Content uploaded by Hyung-sin Kim
Author content
All content in this area was uploaded by Hyung-sin Kim on Jan 02, 2018
Content may be subject to copyright.
34
Reliable and Energy-Efficient Downward Packet Delivery
in Asymmetric Transmission Power-Based Networks
HYUNG-SIN KIM and MYUNG-SUP LEE, Seoul National University
YOUNG-JUNE CHOI and JEONGGIL KO, Ajou University
SAEWOONG BAHK,SeoulNationalUniversity
In low-power wireless networks, maintaining multihop connectivity is considered effective in constructing
communication routes between individual nodes to a gateway. Since sensor networks are typically used for
data collection, multihop routing protocols are designed to find routes optimal in upward directions. As sen-
sor networks become widely applied to diverse applications, efficient downward traffic delivery also becomes
important. To achieve this, we consider an asymmetric transmission power-based network (APN), where
a power-supplied gateway uses high-power radios to cover the entire network via single-hop transmission,
whereas common nodes use low-power transmissions. For effective APN operations, we propose a single-hop
downlink protocol (SHDP) that consists of direct downlink transmission, local acknowledgment, neigh-
bor forwarding, and contention resolution among the destination’s neighbors. We evaluate SHDP through
mathematical analysis, simulations, and testbed experiments. Our proposal outperforms other competitive
multihop routing protocols. Specifically, SHDP shows high packet delivery performance and lowers the duty
cycle greatly while reducing the packet transmission overhead by >50%.
CCS Concepts: rNetworks →Network design principles
Additional Key Words and Phrases: Low power and lossy network, wireless sensor network, asymmetric
power network, RPL, downward packet delivery, energy efficiency
ACM Reference Format:
Hyung-Sin Kim, Myung-Sup Lee, Young-June Choi, Jeonggil Ko, and Saewoong Bahk. 2016. Reliable and
energy-efficient downward packet delivery in asymmetric transmission power-based networks. ACM Trans.
Sen. Netw. 12, 4, Article 34 (September 2016), 25 pages.
DOI: http://dx.doi.org/10.1145/2983532
For H.-S. Kim, M.-S. Lee, and S. Bahk, this work was supported in part by the National Research Foundation
of Korea (NRF) grant funded by the Korea government (MSIP) (No. 2015R1A2A2A01008240) and in part by
the ICT R&D program of MSIP/IITP [B0717-16-0026].
For Y.-J. Choi, this research was supported by the Basic Science Research Program through the National
Research Foundation of Korea (NRF) funded by the Ministry of Education (2015R1D1A1A01058025).
For J. Ko, this research was supported by a grant of the Korea Health Technology R&D Project through
the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare,
Republic of Korea (grant number: HI16C0982).
S. Bahk is the corresponding author.
An earlier version of this article appeared in the Proceedings of the IEEE International Conference on
Communications (ICC’14), June 2014 [Kim et al. 2014].
Authors’ addresses: H.-S. Kim, M.-S. Lee, and S. Bahk, Department of Electrical and Computer Engineer-
ing, Seoul National University, 1 Gwanak-ro, Gwanak-gu, Seoul, Korea 151-742; emails: {hskim, mslee}@
netlab.snu.ac.kr, sbahk@snu.ac.kr; Y.-J. Choi and J. Ko, Department of Software and Computer Engineering,
206 Worldcup Ro, Yeongtong-Gu, Suwon, Korea 16499; emails: {choiyj, jgko}@ajou.ac.kr.
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted
without fee provided that copies are not made or distributed for profit or commercial advantage and that
copies show this notice on the first page or initial screen of a display along with the full citation. Copyrights for
components of this work owned by others than ACM must be honored. Abstracting with credit is permitted.
To copy otherwise, to republish, to post on servers, to redistribute to lists, or to use any component of this
work in other works requires prior specific permission and/or a fee. Permissions may be requested from
Publications Dept., ACM, Inc., 2 Penn Plaza, Suite 701, New York, NY 10121-0701 USA, fax +1 (212)
869-0481, or permissions@acm.org.
c
2016 ACM 1550-4859/2016/09-ART34 $15.00
DOI: http://dx.doi.org/10.1145/2983532
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:2 H.-S. Kim et al.
1. INTRODUCTION
Over the past decade, various wireless sensing systems have been designed, and one
of the common choices in system design was to use a homogeneous radio module for
all the nodes in the network. By doing so, wireless links could be considered to have
“close to” symmetric performance. This symmetry enables researchers to easily apply
techniques developed for various mobile ad hoc networks (MANETs) in the research
domain of wireless sensor networks (WSNs). Furthermore, this symmetric property
helped develop various low-power multihop routing protocols that operate under lim-
ited wireless capacity and energy for data delivery from sensor nodes to gateways that
are typically distantly located.
Examples of such low-power multihop routing protocols include the collection tree
protocol (CTP) [Gnawali et al. 2008] and the routing protocol for low-power and lossy
networks (RPL) [Ko et al. 2011; Winter et al. 2012], which are widely applied in sensor
network deployments [Ancillotti et al. 2013; Ceriotti et al. 2011; Ko et al. 2010]. These
protocols mostly focus on serving uplink traffic where data reporting takes up a major
portion. While downlink traffic is considered in protocols such as RPL, the efficiency
is usually dependent on the uplink quality, since the reverse of them is used as the
downlink.
However, as wireless sensing system applications become diverse, their traffic types
also start to vary. Considering various application services, simply exploiting symmet-
ric wireless links may not be considered a good design choice. For example, with an
application that mostly generates downward traffic (e.g., from the gateway to sensor
nodes), managing multihop routes at each node in dynamic wireless environments will
result in a significant amount of control overhead. Moreover, multihop routing protocols
naturally require each node to have a designated memory space to store and manage
these routes [Clausen et al. 2011].
In this article, we address such inefficiencies in multihop routing protocols for appli-
cations with a high emphasis on downward traffic and propose an asymmetric trans-
mission power network (APN). In the APN, a gateway node with a high power radio
maintains single-hop connectivity to many low-powered nodes, which are deployed
distantly for application-specific purposes. The gateway node is typically connected
to a power source and takes the role of distributing information to the low-power
nodes.
Nevertheless, in order to ensure reliable packet delivery in APNs, a low-powered
destination node needs to send an acknowledgment (ACK) packet toward the gateway
for each received packet in a multihop manner. This action of multihop ACK transmis-
sions incurs severe uplink traffic and high energy consumption. As a way of achieving
both reliable and energy-efficient packet delivery with the single-hop downlink connec-
tivity, we propose a protocol, termed as the single-hop downlink protocol (SHDP)[Kim
et al. 2014], which exploits the single-hop downward transmission capability of the
gateway, local ACK exchange between the destination and its neighbors, and neighbor
forwarding to ensure best-effort packet delivery.
In SHDP, the destination node is not required to send ACK packets to the gateway
directly. Instead, SHDP achieves reliability by allowing the destination and its neigh-
bors to exchange ACK packets locally, and then the neighbors forward any missing
packets opportunistically upon detecting a transmission failure. As a result, each low-
power node transmits an ACK packet toward its neighbors upon a successful packet
reception destined for itself, and keeps track of neighbor-initiated ACK packets by
overhearing the wireless channel. When a transmission failure for a node is detected
(e.g., lack of ACK packet), neighbors participate in local retransmission after taking
random backoffs.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:3
In our testbed experiments, SHDP successfully achieves a 99.4% downward packet
reception ratio (PRR) in dynamic wireless environments. While this PRR is only a
4% increase over the RPL routing protocol, SHDP reduces the relative transmission
overhead by more than 50%, leading to a lower radio duty cycle. Furthermore, our
experimental results show that the performance of the “worst-case” node is significantly
improved with SHDP, implying that the fairness among nodes in the network can be
achieved to provide high-quality networking services for the longest duration possible.
The contributions of this work are threefold:
—We consider applications that mainly generate downward traffic and quantify limi-
tations of existing multihop routing protocols in delivering downward traffic.
—We design an APN system architecture where a gateway node uses a high-power
radio in communicating with many low-power sensor nodes that are deployed over
a wide geographical region. To ensure highly reliable and energy-efficient packet
delivery, we propose SHDP, where the gateway delivers packets to sensor nodes via
single-hop transmissions.
—Through mathematical analysis, simulations, and indoor testbed experiments, we
evaluate SHDP against existing multihop routing protocols extensively, considering
various channel and traffic conditions. By exploiting high transmission power of the
gateway, SHDP can lower the transmission overhead of other low- power nodes by
50%.
While this work is an extension of our previous work [Kim et al. 2014], the tech-
nical contribution differs in several ways. First, we provide a feasibility analysis and
introduce a number of application scenarios in which APNs can benefit the networking
performance. Second, we redesign SHDP so that it interoperates with the RPL rout-
ing protocol, allowing our APN to support single-hop downward and multihop upward
data transmissions. Third, we implement SHDP on real resource-constrained embed-
ded devices and evaluate its performance using an indoor testbed under various traffic
scenarios. Lastly, we present a mathematical analysis on the packet delivery ratio and
latency performance.
The rest of this article is structured as follows. In Section 2, we introduce the concept
of APN and present applications that can benefit from asymmetric transmission power
capabilities of nodes. We propose an APN architecture in Section 3 and provide its an-
alytical evaluations in Section 4. Section 5 shows our simulation results, and Section 6
presents the results obtained from the indoor testbed implementation. We conclude
with a discussion of our results in Section 8.
2. ASYMMETRIC TRANSMISSION POWER NETWORKS AND THEIR APPLICATIONS
Before discussing the details of SHDP, we start with introducing the APN architecture
and consider some applications that can benefit from utilizing the APN architecture.
This work targets a specific APN architecture where the gateway node uses high-power
transmissions and energy-limited sensor nodes operate with low-power constraints.
While the hardware configuration of a sensor node is simple, the heterogeneity between
the gateway and common sensor nodes requires changes in the protocol design.
2.1. APN Characteristics
Assume that PL(d)represents the path loss of a channel as a function of the distance d
between a sender and a receiver. Let us denote the transmission power at the gateway
and the received power at a receiver as Pand pr, respectively. Then, we have
pr=P
PL(d).(1)
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:4 H.-S. Kim et al.
Fig. 1. Enabling factor for an APN.
Representing the bit error rate (BER) according to the received signal-to-noise ratio
(SNR) γas BER (γ), we can express the packet error rate PER as
PER (γ)=1−(1−BER (γ))B,(2)
where Bis the packet size in bits, γ=pr/n0,andn0is the magnitude of noise. Denoting
a given PER requirement and minimum received power required to meet the receiver
sensitivity as eth and pr,th, respectively, we have PER (γth)=eth, where γth =pr,th/n0.
We use high transmission power for a WSN, which mainly considers low-power com-
munications. This requires checking whether low-power transceivers, widely used in
WSNs (e.g., CC2420 [Texas Instruments 2006]), can accept high-power packet trans-
missions without receiver malfunctioning caused by power saturation. Let us denote
the receiver saturation power as psat, which indicates that the receiver malfunctions
when its receiving power is greater than psat (i.e., receiver saturation). Then, the con-
dition for the receiver to decode a packet from the gateway successfully satisfies
pr,th ≤pr≤psat.(3)
We define the feasible transmission distance Rfs and the transmission range Rtx as
Rfs =min{d|pr≤psat },(4)
Rtx =max{dpr≥pr,th }.(5)
Figure 1 illustrates Rfs and Rtx according to the transmission power P.Inthis
example, we assume that psat =10dBm following the data sheet of CC2420 [Texas
Instruments 2006]. The values of eth =0.1and pr,th =-87dBm have been confirmed
through extensive experiments using CC2420 [Chen and Terzis 2011; Srinivasan et al.
2010]. In addition, we assume that PL(d) follows the indoor path loss model of the
IEEE 802.15.4 standard [IEEE Std 802.15.4-2003 2003] given by
PL(d)=104.02+2log
(d),for d≤8
105.85+3.3log
(d/8),otherwise.(6)
We consider transmission power in the range from 0dBm to 30dBm in accordance with
the maximum transmission power of CC2420 [Texas Instruments 2006] and the FCC
regulations for the 2.4GHz band [Federal Communications Commision 1993].
Figure 1(a) shows that the feasible distance is negligibly small since it is shorter
than 10cm even when the maximum transmission power is applied. Figure 1(b) shows
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:5
that the transmission range of the gateway that uses the Wi-Fi transmission power of
17dBm [IEEE Std 802.11-2009 2009] (i.e., ∼191 meters ) is 3.3 times longer than that
of a low-power node (i.e., ∼58 meters).
These results confirm that high-power transmissions at the gateway can connect a
wide area via a single hop, while not power saturating the bandwidth for low-power
transceivers. Overall, this result validates the applicability of APNs in low-power WSN
deployments. We emphasize that, although this physical feasibility check does not
guarantee the usefulness of APNs in practice, it is necessary to have a solid motivation
before designing an APN-based protocol.
2.2. APN Applications
Compared to conventional WSNs in which all nodes use the same transmission power,
the main benefit of an APN is in providing single-hop delivery for downward traffic
without configuring a multihop downward route. To follow are several applications that
can take advantage of the APN architecture.
—Electronic Price Management: Product prices at supermarkets and retail stores
may change frequently depending on the time of sale and prices at competing vendors.
Currently, most of the retagging processes are done manually, which results in added
labor cost and frequent customer claims due to mis-tagging [Aruba Networks 2012].
Electronic price tagging systems send a unicast packet containing updated price
information to each target node (i.e., price tag with a radio module). Within an APN,
price updates can be performed with minimal effort by broadcasting through a high-
power gateway node.
—Emergency Message Broadcasting: Various applications are designed to alert
inhabitants in a target environment where emergency situations arise [Lorincz et al.
2004]. To achieve rapid delivery of emergency messages, flooding can be employed,
which requires that all nodes participate in forwarding [Marot et al. 2004; Werner-
Allen et al. 2006]. The resultant multihop latency, coupled with duty cycling of nodes
to conserve energy, can degrade overall system performance [Wang and Liu 2009;
Yildirim and Kantarci 2014]. APN provides an efficient alternative for propagating
alert messages.
—Mobile Applications: With emerging cyber physical systems (CPSs) and the In-
ternet of Things (IoT), mobility becomes an integral part of the low-power wireless
networks. These applications include robotic networks and military-related sensing
applications. While various network protocols have been proposed and designed, fre-
quent loss of link connectivity due to node mobility makes the bidirectional route
establishment challenging [Fink et al. 2012]. In APNs, mobile nodes need to keep
their routing entries only toward the gateway, since downward transmissions are
completed using a high-power single-hop link.
While the concept of APN is not new, design of an effective network architecture
and performance evaluations in this work introduce new challenges when compared to
traditional single- or multihop networks. Next we introduce our SHDP scheme, which
aims to provide an efficient best-effort packet delivery for APN architectures.
3. DESIGN OF SHDP
3.1. SHDP Overview
In the proposed SHDP, each node (except the gateway) is responsible for the lo-
cal ACK, neighbor forwarding, and forwarding contention mitigation processes. In a
multihop environment with a high-power transmitting gateway, an ACK from a low-
power destination node is typically unable to reach the gateway directly over single-hop
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:6 H.-S. Kim et al.
Fig. 2. An example of BoX-MAC-2 operation. BoX-MAC-2 achieves asynchronous packet delivery by exploit-
ing periodic wakeup at the receiver and repetitive packet transmission at the sender, incurring a tradeoff
between a sender and a receiver in energy consumption according to the wakeup interval.
transmission due to the physical distance. The problem is that without ACK reception,
the gateway cannot confirm whether the packet was delivered successfully.
It can be the case that the destination node transmits ACKs to the gateway over
multiple hops. However, this can take a long time to reach the gateway and incur
significant traffic overhead when issued for all data packets. To address this issue
while maintaining a low networking overhead, we propose to use local-ACKs, which
confirm packet delivery using the help of neighbors near the destination. For local-
ACKs, the destination’s neighbors overhear the downlink packet transmissions from
the gateway and also the (single-hop-transmitted) ACK from the destination node if
the packet is successfully delivered. If a transmission failure occurs and an ACK cannot
be overheard from the destination node, these neighbors who overhear the data packet
from the gateway retransmit the packet for the destination node. To minimize the
contention from this retransmission process as well, if a neighbor node realizes that
a different neighbor already retransmitted the data packet (again from overhearing),
the retransmission is suppressed.
Through such a procedure, the gateway can deliver packets to destinations reliably
despite failures in direct downlink transmissions. As a result of local-ACKs, SHDP
prevents the network from making multihop downward transmissions (with a high-
power root) and also prevents ACK packets from traveling over multiple upward hops
to reach the gateway.
3.2. SHDP Architecture and Implementation
3.2.1. Baseline MAC Protocol.
The design of SHDP is based on BoX-MAC-2 [Moss and
Levis 2008], which is the default MAC protocol of TinyOS and is widely used in various
WSN systems [Landsiedel et al. 2012; Puccinelli et al. 2012]. Figure 2 illustrates an
example of the asynchronous BoX-MAC-2 operations. Here, each node periodically
wakes up and checks for the channel-sensing period on whether the channel is busy or
not. At this point, if the node identifies no traffic on the channel, it goes back to sleep.
In case there are activities on the wireless channel, the node confirms if the packet is
intended to itself, and, if so, it receives that packet, sends an ACK, and goes back to
sleep to continue its periodic wakeup process.
A node with a packet to send performs channel sensing after a random backoff tbo
(0 ≤tbo ≤tbo,max ). If the channel is busy, it accesses the channel again after waiting for
an extended backoff time. If the channel is free, it repeats the process of random backoff
and packet transmissions continuously until an ACK is received. In other words, the
data packet is repetitively transmitted during a full wakeup interval.
Therefore, the channel-sensing period of the receiver should be longer than or equal
to the maximum interval between repetitive packet transmissions at the sender (i.e.,
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:7
Fig. 3. An example of direct transmission in SHDP. A forwarder node can opportunistically retransmit an
overheard downward packet to increase network reliability. To mitigate contention, each potential forwarding
node suppresses forwarding when detecting an ongoing forwarding process of the same packet by another
node.
(tack +tbo,ma x ), where tack is the ACK length) in order to allow the receiver to detect
whether the channel has an ongoing transmission or not.
In Figure 2, the receiver node receives and acknowledges the packet in the second
wakeup period. As we can see in the “neighbor of receiver” case in Figure 2, when
a node starts listening to the wireless channel in the middle of a transmission, it
continues to turn on its radio until the next repetitive transmission begins. As a result,
the maximum idle listening period becomes (tpacket +tack +tbo,max ), where tpacket is the
length of the data packet.
This repetitive transmission of a sender and the periodic wakeup of a receiver leads
to a successful packet delivery without synchronization. If the sender fails to receive
an ACK, it tries again to transmit the packet up to a maximum number of ntx times.1
BoX-MAC-2 has a tradeoff relation between a sender and a receiver in energy con-
sumption based on the wakeup interval. As the wakeup interval increases, the sender
consumes more energy due to the increase in the number of repetitive packet trans-
missions, while the receiver consumes less energy owing to the reduced frequency in
channel sensing. The proposed SHDP is designed to have low transmission overhead
by eliminating multihop transmission for downward traffic.
3.2.2. Gateway’s Direct Transmissions.
We aim to achieve reliable downward packet trans-
missions by allowing a destination’s neighbor nodes (rather than the gateway) to con-
firm the packet delivery by receiving ACK packets from the destination node. However,
BoX-MAC-2 cannot support ACK exchange between the destination and its neighbor
nodes since each node has no knowledge of when its neighbor nodes will be awake (or
asleep). To alleviate the problem, we design a new transmission mechanism for the
gateway, allowing nodes to wake up simultaneously and exchange local ACKs.
We describe the operation of our high-transmission power gateway using Figures 3
and 4, which show examples of a gateway’s successful direct transmission and neighbor
forwarding, respectively. The gateway node “0” transmits a packet to the destination
with a high transmission power. To allow the randomly waking-up destination node to
receive the packet, the gateway repetitively transmits the same packet for the entire
wakeup interval. Of course, due to the asymmetric nature of APNs, the gateway does
not expect to receive an ACK from the destination.
1One packet retransmission comprises multiple repetitive transmissions during a single wakeup interval.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:8 H.-S. Kim et al.
Fig. 4. An example of neighbor forwarding in SHDP. A forwarder node can opportunistically retransmit an
overheard downward packet to increase network reliability. To mitigate contention, the (potential) forwarder
node suppresses forwarding when detecting an ongoing forwarding process of the same packet by another
node.
To support local ACK exchange, the gateway includes a time index sfor each repeti-
tive transmission for the same packet. The index srepresents the time interval between
the start of the current repetitive transmission to the end of the full repetitive trans-
mission batch. To set s, we use a timer designed to manage repetitive transmissions
in BoX-MAC-2. This timer runs for the wakeup interval twakeup. The repetitive trans-
mission process starts with the timer and ends when it expires. For each repetitive
transmission, the gateway sets the index sas the currently remaining time for the
timer to stop and inserts sinto the packet header. The last repetitive packet transmis-
sion has s=0. To minimize the idle listening period of a sensor node, the gateway does
not use a random backoff between two repetitive packet transmissions (i.e., tbo =0).
3.2.3. Local ACK Exchange.
Due to the use of time index sin the packet header from
the gateway, the destination node can decide to exchange local ACKs with its neighbor
nodes as follows. When the destination wakes up and receives a packet, it first checks
the packet’s destination address. If the node is the packet’s destination (node 1 in
Figures 3 and 4), the node checks whether s=0 (i.e., the end of the gateway’s repetitive
packet transmission process). If s>0, the destination sleeps to save energy and wakes
up when s=0 is expected. The sleep interval here tsleep(s) is obtained by the interval
between the end time of the current packet transmission duration and the start time
of the last packet to be transmitted (i.e., s=0)2. This sleep interval can be represented
as
tsleep(s)=s−tpacket.(7)
Upon receiving the s=0 packet, the destination broadcasts an ACK for its neighbors
to overhear. Thus, the main difference between our proposal and the baseline BoX-
MAC-2 is that the destination waits until it receives the latest of the sender’s repetitive
transmissions before issuing an ACK packet.3
If a node is not the packet’s intended destination, the node checks whether the
destination of the packet is one of its neighbors. If so, the node operates identically to
2A small guard time is added in real implementations due to practical issues such as clock drift and per-node
processing delays [Pak et al. 2008].
3With the sequence number, we can allow the destination and its neighbors to simultaneously wake up for
receiving packet s=0 and stay awake until the exchange of an ACK. We can also configure the destination
to send an ACK when receiving the first packet with s>0, which allows neighbor nodes who receive this
ACK (early) to go into sleep mode as a way of conserving power.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:9
the destination node up to the point when it receives packet s=0. Once receiving the
last packet, the node listens to the channel to overhear the ACK from the destination
and goes to sleep once receiving the ACK (nodes 2 and 3 in Figure 3).
In the case in which the node is neither the destination nor one of the destination’s
neighbors, the node puts to sleep its radio for the entire wakeup interval (e.g., nodes 4–6
in Figure 3). Therefore, these nodes conserve their energy in the meantime.
3.2.4. Neighbor Forwarding.
As we show in Figure 4, when the destination fails to re-
ceive the last packet transmission from the gateway, neighbor nodes detect trans-
mission failure via a missing local ACK and try to forward the received packet. The
neighbor forwarding procedure is the same as a normal transmission of our baseline
MAC protocol (BoX-MAC-2). That is, a neighbor node forwards the packet repetitively
using a random backoff mechanism, but without using the time index s. From this,
the destination node realizes that the packet is a “forwarded packet” from one of its
neighbors. Once received at the final destination, an ACK exchange takes place. When
failing to receive an ACK from the destination (for the neighbor-forwarded packet), the
forwarder node retransmits the packet for a maximum number of ntx times, resulting
in a best-effort approach.
If multiple neighbors attempt packet forwarding simultaneously, a contention is-
sue occurs [Choi et al. 2006]. To resolve this, a forwarding node carrier senses for
ongoing transmissions before attempting the packet forwarding. Once a neighbor re-
alizes that a different node has already started a packet-forwarding attempt (e.g., via
overhearing before its transmissions), the node suppresses its message transmissions
under the assumption that the other node (already occupying the channel) will perform
its best effort in delivering the message. This simple forwarding suppression scheme
mitigates a large portion of contention that can occur in the neighbor-forwarding
phase.
4. PERFORMANCE ANALYSIS
In this section, we analyze SHDP with respect to the packet reception ratio, latency, and
power consumption. Consider a network of Ntot nodes that are randomly distributed
over a given area and ready to receive downward packets from the gateway that is
located at the center. Assume that each node has transmission range rand average
number of neighbor nodes Nne within the transmission range. We also assume that
each node can receive packets transmitted from another node within rand suffers
from PER e(≤eth). For mathematical tractability, we assume that each node, includ-
ing the gateway, transmits a packet without contention, collision, and queuing delay.
Thus, the analysis is for light traffic conditions with the general case addressed in the
experimental performance evaluation.
In SHDP, a node does not know when the gateway starts its transmission, so we
assume the time index sof the packet transmitted from the gateway is uniformly ran-
domly distributed between 0 and twakeup, in the perspective of each node. The probability
that the first received packet is the last one (i.e., s=0) becomes small as the wakeup
interval increases. This means that, in a low duty-cycle network, the destination and
its neighbor nodes are likely to receive two repetitive packet transmissions from the
gateway before local ACK exchange.
As a point of reference, we consider an ideal multihop downlink protocol (MHDP). In
MHDP, the gateway is assumed to know a downward (multihop) route with minimum
hop to each sensor node without running route discovery. Since MHDP is assumed to
have no control overhead, it achieves the best performance compared to conventional
multihop routing protocols. We compare SHDP against this optimal MHDP.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:10 H.-S. Kim et al.
4.1. Packet Reception Ratio
Let dhop (k) be the hop distance between the gateway and node kwhen the gateway uses
low power, which is given by dhop (k)=d(k)/r, where d(k) is the distance between the
gateway and node k.
Assuming that ntx is the maximum allowed number of retransmissions, the packet
reception ratio (PRR) over one-hop transmission, PRRone, is given by PRRone =1−entx.
Then the PRR at node kin MHDP becomes
PRRMHDP (k)=PRRdhop (k)
one ,(8)
which shows that PRRMHDP (k)exponentially decreases with dhop (k).
Unlike MHDP, all the nodes in SHDP are within one hop from the gateway. Thus,
the PRR in SHDP is the same for all the nodes, which is given by
PRRSHDP =PRRdirect +PRRindirect.(9)
PRRdirect is the probability that a packet from the gateway is delivered in one-hop
transmission, that is, PRRdirect =1−e.PRRindirect represents the probability that the
packet is indirectly delivered. Indirect packet delivery happens when the destination
failed to receive the gateway’s transmission but at least one of its neighbors successfully
overheard two repetitive packet transmissions from the gateway. Hence, its probability
is 1 −(1 −(1 −e)2)Nne . By mitigating forwarding contention, a forwarding neighbor
accesses the medium and transmits the packet to the destination successfully with
probability PRRone. Thus, PRRindirect can be written as
PRRindirect =e1−(1 −(1 −e)2)Nne PRRone.(10)
Obviously, indirect transmission significantly enhances reliability when Nne is large.
In a fully connected network, each node has at least one neighbor node, which implies
PRRindirect ≥e(1−e)21−entx .(11)
This shows that SHDP can achieve reliable packet delivery with high probabil-
ity through local ACK exchange and neighbor forwarding without the gateway’s
confirmation.
4.2. Latency
A successful packet transmission incurs average latency twakeup/2 to receive an ACK
successfully, which increases by twakeup whenever a retransmission happens. If a packet
is delivered after having jtransmissions, the latency is given as twakeup(j−0.5). Since
the probability that a packet is transmitted jtimes for successful delivery is (1 −e)ej−1,
the average latency for each hop is given as
Lone =twakeup(1 −e)ntx
j=1j−1
2ej−1
PRRone
.(12)
For packet delivery from the gateway to node kin MHDP, the average latency becomes
LMHDP (k)=dhop (k)Lone.(13)
Notice that LMHDP(k) linearly increases with dhop (k)andtwakeup .
In SHDP, the latency is the same for all the nodes. That is,
LSHDP =PRRdirect Ldirect +PRRindirect Lindirect
PRRSHDP
.(14)
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:11
Here Ldirect is the latency when the destination node succeeds in receiving the direct
transmission, denoted as Ldirect =twakeup/2, and Lindirect is the latency from the destina-
tion receiving the packet via neighbor forwarding. Furthermore, Lindirect =Lone +twakeup,
where Lone is the forwarding latency. We can notice here that LSHDP linearly increases
with twakeup. In addition, with a high PRRdirect,SHDP will result in low latency, given
that a lower PRRindirect can be expected and Ldirect <Lindirect due to the smaller number
of transmission attempts.
4.3. Power Consumption
We analyze the basic characteristics of the baseline protocol first, followed by the power
consumption performance of MHDP and SHDP.
4.3.1. Basic Protocol Analysis.
Assume that ptx,prx,pcs,and pidle represent the required
power for transmission, reception, channel sensing, and backoff time (or idle time),
respectively. Each node in BoX-MAC-2 periodically wakes up and consumes energy εcs
for channel sensing, which is
εcs =pcs(tack +tbo,max).(15)
The total power consumption for checking the wireless medium, pcs,tot, is given as
pcs,tot =Ntotεcs
twakeup
.(16)
pcs,tot represents the basic power consumption with no packet transmission or reception.
A node that has a packet to send performs channel sensing first, which consumes εcs.
The number of repetitive transmissions for successful and erroneous transmissions,
ntx,sand ntx,f, is given by
ntx,s=twakeup
2tpacket +tack +tbo,max/2,(17)
ntx,f=twakeup
tpacket +tack +tbo,max/2(18)
where xis the smallest integer not less than x. The energy consumptions for success-
ful and erroneous packet transmissions, εtx,sand εtx,f, are given by
εtx,s=εcs +ntx,sεtx,(19)
εtx,f=εcs +ntx,fεtx.(20)
εtx is the energy consumption for each (repetitive) transmission, which is given by
εtx =pidletbo,max
2+ptxtpacket +prxtack,(21)
since each (repetitive) transmission period comprises random backoff, packet trans-
mission, and ACK reception. Ignoring the collision factor, we obtain the total energy
consumed for a one-hop transmission (including retransmission) as
εtx,one =(1−e)
ntx−1
i=0εtx,s+iεtx,fei+ntxεtx,fentx .(22)
εtx,one increases with twakeup like εtx,sand εtx,f.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:12 H.-S. Kim et al.
We now consider the power consumption of a receiver. When a receiver wakes up and
performs channel sensing while the channel has an ongoing transmission, it can detect
packet transmission but cannot decode it. Thus, its idle listening should last until the
next repetitive transmission begins. This results in the average time for idle listening
of
tlisten =(tpacket +tack)(tbo,max +tpacket/2+tack /2)
tbo,max +tpacket +tack
.(23)
The total energy consumed until the receiver finishes its one-hop reception is
εrx,one =ptxtack(1 −entx )+prx(tpacket +tlisten)(1−e)
ntx
i=1
iei−1+ntxentx .(24)
If a sender transmits a packet, its neighbor nodes overhear it. Since a successful
packet transmission lasts for the period of twakeup
2on average, each neighbor of the sender
can overhear the packet with probability 0.5. For an erroneous packet transmission, all
the neighbors can overhear it. Thus, the energy consumptions required for overhearing
successful and erroneous packet transmissions, denoted as εoh,sand εoh,f, respectively,
can be given as
εoh,s=prxNne (tpacket +tlisten)
2,(25)
εoh,f=prxNne (tpacket +tlisten).(26)
The total energy consumption for overhearing a one-hop transmission becomes
εoh,one =(1 −e)
ntx−1
i=0
(εoh,s+iεoh,f)ei+ntxεoh,fentx.(27)
4.3.2. MHDP Analysis.
For MHDP, when a gateway transmits a packet to node k,Nho p(k)
senders and Nhop (k) receivers are involved in packet delivery. We will ignore the energy
consumption of the gateway since it is assumed to be connected to the power supply.
Then the total energy consumption for (Nhop (k)−1) senders and Nhop(k) receivers,
denoted as Etx
MHDP (k)and Erx
MHDP (k), respectively, can be expressed as
Etx
MHDP (k)=εtx,one
Nhop(k)−1
i=1
(1 −entx)i,(28)
Erx
MHDP (k)=εrx,one
Nhop(k)−1
i=0
(1 −entx)i.(29)
Each node within the sender’s transmission range overhears the packet transmission
by incurring more energy consumption. The total energy consumption for overhearing
a one-hop transmission while the gateway transmits a packet toward node kbecomes
Eoh
MHDP (k)=εoh,one
Nhop(k)−1
i=0
(1 −entx)i.(30)
Obviously, Etx
MHDP (k),Erx
MHDP (k),andEoh
MHDP (k)increase with Nhop(k) since the number
of relaying nodes increases. Nodes near the gateway consume more energy than those
far from the gateway due to the heavy burden imposed by relaying.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:13
Let us denote the packet interarrival time as tint, and assume that the gateway
generates packets toward each destination randomly. Then, the total power consumed
by all the nodes becomes
PMHDP =pcs,tot +1
tintNtot
Ntot
k=1Etx
MHDP (k)+Erx
MHDP (k)+Eoh
MHDP (k).(31)
εtx,one in Etx
MHDP(k) increases, pcs,tot decreases with twakeup , and the weight for εtx,one
increases with the network size. Thus, the MHDP is not appropriate for supporting
low-duty-cycled large-scale WSNs efficiently.
4.3.3. SHDP Analysis.
In SHDP, the energy consumption of all nongateway nodes is
the same given that they are within single-hop range of the gateway and have equal
chances for transmissions, receptions, and overhearings in the long term.
For each packet transmission from the gateway, while all nodes in the network over-
hear the packet, only neighbors of the destination node try to forward the overheard
packets, and only one of them finally participates in the packet delivery as a forwarder
node (i.e., the neighbor of the destination occupying the channel first for packet for-
warding). The total energy consumption for a complete transmission follows:
Etx
SHDP =(1 −(1−e)2)1−(1 −(1 −e)2)Nne εtx,one+
(1−e)2Nne −1pidletbo,max
2+εcs.(32)
The first term in brackets represents the energy consumption at the forwarding node
that occupies the medium first, and similarly the second term is the other neighbor
nodes, which initially try to participate in local retransmission but suppress its packets
after detecting an ongoing transmission attempt. While the latter do not transmit
packets, they consume energy from random backoff and channel sensing until packet
forwarding is detected.
Since the gateway exploits no backoff between repetitive transmissions, the mean idle
listening period at each node becomes tpacket/2. If a destination receives a packet whose
time index sis larger than 0, it sleeps again until the gateway transmits the packet
with s=0. The total energy consumption for a packet reception at the destination
becomes
Erx
SHDP =3prxtpacket
2+(1−e)prxtpacket +(1 −e)2ptxtack (33)
+(1 −(1 −e)2)1−(1 −(1 −e)2)Nne εrx,one.
The first, second, and third terms come from direct packet receptions at the destination,
and the last term results from the packet reception through neighbor forwarding.
In SHDP, all nodes overhear the gateway’s transmission first, and then, only the des-
tination’s neighbors continue sensing to confirm packet delivery. Therefore, we obtain
the total energy consumption for overhearing as
Eoh
SHDP =3prxtpacket Ntot
2+(1−e)Nne prx(tpacket +tack) (34)
+(1 −(1 −e)2)1−(1 −(1 −e)2)Nne εoh,one.
The first term here represents the energy consumption of all the nodes that overhear
the gateway’s transmission. The second term denotes the energy consumption of the
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:14 H.-S. Kim et al.
Fig. 5. Simulation network topology of 100 nodes.
destination’s neighbors, which overhear the s=0 packet and its ACK after success-
fully overhearing packet of index s(>0). The last term is the energy consumption for
overhearing the packet forwarded by a neighbor.
The total power consumption of all the nodes in SHDP can be written as
PSHDP =pcs,tot +Etx
SHDP +Erx
SHDP +Eoh
SHDP
tint
.(35)
As in the analysis of MHDP energy consumption, PSHDP is directly related with twakeup
due to pcs,tot and εtx,one. The weight for εtx,one in SHDP is lower than that of MHDP,
since SHDP works independently from the number of nodes and its relaying occurs only
when the gateway’s transmission fails. The low transmission overhead allows SHDP
to consume much lower energy with a large wakeup interval, especially in large-scale
WSNs, compared to other multihop protocols.
5. SIMULATION RESULTS
Based on our mathematical analysis, we now move on to performance evaluation of
SHDP and MHDP through simulation. Specifically, we developed an event-driven sim-
ulator that considers real channel environments according to the path loss model of
Equation (6), which is suitable for indoor environments [Kim 2013]. Simulation results
help us observe the performance of competitive schemes in terms of latency, packet
delivery ratio, and power consumption for a large network size, which is not easy for
experimental tests.
We configured a simulation environment similar to Kim et al. [2014]. We assume
that each low-power node uses a transmission power of 0dBm (i.e., the maximum
transmission power of CC2420), while the gateway in SHDP uses 17dBm as a
WiFi transceiver [IEEE Std 802.11-2009 2009]. Our feasibility analysis presented in
Section 2.1 shows that these transmission power settings allow each node to have
transmission ranges of ∼58 meters and the gateway to have a range of ∼191 meters.
As we show in Figure 5, we randomly deploy 100 nodes in a circular area with a radius
of 191 meters and locate the gateway at the center, which is depicted as a red dotted
circle. Considering that we apply the offset quadrature phase-shift keying (O-QPSK)
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:15
modulation of IEEE 802.15.4 [IEEE Std 802.15.4-2003 2003], we obtain the BER as
BER (γ)=8
15 ×1
16 ×
16
k−2
(−1)k16
kexp 20 1
k−1γ.(36)
We set ptx =70mW,4prx =78mW, pcs =30mW, and pidle =3.7mW following the
CC2420 specification [Texas Instruments 2006].
We assume that the channel-sensing range is the same as the transmission range,
and do not consider the capture effect. As a result, a hidden-node collision always results
in a packet reception error. Based on an extensive set of experiments, we configure
each node to have a FIFO buffer size of 10 packets and a maximum of 10 transmission
attempts for each packet delivery (i.e., ntx =10). Since we target downward traffic, the
gateway generates a packet toward a randomly chosen destination every 15 seconds.
Each simulation runs for 4 hours.
In Figure 5, the same mark represents the same hop distance from the gateway.
Assuming that each node in MHDP knows its neighbor nodes and their depth from
the gateway, it selects a neighbor with the minimum depth as its parent node. The
gateway has the topology information and an ideal routing path to each destination.
Each node is assumed to have routing information for each destination in advance
without incurring any control overhead.
5.1. Latency
Figure 6 plots the average packet delivery latency observed by varying the wakeup
interval. Here, the horizontally dotted red line indicates the interpacket interval (IPI)
of 15 seconds. As expected, infrequent wakeup naturally results in increased delivery
latency. SHDP achieves significantly lower latency than MHDP as the wakeup interval
increases. This is mainly due to the fact that in SHDP, the higher transmission power
of the gateway leads to direct one-hop delivery most of the time, whereas the same
packet delivery requires multiple-hop transmission in MHDP.
Furthermore, we notice that the performance gap between analysis and simulation is
larger with the wakeup interval in MHDP (especially when the latency becomes larger
than IPI). This is because our analysis has not incorporated contention and collision
effects on performance.
5.2. Packet Delivery Ratio
The packet delivery ratio is an important metric in many applications, which is shown
in Figure 7. Although the analytical results show that MHDP is on par with SHDP in
packet delivery performance, our simulation results differ a bit from those of analysis.
Deep investigation on the simulation traces reveals that packet loss in MHDP is higher
than that in SHDP when the wakeup interval is large, mainly due to the queueing loss.
The queueing delay in MHDP results in high packet latency as presented in Figure 6.
Therefore, buffer provisioning in MHDP is an important concern that SHDP avoids by
transmitting downward packets in one hop.
5.3. Power Consumption
In Figure 8, we show the average power consumption per node with respect to the
wakeup interval. The power consumption of MHDP first decreases and then increases
with the wakeup interval. This is caused by the increased transmission burden from
the wakeup interval. In contrast, SHDP consumes less power than MHDP, and its
power consumption continuously decreases with the wakeup interval.
4ptx is the total power consumption of a node in transmit mode.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:16 H.-S. Kim et al.
Fig. 6. Latency versus wakeup interval. Fig. 7. Packet delivery failure ratio versus wakeup
interval.
Fig. 8. Power consumption versus wakeup interval. Fig. 9. Fairness in power consumption versus
wakeup interval.
We break down the power consumption factors in detail using Figure 10. As ex-
pected, power consumption for channel sensing in MHDP and SHDP decreases with the
wakeup interval. Regarding other power consumption factors, the transmission power
consumption in MHDP increases with the wakeup interval due to prolonged repetitive
transmission. On the other hand, since SHDP finishes its transmission mostly via sin-
gle hop (from the gateway), the power consumption of the low-power nodes remains
very low.
Lastly, we present the fairness in power consumption using the Jain’s Fairness In-
dex [Kim et al. 2014]:
J(p)=Ntot
i=1pi2
Ntot Ntot
i=1p2
i
,(37)
where piis the power consumption of node i.InFigure9,SHDP shows an overall fair
performance compared to MHDP. This is mainly because in a multihop network, the
routing burden on nodes that are closer to the gateway is unavoidably higher than that
of the other nodes distant from the gateway [Kim et al. 2015b]. In SHDP, fair power
consumption can be achieved since the gateway covers the whole network directly.
6. TESTBED EXPERIMENTS
We implemented SHDP on top of RPL and BoX-MAC-2 in TinyOS [Winter et al. 2012;
Moss and Levis 2008]. RPL is an IP routing protocol for low-power and lossy networks
that is widely used [Ko et al. 2010; Ancillotti et al. 2013]. Unlike the ideal (and unrealis-
tic) MHDP, RPL requires control overhead in finding routes and suffers from limited or
outdated routing information. Figure 11 summarizes some of the major modifications
made to the software stack of TinyOS. Specifically, we removed the downlink route
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:17
Fig. 10. Detailed power consumption breakdown.
Fig. 11. SHDP architecture. Fig. 12. Left: MTM-CM3300MSP
for the high-power gateway. Right:
Kmote for the low-power nodes.
discovery procedures and downward routing table in the IP-support stack. This allows
downward packets to be sent directly to the destination within the software stack.
Furthermore, we swap the parent table in RPL with a neighbor table, since SHDP
uses neighbor information for local ACK exchange and neighbor forwarding. If a node
with an inconsistent link is added to the neighbor table, SHDP experiences a large
neighbor-forwarding overhead since local ACK exchange becomes difficult. For efficient
neighbor forwarding, each node adds a node to the neighbor table only when its observed
RSSI is higher than a predefined threshold pr,th. The upward route loss problem, which
is potentially caused by the RSSI-based neighbor filtering in a low-density network,
can be mitigated by activating the filtering process only when a node has a proper
parent node already. SHDP adopts a cross-layer design approach in the sense that the
link layer uses the routing layer information to confirm whether the received packet’s
destination is one of its neighbors.
Figure 13 shows the testbed topology map consisting of 20 nodes and one gateway,
marked with a star, in an office environment (c.f., Figure 12 presents the hardware
platforms used in this experiment). For a low-power node, we use the TelosB-clone
device [Moteiv Corporation 2006], which combines an MSP430 microcontroller [Texas
Instruments 2006] with a TI CC2420 transceiver [Texas Instruments 2006]. For the
high-powered gateway, we use MTM-CM3300MSP [MAXFOR Technology 2014], which
is similar to TelosB [Moteiv Corporation 2006] but includes a 10dB power ampli-
fier. In our experiments, the high-power gateway and common low-power nodes use
transmission power 10dBm and -15dBm, respectively, with an antenna gain of 5dB.
To maintain both multihop uplink routes and single-hop downlink connectivity, the
high-power gateway transmits a data packet with 10dBm and other packets (e.g.,
routing packets and ACKs) with -15dBm. Given this testbed configuration, the RPL
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:18 H.-S. Kim et al.
Fig. 13. Testbed topology map. Fig. 14. Testbed architecture.
Fig. 15. PRR performance of RPL, APNhp,APN
hpsl ,andSHDP for 19 hours (from 2AM to 9PM), where
APNhp and APNhpsl are variants of SHDP.
implementation connects all the nodes in a maximum of three hops. From empiri-
cal results, we set the threshold pr,th as -87dBm when deciding to include a node
in the neighbor table. Finally, each node has a FIFO buffer size of 10 packets.
Finally, as we illustrate in Figure 14, each WSN node is connected to a PC via USB
and sends log messages to the PC through the UART back-channel. We gather the log
messages from each PC through the ethernet to obtain various performance measure-
ments and real-time operation statuses. Furthermore, we remotely reprogram each
node through the UART and ethernet back-channels. The two connections are only
used for debugging and statistics gathering, and are not used for data communication
between nodes. With this testbed architecture, we obtain various performance metrics
by allowing each node to calculate its routing overhead, up/downward transmission
overhead, and duty-cycle performance to be included in its log messages. Furthermore,
we piggyback routing path information such as hop distance and end-to-end retrans-
missions in the application payload of each data packet and configure each relay node
to update the information.
6.1. Downward Traffic Scenario
As a first step in our testbed evaluation, we consider a downward traffic scenario. We
configure the gateway to generate packets toward each destination at an interval of
75 seconds (i.e., 3.75 seconds from the network’s perspective) and set the asynchronous
wakeup interval of each node to 0.5 second. To verify the effectiveness of each design
element in SHDP, we also evaluate the performance of two variants of SHDP, termed
as APNhp and APNhpsl .APN
hp includes the gateway’s high-power transmission but does
not allow each node to participate in neighbor forwarding nor sleep after overhearing
the gateway’s transmissions with s>0. On the other hand, APNhpsl includes the
gateway’s high-power transmission and sleep-after-overhearing mechanism but does
not use neighbor forwarding of SHDP.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:19
Fig. 16. Duty-cycle performance of RPL, APNhp,APN
hpsl ,andSHDP for 19 hours (from 2AM to 9PM), where
APNhp and APNhpsl are variants of SHDP.
In Figure 15(a), we present the per-hour average PRR throughout per day. During the
daytime, the PRR performance of RPL fluctuates due to the channel dynamics created
by the movement of people. We can observe that APNhp and APNhpsl are more vul-
nerable to channel dynamics, and their PRR performances are fluctuating and poorer
compared to RPL. This reveals that only using a high-transmission-power gateway is
not sufficient for achieving reliable downward packet delivery in APNs. On the other
hand, in SHDP, we always observe not only the most stable but also the highest PRR
performance among the competitive protocols mainly owing to the combination of high-
power single-hop transmissions at the gateway and local retransmissions at neighbor
nodes of the packet’s destination. This implies that the link instability due to channel
dynamics significantly impacts the performance of both multihop and high-powered
single-hop communications, while this is not an issue in SHDP.
Figure 15(b) plots the PRR performance of each node for the four protocols of our
interest. The results show that APNhp and APNhpsl significantly suffer from unfair
PRR performance among the nodes. By matching the results with the physical topology
depicted in Figure 13, we confirm that nodes experiencing very low PRR are far from the
gateway (e.g., nodes 18–20) or hidden behind obstacles (e.g., node 11). This reveals that,
in practice, high-powered single-hop transmission is difficult to guarantee reliability
for nodes placed at the boundary of the transmission range. Lastly, SHDP provides
better, more stable, and fairer PRR performance than the others by overcoming both
path loss and channel dynamics, which verifies that neighbor forwarding in SHDP is
beneficial to maximizing the strength of APNs and leads to reliable packet delivery.
We further focus on the duty-cycle performance in Figures 16(a) through 16(c), which
indirectly represents the nodes’ energy consumption. First, Figure 16(b) shows that
RPL’s duty-cycle performance is significantly unfair among nodes. The unbalanced
transmission overhead among nodes in RPL is unavoidable due to its multihop nature
since nodes near the gateway are asked to relay more packets compared to close-to-
leaf nodes. Furthermore, RPL inherits the load-balancing problem [Kim et al. 2015b],
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:20 H.-S. Kim et al.
Fig. 17. Average PRR performance versus wakeup interval.
which worsens the unfair energy consumption among nodes. From Figures 16(a) and
16(c), we can observe that the duty-cycle performance of RPL fluctuates in the time
domain, which becomes severe for the worst-performing node. This is because RPL
requires a high number of retransmissions and control overhead for achieving reliable
packet delivery and route maintenance with fluctuating links. Despite the effort, RPL
fails to overcome link instability as presented in Figure 15(a).
Interestingly, the results here reveal that APNhp consumes more energy than RPL
since each high-powered transmission triggers idle listening at all nodes, leading to a
waste of energy. On the other hand, APNhpsl provides the lowest duty-cycle performance
by allowing each node to sleep after overhearing the gateway’s packet transmission
with s>0 without idle listening. Lastly, we can see that the average per-hour duty-
cycle performance of SHDP is slightly lower than that of RPL, and for the worst case,
the differences become far more prominent. This is because SHDP provides low and
fair transmission overhead for nodes by allowing the gateway to take most of the
transmission burden. Quantitatively, SHDP’s average transmission overhead is only
one-third of RPL’s. For the worst-case node, this gap increases by 1
18 . Compared to
APNhpsl ,SHDP provides only slightly higher radio duty cycles due to the neighbor-
forwarding overhead. Given other performance metrics, we find this as a reasonable
cost to pay.
From the experimental results in this section, we can confirm that, while we can eas-
ily obtain the APN architecture in terms of hardware configurations, its performance
heavily depends on how the network protocol is designed. SHDP successfully takes
the advantages of an APN architecture and significantly improves the networking
performance over RPL.
6.2. Mixed Traffic Scenario
In most cases, even if downward traffic takes a major portion of network traffic, some
upward traffic coexists. For this, we vary the upward traffic generation interval at each
node from 100 seconds to 300 seconds and adjust the downward traffic generation inter-
val to vary between 100 seconds and 300 seconds. We test three cases of downward and
upward traffic generation intervals of [100 seconds, 300 seconds], [150 seconds, 150 sec-
onds], and [300 seconds, 100 seconds], respectively. We consider wakeup intervals of
0.5, 2, and 4 seconds.
In Figure 17, we plot the PRRs for the three offered traffic cases. While the perfor-
mance details are different in each case, we notice that the downward packet delivery
performance in SHDP is superior to that of RPL in all cases. For upward traffic, SHDP is
on par with RPL for short wakeup intervals and shows slightly decreased performance
with the wakeup interval. This is mainly because high-power downward broadcast
packets can be queued due to infrequent wakeups, and this congestion causes packet
loss at low-power nodes. SHDP provides nearly perfect downward PRR regardless
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:21
Fig. 18. Duty-cycle performance versus wakeup interval.
Fig. 19. Per-hour average PRR performance for downward and upward traffic.
of traffic patterns, but its upward PRR decreases as the downward traffic becomes
dominant. Each high-powered transmission in SHDP incurs wireless interference
throughout the whole network, which causes frequent packet collisions at low-power
nodes when the network generates heavy downward traffic.5
In Figure 18, we note an interesting observation for the duty cycle of each node in
both the average case and the worst case. For all the nine instances, the average duty
cycle of SHDP is lower than that of RPL. The duty cycle observed from the worst-case
node in SHDP is significantly lower compared to that of RPL. This implies that SHDP
has a significantly longer network lifetime than RPL since the battery lifetime of the
first dead node impacts the usability of the entire system [Ganesh and Amutha 2013].
Next we turn our attention to the per-hour performance of the test network. For
this, we take the traffic case of [100 seconds, 300 seconds] with the wakeup interval of
0.5 second and plot the PRR performance for each hour in Figure 19. In Figure 19(a),
SHDP shows steady downward performance over time. This result implies that the
high-power transmission and local retransmission features of SHDP help to achieve
steady packet delivery performance despite the varying channel condition throughout
the day. For upward PRR performance shown in Figure 19(b), RPL shows slightly
higher PRR than SHDP. This is because RPL is optimized for data collection, and
downward routes are simply set as the reverse of upward routes.
Figure 20 shows the overhead for each packet transmission. The overhead in SHDP
is significantly lower than that in RPL for both downward and upward transmissions.
The main reason for the reduced upward transmission overhead is that RPL forces
each node to transmit destination advertise object (DAO) packets to the gateway as
a way of maintaining downward routes, while SHDP does not generate such control
packets.
5We consider improving upward packet delivery performance in APNs as part of our future work.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:22 H.-S. Kim et al.
Fig. 20. Per-hour average transmission overhead. Fig. 21. Per-hour duty-cycle performance.
Lastly, we observe the average per-hour duty cycle of each node and that of the
worst-case node over time in Figure 21. When combining the results of Figure 16 and
Figure 21, we observe that each node in SHDP has a lower duty cycle compared to
RPL under various traffic scenarios. Overall, our experimental results show that in a
real-world test environment, SHDP outperforms the multihop routing protocol of RPL
when reliable and energy-efficient delivery of downward packets is considered.
7. RELATED WORK
Ko et al. [2011] experimentally evaluated the performance of RPL using the TinyRPL
implementation in TinyOS and have shown that its performance is similar to that
of the collection tree protocol (CTP), the de facto data collection protocol in TinyOS,
while having the benefits of an IPv6-based architecture. Kim et al. [2015a] evaluated
the performance of TCP over RPL on a multihop WSN testbed. These works revealed
that RPL’s downward packet delivery performance is less efficient than the upward
performance.
A number of works revealed issues related to RPL and tried to alleviate these prob-
lems. Ancillotti et al. [2014] proposed a cross-layering design for RPL, which provides
enhanced link estimation and efficient management of neighbor tables. They used AMI
as a case study and employed the Cooja emulator to evaluate their proposal. The work
in Kim et al. [2015b] investigated the load-balancing problem of RPL and revealed
that the performance of RPL severely degrades in a heavy traffic environment due
to queue loss. To alleviate the issue, QU-RPL was proposed to allow each node to
smartly use queue utilization information for its routing parent selection. Even though
these works provided better upward routes, they did not consider the downward traffic
delivery performance of RPL.
In the perspective of downward route management, Ko et al. [2015] showed that
RPL has a serious connectivity problem when two modes of operation (MOPs) are
mixed within a single network. To address this issue, the authors proposed DualMOP-
RPL, which supports nodes with different MOPs to communicate gracefully in a single
network while preserving the high bidirectional data delivery performance. However,
this work focused on addressing interoperability problems between two MOPs rather
than finding better downward routes.
There are several deployment examples of multihop WSNs. Ko et al. [2010] deployed
a CTP-based WSN at Johns Hopkins Hospital and showed that multihop WSNs can
provide reasonable performance for monitoring the status of patients. Cisco [2015]
designed and deployed a field area network (FAN) for smart grids (CG-Mesh), which
uses 6LoWPAN [Montenegro et al. 2007], RPL, and IPv6 on top of IEEE 802.15.4 to
provide end-to-end two-way communication to each smart metering endpoint. Buevich
et al. [2014] presented the architecture, design, and experiences from a wirelessly
managed microgrid deployment in a rural area. They constructed a multihop network
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:23
by using their own flooding mechanism mPCF. However, none of these works consider
downward-centric traffic patterns. Most importantly, all of the aforementioned works
assume that all nodes use the same transmission power, which is different from our
work.
Several works considered asymmetric (heterogeneous) transmission power usage
among nodes. Li and Hou [2004] proposed a topology control algorithm when each node
exploits different transmission powers. Lin et al. [2006] designed each node to con-
trol its transmission power considering wireless network connectivity. Although these
works considered link asymmetry among nodes, they still constructed homogeneous
networks in the routing perspective. That is, they constructed multihop networks in
which bidirectional routes have the same hop distance, instead of creating multihop up-
ward and single-hop downward connectivity. Li and Fang [2012] analyzed throughput
performance in a heterogeneous wireless network such as a city-wide wireless network
and battlefield network, where some powerful nodes are deployed along with normal
nodes. This work provided a theoretical approach with high algorithm complexity,
thereby leaving some distance from being implemented on real devices.
8. CONCLUSION
In this article, we present the SHDP protocol, which exploits asymmetric transmission
power capabilities of heterogeneous nodes. Specifically, in our network environment,
the gateway node uses a strong transmission power to cover the entire network in a sin-
gle hop while common sensor nodes maintain low-transmission-power profiles. SHDP
comprises subcomponents such as the gateway’s high-power transmission, packet over-
hearing by neighbor nodes of a destination, local ACK exchange between a destination
and its neighbors, neighbor forwarding in the case of direct transmission failure from
the gateway, and contention mitigating between neighbors.
Through mathematical analysis, extensive simulation, and empirical testbed exper-
iments, we confirm that SHDP outperforms other competitive multihop routing pro-
tocols with respect to PRR performance, radio duty cycle, and transmission overhead.
We see that our work on asymmetric transmission power-based nodes will bring new
perspectives in designing sensing systems for various application domains: with low-
cost radio modules and reduction in hardware development costs, the assumption of
ubiquitously using homogeneous transmission power-based networks may not be the
most efficient design choice anymore.
REFERENCES
E. Ancillotti, R. Bruno, and M. Conti. 2013. The role of the RPL routing protocol for smart grid
communications. IEEE Communications Magazine 51, 1 (Jan. 2013), 75–83. DOI:http://dx.doi.org/
10.1109/MCOM.2013.6400442
E. Ancillotti, R. Bruno, and M. Conti. 2014. Reliable data delivery with the IETF routing protocol for low-
power and lossy networks. IEEE Transactions on Industrial Informatics 10, 3 (Aug. 2014), 1864–1877.
DOI:http://dx.doi.org/10.1109/TII.2014.2332117
Aruba Networks. 2012. Technology advances in retail: Improving margins using wireless networks. Available
at: http://www.arubanetworks.com/pdf/technology/whitepapers/wp_Retail_advances.pdf. (2012).
M. Buevich, D. Schnitzer, T. Escalada, A. Jacquiau-Chamski, and A. Rowe. 2014. Fine-grained remote mon-
itoring, control and pre-paid electrical service in rural microgrids. In Proceedings of the 13th Interna-
tional Symposium on Information Processing in Sensor Networks (IPSN’14). 1–11. DOI:http://dx.doi.org/
10.1109/IPSN.2014.6846736
M. Ceriotti, M. Corra, L. D’Orazio, R. Doriguzzi, D. Facchin, S. T. Guna, G. P. Jesi, R. Lo Cigno, L. Mottola,
A. L. Murphy, M. Pescalli, G. P. Picco, D. Pregnolato, and C. Torghele. 2011. Is there light at the ends of
the tunnel? Wireless sensor networks for adaptive lighting in road tunnels. In 2011 10th International
Conference on Information Processing in Sensor Networks (IPSN’11). 187–198.
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
34:24 H.-S. Kim et al.
Y. Chen and A. Terzis. 2011. On the implications of the log-normal path loss model: An efficient method
to deploy and move sensor motes. In Proceedings of ACM Conference on Embedded Networked Sensor
Systems (SenSys’11). 26–39. DOI:http://dx.doi.org/10.1145/2070942.2070946
Y.-J. Choi, S. Park, and S. Bahk. 2006. Multichannel random access in OFDMA wireless networks.
IEEE Journal on Selected Areas in Communications 24, 3 (Mar. 2006), 603–613. DOI:http://dx.doi.org/
10.1109/JSAC.2005.862422
Cisco. 2015. Connected grid networks for smart grid - field area network. http://www.cisco.com/web/
strategy/energy/field_area_network.html. (2015).
T. Clausen, U. Herberg, and M. Philipp. 2011. A critical evaluation of the IPv6 routing protocol for low power
and lossy networks (RPL). In 2011 IEEE 7th International Conference on Wireless and Mobile Com-
puting, Networking and Communications (WiMob’11). 365–372. DOI:http://dx.doi.org/10.1109/WiMOB.
2011.6085374
Federal Communications Commission. 1993. FCC power regulation. Available at: http://transition.fcc.gov/
Bureaus/Engineering Technology/Documents/bulletins/oet63/oet63rev.pdf. (1993).
J. Fink, A. Ribeiro, and V. Kumar. 2012. Robust control for mobility and wireless communication in cyber
physical systems with application to robot teams. Proceedings of the IEEE 100, 1 (Jan. 2012), 164–178.
DOI:http://dx.doi.org/10.1109/JPROC.2011.2161427
S. Ganesh and R. Amutha. 2013. Efficient and secure routing protocol for wireless sensor networks through
SNR based dynamic clustering mechanisms. Journal of Communications and Networks 15, 4 (Aug.
2013), 422–429. DOI:http://dx.doi.org/10.1109/JCN.2013.000073
O. Gnawali, R. Fonseca, K. Jamieson, and P. Levis. 2008. CTP: Robust and efficient collection through control
and data plane integration. Stanford Information Networks Group Technical Report SING-08-02. (2008).
IEEE Std 802.11-2009 2009. IEEE Standard for Information Technology– Local and Metropolitan Area
Networks– Specific Requirements– Part 11: Wireless LAN Medium Access Control (MAC) and Phys-
ical Layer (PHY) Specifications Amendment 5: Enhancements for Higher Throughput. (Oct 2009).
DOI:http://dx.doi.org/10.1109/IEEESTD.2009.5307322
IEEE Std 802.15.4-2003 2003. Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer
(PHY) Specifications for Low-Rate Wireless Personal Area Networks (LR-WPANs). Available at
http://www.ieee802.org/15/pub/TG4.html. (May 2003).
H.-S. Kim. 2013. SHDP simulator. Available at: http://netlab.snu.ac.kr/∼hskim. (2013).
H.-S. Kim, Y.-J. Choi, and S. Bahk. 2014. Elimination of multi-hop transmission form downlink in low power
and lossy networks. In Proceedings of IEEE Conference on Communications (ICC’14).
H.-S. Kim, H. Im, M.-S. Lee, J. Paek, and S. Bahk. 2015a. A measurement study of TCP over RPL in low-power
and lossy networks. Journal of Communications and Networks 17, 6 (Dec. 2015), 647–655.
H.-S. Kim, J. Paek, and S. Bahk. 2015b. QU-RPL: Queue utilization based RPL for load balancing in large
scale industrial applications. In IEEE International Conference on Sensing, Communication, and Net-
working (SECON’15).
J.-G. Ko, S. Dawson-Haggerty, O. Gnawali, D. Culler, and A. Terzis. 2011. Evaluating the performance of RPL
and 6LoWPAN in TinyOS. In Workshop on Extending the Internet to Low Power and Lossy Networks
(IP+SN’11).
J. Ko, J. Jeong, J. Park, J. A. Jun, O. Gnawali, and J. Paek. 2015. DualMOP-RPL: Supporting multiple modes
of downward routing in a single RPL network. ACM Transactions on Sensor Networks 11, 2, Article 39
(2015), 20 pages. DOI:http://dx.doi.org/10.1145/2700261
J.-G. Ko, J.-H. Lim, Y. Chen, R. Musaloiu-E., A. Terzis, G. Masson, T. Gao, W. Destler, L. Selavo, and
R. Dutton. 2010. MEDiSN: Medical emergency detection in sensor networks. ACM Transactions on
Embedded Computing Systems (TECS), Special Issue on Wireless Health Systems (2010).
J.-G. Ko, A. Terzis, S. Dawson-Haggerty, D. E. Culler, J. W. Hui, and P. Levis. 2011. Connecting low-
power and lossy networks to the Internet. IEEE Communications Magazine 49, 4 (Apr. 2011), 96–101.
DOI:http://dx.doi.org/10.1109/MCOM.2011.5741163
O. Landsiedel, E. Ghadimi, S. Duquennoy, and M. Johansson. 2012. Low power, low delay: Opportunistic
routing meets duty cycling. In Proceedings of International Conference on Information Processing in
Sensor Networks (IPSN’12). 185–196. DOI:http://dx.doi.org/10.1145/2185677.2185731
N. Li and J. C. Hou. 2004. Topology control in heterogeneous wireless networks: problems and solutions.
In 23rd Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM’04),
Vol. 1. 243. DOI:http://dx.doi.org/10.1109/INFCOM.2004.1354497
P. Li and Y. Fang. 2012. On the throughput capacity of heterogeneous wireless networks. IEEE Transactions
on Mobile Computing 11, 12 (Dec. 2012), 2073–2086. DOI:http://dx.doi.org/10.1109/TMC.2011.239
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
Reliable and Energy-Efficient Downward Packet Delivery in APN 34:25
S. Lin, J. Zhang, G. Zhou, L. Gu, J. A. Stankovic, and T. He. 2006. ATPC: Adaptive transmission power
control for wireless sensor networks. In Proceedings of the 4th International Conference on Em-
bedded Networked Sensor Systems (SenSys’06). ACM, New York, NY, 223–236. DOI:http://dx.doi.org/
10.1145/1182807.1182830
K. Lorincz, D. J. Malan, T. R. F. Fulford-Jones, A. Nawoj, A. Clavel, V. Shnayder, G. Mainland, M. Welsh,
and S. Moulton. 2004. Sensor networks for emergency response: challenges and opportunities. IEEE
Pervasive Computing 3, 4 (Oct. 2004), 16–23. DOI:http://dx.doi.org/10.1109/MPRV.2004.18
M. Marot, B. Kusy, G. Simon, and A. Ledeczi. 2004. The flooding time synchronization protocol. In Proceedings
of the 2nd Conference on Embedded Networked Sensor Systems (SenSys’04). 39–49.
MAXFOR Technology. 2014. WSN communication module, MTM-CM3300MSP. Available at: http://www.
maxfor.co.kr/datasheet/MAXFOR_Digital_Brochure.pdf. (2014).
G. Montenegro, N. Kushalnagar, J. Hui, and D. Culler. 2007. Transmission of IPv6 packets over IEEE 802.15.4
networks. RFC 4944 (Sept. 2007).
D. Moss and P. Levis. 2008. BoX-MACs: Exploiting Physical and Link Layer Boundaries in Low-Power
Networking. Technical Report SING-08-00. Stanford Information Networks Group.
Moteiv Corporation. 2006. Tmote Sky Datasheet. http://www.moteiv.com/products/docs/tmote-sky-datasheet.
pdf. (2006).
W. Pak, K.-T. Cho, J. Lee, and S. Bahk. 2008. W-MAC: Supporting ultra low duty cycle in wireless sen-
sor networks. In Proceedings of IEEE Global Communications Conference (GLOBECOM’08). 1–5.
DOI:http://dx.doi.org/10.1109/GLOCOM.2008.ECP.79
D. Puccinelli, S. Giordano, M. Zuniga, and P. J. Marr´
on. 2012. Broadcast-free collection protocol.
In Proceedings of ACM Conference on Embedded Networked Sensor Systems (SenSys’12). 29–42.
DOI:http://dx.doi.org/10.1145/2426656.2426660
K. Srinivasan, P. Dutta, A. Tavakoli, and P. Levis. 2010. An empirical study of low-power wireless. ACM
Transactions on Sensor Networking 6, 2, Article 16 (March 2010), 49 pages. DOI:http://dx.doi.org/
10.1145/1689239.1689246
Texas Instruments. 2006. 2.4 GHz IEEE 802.15.4/ZigBee-ready RF transceiver. Available at: http://focus.
ti.com/lit/ds/symlink/cc2420.pdf. (2006).
Texas Instruments. 2006. MSP430 Ultra-low-power microcontrollers. (2006).
F. Wang and J. Liu. 2009. Duty-cycle-aware broadcast in wireless sensor networks. In IEEE
Conference on Computer Communications (INFOCOM’09). 468–476. DOI:http://dx.doi.org/10.1109/
INFCOM.2009.5061952
G. Werner-Allen, K. Lorincz, J. Johnson, J. Lees, and M. Welsh. 2006. Fidelity and yield in a volcano
monitoring sensor network. In Proceedings of the 7th Symposium on Operating Systems Design and
Implementation (OSDI’06). USENIX Association, Berkeley, CA, 381–396. http://dl.acm.org/citation.
cfm?id=1298455.1298491
T. Winter, P. Thubert, A. Brandt, J. Hui, R. Kelsey, P. Levis, K. Pister, R. Struik, J. P. Vasseur, and R.
Alexander. 2012. RPL: IPv6 routing protocol for low-power and lossy networks. RFC 6550 (2012).
K. S. Yildirim and A. Kantarci. 2014. Time synchronization based on slow-flooding in wireless sen-
sor networks. IEEE Transactions on Parallel and Distributed Systems 25, 1 (Jan. 2014), 244–253.
DOI:http://dx.doi.org/10.1109/TPDS.2013.40
Received October 2014; revised October 2015; accepted August 2016
ACM Transactions on Sensor Networks, Vol. 12, No. 4, Article 34, Publication date: September 2016.
