EkhoNet: High Speed Ultra Low-power Backscatter For Next .

2y ago
18 Views
2 Downloads
4.74 MB
12 Pages
Last View : 2m ago
Last Download : 2m ago
Upload by : Cannon Runnels
Transcription

EkhoNet: High Speed Ultra Low-power Backscatter forNext Generation SensorsPengyu Zhang, Pan Hu, Vijay Pasikanti, Deepak GanesanSchool of Computer ScienceUniversity of Massachusetts, Amherst, MA 01003{pyzhang, panhu, vijaykp, dganesan}@cs.umass.eduABSTRACTThis paper argues for a clean-slate redesign of wireless sensor systems to take advantage of the extremely low powerconsumption of backscatter communication and emergingultra-low power sensor modalities. We make the case thatexisting sensing architectures incur substantial overhead fora variety of computational blocks between the sensor andRF front end — while these overheads were negligible onplatforms where communication was expensive, they becomethe bottleneck on backscatter-based systems and increasepower consumption while limiting throughput. We presenta radically new design that is minimalist, yet efficient, anddesigned to operate end-to-end at tens of µWs while enabling high-data rate backscatter at rates upwards of manyhundreds of Kbps. In addition, we demonstrate a complexreader-driven MAC layer that jointly considers energy, channel conditions, data utility, and platform constraints to enable network-wide throughput optimizations. We instantiate this architecture on a custom FPGA-based platformconnected to microphones, and show that the platform consumes 73 lower power and has 12.5 higher throughputthan existing backscatter-based sensing platforms.Categories and Subject DescriptorsC.2.1 [Computer-Communication Networks]:work Architecture and DesignNet-KeywordsArchitecture, Backscatter, Sensor, Wireless1.INTRODUCTIONA fundamental assumption that has driven the design ofsensor networks for decades is that communication is themost power-hungry component of an individual sensor system. The power consumption gap between communicationand other modules has driven a plethora of design choicesin sensor networks, primarily by encouraging designers toPermission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others thanACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permissionand/or a fee. Request permissions from permissions@acm.org.MobiCom’14, September 7-11, 2014, Maui, Hawaii, USA.Copyright 2014 ACM 978-1-4503-2783-1/14/09 . ble 1: Power consumption of accelerometer, audio, ecg, and image sensors.Accel [1] Audio [2] ECG [3] Camera [14]Power6µW15.3µW60µW0.7µWreduce data at the source, thereby minimizing the amountof data that needs to be communicated.We argue that this assumption does not hold when itcomes to passive radios such as backscatter. Backscatter requires extraordinarily simple circuitry since the carrier waveis generated by a reader, and a sensor only needs to modulate the signal to transmit information, thereby eschewingpower-hungry components of a typical active radio. Thesimplicity and inherent efficiency of backscatter means thatthe energy gap between communication and other components of a system has narrowed dramatically.These observations have profound implications on the design of next-generation wireless sensing systems that operateusing backscatter. The primary implication is that the bottleneck in terms of power consumption has shifted away fromcommunication to computation and sensing. But sensing isoften not the bottleneck as well — the past decade has seendramatic reductions in the power consumption of sensorssuch as microphones, cameras, ECG, accelerometers, andothers, many of which consume only µWs of power whilesampling at high rates (Table 1). Thus, both backscattercommunication and a variety of low-power sensors can operate at µWs of power, and the key question becomes one ofoptimizing the rest of the system to match these numbers.This requires that we re-think every component between thesensor and RF interface — data acquisition, data processing,buffering, packetizing, MAC, and many others now becomethe bottleneck for achieving ultra-low power operation.In this paper, we overturn the design principle governingwireless sensor design from one that is focused on minimizing communication to one focused on optimizing the computational elements between the sensor and RF interface.But optimizing computation is easier said than done, andrequires an understanding of every module of the sensingplatform, in-depth analysis of how to eliminate overheadfrom these modules, and design of a modified architectureto support an optimized design.But our efforts to optimize computation raises an unexpected problem. If we do nothing to reduce data at thesource, we need the bandwidth to be able to transfer rawdata from the sensor to infrastructure. While backscattercommunication is efficient in terms of power, throughputsachieved by practical backscatter-based systems have been

2.CASE FOR EkhoIn this section, we make the case that backscatter communication is extremely cheap and overturns the widely heldpremise that communication is more expensive than computation. We focus on the tradeoff between computation andcommunication since many commonly used sensors are already extremely efficient in terms of power. We begin witha discussion of why backscatter is efficient.ReaderTX antennacurrentreader-- tag SensorRX antennacurrent21.5Throughput (kbps)abysmal. Despite several efforts at improving throughputsof backscatter [13, 27, 8, 22, 12], the best case throughput is still only around 20 kbps even when only a singlenode is present, and drops dramatically to barely hundredsof bits/second when there are multiple devices sharing thenetwork. These numbers are not encouraging — for example, a microphone sampled at 8-44 KHz requires transmitrates upwards of 704 kbps, a far cry from the throughputthat backscatter platforms are able to support today.This leads us to the central question that we address inthis paper: how can we design a backscatter-based wirelesssensor system that achieves whole-system power consumption of µWs, while simultaneously increasing data rates tosupport raw data transfer from sensors at several hundreds ofkilobits/second. Our goal is aggressive — as a point of comparison, an existing backscatter-based sensor, the UMassMoo (or the UW WISP) consumes about 2mW of powerwhile transmitting at a few kilobits/second when there aremultiple devices present. Thus, we seek to drop the systemwide power consumption by more than two orders of magnitude while simultaneously enabling two orders of magnitudeincrease in the data rates.Our contributions are two-fold. First, we present a novelbackscatter-based sensor platform, Ekho, that achieves ourdesign goal to optimize power by eliminating computationaloverhead from the sensor to RF pipeline. We start with adeep dive at what computational modules are present between the sensor and RF interface on a typical low-powersensor platform, and measure their power consumption, before launching into a minimalist design that is optimizedfor power. Our second contribution is a network stack,EkhoNet, that is designed to be minimalist and enable bandwidth scale up to support data rates of hundreds of Kbpswhile supporting tens of nodes. While each Ekho node isminimalist, our MAC layer leverages resources at the readerto enable utility-energy and channel-aware optimization ofbit rates and slot sizes across nodes.Our results on a USRP reader and Ekho nodes show that:I For operating an accelerometer at 400Hz, Ekho consumes 35µW of power, 7.6 lower than the 266µW ofthe Moo and 3.3 lower than the 118µW of WISP5.0.For operating an audio sensor at 44kHz, Ekho consumes 37µW of power, 76 lower than the Moo and13.5 lower than the WISP5.0.I We show that EkhoNet can scale to a network of several high bandwidth sensors. When a network of tenEkho nodes equipped with microphones transmit simultaneously to a reader, we achieve a throughput of780 kbps as a result of interleaving the data streams atthe MAC layer. We also use an energy-utility-channelaware scheduler, and show that over 50% of the audiosensors achieve a median MOS score larger than 2, significantly higher than a baseline scheme that assignssampling rates evenly across all nodes.10.50-0.5-1-1.5-205101520Time (seconds)reader --tag2530transistor openbackscatteredsignaltransistor closeFigure 1: Backscatter communication basics.2.1Backscatter radio RF front endBackscatter radios are designed to enable ultra low powerwireless communication. As shown in Figure 1, a readerprovides a carrier wave, which can be modulated with information to enable ultra low power wireless communication.While the carrier wave can also be rectified by a sensor forenergy harvesting, our focus in this paper is on backscatteras a low-power radio, whether energy is obtained via harvesting or a battery, hence we focus on the communicationrather than harvesting aspects of backscatter.To transmit data, a sensor toggles the state of a transistorto detune its antenna and reflect the carrier wave back tothe reader with its own information bits. Because the sensordoes not actively generate RF signal as active radio systems,the power consumption of the backscatter radio is very low.In addition, the on-off transition overhead of backscatter radios is very short because backscatter radios do not have towarm up the RF analog circuits for data transmission unlikeactive radio systems. As a result, there is little overhead incurred while transmitting via backscatter, even when transmitting at a high rate. For example, one key component ofthe backscatter analog RF front end of the WISP [6] is aMOSFET transistor (BF1212WR). Its power consumptionfollows the equation of CV 2 F where C is the capacitance ofthe transistor, V is the digital drain-source voltage, and Fis the frequency of operating the transistor. When this transistor is toggled at a slow rate of 10Hz, it consumes 55pWof power, and even when toggled at a high rate of 1MHz,it only consumes 5.5µW of power. Thus, backscatter radiosconsume of the order of µWs of power, even for high ratedata transfer.2.2Why compute if its cheaper to transmit?The power consumption of backscatter radio has surprising implications on sensor system design, and challengeslong-held views about communication vs computation tradeoffs in these systems.Computation vs Communication: A common assumption in designing sensor systems has been that computationis significantly cheaper than communication, often by manyorders of magnitude. This view has shaped a plethora ofefforts for in-network processing, signal compression, subsampling, and other such approaches to reduce data at thesource prior to communication. Indeed, this tradeoff hasbeen reinforced by performance/power trends over the pastdecade — power consumption of embedded processors have

system is minimal asumption of bapbackscatter-basedmore efficient tobottleneck.just raisesconnectThisan theintFigure 2: 1 bit adder and 1 bit shift register circuit.woulditbebet(b) 1 bit shift register circuits forsumptionofbackscA"circuitry that even matching the efficiency of backscattersensed directlybackscatter. ofbecomes a challenging architectural design problem.more efficient to eliWe take a mesource prior to communication.Indeed,this tradeo has justThus, the cruxof our argumentis the following:backscatconnect the the past would ittheseFigure2: 1 bitbyadderand 1 bit cation,suchthatcommunicationofrawC "blocks betweendecade — power consumptionprocessorsinhavedata may ofbe embeddedpreferable to computationa wider sensedspectrumdirectly throC"backscatter-basdropped dramatically, whilepowerscenarios.reduction in active radiosof real-worldWetake theya measusourcepriortocommunication.Indeed,thistradeo haspowerconbeen relatively slower.(a) 1 bit adder circuits forhascomputation.Implications on architecture design:This asedsensorprocessorsplatform. Traditionalplat-betweenblocksdecadeof embeddedhave sensinganddesign asenrD"Q — power consumptionforms add a lot of computational modules between the owerreductioninactiveradiosformthataddreHowever, backscatterandcommunicationchallengesthis longthe radio for sensordata acquisition,processing, filterpowertheyconsuming,buffering, etc.The contribution ofthese components tohas beenslower.heldview.relativelyBackscatteris inherentlyextraordinarilyefficientoverall power consumption of an active radio-basedof sensorthroughput. SeQ the carrier wave issystemsincegeneratedby thereader,andthetag However,3.on INVESTisminimalandcanlargelybeignored.E"and designa radicQ backscatters the signalonlywithout platforms,any additionalampli- becomebackscatter-basedthese ation.Thus,each bitcommunicationof backscatterchallengesis extremelysimple,However,backscatterthis long1"bit"shi*"register"Backsca2er"RF"This raises an intriguing question — with the power conIn this rilyefficient(b) 1 bit shift register controls a backscatter transumption of backscatter being so low, would it in fact besistor.basedthatbe cheapermoreefficienteliminateall andof thethesemodulesand platformssinceforthecomputationcarrier wave istogeneratedbytoenergy-wise,thereader,thecomputag en-bloc,3.INVESTIGsumptionof µWjust connectthe sensordirectlythe radio?In other words,tationaloperationson eachbitwouldhaveto usetofewergatesonly e 2: 1 bit adder and 1 bit shiftcircuit.the signalwould it be better to just stream every bit of data hebit.This is often afication.eachtobitcommunicateof backscatteris extremelysenseddirectly throughthe radio? simple,talldue to athesimplicityofbackscatter.Weoftakea measurement-drivenIncommunicationthis section, wand orderonly requireshandfulgates(Figure 2). Thisapproachimplies towards answerTo platformsempiricallying thesequestions.First, we lookat the noperationbasedareuthat forincomputationto blocksbe cheaperenergy-wise,compubetweensensing and thethe RFinterface on existingdropped dramatically, while power reductionactiveradiosMoo/UWWISPthatsumsten sensorbeforehavetransmittingtheaggre- derstand howmuch of µWs fotationaloperationsonreadingseachbit wouldto usefewerhas been relatively slower.sors, a rmsHowever, backscatter communicationchallengesthislonginvestigatewhy thethan that required to communicatethebit. Thisis oftenoura empirical studyradio.of throughput.Second,we build on powerdata reductionhave directand significantheld view. Backscatter is inherently suchextraordinarilyefficient wouldcommunication,patallorder anddue totagthe simplicityof abackscatter.and designradically new backscatter-based sensorplatsince the carrier wave is generated bybenefitsthe gforma thataddressesthese limitations.To empiricallyundeonly backscatters the signal without ationoperation3.1 Poorengregationschemefication. Thus, each bit of backscatteris extremelysimple,cuts this cost by a factor of ten. The ansmittingtheaggreWe start witoperationon impliesa backscatter-basedplatform has EXISTINGdubious benand only requires a handful of gates (Figure2). This3. theradio.Ontraditionalsensorplatforms,that for computation to be cheaper energy-wise,thecomputhreekey compefits. Figure 2 shows that thenumberARCHITECTURESof gates required forSENSINGtational operations on each bit wouldsuchhave touse fewergatesdatareductionhave directbutand significantpowerare radio.ure 3): 1) the sebitsa iswouldroughlyfour whygatesIn thisnine,section, we onlyinvestigatecurrent backscatterthan that required to communicate summingthe bit. Thistwois rcon- the protocodlesneeded to transmit the same data via the shift register cirtall order due to the simplicity of dtransfer.WealsoPoor enerConsider, for example, a simplecuitaggregationoperationgregationscheme cutsAsthispowercost bya factor of ten.The samesubsystemon aof backscatter!consumptionisproportionalinvestigate why they are unable to achieve high-data ratethat sums ten sensor readings before transmitting the formhasdubiousbenprocessed(ifneto numberof gates, a ninegate adderconsumes2.2 moreat low power.communication,particularlywhileoperatinggate value over the radio. On traditionalsensor platforms,threekey of circuit.gatestheserequirednetworkstack,To empiricallyunderstandfactors, forwe look at theUMasspowerthan thea fourthatgatesuch data reduction would have directand withsenbenefits since communication ofsummingtwo bitstoisaddroughlybuttoonlygates areIt is necessarya fewnine,caveatsourfoursimplifiedcomsors, a low-power MCU (MSP 430 family) and a backscattergregation scheme cuts this cost by a factor of ten. Theparisonof computationand communication.clockcirrates dles the protocols foneededtotransmitdata via the shiftTheregisterradio.same operation on a backscatter-basedplatformhas dubi- the same3.1.1 onSensoa micare limited bysignal to noise subsystemous benefits. Figure 2 shows that ofthecommunicationnumberof NAND subsystemscuitof backscatter!As power consumptionis proportional3.1Poorenergyefficiencygates required for summing two stheclockrates ofprocessorsSensordata aprocessed(if neededtonumberofgates,aninegateadderconsumes2.2 moreWe start with a break down of the power consumed bysix transistors), but only four NAND gates (sixteen tranbe thanhigher,therebyreducepower.Inaddition,low- Moosomesensorsandhanetworkpowerthe andathefourgatecircuit.threebackscatterkey computationalmoduleson a UMass(Fig- stack,sistors) and an additional transistorcanfor anvia a protoco3):1) thetosensordatasimplifiedacquisitionsubsystem whichsignal are needed to transmit the samevia the shiftcombinationof hardItdatais les signalthe protocolsfor operatingsensors, di er2) the data handlingregister controlled backscatter RF! tionAs ts,justprovideanparisonof computationandcommunication.The clocksubsystemon a micro-controllerwhereratessensor data is stored,is proportional to number of transistors,a ninegate adderentpower domains,extremelytightduty-cycling,andso on.micro-controller3.1.1processed(if limitedneeded),formattedand senttothe Sensor dacommunicationarebysignalintoto packet,noiseconsumes 2.1 more power than the timizations,thecardsarestackedagainstoperationsare ckscatter RF.ratio considerations, whereas

EkhoNet: High Speed Ultra Low-power Backscatter for Next Generation Sensors Pengyu Zhang, Pan Hu, Vijay Pasikanti, Deepak Ganesan School of Computer Science University of Massachusetts, Amherst, MA 01003 {pyzhang, panhu, vijaykp, dganesan}@cs.umass.edu ABSTRACT This paper argues for a clean-slate redesign of wireless sen-

Related Documents:

behringer ultra-curve pro dsp 24 a/d- d/a dsp ultra-curve pro ultra- curve pro 1.1 behringer ultra-curve pro 24 ad/da 24 dsp ultra-curve pro dsp8024 smd (surface mounted device) iso9000 ultra-curve pro 1.2 ultra-curve pro ultra-curve pro 19 2u 10 ultra-curve pro ultra-curve pro iec . 7 ultra-curve pro dsp8024 .

47 117493 SCREW, mach, hex washer hd 2 48 BOX, control 276868 Ultra 395/495 1 15D313 Ultra 595 1 49 CONTROL, board, 110V 1 246379 Ultra 395/495 1 248179 Ultra 595 1 50 276882 COVER, control 1 51 15K393 LABEL, control, Graco 1 56 CORD, power 1 15J743 Ultra 395/495, Stand 1 15D029 Ultra 595, L

TABLE OF CONTENTS: XPRESS ULTRA FIELD TERMINATION CONNECTORS SC Xpress Ultra Fiber Connectors Page 2 LC Xpress Ultra Fiber Connectors Page 3 ST Xpress Ultra Fiber Connectors Page 3 CLEANING & INSPECTION Optical Connector Contamination Pages 5 - 6 Xpress Ultra Cleaner Pages 7 - 8 Xpress Ultra Cleaner Replaceable Cartridge Page 8 Xpress Ultra

Table 3 summarizes these limitations and their implications on the behavior of the STM32L1xxxx ultra-low-power devices. Table 3. Cortex 1.1 Cortex -M3 limitation description for the STM32L1xxxx ultra-low-power devices Only the limitations described below have an impact, even though minor, on the implementation of STM32L1xxxx ultra-low-power .

General Purpose Ultra-low freezer with factory installed chart recorder and two shelves of sliding drawer racks 6 Cardinal Health Ultra-Low Freezers To order, call: 800.964.5227 Options for Ultra-Low Temperature Freezers

A: Yes, High-Speed option may reduce the performance of other network service, but High-Speed option controls QoS(Quality of Service) and protects other important services. Q3: Is every file transfer speed improved with High-Speed option? A: High-Speed option improves only for users who have High-Speed option.

Low Speed and High Speed Typically, a distinction is made between high-speed CAN transceivers and low-speed CAN transceivers. High-speed CAN transceivers support data rates up to 1 Mbit/s. Low-speed CAN transceivers only support data rates up to 125 kbit/s. ISO 11898-2 assigns logical "1" to a typical differential voltage of 0 Volt.

Meekings 2 Contents 3 Declaration 4 Abstract 6 A Mist that Rises from the Sea 215 How A Personal History is Constructed: An Annotated Index of the Past 215 A. Aberdeen Bestiary 226 B. Beginnings 233 C. Chronophobia 237 H. Happiness