research article a low-power microcontroller with accuracy

Research ArticleA Low-Power Microcontroller withAccuracy-Controlled Event-Driven Signal ProcessingUnit for Rare-Event Activity-Sensing IoT Devices

Daejin Park1 Jonghee M Youn2 and Jeonghun Cho1

1School of Electronics Engineering Kyungpook National University 80 Daehakro Bukgu Daegu 702-701 Republic of Korea2Department of Computer Engineering Yeungnam University 280 Daehak-Ro Gyeongsan Gyeongbuk 712-749 Republic of Korea

Correspondence should be addressed to Jonghee M Youn younyuackr and Jeonghun Cho jchoeeknuackr

Received 8 January 2015 Accepted 23 March 2015

Academic Editor Young-Sik Jeong

Copyright copy 2015 Daejin Park et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

A specially designed microcontroller with event-driven sensor data processing unit (EPU) is proposed to provide energy-efficientsensor data acquisition for Internet of Things (IoT) devices in rare-event human activity sensing applications Rare-event sensingapplications using a remotely installed IoT sensor device have a property of very long event-to-event distance so that the inaccuratesensor data processing in a certain range of accuracy error is enough to extract appropriate events from the collected sensing dataThe proposed signal-to-event converter (S2E) as a preprocessor of the conventional sensor interface extracts a set of atomic eventswith the specific features of interest and performs an early evaluation for the featured points of the incoming sensor signal Theconventional sensor data processing such as DSPs or software-driven algorithm to classify the meaningful event from the collectedsensor data could be accomplished by the proposed event processing unit (EPU) The proposed microcontroller architectureenables an energy efficient signal processing for rare-event sensing applicationsThe implemented system-on-chip (SoC) includingthe proposed building blocks is fabricated with additional 7500 NAND gates and 1-KB SRAM tracer in 018 um CMOS processconsuming only 20 compared to the conventional sensor data processing method for human hand-gesture detection

1 Introduction

Nowadays sensor devices with wireless connectivity such asWi-Fi Bluetooth and ZigBee are becoming important in IoTapplications Human activity monitoring based on sensedsignal analysis [1 2] is becoming a popular application usingbiomedical computing technology Body-attached sensorsmeasure the electrical signal converted by a biomedicalinterface [3] Observation-based remote sensors are also usedto monitor the movement of the human body [4]

As described in Figure 1 the remotely installed smart sen-sor systemmonitors the environment periodically and trans-fers the collected sensor data via wireless interface to the hostsystem such as aWeb serverThese wireless sensor devices arepowered by the on-board battery with limited capacity

Battery recharging and replacement are very inconve-nient procedures and a major obstacle in extending variousIoT-based applications Long operating lifetime of the sensor

systems is therefore an important requirement in designingthe system architecture and the sensor data processing algo-rithm

The energy consumption in the sensor systems is causedby sampling the sensor signal processing the sampled dataand transferring the collected data to the host machine viawireless interface Traditional sensor devices sample the sig-nal periodically and analyze the sensed data

This approach has advantages requiring the simple pro-cessing unit in terms of hardware resource and the general-ized algorithm in terms of software development Syntacticactivation of the entire sensor system results in more powerconsumption especially in wireless connectivity

The conventional sensing platform for human activitymonitoring uses general purpose microcontrollers (MCUs)including an analog sensor interface discrete-time analog-to-digital converter (ADC) as a data sampler and a sensor dataprocessing unit to analyze collected data

Hindawi Publishing CorporationJournal of SensorsVolume 2015 Article ID 809201 10 pageshttpdxdoiorg1011552015809201

2 Journal of Sensors

Sens

or

Sampler Processor

Wire

less

IF

Change of stateInterested

state change

Recognized event

Monitoring Notify

Electrical signal

Sampled signal

Recognized event

(i) Event arrival time delayskip allowable

(ii) Event presence is more important than data itself

ObserverGenerator

Intended event

Battery-operated sensor-driven IoT device

Observer

rarr Long-term operation is more important than accurate operation

(i) Low frequency event generation

(ii) Secondary retry allowable (average)

(iii) Event type is predefined (signal shape is expected)

Generator

Triggered event

Figure 1 Sensor-based IoT device concept with sensor interface processing unit and wireless connectivity

However this approach by which the sensed data isanalyzed on the microlevel of data-to-data has operatingpower overhead because it is not optimized to consider thelong event-to-event distance of human activity signals whichis easily observed in rare-event sensing applications [4 5]

In this paper for efficient sensor data processing in theenergy consumption a semantic sampling method is intro-duced to capture the signal with the features of interest and isimplemented as a preprocessor unit named to signal-to-eventconverter (S2E) which generates the atomic events insteadof the sampled data itself The extracted atomic events are arelatively small number of samples compared to the syntac-tic sample data by conventional analog-to-digital converter(ADC) as a signal-to-data converter (S2D)

The proposed S2E replaces the conventional S2D to ext-ract atomic events from the incoming sensor signalThe eventidentification from human activity monitoring is performedby the event-driven sensor data processing unit (EPU) for thesmall set of extracted atomic events

This paper is organized as follows In Sections 2 and3 our research motivation and related work are discussedSection 4 describes the details of the proposed architectureThe implementation and experimental results are presentedin Section 5 Finally we conclude the paper by summarizingour contributions in Section 6

2 Motivation

The key motivation of the proposed method begins with thetransition to macrolevel processing of the sensor signal byS2E instead of the conventional microlevel analysis for thesensed dataset Figure 2 shows the method we used to repre-sent the sensor signal acquisition result in the event dataspace The human activity sensor signals are represented bythe attributes of interest and the elapsed time relationshipbetween the atomic events in Figure 2(a) The monitoredhuman activity is encoded as a set of atomic events by theevent quantization in Figure 2(b)

To address this limitation of the conventional digitalsystem architecture by using the discrete time-based sensordata processing method we propose an event-driven systemarchitecture that modifies traditional digital system designWe present a theoretical framework to implement an event-driven sensor processor for general rare-event sensing appli-cations by analyzing the system operations

21 Event-Space Signal Representation Our main researchbegins with an event-space representation of the signalinstead of the digital data space domain The extracted fea-tures of the sensed signal are encoded into the elapsed timebetween events and informative value such as voltage leveland edge phase crossing the trigger point of the signal Thefundamental event defined which is defined as an atomicevent with the most important information provides a signalrepresentation on an abstract level and reduces the com-putational complexity in performing basic data processingfor extracted informative features of interest The collectedatomic events include partial information in the originalsignal that specifies whether the desired featured points of thesignal are present

22 Accuracy-Controlled Event Quantization The event-quantization concept extends the time-quantization methodfor signal representation that uses elapsed time to enhance theconventional data-sampling and processing method Timequantization monitors only the specific conditions of thesignal transition and captures the time-stamps The event-quantization method also determines whether the specifiedcharacteristics of the signal exist

23 Event-Driven Sensor Data Processing The event-basedapproach with a certain amount of accuracy error is des-cribed by the proposed event-driven sensor data processingflowThe input signal is monitored with specified interest-of-signal characteristics to generate the specific atomic events ofthe signalThe set of atomic events during the specified region

Journal of Sensors 3

Activity raw dataTarget Event quantization

Human activity signal acquisition and event quantization

attribute and time of interestrepresentation

Distances(t)

t

s(t)

(1) thk attribute of interest(2) Φ edge phase(3) eti elapsed time

thc

thb

tha

aev 1

aev 2

aev 3

aev 4

aev 5

aev 0

AEV997888997888997888rarr= aevi | aevi = ⟨(thk Φ) eti⟩

(a) Attribute and its corresponding elapsed time representa-tion

tSens

or si

gnal

sp

ace

Even

t dat

a spa

ce

t

et0

= ⟨Ea et0⟩

et2

= ⟨Ea et2⟩

et1

= ⟨Eb et1⟩

et3

= ⟨Eb et3⟩

et4

= ⟨Ec et4⟩

et5

= ⟨Ed et5⟩

Ea

Eb

Ec

Ed

aev1

aev2

aev3

aev4

aev5

aev0

(b) An example of event-space representation for incoming sensor signal

Figure 2 Human activity sampled data representation in event data space

of the signal are traced into the tracer memory as an eventvector which contains the sequence of the atomic events andthe time-distance relationship between the atomic eventsThetraced event vector identifies the approximate result as a finalevent by comparing it to the expected rules of the atomicevents

These approximation approaches enable us to reduce thecomputational complexity in order to manipulate a largeamount of collected sensing data As a result power con-sumption will be reduced For applications related to humaninteraction an approximation approach enables developersto design the computational block using smaller hardwareresources while providing sufficient performance in limitedresolution of the accuracy

In accuracy-controlling approaches defined from thespecifications our study focused on the data-representationresolution the timing-resolution of the sampling frequencyand the response time as a delay time [6] This enables theconfiguration of the operation accuracy in the processorarchitecture level according to the abstraction level of the pro-posed event-quantization approach

3 Related Work

To overcome the weakness of inefficient power consumptionby the frequent CPU wake-up for the discrete time samplingcontinuous-time signal processing techniques [7 8] havebeen proposed in previous literature If a certain conditionof the signal status such as the voltage level at a specific timeis matched with the user-defined condition [9 10] the timevalue at the triggered condition is sampled and quantized[5] by the selective method which also helps to reduceoperational power [11]

The continuous-time samplingmethodwas introduced toimprove the syntactic sampling and processing approach in

terms of power consumption but it requires additional hard-ware resources and more computational time for the time-distance calculation which gives rise to additional powerconsumption The required power and hardware resourceoverhead which are needed to compensate for reducedwake-up power consumption must be considered in order toachieve benefits in total energy efficiency due to hardware-energy trade-off

The trade-off in terms of energy and accuracy has beenstudied widely [12 13] To obtain long lifetime operationsunder limited battery power [14] the latest research intro-duces inaccurate computation techniques [15 16] with appro-ximation-based hardware designs

The proposed sensor processor for the rare-event sensingapplications adopts the event-driven approach of the conti-nuous-time sampling method Inaccurate time-data manip-ulation reduces computational complexity and sampling res-olution by determining the presence of featured events in thespecific range The event-detection accuracy can be adjustedby making the trade-off between the processing energy con-sumption and the operating specification

4 Proposed Architecture

41 Application-Dependent Constraints Figure 4 shows thedifference between the discrete time samples and the featuredevents of interest with the common shape of the rare-eventsensor signal Event sources such as hand gestures proximityand object activity generate signal pulses for which thedistance between featured points of the signal is very longThe number of data samples (119899) is greater than the numberof events (119898) In this work we assumed an application-specific constraint of rare-event characteristics which resultin a small number of events compared to the number of datasamples


s(t) Data tracing Digital data

Signal features of interest

Data processingData sequenceS2D

(ADC)Final event

Expected data sequence

di di

evj

(a) Data sampling and lazy evaluation for syntactic data processing

Signal feature

Sensor event (w error)

S2E (AEG)

s(t) Event tracing

Signal feature of interest

Atomic eventprocessing

(EPU)Atomic event sequence

Expected atomic event sequence

Identified final eventEvent data

Error bound is allowable

ev

Δeaevi aevi

(b) Event sampling by early evaluation and event processing

Figure 3 Event sampling based on signal-to-event (S2E) and event-driven data processing

Time

Time

s(t)

s(t)

s(t)

Long-term no activity

TimeLong-term no activity

Time

n gt m n asymp mthi = Δ lowast i

thi = Δ lowast i

thi = Δ lowast i

Li = Δ lowast i

tk = Δt lowast k dmktm = Δt lowast m

em998400 m

tm998400 = Δt lowast m998400

Rare-event applications dmk ≫ em998400 m

n ≫ m rare event applications

Figure 4 Wake-up frequency for data sampling and event-drivensampling

The event-quantization accuracy depending on the res-olution of the elapsed time-stamp is described as 119890

1198981015840119898

inFigure 4 The rare-event sensing applications in which theevent-to-event duration is relatively larger than the accuracyerror have the following application-specific constraints

119889119898119896

≫ 1198901198981015840119898

(1)

With these application-specific constraints in (1) theevent identification accuracy error caused by the inaccuratetime-stamp measurement clock is relatively insensitive Therecognized event observer such as the human eye allows fora certain amount of inaccuracy in identifying the meaning of

the events which are constructed by the proposed inaccurateevent-driven sensor processor

The proposed sensor processor is designed with theseapplication-specific constraints by reducing the accuracy ofthe time-stamp measurement clock decreasing the bit widthof the timer block to capture the time-stamps and decreasingthe operational complexity of the time-to-time distancemea-surement blocks which are specially implemented as a ded-icated accelerator for event recognition in the implemen-ted hardware

42 Atomic Event Quantization The conventionalMCU per-forms data sampling in the ADC unit data tracing in buffermemory and digital data processing to identify the originalevent generated by event sources such as a swipe gestureThesyntactic sampling is performed without the consideration ofthe incoming signal propertyThen the lazy evaluation usingthe features of interest is performed to generate the final eventev119894using a large number of sampled data 119889

119894 This syntactic

data sampling and lazy evaluation in conventional MCU isillustrated in Figure 3(a)

The proposed EPU can perform the event relationshipanalysis with a reduced computation overhead for the smallerset of atomic events The signal abstraction by extractingatomic events as signal shape in S2E leads to accuracy errorin identifying the final event The overall procedure of theevent-driven processing in the modified MCU is describedin Figure 3(b)

The event-driven signal sampling in the proposed archi-tecture captures the signal shapes of interest using the featurescanning window which determines the presence of theexpected features of the signal The feature scanning windowin Figure 5(a) is configured to capture the specific signal


Accuracy-controlled (approximation)

Lf

TrTstart Tend

Configuration of feature scanning window

Dmax

Ωi = (Lf Tr Dmax Tstart Tend)

(a) Configuration of signal scanning window foratomic event extraction

AEGSensor analog signal s(t)

Atomic event

f Ω rarr E

Ω = Ωi set of signal segmentsE = set of atomic eventsaevi

aevi

(b) Atomic event generator definition

Timer window

Timer Timer endRr Rr RrRf

Rf RfRf gtDmax

Rf gtDmax

Dmax

Rr gtDmax Rr gtDmax

Rf gtDmaxRr gtDmax

infininfin

Lprobe Lprobe LprobeLprobe

Lprobe Lprobe

Time measurementwindow

start

Step-up (Rsu) Step-down (Rsd) Down-pulse (Rdp) infin-step-up (Risu) infin-step-down (Risd)Up-pulse (Rup)

(c) Examples of set of signal segmentΩ

Elapsed-time sweep of feature point

Swee

p of

feat

ured

leve

l

Equivalent to discrete timed sampling

Signal segmentation

Lr1

Lr1

Lr1

Lr1

Lr1

Lr1

Tr1

Tr1

Tr1

Tr1Tr2

Tr2 Tr3

Tr3

Tstart TstartTend Tend

middot middot middot

⋱

(d) Representing various atomic events according to featured points

Figure 5 Atomic event generator (AEG) based on feature scanning window and signal segmentation

shapeThis configuration is represented with the set of signalsegments Ω in Figure 5(b)

The S2E includes the atomic event generator (AEG) unitto generate a set of atomic events by using the user-definedset of signal segmentsΩ Examples of the user-defined signalsegments Ω are introduced in Figure 5(c)

Figure 7 describes our approach to capture the signalshape as an atomic event crossing a certain range of arrivaltime To more formally define our approach we begin ourexplanation by first presenting the following definitions

Definition 1 Given continuous signal 119904(119905) let AEV = aev119894|

aev119894= (aev

119894minus1 value phase et) be a sequence of an atomic

event aev119894crossing the specific level and time condition with

a relationship of previous atomic event aev119894minus1

where aev119894sdot

value is a result of the approximation-based data quantization

function ADQ and aev119894sdot et is a result of the approximation-

based time quantization function ATQ described as follows

119889119896= ADQ (119904 (119905

119896) 119871119898 Δ119889 119906 V)

forallΔ119889lowast 119906 lt

1003816100381610038161003816119871119898

minus 119889119896

1003816100381610038161003816lt Δ119889lowast V

119905119896= ATQ (aev

119894sdot et119879clk) where

119879clk = 119879clk + Δ119905

(2)

The meet condition 119896 when the expected crossing ispresent is described in the following equation

119905119896= et +

119879clk lowast 119896 forallDQ (119904 (119905119896)) =

119889119896 (3)

As shown in Figures 5(a) and 5(b) the AEG builds an ele-mentwith the attributes which are encodedwith the digitized


Begin with initial configuration

Adjusting sampling frequency

Signal feature determination

Sample frequencytrigger threshold

level configuration

Sampled data signalprocessing

Itera

tive c

onfig

urat

ion

of sa

mpl

ing

spee

d th

resh

old

leve

l

Adjusting reasonable sampleand processing method

s(t) incoming sensor data signal 1

2

(a) Syntactic procedure to determine conventional data sampling frequency

Signal shape segmentation

Signal segment region signal-to-event set

configuration (features)

Sampled event data processing

Gro

upin

gun

grou

ping

of

sign

al se

gmen

ts

Signal segment selection (S2E configuration)

Begin from fundamental atomic event set1

Extend the event sampling window by grouping the adjacent event

2

Ω0

Ω0

Ω1 Ω1Ω2 Ω2 Ω3

Ω3Ω4Ω4

Lf0

Lf0Lf0

Lf1Dr0

Dr0

Dr1

Dr1

Tr0

Tr0

Tr1

Tr1Tstart0

Tstart0 Tstart1

Tstart1

Tend0

Tend0

Tend1

Tend1

ER0

ER0

ER1

ER1

ER2

ER2

ER3

ER3

ER4 ER5

(b) Iterative procedure to determine appropriate event segmentation set

Figure 6 Iterative procedure to determine sampling method and signal segments for the sensor signal

signal level elapsed time and edge phase in the followingequation

AEG (119904 (119905) 119871) = aev119894| aev119894= ⟨aev

119894minus1119889119896 120601edge 119905119896⟩ (4)

From (4) the extracted information as an atomic eventis encoded with the approximation value of the signal levelthe reduced time quantization value of the elapsed time andthe relationship of the previous atomic event aev

119896minus1

43 Atomic Event Extraction The event-quantized signalrepresentation is dependent on the event slice resolution ofthe configured set of signal segments which is described inFigure 5(d) The number of feature points and the windowsize determines the accuracy of the signal representation bythe extracted atomic events Figure 5(d) shows the capabilityto represent various signal shapes with the configuration of119871119903 119863max 119879119903 119879start 119879end in the feature scan window

Definition 2 Given the configured feature scanning windowto extract the atomic events from 119904(119905) let 119879start be a start timemonitoring the signal let 119879end be the end of monitoring thesignal let 119871

119903be a rising signal level at which the time-stamp

is 119879119903 let 119871

119891be a falling signal level at which the time-stamp

is 119879119891 let the pair of 119871

119909and 119879

119910be featured point and let119863max

be a maximum time value in which the featured points are

present The set of signal segments described by the config-uration Ω = Ω

119894| Ω119894

= (119879start 119879end 119871119903 119871119891 119879119903 119879119891 119863max)of the featured scanning window is defined as Ω and theyare used to extract the atomic events of interest for the AEGfunction which is defined as follows

aev119894 = AEG (119904 (119905) Ω) (5)

Ωup defines a signal segment of the feature scanning win-dow with the ldquoup-pulserdquo type in the first part of Figure 5(c)In our applications Ωtype | type = ldquouprdquo ldquosurdquo ldquosdrdquo ldquodprdquo ldquoisurdquoldquoisdrdquo is used

One signal shape can be divided into several slices byuser-defined signal segmentation If the time window forsignal segmentation is the same as the fixed width 119905

119904in the

discrete time sampling method the result of the atomic eventgeneration is equivalent to that of the discrete time samplingThe proposed atomic event generation approach enables atrade-off between the signal extraction accuracy and its pro-cessing power consumption

The application-specific constraints in configuring theset of signal segments must be considered for the accuracy-energy trade-off to provide reasonable accuracy of eventidentification with limited energy consumption Figure 6shows the determination procedure of the signal segmentsto represent the sensed signal with small set of signal seg-ments A reasonable slice of the signal segmentation can be


Time

Mag

nitu

de

Event capture approach by determining the presence of nextexpected atomic event in error range (active + passive)

Event quantizationElapsed time et

sample2nd detail

1st wait

tk = et + Tclk lowast k

= ⟨ minus1 dk Φedge tk⟩aeviminus1 aevi aevi

Δd middot u Δd middot lt |Lm minus dk| lt

Figure 7 Event sample by capturing the specific features of interestand elapsed time

determined by the iterative configuration to provide enoughevent identification performance with reasonable energyconsumption

Figure 6(a) shows an example of searching reasonablesampling frequency The red colored sample can be obtainedby adjusting the sampling frequency after the specified activ-ity signal is analyzed Figure 6(b) describes the procedure ofgrouping a set of signal segments into another signal segmentwhich can represent the activity signal with a smaller numberof atomic events

44 Event-Driven Sensor Data Processing TheAEG scans thecontinuous signal 119904(119905) passing through the configured featurescan window to determine the presence of the signal shapesof interest as shown in Figure 8 The set of atomic events isgenerated with a pair of attributes and time-stamps as a resultof the time quantization shown in Figure 9

aev = aev119894| aev0 aev1 aev

119894= (ldquo119871

119894rdquo 119905119904119894) (6)

The signal representation by a set of atomic events with acertain amount of error is denoted in the following equation

ae = ae119894| ae0 ae1 ae

119894= (ldquo119871

119894rdquo 119905119904119894plusmn Δ) (7)

aev119894 which is matched with the configured scan window

AE119894 is represented as an abstracted atomic event index in

Figure 9 which indirectly addresses the detailed attributesin the constant dictionary The continuous analog signal isconverted into a set of event quantized data aev

119894 and its

index value is traced only into the atomic event tracer bufferTherefore the traced event data processing manipulates theindex value and its relationship to the representative atomicevents to generate the final event EV The proposed EPUwhich is based on event quantization provides the followingadvantages compared to conventional sensor data processing

45 Event Bus Architecture Themodified architecture of theproposed MCU includes S2E to extract atomic event aev

119894

from the activity signal instead of using ADC event tracingto archive the atomic events aev

119894 and the EPU to analyze

the relationship between the archived atomic events

The sensor signal in rare-event applications is describedwith an example in Figure 2(a) which is represented with thethreshold level edge phase type and elapsed time betweenthe previously recognized signal points The signal featuresof interest are used early to extract the atomic events in theS2E unit

The path from S2E to the event tracer is designed withthe event bus on which the atomic event transactions areloaded The predefined event types are configured in EPUconfiguration by the user knowing the signal characteristicsfor which attributes are represented The EPU handles theindex to the events in the event table which is stored in theEPU configuration Figure 9 shows data flow of the event-quantized atomic eventsThe atomic event aev

119894only contains

a pointer to address the detailed attributes in the attributetable and elapsed time table to save the limited tracermemoryarea

5 Implementation and Experimental Results

Figure 10(a) is the data path of the implemented S2E cir-cuit The proposed S2E-based signal conversion and eventsignal processing architecture requires additional hardwareoverhead including a level comparator AEG timer tracermemory and EPU which are distinguished with a red dottedline in Figure 10(b)

The hardware implementation based on the proposedconcept requires the additional 7500 NAND gates and 1 KBSRAM tracer in 018 um CMOS process The implementeddesigns are integrated in an 8051-based microcontrollerFigure 10(b) shows themodified event-bus architecture of theimplementedMCU inwhich the atomic event (aev

119894) is loaded

from S2E The attributes of the user-defined atomic eventincluding signal features and elapsed time ranges are storedas a constant table in the on-chip flash memory

For power consumption measurement the raw dump ofthe electrical signal generated by hand gesture is gathered intothe host computer as shown in Figure 10(c)The input stimu-lus of the activity signal is loaded into the circuit-level simula-tion environment inwhich the accuracy-energy trade-off canbe easily performed to evaluate the energy consumption ofthe proposed MCU architecture

Figure 10(d) shows energy consumption reductionaccording to the accuracy by configuring the S2E for specificsignal segments Using a timer and oscillator unit with 10accuracy error in the swipe-gesture recognition applicationthe implemented MCU could still identify the gesture eventalthough consuming only 20 energy compared to the resultof the accurate discrete time sampling method

The elapsed time resolution for the time quantizationreduces directly the power consumption which is constantlyrequired to monitor the incoming signal shape Trade-offbetween the time quantization error and the power consump-tion reduction is performed to determine the error boundallowing the appropriate signal detectionThe event segmen-tation size also affects the power consumption reductionslightly which is showed with an example of 168 events and104 events in Figure 10(d)The power consumption reduction


s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR

AEG atomic event generatorCMP level comparatorTMR time-stamp timer

Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef

ISR processingInput ef s(0) s(1) s(2) s(3) s(4) s(5)Output ev-final event

Final event

Final event

Final event

ISR processing

Output ev(final event)

Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5

Figure 8 Comparison of conventional digital signal processing versus event-driven signal processing

AEGSensor analog

IF

AEG atomic event generatorESP event-signal processingfet feature attributeelt elapsed time

idx012

k

FeatureAdd new feature

idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi

AEG(saΔi) rarr aevi fetk eltm

idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m

[range]0[range]1[range]2

⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn

FET = fetk fet0 fet1 fet2 fetk = ⟨idxk attributek⟩

ELT = eltm elt0 elt1 elt2 eltm = ⟨idxm [range]m⟩

evn

fetk

eltm

ESP(|

aevi = ⟨FTBL middot idxx ELTBL middot elty⟩aev1 aevn

k

AttributeAttributeAttribute

Attribute

0

1

2

| |

Figure 9 Index-based feature table including attributes and elapsed time range

is dependent on the event-quantization accuracy controlledby timemeasurement resolution and event segmentation size

6 Conclusion

Themacrolevel signal processing concept is based on the earlyevaluation of incoming sensor signal data by the S2E Thesignal-specific signal segmentation with the features of inter-est enables the atomic event extraction from the continuoussensor data signal The early evaluation of the signal featuresenables the entire system in sleep mode with the exception

of the S2E to consume relatively little current The extractedsmall number of atomic events is analyzed by the EPU whichwill traverse the reduced state space The proposed methodrequires the additional hardware by modifying the conven-tional MCU bus architecture and the user must perform theiterative configuration on the S2E and EPU carefully afteranalyzing the signal characteristics for rare-event activity-sensing applications until the reasonable power reductionis accomplished The event-space representation and signalabstraction of atomic events extracted by S2E could reducethe data processing cost in terms of the energy consumption


ADCCMP

TMR OSC

Tracer memory aevSensor

analog front-end

Atomic event data processing

Wai

t-tim

e

OnOff

On

Off Trigger

1st signal-to-event conversion (S2E)

2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel

(a) S2E circuit data path

MCU CPU Buffer Code memory (flash) Signal-to-event

converter (S2E)

Analog front-end

Event tracer

Event type

Event signal processor

EPU configuration

(flash)

Sensor signal

Time counter(OSC timer)

CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe

dictionarylowastaevk = ⟨lowastetk

lowasttak⟩

aev0 aev1 aevk

lowastaev0lowastaev1

lowastaevk

et0 et1 etk ta0 ta1 tak

lowastPointer index (lowasth lowastc)

(b) Modified microcontroller bus architecture

Circuit-level simulator

Sensor processor

Analog front-end

Sensor device + MATLAB $fread()Circuit netlist

Hand gestureevent

Dump raw data of sensed signal

Loading strobe vector

(nanosim primetime copy)

(c) Measurement environment

01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)

Operating current and accuracy of time quantizer block

Energy consumption and lifetime comparisontime-stamps measurement (OSC + time) current sweep

Energy (168 eventss)Energy (104 eventss)

Lifetime (168 eventss)Lifetime (104 eventss)

120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845

(d) Energy consumption according to event quantization error

Figure 10 Implemented circuit and experimental results

by considering specific characteristics of signals observedin rare-event sensing applications The experimental resultshows that the proposedmethod is an effectiveway to providethe power reduction

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This research was supported by Basic Science ResearchProgram through the National Research Foundationof Korea (NRF) funded by the Ministry of Education(2014R1A6A3A04059410) the MSIP (Ministry of ScienceICT amp Future Planning) Korea under the C-ITRC (Conver-gence Information Technology Research Center) support

program (NIPA-2014-H0401-14-1004) supervised by theNIPA (National IT Industry Promotion Agency) and the2013 Yeungnam University Research Grant

References

[1] L Chen J Hoey C D Nugent D J Cook and Z Yu ldquoSensor-based activity recognitionrdquo IEEE Transactions on Systems Manand Cybernetics Part C Applications and Reviews vol 42 no 6pp 790ndash808 2012

[2] O D Lara and M A Labrador ldquoA survey on human activityrecognition using wearable sensorsrdquo IEEE CommunicationsSurveys and Tutorials vol 15 no 3 pp 1192ndash1209 2013

[3] K Van Laerhoven H-W Gellersen and Y G Malliaris ldquoLong-term activity monitoring with a wearable sensor noderdquo inInternational Workshop on Wearable and Implantable BodySensor Networks (BSN rsquo06) pp 171ndash174 April 2006

[4] K Leuenberger and R Gassert ldquoLow-power sensor modulefor long-term activity monitoringrdquo in Proceedings of the 33rd


Annual International Conference of the IEEE Engineering inMedicine and Biology Society (EMBS rsquo11) pp 2237ndash2241September 2011

[5] Y Tsividis ldquoEvent-driven data acquisition and digital signalprocessingmdasha tutorialrdquo IEEE Transactions on Circuits andSystems II Express Briefs vol 57 no 8 pp 577ndash581 2010

[6] P Panek ldquoError analysis and bounds in time delay estimationrdquoIEEE Transactions on Signal Processing vol 55 no 7 pp 3547ndash3549 2007

[7] Y Tsividis ldquoEvent-driven data acquisition and continuous-timedigital signal processingrdquo in Proceedings of the 32nd AnnualCustom Integrated Circuits Conference (CICC rsquo10) pp 1ndash8September 2010

[8] B Schell and Y Tsividis ldquoAnalysis of continuous-time digitalsignal processorsrdquo in Proceedings of the IEEE InternationalSymposium on Circuits and Systems (ISCAS rsquo07) pp 2232ndash2235May 2007

[9] Y Yilmaz GMoustakides and XWang ldquoSpectrum sensing viaevent-triggered samplingrdquo in Proceedings of the 45th AsilomarConference on Signals Systems and Computers (ASILOMAR rsquo11)pp 1420ndash1424 November 2011

[10] M Miskowicz ldquoThe event-triggered integral criterion forsensor samplingrdquo in Proceedings of the IEEE InternationalSymposium on Industrial Electronics (ISIE rsquo05) vol 3 pp 1061ndash1066 June 2005

[11] B French D P Siewiorek A Smailagic and M DeisherldquoSelective sampling strategies to conserve power in contextaware devicesrdquo in Proceedings of the 11th IEEE InternationalSymposium on Wearable Computers (ISWC rsquo07) pp 77ndash80October 2007

[12] W Li M Bandai and T Watanabe ldquoTradeoffs among delayenergy and accuracy of partial data aggregation in wirelesssensor networksrdquo in Proceedings of the 24th IEEE InternationalConference on Advanced Information Networking and Applica-tions (AINA rsquo10) pp 917ndash924 April 2010

[13] K Choi R Soma and M Pedram ldquoFine-grained dynamicvoltage and frequency scaling for precise energy and perfor-mance tradeoff based on the ratio of off-chip access to on-chip computation timesrdquo IEEETransactions onComputer-AidedDesign of Integrated Circuits and Systems vol 24 no 1 pp 18ndash28 2005

[14] Y Lee J Kim and C-M Kyung ldquoEnergy-aware video encodingfor image quality improvement in battery-operated surveillancecamerardquo IEEE Transactions on Very Large Scale Integration(VLSI) Systems vol 20 no 2 pp 310ndash318 2012

[15] V Gupta D Mohapatra A Raghunathan and K Roy ldquoLow-power digital signal processing using approximate addersrdquoIEEE Transactions on Computer-Aided Design of IntegratedCircuits and Systems vol 32 no 1 pp 124ndash137 2013

[16] A B Kahng and S Kang ldquoAccuracy-configurable adder forapproximate arithmetic designsrdquo in Proceedings of the 49thAnnual Design Automation Conference (DAC rsquo12) pp 820ndash825ACM New York NY USA June 2012

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



Sens

or

Sampler Processor

Wire

less

IF

Change of stateInterested

state change

Recognized event

Monitoring Notify

Electrical signal

Sampled signal

Recognized event

(i) Event arrival time delayskip allowable

(ii) Event presence is more important than data itself

ObserverGenerator

Intended event

Battery-operated sensor-driven IoT device

Observer

rarr Long-term operation is more important than accurate operation

(i) Low frequency event generation

(ii) Secondary retry allowable (average)

(iii) Event type is predefined (signal shape is expected)

Generator

Triggered event

Figure 1 Sensor-based IoT device concept with sensor interface processing unit and wireless connectivity

However this approach by which the sensed data isanalyzed on the microlevel of data-to-data has operatingpower overhead because it is not optimized to consider thelong event-to-event distance of human activity signals whichis easily observed in rare-event sensing applications [4 5]

In this paper for efficient sensor data processing in theenergy consumption a semantic sampling method is intro-duced to capture the signal with the features of interest and isimplemented as a preprocessor unit named to signal-to-eventconverter (S2E) which generates the atomic events insteadof the sampled data itself The extracted atomic events are arelatively small number of samples compared to the syntac-tic sample data by conventional analog-to-digital converter(ADC) as a signal-to-data converter (S2D)

The proposed S2E replaces the conventional S2D to ext-ract atomic events from the incoming sensor signalThe eventidentification from human activity monitoring is performedby the event-driven sensor data processing unit (EPU) for thesmall set of extracted atomic events

This paper is organized as follows In Sections 2 and3 our research motivation and related work are discussedSection 4 describes the details of the proposed architectureThe implementation and experimental results are presentedin Section 5 Finally we conclude the paper by summarizingour contributions in Section 6

2 Motivation

The key motivation of the proposed method begins with thetransition to macrolevel processing of the sensor signal byS2E instead of the conventional microlevel analysis for thesensed dataset Figure 2 shows the method we used to repre-sent the sensor signal acquisition result in the event dataspace The human activity sensor signals are represented bythe attributes of interest and the elapsed time relationshipbetween the atomic events in Figure 2(a) The monitoredhuman activity is encoded as a set of atomic events by theevent quantization in Figure 2(b)

To address this limitation of the conventional digitalsystem architecture by using the discrete time-based sensordata processing method we propose an event-driven systemarchitecture that modifies traditional digital system designWe present a theoretical framework to implement an event-driven sensor processor for general rare-event sensing appli-cations by analyzing the system operations

21 Event-Space Signal Representation Our main researchbegins with an event-space representation of the signalinstead of the digital data space domain The extracted fea-tures of the sensed signal are encoded into the elapsed timebetween events and informative value such as voltage leveland edge phase crossing the trigger point of the signal Thefundamental event defined which is defined as an atomicevent with the most important information provides a signalrepresentation on an abstract level and reduces the com-putational complexity in performing basic data processingfor extracted informative features of interest The collectedatomic events include partial information in the originalsignal that specifies whether the desired featured points of thesignal are present

22 Accuracy-Controlled Event Quantization The event-quantization concept extends the time-quantization methodfor signal representation that uses elapsed time to enhance theconventional data-sampling and processing method Timequantization monitors only the specific conditions of thesignal transition and captures the time-stamps The event-quantization method also determines whether the specifiedcharacteristics of the signal exist

23 Event-Driven Sensor Data Processing The event-basedapproach with a certain amount of accuracy error is des-cribed by the proposed event-driven sensor data processingflowThe input signal is monitored with specified interest-of-signal characteristics to generate the specific atomic events ofthe signalThe set of atomic events during the specified region





Distances(t)

t

s(t)


thc

thb

tha

aev 1

aev 2

aev 3

aev 4

aev 5

aev 0



tSens

or si

gnal

sp

ace

Even

t dat

a spa

ce

t

et0

= ⟨Ea et0⟩

et2

= ⟨Ea et2⟩

et1

= ⟨Eb et1⟩

et3

= ⟨Eb et3⟩

et4

= ⟨Ec et4⟩

et5

= ⟨Ed et5⟩

Ea

Eb

Ec

Ed

aev1

aev2

aev3

aev4

aev5

aev0






3 Related Work












(ADC)Final event


di di

evj


Signal feature


S2E (AEG)

s(t) Event tracing







ev

Δeaevi aevi



Time

Time

s(t)

s(t)

s(t)



Time


thi = Δ lowast i

thi = Δ lowast i

Li = Δ lowast i


em998400 m

tm998400 = Δt lowast m998400





1198981015840119898


119889119898119896

≫ 1198901198981015840119898

(1)











Lf

TrTstart Tend


Dmax




Atomic event

f Ω rarr E


aevi


Timer window


Rf RfRf gtDmax

Rf gtDmax

Dmax

Rr gtDmax Rr gtDmax

Rf gtDmaxRr gtDmax

infininfin


Lprobe Lprobe


start




Swee

p of

feat

ured

leve

l


Signal segmentation

Lr1

Lr1

Lr1

Lr1

Lr1

Lr1

Tr1

Tr1

Tr1

Tr1Tr2

Tr2 Tr3

Tr3



⋱







aev119894= (aev




where aev119894sdot




119889119896= ADQ (119904 (119905

119896) 119871119898 Δ119889 119906 V)


1003816100381610038161003816119871119898

minus 119889119896

1003816100381610038161003816lt Δ119889lowast V

119905119896= ATQ (aev


119879clk = 119879clk + Δ119905

(2)


119905119896= et +


119889119896 (3)







level configuration


Itera

tive c

onfig

urat

ion

of sa

mpl

ing

spee

d th

resh

old

leve

l



2






Gro

upin

gun

grou

ping

of

sign

al se

gmen

ts




2

Ω0

Ω0

Ω1 Ω1Ω2 Ω2 Ω3

Ω3Ω4Ω4

Lf0

Lf0Lf0

Lf1Dr0

Dr0

Dr1

Dr1

Tr0

Tr0

Tr1

Tr1Tstart0

Tstart0 Tstart1

Tstart1

Tend0

Tend0

Tend1

Tend1

ER0

ER0

ER1

ER1

ER2

ER2

ER3

ER3

ER4 ER5




AEG (119904 (119905) 119871) = aev119894| aev119894= ⟨aev

119894minus1119889119896 120601edge 119905119896⟩ (4)


119896minus1




is 119879119903 let 119871


is 119879119891 let the pair of 119871

119909and 119879




119894| Ω119894


aev119894 = AEG (119904 (119905) Ω) (5)



119904in the




Time

Mag

nitu

de



sample2nd detail

1st wait









119894= (ldquo119871

119894rdquo 119905119904119894) (6)



119894= (ldquo119871

119894rdquo 119905119904119894plusmn Δ) (7)




119894 and its



119894






119894only contains





119894) is loaded






s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR


Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef


Final event

Final event

Final event

ISR processing


Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5


AEGSensor analog

IF


idx012

k


idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi


idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m


⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn



evn

fetk

eltm

ESP(|


k


Attribute

0

1

2

| |



6 Conclusion




ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation













Distances(t)

t

s(t)


thc

thb

tha

aev 1

aev 2

aev 3

aev 4

aev 5

aev 0



tSens

or si

gnal

sp

ace

Even

t dat

a spa

ce

t

et0

= ⟨Ea et0⟩

et2

= ⟨Ea et2⟩

et1

= ⟨Eb et1⟩

et3

= ⟨Eb et3⟩

et4

= ⟨Ec et4⟩

et5

= ⟨Ed et5⟩

Ea

Eb

Ec

Ed

aev1

aev2

aev3

aev4

aev5

aev0






3 Related Work












(ADC)Final event


di di

evj


Signal feature


S2E (AEG)

s(t) Event tracing







ev

Δeaevi aevi



Time

Time

s(t)

s(t)

s(t)



Time


thi = Δ lowast i

thi = Δ lowast i

Li = Δ lowast i


em998400 m

tm998400 = Δt lowast m998400





1198981015840119898


119889119898119896

≫ 1198901198981015840119898

(1)











Lf

TrTstart Tend


Dmax




Atomic event

f Ω rarr E


aevi


Timer window


Rf RfRf gtDmax

Rf gtDmax

Dmax

Rr gtDmax Rr gtDmax

Rf gtDmaxRr gtDmax

infininfin


Lprobe Lprobe


start




Swee

p of

feat

ured

leve

l


Signal segmentation

Lr1

Lr1

Lr1

Lr1

Lr1

Lr1

Tr1

Tr1

Tr1

Tr1Tr2

Tr2 Tr3

Tr3



⋱







aev119894= (aev




where aev119894sdot




119889119896= ADQ (119904 (119905

119896) 119871119898 Δ119889 119906 V)


1003816100381610038161003816119871119898

minus 119889119896

1003816100381610038161003816lt Δ119889lowast V

119905119896= ATQ (aev


119879clk = 119879clk + Δ119905

(2)


119905119896= et +


119889119896 (3)







level configuration


Itera

tive c

onfig

urat

ion

of sa

mpl

ing

spee

d th

resh

old

leve

l



2






Gro

upin

gun

grou

ping

of

sign

al se

gmen

ts




2

Ω0

Ω0

Ω1 Ω1Ω2 Ω2 Ω3

Ω3Ω4Ω4

Lf0

Lf0Lf0

Lf1Dr0

Dr0

Dr1

Dr1

Tr0

Tr0

Tr1

Tr1Tstart0

Tstart0 Tstart1

Tstart1

Tend0

Tend0

Tend1

Tend1

ER0

ER0

ER1

ER1

ER2

ER2

ER3

ER3

ER4 ER5




AEG (119904 (119905) 119871) = aev119894| aev119894= ⟨aev

119894minus1119889119896 120601edge 119905119896⟩ (4)


119896minus1




is 119879119903 let 119871


is 119879119891 let the pair of 119871

119909and 119879




119894| Ω119894


aev119894 = AEG (119904 (119905) Ω) (5)



119904in the




Time

Mag

nitu

de



sample2nd detail

1st wait









119894= (ldquo119871

119894rdquo 119905119904119894) (6)



119894= (ldquo119871

119894rdquo 119905119904119894plusmn Δ) (7)




119894 and its



119894






119894only contains





119894) is loaded






s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR


Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef


Final event

Final event

Final event

ISR processing


Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5


AEGSensor analog

IF


idx012

k


idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi


idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m


⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn



evn

fetk

eltm

ESP(|


k


Attribute

0

1

2

| |



6 Conclusion




ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation













(ADC)Final event


di di

evj


Signal feature


S2E (AEG)

s(t) Event tracing







ev

Δeaevi aevi



Time

Time

s(t)

s(t)

s(t)



Time


thi = Δ lowast i

thi = Δ lowast i

Li = Δ lowast i


em998400 m

tm998400 = Δt lowast m998400





1198981015840119898


119889119898119896

≫ 1198901198981015840119898

(1)











Lf

TrTstart Tend


Dmax




Atomic event

f Ω rarr E


aevi


Timer window


Rf RfRf gtDmax

Rf gtDmax

Dmax

Rr gtDmax Rr gtDmax

Rf gtDmaxRr gtDmax

infininfin


Lprobe Lprobe


start




Swee

p of

feat

ured

leve

l


Signal segmentation

Lr1

Lr1

Lr1

Lr1

Lr1

Lr1

Tr1

Tr1

Tr1

Tr1Tr2

Tr2 Tr3

Tr3



⋱







aev119894= (aev




where aev119894sdot




119889119896= ADQ (119904 (119905

119896) 119871119898 Δ119889 119906 V)


1003816100381610038161003816119871119898

minus 119889119896

1003816100381610038161003816lt Δ119889lowast V

119905119896= ATQ (aev


119879clk = 119879clk + Δ119905

(2)


119905119896= et +


119889119896 (3)







level configuration


Itera

tive c

onfig

urat

ion

of sa

mpl

ing

spee

d th

resh

old

leve

l



2






Gro

upin

gun

grou

ping

of

sign

al se

gmen

ts




2

Ω0

Ω0

Ω1 Ω1Ω2 Ω2 Ω3

Ω3Ω4Ω4

Lf0

Lf0Lf0

Lf1Dr0

Dr0

Dr1

Dr1

Tr0

Tr0

Tr1

Tr1Tstart0

Tstart0 Tstart1

Tstart1

Tend0

Tend0

Tend1

Tend1

ER0

ER0

ER1

ER1

ER2

ER2

ER3

ER3

ER4 ER5




AEG (119904 (119905) 119871) = aev119894| aev119894= ⟨aev

119894minus1119889119896 120601edge 119905119896⟩ (4)


119896minus1




is 119879119903 let 119871


is 119879119891 let the pair of 119871

119909and 119879




119894| Ω119894


aev119894 = AEG (119904 (119905) Ω) (5)



119904in the




Time

Mag

nitu

de



sample2nd detail

1st wait









119894= (ldquo119871

119894rdquo 119905119904119894) (6)



119894= (ldquo119871

119894rdquo 119905119904119894plusmn Δ) (7)




119894 and its



119894






119894only contains





119894) is loaded






s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR


Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef


Final event

Final event

Final event

ISR processing


Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5


AEGSensor analog

IF


idx012

k


idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi


idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m


⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn



evn

fetk

eltm

ESP(|


k


Attribute

0

1

2

| |



6 Conclusion




ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Lf

TrTstart Tend


Dmax




Atomic event

f Ω rarr E


aevi


Timer window


Rf RfRf gtDmax

Rf gtDmax

Dmax

Rr gtDmax Rr gtDmax

Rf gtDmaxRr gtDmax

infininfin


Lprobe Lprobe


start




Swee

p of

feat

ured

leve

l


Signal segmentation

Lr1

Lr1

Lr1

Lr1

Lr1

Lr1

Tr1

Tr1

Tr1

Tr1Tr2

Tr2 Tr3

Tr3



⋱







aev119894= (aev




where aev119894sdot




119889119896= ADQ (119904 (119905

119896) 119871119898 Δ119889 119906 V)


1003816100381610038161003816119871119898

minus 119889119896

1003816100381610038161003816lt Δ119889lowast V

119905119896= ATQ (aev


119879clk = 119879clk + Δ119905

(2)


119905119896= et +


119889119896 (3)







level configuration


Itera

tive c

onfig

urat

ion

of sa

mpl

ing

spee

d th

resh

old

leve

l



2






Gro

upin

gun

grou

ping

of

sign

al se

gmen

ts




2

Ω0

Ω0

Ω1 Ω1Ω2 Ω2 Ω3

Ω3Ω4Ω4

Lf0

Lf0Lf0

Lf1Dr0

Dr0

Dr1

Dr1

Tr0

Tr0

Tr1

Tr1Tstart0

Tstart0 Tstart1

Tstart1

Tend0

Tend0

Tend1

Tend1

ER0

ER0

ER1

ER1

ER2

ER2

ER3

ER3

ER4 ER5




AEG (119904 (119905) 119871) = aev119894| aev119894= ⟨aev

119894minus1119889119896 120601edge 119905119896⟩ (4)


119896minus1




is 119879119903 let 119871


is 119879119891 let the pair of 119871

119909and 119879




119894| Ω119894


aev119894 = AEG (119904 (119905) Ω) (5)



119904in the




Time

Mag

nitu

de



sample2nd detail

1st wait









119894= (ldquo119871

119894rdquo 119905119904119894) (6)



119894= (ldquo119871

119894rdquo 119905119904119894plusmn Δ) (7)




119894 and its



119894






119894only contains





119894) is loaded






s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR


Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef


Final event

Final event

Final event

ISR processing


Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5


AEGSensor analog

IF


idx012

k


idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi


idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m


⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn



evn

fetk

eltm

ESP(|


k


Attribute

0

1

2

| |



6 Conclusion




ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation














level configuration


Itera

tive c

onfig

urat

ion

of sa

mpl

ing

spee

d th

resh

old

leve

l



2






Gro

upin

gun

grou

ping

of

sign

al se

gmen

ts




2

Ω0

Ω0

Ω1 Ω1Ω2 Ω2 Ω3

Ω3Ω4Ω4

Lf0

Lf0Lf0

Lf1Dr0

Dr0

Dr1

Dr1

Tr0

Tr0

Tr1

Tr1Tstart0

Tstart0 Tstart1

Tstart1

Tend0

Tend0

Tend1

Tend1

ER0

ER0

ER1

ER1

ER2

ER2

ER3

ER3

ER4 ER5




AEG (119904 (119905) 119871) = aev119894| aev119894= ⟨aev

119894minus1119889119896 120601edge 119905119896⟩ (4)


119896minus1




is 119879119903 let 119871


is 119879119891 let the pair of 119871

119909and 119879




119894| Ω119894


aev119894 = AEG (119904 (119905) Ω) (5)



119904in the




Time

Mag

nitu

de



sample2nd detail

1st wait









119894= (ldquo119871

119894rdquo 119905119904119894) (6)



119894= (ldquo119871

119894rdquo 119905119904119894plusmn Δ) (7)




119894 and its



119894






119894only contains





119894) is loaded






s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR


Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef


Final event

Final event

Final event

ISR processing


Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5


AEGSensor analog

IF


idx012

k


idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi


idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m


⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn



evn

fetk

eltm

ESP(|


k


Attribute

0

1

2

| |



6 Conclusion




ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Time

Mag

nitu

de



sample2nd detail

1st wait









119894= (ldquo119871

119894rdquo 119905119904119894) (6)



119894= (ldquo119871

119894rdquo 119905119904119894plusmn Δ) (7)




119894 and its



119894






119894only contains





119894) is loaded






s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR


Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef


Final event

Final event

Final event

ISR processing


Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5


AEGSensor analog

IF


idx012

k


idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi


idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m


⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn



evn

fetk

eltm

ESP(|


k


Attribute

0

1

2

| |



6 Conclusion




ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










s(t) ADCs(n)

s(n)

Tracing Processing

AEG

s(t)

CMP

TMR


Tracer

Matching

OSC

S2E

InOut

Processing

s(0)

s(1)

Signal feature ef

s(2)

s(3)

s(5) Final event

s(4)

Signal feature ef


Final event

Final event

Final event

ISR processing


Decision making

zminus1

zminus1

zminus1

zminus1

s(n minus 1)

s(n minus 2)

s(n minus c)

Input

Li

⟨⟩

⟨⟩

⟨⟩

⟨⟩aei

ae0

ae1

ae2

ae5

ae0

ae0

ae1

ae1

ae2

ae2

ae5

ae0 ae1 ae2

AE0

AE1

AE2

AE5


AEGSensor analog

IF


idx012

k


idx012

m

Elapsed time

ESP

Event tracer


⟨ ⟨⟩

saΔi


idx m

[ra

nge]

idx ka

ttrib

ute km

[range]m


⟩

AEV = aevi aev0

aev i

aev 3

aev 2

aev 1

aev 0

aevi

aevi

fetk eltm) rarr evn



evn

fetk

eltm

ESP(|


k


Attribute

0

1

2

| |



6 Conclusion




ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










ADCCMP

TMR OSC


analog front-end


Wai

t-tim

e

OnOff

On

Off Trigger


2nd detail sampling

OnOff

Repo

rtRe

port

Mai

n pr

oces

sor

AEG

Phas

e le

vel



converter (S2E)

Analog front-end

Event tracer

Event type


EPU configuration

(flash)

Sensor signal


CPU bus

Event bus

DM

A

DM

A

Event report

core (SRAM)

(EPU)

Δe

Δe


lowasttak⟩

aev0 aev1 aevk


lowastaevk





Sensor processor

Analog front-end


Hand gestureevent





01002003004005006007008009001000

5000

10000

15000

20000

25000

Ope

ratin

g lif

etim

e (ho

urs)





120583A

(15

)

612

120583A

(2

)

480

120583A

(3

)

222120583

A (5

)

145120583

A (1

0

)

91120583

A (2

0

)

66120583

A (2

5

)Ener

gy co

nsum

ptio

n ( 120583

J) d

urin

g1

seco

nd

845






Acknowledgments



References





















RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation

























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









research article a low-power microcontroller with accuracy

Documents