Poster: Visible Light Communication in the Darktianzhao/papers/vlc_dark.pdf · We present a new visible light communication (VLC) primitive: VLC in the dark, where the communication

Poster: Visible Light Communication in the Dark

Zhao Tian, Andrew T. Campbell, and Xia ZhouDepartment of Computer Science, Dartmouth College

tianzhao, campbell, [email protected]

ABSTRACT

We present a new visible light communication (VLC) primitive:VLC in the dark, where the communication sustains even whenthe LED light appears dark. We achieve the goal by reducing theduty cycle of the light source to an extremely low level (0.0019%)such that the illuminance is imperceptible to human eyes and yetcan still be sensed by photodiodes to decode data. VLC in thedark consumes much lower energy than the conventional VLC linksand thus is appealing for enabling VLC uplink or communicationamong Internet-of-Things devices with tight energy budget. Webuild a preliminary prototype to demonstrate its feasibility. Ourcurrent single link achieves 190.7 bps data rate with 46.8 µW powerconsumption at the LED.

Categories and Subject Descriptors

C.2.1 [Network Architecture and Design]: Wireless communica-tion

Keywords

Visible light communication; Energy efficiency; Internet of Things

1. INTRODUCTIONOperating on the visible light spectrum, Visible Light Commu-

nication (VLC) holds a great potential to mitigate the spectrumcrunch problem. All existing designs of VLC links commonly re-quire the lights to be always on — the VLC link ceases when thelights are off. This constraint imposes a fundamental limit on theapplicable scenarios of VLC. Furthermore, keeping LED lights onentails a high energy consumption, making VLC less appealing forsmart devices with tight energy budget to transmit data.

In this poster, we design and implement VLC in the dark, a newVLC primitive that can operate even if the LED lights appear dark.This new primitive enables VLC links to maintain always-on com-munication, greatly broadening its applicable scenarios and low-ering the energy consumption of LED lights. In particular, weachieve VLC in the dark by reducing the duty cycle of the light

Permission to make digital or hard copies of part or all of this work for personal or

classroom use is granted without fee provided that copies are not made or distributed

for profit or commercial advantage and that copies bear this notice and the full cita-

tion on the first page. Copyrights for third-party components of this work must be

honored. For all other uses, contact the Owner/Author(s). Copyright is held by the

owner/author(s).

MobiCom’15, September 7–11, 2015, Paris, France.

ACM 978-1-4503-3619-2/15/09.

http://dx.doi.org/10.1145/2789168.2795181.

PD

LED

(a) System Overview

0

2

4

6

8

10

12

2 4 6 8 10 12 14 16 18

MO

SF

ET

Gate

Voltage (

V)

Time (ms)

OFF

ON

(b) Pulse Width Modulation

Figure 1: VLC in the dark. (a) A picture of our system. In the picture,

the duty cycle of the LED is so small (0.0019%) that the light appears

off. (b) Duty cycle d = tON/(tON + tOFF). With the peak forward

current fixed, the illuminance of the LED light is proportional to its

duty cycle. The switching frequency f = 1/(tON + tOFF).

source to an extremely low level such that it becomes impercep-tible (Figure 1) to human eyes. VLC in the dark is inspired by aprior study [1] that discusses the standard for the lights to appear“off” and the required receive power for a target data rate usingsimulations. Our work differs in that we go beyond the analyticalresults and simulations, tackle the practical challenges to actuallyrealize this concept, and explore its potential applications. We builda prototype of VLC in the dark using off-the-shelf LEDs, photodi-odes, and circuit components. The implementation of the prototypeinvolves deliberate trade-off of system parameters (Section 2) anddesign of the circuits (Section 3).

VLC in the dark is imperceptible to human eyes and is energy-efficient. In our testbed, the minimum illuminance shift incurredby the LED is 0.09 lx and the LED consumes only 46.8 µW. Thisnew VLC primitive can enable many novel VLC applications. Itcan be used to build the VLC uplinks for mobile devices. A recentproposal implements VLC uplink using backscatter [5], which re-flects the LED light coming from the ceiling. However, this designstill requires the lights to be on. VLC in the dark can also enablecommunication for Internet of things (IoT) devices, through eitheralways-on ceiling-to-device or device-to-device VLC links. VLCin the dark consumes ultra-low energy and is appealing for IoT de-vices with energy constraints.

2. VLC IN THE DARK: THE DESIGNWe make the LED light appear off by decreasing its flashing duty

cycle. We do not completely switch off the light; instead, we reduce

FPGA MAX4427

USRP

G

S

D

Vs

Vdd

Vdd

Vdd

-Vdd

4.7 μF

0.1 μF

TX RX

(a) Circuit design (b) Testbed

Figure 2: Implementation of the transmitter and receiver. The LED (Cree CXA2520) is driven by a Xilinx Artix-7 FPGA through a MOSFET

(STF5N52U) and a gate driver (MAX4427). The photocurrent generated by the photodiode (Honeywell SD5421) is transformed to voltage signal by

a resistor, and the signal is sampled by USRP with an LFRX daughterboard. The voltage follower (BUF634) between the resistor and the USRP is

for impedance matching.

its illuminance to a negligible level but still maintain its ability ofcommunication.

For LEDs, the luminous flux, which determines the illuminance,is nearly proportional to the forward current. In practice, the LEDsare not always switched on but periodically on and off. The illu-minance is determined by both the peak intensity and duty cycle(Figure 1(b)). We can decrease the LED’s illuminance by reduc-ing the peak current flowing through the LED. But this method willcurtail the communication distance according to the inverse-squarelaw. So we keep the peak intensity high but reduce the duty cycle:

d =tON

tON + tOFF

=tON

tperiod

.

To reduce the duty cycle, we should narrow tON and widen tperiod.The configuration of these two parameters is subject to the capacityof the electrical components and human’s perception (e.g. flicker-ing effect). The minimal tON is determined by the turn-on time ofthe LED as well as the rise- and fall-time limits of the switchingcircuitry [2]. No matter how small we choose tON, the communica-tion is always possible as long as the photodiodes are able to sensethe pulse. So meanwhile, we should also consider the bandwidthof the receiver photodiodes. We can set tON to the physical limit ofthe system. But there is a dilemma when choosing tperiod. We wanttperiod to be large, in order for the duty cycle to be small. But thereare two drawbacks if we set it too large: one is the flickering effectand the other is the low potential data rate. Note that the switchingfrequency is

f =1

tperiod

.

Human eyes can perceive the light flashes if f is too small (120Hz or 160 Hz according to different studies [6]). Suppose we canencode N bits in each flashing cycle, then the potential data rateis Nf . There is no one fixed choice of tperiod; it depends on theapplication scenario. Actually, we can adapt: in the day when theambient illuminance is high, we can squeeze tperiod for higher datarate, while at night when there is little ambient light, we can extendtperiod for lower illuminance.

3. IMPLEMENTATIONIn order to facilitate the slim ON pulse, we need deliberate de-

sign of the transmitter and receiver circuits (Figure 2(a)). Switchingspeed is the major concern when we develop the circuit (Figure 2).The high switching speed requires the driver circuit, wiring, andthe micro-controller all to be fast.

Conventionally, we use an electrical switch (e.g. MOSFET) tomodulate the LED (e.g. PWM) and a micro-controller to drive thegate of MOSFET directly. We need a high current to charge thecapacitor at the gate because we want to switch the MOSFET onand off fast. But common micro-controllers are not able to providesuch a high current. For example, Arduino Uno can provide up toonly 40 mA. So we need to insert a dedicated gate driver betweenthe micro-controller and the MOSFET, which can supply a burstcurrent when switching on and off the MOSFET. The gate driveralso acts as a voltage amplifier if the output voltage of the micro-controller cannot reach the threshold voltage of the MOSFET.

Besides the switching speed of the electrical devices, wiring canalso hinder the speed of circuit. If the wires are too long and be-come so-called transmission line [4], the reflection of the signalwould occur. The signal can oscillate at the falling edges. The ef-fect is insignificant in low frequency circuit but it occurs during oursystem implementation as the speed is high enough. So the wiresshould be kept as short as possible.

We use an FPGA as our micro-controller because it offers thecontrol granularity at the clock-cycle level. We use a DigilentBASYS 3 board equipped with a Xilinx Artix-7 FPGA. The clockcycle of Artix-7 can reach 10 ns. Because the capacity of the PmodI/O interface on the board is only 24 MHz [3], the minimum pulsewidth that the FPGA board can achieve is approximately 40 ns. Butconsidering the speed of our driver circuit and the bandwidth of thephotodiodes we use, we set tON as 100 ns in our current testbed.The current bottleneck is the bandwidth of the photodiode ($ 6).We plan to examine high-end photodiodes in our future work tofurther reduce the pulse width.

At the receiver end, in order to detect the fleeting pulse signal, weneed an analog-to-digital convertor (ADC) with a very high sam-pling rate. In our implementation we use USRP N200, whose sam-pling rate can reach 100 MS/s. We equip the USRP with an LFRXdaughter board, which has a 50-Ω input impedance. Because theresistor that we connect in series with the photodiode is 100 kΩ,we need a unity gain voltage amplifier (a.k.a voltage follower) totransform the impedance, such that the USRP will not load the sig-nal source [7].

4. PRELIMINARY RESULTSWe investigate the relation between the illuminance and the switch-

ing period time (Figure 3(a)), which provides us the basis to choosethe period in different scenarios. The experiment was conducted atnight. We first measured the environment illuminance (I0) withoutthe LED light source using a light meter (Extech 401036), and thengradually increased tperiod (with fixed tON) thus decreasing the duty

0

2

4

6

8

10

10 100 1000 10000

Illu

min

an

ce

(lx

)

tperiod (µs)

w/ LEDw/o LED (I0)

(a) Illuminance vs period time

0.01

0.1

1

10

100

1000

10000

100000

10 100 1000 10000

Po

we

r (m

W)

tperiod (µs)

Normal LEDVLC in the dark

(b) Power vs period time

-0.04

0

0.04

0.08

0.12

0 5 10 15 20 25 30 35

US

RP

re

ad

ing

Time (ms)

1 1 0 0 1 1

(c) On-Off Keying (OOK)

Figure 3: (a) Illuminance of the LED with fixed tON (100 ns) but varying tperiod. The illuminance is measured at night and the environment

illuminance (I0) is 0.25 lx. The distance between the LED and light meter is 15 cm. The increase of the period time causes the illuminance to

approach the ambient illuminance. (b) We measure the average forward current using a multimeter and calculate the power by P = VsI . The power

of the LED used in common mode is 19.8 W, while the power of the LED used in VLC in the dark is as low as 46.8 µW. (c) On-Off Keying (OOK).

The occurrence and absence of peak represent 1 and 0 respectively. The switching frequency is 190.7 Hz, so the potential data rate using OOK is

190.7 bps.

cycle. We can see that the illuminance gradually drops to I0. Theminimum illuminance shift incurred by the LED is only 0.09 lx.For reference, the full moon on a clear night produces about 3 lxof illuminance outside [1]. The same experiment conducted in theday shows that the variation of the ambient illuminance (60–110lx) due to the movement of the clouds can already exceed the theilluminance of the LED. So the LED light is even harder to perceivein the day.

We measure the power of the LED when using VLC in the dark(same setting as the illuminance measurement) and compare it withthe power using in normal mode (Figure 3(b)). The low transmitpower does not make VLC in the dark prone to interference. On theone hand, the flashing frequency of potential interference is low, nomatter it is natural (sunlight) or artificial (fluorescent or LED lamp).Our experiment shows the sunlight is almost a DC signal, while thefrequency of a commercial LED lamp (Cree A19) is only 120 Hz(8 ms period). The intended 100-ns pulse signal can still be reliablydetected in the presence of those interferences. Furthermore, a highpass filter can remove those low frequencies. On the other hand,most photodiodes have a limited field of view (FoV), resembling adirectional antenna. The receiver gain of signal with an incidenceangle exceeding half of the FoV is very small. For SD5421, at 20

incidence angle, the received signal strength is only 10% of that at0. In this way, the interference is largely attenuated.

Our transmitter can send bits using On-Off Keying (OOK) mod-ulation, and our receiver can successfully detect the 100-ns pulseemitted by the LED (Figure 3(c)). The potential data rate (withoutconsidering the preamble overhead) using OOK is 190.7 bps witha period time of 5.24 ms, the longest tperiod shown in Figure 3(a).

We also consider the visibility when humans view the light sourcedirectly. In the day, with the longest tperiod mentioned above, thelight is invisible. But at night, because human eyes are very sensi-tive to light, we could always feel the light, with any period timethat yields a reasonable data rate and will not lead to flickering.

The threshold duty cycle varies with the LED as well as the am-bient light. If the peak intensity of the LED is higher, we shoulddecrease the duty cycle accordingly. If the ambient light becomesbrighter, human eyes become more tolerant toward the “dark” LED,thus the threshold duty cycle rises.

5. DISCUSSION AND FUTURE WORK

Boosting the Link Rate. We plan to examine photodiodes withhigher bandwidth so that we can further decrease the ON duration.With the same duty cycle, the period time can shrink thus increas-ing the switching frequency and achieving a higher data rate. Wewill also implement more sophisticated modulation schemes to im-

prove the data rate. Pulse Position Modulation (PPM) is a goodcandidate for VLC in the dark. It encodes more bits in one cycle byvarying the position of the pulse.

Supporting Distance. Our current prototype has a limited com-munication distance (15 cm), which is mainly constrained by thesensitivity of the photodiodes. To increase the supporting distance,we plan to test more sensitive photodiodes that are able to senseweaker light signals.

Adapting to Ambient Light. We also plan to study the impactof the system parameters (e.g. ON duration, period time, and mod-ulation scheme) on the system performance (e.g. communicationdistance, data rate, and reliability). Based on the understanding,we can adapt the LED light’s duty cycle and modulation schemesbased on the ambient light condition to achieve the best tradefoffbetween user perception and link data rate.

Potential Applications. We will also implement applicationsthat can leverage this new VLC primitive, e.g., building device-to-device and device-to-ceiling IoT network. VLC in the dark candemonstrate its ability in the Near Field Communication (NFC).For example, thanks to the high directionality of visible light, VLCin the dark can serve as a secure link during contact-less payment.

6. ACKNOWLEDGMENTSWe sincerely thank the reviewers for their insightful feedback.

We also thank Kevin Wright (Dartmouth College) for the adviceand support on building the hardware prototype. This work is sup-ported in part by the National Science Foundation under grant CNS-1421528.

7. REFERENCES[1] T. Borogovac, M. B. Rahaim, M. Tuganbayeva, and T. D. Little.

“Lights-off” visible light communications. In Proceedings of 2nd

IEEE Globecom 2011 Workshop on Optical Wireless

Communications, pages 797–801. IEEE, 2011.

[2] Cree, Inc. Pulsed over-current driving of Cree XLamp LEDs:

information and cautions. REV 1B.

[3] Digilent, Inc. Digilent Pmod interface specification, November 2011.

[4] R. Gilmore and L. Besser. Practical RF circuit design for modern

wireless systems, volume 1. Artech House, 2003.

[5] J. Li, A. Liu, G. Shen, L. Li, C. Sun, and F. Zhao. Retro-VLC:Enabling battery-free duplex visible light communication for mobileand iot applications. In Proceedings of HotMobile’15.

[6] L. Li, P. Hu, C. Peng, G. Shen, and F. Zhao. Epsilon: A visible lightbased positioning system. In Proceedings of NSDI’14).

[7] Texas Instruments, Inc. Amplifiers and bits: an introduction to

selecting amplifiers for data converters, May 2015.