arX

iv:1

701.

0467

3v1

[cs.

NI]

17

Jan

2017

1

HARE: Supporting efficient uplink multi-hopcommunications in self-organizing LPWANs

Toni Adame, Sergio Barrachina, Boris Bellalta, Albert BelDepartment of Information and Communication Technologies,

Universitat Pompeu Fabra, BarcelonaEmail: {toni.adame, sergio.barrachina, boris.bellalta, albert.bel}@upf.edu

Abstract—The emergence of low-power wide area networks(LPWANs) as a new agent in the Internet of Things (IoT) willresult in the incorporation into the digital world of low-au tomatedprocesses from a wide variety of sectors. The single-hop concep-tion of typical LPWAN deployments, though simple and robust,overlooks the self-organization capabilities of network devices,suffers from lack of scalability in crowded scenarios, and payslittle attention to energy consumption.

Aimed to take the most out of devices’ capabilities, the HAREprotocol stack is proposed in this paper as a new LPWAN technol-ogy flexible enough to adopt uplink multi-hop communicationswhen proving energetically more efficient. In this way, resultsfrom a real testbed show energy savings of up to 15% whenusing a multi-hop approach while keeping the same networkreliability. System’s self-organizing capability and resilience havebeen also validated after performing numerous iterations of theassociation mechanism and deliberately switching off networkdevices.

I. I NTRODUCTION

In the coming years, electronic equipment will be inter-connected and consequently every person and every industrywill become simultaneously data generators and consumers.Internet of Things (IoT) paradigm is a key enabler of thisvision by delivering machine-to-machine (M2M) and machine-to-person communications on a massive scale.

As more and more things are connected to the Internet,low-cost and low-traffic devices are starting to be demanded.However, traditional cellular networks do not deliver a goodcombination of technical features and operational cost forthose IoT applications that need wide-area coverage combinedwith relatively low bandwidth, long battery life, low hardwareand operating cost, and high connection density [1].

Low-power wide area networks (LPWANs) are intended tobecome the engine of long-range, low-bandwidth IoT applica-tions (see Figure 1), which until now have been constrained bydeployment costs and power issues. The goal of these networksis to deliver small amounts of data over long ranges, at ratesof up to tens of kilobits per second (kbps), with a batterylifetime of up to several years, supporting thousands of devicesconnected to a base station, and facilitating online integration.

Existing LPWAN technologies can be categorized into twotypes [2]:

1) Dedicated LPWANs consisting of the purposely de-signed technologies such as LoRaTM [3], SIGFOXTM [4],IngenuTM [5], WeightlessTM [6], DASH7 [7], and ETSI-LTN [8].

High BW

Short range Medium range Long range

Medium BW

Low BWRFID / NFC

BluetoothBLE

802.11

VHT 802.11

WBAN802.15.6

ZigBee /802.15.4

WPAN802.15.3

4G

5G

LPWAN

Fig. 1. Localization of LPWAN technologies according to range capabilityand bandwidth required.

2) Evolutionary LPWANs covering the alternatives that havebeen developed as upgrades to well-established protocolslike IEEE 802.11ah (also known as Wi-Fi HaLow) [9],EC-GSM-IoT [10], LTE-M [11], and NB-IoT [12].

LPWAN architecture is characteristically single-hop, whereend devices are connected directly to the base station, greatlysimplifying the network and endowing it with robustness andcentralized control. And yet this single-hop massive channelaccess sets out some inherent challenges: reliability, scalability,flexibility, and quality of service (QoS). In fact, the channelaccess mechanism of some LPWAN technologies resorts to theuse of ALOHA [13] [14], a random medium access control(MAC) protocol in which end devices transmit without doingany carrier sensing to check the channel state in advance.Although simple, this uncontrolled medium access leads tointerference or packet collisions among uncoordinated devices,acutely affecting reliability in dense networks. In addition,LPWAN devices located far away from the base station mustmake use of high transmission power levels, resulting in severeenergy consumption and reduced battery lifetime [15].

In this article, the HARE protocol stack is proposed asa new LPWAN technology flexible enough to adopt uplinkmulti-hop communications when proving energetically moreefficient than single-hop. A full set of advanced techniquesbelonging to different communication layers has been designedfor this purpose, while ensuring data transmission reliability:

• Inherent clock synchronization, with nodes being period-

2

ically set in time by means of beacons.• TDMA-like channel access for groups of contenders with

multiple transmission opportunities.• Adaptive transmission power level.• Flexible and scalable network association process.• Energy-aware, adaptive and resilient routing protocol.• Regular use of deep-sleep states.

Furthermore, HARE protocol stack has been implementedand tested in real hardware platforms. Results evaluation fromdifferent network configurations (single-hop vs. multi-hop, useof different MAC layers within the TDMA slots, channelerror injection) show very high reliability while maintaininglow energy consumption (particularly in multi-hop topologies).Lastly, we have observed a better overall system behaviourwhen using multi-hop topologies in non error-prone scenarios.

The remainder of this paper is organized as follows: SectionII introduces the main requirements of feasible use cases forHARE. Next, Section III describes the general operation of theprotocol stack and Section IV provides detailed informationof the developed mechanisms. Section V describes the pro-posed testbed and Section VI compiles the obtained resultsfrom different experiments. Lastly, Section VII presents theconclusions and discusses open challenges.

II. SCENARIOS AND REQUIREMENTS

According to their own characteristic range and bandwidthcapabilities with respect to other technologies, the main usecases to which LPWANs are addressed include security alarms,car park spaces, agricultural applications, smart metering,consumer electronics, and intelligent buildings. By way ofillustration, in the ENTOMATIC EU-project1 a network ofwireless sensor nodes [16] periodically report information onpest population density and environmental parameters, such astemperature and relative humidity.

HARE is clearly aligned with typical LPWAN applicationsand circumscribes its suitability to those scenarios with specialconcern for energy efficiency, where device batteries are solimited that the establishment over time of a direct connectionto the base station, or gateway (GW), would greatly affecttheir lifetime.

In this sense, Table I offers a comprehensive list of commonrequirements from use cases to which HARE protocol stackgives response in combination with the appropriate hardware.Assuming that this hardware provides good signal penetration,a single GW can serve up to thousand devices within its givencoverage range. Applications executed by stations (STAs),inturn, follow a continuous data delivery model [17] for theirsensed information, periodically delivering small amounts ofnon-delay sensitive data.

As sensor nodes are scattered over large areas, sometimeswith problematic access, self-maintenance of the system shallbe a priority, capable of giving response to the followingchallenges:

1ENTOMATIC main webpage: https://entomatic.upf.edu/2Although not considered in the current article, future HAREdevelopments

will consider offering QoS in scenarios with miscellaneoussensors (continu-ous, event-driven, query-driven, and hybrid)

TABLE ICOMMON REQUIREMENTS OFHARE USE CASES

Requirement ValueCoverage range Up to several km.Geographic coverage Excellent even in remote and rural areasPenetration Good in-building and in-ground penetrationDevice density(per base station) High (up to thousand)

Power profile Unassisted, battery-powered devicesBattery lifetime From some months up to several yearsThroughput <100 bits/sLatency Non-delay sensitiveMobility Static devicesCost Low hardware and operating costMaintenance Unassisted and self-organizing networkDelivery model Continuous data delivery model2

1) Node initiated network connection: Once installed forthe first time or relocated, any node shall initiate itsassociation process through a simple action (for instance,pressing a button).

2) Self-configuration and management: With the aim ofbuilding a robust network, it shall adapt itself to envi-ronmental and/or topology changes without human inter-vention.

3) Battery lifetime maximization: LPWANs replace oldmonitoring systems consisting of assigning human re-sources to study in situ the behavior of one or more physi-cal parameters. Therefore, maximizing battery lifetime insuch systems is vital in order to justify their usage aheadof other methods.

4) Firmware distribution: Any change in the network con-figuration or in the application purpose shall be remotelyand easily distributed by the GW.

Lastly, the operating system of embedded sensor nodes istypically less complex than general-purpose operating systems[18]. However, the high variety of resources to manage inthis kind of devices (processors, memories, clocks, networkinterfaces, etc.) and the demand of support for concurrentexecution of processes (time synchronization, data acquisition,task scheduling, channel access, routing parameters, etc.) makeessential the use of a real time operating system (RTOS).

Under these premises, Table II compiles some of the mainuse cases supported by the HARE protocol stack in five IoTrepresentative sectors: home and industrial automation, publicinfrastructure, natural resources, and smart agricultureandfarming.

III. HARE OPERATION

The HARE protocol stack conceives end devices as ele-ments controlled by the GW by means of beacons. This cen-tralized approach allows STAs to remain asleep the majorityofthe time, so that their single concern is to be awake enough inadvance to listen to the next beacon. Network synchronizationis thus achieved and allows the GW to ask for specific dataand/or distribute configuration changes in just one hop.

The GW is considered to be appropriately placed closeto a power source. Thus it may always stay in an activestate and is provided with the ability to directly communicate

3

TABLE IIUSE CASES SUPPORTED BYHARE PROTOCOL STACK

Sector Use cases

Home automationDomoticsChild/elderly trackingSmart metering

Industrial automation

Remote maintenance/controlLogisticsLocal asset tracking managementDistribution automation (smart grid)Smart metering

Public infrastructure

City smart lightingSmart parkingIntelligent buildingsPredictive maintenance

Natural resourcesEnvironmental monitoringNatural disasters detection

Smart agricultureand farming

Agriculture monitoringAnimal monitoringSilo stock monitoring

(i.e., via single-hop communications) with any node of thenetwork throughunicast and/or broadcast messages as wellas to redirect gathered data from the WSN to other networksor the Internet.

Conversely, STAs can take advantage of their neighbors tocreate multi-hop paths over which data is transmitted to theGW by means of lower transmission power levels. Dependingon their position within these paths, STAs are ideally organizedinto rings, as shown in Figure 2. The number of hops to reachthe GW determine the ring number (i.e., STAs fromring 2need two hops to reach the GW).

Each uplink data transmission phase (consisting of one ormoretransmission windows) begins with a beacon signal fromthe GW. Transmission windows are in turn virtually split intoas many TDMA slots as network rings, so that STAs are onlyactive during their own slot (for transmitting data) and theprevious one belonging to their children3 (for receiving data).

The first slot is allocated to the highest ring and therest are scheduled consecutively. Data received by STAs isaggregated to that generated by themselves, and finally sentto the corresponding parent at the minimum power level whichensures reliable communications. This process is repeatedasmany times as rings the network has.

The correct reception of data transmissions at the GW isacknowledged with a broadcast message, so that STAs are notonly aware of their own end-to-end reliability, but also of thoseSTAs in the same path to the GW. These acknowledgmentbeacons, together with the information obtained from theiradjacent nodes, allow STAs to decide whether they shouldremain awake to perform retransmissions of lost networkpackets.

Network association (also started by a beacon) remainsstable until a change in the topology is detected or themechanism is reset by the GW. Nevertheless, the agreedtransmission power between adjacent nodes in the associationphase is constantly monitored and adjusted in a decentralized

3Children refers to all STAs of an adjacent higher ring from which an STAreceives packets. Similarly,parent refers to that STA from an adjacent lowerring to which an STA transmits its own packets (after aggregating the onesfrom its children) in its way to the GW.

GW

R1 R

2R3

N18

N3

N2

N5

N6

N7

N8

N11

N14

N17

N10

N4

N1

N13

N9

R4

N16

N15

N26

N25

N24

N23

N22

N12

N28

N29

N30 N

19

N20

N21

N27

(a) Multi-hop LPWAN with a gateway (GW) and 30 stations (N1 −

N30) deployed in 4 rings (R1 − R4).

GW N3

N1 N

2

N7

N8

N4

N5

N6

N9

R1

R2

R3

R4

obstacle

N10

(b) Multi-hop LPWAN affected by an obstacle, with a gateway (GW)and 10 stations (N1 −N10) deployed in 4 rings (R1 − R4).

Fig. 2. Network topology of typical LPWANs

way in order to reduce the energy consumption.

IV. PROTOCOL STACK

The main features of the HARE protocol stack are shownin Table III; a complete description of them is offered next.

A. PHY layer

HARE protocol stack is intended to be used over anywireless PHY layer fulfilling a minimum set of functions;namely, availability of different operational states bothinthe microprocessor (processing, low power mode) and in theradio module (receiving, transmitting, and sleeping), selectionof different transmission levels in the radio transceiver,andability to execute low level tasks required by typical sharedmedium access techniques.

4

TABLE IIIMAIN FEATURES OFHARE PROTOCOL STACK

Layer Features

Transport

End-to-end ACKPoisoning mechanismTransmission windows

Distributed caching

NetworkAddressing system

AssociationRouting

Link

Beaconing systemWakeup patterns

Data transmission, aggregation & segmentationPower regulation mechanism

Physical Hardware dependant

B. Link layer

The MAC layer is a combination of a time division multipleaccess (TDMA) scheme, where time slot duration is managedby the GW, and an underlying carrier sense multiple accesswith collision avoidance (CSMA/CA) technique with packetacknowledgment (ACK), performed by the group of STAsallocated into each generated time slot. At this point it isworth noting that HARE is not only limited to CSMA-basedaccess techniques, but also can properly work with other MACprotocols for WSNs [19].

1) Beaconing system: The designed beaconing system hasa double function: synchronizing the network devices andscheduling the different actions to be performed. Two typesof beacons are used for this purpose:primary and secondarybeacons (see Figure 3).

Both beacons include a timestamp, the time until thenext primary beacon, and the next action to be taken bythe network: for instance, an association phase (networkassociation primary beacon), or an uplink data transmissionphase (data primary beacon). Secondary beacons include thesame information asprimary ones, and are used to guaranteeinformation redundancy for already associated STAs as wellas to accelerate network discovery for non-associated ones.However, no action is performed by STAs after asecondarybeacon.

Time between two consecutiveprimary beacons and twoconsecutivesecondary beacons is defined asTp and Ts,respectively. WhereTp = (ks + 1) · Ts, beingks the numberof secondary beacons transmitted after everyprimary beacon.

2) Wakeup patterns: A wakeup pattern is a set of instruc-tions generated by the GW which define the wakeup planof its associated STAs over time periods. With the goal ofminimizing the time STAs remain active (and, consequently,their energy consumption), two different wakeup patternscontrolled by the GW are proposed according to the network’straffic flow [20].

The periodic wakeup pattern is suitable for listening tobroadcast downlink communications from the GW, as it makesall STAs wake up at the same time. On the other hand,uplink communications follow astaggered wakeup pattern,which allocates different active periods to nodes belonging toadjacent rings with partial overlapping (as shown in Figure4). Apart from reducing time STAs are awake during uplinkcommunications, this method facilitates the implementation of

data aggregation mechanisms.Even though STAs have predetermined active periods, they

can go to sleep even earlier in the transmitting (TX) timeperiod if their parent has acknowledged all their data, or inthe receiving (RX) time period after having received all datafrom their children.

awakeawake

sleep

sleep

sleep

sleep

sleepsleep

RX Time Period

TX Time Period

TX Time Period

RX Time Period

TX Time Period

RX Time Period

STAs from ring 1

GW

STAs from ring 2

STAs from ring 3

transmission window

Fig. 4. Example of astaggered wakeup pattern in a 3-ring LPWANperforming uplink communications.

3) Data transmission, aggregation, and segmentation:Downlink communications are generally executed throughbroadcast messages from the GW. Conversely, uplink com-munications are unicast and follow a multi-hop route.

The staggered wakeup pattern fits here perfectly with theapproach of data aggregation in WSN. Thus nodes attachtheir own data to that received from their children and allthe information is jointly sent to the next hop (i.e., parent). Ifthe total amount of data aggregated by an STA exceeds themaximum payload supported by the hardware, it is split intosegments4 sent consecutively.

A selective ACK mechanism has been developed, so thatbefore the end of the allocated time slot, the receiver explicitlylists which segments in a stream coming from the same childare acknowledged. Upper layers are therefore responsible formaking the sender retransmit only the missing segments insuccessive transmission windows.

4) Power regulation mechanism: The selection of the min-imum suitable transmission power level for outgoing packetsis managed through a mechanism based on the received signalstrength indicator (RSSI). For this purpose, a safety margin forreliable communications is defined by RSSImin and RSSImax.If a node is transmitting data packets (ACKs) to its parent(child) at a power level making the received RSSI higherthan RSSImax, it will be asked to decrease it for the nexttransmission. Similarly, if the received RSSI is lower thanRSSImin, it will be asked to increase it.

Power regulation requests are included in an RSSI controlfield of data packet and ACK headers. Possible values of thisfield are: increase, keep, and decrease. Once computed therequests from parent and children, the STA determines whether

4The amount of data aggregated by an STA (from itself and from itschildren) is calledpacket. If this packet is split into different parts, eachone of these parts is calledsegment. In case both terms can be indistinctlyused, the current article will usepacket.

5

STA

asso

cia

tio

n p

ha

se

UL d

ata

tra

nsm

issio

n p

ha

se

Ne

two

rk

asso

cia

tio

n p

ha

se

...

Network association primary beacon

Data primary beacon

Data primary beacon

STA

asso

cia

tio

n p

ha

se

UL d

ata

tra

nsm

issio

n p

ha

se

Tp = (ks+1)·Ts

Ts

Secondary beacons Secondary beacons Secondary beacons

Network association primary beacon

#1 #2 #3 #4

Ne

two

rk

asso

cia

tio

n p

ha

se

Fig. 3. HARE beaconing system consisting ofnetwork association primary beacons, data primary beacons, andsecondary beacons.

and how to regulate its own power level depending on thefollowing considerations:

• If one or more STAs ask for a higher value, increase thepower level.

• If all STAs ask for a reduction, decrease the power level.• Otherwise, keep the current power level.In addition, if an STA needs to retransmit a packet to

its parent, it will also increase the power level in eachnew transmission window. Regarding the association process,whenever an STA listens to a discovery request, it will answerat maximum power. The STA selected as parent will keepthe maximum power level at the beginning and regulate itfollowing the previously described procedure. Instead, thoseSTAs not selected as parents will set their power back to thelevel they had before answering to the discovery request.

Consequently, the main advantages of using such a MAClayer scheme are:

• Clock synchronization is inherent to TDMA, with nodesbeing periodically set in time by means of beacons.

• Groups of nodes have their time slots clearly allocated,and collisions within groups are sensibly reduced or evenavoided by using CSMA/CA.

• Network overall lifetime is increased by putting nodesin non-active modes for most of the time and onlyperiodically waking up to check for activity.

• Association and routing mechanisms are also fit for thisscheme, so that intermediate and already associated nodesdo not have to constantly listen to hypothetical networkdiscovery requests.

• The scheme is also suitable for uplink data aggregation.• Changes in the network configuration or even new

firmware can be easily distributed in a coordinated man-ner.

C. Network layer

Network communications follow a centralized scheme,where the GW adopts the main role and assumes the respon-sibility of managing network associations, delivering networkaddresses, and periodically notifying the start of new routingprocesses.

STAs adopt a subordinated role waiting for orders comingfrom the GW. In the routing process, they organize themselves

in paths autonomously, but all subsequent data transmissionsare addressed to the GW, directly or through other STAs.Conversely, the GW can make use of its greater transmissionpower to periodically send broadcast messages to all networkSTAs, or send unicast messages to single STAs.

1) Addressing system: The addressing system is managedby the GW, which allocates a unique network address to eachnode during the association process. Nodes will maintain thesame network address as long as they do not leave the network.A dynamic record matching the MAC and the network addressof all STAs is stored in the GW. The size of the networkaddress is configurable and its value determines the addressingrange.

2) Association: To cope with multiple association requestsin a short period of time, the system is able to admit newSTAs through two different mechanisms: an active, global,scheduled one, callednetwork association mechanism; and apassive, singular one, calledSTA association mechanism.

• Network association mechanismThe network association mechanism allows a largeamount of STAs to associate to the network in a shortperiod of time. Once the GW is activated, or after a pre-determined number ofprimary beacons (Npr), the GWbroadcasts anetwork association primary beacon.Depending on the RSSI value received in thenetworkassociation primary beacon as well as some other config-uration parameters, STAs determine their turn to initiatethe association process (generally, the greater the RSSIreceived, the earlier association turn is selected).STAs then follow with a discovery message sent viabroadcast, which is responded by the GW and all thealready associated STAs, provided they are within thecoverage range. The process of selecting the best path toreach the GW is detailed in theRouting subsection.Once the routing mechanism is completed, the GWnotifies the joining of new STAs by means of a summarybroadcast message sent immediately after every associa-tion turn.

• STA association mechanismThe STA association mechanism provides a solution tothose specific nodes that (i) have not found a path tothe GW during the network association mechanism, (ii)have been powered on between two consecutivenetwork

6

association primary beacons, or (iii) have simply sufferedrouting problems in their path to the GW.This mechanism follows the same pattern as thenetworkassociation one, with the single exception that there isonly one association turn located immediately after eachdata primary beacon to be used by non-associated STAs.

Inactive or erratic STAs are removed from the network andthe GW’s routing table to create, if necessary, new routingpaths that ensure correct packet reception from remainingnetwork STAs. Disassociations can be controlled by the GWthrough thedisassociation mechanism or by the STAs them-selves through theself-disassociation mechanism:

• Disassociation mechanismThe GW removes an STA from the network if not re-ceiving any data packet during a pre-determined numberof consecutiveprimary beacons (Npd). A roster with thelatest disassociated STAs is included in everyprimarybeacon. This information is not only useful for malfunc-tioning STAs, which can make immediate use of theSTAassociation mechanism, but also for their parents, as theycan check the current state of their children. Hence, if allits children became disassociated, a parent would go tosleep during the RX time period allocated to its ring.

• Self-disassociation mechanismThe goal of this mechanism is to avoid repetitive associa-tion requests and other energy consuming procedures thatcould make STAs run out of battery when no connectionwith the GW is possible. All STAs have a timer that isactivated after being switched on or when receiving apri-mary beacon. From that moment on, if an STA does notreceive any other beacon during a predetermined period(Td), it turns itself off. Thus the STA is considereddeadand it will need to be reactivated by manual procedures.

3) Routing: The routing protocol has been designed as anintrinsic part of the association process. Thus, accordingtothe responses to the discovery message coming from othernodes, each STA determines which candidate is the best oneto become its parent; i.e., the one with the minimumS valuefrom:

S = a1 · (PTXmax− RSSITX) + a2 · (PTXmax

− RSSIRX) + a3 · r + a4 · c, (1)

where PTXmaxis the maximum transmission power of the

transceiver (in dBm), RSSITX is the RSSI received at thecandidate (in dBm), RSSIRX is the RSSI received at the STAitself (in dBm), r is the ring to which the candidate belongs,and c is the current number of candidate’s children. Theaweights are attached to everyprimary beacon, and can betuned by the GW according to environment requirements.

Once computed the best parent, the STA sends it a specificrequest. This request will be forwarded by the parent throughits own path until reaching the GW, which will send a packetvia broadcast confirming the association and providing theSTA with its new address. This way, both the newly associatedSTA and its parent are informed of the establishment of thenew path.

When the association process is finished, the STA exactlyknows the next hop its messages must follow to reach the

GW

R1 R

2R3

N1

N3

N2

N6

N7

R4

N4

N5

N9

N10

N12

N13

N14

N8

N11

X

Fig. 6. Network topology of the multi-hop LPWAN from Figure 5, with agateway (GW) and 14 stations (N1 −N14) deployed in 4 rings (R1 −R4).

GW. As long as the STA is associated to the network, ituses the same routing path, which is only recomputed afteran internal or external (i.e., from its parent) failure. Indeed,no new routing process is initiated unless it is part of a newnetwork association mechanism.

D. Transport layer

Reliable end-to-end communications from the STAs to theGW, where retransmissions are only executed when neededand by the minimum number of involved STAs, are achievedin HARE by using the following mechanisms:

1) End-to-end ACK: According to thestaggered wakeuppattern, STAs from ring 1 are the last ones to access to thechannel and transmit their information. Once compared thedata sources with the expected uplink traffic, the GW emitsa broadcast message calledend-to-end ACK (e2e ACK) witha list of acknowledged STAs. Figure 5 shows the e2e ACKoperation at the end of every transmission window. Apart frombeing simple, quick and simultaneously listened by all networkelements, end-to-end ACKs allow STAs to evaluate the stateof their path to the GW and act consequently.

2) Poisoning mechanism: The poisoning mechanism identi-fies which specific nodes experience communication problemsin their path to the GW, so that they can perform subsequentretransmissions. Nodes having problems with their childrentransmit packets with the poison flag activated. An STA isconsideredpoisoned if, before transmitting an outgoing datapacket, one of the following conditions is satisfied:

• The STA is part of a poisoned path; i.e., it has receivedone or more packets with the poison flag activated duringthe current transmission window.

• The STA has not received any data packet from one ormore of its children.

• The STA has not received all the expected segments fromone or more of its children.

In Figure 6, nodeN3 activates its poison flag after notreceiving data from its childN6. In its way to the GW, a datapacket fromN3 poisons its next hop:N1. Therefore, nodesN6, N3, andN1 form a poisoned path, as shown in Figure 7.

3) Transmission windows: A number of transmission win-dows (w) with their corresponding e2e ACK are included in

7

sleep

sleep

sleep

awake

sleep

sleepsleep

sleepsleep

awake

sleep

sleep

sleep

sleep

sleep sleep

sleep

awake

sleep

RX e2eACK

RX e2eACK

TXe2e ACK

RX Time Slot

TX Time Slot

RX e2eACK

RX e2eACK

RX e2eACK

TXe2e ACK

RX e2eACK

RX e2eACK

ACK

ACK

RX Time Slot

TX Time Slot

ACK

ACK

RX Time Slot

TX Time Slot

ACK

ACK

RX Time Slot

TX Time Slot

ACK

ACK

RX Time Slot

TX Time Slot

ACK

ACK

RX Time Slot

TX Time Slot

ACK

ACK

RX Time Slot

TX Time Slot

ACK

ACK

N1 (ring 1)

GW

N3 (ring 2)

transmission window #1 transmission window #2

N6 (ring 3)

N12 (ring 4)

X

correct transmission

poisoned transmission

incorrect transmissionX

Unicast transmission

Unicast reception

Broadcast transmission

Broadcast reception

Fig. 5. Uplink data transmission phase in a multi-hop LPWAN running HARE protocol stack with the network topology from Figure 6. Note the communicationproblems in the first transmission window between nodesN6 andN3.

GW

R1 R

2R3

N1

N3

N2

N6

N7

R4

N4

N5

N9

N10

N12

N13

N14

N8

N11

poisoned path

Fig. 7. State of the network from Figure 6 after the corresponding e2e ACK.Note thepoisoned path passing through nodesN6, N3, andN1. Togetherwith the GW, these nodes (colored in red) stay awake during the secondtransmission window. The rest of nodes (colored in green) goto sleep as theyare not involved in the new transmission process.

each uplink data transmission phase to ensure correct packetreception. Within these windows, not all STAs remain awake,but only the ones directly involved in the transmission process.Before the start of a new transmission window, STAs evaluatewhether they shall stay awake or go to sleep.

This decision takes into account if the STA has beenpreviouslypoisoned by one of its children as well as severalother conditions according to the decision flowchart fromFigure 8. Whenever an STA decides to go to sleep, it willremain in this state until the nextprimary beacon.

4) Distributed caching: Due to the structure of multi-hopnetworks, lost packets cause expensive retransmissions alongevery hop of the path between the sender and the receiver [21].

Poisoned?

Stay awake

All segments acknowledged by my parent?

NOYES

NOYES

Appeared in the e2e ACK?

NOYES

More than one segment sent in the last transmission

window?

YES NO

Stay awake

Stay awake Go to sleep

Go to sleep

*Retransmit segments not yet acknowledged by my parent

Fig. 8. STA’s decision flowchart to stay awake or go to sleep before the startof a new transmission window.

To alleviate this problem, a distributed caching system is usedin HARE, so that parents acknowledge the correct reception ofpackets from children and cache their data until it is properlyreceived in the GW.

As it can be seen in Figure 7, nodesN12 and N13 cango to sleep after the first transmission window, because theirdata packets have been acknowledged by nodeN6, which willcache them in memory together with its own data to be sentin the next transmission window.

V. TESTBED

Contiki 3.0 OS [22] was the selected RTOS to validatethe HARE protocol stack, mainly due to its ability to eas-

8

ily execute multiple processes concurrently and its powerfulCOOJA network simulator [23]. Apart from simulations, twodifferent real platforms5 were used for preliminary testingand operational validation: MEMSICTM TelosB 2.4 GHz nodes[24] and ZolertiaTM RE-Mote 868 MHz nodes [25], whosemain features are depicted in Table IV.

HARE protocol stack has been programmed as an additionalmodule for Contiki 3.0 OS which interacts with the alreadyavailable upper communication layers of the system (MACand Network), regardless the employed hardware. Specificinteractions of HARE with PHY layers of the aforementionedhardware were separately programmed.

TABLE IVMAIN FEATURES OF THE HARDWARE EMPLOYED IN THEHARE

OPERATIONAL VALIDATION

Platform MEMSIC TelosB Zolertia RE-MoteMicroprocessor TI MSP430 ARM Cortex-M3Radio Module TI CC2420 TI CC1200Frequency Band 2.4 GHz 868/915 MHz

Performance evaluation of HARE protocol stack was per-formed in a testbed located on the 2nd floor, right wing of theTanger building at UPF facilities6. The testbed consisted of 13ZolertiaTM RE-Mote nodes (one of them acting as a gatewayand connected to a PC) running the HARE protocol stack.

All tests were executed considering no mobility and with thesame STAs’ placement (see Figure 9). All STAs were poweredby an 800 mAh battery except the gateway, which was perma-nently powered by the PC. Results were directly obtained fromthe GW, or thanks to thestatistics messages periodically sentby STAs. These messages contain information about differentmetrics such as the number of packets sent and acknowledged,RTT delays, as well as power profiles of microprocessor andradio module.

The calculation of total energy consumption(ET) is basedon these two power profiles:EµP and ERADIO, for the mi-croprocessor and the radio module, respectively, as shown inEquation (2).VDD is the supply voltage, whilet and I are,respectively, the time and the current corresponding to theoperational states of the microprocessor and the radio moduleof the employed hardware, whose values are summarized inTable V. Notice that theITX value of the radio module dependson the transmission power level.

ET = EµP + ERADIO

EµP = VDD · (tCPU · ICPU+ tLPM · ILPM)

ERADIO = VDD · (tRX · IRX + tTX · ITX + tSL · ISL) (2)

In addition, different network configurations were applied.Firstly, two different MAC layers inherent to Contiki OSwere tested: NULLMAC and X-MAC [27]. While NULLMACmaintains STAs continuously awake duringactive periods,X-MAC combines the introduction of sleeping periods forreceivers with the use of strobed preambles for senders.

5See Contiki main web page (http://www.contiki-os.org/) for a comprehen-sive table of hardware compatible with Contiki 3.0 OS

6UPF communication campus main website:https://www.upf.edu/campus/en/comunicacio/tanger.html

TABLE VCURRENT VALUES OF THEZOLERTIA RE-MOTE DIFFERENT

OPERATIONAL STATES(FROM [26])

Operational state CurrentMicroprocessorARM Cortex-M3

Processing (CPU) ICPU = 13mALow power mode (LPM) ILPM = 0.4µA

Radio ModuleTI CC1200

Receiving (RX) IRX = 19mATransmitting (TX) ITX = 39− 61mA

Sleeping (SL) ISL = 0.12µA

Secondly, and always over the same node deployment, twodifferent network topologies were tested: single-hop and multi-hop. In the first case, all nodes were directly connected to theGW, while in the second case, STAs were free to establish theirown routes to the GW with the single limitation of having 5children per STA.

And thirdly, the whole system was altered with the arbitrar-ily introduction of a certain error probability when sendingboth application packets and their corresponding ACKs (itis worth noting here that neither messages implied in theassociation process norstatistics packets were affected byarbitrary generated errors). Errors were generated througha uniformly distributed random variable according to meanerror values from Table VI. Before sending a message, STAscomputed this value and discarded messages accordingly. Forthis purpose, four different error configurations were defined.

The addressing system followed the Rime format [28]consisting of two 8-bit numbers. Similarly to IP addressing,the use of netmasks leads to flexible subnetting configurationswith up to (216 − 2) STAs. In our particular case, the first 8-bit number identified the network prefix shared by all devices,and the second one the host part, whose value for GWs was0 and for STAs was selected from 1 to 255.

All tests began with anetwork association primary beaconin which all nodes tried to associate to the network. Fromthen on, the GW emitted a new (network association or data)primary beacon every Tp = 3 min. Data primary beaconscould ask STAs for a newapplication or statistics packet. In allour tests,application andstatistics packets generated by STAscontained, respectively, 10 and 20 bytes of net information7.

VI. RESULTS

A. Association process

To show the performance and the coherence of the pro-posed association process and its underlying routing, all STAswere forced to repeatedly renew every twoprimary beacons

7Implementation of IEEE 802.15.4 in Contiki OS increases theminimumlength of any transmitted packet up to 43 bytes after including headers and,if necessary, applying padding

TABLE VIDEFINITION OF ERROR CONFIGURATIONS FOR THE PROPOSED TESTBED

Error Config. Data Error ACK ErrorE0/0 0% 0%E10/5 10% 5%E20/10 20% 10%E30/15 30% 15%

9

Fig. 9. Nodes’ placement at UPF facilities and association diagram when each STA admits up to 5 children.

(Npr = 2) their association to the network and compute theirbest parent according to (1) with the following parameters:a1 = a2 = 10, a3 = 1, anda4 = 5. In addition, the numberof children per STA was artificially limited to5 to guaranteemultiple paths towards the GW. Intersperseddata primarybeacons were used to check the reliability of routing paths andto allow not yet associated STAs to have another opportunityto join the network.

The selected underlying MAC for all STAs was X-MACand no error was introduced in the network (i.e.,E0/0 errorconfiguration was used). Under these premises, and after 200repetitions, an average number of11.97 STAs were associatedto the network after thedata primary beacon of the givensequence (i.e.,99.75% of success). As for the packet deliveryratio (PDR), it achieved100% in all the associated STAs.

Routing tables compiled by the GW were processed andadapted to graphical representation in Figure 9, where line’sthickness is proportional to link’s frequency appearance.Pref-erence of STAs for establishing paths with closer neighbours intheir way to the GW becomes evident, just like the importanceof clear paths (i.e., without obstacles) such as the formed bythe corridor walls.

The limitation of 5 children can be clearly appreciated inSTAs #6, #8, #9, #10, and#11 being almost always di-rectly connected to the GW in ring 1. The rest of STAs (princi-pally #7) could only access to that ring when circumstantiallyhaving better channel conditions than the aforementioned ones.

B. Reliability

Once all STAs are associated to the network and their pathsto the GW properly established, the next goal is to analyzethe reliability and the cost (in terms of energy consumption)of sending data. To do that, the GW was programmed tosend 20 beacons with the following sequence: beacon#1was a network association primary beacon, beacons#10and #20 were data primary beacons asking for statisticspackets; the rest of beacons weredata primary beacons asking

for application packets. To send their packets, STAs had 5available transmission windows(w = 5).

The results with the obtained PDR in all these configurationsare compiled in Figure 11. After 5 transmission windows,PDR is in any configuration above 95%, and it even achievesvalues above 90% after 3 and 4 transmission windows whenusing X-MAC and NULLMAC, respectively. In this case,NULLMAC specially suffers from the effect of collisions,due to the backoff implementation8 and the higher numberof concurrently active STAs compared to X-MAC.

Another insight from obtained results is how multi-hoptopology outperforms single-hop in all possible configurationsexcept when using X-MAC withE30/15. Again, the inherentreduction of concurrently active STAs competing for thechannel during the same time period (in this case, due to theallocation of STAs to different slots according to their ring)proves beneficial for system’s reliability.

The network’s ability to properly deliver data packets toits destination was also analyzed by computing the quotientbetween the total number of packets sent by STAs andthose properly received by the GW. As shown in Figure 10,multi-hop schemes still have better performance than single-hop in low-error configurations. On the contrary, in highlyunfavorable channels, parents usually do not receive all theirexpected payloads at once, so that they tend to send severalpackets in successive transmission windows with only partialinformation.

C. Energy consumption

The effect of this interdependence can also be observedin total energy consumption (Figure 12), computed after 20transmitted beacons (i.e., a 1-hour test). Important savings(up to 15%)9 can be achieved when using multi-hop schemes

8Main values of the NULLMAC CSMA/CA default backoff implementationin Contiki OS: minimum value of the backoff exponent (macMinBE = 0),maximum value of the backoff exponent (macMaxBE = 4), and maximumnumber of backoff attempts (macMaxCSMABackoffs = 5).

9From previous studies [15], we believe that in larger networks, these gainswill be much higher.

10

TABLE VIIAVERAGE LIFETIME OF AN 800MAH BATTERY IN THE PROPOSED TESTBED

Battery lifetime (days)Tp = 3min Tp = 1h Tp = 4h

NU

LLM

AC Single-hop

E0/0 2.37 47.35 187.87E10/5 2.21 44.19 175.41E20/10 2.18 43.46 172.53E30/15 2.14 42.70 169.53

Multi-hop

E0/0 2.63 52.51 208.17E10/5 2.44 48.60 192.79E20/10 2.24 44.66 177.27E30/15 2.07 41.35 164.23

X-M

AC

Single-hop

E0/0 4.46 88.75 349.64E10/5 4.27 85.00 335.07E20/10 4.52 89.99 354.46E30/15 4.56 90.79 357.55

Multi-hop

E0/0 5.29 105.17 413.16E10/5 5.08 101.04 397.21E20/10 4.53 90.10 354.85E30/15 4.39 87.38 344.31

with respect to single-hop ones in low-error configurations(E0/0 − E20/10) and similar or slightly worse values (lessthan 4% of extra consumption) inE30/15.

Time percentage of STAs’ microprocessor in low powermode is, in all studied cases, above 97% for X-MAC and99% for NULLMAC, due to the higher number of operationsinvolved in the first case. However, the impact of radio modulesleeping periods introduced by X-MAC layer reduces totalenergy consumption in up to 50% with respect to NULLMAC.In this case, values of energy consumed per bit of payloaddelivered are confined between50 − 65 mJ/bit for X-MAC,and105− 140 mJ/bit for NULLMAC.

As for the battery lifetime, Table VII compiles the durationin days of the 800mAh battery included in the Zolertia RE-Mote for the current testbed withTp = 3 min, as well astwo estimations withTp = 1 h andTp = 4 h. The temporalflexibility of the TDMA-based system employed in HAREallows this kind of extrapolations, by assuming that, in non-active time periods, both the microprocessor and the radiomodule remain asleep.

D. Resilience against failures

To prove the adaptability and resilience of the routingprotocol implemented in HARE, the network was subjectedto the deliberate shutdown of two of its STAs. In this way, theGW was programmed to send 50 beacons with the followingsequence: beacon#1 was a network association primarybeacon, beacons multiple of 10 weredata primary beaconsasking for statistics packets; the rest of beacons weredataprimary beacons asking forapplication packets. In addition,the disassociation mechanism was programmed in the GW toremove an STA from the network if not receiving any datapacket during oneprimary beacon (Npd = 1).

(a) Logical network topology afterthenetwork association primary bea-con

(b) Logical network topology frombeacon #15 until beacon #50

Fig. 13. Network topology before and after shutdown of nodes#1 and #4.

Once finished the initialnetwork association mechanism,the network was organized in four rings, as shown in Figure13(a). After beacon #4(A), STA #1 was switched off, but itdid not imply further problems to the network, as this STAdid not have any children. However, after beacon #12(B),STA #4 was also switched off, and it forced the network toreconfigure itself. The path to the GW of STAs #2, #3 and #5was broken, and they had to look for a new route by usingthe STA association mechanism of successivedata primarybeacons. After beacon #15, all active STAs (i.e., all of themexcept #1 and #4, which remain off) had a path to the GWand the network was again stable (see Figure 13(b)).

This test was also useful to analyze the performance of theproposedpower regulation mechanism when setting it withRSSImin = −110 dBm and RSSImax = −100 dBm. It isworth noting here that Zolertia RE-Mote devices use up to31 different power levels (from−16 dBm to 14 dBm withsteps of 1 dB [29]) and are programmed by default with themaximum transmission power level.

Figure 14 shows the clear reduction of transmission powerin most of the analyzed STAs during 50primary beacons,being the most significant examples STAs #7, #8, #9 and #10;the nearest ones to the GW. This fact results in a lower energyconsumption, asITX = 61 mA when transmitting at 14 dBm,but almost half (ITX = 39 mA) when doing it at -16 dBm.

The effects of switching off nodes are also visible in thetransmission power, as shown in(A) and (B) from Figure14. While STA #1 in (A) simply stopped working, nodesinvolved in the shutdown of STA #4 in(B) experienced notablechanges. Thus, STAs #2, #3 and #5disappeared along with theshutdown of STA #4. However, they became associated againbetween beacons #13 and #15 with maximum transmissionpower. For its part, when STA #6 became parent of STA #2,it set the maximum power level to establish connection withits new child.

Lastly, the power regulation mechanism proved its goodperformance against channel alterations as shown in area(C). In this case, and due to the test execution on a realscenario, the presence of people in the floor corridor mayhave disturbed channel conditions. To overcome this situation,some STAs (#9, #10, #11 and #12) selected temporarily greatertransmission power levels that were reestablished once finishedthe detected channel issues.

11

VII. C ONCLUSIONS

Single-hop topologies have become thede facto system totransmit data in current LPWANs mainly due to the need fornetwork simplicity and robustness, and the fear of consum-ing too much energy in processing tasks and/or maintainingcomplex routing mechanisms. However, the HARE protocolstack presented in the current article proves the suitabilityof alternative uplink multi-hop communication approachesinLPWANs. Distributed among three OSI layers (MAC, networkand transport), the multiple mechanisms contained in HAREensure network reliability and resilience against failures inuplink transmissions while keeping low energy consumption.

Results from a real testbed show uplink PDR values above95% for all considered configurations, with faster achievementof this level when using multi-hop topologies with multipletransmission windows. In addition, multi-hop topologies out-perform single-hop ones in terms of energy consumption inthe considered non error-prone scenarios, with up to 15%improvement (which could be even much higher in largernetworks) and values as low as50mJ/bit when employing anunderlying duty-cycled MAC layer. Similarly, network auto-configuration and resilience have been successfully put to thetest after forcing the shutdown of some network STAs.

In the near future, LPWANs are foreseen to occupy a centralrole in applications requiring to interconnect low-bandwidthdevices, focusing on range and power efficiency. While rangecoverage is mostly an issue from the physical layer, futurechallenges regarding power efficiency will surely encompassthe coordination of different layers and even the inclusionofnovel cross-layer mechanisms.

ACKNOWLEDGMENTS

This work was partially supported by the Catalan govern-ment through the project SGR-2014-1173 and also receivedfunding from the European Unions Seventh Framework Pro-gramme for research, technological development and demon-stration under grant agreement no 605073 (ENTOMATIC).

REFERENCES

[1] N. Jones, “Top 10 IoT technologies for 2017 and 2018,” Gartner, Inc.,Tech. Rep., 2016.

[2] A. Markkanen, “LPWA: Disruptive new networks for IoT,” M2MStrategies Reports, Machina Research, Tech. Rep., 2015.

[3] LoRa Alliance. LoRa Alliance Wide Area Networks for IoT.https://www.lora-alliance.org/. Accessed: 2016-09-14.

[4] SIGFOX. Sigfox main webpage. https://www.sigfox.com/. Accessed:2016-09-14.

[5] Ingenu. Ingenu main webpage. http://www.ingenu.com/.Accessed:2016-12-29.

[6] Weightless SIG. Weightless - Setting the Standard for IoT.http://www.weightless.org/. Accessed: 2016-09-14.

[7] DASH7 Alliance. DASH7 Alliance main webpage. http://www.dash7-alliance.org/. Accessed: 2016-09-14.

[8] ETSI. ETSI Low Throughput Networks.http://www.etsi.org/technologies-clusters/technologies/low-throughput-networks. Accessed: 2016-12-30.

[9] T. Adame, A. Bel, B. Bellalta, J. Barcelo, and M. Oliver, “IEEE802.11 ah: the WiFi approach for M2M communications,”WirelessCommunications, IEEE, vol. 21, no. 6, pp. 144–152, 2014.

[10] GSMA. Extended Coverage GSM Internet of Things (EC-GSM-IoT).http://www.gsma.com/connectedliving/extended-coverage-gsm-internet-of-things-ec-gsm-iot/. Accessed: 2016-12-29.

[11] ——. Long Term Evolution Machine Type Communication: LTE-MTC(CatM1). http://www.gsma.com/connectedliving/long-term-evolution-machine-type-communication-lte-mtc-cat-m1/. Accessed: 2016-09-14.

[12] ——. Narrow Band Internet of Things (NB-IoT).http://www.gsma.com/connectedliving/narrow-band-internet-of-things-nb-iot/. Accessed: 2016-12-29.

[13] C. Goursaud and J.-M. Gorce, “Dedicated networks for IoT: PHY/MACstate of the art and challenges,”EAI endorsed transactions on Internetof Things, 2015.

[14] A. Laya, C. Kalalas, F. Vazquez-Gallego, L. Alonso, andJ. Alonso-Zarate, “Goodbye, aloha!”IEEE Access, vol. 4, pp. 2029–2044, 2016.

[15] S. Barrachina, B. Bellalta, T. Adame, and A. Bel, “Multi-hop Commu-nication in the Uplink for LPWANs,”arXiv preprint arXiv:1611.08703,2016.

[16] I. Potamitis, I. Rigakis, and N.-A. Tatlas, “Automatedsurveillance offruit flies,” Sensors, vol. 17, no. 1, p. 110, 2017.

[17] D. Chen and P. K. Varshney, “QoS Support in Wireless Sensor Networks:A Survey.” in International Conference on Wireless Networks, vol.13244, 2004, pp. 227–233.

[18] M. O. Farooq and T. Kunz, “Operating systems for wireless sensornetworks: A survey,”Sensors, vol. 11, no. 6, pp. 5900–5930, 2011.

[19] C. Cano, B. Bellalta, A. Sfairopoulou, and M. Oliver, “Low energyoperation in WSNs: A survey of preamble sampling MAC protocols,”Computer Networks, vol. 55, no. 15, pp. 3351–3363, 2011.

[20] A. Keshavarzian, H. Lee, and L. Venkatraman, “Wakeup scheduling inwireless sensor networks,” inProceedings of the 7th ACM internationalsymposium on Mobile ad hoc networking and computing. ACM, 2006,pp. 322–333.

[21] A. Dunkels, J. Alonso, T. Voigt, and H. Ritter, “Distributed TCP cachingfor wireless sensor networks,”SICS Research Report, 2004.

[22] A. Dunkels, B. Gronvall, and T. Voigt, “Contiki-a lightweight andflexible operating system for tiny networked sensors,” inLocal ComputerNetworks, 2004. 29th Annual IEEE International Conference on. IEEE,2004, pp. 455–462.

[23] F. Osterlind, A. Dunkels, J. Eriksson, N. Finne, and T. Voigt, “Cross-level sensor network simulation with COOJA,” inProceedings. 200631st IEEE Conference on Local Computer Networks. IEEE, 2006, pp.641–648.

[24] Memsic. (2016) Telosb mote platform datasheet.http://www.memsic.com/userfiles/files/Datasheets/WSN/telosb datasheet.pdf.Accessed: 2016-07-19.

[25] Zolertia. (2016) Zolertia remote platform datasheet.https://github.com/Zolertia/Resources/blob/master/RE-Mote/Hardware/Revision%20A/Datasheets/ZOL-RMA001%20-%20RE-Mote%20Rev.A%20datasheet%20(draft)%20Dec%202015.pdf.Accessed: 2016-07-19.

[26] Texas Instruments. CC2538 Powerful Wireless Microcontroller System-On-Chip for 2.4-GHz IEEE 802.15.4, 6LoWPAN, and ZigBee Applica-tions. http://www.ti.com/lit/ds/symlink/cc2538.pdf. Accessed: 2016-12-13.

[27] M. Buettner, G. V. Yee, E. Anderson, and R. Han, “X-MAC: ashortpreamble MAC protocol for duty-cycled wireless sensor networks,” inProceedings of the 4th international conference on Embedded networkedsensor systems. ACM, 2006, pp. 307–320.

[28] A. Dunkels, “Rime, a lightweight layered communication stack for sen-sor networks,” inProceedings of the European Conference on WirelessSensor Networks (EWSN), Poster/Demo session, Delft, The Netherlands.Citeseer, 2007.

[29] Texas Instruments. AN125 - CC2538 + CC1200 Evaluation Module.http://www.ti.com/lit/an/swra437/swra437.pdf. Accessed: 2016-12-13.

12

1

1,25

1,5

1,75

2

2,25

2,5

# T

rans

mis

sion

s pe

r pa

cket

rec

eive

d

Error configuration

NULLMAC (Single−hop)NULLMAC (Multi−hop)X−MAC (Single−hop)X−MAC (Multi−hop)

E0/0 E

10/5 E20/10

E30/15

Fig. 10. Number of transmissions per packet received in the proposed testbed.

1 2 3 4 53035404550556065707580859095

100

# Transmission window

PD

R (

%)

E0/0

E10/5

E20/10

E30/15

Single−hopMulti−hop

(a) NULLMAC

1 2 3 4 53035404550556065707580859095

100

# Transmission windowP

DR

(%

)

E0/0

E10/5

E20/10

E30/15

Single−hopMulti−hop

(b) X-MAC

Fig. 11. Packet delivery ratio in the proposed testbed for different MAC layers, network topologies, and error probabilities.

0

25

50

75

100

125

150

175

200

225

250

Error configuration

Tot

al e

nerg

y co

nsum

ptio

n pe

r S

TA

(J)

NULLMAC(Single−hop)NULLMAC(Multi−hop)X−MAC(Single−hop)X−MAC(Multi−hop)

E0/0

E10/5

E20/10

E30/15

Fig. 12. Average total energy consumption per STA after 20 beacons whenTp = 3min.

Maximum transmission power (14 dBm)

Minimum transmission power (-16 dBm)

C

Fig. 14. Temporal evolution of STAs’ transmission power level.