Comparison between Channel Hopping and Channel Adaptation for

Industrial Wireless Sensor Networks

Ruan D. Gomes

1,2,3

, Marcelo S. Alencar

, Diego V. Queiroz

Iguatemi E. Fonseca

and Cesar Benavente-Peces

Post-Graduate Program in Electrical Engineering, Federal University of Campina Grande, Campina Grande, Brazil

Informatics Coordination, Federal Institute of Paraiba, Guarabira, Brazil

Signal Theory and Communications Department, Universidad Politecnica de Madrid, Madrid, Spain

Informatics Center, Federal University of Paraiba, Joao Pessoa, Brazil

Keywords:

Industrial Wireless Sensor Networks, Channel Diversity, Dynamic Channel Allocation.

Abstract:

One of the differences between the new standard IEEE 802.15.4e, in comparison to the previous IEEE 802.15.4

standard, is the use of multiple channels. The Time-Slotted Channel Hopping (TSCH) mode employs channel

hopping, and the Deterministic and Synchronous Multi-channel Extension (DSME) mode employs channel

hopping or channel adaptation, during the contention free periods. When using the channel adaptation as the

channel diversity technique, a pair of nodes communicate using the same channel while the channel quality

is good enough in terms of signal-to-noise ratio. Thus, it is necessary to evaluate the quality of the links, in

order to proper use this mechanism. In this paper, three different approaches, based on the DSME protocol,

were implemented and evaluated through a simulation study. The ﬁrst one (CH-DSME) is based on a simple

channel hopping mechanism, the second one (CA-DSME) employs channel adaptation, and the third one is

a novel hybrid approach (H-DSME), that uses both channel hopping and channel adaptation. The H-DSME

outperformed the other two approaches for the scenario in consideration, which shows that the use of channel

adaptation is better than channel hopping for the transmission of unicast packets, when the quality of the

links are monitored continuously. However, for packets transmitted in broadcast by the coordinator, the use of

channel hopping is a good alternative to deal with the spatial variation in the quality of the channels.

1 INTRODUCTION

The use of Wireless Sensor Networks (WSN), to im-

plement monitoring and control systems in industrial

environments, has some advantages when compared

with the use of wired networks, such as low cost and

high ﬂexibility. However, it is necessary to deal with

typical problems of wireless systems, such as electro-

magnetic interference (Lima-Filho et al., 2012), and

industrial environment high attenuation, due to the

presence of many objects and obstructions (Tanghe

et al., 2008). Many industrial environments also

present characteristics that make the wireless channel

non-stationary, for long time periods, which can cause

abrupt changes in the characteristics of the channel

over time (Agrawal et al., 2014b).

To overcome these limitations, mechanisms that

allow the network to self-adapt to the variations that

occur in the link quality over time need to be im-

plemented, such as adaptive routing (Gnawali et al.,

2009) or dynamic channel allocation (Gomes et al.,

2014). Other important aspect is the energy consump-

tion and management, which is a key issue in indus-

trial WSN, given network motes are usually powered

by batteries. Some appropriate strategies, as energy

aware geographic routing in lossy WSN, can be de-

veloped (Anastasi et al., 2009).

Some standards have been proposed in the last

years with a focus on industrial applications, such as

the WirelessHART and the ISA100.11a. Both Wire-

lessHART and ISA100.11a are based on the physi-

cal layer of IEEE 802.15.4, but deﬁnes its own MAC

layer. Instead of using CSMA/CA, as deﬁned by the

IEEE 802.15.4 standard, they use a MAC layer with

Time Division Multiple Access (TDMA). By using

TDMA, collisions are avoided and the power con-

sumption can be optimized. They also use frequency

hopping and blacklisting, to mitigate the problems re-

lated to interference and fading. However, without

an adequate management of the blacklist, the com-

D. Gomes R., S. Alencar M., V. Queiroz D., E. Fonseca I. and Benavente-Peces C.

Comparison between Channel Hopping and Channel Adaptation for Industrial Wireless Sensor Networks.

DOI: 10.5220/0006206800870098

In Proceedings of the 6th International Conference on Sensor Networks (SENSORNETS 2017), pages 87-98

ISBN: 421065/17

munication performance may be lower for these stan-

dards (Petersen and Carlsen, 2009).

More recently, the IEEE 802.15.4e standard was

released. The goal of this standard is to propose

solutions for applications that require high reliabil-

ity, such as industrial applications (Guglielmo et al.,

2016). Five modes of operation are deﬁned, that

is: Time-Slotted Channel Hopping (TSCH), De-

terministic and Synchronous Multi-Channel Exten-

sion (DSME), Low Latency Deterministic Network

(LLDN), Asynchronous Multi-Channel Adaptation

(AMCA), and Radio Frequency Identiﬁcation Blink

(BLINK). However, only the modes TSCH, DSME,

and LLDN have been explored in the literature until

recently. In general, the modes of the IEEE 802.15.4e

are based on TDMA or frequency hopping to reduce

collisions and mitigate the effects of interference and

fading, and to satisfy the requirements of industrial

applications in terms of reliability and determinism.

One of the main differences between the new

standard IEEE 802.15.4e in comparison to the pre-

vious IEEE 802.15.4 standard is the use of multiple

channels. The TSCH mode employs channel hop-

ping. When using this mechanism, the nodes usu-

ally switch to a new channel before each transmission,

which makes the network more robust against prob-

lems that affect only a subset of the channels. How-

ever, if a proper management of the blacklist is not

made, the network performance can be signiﬁcantly

degraded (Grsu et al., 2016). In the DSME mode,

channel hopping can also be used in the contention

free periods. It is also possible to use a channel adap-

tation mechanism instead of channel hopping. When

using channel adaptation, a pair of nodes communi-

cate using only one channel during a large time pe-

riod. A channel switch only occurs when the channel

in use starts to present low quality. Thus, a procedure

is necessary to evaluate the quality of the links contin-

uously, in order to use the channel adaptation mecha-

nism properly. The implementation of this procedure

is not deﬁned by the standard.

In this paper, the performance of three different

approaches for channel diversity, based on the DSME

mode of the IEEE 802.15.4e standard, were evalu-

ated. The ﬁrst one is based on a simple channel

hopping mechanism, the second one employs chan-

nel adaptation, and the third one is a novel approach

that uses both channel hopping and channel adap-

tation. The simulations were performed using the

Castalia simulator, which is an event-driven simula-

tor for WSN. A realistic channel model was used,

which includes the effects of fading, shadowing, and

the non-stationary characteristics of the channel in in-

dustrial environments. When using this model, which

was ﬁrst described in (Gomes et al., 2015b), it is

possible to observe the performance of the protocols,

considering the non-stationary behavior of the wire-

less channel in industrial environments. This simula-

tion model was also used in (Gomes et al., 2016) to

evaluate link quality estimators for industrial WSN.

2 IEEE 802.15.4e STANDARD

In networks that use the IEEE 802.15.4 standard, it is

difﬁcult to establish strict latency boundaries, due to

the CSMA/CA protocol used in the MAC layer, since

the access to the communication medium occurs in

a distributed and random way. Hidden and exposed

terminal problems can also affect the performance of

the network, making it even more unpredictable. The

MAC protocols deﬁned by the IEEE 802.15.4 use a

single channel for communication, which is a single

point of failure, and increases the number of collisions

in the network. Due to these limitations, the new stan-

dard IEEE 802.15.4e was proposed, for applications

with more stringent requirements of reliability and de-

terminism.

Five modes of operation are deﬁned by the

standard, but only modes TSCH, DSME, and

LLDN have been more extensively studied until re-

cently (Guglielmo et al., 2016). Modes TSCH, DSME

and LLDN use TDMA as the channel access method,

which allows to reduce the number of collisions, and

increases the determinism of the network, since each

node has speciﬁc time slots allocated to it, and the ac-

cess can be done without contention. Besides, it is

possible to reduce the energy consumption, since the

nodes can sleep during time slots in which they are

not transmitting or receiving packets. Modes TSCH,

and DSME also employ multichannel communication

and three types of topology (star, mesh or tree), while

the LLDN deﬁne only star networks that use a sin-

gle channel to communicate. However, some works

have proposed the use of a tree topology and multiple

channels in LLDN networks (Patti et al., 2014; Patti

and Bello, 2016).

In the next sub-section more details about the

DSME mode are provided, since the protocols pro-

posed and evaluated in this paper are based on this

mode.

2.1 The Deterministic and Synchronous

Multi-channel Extension Mode

The DSME mode is the most complex and ﬂexible

mode of the IEEE 802.15.4e standard (Guglielmo

et al., 2016). It extends the beacon-enabled mode

SENSORNETS 2017 - 6th International Conference on Sensor Networks

Figure 1: Example of a superframe structure for a DSME

network.

of the IEEE 802.15.4, which is based on a super-

frame structure, that is managed by the network coor-

dinator. The coordinator sends beacon frames to de-

limit two consecutive superframes. The main differ-

ences between the beacon-enabled mode of the IEEE

802.15.4 and the DMSE mode is that the DSME mode

allows the use of a higher number of Guarantee Time

Slots (GTS), and allows communication using multi-

ple channels during the contention free period. Thus,

multiple nodes can transmit simultaneously during

the same GTS in different channels, increasing the

overall WSN throughput.

Figure 1 shows an example of a superframe struc-

ture deﬁned according to the DSME mode of the

IEEE 802.15.4e standard (802, 2012). Each super-

frame is composed by a Contention Access Period

(CAP), and a Contention Free Period (CFP). Different

from the beacon-enabled mode of the IEEE 802.15.4,

there is no inactive period. In the CAP, the nodes can

access the communication medium using CSMA/CA

or ALOHA (Guglielmo et al., 2016). The beacon

frames, referred in the standard as Enhanced Bea-

cons (EB), delimit the superframes. The EB frames

are transmitted using the same channel, deﬁned in the

starting process of the network (802, 2012), which is

also used in the CAP period.

Multiple superframes can be grouped inside a

multi-superframe, and multiple multi-superframes

can be grouped inside the same Beacon Interval (BI).

This time structure is conﬁgured using some param-

eters, that is: macSuperframeOrder (SO), macMulti-

superframeOrder (MO), and macBeaconOrder (BO),

in which 0 ≤ SO ≤ MO ≤ BO ≤ 14. The parame-

ter SO deﬁnes the size of the superframes, MO de-

ﬁnes the size of the multi-superframes, and BO de-

ﬁnes the BI. In the example shown in Figure 1 the

parameters MO = SO + 1, and BO = MO + 1. Each

superframe has 16 time slots. Thus, the value of SO

deﬁnes the duration of each slot. Optionally, the num-

ber of CAP can be reduced in a multi-superframe,

through a mechanism called capReduction. When the

capReduction is enabled, only the ﬁrst superframe in

Figure 2: Channel diversity mechanisms deﬁned for DSME

networks.

the multi-superframe has the CAP.

In the DSME networks there are three types of

nodes: PAN coordinator, coordinator, and end node.

The PAN coordinator sends a EB every BI. The co-

ordinator is the sink node for some of the network

nodes. A coordinator sends a beacon at least once per

multi-superframe, in the beacon slot, in order to regis-

ter its presence in the network (Alderisi et al., 2015).

In Figure 1, Node 1 is the PAN coordinator, Node

2 is a coordinator, and the remaining are end nodes.

The coordinators can forward packets from end nodes

that do not reach the PAN coordinator directly. In

the same network multiple coordinators nodes are al-

lowed.

The DSME mode deﬁnes two types of channel di-

versity, that is: channel hopping, and channel adap-

tation. Figure 2 shows two examples of scheduling

for the CFP period, using channel hopping in (a), and

channel adaptation in (b). When using the channel

hopping, the nodes receive packets in different chan-

nels depending on the channel offset of the node, the

slot ID, the superframe ID, and the sequence num-

ber of the EB sent by the coordinator. For example,

in the example illustrated in Figure 2(a), Node 1 re-

ceives a packet using Channel 0 in the ﬁrst time slot,

using Channel 1 in the second time slot, and so on.

The nodes that receive packets inside the same super-

frame need to have different channel offsets, in order

to avoid collisions.

When using channel adaptation, a ﬁxed channel is

allocated for a given time slot inside the superframe,

and a given pair of nodes. For example, in the ex-

ample illustrated in Figure 2(b), Channel 0 was allo-

cated for the communication from node 2 to node 1,

in all slots. The channel is only changed if the allo-

cated channel starts to present bad quality. The stan-

dard does not deﬁne how to select the communication

channels when using channel adaptation (Guglielmo

et al., 2016). In this paper, an algorithm to estimate

the link quality is proposed, and used to implement

the channel adaptation mechanism.

Another interesting characteristic of the DMSE

is the possibility of using group acknowledgment

(GACK), in which two time slots of the multi-

Comparison between Channel Hopping and Channel Adaptation for Industrial Wireless Sensor Networks

superframe are allocated for the GACK frames (G1

and G2). The coordinator uses G1 to acknowledge

all packets received until the G1 time slot. The G2 is

used to acknowledge all packets received after the G1

time slot and before the G2 time slot. If the GACK

is not used, all packets transmitted to the coordinator

are acknowledged individually (802, 2012). With this

mechanism, a node can retransmit a lost packet inside

the same multi-superframe, if one slot before the G1

and other slot between G1 and G2 are allocated to the

node.

2.2 Related Research

Some authors have proposed mechanisms to im-

prove the performance of IEEE 802.15.4e networks,

through the use of dynamic channel allocation or dy-

namic conﬁguration of the blacklist for TSCH net-

works. In (Grsu et al., 2016) an experiment was per-

formed to analyze the performance of a TSCH net-

work inside an aircraft cabin, with external interfer-

ence caused by Wi-Fi networks. In the experiments

described in (Grsu et al., 2016) the Packet Error Rate

(PER) was 35%, when using the 16 available chan-

nels, due to interference problems. In general, when

fewer channels were used, the performance was bet-

ter, as the interference is lower. For example, when

using only one channel, the less affected by the inter-

ference sources, the PER was 5%. However, a mecha-

nism is needed to estimate the quality of the channels

and to dynamically conﬁgure the blacklist.

In (Du and Roussos, 2011; Du and Roussos, 2013)

the use of adaptive frequency hopping for TSCH net-

works was proposed, in order to avoid using channels

affected by interference sources. In this approach,

two time slots in each cycle are used to perform read-

ings of RSSI values, in order to identify interference

sources. Based on these measurements, the black-

list is updated to avoid the channels with a high level

of interference. In (Du and Roussos, 2013) experi-

ments were conducted considering different sizes for

the blacklist. It was observed that the higher the size

of the blacklist, the better the communication perfor-

mance. This result corroborate the results presented

in (Grsu et al., 2016). However, this type of behav-

ior only occurs if an adequate monitoring of the qual-

ity of the channels is performed, in order to properly

conﬁgure the blacklist in real time. One limitation of

the approach presented in (Du and Roussos, 2011)(Du

and Roussos, 2013) is that only interference problems

are considered. Other aspects that can affect the qual-

ity of the links are not considered, such as shadowing

and fading. Besides, the channel quality monitoring

is performed by all nodes and using time slots that

could be used for communication, which incurs in a

high overhead, and in an increase of latency.

Some authors have proposed the use of techniques

for channel diversity and multi-channel communica-

tion based on the IEEE 802.15.4e standard for LLDN

networks, which use originally only one channel.

In (Patti et al., 2014) a multi-level and multichannel

protocol based on the LLDN mode, called the MC-

LLDN, was proposed. The goal is to increase the scal-

ability of the network through the use of a multi-level

topology, data aggregation, and multi-channel com-

munication. The drawback is that the channels are al-

located to the sub-networks in a static way. Thus, it is

not capable of dealing with the variations that occur in

the channel quality over time. The protocol described

in (Patti and Bello, 2016) is an evolution of the MC-

LLDN, called PriMuLa, which incorporates adaptive

channel selection. One limitation of the proposed pro-

tocol, which is due to the characteristics of the LLDN,

is that a same channel is allocated to all nodes in the

sub-network. However, spatial variations in the chan-

nel quality can occur, as well as asymmetry problems.

In the approaches developed for the present paper, the

channel quality is assessed in a per-link basis, as well

as the channel allocation.

The experiments described in (Jeong and Lee,

2012) and (Lee and Jeong, 2012) evaluated the per-

formance of the DSME mode in comparison to the

beacon-enabled mode of the IEEE 802.15.4. The

experiments veriﬁed that, in some scenarios, the

throughput of the IEEE 802.15.4e DSME network can

be 12 times higher than the IEEE 802.15.4 beacon-

enabled network, and with a lower energy consump-

tion, due to the use of a TDMA-based medium ac-

cess. In the experiments frequency hopping was used,

and no dynamic management of the blacklist was em-

ployed. In (Lee and Jeong, 2012) the inﬂuence of in-

terference caused by Wi-Fi networks was evaluated,

but other problems that can affect the channel qual-

ity in industrial environments, such as shadowing and

fading, were not considered.

In (Capone et al., 2014) simulation studies to

verify the performance of DSME networks are de-

scribed, and some enhancements to optimize the en-

ergy consumption are proposed. However, the paper

focuses mainly on energy consumption, and did not

consider in the experiments the problems that can af-

fect the channel quality, such as interference and fad-

ing. Besides, although in the simulations described

in (Capone et al., 2014) the channel adaptation mech-

anism was considered, the details about the imple-

mentation of this mechanism are not provided.

In (Alderisi et al., 2015) a comparison between

DSME and TSCH in process automation scenarios is

described. Simulations were performed to verify the

SENSORNETS 2017 - 6th International Conference on Sensor Networks

delay, reliability, and scalability of each mode. The

TSCH presented better results for small networks,

with up to 30 nodes. For larger networks, with more

than 30 nodes, the DSME presented better results.

The simulations described in (Alderisi et al., 2015)

used realistic parameters for the log-normal shadow-

ing, but the effect of fading and the non-stationary

characteristics of the wireless channel were not con-

sidered. In addition, only the channel hopping mech-

anism of the DSME were analyzed. The simulation

model used for the present paper considers more as-

pects that can affect the channel quality, that is: shad-

owing, fading, asymmetry, and the non-stationary

characteristics of the channel in long time periods.

Besides, a comparison between channel hopping and

channel adaptation is performed.

In (Juc et al., 2016) a comparison between TSCH

and DSME is described, in terms of energy consump-

tion and performance. In the scenarios under consid-

eration, the energy consumption of DSME was slight

better than TSCH, as well as the performance. For ap-

plications that send less data, the TSCH under-utilize

the bandwidth, due to the ﬁxed size of the time slots.

In the experiments described in (Juc et al., 2016) only

channel hopping was considered, and without group

ACK. In the present paper, channel adaptation are also

considered, as well as group ACK.

3 PROPOSED DSME-BASED

PROTOCOLS

In this paper, three approaches for the DSME mode

are implemented and evaluated, called CH-DSME,

CA-DSME, and H-DSME. The CH-DSME is based

on the channel hopping mechanism, and without

blacklist. Most papers, described in Section 2.2, eval-

uated the DSME with this type of channel diversity.

In (Capone et al., 2014) the channel adaptation mech-

anism was considered, but the details about the imple-

mentation of this mechanism were not provided. The

CA-DSME is based on the channel adaptation mech-

anism for the CFP periods of the superframes. An

algorithm to estimate the quality of the links is used

to decide when a channel switch is needed. The H-

DSME is a hybrid approach that uses channel adapta-

tion in the CFP periods, and channel hopping for the

beacons, and GACK frames.

This paper focuses in channel diversity tech-

niques. Thus, to simplify the analysis at this point, a

star topology was considered. In future works, the ex-

tension of the proposed approaches for tree and mesh

networks, and considering a larger number of nodes,

will be studied.

3.1 CH-DSME

Figure 3 shows the frame structure implemented for

the CH-DSME. A network with 10 nodes, and star

topology was considered, in which nine end nodes

(Node 1 to Node 9) are connected, and transmit pack-

ets directly to the PAN coordinator (Node 0), and only

one transmission occurs in each time slot. In Figure 3

the numbers in the time slots indicate the ID of the

end node that performs a transmission in each time

slot. The capReduction was enabled, and thus only

the ﬁrst superframe has the CAP.

Each node has two time slots to transmit pack-

ets. The ﬁrst one is before the ﬁrst GACK (G1), and

the second one is placed between G1 and the second

GACK (G2). Thus, if the transmission in the ﬁrst at-

tempt fails, the end node can try again, using other

channel, inside the same multi-superframe. The sec-

ond time slot of each node is used only for retrans-

mission. If the ﬁrst transmission occurs successfully,

the second time slot is not used. The beacons, and

GACK frames are always transmitted using the same

channel (Channel 0 in Figure 3). For transmission of

data packets, the channel to be used in a given time

slot i is determined using

c(i) = (i + j × l + macChannelO f f set + BSN)%16,

where j is the superframe index, macChannelOffset

is the channel offset of a receiver node, BSN is a se-

quence number of the beacon sent by the PAN coor-

dinator. The value of l is equal to 15 if CAP reduction

is enabled and j is not equal to zero, or 7 otherwise.

In the implementation built for this paper, the mac-

ChannelOffset is equal to the ID of the receiver node.

All the 16 channels were considered, thus the value of

c(i) is an integer between 0 and 15, which represents

the channel identiﬁcation.

3.2 CA-DSME

The CA-DSME also uses the frame structure shown in

Figure 3, but with a different channel diversity mech-

anism. In CA-DSME all nodes transmit to the coor-

dinator using only one channel, without channel hop-

ping. All the nodes use the same channel at the be-

ginning. Since only one transmission occurs in each

time slot, the end nodes can use the same channel in

the transmissions. In scenarios with more than one

transmission at the same time, the channels need to

be allocated in order to avoid collisions. A channel

switch only occurs when the channel of a given link

starts to present bad quality, and only the channel of

the affected link is changed.

When using channel adaptation in the CFP peri-

ods, it is possible to pick good channels for all links, if

Comparison between Channel Hopping and Channel Adaptation for Industrial Wireless Sensor Networks

Figure 3: The time structure used in the simulations.

the link quality estimator is able to quickly and accu-

rately estimate the link quality. The experiments de-

scribed in (Du and Roussos, 2013; Grsu et al., 2016)

showed that by using only one channel, the network

performance is higher than using a channel hopping

mechanism with a larger set of channels. However,

the channels to be used need to be properly chosen.

Since the nodes try to retransmit a lost packet only

once, and inside the same multi-superframe, it is pos-

sible to calculate the Packet Reception Rate (PRR),

using the information obtained from a set of received

packets. The Algorithm 1 was used to calculate the

Packet Reception Rate of each link, in which each

packet has an ID (a sequence number that identiﬁes

the packet) and the information about in which slot it

was received, that is, before G1 (slot 1) or after G1

(slot 2). If the same packet is received twice due to

a fail in the reception of G1, only the ﬁrst received

packet is put on the packet list to be analyzed by the

algorithm. This algorithm also considers that the sec-

ond time slot is used only for retransmission.

Using the Algorithm 1 the coordinator calculates

the PRR for each link using a window of N packets. A

Algorithm 1: Algorithm to calculate the Packet Reception

Rate.

Input: a list of packets packet list with N packets, and the

expected id for the ﬁrst packet in the list f p

Output: the Packet Reception Rate (PRR)

1: f ail cont := 0

2: for each packet in packet list do

3: if packet.slot = 1 then

4: f ail cont := f ail cont + 2 · (packet.id − f p

)

5: else

6: f ail cont := f ail cont + 2 · (packet.id − f p

) +

7: end if

8: f p

:= packet.id + 1

9: end for

10: PRR :=

N+ f ail cont

11: return PRR

threshold can be deﬁned for each link, in order to trig-

ger the channel switch procedure. In the implementa-

tion built for this paper, N = 10, and a threshold equal

to 0.9, for all nodes, were used. Each new PRR value

obtained using the Algorithm 1 is combined with the

last calculated value using an Exponentially Weighted

Moving Average (EWMA) ﬁlter with history control

factor α = 0.3, to make the calculated PRR more sta-

ble over time. Higher values for α can make the esti-

mator more stable, but the reactivity becomes smaller.

In the beacon frame, there is a bitmap to indicate

to each node if it needs to perform a channel switch.

When the PRR calculated for a given link is below its

threshold, the coordinator sets the corresponding bit

on the bitmap to 1. When the node receives a beacon

indicating the need of a channel switch, it switches

to the next channel, in a round-robin fashion. While

the coordinator does not receive a packet in the new

channel, it continues to send the beacon with the bit

equal to 1 in the bitmap. After receiving the ﬁrst

packet in the new channel, the coordinator clears the

bit. Since the beacon frames are always sent using the

same channel, if the transmission of a beacon fails, the

nodes can wait for the next beacon to re-synchronize.

A mechanism to identify deep fading problems

was also implemented. As the coordinator needs to

receive data packets to calculate the PRR using the

Algorithm 1, when the link between a given end node

and the coordinator enters in a deep fading state, no

packets can be received while the channel remains in

that state. Thus, when the coordinator does not re-

ceive packets from a given end node during a long

period, it starts the channel switch procedure for that

node. In the implementation built for this paper, the

coordinator starts a channel switch procedure when

no packet is received from a given end node during

10 consecutive BI.

In both CH-DSME and CA-DSME the channel

used to transmit the beacons and the GACK frames

SENSORNETS 2017 - 6th International Conference on Sensor Networks

is a single point of failure. In these protocols, it is

possible to deal with problems that affect the qual-

ity of a subset of channels through channel hopping

or channel adaptation, but they are not capable of

dealing with problems that affect the channel used to

transmit the beacons and the GACK frames. Thus,

in this paper a new hybrid approach is proposed (the

H-DSME), which is better explained in Section 3.3.

3.3 H-DSME

The beacon and GACK frames are transmitted in

broadcast mode to all the end nodes connected to the

coordinator. Therefore, the channel used to trans-

mit these frames needs to present good quality for

all links between the coordinator and the end nodes.

However, spatial variations in the channel quality can

occur. The coherence length is used to quantify the

maximum change in distance that will result in the

channel being highly correlated. In experiments de-

scribed in (Watteyne et al., 2010), it was veriﬁed a

coherence length of 5.5 cm for IEEE 802.15.4 radios

operating in the 2.4 GHz band. Thus, two nodes po-

sitioned more than 5.5 cm apart from each other, and

using the same channel, can be considered uncorre-

lated, and thus the channel can present high quality

for one node, and low quality for the other.

Although the use of only one channel during a

large time period can be advantageous for the CFP

periods, as explained in Section 3.2, it may be dif-

ﬁcult to guarantee a good qualify of service for all

end nodes when using only one channel for the trans-

mission of beacons and GACK frames. Thus, the

H-DSME uses channel adaptation for the CFP peri-

ods (in the same way of CA-DSME), and channel

hopping for the transmission of beacons and GACK

frames. The channels are used in a round-robin fash-

ion to transmit the beacons and GACK frames. Using

this mechanism, the end nodes do not remain discon-

nected for a large time period, when one channel be-

gins to decrease its quality regarding the coordinator

link.

The IEEE 802.15.4e standard deﬁnes that the

channel used in the set-up of the network needs to be

used to transmit the beacons and for the transmissions

in the CAP period (802, 2012). However, the modi-

ﬁcation to use multiple channels can be done with no

interference with the other parts of the protocol.

When dynamic addition of nodes is considered,

the nodes that want to join the network need to listen

in some channel during up to 16 BI. If no beacon is re-

ceived, the node can start to listen in another channel

during other 16 BI. When only one channel is used to

transmit the beacon frames, as deﬁned originally by

the standard, the node listens in one channel during

only one BI. However, if the end node doesn’t know

the channel used to transmit the beacons a priori, in

some cases it will be necessary to wait for multiple

BI (16 in the worst case) until a beacon is received. In

addition, if the channel in use to transmit the beacons

presents a very low quality for the link between the

coordinator and the new node, the delay to join the

network can be very high. This aspect will be better

evaluated in future works.

To accommodate the use of channel hopping in the

transmission of the beacons, it is necessary to have a

mechanism to maintain the network synchronized in

case of failures during the reception of a beacon. To

do this, a timer is used in the end nodes to identify

that a beacon has been lost. The coordinator sends a

new beacon for each BI, thus the timer is conﬁgured

to expire after a time equal to BI +

ms, where

is the duration of a time slot. The values of BI and

SD depend on the values of the parameters BO and

SO, respectively. If a node receives a new beacon be-

fore the timer expires, the timer is reseted. Otherwise,

the node switches the channel, and waits for the next

beacon, which maintains the synchronization.

4 EVALUATION

METHODOLOGY

The wireless channel can be considered as stationary

for a short term, despite the moving parts around the

transmitter and the receiver. However, the properties

of the channel can change signiﬁcantly over time due

to changes in the topology of the environment, which

are not considered in the distributions used to model

the fading. This may require the recalculation of the

distribution parameters, since these parameters may

become obsolete over time (Agrawal et al., 2014b).

A characterization of the wireless channel in in-

dustrial environments was performed for a long term

(20 hours) in (Agrawal et al., 2014a). The experiment

demonstrated that abrupt changes in the channel char-

acteristics can occur when the channel is analyzed for

a long time, and differences on the mean value of the

received power are observed, although the transmit-

ter and receiver remain static. For example, in the

experiment described in (Agrawal et al., 2014a), the

received power varied about -55 dBm during seven

hours, and after this period the mean value of the re-

ceived power changed abruptly to -46 dBm. An ex-

periment described in (Olofsson et al., 2016) also pre-

sented similar behavior, showing the special nature of

these environments.

To allow the simulation of protocols for industrial

Comparison between Channel Hopping and Channel Adaptation for Industrial Wireless Sensor Networks

Figure 4: Asymmetry and temporal variations in the re-

ceived power.

WSN, it is necessary to use a model that takes into

account the channel characteristics for a long period

of time. In a previous article (Gomes et al., 2015b),

a simulation model was developed, which includes

the effects of fading, log-normal shadowing, and the

non-stationary characteristics of the channel. In this

model, different channels can present different char-

acteristics, since the channels deﬁned by the physi-

cal layer of the IEEE 802.15.4e are uncorrelated in

frequency. Experiments described in (Amzucu et al.,

2014) have demonstrated that changing the communi-

cation channel can lead to a difference of up to 30 dB

in the received power, in an ofﬁce environment. Ex-

periments described in (Watteyne et al., 2010), in an

ofﬁce environment, showed that for distances greater

than 6.5 m between transmitter and receiver, even the

adjacent channels are uncorrelated.

In the current implementation, two instances of

the model are used to model the wireless channel in

the two directions of a link, to capture the asymme-

try. In the model, abrupt changes in the channels

characteristics can occur. A mean time of change is

deﬁned for the model, which is used to deﬁne the

value of a parameter p, the probability that a change

in the characteristics of the channels occur. Thus, it

is possible to simulate environments that remain un-

changed for a long period of time and environments

that present frequent changes in the topology. The

simulation result obtained using the model is compat-

ible with results from experiments performed in in-

dustrial environments (Agrawal et al., 2014b) (Gomes

et al., 2015a) (Olofsson et al., 2016).

Figure 4 shows the reception power at a receiver

(obtained from received packets) and a transmitter

(obtained from received ACKs) during ﬁve hours of

simulation to test the model. It is possible to notice

the abrupt changes that occur in the channel charac-

teristics over time, and the asymmetry between the

two directions of the link.

Table 1: Parameters used in the simulation.

Area 60 x 60 meters

Physical layer IEEE 802.15.4

Bit rate 250 kbit/s

Simulation Time 7200 s (2 hours)

Transmission power 0 dBm

Packet transmission rate 1 packet/s

Mean time of change 40 minutes

Table 2: Position of the nodes in the simulations.

Node

Coordinates

(X, Y, Z)

Distance to the

coordinator (node 0)

0 (-8.13, 7.66, 2) -

1 (-14.53, 2.66, 2) 8.12 meters

2 (-22.83, 8.91, 2) 14.75 meters

3 (-12.25, -19.79, 2) 27.76 meters

4 (16.66, -11.84, 2) 31.54 meters

5 (-12.46, -15.26, 2) 23.33 meters

6 (-1.93, 1.65, 2) 8.63 meters

7 (-13.60, -20.99, 2) 29.17 meters

8 (22.60, -5.45, 2) 33.41 meters

9 (-15.30, 9.73, 2) 7.46 meters

To evaluate the performance of the three ap-

proaches, ﬁve replications of the experiment were

made. Table 1 shows the parameters considered in the

simulations for each replication. For the lognormal

shadowing model the values of n = 1.69, d

= 15 m,

L(d

) = 80.48 dB, and X

= 6.62 dB were used.

These values were obtained from experiments in an

industrial environment described in (Tanghe et al.,

2008). The mean time of change deﬁnes the average

time between two changes in the characteristics of the

channel.

To perform a fair comparison, for each replication

the same seed was used to evaluate each approach,

and different seeds were used for different replica-

tions. Thus, the three approaches were evaluated con-

sidering the nodes positioned at the same position and

with the same channel characteristics during the repli-

cations. The positions of the nodes are shown in Ta-

ble 2.

The frame structure used in the simulations is

shown in Figure 3. Table 3 shows the values of the pa-

rameters that were used to conﬁgure the frame struc-

ture. In this conﬁguration, the SO is equal to the BO,

thus each BI has only one multi-superframe. Each

multi-superframe has two superframes, in which only

the ﬁrst one has the CAP. With this conﬁguration, the

beacon interval has approximately 0.246 s, which is

enough to accommodate the application implemented

for the simulations, that transmits one packet per sec-

ond.

SENSORNETS 2017 - 6th International Conference on Sensor Networks

Table 3: Parameters of the frame structure.

macBeaconOrder (BO) 4

macMultisuperframeOrder (MO) 4

macSuperframeOrder (SO) 3

Time slot duration 7.68 ms

capReduction enabled

Group Ack enabled

5 RESULTS

Four metrics were used to evaluate the approaches,

that is: the Packet Reception Rate (PRR) at the ap-

plication layer, the PRR at the MAC layer, the de-

lay, and the maximum time between the reception of

two consecutive packets. The PRR at the application

layer considers the relation between the packets re-

ceived and transmitted without considering the num-

ber of retransmissions at the MAC layer, while the

PRR at the MAC layer considers the retransmissions.

The delay is the time between the transmission of a

packet and the reception of the packet at the applica-

tion layer. The maximum time between the reception

of two packets was analyzed to investigate the time

in which the nodes remain disconnected due to low

channel quality.

Figure 5 shows the PRR at the application layer

for the nine end nodes. The averages and conﬁdence

intervals were calculated considering the data ob-

tained from all replications. For the nodes positioned

further apart from the sink node, the PRR was smaller

when using CH-DSME and CA-DSME. When using

the H-DSME the PRR at the application layer was al-

most 100% for all nodes. When using CH-DSME and

CA-DSME, most errors occurred due to failures in the

transmission of the beacons.

Figure 5: PRR at the application layer.

In general, the performance of the CA-DSME was

better than the CH-DSME, except for the End Node 8.

This shows that in most cases the use of only one

Figure 6: PRR at the MAC layer.

channel during a larger time period provides a bet-

ter quality than the use of all available channels with

channel hopping. However, in scenarios in which the

quality of the link between the coordinator and the

end node for beacon transmissions is very low, the

channel adaptation mechanism implemented for this

research can delay a long time to perform the chan-

nel switch. This was the case of the End Node 8.

When using H-DSME the channel adaptation mech-

anism was more reactive, since multiple channels are

used to transmit the beacons, which eliminates the

single point of failure.

Figure 6 shows the PRR at the MAC layer. A

transmission of a data packet in the MAC layer only

occurs when a beacon is received at the end node.

Thus, the failures in the beacon transmissions do not

inﬂuence the PRR calculated at this layer, since only

the transmissions of data packets from the end nodes

to the coordinator are considered. It is worthy to note

that when using CH-DSME the PRR at the MAC layer

is signiﬁcantly lower.

Since in the evaluated protocols two attempts are

possible per packet, and in the CH-DSME different

channels are used in each attempt, in most cases the

packet is delivered at the application layer. How-

ever, when more retransmissions are needed the en-

ergy consumption of the end nodes can increase sig-

niﬁcantly. Again, the End Node 8 was the only ex-

ception, and the PRR was lower for the CA-DSME

than for the CH-DSME for this end node. Even both

CA-DSME and H-DSME use channel adaptation, the

PRR at the MAC layer for the H-DSME was higher,

since the channel adaptation mechanism was more re-

active when using channel hopping for the beacons, as

observed also in Figure 5.

Figure 7 shows the cumulative distribution func-

tion of the delay. Since the beacon period is approxi-

mately 0.25 seconds, 93% of the packets were trans-

mitted with a delay lower than 0.25 seconds, and 99%

of the packets were transmitted with delay lower than

Comparison between Channel Hopping and Channel Adaptation for Industrial Wireless Sensor Networks

Figure 7: Delay at the application layer.

Figure 8: Maximum time between the reception of two con-

secutive packets.

0.5 seconds, which corresponds to two beacon inter-

vals. Since the three evaluated approaches use the

same superframe structure, the delay distribution was

very similar. However, the delay is only computed for

the delivered packets, and does not account for lost

packets.

Figure 8 shows the maximum time lapse between

the reception of two packets, which represents the

maximum time interval in which a node remains dis-

connected. Since CA-DSME, and CH-DSME use

only one channel for beacon transmissions, some-

times this channel can present low quality, for some

nodes, during a long time period, due to a deep fad-

ing. The same channel can present good quality for

other nodes, due to the spatial variation in the channel

quality.

When using CH-DSME and CA-DSME, End

Nodes 3, 4, and 5 presented long disconnection times

(88, 29, and 72 minutes) due to problems in the chan-

nel used to transmit the beacon frames. Since the H-

DSME uses channel hopping to transmit the beacons,

this protocol is more robust against deep fading prob-

lems that affects only some channels. Besides, it is

possible to deal with the spatial variation in the qual-

ity of the channels. For the H-DSME the maximum

time of disconnection was 1.6 minutes, for the End

Node 4.

As an example of the implemented channel adap-

tation mechanism, Figure 9 shows the reception

power for End Node 8 in the ﬁrst hour of simulation of

the ﬁrst replication, when using the H-DSME. In the

charts, some moments in which the channel switch

procedure is triggered are highlighted. It is possible

to notice the difference in the characteristics of the

different channels when the channel switch occurs.

Figure 9: Received power at the End Node 8 in (a) and the

calculated PRR using the Algorithm 1 (b), using H-DSME.

Sometimes, the channel switch procedure is trig-

gered several times in a short time interval, until the

node picks a good channel. For example, between

1.38 and 3.13 minutes, the End Node 8 switched its

channel ﬁve times. However, when the End Node

picks a good channel, it can remain for a long time

using the same channel. In some cases, as demon-

strated in the experiments described in (Agrawal

et al., 2014a), the channels can maintain the same

characteristics during several hours before an abrupt

change in its characteristics occurs. In the simulations

a threshold of 0.9 was deﬁned. If the application sup-

ports lower thresholds, the channel switch procedure

is triggered less often.

SENSORNETS 2017 - 6th International Conference on Sensor Networks

6 CONCLUSIONS

This paper presented a comparison between chan-

nel hopping and channel adaptation for industrial

WSN. Three protocols, based on the DSME mode of

the IEEE 802.15.4e standard, were implemented and

evaluated using a realistic simulation model. This

model permitted to observe the performance of the

protocols, considering the non-stationary behavior of

the wireless channel in industrial environments.

A novel hybrid approach (H-DSME), that uses

channel hopping for the transmission of beacons and

ACK frames, and channel adaptation for the trans-

mission of data packets, was proposed. The results

showed that the use of channel adaptation is better

than channel hopping, when the quality of the links is

monitored continuously by the network coordinator.

Since the beacon frames and ACK frames are

transmitted in broadcast mode to several nodes, it is

difﬁcult to guarantee that a single channel will present

good enough quality for all nodes. Thus, the use of

channel hopping to transmit the frames is a good al-

ternative to deal with the channel spatial variation in

quality.

The H-DSME outperformed the other two ap-

proaches. In the performed simulations, even for

nodes that were positioned further apart from the co-

ordinator, and even considering the variations in chan-

nel quality over time, the PRR at the application layer

was almost 100% when using the H-DSME, and the

number of retransmissions was also reduced. Be-

sides, 99% of the packets were delivered with a delay

smaller than 0.5 s, which corresponds to two beacon

intervals, and the maximum time of disconnection of

the nodes was only 1.6 minutes.

This paper describes an analysis of the use of

channel adaptation to implement industrial WSN

based on the DSME mode. However, the algorithm

used to estimate the channel quality considers that all

the operation of the transmitter is predictable, that is,

only one retransmission is allowed, and there is a slot

dedicated to transmissions, and other slot dedicated

only for retransmissions.

A more ﬂexible use of the time slots, for exam-

ple, the use of two transmissions of different packets

in the same multi-superframe, and the use of a vari-

able number of retransmission attempts, can lead the

algorithm to present a low accuracy, unless more in-

formation obtained at the end nodes are provided to

the coordinator, such as the average number of trans-

mission attempts per packet.

Other aspect to be considered is the overhead

caused by the algorithm. Experimental studies will

be performed to investigate if the continuous monitor-

ing of the quality of the links is viable using low-cost

sensor nodes. The use of dedicated nodes to moni-

tor channel quality, as proposed initially in (Gomes

et al., 2014), as well as the use of estimators based on

physical layer metrics, will be studied. Experiments

considering different network topologies will also be

performed.

ACKNOWLEDGEMENTS

The authors would like to thank the support of the

COPELE-UFCG, the Institute for Advanced Studies

in Communications (Iecom), the Brazilian Council

for Research and Development (CNPq), the Coordi-

nation for the Improvement of Higher Education Per-

sonnel (Capes), and the SMART 2 Project of the Eras-

mus Mundus Programme.

REFERENCES

(2012). Ieee standard for local and metropolitan area

networks–part 15.4: Low-rate wireless personal area

networks (lr-wpans) amendment 1: Mac sublayer.

IEEE Std 802.15.4e-2012 (Amendment to IEEE Std

802.15.4-2011), pages 1–225.

Agrawal, P., Ahlen, A., Olofsson, T., and Gidlund, M.

(2014a). Characterization of long term channel varia-

tions in industrial wireless sensor networks. In IEEE

International Conference on Communications, pages

1–6.

Agrawal, P., Ahl

en, A., Olofsson, T., and Gidlund, M.

(2014b). Long term channel characterization for en-

ergy efﬁcient transmission in industrial environments.

IEEE Trans. on Communications, 62(8):3004–3014.

Alderisi, G., Patti, G., Mirabella, O., and Bello, L. L.

(2015). Simulative assessments of the ieee 802.15.4e

dsme and tsch in realistic process automation scenar-

ios. In 2015 IEEE 13th International Conference on

Industrial Informatics (INDIN), pages 948–955.

Amzucu, D. M., Li, H., and Fledderus, E. (2014). Indoor

radio propagation and interference in 2.4 ghz wireless

sensor networks: Measurements and analysis. Wire-

less Personal Communications, 76:245–269.

Anastasi, G., Conti, M., Francesco, M. D., and Passarella,

A. (2009). Energy conservation in wireless sensor net-

works: A survey. Ad Hoc Networks, 7(3):537 – 568.

Capone, S., Brama, R., Ricciato, F., Boggia, G., and Mal-

vasi, A. (2014). Modeling and simulation of energy

efﬁcient enhancements for ieee 802.15.4e dsme. In

2014 Wireless Telecommunications Symposium, pages

1–6.

Du, P. and Roussos, G. (2011). Adaptive channel hopping

for wireless sensor networks. In Mobile and Wireless

Networking (iCOST), 2011 International Conference

on Selected Topics in, pages 19–23.

Comparison between Channel Hopping and Channel Adaptation for Industrial Wireless Sensor Networks

Du, P. and Roussos, G. (2013). Spectrum-aware wireless

sensor networks. In 2013 IEEE 24th Annual Inter-

national Symposium on Personal, Indoor, and Mobile

Radio Communications (PIMRC), pages 2321–2325.

Gnawali, O., Fonseca, R., Jamieson, K., Moss, D., and

Levis, P. (2009). Collection tree protocol. In Pro-

ceedings of the 7th ACM Conference on Embedded

Networked Sensor Systems, SenSys ’09, pages 1–14,

New York, NY, USA. ACM.

Gomes, R. D., Alencar, M. S., Queiroz, D. V., and Fon-

seca, I. E. (2016). Evaluation of link quality estima-

tors for industrial wireless sensor networks. In XXXIV

Simp

osio Brasileiro de Telecomunicac¸

oes e Processa-

mento de Sinais, pages 1–5.

Gomes, R. D., Fonseca, I. E., and Alencar, M. S. (2015a).

Protocolos multicanais para redes de sensores sem ﬁo

industriais (in portuguese). Revista de Tecnologia da

Informac¸

ao e Comunicac¸

ao, 5(2):25–32.

Gomes, R. D., Queiroz, D. V., Fonseca, I. E., and Alen-

car, M. S. (2015b). Modelo para simulac¸

ao realista de

redes de sensores sem ﬁo industriais (in portuguese).

In XXXIII Simp

osio Brasileiro de Telecomunicac¸

oes,

pages 1–5.

Gomes, R. D., Rocha, G. B., Filho, A. C., Fonseca, I. E.,

and Alencar, M. S. (2014). Distributed approach for

channel quality estimation using dedicated nodes in

industrial wsn. In Personal, Indoor, and Mobile Ra-

dio Communication (PIMRC), 2014 IEEE 25th An-

nual International Symposium on, pages 1943–1948.

Guglielmo, D. D., Brienza, S., and Anastasi, G. (2016).

{IEEE} 802.15.4e: A survey. Computer Communi-

cations, 88:1 – 24.

Grsu, M., Vilgelm, M., Zoppi, S., and Kellerer, W. (2016).

Reliable co-existence of 802.15.4e tsch-based wsn

and wi-ﬁ in an aircraft cabin. In 2016 IEEE Inter-

national Conference on Communications Workshops

(ICC), pages 663–668.

Jeong, W.-C. and Lee, J. (2012). Performance evaluation

of ieee 802.15.4e dsme mac protocol for wireless sen-

sor networks. In Enabling Technologies for Smart-

phone and Internet of Things (ETSIoT), 2012 First

IEEE Workshop on, pages 7–12.

Juc, I., Alphand, O., Guizzetti, R., Favre, M., and Duda,

A. (2016). Energy consumption and performance of

ieee 802.15.4e tsch and dsme. In 2016 IEEE Wireless

Communications and Networking Conference, pages

1–7.

Lee, J. and Jeong, W. C. (2012). Performance analysis of

ieee 802.15.4e dsme mac protocol under wlan interfer-

ence. In 2012 International Conference on ICT Con-

vergence (ICTC), pages 741–746.

Lima-Filho, A., Gomes, R., Adissi, M., Borges da Silva,

T., Belo, F., and Spohn, M. (2012). Embedded sys-

tem integrated into a wireless sensor network for on-

line dynamic torque and efﬁciency monitoring in in-

duction motors. IEEE/ASME Trans. on Mechatronics,

17(3):404–414.

Olofsson, T., Ahln, A., and Gidlund, M. (2016). Model-

ing of the fading statistics of wireless sensor network

channels in industrial environments. IEEE Trans. on

Signal Processing, 64(12):3021–3034.

Patti, G., Alderisi, G., and Bello, L. L. (2014). Introduc-

ing multi-level communication in the ieee 802.15.4e

protocol: The multichannel-lldn. In Proceedings of

the 2014 IEEE Emerging Technology and Factory Au-

tomation (ETFA), pages 1–8.

Patti, G. and Bello, L. L. (2016). A priority-aware multi-

channel adaptive framework for the ieee 802.15.4e-

lldn. IEEE Transactions on Industrial Electronics,

PP(99):1–1.

Petersen, S. and Carlsen, S. (2009). Performance evalua-

tion of wirelesshart for factory automation. In IEEE

Conference on Emerging Technologies & Factory Au-

tomation, pages 1–9.

Tanghe, E., Joseph, W., Verloock, L., Martens, L., Capoen,

H., Herwegen, K. V., and Vantomme, W. (2008). The

industrial indoor channel: Large-scale and temporal

fading at 900, 2400, and 5200 mhz. IEEE Trans. on

Wireless Communications, 7:2740–2751.

Watteyne, T., Lanzisera, S., Mehta, A., and Pister, K. S. J.

(2010). Mitigating multipath fading through channel

hopping in wireless sensor networks. In 2010 IEEE

International Conference on Communications, pages

1–5.

SENSORNETS 2017 - 6th International Conference on Sensor Networks