rapid increase in the quan- tity of embedded systems leads to sustainability problems in the environmental aspect, primarily since modern batteries require scarce natural resources and need to be manufactured at a larger scale to accommodate for the increasing number of devices used in NES.
1.1 Motivation
In order to accommodate for the increase in NES to allow for environmental sus-
tainability, it would require new forms of sensing systems and mechanisms that are
energy-efficient in practice and can last for a long time without maintenance. The two main bottlenecks of most existing state-of-the-art systems are the energy- expensive operations of sensor-local processing units such as micro-controllers (MCUs) or field-programmable gate arrays (FPGAs) [1, 18, 19], and the power consumption of the active components of the transmitting radios [20]. These units force the systems to operate by mains or by on-board batteries - which has to be frequently replaced - and are therefore not sustainable. If these power-hungry units could be moved away from all the sensors within a network and converge to a base station or a common computing device without compromising on the functionality at sensor-level, then these devices would potentially be able to operate at minus- cule amounts of energy which could easily be harvested from ambient sources [21].
The harvested energy could be used to charge on-board capacitors that power the rest of the circuitry long enough to achieve the desired functionality. Subsequently, if the amount of charge and recharge cycles of the on-board capacitors are kept to the absolute minimum, these systems could potentially operate well beyond our lifetime.
1.1.1 Ultra-low Power Sensing Using Visible Light
Visible light is an ubiquitous medium that can be sensed using simple and inex- pensive photodiodes or solar cells. Small or thin-film solar cells can be used to both sense changes in light intensity and to harvest energy to power the sensing device [2]. Solar cells are designed for maximum power yield, and unlike photo- diodes do not require energy-expensive amplification components such as a tran- simpedance amplifier (TIA) to output a sufficiently strong signal [22]. Thus, solar cells can potentially bring down the power consumption to near zero for sensing.
23
1.1.2 Ultra-low Power Communication
Most existing NES communicate within the industrial, scientific and medical (ISM) band, which in Europe occupies frequencies between 868 MHz to 5.8 GHz, and in- cludes Bluetooth, ZigBee, and WiFi. Many of these devices use on-board trans- mitters to generate RF signals to establish communication links to communicate sensor data or node identifiers. However, using on-board transmitters to generate signals require amplifiers which are energy-expensive [23]. There are a number of passive radio-frequency identification (RFID) systems that utilize passive tags, which are essentially node identifiers that can communicate their unique identi- fiers without generating their own signal. Instead, these devices can absorb energy from the RFID reader (transmitter) and reflect a modulated signal with a unique identifier back to the reader (also acting as a receiver). The phenomenon of re- flecting ambient wireless signals is called backscattering. Recent work has explored encoding information on backscattered signals from ambient RF sources such as Bluetooth [24], ZigBee [25, 26], WiFi [27], TV [23], and FM [28]. This would al- low future NES to be able to use incident RF signals from any transmitting device nearby to be modulated with their own signal. Further work have improved on this by backscattering at a frequency shift away from the source carrier signal to avoid self-interference from the carrier and improve the signal-to-noise ratio (SNR) at re- ception [29, 30]. Backscatter transceivers have been shown to operate at a few tens of microwatts of power for communication, which is three orders of magnitude less power consumption compared to conventional RF transceivers [21, 23, 27, 30, 31].
1.2 Problem Statement
In this thesis, we answer the following question: Is it possible to design a battery-
free wireless sensing system that can be deployed both sustainably and pervasively?
This thesis aims to provide a solution to enable pervasive deployment of NES by designing a battery-free system that can both sense and communicate sensor readings wirelessly without the use of any energy-expensive computational blocks.
To achieve a very low power consumption which is necessary for battery-free oper- ation, the system should be able to delegate all of the necessary processing tasks of sensor readings to a powerful end device. Furthermore, all the energy required to power the system should be harvested from the surrounding environment, which would make the system self-powered. To address these challenges, this thesis work provides three main contributions: utilizing ambient light sources for sensing and energy harvesting, using RF backscatter to achieve ultra-low power communica- tion, and avoiding computational blocks at sensor level by modulating backscatter signals with analog information. In addition, a gesture recognition system will be designed to demonstrate a use case of the battery-free wireless sensing system.
25
Chapter 2 Background
This section introduces the relevant theory for this thesis. Overall, it covers con- cepts behind visible light sensing and RF backscatter.
2.1 Visible Light Sensing
The definition of light is usually interchangeable with visible light and refers to the set of electromagnetic waves that can be perceived by the human eye. These waves make up the electromagnetic spectrum, which consists of wavelengths be- tween around 390 to 700 nm or frequencies between 430 to 770 THz. With the exception of sunlight, visible light can be provided by artificial light sources such as incandescent bulbs, fluorescent bulbs, halogen bulbs or light-emitting diodes (LEDs). Light can be sensed using photodiodes, photoresistors or photovoltaic cells (solar cells), which convert light energy to electrical energy through the pho- toelectric effect.
Visible light sensing (VLS) is a mechanism that consists of sensing changes in
light illumination to detect activity. The light source could for instance be ambient
light in indoor or outdoor environments, or modulated visible light beacons. The
principle behind VLS is simple; light sensors can be used to record variations in light levels, which when recorded over a period of time can be used to infer information about the surrounding environment. For example, a moving object between a light source and a light sensor will cast a shadow over the sensor, which corresponds to a significant drop in light intensity. If the light readings are recorded and processed, the collected data could infer whether the shadow of the moving object has obstructed the sensor. While the object stays near the sensor, the recorded sensor data could give information about how the object behaved during that period of time. This principle in effect allows for tracking of shadows cast by objects or people through their unique patterns in the perceived light readings.
Figure 2-2 demonstrates how VLS can be accomplished.
Ultraviolet Infrared
10 400 700 Wavelength (nm)
Visible light
Figure 2-1: The visible spectrum. The figure shows the region of the electro- magnetic spectrum at which radiation, with range of wavelengths between approx- imately 400 to 700 nm, is visible to the human eye.
2.2 Radio Frequency Backscatter
2.2.1 Theory
RF backscatter is widely used to provide ultra-low power links between NES used
in Internet of Things (IoT) networks. Backscatter devices can transmit data by re-
flecting incident RF signals and therefore they only consume microwatts of power
Voltage
Time Light Source
Signal Representation
Light Sensor
Figure 2-2: Demonstration of the basic principle of visible light sensing.
In this scenario, a moving hand obstructs a light sensor by casting a shadow over it, causing a unique pattern in the perceived light readings. The right image shows a signal representation of the event.
for operation. They are commonly used in RFID systems in which a reader inter- acts with a backscatter tag by transmitting a carrier signal. The carrier signal is first absorbed by the tag to provide energy for waking the tag up, then the tag reflects a modulated version of the signal back to the reader containing its unique identifier. The backscatter tag determines whether to absorb or reflect the carrier signal by switching the impedance of its antenna between two states. By changing the impedance of the antenna, the backscatter device in turn controls the radar cross section (RCS), which governs the level of reflection or absorption [30].
The working principle of backscattering as perceived by a receiving device can be modeled as follows: let S
c(t) be the carrier signal generated by an RF transmitter that reaches the antenna of the backscatter device. The backscatter device reflects a signal R(t), which consists of S
b(t) and a signal S
b(t) caused by varying the impedance of the backscatter antenna. The resulting backscattered signal can then be observed by an RF receiver. Thus, the received signal R(t) can
29
be expressed in the following way:
R(t) = S
c(t) + σB(t)S
b(t) , (2.1)
where σ is the RCS of the antenna, and B(t) is a binary coefficient that is 1 when the backscatter antenna is at a reflecting state, and 0 when absorption occurs.
Specifically, by changing the impedance of the antenna, the backscatter trans- mitter maps a bit sequence to RF waveforms. The reflection coefficient of the antenna can be computed as follows:
Γ
i= Z
i− Z
a∗Z
i+ Z
a, (2.2)
where Z
ais the impedance of the backscattering antenna, ’*’ the complex conju- gate, and i = 1, 2 represents the switching states, i.e. absorption or reflection [32].
At a high level, this means that by switching between Z
1and Z
2, the reflection coefficient can be shifted between absorbing or reflecting states, as shown in Fig- ure 2-3. The absorbing state represents bit ’0’, and the reflecting state represents bit ’1’.
The power of the backscatter signal can be modeled in the following way, given an incident carrier signal at the backscatter antenna with power P
c:
P
b= P
c|∆Γ|
24 , (2.3)
where P
bis the power of the backscattered signal, and |∆Γ|
2= |Γ
∗1− Γ
∗2|
2consists
of the complex conjugates of the reflection (Γ
1) and absorption (Γ
2) coefficients,
respectively [27]. The maximum power of the backscattered signal is achieved
when P
b= 1, i.e. when |∆Γ|
2= 4. In this case, the power of the backscatter signal
is equivalent to the incident carrier signal. This is achieved when ideal absorption
or reflection occurs, i.e. when Z
1and Z
2vary between +1 and −1. However, in
Z1
Z2
Incident RF Signal
Backscaered Signal Za
Impedance Switching
Figure 2-3: Backscatter transmitter frontend. By switching the impedance states between Z
1and Z
2, corresponding to bit ’0’ and bit ’1’, respectively, the backscatter transmitter modulates the RF signal. In this illustration, the incident RF signal is modulated in amplitude and in frequency.
practice, these coefficients deviate to a large extent from this behaviour, which in turn results in backscatter signals that are weaker than the carrier signals.
2.2.2 Self-interference
Self-interference is the phenomenon that occurs when a radio transceiver trans- mits and receives a signal at the same frequency. In traditional RF backscatter systems, this occurs when the carrier signal interferes with the backscattered sig- nal in a communication link. As a result, it negatively impacts the quality of reception at the RFID reader. Existing RFID readers attempt to get around this problem by employing self-interference cancellation or various techniques to isolate the backscatter signal from the carrier [33, 34, 35, 36, 37]. However, these tech- niques are complex and energy-expensive, and severily restricts the communication link to a few meters.
One alternative approach to mitigate this effect is to backscatter a signal at a frequency shift away from the carrier signal [30, 27, 38]. The resulting backscat-
31
tered signal can be modeled in the following way: first, the reader is assumed to generate a tone or a pure sinusoidal at frequency f
c, i.e. S
r= sin(2πf
ct ). Next, the backscatter tag varies the RCS at a frequency ∆f . Thus, the term B(t) in equa- tion 2.1 is a square wave with frequency ∆f , and the expression for the resulting backscatter signal σB(t)S
bcan be simplified to the following form:
2 sin(f
ct) sin(∆f t) = cos[(f
c+ ∆f )t] − cos[(f
c− ∆f )t] , (2.4)
which gives a backscatter signal that appear at a frequency shift ∆f on the posi- tive and negative sides of f
c. By appearing at different frequencies, the resulting backscatter signal avoids interference from the carrier to the backscatter transmis- sion. The choice of the minimum ∆f to avoid self-interference is dependent on the type of transceiver used and the bandwidth of carrier signal [30]. Typically, a good heuristic is to choose a ∆f which is outside of the occupied bandwidth of the carrier signal to avoid attenuation of the backscattered signal.
2480.5 2480.0
2479.5
Carrier signal Backscaered signals
[MHz]
Figure 2-4: Frequency-shifted backscatter. The captured image shows the fre- quency spectrum with a carrier signal centered at 2.48 GHz and the backscattered signal with its two mirrored frequencies located 500 kHz away from the carrier, i.e.
at 2.4795 GHz and 2.4805 GHz.
Tag location
Received signal power
Reader location
(a) Monostatic setup.
Tag location
Received signal power
Receiver
location Carrier generator
location
(b) Bistatic setup.
Figure 2-5: Monostatic vs. bistatic setup. As a backscatter tag moves away from the reader in a monostatic setup, the received signal power decreases in accordance with the inverse square law. In a bistatic setup, the received signal power increases as the tag moves closer to the receiver or the carrier source.
2.2.3 Bistatic Configuration
Conventional RFID systems use monostatic setups in which a RFID reader act as both a transmitter and a receiver. However, this increases the cost and com- plexity of deployment since RFID readers are complex and expensive devices and require dedicated infrastructure for deployment. A bistatic setup is a setup in which the transmitter is separated from the receiver. Recent research has shown that a bistatic setup enables backscatter communication of carrier signals that are generated by commodity devices operating in the 2.4 GHz band such as Bluetooth and WiFi, and received by low-cost receivers [39, 30, 25]. Furthermore, a bistatic setup can also increase the operating range of the backscatter communication [30].
33
Radio tower
Ambient signal
Backscaer Tag Receiver
Backscaered signal
Receiver
Figure 2-6: Ambient backscatter and frequency shifting. An RF source such as a radio tower generates a carrier signal at frequency f
c, which in turn is reflected by a backscatter tag at the frequency f
c+∆f and received by a commodity receiver located at some distance away from the tag.
2.2.4 Ambient Backscatter
The modern society is surrounded by ambient RF signals. These signals could
come from FM or TV broadcasting towers, WiFI routers, or commodity devices
that use ZigBee or Bluetooth communication such as smartphones or fitness track-
ers. These sources are widespread and therefore provide ubiquitous carrier signals
which could be utilized for backscatter communication. Recent work has shown
that reliable ambient backscatter communication can be achieved by reflecting RF
transmissions from TV towers [32], FM stations [28] or commodity WiFi, Blue-
tooth or Zigbee transmitters [25]. Zhang et al. achieves a communication range of
up to 54 m when backscattering WiFi signals [38], and Varshney et al. achieves a
communication range of 3.4 km when backscattering LoRa signals [30]. For exam-
ple, outdoors, within, or near cities, FM and TV signals could be used as carrier
signals. Within indoor office environments, WiFi, Bluetooth or ZigBee sources
could provide the carrier signals. Away from cities, low-power wide-area (LP-
WAN) communication technologies such as LoRa [40] could be utilized as sources
for carrier signals. Thus, ambient backscatter fits well with the vision of pervasive
deployment of NES.
Computational Blocks MCU/FPGA
Light Sensor Communication
(RF Backscatter)
Figure 2-7: Enabling ultra-low power wireless sensing by avoiding energy- expensive computational blocks. The block diagram shows the key design.
The idea is that the light sensor is directly coupled to the communication mecha- nism.
However, there are two main limitations of ambient backscatter systems. First, deployment of backscatter systems is restricted by the location of the transmitting source. Second, certain carrier sources such as WiFi occupy a large bandwidth (22 MHz), and therefore take up a large space of the ISM band. This adds to the challenge of pervasive deployment since a backscatter signal will mirror the WiFi signals and therefore also occupy a large bandwidth. This is infeasible if the intent is to deploy these devices at scale, particularly when frequency shifting is used as a technique.
2.3 Mitigating Computational Overhead
In the design of commonly available NES, on-board processing units such as MCUs or FPGAs are used to perform processing locally at the sensor, and active radios are used to generate wireless transmissions for communication. However, the use of these components for such tasks are adversary to battery-free operation, as these mechanisms are energy-expensive and push the operational power of NES beyond milliwatts. To achieve microwatts of power for consumption at sensor- level, this overhead of computation and communication has to be mitigated or avoided entirely.
First, overcoming the overhead from communication is addressed in the pre- vious section, i.e. using RF backscatter for transmitting wireless signals instead
35
of local active radio to achieve microwatts of power for communication. Here, we
embrace Zhang et al.’s observation that processing is significantly more energy ex-
pensive as compared to backscatter transmissions [41]. Second, to eliminate local
processing, novel mechanisms have to be devised to communicate sensor readings
either directly in raw form, or if digitized at minimal power consumption. Thus,
the sensor would be directly coupled to the communication mechanism, and all the
computational blocks in between would be minimized or eliminated from existing
as we know them in devices. Figure 2-7 demonstrates the high-level design princi-
ple. We explain more in following sections how this can be accomplished through
novel hardware designs.
Chapter 3
Related Work
In this chapter, we discuss the related work in the areas of visible light sensing, gesture recognition, and ultra-low power designs that make use of RF backscatter to communicate analog sensor data.
3.1 Visible Light Sensing
Prior work has explored sensing shadows to track human body movements or to detect hand gestures. Li et al. reconstruct human skeleton postures by sensing shadows using an array of photodiodes under modulated light beacons [42, 43].
They leverage VLC to modulate light beacons from multiple LEDs - each flash-
ing at a unique high frequency imperceptible to the human eye - in order to use
the photodiodes to sense which light beacons are blocked by the shadow. By ag-
gregating the sensed blockage information from multiple photodiodes, they can
infer relative positions of the human shadow to reconstruct the skeleton. In an-
other system, Li et al. reconstruct hand poses by using a similar principle; they
design a customized lighting setup consisting of an array of photodiodes under
modulated light to reconstruct hand skeletons when a hand moves over the pho-
todiodes [18]. Kaholokula et al. detect hand gestures under unmodulated light by tracking changes in light intensity caused by shadows from hand movements over multiple photodiodes [1].
However, all of these VLS systems suffer from a few critical limitations which hinder them from being deployed pervasively. There are several key reasons for this: First, they make use of TIAs to amplify signals from photodiodes. TIAs are susceptible to saturation under bright light conditions, which severely restricts the operating range of detectable illumination levels. The use of a TIA is inim- ical to pervasive deployment, particularly in outdoor environments when being exposed to sunlight which can vary between a few lx to tens of thousands of lx.
Second, they use power hungry commercial off-the-shelf (COTS) platforms such as Arduinos which drive analog-to-digital converters (ADCs) to sample and dig- itize light sensor readings. When combined, these mechanisms push the energy consumption to mW of power at the sensing devices, and hence forces them to be powered by mains or by batteries. Furthermore, the usage of VLC to establish downlink communication between light sources and light sensors requires modifi- cations to the existing light infrastructure, which increases the cost and complexity of deployment. Specifically, this requires that the existing infrastructure of light bulbs is retrofitted with specialized driving circuits to modulate visible light, i.e.
to switch the on and off periods of light bulbs to transmit bits at an unobservable rate. Thus, VLC is an adversary to natural light for sensing and therefore severely restricts pervasive deployment of VLS based systems. In addition, VLC systems require complex circuitry and mechanisms for demodulation [44], for instance, to perform Fast Fourier Transforms (FFTs) of the sensor readings.
Thus, in our design of the light sensor, we enable sensing under unmodulated
light, and we leverage passive sensing using solar cells. Solar cells are optimized
for photovoltaic efficiency and can therefore operate without the usage of any TIA.
MCU/FPGA
Figure 3-1: A typical visible light sensing sensing system. These systems use photodiodes coupled with transimpedance amplifiers which are susceptible to saturation under bright light. Furthermore, they digitize using energy-expensive microcontrollers or field-programmable gate arrays.
Thus, solar cells are not susceptible to saturation and can function in diverse light conditions. Furthermore, as opposed to using energy-expensive MCUs with ADCs for digitization, we make use of analog backscatter to eliminate the power hungry computational blocks from the sensing device. Hence, by delegating the sensor readings to a powerful end device that can perform the required processing oper- ations, we can reduce the power consumption of our sensing device significantly.
In a more recent work by Li. et al, they develop a battery-free wearable eye tracker on glasses, which can detect the position of the eye based on the absorption and reflection properties of the eye when subject to near-infrared illumination [45].
Similar to our system, theirs use a thinfilm solar cell to harvest energy from light to power itself. They use a few near-infrared LEDs around the eye, and an array of photodiodes in a particular arrangement to sense variations in reflected light intensity. This allows the device to infer the position of the eye in a 2D space, and the diameter of the pupil. The sensor data is then fed to a microcontroller where it employs processing algorithms for identification. However, their system is limited to indoor environments where ambient near-infrared light is weak, as otherwise in outdoor environments where the light conditions are very diverse, near infrared light can be very strong and may therefore saturate the photodiodes and interfere with light sensing properties of the device.
39
3.2 Ultra-low Power Gesture Recognition
Kellogg, Talla, et. al design an ultra-low power gesture-recognition system that leverages changes in ambient RF signals to detect the presence of hand ges- tures [46]. They achieve this by making use of the principle that movements close to a receiving antenna will induce significant amplitude changes in the re- ceived RF signals. Furthermore, they use simple analog components to extract the envelope from the received RF signals and digitize them using comparator circuits, thus bringing down the power consumption to tens of microwatts at the sensing device which allows their system to be operated on harvested RF energy from the environment.
We build on a similar principle to detect hand gestures based on changes in ambient light levels. Specifically, we make use of the fact that significant changes in light intensity near a light sensor such as a solar cell causes corresponding changes in the voltage output of the sensor. We can utilize this idea to, for instance, detect hand gestures from shadows cast over a solar cell. Similar to the design by Kellogg, Talla, et. al, we can make use of ultra-low power components at the sensing device to achieve battery-free operation. Furthermore, in the next section, we describe how we can use a comparator circuit to perform digitization at ultra-low power.
Moreover, we show that we can provide sufficient power to our sensing device by harvesting light energy from artificial light sources found in common indoor environments.
Concurrent to our work, Li et. al design a self-powered module that utilizes
changes in ambient light levels to sense finger gestures [2]. The design includes an
array of over 40 photodiodes that sense changes in incident light levels, and harvest
energy to power an on-board TI MSP430 microcontroller [47] which processes
the output sensor data. Furthermore, they employ classification mechanisms to
identify various finger gestures based on how the aggregated analog output of the photodiodes change over time, which can provide information regarding how intensity levels change and the direction at which the finger is moving.
3.3 Radio-Frequency Backscatter
RF backscatter has within the past decade emerged as a form of communication method with high potential, as it can bring the power consumption of existing radio modules to a few microwatts of power. The main reason for this is the fact that devices with backscatter capabilities do not need to generate their own signals for transmissions. Instead, they can leverage incident wireless signals as carrier signals. As a result, these devices could potentially operate on harvested energy from surrounding environmental sources, and free of batteries. Thus, RF backscatter is an exciting prospect for applications where requirements of replacing batteries are cumbersome, for instance, devices that are intended to be deployed in hard to reach places such as concrete or within walls. Furthermore, it enables the design of small, inexpensive, and self-powered devices with the ability to both sense their surroundings and communicate this information to a base station for further processing.
Existing computational RFID (CRFID) tags such as the WISP [48] and MOO [49]
utilize passive backscatter tags to communicate sensor data. These devices oper- ate on harvested energy and have been deployed for localization [12] and wireless microphones [50]. Naderiparizi et al. develop a passive, and Ultra-High Frequency (UHF) RFID camera tag based on the WISP platform, which is battery-free [16].
However, they are severely restricted by the short communication range of RFID readers, and the high cost (≥ 2000 USD) and power requirements of these readers.
In these systems, the reader act as both the carrier generator and as a receiver.
41
Varshney et. al overcome these limitations by decoupling the reader from the carrier generator, which allows the usage of much cheaper commodity devices such as WiFi routers and sensor nodes to act as a source of carrier signals, instead of conventional RFID readers [30]. Furthermore, they make use of narrow-band backscatter transmissions and employ a mechanism which keeps the frequency of the backscatter signal away from the carrier signal. This improves the receiver sensitivity and mitigates self-interference from the carrier, hence allowing this sys- tem to achieve a much longer range. This thesis work is fundamentally based on the previous research by Varshney et. al with a decoupled carrier generator and receiver, and a frequency shifting of backscatter signals. In the next section, we dive into the specifics of the communication mechanisms and hardware that we use to enable wireless communication at ultra-low power.
3.4 Ambient Backscatter
In more recent years, as the level of connectivity increases and the wireless spec- trum gets increasingly crowded, carrier signals also become ubiquitous, which in- creases the potential of backscatter. This is particularly true in urban environ- ments. Liu et. al leverage RF backscatter of ambient signals from TV and cellular transmissions to enable ultra-low power at less than 1 µW between two devices [23].
Furthermore, they show that they can harvest energy from the incident RF signals to power the communicating devices without the use of any on-board batteries.
Wang et. al show that ambient FM signals can be used for backscatter communi-
cation [28]. They achieve this by transforming ambient RF backscatter signals to
audio signals, which can be received and decoded by inexpensive FM receivers used
in cars or in smartphones. This idea particularly interesting, since a backscattered
signal will replicate the shape of the carrier signal, i.e. the FM station, which
allows a simple radio transceiver to receive and decode the audio signal. If ambi- ent backscattering is performed by frequency shifting the incident FM signals, the receiver can simply tune in to the shifted frequency to decode the audio signal.
3.5 Eliminating Computational Blocks
As RF backscatter has emerged as a method of communication within recent years, the amount of time spent on communicating data is not longer the power consumption bottleneck of a sensor system. Zhang et al. argue that the use of RF backscatter at the sensing device implies that most power is consumed by the computational blocks [41]. Thus, instead of minimizing the communication, they optimize the computational block to bring down the power consumption of the entire sensing device to microwatts. We embrace this key principle to minimize the power consumption in our design of our sensor hardware by avoiding any computational blocks.
Talla et. al [31] demonstrate a prototype of a battery-free cellphone that avoids any computational blocks and consumes a few microwatts of power for operation, as shown in Figure 3-2. In their design, they feed analog sensor readings from a microphone directly to a communication module which transmits the analog infor- mation using RF backscatter. They achieve this by varying the impedance of the antenna according to the sensor readings, thus performing amplitude modulation.
However, their system performs backscatter at the same frequency as the carrier signal and hence encounter significant self-interference.
43
Figure 3-2: The Battery-free Cellphone. Their prototype design avoids compu- tational blocks by a mechanism to directly couple the sensor to the RF backscatter communication module. We build on a similar principle in our design.
3.6 Analog Backscatter
Concurrent to our work, Naderiparizi et. al use analog backscatter to enable battery-free HD video streaming [21]. Similar to our design, they also avoid any computational blocks at the sensing device to achieve ultra-low power consump- tion. They perform digitization at sensor level by using a comparator based mech- anism to compare the input analog signal from a camera’s sensor to generate pulse width modulated digital signals. The modulated digital signals in turn control the duty cycle of backscattering, which is proportional to the input analog signal.
Furthermore, they perform up-converting of the digital signal to backscatter at a
frequency ∆f to avoid self-interference from the carrier signal.
Chapter 4
System Design
In this chapter, we present the design and implementation of our battery-free visible light sensing system. It can operate on harvested energy from light, and communicate sensor readings using RF backscatter. Furthermore, we present the design of two forms of battery-free devices that can sense visible light and commu- nicate by backscattering incident RF signals at 868 MHz, 2.4 GHz, and ambient FM transmissions (88 to 108 MHz). Finally, we explain how we can communicate analog information from light sensor readings using backscatter, and how we can receive these signals to reconstruct the analog information at the receiver.
4.1 Overview
We envision an ultra-low power VLS system which we call Visible Light Marker
(VLM). The VLM has the ability to sense changes in light levels, for instance due to
shadows, and communicate this information wirelessly, all at tens of microwatts of
power. Furthermore, the VLM has the ability to harvest energy from indoor light
to power itself, thus eliminating the need for any batteries. Instead, an on-board
capacitor can be used as an energy storage device.
Figure 4-1 illustrates the concept design of the VLM. The VLM makes use of visible light as a sensing and energy harvesting medium, and RF backscatter as a communication method. The system works in the following way: first, a commodity radio or a radio tower generates a carrier signal at frequency f
c. Next, the VLM senses a significant change in the perceived light readings, and enables the backscatter frontend to modulate the carrier signal with the sensor readings and reflect it at a frequency f = f
c+ ∆f . Finally, a receiving device processes the backscattered data to reveal the sensor information.
Sensor reading
Receiver Visible Light Marker
Commodity Radio
Carrier
Figure 4-1: Visible light sensing and communication at µWs of power. In this scenario, a commodity radio generates a carrier signal, and the VLM senses a shadow under sunlight. In another scenario, the VLM can backscatter RF signals from radio towers and sense shadows under artificial light.
We will present two designs of the visible light marker: the thresholding visible light marker (tVLM) and the analog visible light marker (aVLM). Both designs avoid computational blocks to enable battery-free operation. The tVLM trade-offs sensing resolution for ultra-low power consumption, while the aVLM can communi- cate sensor information at a higher resolution at the expense of power consumption.
The idea behind using two different designs for the VLM is that they can be used
to fit the requirements of different application scenarios while being optimized for
ultra-low power consumption. For instance, to enable human occupancy detection,
it may suffice to use the tVLM as it has the capability to sense when the light
levels drop below a certain threshold, and thus detect whether or not a person
is present in proximity to the light sensor. The ultra-low power consumption of
1
2 3
4