BLISP: Enhancing Backscatter Radio with Active Radio for Computational RFIDs
Ivar in 't Veen, Qingzhi Liu, Przemysław Pawełczak, Aaron Parks, Joshua R. Smith
BBLISP: Enhancing Backscatter Radio with ActiveRadio for Computational RFIDs
Ivar in ’t Veen ∗ , Qingzhi Liu ∗ , Przemysław Pawełczak ∗ , Aaron Parks † , and Joshua R. Smith †∗ Delft University of Technology, Mekelweg 4, 2628 CD Delft, The NetherlandsEmail: { qinzhi.q.liu, p.pawelczak } @tudelft.nl † University of Washington, Seattle, WA 98195-2350, USAEmail: [email protected], [email protected]
Abstract —We demonstrate the world’s first hybrid radio plat-form which combines the strengths of active radio (long rangeand robustness to interference) and Computational RFIDs (lowpower consumption). We evaluate the Wireless Identificationand Sensing Platform (WISP), an EPC C1G2 standard-based,Computational RFID backscatter radio, against Bluetooth LowEnergy (BLE) and show (theoretically and experimentally) thatWISP in high channel attenuation conditions is less energyefficient per received byte than BLE. Exploiting this observationwe design a simple switching mechanisms that backs off toBLE when radio conditions for WISP are unfavorable. By a setof laboratory experiments, we show that our proposed hybridactive/backscatter radio obtains higher goodput than WISP andlower energy consumption than BLE as stand-alone platforms,especially when WISP is in range of an RFID interrogator forthe majority of the time. Simultaneously, our proposed platformis as energy efficient as BLE when user is mostly out of RFIDinterrogator range.
I. I
NTRODUCTION
Most low power wireless sensor nodes use active radiotransmission techniques, such as Bluetooth Low Energy [2],to transport data. While active radios are becoming betterwith each year (in terms of throughput and range), the powerconsumption expenditure of radio communication can still bemuch larger than the power expended for computation [3].This indicates that there is still a lot to be done to makewireless sensor nodes more power efficient, despite manyyears of research in low power electronics. One approach toreducing energy consumption of the wireless front end is by notactively transmitting, but instead modulating the reflection ofpower emitted by an external transmitter—as with RFID-basedComputational RFIDs (CRFIDs) [4].
A. Problem Statement and Research Question
Unfortunately the transmission technique used by CRFIDs,i.e. backscatter, has non-ideal characteristics compared to active
Supported by the Dutch Technology Foundation STW under contract 12491and in part by a Google Faculty Research Award, a Google PhD fellowship,the Intel Science and Technology Center for Pervasive Computing, and NSFaward CNS-1305072.Another version of this work is available in [1].c (cid:13) radio. While power efficient, backscatter is susceptible todistortion by the environment [5, Fig. 4]. Additionally, the pathloss for backscatter signals is very different than for activetransmissions. Active transmissions have a signal-to-noise ratiowhich approximately decays with the square of distance. Forbackscatter radio, this decay approximates the fourth power ofdistance [5, Sec. 2.2]. Hence, the energy wasted due to lostdata increases. At the same time, active radios, although moreresistant to interference, consume more energy than backscatterradios. The difference in power consumption is mostly dueto the need to actively emit RF power instead of reflectingpreexisting signals. This robustness/energy efficiency trade-off of active and backscatter radio calls for connecting theseplatforms. Practically, many real-life situations call for anextension of backscatter by active radio.
Example:
It is shown in [6] that cows have preferred regions(hotspots) within the paddock in which they spend the majorityof their time. In [6, Fig. 3] the number of hotspots (coveringless than 20 m ) is limited to six, and is spread over a large area( ≈
230 m ). To monitor cattle movement (C)RFID would coverthe hotspot area, while active radio would cover transitionalmovement.The research question is then: What energy consumption andtransmission reliability improvements can one get by exploitingthe combined benefits of active and backscatter radio?B. Contributions of This Paper
To answer this question we design a new heterogeneousradio sensor node combining both active and passive radioin one device. We call this platform BLISP—a compositionof B luetooth L ow Energy (state-of-the-art Commercial off-the-Shelf (COTS) active radio platform for consumer appli-cations [2]) and W ISP [7] (state-of-the-art CRFID). Thisproposed platform consists both of low-cost experimentalhardware combining the two radios in one system, and a radioselection technique (implemented in software) to choose theappropriate radio for the appropriate situation while trying tooptimize both reliability and energy efficiency. To show thebenefit of BLISP, the complete system is evaluated in replicablestatic and mobile scenarios using a COTS RFID reader and amodified smartphone-attached RFID reader.The contributions presented in this paper are: a r X i v : . [ c s . N I] A p r ontribution 1: we provide a set of simple theories,supported by experiments, showing the benefit of connectingactive and backscatter radio platforms; Contribution 2: we show the benefit of using BLISP as anextension to CRFID applications by demonstrating that it ispossible to transmit more data compared to an out-of-rangeCRFID while only increasing energy consumption per byte by ≈
15 % compared to Bluetooth Low Energy (BLE).
Contribution 3: we show the benefit of using BLISP as anextension to BLE applications by demonstrating the possibilityof transmitting the same amount of data compared to BLEwhile decreasing energy consumption per byte by more than50 %.The rest of this paper is organized as follows. Section IIreviews related work. Research motivation is provided in Sec-tion III, followed by Section IV discussing a simple feedback-less radio switching method for BLISP. Section V presentsexperimental platform used to verify the quality of the proposedswitching mechanism of which the results are discussedin Section VI. A discussion on limitations and future work isgiven in Section VII, and the paper concludes with Section VIII.II. R
ELATED W ORK
We start by reviewing literature pertaining to active andbackscatter radios and connection thereof into an hybrid device.
A. Computational RFID
The use of CRFID for wireless sensor applications hasbeen advocated by many papers including [8], [9]. Theonly stable CRFID [4] implementation currently availableis Wireless Identification and Sensing Platform (WISP).The communication protocol used by WISP is the industrialstandard EPCglobal Class 1 Generation 2 (EPC C1G2) RFIDprotocol. Although completely battery-autonomous, CRFID hasintrinsic limitations: Limited channel robustness, as evaluatedby [5]; and limited RF power transfer efficiency results in anintermittent power supply. A solution to the continuous powersupply problem proposed by [10] exercises a hybrid powersolution based on RF power harvesting and an energy storagedevice. While this significantly improves CRFID energy supplystability, it does not solve the robustness problem.
B. Bluetooth Low Energy
Active (low power) radio systems are less susceptible tointerference compared to backscatter communication. However,they bring the disadvantage of higher energy consumption.There are multitudes of low power active radio platforms, andreviewing all options is not in the scope of this work. However,there is one believed to be broadly adopted, with more than30 billion devices expected to reach the consumer market by2020 [11]: BLE—the newest version of the Bluetooth protocoloptimized for low energy applications . Works by [2], [15] For example, recent standards like SigFox [12], LoRa [13] or IEEE802.15.4k [14] could be used, and are expected to have even lower energyconsumption than BLE. We will not use them in this work as they are not(yet) easily accessible for experimental evaluation, nor broadly adopted. experimentally evaluate the performance of BLE, while [16]shows the energy consumption of BLE compared to otherpopular active radio technologies. No studies comparing theenergy consumption of BLE with a backscatter-based CRFIDhave yet been published to the best of our knowledge.
C. Multi-Radio Systems
A combination of backscatter radio and active radio seemsto be the logical step to solve the imperfections of bothsystems. Again, to the best of our knowledge, no such hybridimplementation exists. One obvious way of using BLE toextend the RFID range is to use multiple RFID readers whichare coupled using BLE, as proposed for different radio types(with node-to-node communication) by [17]. This approach,unfortunately, cannot be used for BLISP because state of the artCRFIDs cannot communicate with other CRFIDs without theinterrogator. The only hybrid active/backscatter platform we areaware of is [18], which uses BLE to reprogram a backscattertestbed, and does not use the active radio to improve reliability.Authors of [19] propose a method of using BLE as aphysical transport layer for an RFID protocol. A backscatter-BLE method is proposed in [20], which allows a backscatterdevice to synthesize BLE packets but which has similar channelconstraints as conventional backscatter. In the non-backscattercontext, an approach to combine multiple heterogeneous radiosby [21] uses acknowledgement delay and machine learningmechanisms to optimize system performance. All above-mentioned multi-radio platforms rely on acknowledgementsfrom the receiving party and/or active radio transmissions.
D. RF Power Harvesting
Considering literature related to energy storage in CRFID,we need to mention [10] again proposing to store energy inbattery/capacitor for future use and [22] where energy storagefrom rectifying Wi-Fi signals has been proposed.III. M
OTIVATION FOR C OMBINING A CTIVE AND B ACKSCATTER R ADIO
To understand why backscatter is not always the mostefficient radio technique, we introduce a simple analyticalbasis to bring insight into the design of BLISP. The theoreticalmodel is followed by experimental results verifying the theory.
A. Difference in WISP and BLE Radio Efficiency
We start with the analytical model.
1) Analysis:
Assume a hybrid radio platform composedof i = { . . . , n } independent radio technologies (such asbackscatter and active radio). We characterize the energyper successful transferred byte for radio i as E byte , i ( d ) = E tx , i /B rx , i ( d ) , where E tx , i is the total amount of energy spentin transmitting data and B rx , i ( d ) is the number of receivedbytes for distance d ∈ [0, d max ) . Generalizing [23, Sec. III-A] B rx , i ( d ) (cid:44) LL + H [1 − erfc ( f i ( d ))] L + H , (1)where L and H are the payload size in bits and the amount ofoverhead in bits, respectively, and erfc(.) is the complementaryrror function. We define the signal quality decay function f i ( d ) = ( d/a i ) − r i , a i as radio-intrinsic correction value and r i as loss coefficient. For example, a typical value of r i = 2 for active radio or r i = 4 for backscatter radio. Now, basedon the above model we pose the following lemma. Lemma 1:
Any hybrid radio composed of n radios haslimited range after which energy consumption per byte isinfinite. Proof: ∀ n lim d → + ∞ B rx , i ( d ) → ⇒ E byte , i = E tx , i /B rx , i ( d ) → + ∞ which completes the proof. Corollary 1:
Defining E( d ) (cid:44) { E byte ,1 ( d ), · · · , E byte , n ( d ) } if ∃ E byte , i ( d ) ∈ E( d ) ∀ E byte , j ( d ), i (cid:54) = j E byte , j ( d ) < E byte , i ( d ), ∀ d , thenradio j can be removed from designing a hybrid radio. Corollary 2:
The maximum range of a system is limited bythe radio with the largest range.
Corollary 3:
At distance d the lower bound of the hybridradio energy consumption per byte is given by the radio withthe lowest energy consumption at that distance.
2) Measurement:
To verify this simple analytical modelwe need to measure the consumed power of each radio asa function of the signal loss. We first introduce the selectedhardware for BLE, WISP and finally the measurement setup. a) Bluetooth Low Energy—Transmitter/Receiver:
Weselected the Nordic Semiconductor PCA10005 evaluationmodule with an NRF51822 BLE System on Chip (SoC) [24]as BLE transmitter. The software used on the BLE radio is acustomized firmware version (source code is available uponrequest or via [25]) transmitting only standard advertisingmessages [11] at a constant rate of 120 Byte/s=0.96 kbit/s,which is comparable to 0.65 kbit/s of [26, Sec. III-B]. BLEhas a maximum packet size smaller than the selected payload(i.e. 24 Byte) therefore each transmission consists of multiplepackets. A second identical NRF51822 module is used as BLEreceiver—continuously logging advertisement messages sendby the BLE transmitter. b) Computational RFID—Transmitter/Receiver:
We se-lect WISP 5 as a state-of-the-art CRFID platform [7]. TheWISP 5 used for experiments has the RF energy harvesterdisabled by desoldering the output pin of the buck converter.This modification simplifies the energy measurement, as theenergy provided to WISP 5 is not fluctuating in time as in thecase of harvested energy. The WISP 5 firmware is adapted (seeagain [25]) to transmit with the same data rate as BLE. Again,as in the case of BLE, since the maximum payload of WISP 5,i.e. 12 Byte, is smaller than 120 Byte each message consistsof multiple packets. The RFID reader is an Impinj SpeedwayR420 [27], controlled via SLLURP Low-Level Reader Protocol(LLRP) library [28], and connected to a panel antenna [29].Based on observations by [30, Sec. 4.1] we have chosento use the EPC C1G2 Electronic Product Code (
EPC ) fieldas our data carrier instead of the
Read command. Usingthe
EPC field cuts down on the protocol overhead because ithalves the amount of roundtrips [31, Sec. 6.3.2.12.3]. Accordingto [31, Sec. 6.3.2.1.2.2] the length of the
EPC field may beset between zero and thirty one words. While it is possible tohave WISP transmit longer
EPC values to reduce the overhead,
30 40 50 60 70 80 90020406080100 Loss d [ dB ] E ne r g y E b y t e [ µ J ] WISP (Battery)WISP (Fitted)BLE (–30 dBm)BLE (Fitted)BLE (4 dBm)
Fig. 1.
Energy per byte over distance for WISP and BLE.
The dasheddata points are extrapolated, the constant power consumption for the BLEradio and all data being received, yields constant energy per byte. Fitted plotsare based upon (1). Parameters for fitted WISP curve: a i = 30 , d i = 4 , E tx , i = ( L + H )5 µJ with L = 96 and H = 320 . Parameters for fitted BLEcurve: a i = 87 , d i = 2 , E tx , i = ( L + H )21 µJ, L and H are equal to WISP. this increases the probability of corrupted messages [23]. c) Measurement Setup: We measure energy per byte at thereceiver (separately for BLE and WISP) as a function of signalattenuation. This is realized with two signal attenuators [32]connected in series. These attenuators limit the signals bi-directionally, and therefore both uplink and downlink areattenuated at the same time. Both BLE transmit/receiveevaluation boards used are equipped with an antenna connectorallowing attenuators to be inserted directly into the transmissionchannel. WISP, on the other hand, does not provide such anantenna connector and therefore it is positioned at a fixeddistance of 50 cm from the interrogator antenna which is thenconnected to the RFID interrogator via the attenuators.The BLE module [24] has an uncalibrated transmissionpower setting via the API of the S110/S120 (transmit-ter/receiver, respectively) softdevice. The highest (4 dBm) andlowest (–30 dBm) transmission power are tested. The RFIDreader is tested at its maximum transmission power (32.5 dBm).We measure the power consumption of both radios using aself developed, buffered, differential, sensing circuit monitoringthe voltage drop over a 100 Ω shunt resistor in series with theDevice Under Test (DUT). This circuit is coupled to a TektronixMDO4054B–3 oscilloscope [33] to measure power over timewhich is used to calculate the energy consumption. Schematicsof this device are available upon request or at [25].
3) Measurement Results:
The relationship between energyper byte and signal loss, as measured for both active andbackscatter radio and complementary fitted plots, is shownin Fig. 1. As expected, the WISP—while more energy efficientin good channel conditions—also has a shorter range ofoperation. Instead of a gradual increase in energy consumptionper received byte, at one point the energy per byte metricstarts to rapidly increase for both platforms. This “brick wall”effect [23, Sec. V] is caused by an increase in bit errors,causing whole packet loss and therefore requiring more transmitattempts per successfully received byte.
B. Do Alternatives Exist to Hybrid Active/Backscatter Radio?
The question remains of whether, in the light of thisobservation, the hybrid radio platform is the only solutionhich improves energy efficiency and transmission range ofCRFID. We review the alternatives and provide our answer.
1) Low Power Active Radio with Battery:
The simplestalternative to the hybrid platform would be a connection of asufficiently large battery and BLE radio.
Limitation:
Unfortunately, all batteries will eventuallydeplete, leading to expensive battery (or even whole device)replacement. For battery replacement, the device must bephysically accessible, as it is impossible to wirelessly restore theenergy level of the empty battery without an energy harvester.
2) Power Harvester with Active Radio:
Wireless RF powerharvesters solve the physical accessibility and battery constraint.
Limitation:
Inefficiencies in RF power harvesters, energystorage, energy conversion, and energy transmission throughRF waves, mean that no power harvester and active radiocombination will be as energy efficient as a backscatter radio.
3) Backscatter Radio with Improved Channel Coding:
The operational reliability and robustness of communicationof CRFIDs could be improved by adding a more extensivechannel coding mechanism. For example: WISP is currentlylimited to the FM0 coding [31, Sec. 6.3.1.3.2.1], in which eachbit is represented by one signal alternation for each symbol.Miller coding methods [31, Sec. 6.3.1.3.2.3] have redundantalternations within each symbol, reducing the possibility oflost messages.
Limitation:
Channel coding would make CRFID morerobust (i.e. shift the WISP curve to the right in Fig. 1),still keeping CRFID susceptible to reflections and destructiveinterference. Finally, we conjecture, this would still not makeWISP as energy efficient as BLE in a broad attenuation range.IV. C
HANNEL E STIMATION M ETHODS FOR H YBRID A CTIVE /B ACKSCATTER R ADIO P LATFORMS
We propose a method of estimating the backscatter channeland use this estimation to select between backscatter and activeradio on-the-fly. We start with revising unsuitable solutions.
A. Backscatter Channel Quality Estimation Methods: Reviewof Unsuitable Solutions
Because backscatter radios behave differently than activeones, typical channel estimation methods do not directly apply.
1) EPC C1G2 Protocol Feedback:
The de facto standardmethod of assessing packet reception rate is to query thereceiving party if it indeed received a packet. Most protocolsrely on receive acknowledgments for (all) packets.
Limitation:
Within the EPC C1G2 [31] protocol there areno standard ways to guarantee the successful reception of
EPC values transmitted by a tag. The default method of awaitingan Acknowledgement (
ACK ) message for each transmitteddata message is therefore not possible. The exclusion of thisfunctionality is logical for standard RFID tags, as they arecomputationally limited, transmit unchanging identifier, andmost likely could not handle retransmissions. Transmitting databack to a CRFID also implies that CRFID should handlecomputationally hard, and a protocol-wise large overheadinducing EPC C1G2 write accesses.
2) BLE Protocol Feedback:
The more responsive BLEchannel could be used to provide a feedback for the receptionof RFID packets transmitted by CRFID.
Limitation:
The use of a separate radio channel couldincrease RFID reliability because the channels might breakdown under different circumstances. However, it might alsodecrease reliability because the BLE channel might be brokenwhile the RFID channel is working. Practically, including aBLE radio in receiving mode will also dramatically decreasethe energy efficiency of a hybrid platform, as the radio has tolisten for an extended (worst case: continuous) time.
3) RSSI Strength Feedback:
Neither CRFID hardwarenor EPC C1G2 protocol has a built-in support for ReceivedSignal Strength Indication (RSSI) measurement on the RFIDtransmission. A coarse method to estimate the vicinity ofRFID reader is by measuring the amount of energy harvestedby the CRFID. If a tag is close to a reader, it is easily possibleto harvest energy, while if a tag is far away it would be almostimpossible to harvest it. The BLE radio has native support forRSSI measurements on the received messages.
Limitation:
Measuring RSSI for the signal originating fromthe interrogator and received by the backscatter radio doesnot directly correlate with the channel quality for backscatterdata (as there is no constructive interference, as explainedin Section I-A). While the interrogator knows the RSSI, thebackscatter device cannot reliably determine it. A CRFID couldquery the RFID reader for its RSSI as measured by the readerbut this would induce a lot of overhead on both sides. I BLEshould be placed into listening mode in order to retrieve RSSIvalues, which is more power consuming than the transmissionmode. Therefore, enabling BLE only for channel estimationwithout using it for data transfer is a loss of energy.
B. Proposed Channel Quality Estimation Method
For the BLISP system we propose a novel, less standard,way of estimating the channel.
Proposition 1:
Tracking the number of EPC C1G2
RN16ACK messages in handshake can be used to estimate thebackscatter channel quality.
Proof: (Sketch) If RFID interrogator and tag performa multipart handshake, the backscatter channel is usable totransfer data. Work of [5] proposed an approach for settingan interrogator to its optimal settings based on both measuredRSSI and packet loss. However, packet loss-based, estimationscan also be performed on the tag instead of the interrogator.Part of this handshake is the tag sending the reader a randomnumber (
RN16 ), which the reader should acknowledge by an
ACK message containing this random number. To reach the
ACK both channels (to and from) the CRFID tag need to be ina state good enough to transmit a payload. By measuring thenumber of handshakes and testing this number to be at leastthe same as the amount of packets we expected to transmit,we are able to estimate quality of the backscatter channel.V. BLISP D
ESIGN
We are now ready to introduce BLISP, our hybrid backscatterand active radio platform, to help exploit the main trade-
ISP 5 EnergyManager × NRF51822ThermalSensor Battery/Capacitor × Notimplementedin this paper
ImpinjR420 † NRF51822 † + EnergyVCC VCCStorage VCCData UART (w/ flow control)LLRP UARTEPC C1G2 BLE
BLISP TransmitterSinkBLISP Receiver
Fig. 2.
Overview of the BLISP system consisting of one transmitter andone receiver.
The temperature sensor providing data is part of WISP butdisplayed separately for clarity and completeness. All displayed connectionsdepict a flow of energy or data and do not directly correspond to physicalconnections. For a detailed description of the physical connections see [25]. † For the mobile reader experiments the Impinj R420 is replaced by an MTIMINI ME, and the NRF51822 by the BLE receiver of a Samsung Galaxy S3.
TI FET Interface WISP5BLE AntennaPCA10005NRF51822 (a) The top side of the BLISP with annotations for the mostimportant components. Please note that the WISP antenna is notfully shown. (b) Bottom side of the BLISP. Symbolicaly illustrated directionalconnections by color, as numbered: (1) white : ground; (2) brown :clear to send; (3) red : power supply; (4) green : WISP to BLEserial channel; (5) orange : BLE to WISP serial channel (unused);(6) yellow : ready to send, and (7) blue : power supply.Fig. 3.
The BLISP PCB as seen from top (Fig. 3(b)) and bottom (Fig. 3(a)).
PCB design files are available upon request or from [25]. offs as proposed in Lemma 1 and Corollaries 1 to 3. TheBLISP infrastructure mainly consists of two parts: (i) a COTSRFID interrogator combined with a BLE receiver; and (ii) ourmulti-radio sensor node—the BLISP. To provide a flexibleplatform we opt to combine two readily available radios insteadof developing our own, single silicon, platform.A complete system level overview of the BLISP is shownin Fig. 2. The main design principle behind BLISP is theabsence of any algorithm on the host side: the host only mergesthe multiple data streams received by the different radios.
A. BLISP Hardware Architecture
The chosen radio modules for this platform are the same asdescribed in Section III-A2. PCB has been designed to ease
TABLE IS
ETUP P ARAMETERS OF D ATA A GGREGATORS
Component Parameter Mobile BLISP RX Static BLISP RXHost Device Model Samsung Galaxy S3 Lenovo T530Software Android 4.3 Linux 3.13.0RFID Reader Model MTI MINI ME Impinj R420TX Power 18 dBm 32.5 dBmRX Sensitivity –84 dBm –82 dBmAntenna Gain 2 dBi 9 dBiLink Frequency 640 kHz 640 kHzCoding FM0 FM0Session 2 2Q-value 5 n/aDuty Cycle 100% 100%BLE Receiver Model Samsung Galaxy S3 Nordic NRF51822Duty Cycle 100% 100% the connection of the two separate radio platforms, see Fig. 3.The PCB connects the active and passive radio, and providesmeans for radio collaboration and energy distribution.
1) Active Radio:
We use the same NRF51822 BLE moduleas described in Section III-A2.
2) Backscatter Radio:
As backscatter radio we also use thesame WISP 5 as described in Section III-A2.
3) Radio Collaboration:
A communication channel isneeded to convey desired state information for the active radioand to share sensor values between the two separate radios. TheNRF51822 BLE module has a silicon bug causing high powerconsumption by perpetually keeping non-vital microcontrollerperipherals enabled [34, Id 39]. This bug unfortunately affectsall conventional (digital) communication channels includingGeneral Purpose Input/Output (GPIO)-interrupts renderingthem useless as low power wake-from-sleep devices. The lowpower analog comparator peripheral is not affected by this bug,therefore this peripheral is used as wake-up signal enabling thehigh throughput Universal Asynchronous Receiver/Transmitter(UART). The BLE radio also uses digital output as CTS signal.
4) BLISP Receiver/Sink:
The receiving side of BLISPconsists of two receiving radios matching the two transmittingradios on the BLISP. In contrast to [21], the BLISP receiveris as simple as possible and only merges the data streamsfrom the receiving radios. Because the host does not makedecisions about which radio to use, the BLISP can switchwithout synchronization mechanism. We present two hostsetups: (i) a fixed receiver; and (ii) a mobile smartphone setup. a) Fixed Receiver:
The fixed receiver consists of ahost computer with an Ethernet connected Impinj SpeedwayR420 [27] and an USB/UART connected Nordic SemiconductorNRF51822 [24]. This setup is again described in Section III-A2. b) Mobile Receiver:
To test BLISP with a mobile reader,comparing to the fixed reader case, we have a prepared thefollowing setup. Smartphone is selected as platform for mobilehost, which consists of BLE and RFID reader. We developedan Android application (available upon request or via [25]) forthe smartphone to scan the BLE channel and log all advertisingdata originating from the BLISP. As a smartphone-attachableRFID reader we selected the MTI MINI ME [35]. Based onthe low level command set Application Programming Interface(API) provided by MTI, we log all inventory data.Unfortunately, the MINI ME can only inventory WISP withfixed power supply up to a maximum range of 2 cm. To increase martphone MINI ME GSM AntennaBLISPTI FET Programmer
Fig. 4.
Mobile receiver BLISP test setup.
BLISP and TI FET program-mer/power monitor are hanging from an overhead crane (the red diagonalwires) [37] as described in Section VI-A. MINI ME mobile RFID reader isshown without plastic housing, easing connection of a different antenna. the inventory range of MINI ME, we replace the embeddedantenna with a 2 dBi GSM band omnidirectional antenna [36].By replacing the antenna, the maximum range is extended to10 cm. Table I shows parameters for the two reader platforms,while Fig. 4 shows the MINI ME reader with GSM antennaconnected to a smartphone running our application.
B. BLISP Software Architecture
The WISP component of the BLISP software consists of1700 lines of C code and 1900 lines of assembly code of which600 lines C and 50 lines assembly were written in the BLISPdevelopment process. The remaining part is based upon [38].The BLE element consists of a 500 line C coded program andthe NRF51822’s API. The fixed BLISP host currently consistsof various Bash and Octave scripts with varying lengths. Themobile host consists of 750 lines of customized Java code.
1) Wireless Identification and Sensing Platform:
Because ofthe low power requirements and therefore our preference forbackscatter communication we choose to have WISP actingas master over BLE radio. Between the periodic sensing andtransmission rounds WISP is put into a low power state.For all following experiments WISP measures temperatureand a timestamp since the startup . The timestamp is includedfor evaluation purposes, as this value enables to evaluatethe number of missing and/or duplicate packets. To ensurea constant data stream in case of radio switching the sensordata is periodically shared with the BLE radio as describedin Section V-A3. The BLE radio and the WISP are both setto have 12 Byte payload per message and ten messages arecombined into a single transmission. As the temperature datacombined with the timestamp only uses 4 Byte the message ispadded with 8 Byte of constant data.Because of incompatibilities between WISP and the MINIME RFID reader used for the mobile host experiments theEPC C1G2 tag select mechanism [31, Sec. 6.3.2.3] isdisabled for all fixed and mobile reader experiments.
2) Bluetooth Low Energy:
BLE module (as decribedin Section III-A2a) is programmed as slave under WISP.As described in Section V-A3 the BLE radio is periodicallyawaken by the WISP to receive new data. When not wirelessly Other possible sensors are the accelerometer, already available on WISP,or any other (low power) electronic sensor.
Algorithm 1
BLISP Control Protocol x ← Maximum backoff window, see Section V-B3a each P ERIOD n do a ← n − (cid:46) Received ACKs r ← RAME n − (cid:46) Frames planned to transmit WISP ok ← ( a = r ) (cid:46) Expect ACK for each frame if WISP ok then backoff ← (cid:46) No backoff on success if backoff then (cid:46) Is (re)try slot? WISP TX ← true (cid:46) Transmit using WISP if ¬ WISP ok then backoff ← U (0, x ) (cid:46) New uniformly random backoff else
WISP TX ← false (cid:46) Not transmit using WISP backoff ← backoff − (cid:46) Shift backoff
BLE TX ← ¬ WISP ok (cid:46) Use BLE if not use WISP transmitting nor receiving (UART) data from WISP BLEmodule is put into a low power sleeping state.
3) Radio Switching:
The software implements feed-forwardchannel estimation as proposed in Section IV-B. The cir-cumstances and environmental influences affecting the RFperformance of the WISP might change in a very irregularand most likely unpredictable way. We therefore propose andevaluate two switching approaches. a) Random ( < x ): Making the switching mechanismdepend on past results will decrease the number of unnecessarybackscatter channel evaluations, thereby reducing overheadand improving energy efficiency. Because we assume theenvironment to have random unpredictable behavior we optthat it does not make sense to include a sophisticated self-learning algorithm. Our random backoff approach implementsan ALOHA-inspired random backoff window with a maximumvalue of x . A low value of x will make the system moreresponsive while a high value will make the system morestable in the long run. A pseudo-code representation of thisswitching algorithm in shown in Algorithm 1. b) Na¨ıve: Limiting the maximum random value to zerowill generate a constant as-short-as-possible backoff windowresulting in the na¨ıve approach. This approach (used as areference) assures that we use WISP as much as possiblewhich increases energy efficiency. At the same time, checkinga perpetually broken WISP communication channel induces anoverhead compared to other maximum backoff window sizes.VI. E
XPERIMENTAL E VALUATION
To test the performance of BLISP we executed the followingexperiments measuring both goodput and energy consumption.
A. Experiment Setup
Our experimental setup consists of hardware componentsand methodologies for replicable and traceable measurements.For this test the BLISP (built as described in Section V-A) wasrunning software as described in Section V-B.
1) Hardware:
The measurement and evaluation setup weuse for these experiments is based on the setup describedin Section III-A2. In addition we use an automatic three-dimensional positioning crane [37] situated in a lab environmentto automate the experiments involving a mobile BLISP. oving Out of range In range0.00.20.4 ∞ E ne r g y E b y t e [ m J ] WISP (Battery) BLE (4 dBm) Na¨ıveRandom ( < ) Random ( < ) (a) Energy per byte comparison for all setups in different scenarios Moving Out of range In range0.01.02.03.0 · D a t a [ B y t e ] WISP (Battery) BLE (4 dBm) Na¨ıveRandom ( < ) Random ( < ) (b) Number of received messages for all setups in different scenariosFig. 5. Results of the WISP, BLE and BLISP evaluation using ImpinjR420 RFID reader.
Because of WISP not being able to transmit data in thelong range, see Fig. 5(b), effectively wasting energy, the energy per byte isinfinite for this situation, see Fig. 5(a). We show the only WISP, only BLE,na¨ıve BLISP and random BLISP for random backoff windows up to threeand ten slots. These experiments have been normalized to unique messageseliminating messages transmitted by both radios around switching moments.
2) Replicability:
According to Fig. 1 wireless radios havetwo main ranges of operation: (i) within the first range mostof the packets get received and therefore the energy per byteratio stays rather constant, (ii) within the second range almostno packets are received and the energy spend on transmitting abyte therefore increases drastically. For the BLISP performancetests we limit the transmission power and sensitivity of RFIDreader and define two static positions, one in WISP-range andone outside WISP-range. The experiments were performed byplacing the BLISP in the in-range spot, placing the BLISPin the out-range spot, and alternating the BLISP locationbetween the in- and out-range positions on a predefinedconstant time interval (10 s). The BLE radio was in rangefor all experiments, otherwise the system would fail accordingto Corollary 2. The time duration for each experiment was2 min and each experiment was repeated five times. We runbaseline experiments with a battery powered WISP and a BLEradio transmitting at 4 dBm as used in Section III-A.
3) Data Collection:
In experiments we log the number ofreceived packets for RFID and BLE receivers. The powerconsumption is measured by the programmer interface usingthe EnergyTrace platform [39]. Due to random startup delaysof each platform, we match the start and stop of an experimentby asynchronously starting all platforms and logging their stateafter a fixed (empirically found) delay of 3 s. B. Static RFID Reader Experiment
Measurements of energy per byte and transferred data,are shown in Fig. 5(a) and Fig. 5(b), respectively. Due to Because of the limited API for the EnergyTrace platform we usesynchronously timed screen shots and Optical Character Recognition (OCR)to log the energy measurements for experiments using the EnergyTrace.
Out of range In range0.00.20.4 · − ∞ E ne r g y E b y t e [ m J ] WISP (Battery) BLE (4 dBm) Na¨ıveRandom ( < ) Random ( < ) (a) Energy per byte comparison for all setups in different scenarios Out of range In range0.01.02.03.0 · D a t a [ B y t e ] WISP (Battery) BLE (4 dBm) Na¨ıveRandom ( < ) Random ( < ) (b) Received messages per radio (BLE or WISP). Note: dark (top)part of bars—BLE messages, light (bottom) part—WISP messagesFig. 6. Results of the WISP, BLE and BLISP evaluation using MiniMeRFID reader.
Again, in the long range the MINI ME reader is not able toreceive data transmitted by the WISP. Fig. 6(b) shows the distribution ofreceived messages per radio for completeness of the illustration. Because ofthe limited logging capabilities on the smartphone the number of messages isnot normalized to number of unique messages. normalization to unique messages, the values in Fig. 5(a) arearound ten times larger than the ones shown in Fig. 1.Our experiments show that BLISP increases goodput almostinfinitely in the long range compared to WISP (see Lemma 1)while not severely increasing power consumption over WISPin the short range. On the other hand BLISP almost halvesenergy consumption in the short range compared to a normalBLE radio while for Random ( x < ) increasing energyconsumption by ≈
25% on the long range. For the remainingtwo switching methods this difference is much larger. This ispresumably caused by the amount of unneeded channel sensingoperations and the overhead of redundant micro-controllers.As we add a mobility to the experiment we see WISP loosinga share of messages corresponding to the relative out of rangetime, this increases the energy per byte to the same level as theactive BLE radio which is able to transfer data in all positions.The combined system cannot be more energy efficient than themost efficient radio for a certain position (see Corollary 3).For an uniformly distributes in-/out-range mobility patternthe energy profit the BLISP has over the BLE radio in shortrange and the energy cost in the long range zero out. BLISPimproves energy efficiency and throughput for situations inwhich the WISP can be used for half of the time.
C. Mobile RFID Reader Experiment
The experiment setup parameters on the BLISP side is thesame as using fixed RFID reader in Section VI-A. The detailsetup parameters of mobile data aggregator are as in Table I.Results for the mobile host experiments as shown in Fig. 6show comparable results among WISP, BLE and BLISPcompared with fixed reader experiments from Section VI-B.The relative improvement from BLE to WISP and na¨ıve -BLISPusing a mobile reader is even larger while in-range. Thiselative improvement is mainly because the performance ofthe smartphone’s BLE module has worse performance than theNRF51822 receiver. Interestingly, for in-range measurements, alarge backoff window shows worse performance than the na¨ıve and small backoff experiments. We suspect that this is caused bythe hardware limitation of MINI ME. Based on our experiments,the MINI ME reader has trouble with rapidly moving, or onlyshortly available, RFID tags. Fortunately, the BLISP algorithmdetects the failing RFID reader and correctly enables the BLEradio which results in continuous data availability.VII. L
IMITATIONS AND F UTURE W ORK
We list the limitations and action items for future workrelated to hybrid active/passive radio platforms:1)
Improving platform switching mechanism:
Non-predictable mobility patterns require further research onlearning mechanism to select the best backoff parameter x of Algorithm 1, or the complete redesign thereof.2) Reducing micro-controller overhead:
The currentBLISP is built using two separate radio modules andtherefore two micro-controllers. One micro-controller is abetter approach, reducing energy consumption of BLISP.3)
Extending to beyond two radio platforms:
By Corol-lary 2 and Corollary 3 adding radios with heterogenouscharacteristics to a hybrid system will increase the perfor-mance of BLISP, requiring research on radio selection.VIII. C
ONCLUSION
In this paper we design, implement, and evaluate a hybridradio platform composed of Wireless Identification and SensingPlatform (WISP) and Bluetooth Low Energy (BLE), denoted asBLISP. Through experiments we show that BLISP, in situationsin which this hybrid platform stays within the reception regionof the lowest power radio, i.e. WISP, the energy efficiency isimproved compared to BLE. At the same time the reliabilityof BLISP is larger than the reliability of WISP alone whenBLISP moves frequently away from the RFID reader range.R
Sensors , vol. 12, no. 9, pp. 11 734–11 753, Aug. 2012.[3] R. Fonseca, P. Dutta, P. Levis, and I. Stoica, “Quanto: Tracking energyin networked embedded systems,” in
Proc. USENIX OSDI , vol. 8, SanDiego, CA, USA, Dec. 8–10, 2008.[4] A. P. Sample, D. J. Yeager, P. S. Powledge, A. V. Mamishev, and J. R.Smith, “Design of an RFID-based battery-free programmable sensingplatform,”
IEEE Trans. Instrum. Meas. , vol. 57, no. 11, pp. 2608–2615,Nov. 2008.[5] P. Zhang, J. Gummeson, and D. Ganesan, “Blink: A high throughputlink layer for backscatter communication,” in
Proc. ACM MobiSys , LowWood Bay, Lake District, UK, Jun. 25–29, 2012.[6] Y. Guo, G. Poulton, P. Corke, G. Bishop-Hurley, T. Wark, and D. L.Swain, “Using accelerometer, high sample rate GPS and magnetometerdata to develop a cattle movement and behaviour model,”
EcologicalModelling , vol. 220, no. 17, pp. 2068–2075, Sep. 2009.[7] (2015) WISP 5.0 wiki. [Online]. Available: http://wisp5.wikispaces.com/ [8] D. J. Yeager, P. S. Powledge, R. Prasad, D. Wetherall, and J. R. Smith,“Wirelessly-charged UHF tags for sensor data collection,” in
Proc. IEEERFID , Las Vegas, NV, USA, Apr. 16–17, 2008.[9] M. Philipose, J. R. Smith, B. Jiang, A. Mamishev, S. Roy, and K. Sundara-Rajan, “Battery-free wireless identification and sensing,”
IEEE PervasiveComput. , vol. 4, no. 1, pp. 37–45, Jan. 2005.[10] Y. Dong, A. Wickramasinghe, H. Xue, S. Al-Sarawi, and D. C.Ranasinghe, “A novel hybrid powered RFID sensor tag,” in
Proc. IEEERFID
Texas Instruments Application Note AN092
Proc. IEEE WCNC Workshops , Paris, France, Apr.1, 2012.[17] A. B. M. A. A. Islam, M. S. Hossain, V. Raghunathan, and Y. C. Hu,“Backpacking: Energy-efficient deployment of heterogeneous radios inmulti-radio high-data-rate wireless sensor networks,”
IEEE Access , vol. 2,pp. 1281–1306, Oct. 2014.[18] E. Kampianakis, J. Kimionis, K. Tountas, and A. Bletsas, “A remotelyprogrammable modular testbed for backscatter sensor network research,”in
Real-World Wireless Sensor Networks , K. Langendoen, W. Hu,F. Ferrari, M. Zimmerling, and L. Mottela, Eds. Springer InternationalPublishing, 2014, vol. 281, pp. 153–161.[19] R. Brideglall, “RFID device, system and method of operation includinga hybrid backscatter-based RFID tag protocol compatible with RFID,Bluetooth and/or IEEE 802.11x infrastructure,” U.S. Patent US 7 215 976B2, May 8, 2007.[20] J. F. Ensworth and M. S. Reynolds, “Every smart phone is a backscatterreader: Modulated backscatter compatibility with bluetooth 4.0 devices,”in
Proc. IEEE RFID , San Diego, CA, USA, Apr. 15–17, 2015.[21] J. Gummeson, D. Ganesan, M. D. Corner, and P. Shenoy, “An adaptivelink layer for heterogeneous multi-radio mobile sensor networks,”
IEEEJ. Sel. Areas Commun. , vol. 28, no. 7, pp. 1094–1104, Sep. 2010.[22] V. Talla, B. Kellogg, B. Ransford, S. Naderiparizi, S. Gollakota, andJ. R. Smith. (2015) Powering the next billion devices with Wi-Fi.[Online]. Available: http://arxiv.org/abs/1505.06815[23] P. Lettieri and M. B. Srivastava, “Adaptive frame length control forimproving wireless link throughput, range, and energy efficiency,” in
Proc. IEEE INFOCOM
Proc. IEEE RFID