LiFi for Low-Power and Long-Range RF Backscatter

A

Solar

cell

Harvester

MCU

LiFi for Low-Power and Long-Range

RF Backscatter

Muhammad Sarmad Mir, Borja Genoves Guzman, Member, IEEE, Ambuj Varshney, and

Domenico Giustiniano, Senior Member, IEEE.

Abstract—Light bulbs have been recently explored to design

Light Fidelity (LiFi) communication to battery-free tags, thus

complementing Radiofrequency (RF) backscatter in the uplink.

In this paper, we show that LiFi and RF backscatter are

complementary and have unexplored interactions. We introduce

PassiveLiFi, a battery-free system that uses LiFi to transmit

RF backscatter at a meagre power budget. We address several

challenges on the system design in the LiFi transmitter, the tag

and the RF receiver. We design the first LiFi transmitter that

implements a chirp spread spectrum (CSS) using the visible

light spectrum. We use a small bank of solar cells for both

communication and harvesting, and reconfigure them based

on the amount of harvested energy and desired data rate.

We further alleviate the low responsiveness of solar cells with

a new low-power receiver design in the tag. We design and

implement a novel technique for embedding multiple symbols

in the RF backscatter based on delayed chirps. Experimental

results with an RF carrier of 17 dBm show that we can generate

RF backscatter with a range of 92.1 meters/µW consumed in the

tag, which is almost double with respect to prior work.

Index Terms—Battery-free, Internet of Things (IoT), RF

backscatter, Visible Light Communication (VLC).

I.

I

NTRODUCTION

Continuous increase in deployment of Internet of Things

(IoT) devices leads to massive use of batteries, as they

energize the IoT devices. Although batteries in a small form

factor may last for a long time, even years, any computation

and active communication can quickly deplete them, and it

calls for solutions that do not need batteries at all. Additionally,

batteries generate hazardous waste due to their chemical

composition and also have a negative environmental impact,

as consumers currently dispose of billions of batteries per year

and battery recycling is a delicate matter [1], [2].

The research efforts in battery-free systems exploit low-

power electronics, power harvesting, communication and pro-

cessing techniques [3]–[6]. RF backscatter is now a mature

technology for transmitting IoT data to the network because of

its energy efficiency and absence of power-hungry active radio

for transmission. In fact, the scarce amount of harvested energy

from the environment limits the communication and process-

ing capabilities. In particular, energy is mainly harvested from

RF [7], light [4], [5], [8], kinetic [9] and thermal [10], [11]

Muhammad Sarmad Mir is with the University Carlos III of Madrid,

Legane´s, 28911, Spain. E-mail: sarmadmir2003@gmail.com

Borja Genoves Guzman and Domenico Giustiniano are with IMDEA

Networks Institute, Legane´s, Spain, 28918.

E-mail: {borja.genoves, domenico.giustiniano}@imdea.org

Ambuj Varshney is with National University of Singapore, Singapore,

119077. E-mail: [email protected]

LiFi

transmitter

Tag top side Tag bottom side

Fig. 1: PassiveLiFi: hardware prototypes of LiFi transmitter

and passive tag. The tag comprises a solar cell array for energy

harvesting and downlink communication (top side), a LiFi

module for downlink, a harvester circuit and an RF backscatter

module for uplink communication (bottom side).

sources. A solar cell is typically used for harvesting energy

from light, and it provides the best trade-off between the level

of energy provided and the availability of sources [12], [13].

Only limited work has been conducted to exploit commodity

solar cells in battery-free IoT devices for communication.

EDISON [13] has shown the design of a battery-free IoT tag

that receives data through light, a concept commonly called

Visible Light Communication (VLC) or Light Fidelity (LiFi)

in a networked system. It then sends data through RF backscat-

ter. EDISON demonstrated that LiFi and RF backscatter are

incomplete as standalone technologies for passive communica-

tion, but have complementary properties that can be exploited

to use LiFi in downlink and RF backscatter in the uplink. LiFi

provides downlink communication with ease of deployment

and delivers optical power to harvest. Whereas for uplink

in battery-free systems, LiFi is unsuitable due to its higher

energy consumption and user visual comfort issues [14]. RF

backscatter, on the other hand, is extremely energy efficient for

uplink but classical radio frequency envelope detectors used as

receivers for IoT are affected by low sensitivity, false detection

alarms, and low energy efficiency, which hinders its use for

downlink [13].

In this work, we introduce PassiveLiFi, shown in Fig. 1,

whose preliminary design was presented in [15] by the same

co-authors. Here we present an extended and renewed version

which includes the novel implementation of uplink modulation

Heat

sink

LED

MOSFET

Current

regulator

0

1

0

1

0

1

Data

Visible Light Chirp

1

0

1

Downlink

Uplink

Fig. 2: Downlink (left): Light intensity is changed to send data to the passive tag at a fixed clock rate. Uplink (right): Carrier

and baseband delegated to the infrastructure. Chirps are complex to generate at the tag, and hence we delegate them to the

light infrastructure. The light intensity is changed to generate visible light chirps at a varying clock rate. This chirp is then

mixed in the tag with the input RF carrier for RF backscatter.

for increasing data rate, besides engineering improvements

such as impedance matching to enhance the range and per-

formance comparison of different types of antennas for RF

backscatter communication. We also integrate VLC and RF

backscatter module into a single PCB and transmit real modu-

lated data to make it closer to IoT market. These improvements

and novelties will be commented along the paper.

PassiveLiFi exploits the unexplored interactions between

LiFi communication in the downlink and RF backscatter in the

uplink. As we will show in this work, these interactions allow

us to significantly increase both the range for RF backscatter

and the energy efficiency of the IoT tag. We use LiFi not

only to transmit downlink data but also to generate the clock

signal needed by the IoT tag to transmit RF backscatter in

the uplink, thus removing the need of a clock in the IoT tag.

A first approach could be to modulate the LiFi bulb with a

simple On-Off Keying (OOK) modulation and use this signal

as a clock in the IoT tag. This approach would already result

in energy saving in the IoT tag. However, the RF backscatter

communication range would be similar to prior design that

used oscillators in the IoT tag for the same purpose [13].

In order to increase both the communication range in RF

backscatter and decode signals drowned by the noise, we

present the first implementation of the chirp spread spectrum

(CSS) using the visible light spectrum. This visible light signal

is received by the solar cells in the IoT tag, and used there

as a baseband signal to communicate with RF backscatter

by turning the chirp on and off based on the bit stream.

Generating chirp spread spectrum in the tag consumes around

10 mW using off-the-shelf components [16], and offloading

it to the infrastructure while completely removing the need

of oscillators for passive chirp spread spectrum has been

not shown so far. A high-level illustration of PassiveLiFi is

presented in Fig. 2, where we show the operations both in

downlink and uplink.

The first problem we have to solve in order to implement

the chirp spread spectrum in LiFi is that commercial light

bulbs could modulate the light intensity at speeds in the

order of a few Mb/s. However, solar cells have not been

designed for communication, and thus they have inefficiencies

as receivers that must be addressed to sustain a sufficiently

high data rate [17]. Furthermore, delegating chirp generation

to the infrastructure requires that the LiFi receiver in the IoT

tag consumes low power, smaller than the one consumed by

the local oscillators for performing CSS modulation. Yet, low-

power LiFi receivers are based on a light power envelope and

are sensitive to any source of light interference, like other light

fixtures and the sun.

A second problem is that prior work used two different

solar cells, one for communication and one for harvesting [13].

However, this has two drawbacks: it increases the size of the

tag or, if we keep the same tag area, it does not exploit all

available light energy for both communication and harvest-

ing. Besides, solar cells are typically designed to work with

solar energy, but their effectiveness with indoor and artificial

lighting conditions is less known [18], [19].

Our contributions can be summarized as follows.

•

We present the first design of the chirp spread spectrum

using LiFi, and propose to use LiFi in two modes of

operation: the first one for communicating downlink data,

and the second one for generating the chirp signal needed by

uplink RF backscatter. In the first mode, it uses a traditional

constant clock rate, while, in the second mode, the clock

rate changes based on the desired bandwidth and spreading

factor;

•

We propose a design that uses a single solar cell both for

communication and harvesting, decoupling the modulated

LiFi signals received from light bulbs from the light energy

that can be used for harvesting. We show that the problem

of optimizing both communication and harvesting with a

solar cell follows a Pareto curve and we propose a criterion

to select the best solar cells for both communication and

harvesting;

•

We show a unique low-power technique to modulate chirps

to transmit uplink data by creating delayed versions of chirp

on the tag. We implement a basic and advance modulation

modes on the tag and demodulation of the backscattered

signal on the RF receiver.

•

We implement PassiveLiFi with customized hardware both

on the LiFi transmitter and IoT tag, and we evaluate our

system in a variety of scenarios. Our experiments show

that PassiveLiFi can transmit RF backscatter signals with

a meter/power consumed metric that is almost doubled with

respect to the state of the art.

The rest of the paper is structured as follows: In Section II,

we present challenges faced by state-of-the-art systems, and

we place our system in context to them. We also provide

a high-level overview of our system. Next, in Sections III

and IV, we describe the design of the LiFi transmitter and

the tag, respectively. In Section V, we evaluate the system in

terms of range, energy harvesting and power consumption in

different scenarios. Next, in Section VI, we present application

scenarios that our system could enable. Finally, we discuss

prior works related to our system, and we conclude the paper.

II.

C

HALLENGES

We discuss the challenges we address in this work and

position them with respect to prior work in the literature.

A.

Delegating oscillators

RF backscatter absorbs and reflects the surrounding radio

waves to communicate with battery-free devices. In contrast,

radios technologies such as Bluetooth, LoRa, WiFi and Zigbee

use active transmission which makes them power-hungry

with consumption in a range of milliwatts as presented in

Table I [20]. An alternative option is to employ LiFi in uplink

but it necessitates the modulation of LED on IoT devices

which again results in increased power consumption, making it

unsuitable for uplink. On battery-free devices, achieving low-

energy consumption for every device is essential to enable its

operation on the small amounts of energy harvested from the

ambient environment which makes RF backscatter an ideal

choice for uplink. On the backscatter tags, the oscillator’s

energy dominates the overall energy consumption, and it is

the order of tens of µW (demonstrated through simulation

or implementation [21]). Further, these oscillators are often

combined with other circuits such as those to generate chirps

for communication, which further pushes the complexity and

energy consumption [22]. It makes it prohibitive to operate

these platforms on the harvested energy. Recent systems over-

come the oscillators’ energy-expensive nature by delegating

oscillations to an external and powered RF infrastructure [21].

This leads to lowering the power consumption and complexity

of the backscatter tag. However, the communication range is

not sufficient for most applications, and it is in the order of

2 m. One possible approach to increase the communication

range is to employ chirps for communication. However, gen-

erating these chirps locally at the tag is an energy-expensive

operation. Prior work has also tried to delegate the energy-

expensive process of generating chirps [22]. However, it still

required an oscillator at the tag to shift this signal by 1-2

MHz to avoid self-interference and backscatter it back to the

RF receiver. This leads to an increased complexity and power

consumption of the backscatter tag.

Offloading chirp spread spectrum signals to the infrastruc-

ture while completely removing the need of oscillators for

passive chirp spread spectrum has been not shown so far. Yet,

the ability to offload chirp signals could result in a much larger

communication range than using simpler modulations which

are prone to error [23]. Delegating oscillations to the infras-

tructure requires that the power budget needed for downlink

reception in the tag is lower than the one consumed by its local

oscillators. This helps to take advantage of chirp modulation

and reduction in power consumption of the tag. Furthermore,

TABLE I: Comparison of RF backscatter against wireless

communication technologies

Technology

Power

consu

mption

BLE

30.03

mW

(CC2651R3)

LoRa

128.37

mW

(RN2483)

Zigbee

192

mW

(AT86RF215)

WiFi

880.6

mW

(TI

WL1801MOD)

RF

backscatter

<

100

µ

W

the received signal must be also sufficiently precise to be used

as an oscillator. This is difficult to achieve because of the

limitation of passive envelope detectors, commonly used as RF

receivers in tags. In fact, passive envelope detectors aggregate

all energy received in the band, and cannot select the desired

frequency as a clock. Besides, any ambient traffic could trigger

simple RF envelope detectors, increasing the consumption of

the tag [13].

Instead of delegating the chirp spread spectrum to the RF

infrastructure, we propose to use light bulbs for generating

visible light chirps that can be detected by low-power LiFi

receivers. These receivers can provide better baseband signals

than their RF counterparts for two reasons:

•

LiFi transmission follows an Intensity Modulation (IM)

baseband procedure, where the modulation of the optical

power of the LiFi transmission carries the information, and

the signal phase does not carry the information. Instead,

the receiver carries out a Direct Detection (DD) to convert

the optical received signal into an electrical signal. In its

simplest form, LiFi requires just turning on and off the Light

Emitting Diode (LED) in the bulb with the desired pattern

to transmit a bit stream. We instead cannot send RF signals

in the baseband and they require an RF carrier.

•

Passive LiFi receivers can be designed with low-power

consumption, yet the light propagation can be much better

controlled than the RF propagation. Light is more confined

than RF and, as a consequence, LiFi receivers may receive

fewer interfering signals. The main source of interference

is the sunlight, which is not modulated, and therefore can

be filtered out at the receiver, and other sources from older

technologies, such as fluorescent lights, are disappearing.

B.

Communication and harvesting

In passive LiFi systems, the receiver relies on solar cells

both for communication and harvesting. Solar cells are ad-

vantageous with respect to other optical receivers such as

photodiodes because they operate fully passive, without the

usage of any active amplifier [13]. In order to use the overall

light-sensitive area, we advocate for a design that uses the

same solar cell for both communication and harvesting. A

simple approach would be to slice the time such that a

certain portion of time is dedicated to harvesting and the

rest to communication. However, this would result in poor

efficiency. More formally, let us define T

com

as the time to

communicate N bits and T

h

as the time to harvest enough

energy to transmit N bits. Because of the latency required

for harvesting, the time left for a single battery-free device to

communicate data would be largely reduced and T

com

would

increase significantly. Furthermore, the time needed to harvest

energy could disrupt any protocol that needs to use the same

solar cell for communication.

Light

power

Average

ON

OFF

ON

OFF

ON

OFF

Rather than using the same solar cell in different slices of

time, we aim to use it at the same time both for communication

and harvesting, without losing any energy that could be useful

for harvesting. However, the photonics community has always

Time

Average

ON

OFF

considered this unrealistic because of how photodetectors (and

solar cells are just one type of them) work. Fundamentally,

in order to receive data, photodetectors are biased in reverse

mode (photoconductive), meaning that there is a higher voltage

to the negative terminal with respect to the positive terminal of

the photodetector. In contrast, when the photodetector operates

in photovoltaic mode to harvest energy, it is positive or

zero bias, and hence the voltage is with the opposite sign

with respect to communication mode. Photoconductive mode

improves the response of the photodetector but it requires

the availability of negative bias voltage on the device and

causes an increase in power consumption. In this work,

we operate solar cell in photovoltaic mode (zero biased) to

eliminate the dark current and maximize the low-illuminance

performance [24]. We present a new low-power LiFi receiver

to solve the problem of different timeslots for harvesting

and communication, leveraging the fact that communication

and harvesting use different frequency components of the

same signal. Therefore, for simultaneous communication and

harvesting, we post-process the voltage signal received by the

solar cell and, by using a low-pass filter and a high-pass filter,

disentangle it into two components, one for harvesting and

another for communication, respectively. We further propose

to use a small set of solar cells instead of a single larger one

to optimize communication and harvesting as per requirement.

C.

Uplink data rate

In battery-free systems, there is a constraint of an acutely

limited energy budget. In passive LiFi systems, to save energy,

we can offload generation of the clock to infrastructure but

still, we need to modulate data in the uplink to send informa-

tion from tag to infrastructure. Simple modulation schemes,

like OOK, are energy efficient but at the cost of short-range

and less robustness to interference and noise.

We instead modulate the uplink signal by creating different

delayed versions of the chirp using low-power electronics. This

method ensures a long-range and robust communication link

with minor changes on the tag as compared to [15]. Our system

also provides the flexibility to select a number of delay lines to

increase the data rate as per requirement or as per the energy

available. This feature makes PassiveLiFi tag suitable for a

number of applications.

In what follows, we address the limitations presented in

this section for low-power battery-free devices and present

PassiveLiFi, composed of:

•

LiFi transmitter to communicate to the tag and generate the

baseband signal for uplink communication (Section III);

•

battery-free tag to receive and process LiFi signal, harvest

energy from the solar cells, and provide uplink mixing

the input RF carrier and the LiFi baseband signal for RF

backscatter communication (Section IV).

The system is complemented by the RF infrastructure to

provide carrier signal for RF backscatter and process the

received backscatter signal.

(a)

LED for light dimming.

(b)

LED as a communication source.

Upchirp Clock (constant frequency)

(c)

LED as a clock generator.

Fig. 3: Varying LED intensity can serve multiple requirements.

Its first application was light dimming (a); with LiFi, it has

been used to transmit downlink data (b); in this work, we pro-

pose to use it as a clock generator, varying its frequency over

time to produce baseband signal for uplink communication (c).

III.

L

I

F

I TRANSMITTER

In PassiveLiFi, the LiFi transmitter provides the illumination

to fulfill the requirements of indoor lighting standards. It

provides the energy to the tag to support battery-free operation

for indoor deployments, and the baseband signals to support

downlink communication and oscillations to support the RF

backscatter-based uplink channel. The prototype we have built

of the LiFi transmitter is shown at the top of Fig. 1.

A.

Multiple roles of light bulbs

We are observing a rapid deployment of LED lighting

in homes, offices and streetlights because of their energy

efficiency and long lifespan. We refer to Fig. 3. Typically

LEDs are driven by a switching power circuitry that operates

at a high frequency. This driver has been first used for light

dimming by controlling the amount of time the light is on with

respect to the time is off

1

. More recently, LEDs have started

to be employed to generate LiFi signals, where the intensity of

light is modulated to convey information. In its simplest form,

LiFi communication associates bit 1 to high light intensity and

bit 0 to low light intensity. In turn, a baseband signal is emitted

by the bulb in the visible spectrum.

In this work, we propose to change the light intensity of

LED bulbs for a third purpose, suitable for creating passive

LiFi communication. We create a baseband signal with LiFi

that can be mixed at the IoT tag with an RF carrier signal.

This super-imposed signal can then be modulated by the tag,

simply turning on and off the RF signal that is reflected. This

clock signal can be used to offload the oscillator in the IoT tag

to the LiFi transmitter. One key advantage of this approach is

the energy saving in the IoT tag thanks to offloading of the

oscillator to the LiFi transmitter and the removal of power-

hungry elements on the tag.

1

Pulse width modulation is typically used for this purpose.

Vs

L

C

in

D

LED

Baseband

Signal (S

1

)

R

SNS

C

o

Gate

Driver

Vin

SW

Switching

Regulator

DIM

CS

DC-DC

Converter

(a)

OpenVLC transmit signals.

5V

14V

S

1

(b)

New LiFi TX signals.

Fig. 4: Comparison of 100 kHz signals transmitted by Open-

VLC1.3 (only 10.9-9.5=1.4 Vpp and also with relevant capac-

itance effect) and our improved LiFi transmitter (peak-to-peak

voltage is 12 V, with a very sharp waveform).

B.

Bandwidth in passive downlink

As discussed in Section II-A, passive downlink communi-

cation requires a very low-power receiver. Any distortion in

the signal received by the tag could inevitably cause errors

in the interpretation of the bit pattern. We study this problem

by measuring the signal transmitted using the open source and

low-cost OpenVLC1.3 board [25]. This platform has been also

used by EDISON as a LiFi transmitter. We transmit a 100 kHz

signal using OpenVLC1.3, measure the voltage at the LED

pins and plot the result in Fig. 4a. We observe that the shape

of the transmitter signal distorts at higher frequencies, with

a transient time from 90% to 10% of about 0.8 µs, which is

16% of the duration of one bit.

An active receiver could easily handle this transition time

and operate up to 1 Msample/sec (as shown in OpenVLC [25]).

On the contrary, passive LiFi communication requires a

baseband signal as ideal as possible with sharp rising and

falling edges, such that a simple comparator of light intensity

could be effective to distinguish high and low light intensity.

Another problem is that OpenVLC operates the LED at a

low forward voltage of 10.9 V and current of 175 mA. As the

relation between LED current and the output luminous flux is

approximately linear, this design leads to poor harvesting and

communication capabilities.

We modify the OpenVLC design with the goal of achieving

a sharper baseband signal with low-cost hardware, and exploit

the full dynamic range of the LED. We use the same LED as in

OpenVLC, but we largely improve the front-end design. We in-

crease the harvesting capabilities and range of communication

by operating the LED at a higher forward voltage. OpenVLC

uses resistance in series to the LED, which wastes energy,

and it cannot work at higher current levels. We instead use a

switching regulator based LED transmitter design, widely used

for commercial LED luminaries. This allows us to operate the

same LED at the highest current possible (550 mA), provide a

sharper transmitted signal at higher frequencies and dissipate

only 10% energy as heat and switching losses, contrary to

51.6% for a linear regulator such as OpenVLC1.3.

The schematic of our LiFi transmitter is shown in Fig. 5a,

and the hardware prototype in Fig. 1. The regulator that we

use operates in continuous conduction mode to maintain a

positive current through the inductor L and rectifies the biggest

(a)

LiFi transmitter design.

Comparator

(b)

Block diagram of battery-free IoT tag.

Fig. 5: LiFi transmitter and battery-free IoT tag.

4000

3000

2000

1000

0

0.5 1 1.5 2 2.5 3 3.5 4

Distance (m)

Fig. 6: Comparison of illumination provided by OpenVLC TX

and our design. Illuminance is multiplied by 10 at a distance

of 1 m (1350 lux vs. 134 lux) and a larger distance can be

achieved while illuminating at typical illuminance values.

delay in turning the LED on and off. The parallel N-channel

MOSFET is used to increase the slew rate of LED to achieve

a high switching frequency. The MOSFET gate driver is used

to provide high current in order to overcome the effect of gate

capacitance in high switching frequencies required to generate

the chirp signal.

We outperform the OpenVLC design. From our tests, we

observe that by setting V

s

at 5V, we achieve a sharper signal

across the LED, as it can be seen in Fig 4b. The experiments in

Fig. 6 show that the measured illuminance with a lux meter is

multiplied by 10 at a distance of 1 m with respect to OpenVLC,

enabling larger scenarios with LED lighting [26].

C.

Visible light chirps

As discussed in Section II-A, we propose to use light bulbs

for delegating oscillations. For instance, with PassiveLiFi,

we can generate the RF signal at 868 MHz and the LiFi

signal at 100 kHz. The IoT tag can passively mix them to

generate an operation frequency of 868.1 MHz for the uplink

RF communication. Yet, this approach would improve only

the energy efficiency, but not the range of communication.

Instead, we propose to delegate the generation of chirp sig-

nals to LiFi, as shown in Fig. 2. Chirp spread spectrum (CSS)

can achieve a longer range with respect to simpler modulations

(e.g., On-Off keying), as successfully shown in LoRa [27],

thanks to the property of CSS to demodulate signals below the

HPF

Delay

Stages

Sensor

Solar

Cell

LPF

MCU

MUX

Harvester

C

Voltage

Regulator

Z

1

Z

2

RF Switch

OpenVLC TX

Switching based Transmitter

Illumination (Lux)

Time to charge (parallel) Vpp (parallel)

Time to charge (series) Vpp (series)

300

200

100

500

400

300

200

S

P

closed

S

N

open

S

P

open

S

N

closed

0 100

1000 1500 2000 2500 3000

Area of Solar cells (mm

2

)

Fig. 7: Effect of solar cell area on communication and harvest-

ing using up to five solar cells in parallel or series. The energy

harvesting ability improves when the solar cells are connected

in parallel, whereas connecting them in series improves their

ability to receive downlink communication.

noise level, being also more robust to multipath. We propose

to use light bulbs for generating visible light chirps, with the

objective of improving both energy efficiency and the range

of uplink communication. In PassiveLiFi, the LiFi transmitter

sends a clock with varying frequency over the visible light

channel that increases over time (up-chirp signal). Note that

there is no light flicker with our implementation of CSS as we

work at a sufficiently high frequency, starting from 40 kHz.

Next, the tag receives these transmissions using a low-power

solar cell-based LiFi receiver and further modulates the signal

based on the information to be transmitted. On the receiver

side, symbols are detected by the energy observed at different

FFT bins, correlating the received signal with a down-chirp

signal (cf. Fig. 2).

IV.

I

O

T

TAG

The core of our end-to-end communication system is the

battery-free IoT tag. The tag operates solely on harvested en-

ergy from the solar cell. Solar cells are preferred for harvesting

because of the widespread availability of light sources and a

higher level of harvested energy with respect to RF [12]. RF

sources are also limited in space and deploying dedicated RF

source has practicality issues. Furthermore, RF power sources

transmitting strong signals are needed to achieve reasonable

harvesting (3 W transmitters to achieve less than 200 µW of

power harvested at 5 m [28]). For the solar cells, we consider

a total size of 30 cm

2

(4.6 inch

2

), which is similar or smaller

with respect to the state of the art [4], [5], [13].

The design goals for the tag include:

•

Use of single solar cell for both harvesting energy and

downlink communication;

•

Energy thresholding circuit design in the tag robust to indoor

lighting and LiFi communication frequency;

•

Use of downlink chirp signal to enable long-range and low-

power uplink backscatter communication;

•

Modulate data by the implementation of frequency transition

on the tag with low power consumption;

•

Impedance matching to enhance the strength of backscatter

signal;

•

Use PCB antenna to reduce the form factor of tag and make

it closer to market demand;

•

Ultra low-power design to enable maximum operation time

on harvested energy.

The block diagram of the IoT tag is shown in Fig. 5b.

Fig. 8: Combination of a solar cell array in series or parallel

depending on the charge level of battery/capacitor.

A.

Trade-offs with solar cells

In this work, we propose to use a small set of solar cells

instead of a single larger one, and use all of them for both

harvesting and communication. Yet, we find that there exists

a dichotomy between harvesting and communication that we

have to solve. We perform an experiment where we measure

the time the solar cell takes to charge a capacitor of a specific

value, the time to charge (T

c

) as well as the peak-to-peak

value of the voltage measured at the receiver after the solar

cells (V

pp

). The experiment is performed at a distance of 1.5 m

between our LiFi transmitter and receiver at 50 kHz frequency

without background light.

We need small time to charge (we can harvest more quickly)

and a high V

pp

(we can operate at a longer range). As

represented in Fig. 7, this can be obtained using a larger total

area of the solar cells, thus using all solar cells for both har-

vesting and communication. However, the harvesting improves

considerably (time-to-charge decreases) with multiple (up to

five) solar cells connected in parallel, while the communication

worsens slightly, due to a lower V

pp

value. On the other

hand, when multiple solar cells are connected in series, the

communication is boosted (larger V

pp

) and the time-to-charge

slightly decreases. The effect on time-to-charge is mainly due

to the harvester, which uses the maximum power point tracking

(MPPT) to efficiently draw power from solar cell to charge the

output capacitor. Solar cells connected in parallel add up the

current to deliver more power and reduce the time-to-charge as

compared to series connections where the voltage difference

of each solar cell sums up and the current slightly decreases.

B.

Reconfiguring the solar cells

The decision of parallel or series connection of solar cells

is based on V

BAT

which is the voltage across the capacitor

to store harvested energy. The harvester BQ25570 generates a

Battery OK digital signal depending on the state of V

BAT

.

When V

BAT

is above the threshold (programmable by re-

sistors), the Battery OK is high and it toggles when V

BAT

drops below the threshold. The configuration of solar cells

can be switched between series and parallel by connecting the

Battery OK signal to gates of N-channel MOSFETs (S

N

) and

P-channels MOSFETs (S

P

) as shown in Fig. 8. We need ’n-

1’ N-channel and ’2n-2’ P-channel MOSFETs for the design

where ’n’ is the number of solar cells used. ADG72X [29]

switches can be used due to their low power dissipation (< 0.1

µW) and tiny package. In this way, the connection among

solar cells is reconfigurable automatically: when harvesting is

Solar

Cell

1

S

P

S

P

S

N

Solar

Cell

2

S

P

S

P

S

N

Solar Cell 3

Battery_OK

Time to charge (sec)

Vpp (mV)

300

250

200

150

2000

1500

1000

300

250

200

150

2000

1500

1000

300

200

100

300

200

100

50

0

200

600

1000

500

0

1400

100

50

0

200

600

1000

500

0

1400

0

-2000 -1000 0

-V

pp

(mV)

(a)

LiFi at 50 kHz.

0

-800 -600 -400 -200 0

-V

pp

(mV)

(b)

LiFi at 100 kHz.

Illuminance (Lux)

(a)

LiFi at 50 kHz.

Illuminance (Lux)

(b)

LiFi at 100 kHz.

Fig. 10: Representation of Pareto fronts for each illuminance

value when considering shortlisted solar cells at different LiFi

Fig. 9: Comparison of peak-to-peak voltage and time to charge

100µF capacitor for shortlisted solar cells at different LiFi

transmission rates. For the three solar cell types, the exposed

area is 30 cm

2

.

the priority due to the low charge on a capacitor, solar cells

are connected in parallel; when harvesting is not a priority, to

boost the communication they are connected in series. Note

that, although harvesting or communication is being prioritized

each time, both actions occur simultaneously.

C.

Comparison of commodity solar cells

transmission rates.

0.1

0.05

0

-0.05

-0.1

200 600 1000 1400

Illuminance (lux)

(a)

f =50 kHz, α = 0.5.

0.1

0.05

0

-0.05

-0.1

200 600 1000 1400

Illuminance (lux)

(b)

f =100 kHz, α = 0.5.

There exist several solar cells in the market for IoT appli-

cations, and we study how to select the best-performing solar

cell in terms of harvesting and communication performance.

Although solar cells in the market are all low cost (4-5

dollars each), their efficiency for harvesting varies largely

(from 3 to 25%) as well as their size. Specifications of the

communication performance are not given, as solar cells are

designed typically only for harvesting. We study a total of six

different commodity solar cells, and shortlisted three based

on good performance both in communication, (V

pp

), and in

harvesting, time-to-charge.

Fig. 9 compares the three best solar cell types under eval-

uation. As the selected solar cells have different sizes, for

carrying out a fair comparison, we connect several solar cells

of each type in order to create the same total area. In total,

we create a solar cell of approximately 3000 mm

2

by unifying

5, 4, and 3 solar cells of ‘SLMD121H04L’, ‘SLMD600H10L’

and ‘SLM141K06L’, respectively. Following our analysis in

Section IV-A, solar cells are connected in series for V

pp

results, as the voltage in the output of each solar cell is

summed up. Differently, they are connected in parallel for

time-to-charge results, as the current in the output of each solar

cell is added to contribute to faster harvesting. We observe

that time-to-charge monotonically decreases with illuminance,

whereas V

pp

monotonically increases with illuminance, which

contributes to faster harvesting and better communication,

respectively. However, the frequency of LiFi transmission does

not affect the time-to-charge, but the V

pp

decreases when

the LiFi rate increases due to the low bandwidth of the

solar cell. In fact, the capacitance of solar cells distorts the

received signal and, as a consequence, the V

pp

value. In the

next section, we search for a Pareto-optimal solution [30],

as there is not a single solar cell type that provides the best

performance in both communication and harvesting.

Fig. 11: Representation of function to minimize versus illumi-

nance. The figure shows how the solar cell ‘SLMD121H04L’

provides the best performance both in communication and

harvesting. Note that curve belonging to ‘SLMD600H10L’ is

zero for all illuminance values because this solar cell achieves

T

c

,

max

(

l,

f

)

and

V

pp

,

max

(

l,

f

)

values.

D.

Criterion to choose the solar cell

The aim of this subsection is to choose the best solar cell

type in terms of communication and harvesting. Communica-

tion is optimized by maximizing V

pp

, i.e., minimizing –V

pp

,

while time to charge (T

c

) is optimized by minimizing it.

Fig. 10 shows the Pareto fronts for fixed illuminance and fre-

quency, which demonstrates that the solar cell ‘SLM141K06L’

is Pareto-dominated by ‘SLMD121H04L’. However, we ob-

serve that both ‘SLMD121H04L’ and ‘SLM600H10L’ are

within the Pareto-front, which means that both are Pareto

efficient. To select a single solar cell type as the best solar

cell for our scenario, we convert the problem into a unique

objective function to be minimized, by using the weighted

sum method as

f

1

=

α

·

T

c

,

norm

(

T, l,

f

)

−

(1

−

α

)

·

V

pp

,

norm

(

T, l,

f

)

,

(1)

where

α

is the weight that is typically set by the de-

cision maker,

T

∈ {

A, B, C

}

=

{

‘SLMD121H04L’,

‘SLM600H10L’, ‘SLM141K06L’

}

represents the solar cell

type, and

l

and

f

are the illuminance and LiFi frequency,

respectively.

The computation of the optimal solar cell can be derived

from the analysis in the Appendix of [15]. From there, Fig. 11

represents f

1

for each solar cell type versus illuminance for

50 kHz and 100 kHz of transmission rate and considering

α = 0.5 (equal importance for communication and harvesting).

In such a figure, solar cell ‘SLMD121H04L’ provides the

SLMD121H04L

SLMD600H10L

SLM141K06L

SLMD121H04L

SLMD600H10L

SLM141K06L

SLMD121H04L

SLMD600H10L

SLM141K06L

SLMD121H04L

SLMD600H10L

SLM141K06L

SLMD121H04L

SLMD600H10L

SLM141K06L

SLMD121H04L

SLMD600H10L

SLM141K06L

Time to charge (sec)

V

pp

(mV)

Time to charge (sec)

V

pp

(mV)

f

1

Time to charge

(sec)

f

1

Time

to

charge

(sec)

Output

R

C

1

R

1

Delay 1

1x8 µs

2x8 µs

Delay 2

nx8 µs

Delay n

RF

switch

nx1

Mux

350

700

0.05

300

600

500

0

-0.05

(a)

EDISON’s thresholding circuit. (b) Proposed thresholding circuit.

Fig. 12: Configuration of the thresholding circuit. We integrate

LPF for harvesting and HPF for communication purposes. This

improves the robustness of the thresholding circuit.

250

200

150

400

300

200

100

0

0 2000 4000 6000 8000

R

2

[ ]

-0.1

-0.15

-0.2

-0.25

-0.3

0 5000 10000 15000

R

2

[ ]

lowest f

1

value for typical lighting conditions in indoor

environments [26] [31]. However, for larger illuminance values

‘SLMD600H10L’ becomes the best solar cell due to the larger

differences in V

pp

(see Fig. 9). Considering the results obtained

(a)

Peak-to-peak voltage

(dashed) and Time to charge

(solid) for different R

2

values and LiFi rates.

(b)

The optimal point for

both communication and har-

vesting is R

2

= 4 kΩ, when

LiFi rate = 50 kHz.

in Fig. 11 and as illuminance values for indoor workplaces are

typically lower than 1200 lux [26], the solar cell with the best

harvesting and communication capability is ‘SLMD121H04L’.

After the selection of the solar cell, we select the number

of solar cells to use and their configuration. As the size of our

prototype tag is 75 x 50 mm, and as the larger the number of

solar cells, the better are communication and harvesting (see

Fig. 7), we place 5 SLMD121H04L solar cells on the back

side of the tag to fully cover the area, as shown in Fig. 1.

E.

Receiver circuitry

As a next step, the DC and AC components at the output

Fig. 13: Calculation of the optimal R

2

value for both commu-

nication and harvesting functionalities when LiFi rate is 1 kHz

and 50 kHz.

Upchirps

of the solar cell are separated using a low pass filter (LPF)

and high pass filter (HPF) respectively, as shown in Fig. 12b.

Delay

t

s

Uplink

Data

For comparison, the thresholding circuit in EDISON uses a

dedicated solar cell, as shown in Fig. 12a. The photocurrent

from the solar cell consists of both AC (i

sc

) and DC compo-

nent (I

sc

). The DC component is blocked by C

1

and passes

through the branch for harvesting energy. The AC component

flows through both the branches but it is highly attenuated

by C

2

[32]. The best communication can be achieved with

a large value of R

2

which acts as an open circuit and all

AC component pass through the HPF. The optimization of R

2

is important for simultaneous communication and harvesting,

as it causes a trade-off between communication range and

time to charge. Larger the value of R

2

greater will V

pp

and

T

c

be as shown in Fig. 13a. However, note that V

pp

is not

improved from a R

2

value on, whereas the time to charge

keeps increasing. In order to find the optimal R

2

value to

operate by optimizing both communication and harvesting

simultaneously, we develop the same method as the one used

for finding the optimal solar cell type. The optimal R

2

value

is 4 kΩ, where V

pp

starts saturating and from this point on the

harvesting (time to charge) worsens dramatically. However, we

identify that the optimum R

2

depends on the data rate: at low

data rates (shown in Fig. 13a), V

pp

is larger than V

pp

,

min

for

all R

2

values.

For the sake of simplicity, unless other data is specified,

from this point on, we will perform with the optimal solar cell

and optimum R

2

value, i.e., solar cell type ‘SLMD121H04L’

and R

2

= 4 kΩ. The values of C

1

and R

1

in HPF are selected

to rectify the low-frequency noise from ambient lighting.

HPF also makes our system robust to commercial LEDs

Fig. 14: Modulation of recovered chirp signal using delay

stages for uplink RF backscatter communication on the tag.

with dimming functionality as their pulse width modulation

(PWM) frequency is normally below 500 Hz. Moreover, the

HPF removes the DC component of the signal and translates

the signal down to the ground as average.

F.

Uplink Modulation

A novel low-power modulation technique is presented for

uplink communication. The recovered chirps on the tag are

delayed by a predefined time to embed information while

keeping the energy consumption to a few µW and achieving

robustness and long-range communication.

For uplink modulation we use two different modes: 1) Basic

and 2) Advance. In Basic mode, OOK is adopted where

bit ’1’ is modulated by an upchirp and bit ’0’ by no-chirp

(or Ground). This mode is simpler to implement by just

controlling the multiplexer in the tag. It is extremely power-

efficient but less robust to multipath. The Advance mode on

other hand is based on frequency jumps in chirps to encode

information, similar to LoRa technology where the chirp signal

is delayed by the multiples of 8µs to generate symbols that

makes it more robust but at the cost of an increase in power

consumption. Thus, to embed multiple symbols in the uplink

channel, the tag must create multiple delayed versions of

the incoming chirp as presented in Fig. 14. For example, in

2-symbol modulation, one upchirp and one delayed version

is required. Similarly, for 4-symbol, one upchirp and three

delayed versions (8µs, 16µs, 24µs) are required.

LiFi rate = 1 kHz

LiFi rate = 50 kHz

LiFi rate = 50 kHz,

= 0.5

Time to charge [s]

V

pp

[mV]

f

1

Power consumption ( W)

B

1

2

400

300

200

100

300

250

200

150

100

50

0

Schmit trigger Shift register Delay line

0

Schmit trigger Shift register Delay line

IoT

Tag

(a)

Power consumption of delay

methods.

(b)

Footprint requirement of de-

lay methods.

Fig. 16: Scheme for mixing recovered chirp signal with RF

carrier to enable uplink backscatter communication on the tag.

Fig. 15: Comparison of power consumption and chip size

required to delay the chirp signal by 8µs to implement multi-

symbol uplink modulation.

Different methods are used in literature to introduce a

delay in the signal: Delay line introduces a fixed delay to

the incoming signal and can be used for phase offset in

transceiver design [33]. It provides high accuracy but it is

only suitable for applications that require a delay in the order

of nanoseconds (up to 500 ns). Multiple delay lines can be

cascaded to achieve higher delays; Shift register with serial

input and parallel output also creates the delayed versions of

the incoming signal [34]. The maximum delay and precision

are governed by the clock frequency that triggers the shift

register, the higher the frequency better the precision but at the

cost of higher power consumption and lower delay per stage

of a shift register; Delay generation with Schmitt trigger are

previously used for VLC based synchronization [35]. Schmitt

trigger offers an energy-efficient and simple solution to delay

the signal but the maximum delay is limited by half of the

time period of the incoming signal.

To implement 8µs delay with 100 kHz incoming square

signal we need 16 delay lines (DS1100-500) in cascade,

shift register (MC74HC164A) with clock (LTC6906), or 2-

channel Schmitt trigger (74LVC2G14) with RC delay circuit.

The experimentally measured power consumption and chip

5

0

-5

-10

-15

-20

850

855

860

865

870

875

880

885

Frequency (MHz)

Fig. 17: Reflection coefficient when RF switch connects with

impedance Z

1

or Z

2

as a function of operating frequency.

are converted to electrical chirps by the solar cell and further

processed by the HPF and comparator. In basic mode, the

recovered chirps are passed as it is and in advanced mode,

the chirps are delayed for multi-symbol uplink modulation

and then fed into the RF switch to toggle the RF antenna

between absorption and reflection state. The antenna mixes

the chirps with RF carrier signal and backscatters the signal

varying from f

c

+f

1

to f

c

+f

2

. In reflection state, the antenna

is connected to ground by setting Z

1

= 0. In the absorption

state, a matched impedance Z

2

is determined using NanoVNA,

a low-cost network analyzer. Impedance matching helps to

maximize the signal amplitude Γ

B

of the backscatter signal

according to the following equation:

size requirement of these methods are presented in Fig. 15.

The result shows that the Schmitt trigger is the most energy-

efficient solution and its footprint fits well in IoT applications.

Therefore, we choose the Schmitt trigger in our design to delay

the incoming chirp signal for LoRa modulation on a battery-

free tag. We may have multiple delayed versions of the chirp

signal, each of them with a delay of 8 µs. The larger the delay,

the larger the power consumption is as there are more circuit

elements to power, but also the larger the modulation order

which involves a data rate increase. Each new delay stage

(i.e., one more symbol in the constellation) means an extra

power consumption of 11 µW. For the m-symbol, where ’m’

is the order of modulation, we need m-1 delay stages. The

data rates for different orders of modulation are presented in

Section V-E.

G.

Backscatter Circuitry

We describe the backscatter circuitry, referring to Fig. 16.

The AC component of the received signal contains LiFi

data and chirps. The chirps are originally transmitted by the

LiFi infrastructure and received by the solar cell-based LiFi

receiver. The optical chirps varying from frequency f

1

to f

2

|Γ | = |Γ | − |Γ | =

Z

1

− Z

A

−

Z

2

− Z

A

, (2)

Z

1

+

Z

A

Z

2

+

Z

A

where Z

A

is the antenna impedance, Γ

1

is the reflection

coefficient when the antenna is connected to impedance Z

1

and

Γ

2

is the reflection coefficient when the antenna is connected

to impedance Z

2

. By setting Z

1

to 0 and matching Z

2

to Z

A

,

we get maximum amplitude of backscatter signal with |Γ

B

|= 1.

The reflection coefficient with Z

1

grounded and Z

2

with both

matched and unmatched impedance is shown in Fig. 17. By

impedance matching, we create a reflection coefficient of 16.5

dB as compared to 5.4 dB in the case of an unmatched state.

The matching improves the performance of the backscatter

system in terms of communication range.

In backscatter systems, antenna selection plays an important

role in communication link performance. Generally, omnidi-

rectional dipole antennas are used to efficiently reflect the

electromagnetic energy and keep the system less sensitive to

orientation. These antennas are robust but sometimes they are

not suitable for IoT applications due to their large size. An

alternative is a PCB trace antenna which is cheaper, smaller in

size and can be built on tag PCB without requiring any external

Optical

signal

(f

1

to

f

2

)

Electrical

signal

(f

1

to

f

2

)

Solar

Cell

Delay

stages

+

HPF

+

MUX

RF

switch

Z

1

Z

2

Z

1

Z

2

(Unmatched)

Z

2

(Matched)

5.4 dB

16.5 dB

Chip size (mm

2

)

Reflection coefficient (dB)

50

40

. . .

30

20

10

0

Dipole

PCB1

PCB2

PCB3

Fig. 19: Transmission of LiFi frame and up-chirp signals (size

of LiFi frame in bytes).

(a)

Dipole and PCB antennas.

(b)

Relative RSSI of backscatter

signal for different antennas.

0.5

Fig. 18: Backscatter performance of different antennas in a

fixed setup.

0.4

0.3

connection, which makes it suitable for IoT applications with

size constraints.

In market several antenna exists for IoT applications, and

we study their performance in backscatter communication.

0.2

0.1

0

500 1000 1500 2000 2500 3000 3500 4000

Background illuminance [lux]

Most antennas are low-cost but vary considerably in their size

and performance. We study four different antenna: 1) dipole

antenna (ANT-868-CW-HW, 855 MHz - 880 MHz), 2) PCB1

(RFPCA7910, 863MHz - 870MHz), 3) PCB2 (Molex212570,

824 - 2170MHZ), and 4) PCB3 (TaoglasPC81, 868MHz -

870MHz) as presented in Fig. 18a. The relative backscattered

signal strength is measured by placing the tag at a distance

of 1 m and the receiver at a distance of 3 m from the RF

carrier generator. Results in Fig. 18b show that dipole antenna

performs better mainly due to its size as compared to PCB

antennas. The performance of the PCB2 antenna is worst as

it is a wide-band, very thin and flexible antenna. Moreover,

we observe that the PCB antennas are highly sensitive to ori-

entation. Finally note that, although the reflection coefficients

between matched and unmatched have 11.1 dB of difference

(Fig. 17), in a communication scenario it is reduced to around

3 dB as seen in Fig. 18b.

V.

E

VALUATION

In this section, we present the experimental evaluation of our

design and comparison with state-of-the-art work. The results

are focused on the following points:

•

Ability of our end-to-end system to detect the chirps in

CSS modulated signal below the noise floor. Our backscatter

receiver shows detection of chirp symbols up to -17 dB

below the noise floor.

•

A basic and advance mode to create CSS modulated data

on tag to improve uplink throughput.

•

A 2x improvement in range of uplink communication in

an outdoor and indoor environment with a 90% decrease

in power consumption as compared to EDISON. Also, the

range of communication with respect to the consumed power

by the tag outperforms works like LoRa backscatter [16] and

Lorea [36].

•

Performance of system in terms of harvesting and LiFi

downlink in an indoor and outdoor environment.

A.

Experimental setup

LiFi transmitter. In our LiFi transmitter, the baseband

signal could be generated using the programmable real-time

unit (PRU) of the Beaglebone used as an embedded processor

Fig. 20: Thresholding circuit evaluation: BER versus back-

ground illuminance provided by an external unmodulated LED

when LiFi transmission rate is 50 kHz and 100 kHz. The LiFi

bulb provided 550 lux illuminance in a dark environment and

at a distance of 1.5 m between transmitter and receiver.

(similarly to OpenVLC). For generating the chirp signal, as

proof of concept, we use the multi-function instrument Analog

Discovery-2 to generate the baseband signal with transmission

structure shown in Fig. 19. The LiFi transmitter provides con-

stant illumination. Downlink transmission implements Manch-

ester coding to guarantee constant light level regardless of

the bit stream. The LiFi transmitter communicates with the

tag sending LiFi frame at the desired data rate with a packet

structure that includes a preamble, start frame delimiter (SFD),

transmitter identifier, receiver identifier, frame length and

payload. After the downlink frame, the transmitter recurrently

sends an upchirp signal varying from a minimum frequency

of 40 kHz to a maximum one of (40 + BW), where BW is the

bandwidth of the chirp signal.

Tag. The received analog signal (LiFi frame and chirp) is

digitized by using 1-bit ADC implemented using TS881 [37]

comparator. The tag uses MSP430FR5969 [38] microcon-

troller unit (MCU) for processing of LiFi received data. The

tag wakes up when a preamble and SFD are detected, similarly

to [13], else the tag stays in sleep mode. The energy harvesting

is performed by solar cell combined with the Texas Instrument

BQ25570 [39] integrated circuit to efficiently extract power

from a solar cell using Programmable Maximum Power Point

Tracking (MPPT). The voltage at the output of the harvester

is regulated to 2.0 V using S-1313 [40] voltage regulator. For

uplink communication, the multiplexer ADG804 [41] selects

between chirp signal and Ground (or delayed chirp) depending

on uplink mode of communication. RF switch ADG902 [42]

is used to vary the impedance of the antenna to backscatter

the 868 MHz carrier signal.

Carrier emitter and RF receiver. The uplink commu-

nication is established by backscattering the 868 MHz tone

transmitted by the carrier wave (CW) generator. Any off-the-

shelf modem or transmitter chipset can be used to generate

EDISON comparator (50 kHz)

EDISON comparator (100 kHz)

Our comparator (50 kHz)

Our comparator (100 kHz)

Dipole

PCB

1

PCB

2

PCB

3

Unmatched

Matched

RSSI of backscatter signal

above noise floor (dB)

BER

t

1

t

2

LiFi frame

Up-chirp

signal

LiFi frame

Up-chirp signal

LiFi frame

Preamble

SFD

TX

ID

RX

ID

Frame

Length

Payload

4

2

1

4

0 to

2

29

10

-

4

Our comparator

2

1

0

0 0.5 1

2

1

0.8

0.6

0.4

0.5

0.4

0.3

0.2

0.1

0

1

2

3

4

0.4

0.3

0.2

0.1

0

5

10

15

1

0.2

Downlink distance [m]

(a)

BER versus distance.

LiFi transmission rate [Hz]

10

4

(b)

BER versus LiFi rate.

0

0 0.5 1

time [s]

10

-4

(a)

Time-domain signals.

0

0 1 2 3

voltage after comparator [V]

(b)

CDF of signal values.

Fig. 22: Threshold circuit evaluation: BER versus distance (at

50 kHz of LiFi transmission rate) and BER versus LiFi trans-

mission rate (at a distance of 1.5 m) with 800 lux background

light.

Fig. 21: Time-domain signals after comparator when 50 kHz

of LiFi transmission rate, 1.5 m of distance and 1800 lux of

background light.

the tone [34], [36]. In our experiments, we use two software-

defined radio USRPs B210 to transmit the RF carrier at

868 MHz and receive the RF backscatter signal, respectively.

We use the open source standard compliant LoRa receiver

for the detection of upchirps [43]. The design is implemented

in Pothosflow software. To exploit the CSS synchronization

method, we modify the receiver to detect one synchronization

word that corresponds to one upchirp. Once synchronized,

later upchirps are considered as data symbols, and we decode

the FFT bins of the received upchirps, meaning different

symbols. Then, note that although we invoke the CSS funda-

mentals of the LoRa standard, we do not transmit standardized

LoRa codewords. We rather exploit the CSS concept for in-

creasing uplink distance. However, this could be implemented

with strict synchronization, at the expense of an increase in

complexity.

B.

LiFi Receiver

For our LiFi receiver, we observe three main findings. First,

it is more robust to background illumination as shown in

Fig. 20 with respect to prior work. The bit error rate (BER) is

plotted against the background illuminance. The plot depicts

the improved performance of our comparator design with

0% BER in the presence of 1000 lux and 3000 lux when

operating at 100 kHz and 50 kHz LiFi transmission frequency,

respectively. Please note that the background illuminance is

generated from an external light source which acts as a noise

to our system.

Second, the output of the comparator is independent of

the input frequency and symmetry is maintained as shown

in Fig. 21a. The cumulative distribution function (CDF) of

voltage after comparator is presented in Fig. 21b. This makes

the sampling of bits at the LiFi receiver less prone to error.

Differently from our thresholding circuit, we notice that the

duty cycle of a signal after the EDISON comparator is not

50%, and sometimes it is even 100%, which introduces a large

number of errors in the decoding process.

The improvement in the range is displayed in Fig. 22a. Our

design can reach up to 3.5 m with 0% BER with a background

of 800 lux. Fig. 22b shows the improvement in terms of data

rate. Our system can achieve a transmission frequency of

0.4

0.3

0.2

0.1

0

0.5 1 1.5 2 2.5 3 3.5 4

Downlink distance [m]

Fig. 23: Evaluation of solar cell configuration for downlink

communication. Communication reliability improves with so-

lar cells connected in series (lower BER, reliable link).

140 kHz corresponding to 280 kbps as compared to 120 kbps

by EDISON design. As the data rate of the LiFi transmitter

increases, PassiveLiFi can better cope with the capacitance

effect from the solar cell, thanks to the higher symmetry

and higher dynamic range of our LiFi transmitter, and higher

robustness to noise of our passive LiFi receiver.

Finally, the BER in LiFi downlink is represented in Fig. 23

when the two configurations of solar cells are established:

series and parallel. Note that, when solar cells are connected

in series the achieved range may be increased due to providing

a larger peak-to-peak voltage in the output of solar cells.

However, at low LiFi rates, this difference is not noticeable

(subject to some minor experimental errors), because the speed

response of solar cell does not clip the peak-to-peak voltage

and then allowing to achieve similar results. As shown before

in Fig. 7, the harvested energy provided by solar cells in

parallel is always better than when they are connected in series.

Then, at lower rates, it is better to configure solar cells in

parallel, while at higher rates, it is convenient that connections

in series and in parallel are switched adaptively to optimize

decoding and harvesting, respectively.

C.

Uplink reception

We evaluate PassiveLiFi in terms of its ability to detect

the received symbols (chirps) below the noise floor. The LiFi

downlink generates the upchirps at the tag which are used by

the backscatter module to vary the antenna impedance. The

distance between the tag and carrier wave generator is fixed

at 1.4 m. Backscatter tags are usually placed either closer to

the carrier wave generator or to the RF receiver. In practical

scenarios, we can alleviate this constraint by distributing a

large number of carrier wave generators, as suggested in [44].

EDISON comparator

Our comparator

EDISON comparator

Our comparator

EDISON comparator

Our comparator

50 kHz - solar cells in series

50 kHz - solar cells in parallel

100 kHz - solar cells in series

100 kHz - solar cells in parallel

voltage after

comparator [V]

voltage after

comparator [V]

cdf

BER

1

0.8

0.6

0.4

0.2

0

-140 -135 -130 -125 -120 -115

Received Signal Power (dBm)

(a)

Effect of SF on SNR limit.

1

0.8

0.6

0.4

0.2

0

-110

Fig. 25: Plan of indoor scenario. The positions of the tag and

CW generator are highlighted, and the positions of the RF

receiver are marked with P1-7.

1

0.8

0.6

0.4

0.2

-140 -135 -130 -125 -120 -115 -110 -105

Received Signal Power (dBm)

(b)

Effect of BW on SNR limit.

Fig. 24: Evaluation of backscatter receiver to detect chirps

below the noise floor.

0

1

0.8

0.6

5

10 15 20 25 30 35 40 45

Distance (m)

(a)

Indoor scenario.

50 55

We present multiple application scenarios of PassiveLiFi in

Section VI.

Note that although we conduct experiments with a relatively

0.4

0.2

0

100 150 200 250

Distance (m)

300 350 400

short range in LiFi link between the LED and the tag, we may

require a long-range for RF backscatter in order to transmit

sensed data to the edge device. This enables to have unique

(or a few) edge devices for multiple rooms (indoors) or a

large coverage area (outdoors). The transmission power of the

carrier generator is varied to evaluate a system for different

received power. The results are shown in Fig. 24 for the

OOK modulation scheme used in EDISON (demodulates only

above the noise floor), and CSS used in our design. Fig. 24a

and Fig. 24b show the effect of spreading factor (SF) and

bandwidth (BW) on SNR limit, respectively. With the increase

of SF, the SNR limit decreases and with an increase in BW,

the SNR limit increases which is consistent with the LoRa

standard. In the best configuration (SF 12 and BW 60 kHz),

our receiver decodes 17 dB below the noise floor.

We are limited by the noise floor of the USRP. However,

we can significantly improve the communication performance

through the usage of commodity transceivers for the reception

which gives up to 25-30 dB lower noise floor when compared

to the SDR [16]. The selection of chirp BW is important here,

as on the lower side we are limited by the interference from

the carrier generator tone and on the higher side limited by

the BW of the solar cell. For generating chirps we use 40 kHz

as a lower limit and the upper limit is selected based on the

value of chirp BW i.e. 100 kHz for chirp with 60 kHz BW.

In Fig. 24b, with 90 kHz BW, it can be seen that the highest

BER is 0.18 due to the limitation of the solar cell’s BW.

In the above experiments, the basic mode is used to modu-

late data which offers a low data rate. In the next section, we

evaluate the performance of advance mode of communication

which provides a robust link and higher throughput.

(b)

Outdoor scenario.

Fig. 26: Range of uplink communication in indoor and outdoor

scenarios. The tag and CW generator are located at a distance

of 1.2 for indoor and 1 m for outdoor, and the RF receiver is

moved away.

D.

Uplink range and energy consumption

We evaluate the uplink range in both indoor and outdoor

scenarios. The experiment is performed with OOK as in

EDISON, basic mode with 2-symbol (bit ’1’ as upchirp and

bit ’0’ as no-chirp) and advance mode with 4-symbol (one

upchirp and three delayed versions). In an indoor environment,

we perform the test inside a building on the ground floor by

keeping the tag and CW generator in a room at a distance

of 1.4 m. We place the RF receiver at position P1 to P16

at distance of 5.5, 13, 18, 22, 24, 27, 34, 36, 38, 39, 40,

43.4, 44.2, 45.5, 46 and 48 m, respectively, from tag as shown

in Fig. 25. In Fig. 26a we observe a significant increase in

range by PassiveLiFi as compared to EDISON in the indoor

scenario. With our design, we get a range as large as 47 m

with a normalized bit loss rate less than 0.22, which is a 2x

improvement over EDISON. Besides, matched circuit means

a gain of around 3 m with respect to an unmatched circuit,

and also note that even transmitting a 4-symbol modulation

we achieve an indoor distance of around 37 m.

In an outdoor scenario, we perform the experiment in an

open space. We use the same configuration for tag and CW

generator as in indoor and we place the backscatter receiver

at different distances in open space. Note that outdoor light

sources such as streetlights may be located at a larger distance,

but their transmission power is also larger than the power of

SF=12, BW=70kHz

SF=11, BW=70kHz

SF=10, BW=70kHz

Noise floor

SF=12, BW=60kHz

SF=12, BW=70kHz

SF=12, BW=80kHz

SF=12, BW=90kHz

Noise floor

P1

Tag







P16 P14

P12

CW

generator

 



   



P15

P13

P11

P9 P8

P7

P6 P5

P4

P10

P3

P2

 



 



BER

OOK - Unmatched

OOK - Matched

4-symbol (Adv.), Unmatched

4-symbol (Adv.), Matched

2-symbol (Basic), Unmatched

2-symbol (Basic), Matched

P4

P6

P9

P11

P14

P16

P5

P10

P8

P13

P15

P1

P2

P3

P7

P12

OOK - Unmatched

OOK - Matched

4-symbol (Adv.), Unmatched

4-symbol (Adv.), Matched

2-symbol (Basic), Unmatched

2-symbol (Basic), Matched

TABLE II: Computation of maximum achieved range versus

uplink power consumption.

1400

1200

1000

800

600

400

200

0

500 1000 1500 2000 2500

Illuminance (lux)

1

0.5

0

Fig. 27: Range of tag

when

it is moved away while CW

Fig. 28: Maximum power harvested by 30 cm

2

solar cell (5x

SLMD121H04L in parallel).

implementation in all these works except LoRa Backscatter

which uses an application-specific integrated circuit (ASIC)

design. As seen in ratio distance-consumption results, our

PassiveLiFi tag shows an uplink efficiency much larger (x2)

than previous works, which makes it much more sustainable

while achieving longer ranges.

The last indoor experiment demonstrates the possibility of

locating the tag in a different room as a CW generator and RF

receiver. Fig. 27 shows that, when placing the CW generator

and RF receiver in the same room at a distance of 2.3 m, the

signal can be decoded when the tag is located in a different

room at a distance of around 7 m. Unlike prior works [13], we

enable the possibility of separating tag from either RF receiver

or CW generator, increasing the flexibility of the setup indoors.

generator and RF receiver are placed in a different room

at a distance of around 7 m from the tag. Basic 2-symbol

modulation order is used.

our LiFi transmitter, which enables these outdoor experiments.

Thus, these outdoor results are still very valuable to evaluate.

The results are presented in Fig. 26b. We obtain around

2x improvement over EDISON, with a range up to 350 m

for our design with basic 2-symbol modulation. Note that,

when doubling the data rate (4-symbol modulation) in advance

mode, the achievable distance is reduced up to around 280 m.

The energy consumption of the backscatter module is sig-

nificantly reduced by offloading the oscillators which are the

most power-hungry components in the backscatter module.

The energy is reduced from 70 µW (as in EDISON) to 3.8 µW

in basic mode. Only comparator, multiplexer and RF switch

are the active elements in the backscatter module with a typical

power consumption of < 1 µW. Similarly, for advance mode

with SF 12, the power consumption is 14.8 µW, 46.2 µW and

112 µW for 2-symbol, 4-symbol and 8-symbol, respectively.

Table II presents a comparison in the ratio between achieved

distance over uplink energy consumption. Note that the re-

ported effective isotropic radiated power (EIRP) is different

in every state-of-the-art work, which is unfair. To make a

fair comparison, we consider Friis’ path model to get the

corresponding sensitivity of receivers. Knowing that and set-

ting up the same EIRP as in our scenario (20 dBm), we are

able to compute the maximum achieved distance in uplink

under the same configuration. Note that the maximum distance

considered for LoRea is the one that provides a BER=10

−

2

and 2.9 kbps, whereas the maximum distance considered for

LoRa Backscatter is the one that obtains 200 bps. The power

consumption is reported for commercial-off-the-shelf (COTS)

E.

Self-sustainability of tag

The tag is powered by a solar cell and the total power

harvested depends on the illumination and the operating point.

The maximum power harvested by the solar cell as a func-

tion of illumination by operating at optimum power point

is shown in Fig. 28. We study the self-sustainability of tag

at different uplink and downlink bit rates and present the

results in Table III and Table IV. We evaluate 2-symbol, 4-

symbol and 8-symbol uplink modulation orders for different

spreading factors from 8 to 12. The data rate increases by the

order of modulation at the expense of an increase in power

consumption. The power consumption of frontend includes

a comparator, delay stages, multiplexer and RF switch. The

MCU generates the control signals for the multiplexer and the

transmitted data in the uplink. The basic 2-symbol mode con-

sumes minimum power as it does not require any delay stages

or precise control signals for multiplexer. In the downlink, the

MCU consumes power in processing the received LiFi packet.

The low-power modes in MCU help to reduce the power

consumption to micro-watts. The results are presented for

continuous downlink reception (100% ON), 25 % ON time and

10 % ON time. The user can select the uplink and downlink

transmission rates based on the illumination and power budget

available. For example, basic mode in uplink with SF 12 and

0.5 kbps in downlink with 100% ON time consumes total

of 3.8 + 183.4 = 187.2µW. From Fig. 28, around 500 lux

illumination is required to achieve self-sustainability of tag,

which is a reasonable lighting value in indoor scenarios.

VI.

A

PPLICATION

S

CENARIOS

There has been interest in IoT and mobile systems that

leverage light and RF for sensing and communication. For

BER

Power harvested ( W)

Lorea

[36]

EIRP

Max. distance

Operation freq.

Uplink consumption

Implementation

28

dBm

3.4 km

868 MHz

70 µW

COTS

Max.

dist.

when 20 dBm:

950 m

Ratio

distance-consumption:

13.6 m/µW

Lora

Backscatter

[16]

EIRP

Max. distance

Operation freq.

Uplink consumption

Implementation

36

dBm

2.8 km

915 MHz

9.25 µW

ASIC

Max.

dist.

when 20 dBm:

435 m

Ratio

distance-consumption:

47 m/µW

EDISON

[13]

EIRP

20

dBm

Ratio

Max.

distance

160

m

distance

-

consumption:

Operation

freq.

868

MHz

2.29

m/

µ

W

Uplink

consumption

70

µ

W

Implementation

COTS

PassiveLiFi

(Proposal)

EIRP

20

dBm

Ratio

Max.

distance

350

m

distance

-

consumption:

Operation

freq.

868

MHz

92.1

m/

µ

W

(basic)

Uplink

consumption

(basic)

3.8

µ

W

23.65

m/

µ

W

(advanced)

Uplink

consumption

(advanced)

14.8

µ

W

Implementation

COTS

TABLE III: Experimental power consumption of tag in uplink communication.

Architecture

Basic

Advance

Modulation

order

2

-

symbol

2

-

symbol

4

-

symbol

8

-

symbol

Spreading

factor

8

10

12

8

10

12

8

10

12

8

10

12

Datarate

(b/s)

488

122

30

488

122

30

976

244

61

1464

366

91

Power consumption (µW)

Frontend

4.09

3.95

3.80

25.2

20.6

14.8

64.3

57.6

46.2

139

127

112

MCU

0

70.1

74.5

78.3

TABLE IV: Experimental power consumption of tag in downlink communication.

Datarate

(kb/s)

0.1

0.5

1

2

5

10

Power consumption (µW)

100%

ON

120

183.4

282

462

908.2

1786.4

25%

ON

62

90.3

126.4

165.9

371.7

642.6

10%

ON

40

59

75.96

111.3

219.1

398.1

example, LiFi systems are currently being deployed in large

numbers to support high-speed downlink communication ap-

plications. These systems predominately use RF to support

uplink transmissions, commonly through energy-expensive

WiFi radios. Our system builds on these efforts and develops

mechanisms to support energy-efficient uplink for battery-free

devices through RF backscatter. LiFi for battery-free devices

is largely unexplored, and our system targets this vital area

and paves the way to enable numerous scenarios. We discuss

some of these application scenarios.

Outdoor deployments. The deployment of sensors in out-

door settings enables numerous applications. For example,

they may be deployed at a large scale to enable the concept

of smart cities. These applications require a large deployment

of sensors and these sensors transmit their information to

a reasonably large range. Our system benefits from these

scenarios, as most outdoor settings provide access to lighting

infrastructure that could be re-purposed for delegation of the

oscillations or to support downlink communication. Further,

our system enables us to lower the complexity and power

consumption of the tags, which is necessary for large-scale

deployments in outdoor settings.

Smart homes. We are automating homes and deploying

IoT devices in large numbers. Today, almost all of these

IoT devices are energy-expensive and are reliant on batter-

ies. Backscatter may help overcome this reliance. However,

backscatter in devices deployed in homes is challenging due

to the lack of downlink communication and limited range.

Our system is well suited for indoor environments as artificial

lighting is omnipresent indoors, providing a downlink channel

to the backscatter tags. Further, the large communication range

due to CSS can enable flexibility in the receiver-equipped

edge device’s placement. For example, PassiveLiFi equipped

with a temperature sensor can be deployed in each room

of a smart home, which takes command from light bulbs

in a room (downlink) to activate and record temperature,

and use backscatter to communicate the reading to a central

edge device or control unit to maintain the air conditioning.

One main limitation of LiFi is that the best communication

range is achieved on a Line-of-Sight (LoS) link. However,

the trend is toward deploying lighting infrastructure composed

of dense light fixtures [45], where every point in the room

is illuminated by more than one fixture in order to comply

with lighting standards (illuminance homogeneity, average

illuminance, etc.). This will ensure receiving a signal from

more than one light fixture, which reduces enormously the

blockage probability.

Farming. Growing plants in an indoor environment such

as in greenhouses and vertical farms are attracting significant

interest. These environments require the deployment of sensors

to track soil moisture, temperature, etc. For example, in

vertical farming setup, LEDs are installed under the height

of 1 m or less [46] and sensors are placed with plants to

measure environmental parameters. The sensor readings are

communicated to a central unit to actuate different tasks such

as watering the plants. PassiveLiFi provides the best solution

to achieve these operations in vertical farming [20]. Further,

artificial lighting is omnipresent to help plants grow. Our

system could benefit such applications by taking advantage of

already present lighting and enabling the low-cost deployment

of sensors with lesser installation efforts.

VII.

R

ELATED WORK

We discuss works that are most related to our system.

Backscatter Communication Recent systems show the

ability to synthesise transmissions compatible with WiFi [23],

ZigBee [34], BLE [47], and LoRa [16], other systems have

achieved an enormous communication range [16], [36]. It

enables new scenarios and possibilities. However, backscatter

systems have a poor ability to receive transmissions. These

tags are limited due to the passive envelope detectors employed

to perform reception. They suffer from poor sensitivity, sus-

ceptibility to cross-technology interference, and their inability

to support complex modulation schemes. In this regard, we

take a step to overcome these limitations by building on

recent systems that advocate LiFi as an alternative to RF

to receive downlink information [13], [48]. When compared

to these systems, we significantly improve design, exploring

the trade-off between solar cell size, energy harvesting and

communication, improve the robustness of the LiFi receiver

through various energy-efficient filters and the RF backscat-

ter ability by leveraging the chirp spread spectrum scheme

enabled through the concept of LiFi as an oscillator.

Offloading Computing, Processing and Oscillations The

past decade has seen a dramatic improvement in the energy

efficiency of sensors, with microphones [3] and cameras [49]

consuming tens of microwatts of power. It has made computa-

tion and communication significantly more energy expensive

than sensing. Backscatter reduces this energy asymmetry, as

it brings the energy cost for performing transmissions to a

level similar to that for performing sensing. Consequently,

computational elements such as FPGAs and MCUs are a

crucial bottleneck. Recent systems have advocated eliminating

computational elements. They couple the sensor directly to a

backscatter transmitter and delegate all the necessary sensor

readings to a powerful edge device. Building on this architec-

ture: [3] designs a battery-free cellphone that transmits audio

signals. Further, recent systems even demonstrate battery-free

video streaming cameras [49].

Recent systems have explored delegating oscillators to ex-

ternally powered infrastructure. [21] generates a twin carrier

tone by re-purposing a WiFi device. This enables them to

provide energy expensive oscillations to a tag. We build on

these insights and delegate the energy expensive oscillations

to the infrastructure. Our work differs in using LiFi signals

to deliver oscillations. Our work is most closely related to

EDISON [13], which has shown in a dedicated experiment the

possibility to deliver clock signals through light. As shown

in our evaluation, we significantly improve their design by

enhancing the LiFi transmitter and receiver and demonstrating

the ability to receive chirps signals, thus broadly improving the

overall performance, also being able of embedding multiple

symbols creating higher modulation orders with low-cost

electrical components.

There have also been efforts to recover clock signals from

optical communication, leading to energy-efficient integrated

circuits (IC). In particular, some of these systems demon-

strate recovery of clock signals from the Manchester encoded

data using low-power digital circuits [50]. Our system is

complementary to these systems, and goes much beyond the

capabilities demonstrated by prior designs. We demonstrate the

recovery of complex baseband signals that employ a complex

chirp spread spectrum (CSS) modulation scheme, which we

then used to modulate an RF carrier. Nevertheless, we can also

employ techniques presented in prior works to improve our

system’s energy efficiency, helping us realise low-power ICs.

Solar cell for LiFi. Solar cells have seen interest beyond

their traditional role of harvesting energy from light. There has

been an effort to repurpose them for LiFi communication. It

has enabled a significant reduction in the energy consumption

of the LiFi frontend. Some works have designed application-

specific integrated circuits [50] [51] which are difficult to repli-

cate or use in a different context, such as low-power backscat-

ter communication. Other systems have only used solar cells

for harvesting or communication, and they lack the necessary

design to optimise for both energy harvesting and commu-

nication [52] [32]. As opposed to these systems, we design

a low-power mechanism that can harvest and communicate

using the LiFi infrastructure and enable various applications.

LiFi Communication for IoT devices Active LiFi aims to

create a networked system that uses modulated light bulbs

and active receivers. More and more often, uplink com-

munication relies on RF [53]. However, these systems use

energy-expensive components, which pushes them beyond

the means of IoT devices. Systems such as OpenVLC [54]

have tackled the issues of supporting LiFi on battery-powered

IoT devices and have even demonstrated streaming video

using this platform [55]. However, these systems are still

energy expensive for an emerging class of battery-free IoT

devices. Recent systems have tackled the challenge of LiFi on

battery-free devices. RetroVLC and PassiveVLC demonstrate

a battery-free tag that can receive downlink transmissions

using LiFi and uplink through visible light backscatter [8],

[14]. However, these systems suffer from the challenge of the

directionality of visible light backscatter links. Further, their

downlink LiFi reception suffered from challenges of ambient

noise. We design an efficient LiFi receiver. Further, we adopt

the EDISON approach of using RF backscatter to support

uplink transmissions and significantly improve their design by

using chirps to improve the range, in addition to a practical

technique to allow higher modulation orders and increase data

rate in the uplink. We expect that RF (backscatter) will likely

become the predominant technology for uplink communication

and passive LiFi.

VIII.

C

ONCLUSION

We have presented PassiveLiFi. It explores the interactions

between LiFi downlink and RF backscatter uplink to achieve

very low-power and long-range uplink communication. Our

design introduces visible light chirps that are sent by the LiFi

transmitter, which are received and mixed by the IoT tag

with the input RF carrier to transmit uplink RF backscatter

signals. We embed symbols in the tag by multiple delay

stages that create delayed upchirps. This allows transmitting

at higher modulation orders that increases the data rate. We

have extensively evaluated our system and shown promising

results in reducing power consumed by the tag (3.8 µW) while

communicating at a distance of up to 350 m using an RF carrier

emitting at 17 dBm.

A

CKNOWLEDGMENTS

This work has been funded at IMDEA Networks in part

by the EU’s Horizon 2020 ITN program under the MSCA

grant agreement ENLIGHTEM (814215), in part by the project

RISC-6G (TSI-063000-2021-59), granted by the Ministry of

Economic Affairs and Digital Transformation and the Euro-

pean Union-NextGenerationEU through the UNICO-5G R&D

program of the Spanish Recovery, Transformation and Re-

silience Plan, and partially funded by Juan de la Cierva

grant (FJC2019-039541-I/AEI/10.13039/501100011033). Fur-

thermore, it has received funds from Vinnova (Sweden Innova-

tion Agency) under the grant (2018-04305) at Uppsala Univer-

sity and a startup grant (A-800027700-00) from the National

University of Singapore, both awarded to A. Varshney.

R

EFERENCES

[1]

M. M. Tentzeris, A. Georgiadis, and L. Roselli, “Energy harvesting and

scavenging,” Proc. IEEE, vol. 102, no. 11, 2014.

[2]

M. Anderson, “Potential Hazards at Both Ends of the Lithium-Ion Life

Cycle,” IEEE Spectrum, 2013.

[3]

V. Talla, B. Kellogg, S. Gollakota, and J. R. Smith, “Battery-free

cellphone,” Proc. of the ACM IMWUT, vol. 1, no. 2, pp. 1–20, 2017.

[4]

A. Varshney, A. Soleiman, L. Mottola, and T. Voigt, “Battery-free visible

light sensing,” in Proc. of the 4th ACM Workshop on Visible Light

Communication Systems, ser. VLCS ’17. NY, USA: ACM, 2017.

[5]

J. de Winkel, V. Kortbeek, J. Hester, and P. Pawełczak, “Battery-free

game boy,” Proc. ACM IMWUT, vol. 4, no. 3, Sep. 2020.

[6]

Y. Li et al., “Self-powered gesture recognition with ambient light,” in

Proc. of the 31st Annual ACM Symposium on User Interface Software

and Technology, 2018.

[7]

A. Saffari, M. Hessar, S. Naderiparizi, and J. R. Smith, “Battery-free

wireless video streaming camera system,” in Proc. IEEE RFID, 2019,

pp. 1–8.

[8]

X. Xu et al., “PassiveVLC: Enabling Practical Visible Light Backscatter

Communication for Battery-free IoT Applications,” in Proc. of ACM

MobiCom, 2017.

[9]

M. Gorlatova et al., “Movers and shakers: Kinetic energy harvesting

for the internet of things,” IEEE J. Sel. Areas Commun., vol. 33, no. 8,

2015.

[10]

Q. Wan, Y.-K. Teh, Y. Gao, and P. K. T. Mok, “Analysis and design

of a thermoelectric energy harvesting system with reconfigurable array

of thermoelectric generators for iot applications,” IEEE Transactions on

Circuits and Systems I: Reg. Paper, vol. 64, no. 9, pp. 2346–2358, 2017.

[11]

M. Shakeel et al., “A low-cost printed organic thermoelectric generator

for low-temperature energy harvesting,” Renewable Energy, vol. 167,

pp. 853–860, 2021.

[12]

Y. Zhang et al., “Sozu: Self-powered radio tags for building-scale

activity sensing,” ser. UIST ’19. NY, USA: ACM, 2019, p. 973–985.

[13]

A. Galisteo, A. Varshney, and D. Giustiniano, “Two to tango: Hybrid

light and backscatter networks for next billion devices,” in Proc. of ACM

MobiSys, 2020.

[14]

J. Li et al., “Retro-VLC: enabling battery-free duplex visible light

communication for mobile and IoT applications,” in Proc. of the 16th

International Workshop on Mobile Computing Systems and Applications.

ACM, 2015, pp. 21–26.

[15]

M. S. Mir, B. G. Guzman, A. Varshney, and D. Giustiniano, “Pas-

siveLiFi: Rethinking LiFi for Low-Power and Long Range RF Backscat-

ter,” in Proc. of ACM MobiCom, 2021.

[16]

V. Talla et al., “LoRa Backscatter: Enabling The Vision of Ubiquitous

Connectivity,” Proc. ACM IMWUT, vol. 1, no. 3, pp. 105:1–105:24, Sep.

2017.

[17]

W. Zhao et al., “Investigation on image signal receiving performance of

photodiodes and solar panel detectors in an underground facility visible

light communication system,” Opt. Express, vol. 29, no. 2, Jan 2021.

[18]

S. Biswas and H. Kim, “Solar cells for indoor applications: Progress

and development,” Polymers, vol. 12, no. 6, 2020.

[19]

M. S. Mir et al., “RGB LED bulbs for communication, harvesting and

sensing,” in Proc. of IEEE Percom, 2022.

[20]

B. Genoves Guzman et al., “Towards sustainable greenhouses using

battery-free lifi-enabled internet of things,” IEEE Commun. Mag., 2023.

[21]

M. Rostami et al., “Redefining passive in backscattering with commodity

devices,” in Proc. of the 26th ACM MobiCom, 2020, pp. 1–13.

[22]

R. Nandakumar, V. Iyer, and S. Gollakota, “3D localization for sub-

centimeter sized devices,” in Proc. ACM SenSys, 2018, pp. 108–119.

[23]

B. Kellogg, V. Talla, S. Gollakota, and J. R. Smith, “Passive Wi-Fi:

Bringing Low Power to Wi-Fi Transmissions,” in Proc. NSDI. Berkeley,

CA, USA: USENIX, 2016.

[24]

M. W. M. Cunico and J. Carvalho, “Determination of degree of

polymerization by low cost photometric system: application of uv led

and photodiode,” in 21st Brazilian Congress of Mechanical Engineering,

ABCM, Natal, 2011.

[25]

A. Galisteo, D. Juara, and D. Giustiniano, “Research in visible light

communication systems with OpenVLC1.3,” in Proc. IEEE WF-IoT,

2019.

[26]

CEN-Light, “Lighting of work places—part 1: Indoor work places,”

European Committee for Standardization, Brussels, Belgium, 2002.

[27]

“LoRa Alliance,” https://lora-alliance.org/.

[28]

X. Lu et al., “Wireless Networks With RF Energy Harvesting: A

Contemporary Survey,” IEEE Commun. Surveys Tuts., vol. 17, no. 2,

pp. 757–789, 2015.

[29]

Analog Devices. ADG72X switches. [Online]. Avail-

able: https://www.analog.com/media/en/technical-documentation/data-

sheets/ADG721 722 723.pdf

[30]

R. Marler and J. Arora, “Survey of multi-objective optimization methods

for engineering,” Structural and Multidisciplinary Optimization, vol. 26,

pp. 369–395, 2004.

[31]

I. E. S. of North America, Lighting handbook: Reference & application.

Illuminating Engineering Society of North America, 2000.

[32]

Z. Wang, D. Tsonev, S. Videv, and H. Haas, “On the design of a solar-

panel receiver for optical wireless communications with simultaneous

energy harvesting,” IEEE J. Sel. Areas Commun., vol. 33, no. 8, 2015.

[33]

P. Zhang, D. Bharadia, K. Joshi, and S. Katti, “HitchHike: Practical

Backscatter Using Commodity WiFi,” in Proc. of ACM Sensys, 2016.

[34]

V. Iyer et al., “Inter-technology backscatter: Towards internet connec-

tivity for implanted devices,” in Proc. of the 2016 ACM SIGCOMM

Conference, 2016.

[35]

A. Torrisi, K. S. Yıldırım, and D. Brunelli, “Autonomous energy status

sharing and synchronization for batteryless sensor networks,” in Proc.

of ACM SenSys, 2021.

[36]

A. Varshney et al., “LoRea: A Backscatter Architecture That Achieves

a Long Communication Range,” in Proc. of ACM SenSys, 2017.

[37]

STMicroelectronics. TS881. [Online]. Available: https://www.st.com/

resource/en/datasheet/ts881.pdf

[38]

Texas Instruments. MSP430FR5949. [Online]. Available: http://www.ti.

com/lit/ds/symlink/msp430fr5949.pdf

[39]

Texas Instruments. BQ25570. [Online]. Available: http://www.ti.com/

lit/ds/symlink/bq25570.pdf

[40]

Ablic. S-1313. [Online]. Available: https://www.ablic.com/en/doc/

datasheet/voltage regulator/S1313 E.pdf

[41]

Analog Devices. ADG804. [Online]. Available: https://www.analog.

com/media/en/technical-documentation/data-sheets/adg804.pdf

[42]

Analog Devices. ADG902. [Online]. Available: https://www.analog.

com/media/en/technical-documentation/data-sheets/ADG901 902.pdf

[43]

LoRa modem with LimeSDR, “https://github.com/myriadrf/lora-sdr.”

[44]

M. Katanbaf, A. Saffari, and J. R. Smith, “Multiscatter: Multistatic

backscatter networking for battery-free sensors,” in Proc. ACM SenSys,

2021.

[45]

J. Beysens et al., “DenseVLC: A Cell-Free Massive MIMO System with

Distributed LEDs,” in Proc. ACM CoNEXT, 2018, p. 320–332.

[46]

J. D. Borrero, “Expanding the level of technological readiness for a

low-cost vertical hydroponic system,” Inventions, vol. 6, no. 4, 2021.

[47]

J. F. Ensworth and M. S. Reynolds, “Every smart phone is a backscatter

reader: Modulated backscatter compatibility with bluetooth 4.0 low

energy (BLE) devices,” in Proc. IEEE RFID, 2015, pp. 78–85.

[48]

D. Giustiniano, A. Varshney, and T. Voigt, “Connecting Battery-Free

IoT Tags Using LED Bulbs,” in Proc. ACM HotNets, ser. HotNets ’18.

New York, NY, USA: ACM, 2018, p. 99–105.

[49]

S. Naderiparizi et al., “Towards battery-free HD video streaming,” in

Proc. 15th USENIX NSDI, 2018.

[50]

X. Wu et al., “A 0.04MM316NW Wireless and Batteryless Sensor Sys-

tem with Integrated Cortex-M0+ Processor and Optical Communication

for Cellular Temperature Measurement,” in Proc. 2018 IEEE Symposium

on VLSI Circuits, 2018, pp. 191–192.

[51]

J. Lim et al., “26.9 A 0.19×0.17mm2 Wireless Neural Recording IC for

Motor Prediction with Near-Infrared-Based Power and Data Telemetry,”

in Proc. IEEE ISSCC, 2020, pp. 416–418.

[52]

S. Ma et al., “Simultaneous lightwave information and power transfer in

visible light communication systems,” IEEE Trans. Wireless Commun.,

vol. 18, no. 12, pp. 5818–5830, 2019.

[53]

H. Haas et al., “Introduction to indoor networking concepts and chal-

lenges in LiFi,” IEEE/OSA J. Opt. Commun. Netw., vol. 12, no. 2, 2020.

[54]

A. Galisteo et al., “Openvlc1.2 for increased data rate with embedded

systems,” in Proc. of the 4th ACM Workshop on Visible Light Commu-

nication Systems. ACM, 2017, pp. 33–33.

[55]

A. Galisteo, D. Juara, H. Cordobes, and D. Giustiniano, “Video trans-

mission using low-cost visible light communication,” in Proc. of ACM

Mobihoc, 2019.

Muhammad Sarmad Mir is a Ph.D. researcher at UC3M, Spain. He was a

recipient of MSCA ITN scholarship, and his research interests include low-

power communication and battery-less devices.

Borja Genoves Guzman was a Postdoctoral researcher at IMDEA Networks

(currently at University of Virginia, USA). He received a Ph.D. degree from

the UC3M in 2019 (Extraordinary PhD award). His current research focuses

on LiFi, IoT and next generation wireless networks.

Ambuj Varshney is a faculty member at the National University of Singapore.

He received postdoctoral training at the University of California, Berkeley, and

earned his doctorate from Uppsala University. His research centers on wireless

embedded systems by designing energy-efficient communication mechanisms.

Domenico Giustiniano is Research Associate Professor (tenured) at IMDEA

Networks Institute and leader of the Pervasive Wireless System Group. His

current research interests cover battery-free IoT systems, large-scale spectrum-

based analytics, and 5G+ localization systems.