Doctoral Colloquium: Clean Slate System for Minimum-PowerMaximum-Reliability Low-Rate Multi-Hop Wireless Sensor

Networks

Eoin O’ConnellTyndall National Institute

University College Cork, Ireland

eoin.oconnell@tyndall.ie

AbstractThis work presents a clean slate, bottom up, cross-layer

approach to construct highly reliable and ultra-low powerlow-rate multi-hop wireless sensor networks. Starting at thephysical layer, new hardware was designed to help meet in-dustry specified requirements. At the MAC layer, a novelprotocol, inspired by numerous techniques reported in theliterature, was designed. Initial results demonstrate that theprotocol developed outperforms the state-of-the-art in termsof reliability and scalability, and is highly competitive withthe lowest power solutions presented in the literature. Aunique system was developed based on a new weighted ob-jective function with multiple fail safe mechanisms to ensureextreme high reliability and robustness. Average reliabilityfor a deployment of 52 nodes has been empirically evaluatedresulting in over 99.9% packet reception reliability with aradio duty cycle of 0.2%.

1 Project BackgroundThis work is industrially co-funded by EI Electronics, Ire-

land, which mass produces wireless smoke and CO alarmsystems. Their current system provides adequate reliabilityand battery lifetime, but it only offers event detection. Theprotocol functions almost identically to BoX-MAC1 [7]. Thepayload is repeated multiple times with no interleaved gapsfor acknowledgments. The network is flooded with broad-casts from nodes that detect abnormal levels of smoke and/orCO. Neighbouring nodes that hear these broadcast Alarmmessages re-broadcast them, thus achieving network-widemulti-hop coverage.

1.1 Problem StatementA strategic objective of the company was to enable pe-

riodic reporting from each node in the network. It was re-quired that each node should report periodic sensor readings

Permission 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 citationon the first page. To copy otherwise, to republish, to post on servers or to redistributeto lists, requires prior specific permission and/or a fee.

ACM

including temperature, and smoke and CO levels (approx 10minutely). The primary design challenge is thus, can a fullyfunctional multi-hop network capable of reporting periodicsensor readings be developed with an average current drawbudget of 10-12µA?1 Another system requirement definedby the company is a packet delivery reliability of no less than99.5%. Therefore, the multi-objective design constraint iscaptured: Can this average current be achieved while guar-anteeing a reliability of greater than 99.5%?

1.2 Literature ReviewThe literature was surveyed to find a benchmark, state-of-

the-art reliability metric. The most common routing protocolfor WSNs is CTP [4], developed by Moss and Culler. Theauthors provide a list of reliability results from experimentscarried out on various testbeds and hardware. For duty cy-cled deployments where receive checks were performed lessoften than every 0.5s, the authors achieved reliability resultsranging between 90.5% and 98.3%. When using always-onnon-duty cycled sensor networks (100% CSMA MAC) run-ning CTP, the authors report reliability results ranging be-tween 99.9% and 94.7%. Taking these results as a bench-mark of the reliability of such systems, it was evident thatimplementing this technology would not achieve the desiredbalance between reliability and low-power operation.

Section 4 revisits the related literature considering thecurrent state-of-the-art (i.e. developments since the begin-ning of this doctoral work, which began in 2010). Specificconsideration is given to recent developments such as ORW,and other cross-layer approaches presented in the literatureincluding Dozer [1], Koala [8] and DISSense [2].

1.3 Thesis StatementThe scope of this work is broad, covering physical layer

design (using standard ICs on custom designed and manufac-tured PCBs), MAC protocol design, routing protocol design,and application layer design. This work adopts a clean-slate, co-design approach to avoid any fundamental implica-tions and limitations (e.g. concurrency models, dependen-cies, behaviors) associated with community developed andopen-source hardware and software systems. WSNs are con-sidered to be notoriously lossy and unreliable. Based onpreliminary results, this work shows that the reliability of a

110µA is the average drain required to provide the required battery lifeof 10 years

(a) 868MHz mote

TXRX

Hop to New Channel, Wait for PayloadSend CTS

Receive RTS Address Match

Receive CTS Address Match Receive Payload ACKWait for Payload ACKReceive CTS Payload Interrupt,

Hop to New Channel & Send Payload

Payload Received Send Payload ACKProcess RX FIFO

Sleep Mode RX OffReceivingNode

TransmittingNode

CTSListen

RTS

(b) Unicast Send

1101001000500010

100

1,000

Packet Inter Arrival Rate (s)

Curren

t Cons

umptio

n (u

A)

Current Consumption vs Packet Inter Arrival Rate

IX−MAC

BoX−MAC2

ContikiMAC

A−MAC

HUI−MAC

WiseMAC

(c) Power Consumption Testing

Figure 1: (a) Hardware Platform (PIC MCU & SX1211 TRX). (b) Illustrates the send process with ACK & channelhopping. (c) IX-MAC current versus packet inter-arrival rate w.r.t. current art. (TW(RCI)=0.5s)

0 1 2 3 4 5 6

0

0.5

1

1.5

2

2.5

3X: 0.9597Y: 3.243

Time (ms)

Current (mA)

Receive Check Profile

(α)

(β)

(γ)

(δ)

(a) Receive check

6.25 6.3 6.35 6.4 6.45 6.5 6.55 6.6 6.65

0

5

10

15

20

X: 6.612Y: 22.97

Time (s)

Current (mA)

Unicast Send Tw = 1s

X: 6.233Y: 3.667

(β)

(ε)

(α)

(γ )

(δ )

(ζ )

(b) Unicast send with learned offset

0 2 4 6 8 10 12 14 16

80

85

90

95

100

Number of Senders

Pe

rce

nta

ge

of

Pa

cke

ts R

ece

ive

d

Reliability vs Number of Contenders 100ms

BoXMAC2

XMAC

ContikiMAC−125

IX−Mac

(c) Reliability TW=100ms

Figure 2: (a) and (b) captured using a DC power analyzer. In (a), (α) is oscillator startup; (β) is PLL lock phase; (γ) isreceiver on time; (δ) is radio shutdown. In (b), communication is shown between two nodes. Captured is the sender’scurrent profile. (α) is the receive check; (β) is a low power delay of 350ms; (γ) is a CCA check; (δ) is RTS/CTS untildestination responds with optimised TX power; (ε) is payload transmission at full power, and (ζ) is receive payloadACK. (c) is Reliability vs. Number of concurrent senders. 1pkt/s to a single duty cycled receiver. (TW=0.1s )

low-power multi-hop WSN can approach 100% using a verysmall power budget (i.e. ≈10µA average current).

When adopting a clean-slate approach, a certain elementof overlap with existing techniques is unavoidable. However,results obtained show this approach has allowed for some ex-cellent performance results to be achieved, particularly withrespect to reliability.

2 Work Done2.1 Hardware

I have designed, manufactured and tested a WSN hard-ware platform, pictured in Figure 1a. 100 nodes consist-ing of a PIC24F microcontroller interfaced with an SX1211868MHz transceiver were produced. The PIC24F XLP fam-ily was chosen because of its excellent sleep mode cur-rent. The SX1211 transceiver was chosen because of itslow receive mode current of 3mA. The hardware is de-signed to physically integrate with existing product housing.The communications protocols presented in the next sectionshave been implemented on a 2.4GHz (CC2520 transceiver,MSP430 microcontroller) platform to demonstrate utility be-yond the project specific hardware.

2.2 Low Power IX-MAC Design & ResultsI designed a novel MAC protocol that includes a number

of reliability and scalability enhancing features, in addition

to power saving techniques. At a glance this MAC is similarto X-MAC, hence the name improved X-MAC, or IX-MAC.A summary of the power saving techniques of the IX-MACprotocol are as follows:

• Optimized Receive Check (Depicted in Fig 2a)

• Neighbor Schedule Learning (Depicted in Fig 2b)

• Dynamic TX Power

• Optimized ACK TimeoutsThe reliability and scalability enhancing techniques can besummarized:

• Efficient RTS/ CTS (Depicted in Fig 1b)

• Two Stage ACK (Depicted in Fig 1b)

• Reduced TX times (Depicted in Fig 2b)

• Optimized CSMA implementation (Depicted in Fig 2c)

• Optimized TimeoutsThe performance of the MAC and physical layer combina-tion was compared against a number of established proto-cols, mostly demonstrated on TelosB hardware.

Presented in Figure 2c are the results of the scalabil-ity/reliability testing. The reliability performance of variousprotocols when multiple concurrent transmitting nodes arecontending for the medium is depicted. This work achieves

100% reliability for 11 concurrent senders when sending onepacket per second to a single duty-cycled receiver (sink),with Receive Check Interval TW =100ms.

Presented in Figure 1c are the results of the power con-sumption testing. Included are results for this work, BoX-MAC2, X-MAC, and Contiki-MAC, and simulation-only re-sults from Wise-MAC. Real traces of the power consumptionof one node forwarding upstream packets to a duty-cycledneighbor are shown.

Significant effort was invested in optimizing the perfor-mance of the MAC protocol. Multiple timing parameterswere swept to find the values which give the best perfor-mance trade-off in terms of reliability and power consump-tion.

2.3 Routing Protocol Design & ResultsI designed an ultra-reliable and low-network overhead

routing protocol, and tested it. Building on the IX-MACprotocol, it includes features to enable ultra low-power, re-liable networking. Prominent features are: 1. Storing Mul-tiple Routing Options, 2. Loop-back avoidance, 3. NovelWeighted Parent Selection Process, 4. Load Balancing, 5.Efficient usage of Beacons, 6. Message Piggybacking, 7.Energy Aware Routing.

A total of 52 nodes were deployed in an old buildingspanning 3 storeys. The dimensions of the building are(L:60m,W:70m,H:20m), thus larger in coverage area thantestbeds such as Indriya [3] and Twist [5]. The metrics undertest were reliability, radio duty-cycle and end-end latency.Results were compared against experiments conducted onother testbed deployments for protocols including ORW [6]and CTP. Two slightly different versions of the protocol werealso tested and compared. Version A optimizes for reliabil-ity and low-power operation, whereas Version B optimizesfor reduced latency network paths. This is achieved by usingdifferent parent selection parameters. Both versions of theprotocol were tuned and perfected after significant periodsof ‘trial and error’ testing. Version A achieves an averagereliability of 99.983% and 0.207% radio duty-cycle, and Bachieves 99.58% at 0.25% duty-cycle. CTP achieves 2.2%duty-cycle and ORW 0.8% sending packets at the same rate.The improvements are evidently noteworthy.

3 Novelty & ContributionThe novelty of the work lies in the integration of multi-

ple power saving and reliability improving techniques intoone overall cross-layer solution. The results of the MAClayer testing and the results for the deployment (whichtests the overall performance of the system) show signifi-cant improvements compared to the state-of-the-art). Withthe exception of Wise-MAC at low packet forward rates(<1pkt/100s), this work outperforms all evaluated protocolsin terms of power consumption. It also outperforms the state-of-the-art in terms of scalability and reliability (See Figure2c). In terms of radio duty-cycle this work achieves 0.207%sending packets every 4 minutes and with TW =1s.

This work allows WSNs to provide long maintenance freeoperation and more frequent, reliable sensor readings fromeach networked device.

4 Proposed Final StepsEven though this work shows improvements over CTP

and ORW in terms of the metrics of interest, it is plannedthat results from other SoA systems such as Dozer, Koalaand DISSense can be compared against this body of workto fully verify the contribution and improvements. It wouldalso be of interest to compare my parent selection process tothat used in IETF RPL.5 Biographical Sketch

O’Connell earned hisB.Eng. in Electrical andElectronic Engineering atUniversity College Cork,Ireland, graduating in 2010with First Class Honors. Hecompleted many wirelessembedded systems projectssince a young age, and waskeen to pursue a researchcareer in the field of WSNs.His initial research studiedreliability and energy prop-erties of WSNs as reportedin the literature. Thereafterhe proposed and executed

a clean-slate approach to building ultra-low power, ultra-reliable, duty cycled sensor networks. He is academicallysupervised by Prof.Cian O’Mathuna, Tyndall National In-stitute, University College Cork, Ireland. Dr. David Boyle,Dept. of Electrical and Electronic Engineering, ImperialCollege London, UK, has co-supervised and mentored hisPhD project. I hope to submit my doctoral dissertation anddefend my thesis for consideration by the University CollegeCork Summer Examination Board in 2014.6 References[1] N. Burri, P. von Rickenbach, and R. Wattenhofer. Dozer: ultra-low

power data gathering in sensor networks. In Proceedings of the 6thinternational conference on Information processing in sensor networks.ACM, 2007.

[2] U. Colesanti, S. Santini, and A. Vitaletti. Dissense: An adaptiveultralow-power communication protocol for wireless sensor networks.In Distributed Computing in Sensor Systems and Workshops (DCOSS),2011 International Conference on, pages 1–10, 2011.

[3] M. Doddavenkatappa and M. C. Chan. Aal an experience of buildingindriya. National University of Singapore, 2009.

[4] O. Gnawali, R. Fonseca, K. Jamieson, D. Moss, and P. Levis. Col-lection Tree Protocol. In Proceedings of the 7th ACM Conference onEmbedded Networked Sensor Systems (SenSys’09), November 2009.

[5] V. Handziski, A. Kopke, A. Willig, and A. Wolisz. Twist: A scalableand reconfigurable wireless sensor network testbed for indoor deploy-ments. Technical University Berlin, TKN Technical Report TKN-05-008, 2005.

[6] O. Landsiedel, E. Ghadimi, S. Duquennoy, and M. Johansson. Lowpower, low delay: opportunistic routing meets duty cycling. In Pro-ceedings of the 11th international conference on Information Process-ing in Sensor Networks, pages 185–196, Beijing, China, 2012. ACM.

[7] D. Moss and P. Levis. Exploiting physical and link layer boundaries inlow-power networking. Technical report, Stanford, 2008.

[8] R. Musaloiu-E., C.-J. M. Liang, and A. Terzis. Koala: Ultra-low powerdata retrieval in wireless sensor networks. In Proceedings of the 7thinternational conference on Information processing in sensor networks.IEEE Computer Society, 2008.