MAC SCHEDULING IN LARGE-SCALE UNDERWATER

ACOUSTIC NETWORKS

Wouter van Kleunen, Nirvana Meratnia and Paul J. M. Havinga

Pervasive Systems, University of Twente, 7522 ND, Enschede, The Netherlands

Keywords:

MAC Scheduling, Large-scale underwater acoustic networks, Spatio-temporal scheduling.

Abstract:

The acoustic propagation speed under water poses signiﬁcant challenges to the design of underwater sensor

networks and their medium access control protocols. Scheduling allows reducing the effects of long propaga-

tion delay of the acoustic signal and has signiﬁcant impacts on throughput, energy consumption, and reliability.

In this paper we propose two approaches for scheduling large-scale underwater networks. One approach is a

centralized scheduling approach, which yields the most efﬁcient schedules but has signiﬁcant communication

and computational overhead. Our second approach uses clustering to split up the network in independent

clusters. This approach provides signiﬁcant beneﬁts in terms of communication and computation, but yields

less efﬁcient schedules than the centralized approach.

We evaluate both approaches in terms of efﬁciency, communication overhead and computation overhead of

the resulting schedule. We show that the centralized approach yield the best schedule at the cost of signiﬁcant

communication and computation overhead. The distributed scheduling approach yield less efﬁcient schedules

but offers signiﬁcant communication and computational complexity advantages.

1 INTRODUCTION

Design of an efﬁcient medium access control proto-

col for underwater acoustic networks is difﬁcult. Un-

derwater communication suffers from the slow acous-

tic propagation speed. Several underwater MAC pro-

tocols exist, examples include T-Lohi (Syed et al.,

2007) and Slotted-Fama (Molins, 2006). However

these protocols do not provide very high bandwidth,

while there is an interest in scheduling transmissions

to provide higher bandwidths. Examples of scheduled

MAC protocols include ST-MAC (Hsu et al., 2009),

STUMP (Kredo II and Mohapatra, 2010). In our pre-

vious paper (van Kleunen et al., 2011) we have pro-

posed a MAC scheduling approach which is far much

simpler than existing approaches. We have achieved

this simplicity by ﬁrst deriving a set of simpliﬁed

scheduling constraints and using these constraints in

design of a simple scheduling algorithm for underwa-

ter communication.

Scheduling communication brings signiﬁcant ad-

vantages to underwater acoustic networks in that it ex-

ploits spatio-temporal uncertainty and reduces the im-

pact of long propagation delays. Figure 1 illustrates

how scheduling can exploit the long propagation de-

lays to provide higher throughputs.









  







(a) Exclusive access









  



(b) Scheduled

Figure 1: Exploiting spatio-temporal uncertainty of under-

water communication.

Centralized MAC scheduling approaches are able

to calculate efﬁcient schedules but this comes with

signiﬁcant costs. To run a centralized algorithm all

link and node information should be collected at a

central place, which introduces a signiﬁcant commu-

nication overhead. If the network is large or dynamic,

it may not be feasible or desired to collect all link in-

formation at a central place. The computational com-

plexity of a centralized algorithm can also be high.

To this end, the ability to run the algorithm distribut-

edly on several nodes can reduce the overall compu-

tation time. Using a distributed approach will come

at the cost of suboptimal schedules, which should be

reduced as much as possible.

In this paper we present how the complexity of

van Kleunen W., Meratnia N. and J. M. Havinga P..

MAC SCHEDULING IN LARGE-SCALE UNDERWATER ACOUSTIC NETWORKS.

DOI: 10.5220/0003438600270034

In Proceedings of the International Conference on Wireless Information Networks and Systems (WINSYS-2011), pages 27-34

ISBN: 978-989-8425-73-7

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

given j for all i < j,











.start ≥ δ

.start + δ

.duration if δ

.src = δ

.src

.start ≥ δ

.start +δ

.duration+max(

T(δ

.src,δ

.dst) −T(δ

.src,δ

.dst),

T(δ

.src,δ

.dst) −T(δ

.src,δ

.dst))

if δ

.src 6= δ

.src

Figure 2: Set of simpliﬁed scheduling constraints.

the algorithm proposed in (van Kleunen et al., 2011),

which allows exploiting the long propagation delays

in underwater acoustic communication, can be re-

duced to allow scheduling of large-scale underwater

networks. This reduced complexity centralized ap-

proach can be used in scenarios in which the com-

munication overhead is not an issue and the aim is to

achieve the best possible schedule.

We will also present how a clustered approach can

signiﬁcantly reduce the computational and communi-

cation overhead of scheduling large-scale underwater

networks. A large part of the communication of our

distributed clustered approach is done locally (sin-

gle hop) and the computational communication is low

enough to be implemented on embedded processors.

The distributed clustered approach calculates less ef-

ﬁcient schedule but the communication and computa-

tional overhead scales much beter to larger-scale un-

derwater networks.

The outline of the paper is as follows: in Sec-

tion 2 the related work will be discussed. An extended

set of simpliﬁed scheduling constraints for scheduling

multi-hop networks will be presented in Section 3.

In Section 4 a centralized algorithm with reduced

complexity is proposed. Section 5 describes our dis-

tributed scheduling technique, which uses a clustering

concept to split up the scheduling problem. Evalu-

ation of the communication and computational com-

plexity of different approaches is presented in Sec-

tion 6, while performance evaluations of the different

approaches will be presented in Section 7. In Sec-

tion 8 conclusions are drawn and future directions are

highlighted.

2 RELATED WORK

Scheduling approaches for underwater acoustic trans-

missions, which allow mitigating the effects of the

propagation delay, already exist in literature. These

approaches range from centralized to distributed ap-

proaches. ST-MAC (Hsu et al., 2009) is a centralized

scheduling approach, which uses timeslots to form a

collision free schedule. In (van Kleunen et al., 2011)

we have shown that the slotted approach used by ST-

MAC leads to suboptimal results.

STUMP-WR (Kredo II and Mohapatra, 2010)

provides a fully distributed approach to schedule un-

derwater communication. It also uses timeslots, sim-

ilar to the approach of ST-MAC, which has been

shown to lead to suboptimal results. STUMP-WR de-

rives a node schedule from local interference patterns

and link schedules from neighbouring nodes. Nodes

broadcast their route as well as link schedule updates

during control frames until the network converges.

This approach is quite interesting because it is fully

distributed. Nodes are able to schedule their transmis-

sions based on their own information and schedules of

neighbours. This may, however, require a signiﬁcant

amount of communication between nodes and special

attention should be paid to ensure that the nodes con-

verge to a schedule in networks facing packet loss.

Because of the way STUMP-WR works, the pos-

sibilities of ordering the transmissions are limited.

The ﬁrst schedules will be formed using a limited

amount of transmissions. Other nodes will extend

their schedules using their transmissions but will

not move existing transmissions to improve schedule

times. This will result in less optimal schedules from

a throughput point of view.

Another problem with STUMP-WR (Kredo II and

Mohapatra, 2010) might be that transmissions still

cause interference at certain nodes. This is because

in a real network setup, nodes may not always be able

to communicate with nodes that they might interfer.

The interference range might actually be larger than

the communication range. The completely distributed

approach is an interesting approach nonetheless.

In (van Kleunen et al., 2011), we have already

shown how the scheduling constraints for underwa-

ter communication can be reduced to a simpliﬁed set.

Using these constraints we introduced two scheduling

algorithms: one algorithm assumes a given order of

transmissions and the other one selects the most op-

timal order of transmissions and derives a schedule.

The set of scheduling constraints assumes all nodes

are within transmission range of each other and all

schedulings are done in a centralized manner. This

necessitates availability of knowledge from all nodes

and transmissions at a central place in the network,

which introduces a signiﬁcant communication over-

head. In Figure 2 the set of simpliﬁed scheduling

constraints are given. The set gives a constraint be-

tween two transmissions (δ

and δ

), which both have

WINSYS 2011 - International Conference on Wireless Information Networks and Systems

a source (δ

.src and δ

.src), destination (δ

.dst and

.dst), and duration (δ

.duration and δ

.duration).

The function T gives the propagation time between

two positions. The simpliﬁed set gives a constraint

between transmission δ

and all previously scheduled

transmissions ( for all δ

where i < j). A minimum

start time can be calculated which satisﬁes the con-

straint with all earlier schedule transmissions. The

proposed algorithm in (van Kleunen et al., 2011),

which minimizes the schedule length, works by ﬁrst

calculating a two-dimensional table with the minimal

delay between all pairs of transmissions. The algo-

rithm considers every transmission as ﬁrst transmis-

sion, calculates a minimal schedule with every trans-

missions as ﬁrst transmission, and selects the best

schedule. The schedule is calculated by scheduling

transmissions iteratively. The transmission which has

the minimum delay with respect to all previous trans-

missions is selected as next transmission. This is a

greedy approach. The delays for all transmissions

with respect to all previous scheduled transmission

are updated after each iteration. The greedy approach

of trying to minimize the delays leads to the most ef-

ﬁcient schedule, in terms of schedule length. The re-

sulting schedule is collision-free for all nodes in the

network and it considers all nodes to be within com-

munication range (one-hop). The transmission times

are continous, although the algorithm can also be used

for calculating slotted transmission times.

In (Stojanovic, 2008) and (Peleato and Stojanovic,

2007) the possibility of underwater cellular networks

has been investigated. In the latter a protocol is

proposed for channel sharing using cellular network,

which only considers communication from and to the

base-station.

i = 1

j = 1

D = R

√

Figure 3: Cellular network example.

Figure 3 shows an example of a cellular network.

The whole area is split into hexagonal cells and every

cell is assigned a frequency. A group of cells together

form a cluster, within which a frequency is used only

once. The number of cells within a cluster determines

the reuse distance (D), i.e, the minimum distance be-

tween two cells that share the same frequency.

The work of Peleato et al. (Peleato and Stojanovic,

2007) does some form of scheduling by splitting up

the time into two phases. In the ﬁrst phase, nodes

that are the furthest away from the base-station are

allowed to transmit. In the second phase, the inner

nodes will transmit. These phases should run more-

or-less synchronized between the cells. The cells ex-

perience the most interference from surrounding cells

during the second phase, while the ﬁrst phase will be

more or less free of interference. The nodes in the sec-

ond phase are closer to their destination (which will

be the base-station) or source (also the base-station).

Therefore the signal will experience less attenuation

and is able to achieve the desired signal to noise ratio

even when the reuse distance (D) is small. Due to the

fact that the reuse distance is small a higher through-

put can be achieved.

3 EXTENDING THE SET OF

SIMPLIFIED RULES

Before we begin describing our proposed schedul-

ing algorithm, we ﬁrst extend the set of simpliﬁed

scheduling rules and explain how multihop schedul-

ing can be added.

Two nodes are outside of interference range of

each other if the signal of one node results in a re-

ceived signal strength on the other node which is be-

low a certain threshold (TH

). The value of this

threshold (TH

) should be chosen in such a way that

interfering signals are always below the receiver sen-

sitivity of the node or the interfering signal can be

guaranteed to be captured by the wanted transmission.

The received signal strength is dependant on the

output power of the sender and the attenuation be-

tween the sender and the receiver. The attenuation

between nodes is dependent on the absorption rate of

the water and the spreading of the signal. This path

loss equation can be written as follows:

10log(d, f) = k·10logd + d ·10loga( f) (1)

The path loss is dependant on the carrier fre-

quency (f) of the signal as well as the distance (d) be-

tween sender, and receivery. The spreading factor is

MAC SCHEDULING IN LARGE-SCALE UNDERWATER ACOUSTIC NETWORKS











.start ≥ δ

.start +δ

.duration

if δ

.src = δ

.src

.start ≥ δ

.start +δ

.duration+max(

T(δ

.src,δ

.dst) −T(δ

.src,δ

.dst),

T(δ

.src,δ

.dst) −T(δ

.src,δ

.dst))

if δ

.src! = δ

.src and (Inter f er(δ

.src,δ

.dst)

or Inter fer(δ

.src,δ

.dst))

.start ≥ δ

.start otherwise

(2)

Figure 4: Extended set of simpliﬁed scheduling constraints allowing multi-hop scheduling.

constant, which can either be spherical (k = 2), cylin-

drical (k = 1), or somewhere in between.

Using this formula we can calculate whether two

nodes interfer with each other. Consider two trans-

missions δ

and δ

, both transmissions have a source

(δ

.src and δ

.src) and destination (δ

.dst and δ

.dst).

We will use the path loss function (PL) to calculate

the difference of the received signal strengths at the

destination of transmissions (δ

Inter fer(δ

,δ

) = TRUE if

PL(δ

.src,δ

.dst) −PL(δ

.src,δ

.dst) ≤ TH

(3)

Function (3) will return false if transmission δ

does not cause interference for transmissions δ

. We

will now show how this equation can be applied to

the set of simpliﬁed scheduling rules. The interfer-

ence rule only applies when two nodes are able to in-

terfer with each others transmissions. If δ

.src is out

of range of δ

.dst and δ

.src is out of range of δ

.dst,

there is no constraint between the two transmissions.

In Figure 4 the set of extended scheduling con-

straints is shown. We added the interference condition

to the interference rule and added a scheduling rule in

case two transmissions are outside interference range.

This set of constraints can be used in large networks

where nodes can be outside of each others interfer-

ence range.

4 A CENTRALIZED

SCHEDULING APPROACH FOR

HIGH-EFFICIENCY

SCHEDULES

The extended set of simpliﬁed constraints from Sec-

tion 3 can be applied to design a scheduling algo-

rithm with low complexity for large-scale underwater

networks. The algorithm from (van Kleunen et al.,

2011), which has O(n

) complexity, considers every

transmission as the ﬁrst transmission. To reduce the

complexity, we can take a random or the ﬁrst trans-

mission as the transmission to be scheduled at time 0.

This will reduce the complexity of the algorithm from

O(n

) to O(n

When we calculate the schedule only once, there

is also no need anymore to precalculate a table of de-

lays for all transmission pairs. Any transmission pair

will be considered at most once, but some will never

be calculated. At the ﬁrst iteration the algorithm will

calculate the delays for n −1 pairs, the second itera-

tion n −2, and so forth. This will further reduce the

complexity from O(n

) to O(

). Because we do

not calculate the delay table, the memory space com-

plexity can also be reduced to O(n).

The full algorithm can be seen in Figure 5. The al-

gorithm initially schedules the ﬁrst transmission. In-

side the scheduling loop ﬁrst all the minimum start-

ing times for the remaining transmissions are calcu-

lated. The loop also ﬁnds the transmission with the

minimum schedule time and removes this transmis-

sion from the set of to be scheduled transmissions.

This is repeated until all transmissions are scheduled.

In this way not only the computational complexity re-

duced, but also the algorithm is now small and easy to

understand.

5 A DISTRIBUTED SCHEDULING

APPROACH WITH LOW

COMPUTATIONAL AND

COMMUNICATION

COMPLEXITY

In Section 4 we have presented an algorithm for

scheduling large-scale underwater networks in a cen-

tralized manner. However this algorithm requires

multi-hop communication to gather informationabout

all required transmissions within the network. This

has a signiﬁcant overhead and because it is done be-

fore scheduling, this communication will be done in

an unscheduled way.

To solve this communication overhead problem

we propose a distributed scheduling approach based

on a clustering concept. We propose a technique in

which cluster-heads are time-schedule arbriters for a

cluster and nodes will send a request to the cluster-

head to do a communication. The clusters are as-

signed a timeslot, which can span up to several sec-

onds and will schedule all the requested transmissions

WINSYS 2011 - International Conference on Wireless Information Networks and Systems

V ←transmissions {Set of all transmissions}

schedule ← [N] = 0 {Resulting schedule}

schedule[0] = 0 {Schedule the ﬁrst transmission}

time = 0

last = 0

V ←V \δ

{Remove transmission from set}

{Scheduling loop schedules transmissions greedy}

while !empty(V) do

time

min

← in finity

{Calculate minimum starting time for remaining trans-

missions}

for δ ∈V do

schedule[δ

index

] = max(schedule[δ

index

],time +

constraint(δ

last

,δ

index

)]

{See if this transmission has the smallest starting

time}

if schedule[δ

index

] < time

min

then

time

min

← schedule[δ

index

]

index ←δ

index

end if

end for

{Schedule transmission with smallest starting time

ﬁrst}

time = time

min

last = index

V ←V \δ

index

end while

Figure 5: Reduced complexity algorithm for scheduling

transmissions.

Figure 6: Example of a deployment.

in their timeslot. The timeslots can be reused in other

clusters and this will ensure that no interference or

minimal interference occurs between clusters.

In Figure 6 an example deployment is shown. The

cluster-heads are in the center of their cluster and the

numbers shown in the cluster indicate the used times-

lot of the cluster. The small dots are sensor nodes

scattered across the complete deployment area and the

lines between nodes indicate communication links.

Communication does not necessairly have to be done

from or towards the cluster-heads and can be done to

any node within the communication range. The links

are set up in such a way that information is collected

at a central sink.

The size of the clusters is dependant on the com-

munication range of the nodes. We assume that all

nodes in the network use the same output power for

transmissions and will therefore have the same com-

munication range. All nodes within the cluster should

be able to communicate with the cluster-head, there-

fore the cluster size should not be bigger than the

communication range. We assume the radius of the

cluster is exactly the size of the maximum communi-

cation range. The actual size can be calculated using

the path loss Equation (1).

The clusters in our approach are similar to cells

in a cellular network. If we assume the shape of a

cluster in our approach is hexagonal, we can then use

the equations from cellular networks to calculate the

number of timeslots required. The number of times-

lots determines the reuse distance, one may recall that

the reuse distance is the minimum distance between

two clusters that share the same timeslot, see Figure 3

for an example.

The number of timeslots can not arbitrarly be cho-

sen and is determined from the following formula:

N = i

+ ij + j

(4)

The i and j parameters determine the reuse dis-

tance of a timeslot along two axis. The reuse distance

(D) can be calculated from the number of cells per

cluster (N) and the cell radius (R):

D = R

√

3N (5)

The reuse distance is the minimum distance be-

tween two interfering senders in the network. The

larger the distance between two interferers, the less

interference experienced during communication. If a

total of 3 timeslots are used, the closest distance be-

tween two interfering nodes is exactly the radius of

the cluster. If more timeslots are used, the distance

between two interfering nodes will be larger, result-

ing in less noise from neighbouring clusters.

The nodes within a cluster all register their trans-

missions to the closest cluster-head. The cluster-

head is therefore able to schedule all the transmis-

sions within its cluster. After doing so, it will send

the minimum length of its local schedule to the cen-

tral cluster-head. The central cluster-head will assign

timeslots to the clusters and determine the length of

each timeslot. The timeslots do not necessairly have

MAC SCHEDULING IN LARGE-SCALE UNDERWATER ACOUSTIC NETWORKS

Cluster

1 2 3 4 5 6 Max

Slot 1 1.33 1.57 1.57

Slot 2 1.61 1.43 1.61

Slot 3 1.37 1.45 1.45

Slot length 1.57 1.61 1.45 1.57 1.61 1.45

Figure 7: Results of calculating slot length based on cluster

schedule lengths.

to be of equal time. The central cluster-head will as-

sign the maximum schedule length of all clusters that

share the same timeslot.

The cluster-heads will determine the order of

transmissions within their cluster. This can be

done using different optimization criteria as presented

in (van Kleunen et al., 2011). We will be using the

greedy approach in which transmissions are sched-

uled based on minimum delay.

For scheduling the transmissions within a cluster

we can use the algorithm from (van Kleunen et al.,

2011) or the reduced complexity algorithm from Sec-

tion 4. The second algorithm will yield a smaller com-

putational and memory space complexity, but because

the number of transmissions per cluster is in practice

limited, the ﬁrst algorithm may as well be a good op-

tion.

Figure 7 shows an example of how the algorithm

works. The table shows for all clusters the calculated

cluster schedule lengths. The cluster-head schedules

all transmissions within its cluster and determines the

clusters schedule length. The central scheduler deter-

mines the maximum of all schedule lengths per slot

and assigns the maximum schedule length to the slot.

The cluster-heads and the central scheduler only need

to communicate the schedule length and slot lengths.

6 EVALUATION OF

COMMUNICATION AND

COMPUTATION COMPLEXITY

We evaluate the different centralized and distributed

scheduling approaches. We will discuss brieﬂy their

complexity in terms of number of communications

required as well as computational complexity of dif-

ferent approaches. The complexity overview of all

scheduling approaches can be seen in Figure 8.

• Centralized Scheduling: In this case we assume

all transmission as well as position information is

collected in a central location. The communica-

tion complexity is n·hops

avg

(The average num-

ber of hops), because all transmission information

needs to be sent over a multi-hop link to the cen-

tral scheduler. For scheduling the links we will

use the algorithm described in (van Kleunen et al.,

2011), whose complexity is O(n

• Reduced Complexity Centralized Scheduling:

This is the algorithm described in Section 4.

The computational complexity of this algorithm

is O(

). The communication complexity is

the same as the other centralized scheduling ap-

proach, namely O(n

• Distributed Scheduling: In the distributed sit-

uation, the transmissions are sent only to the

cluster-head (O(n) communications). The cluster-

head will calculate a schedule for its own clus-

ter and will forward the length of its sched-

ule over a multi-hop link to the central sched-

uler (O(hops

avg

k) communications). On average,

the number of transmissions per cluster is n/k,

which results in a computational complexity of

O((n/k)

) per cluster, but also for the whole net-

work.

• Distributed Reduced Complexity Scheduling:

It is similar to the distributed approach, but the

scheduling per cluster uses the reduced com-

plexity centralized scheduling algorithm. This

reduces the scheduling algorithm complexity to

(n/k)

) per cluster. The communication com-

plexity remains O(hops

avg

k).

The packet size of all approaches is constant and

does not grow with respect to the number of nodes

in the network. From the evaluation of the complex-

ity of the different approach, we can see that the dis-

tributed approaches have a much lower computational

and communication overhead compared to the cen-

tralized approaches. The scalability of the distributed

approaches is therefore much better than the central-

ized approaches.

7 EVALUATION OF

SCHEDULING EFFICIENCY

To evaluate the scheduling efﬁciency of the different

approaches, we implement them in c++. We evalu-

ate the algorithms for different sizes of deployments.

The parameters can be found in Figure 9(a). The net-

work size ranges from 500 up to 8000 nodes scattered

randomly over an area. The communications are set

up in such a way that all data is collected at a central

sink, similarly to the deployment in Figure 6.

For the different distributed scheduling ap-

proaches a reuse distance should be selected. We

evaluated the distributed algorithms with both 3 as

well as 7 timeslots.

WINSYS 2011 - International Conference on Wireless Information Networks and Systems

Scheduling approach Computational Communication Packet size

Centralized O(n

) 2(n·hops

avg

) O(1)

Reduced Complexity Centr. O(

) 2(n·hops

avg

) O(1)

Distributed O((n/k)

) 2(n+ k·hops

avg

) O(1)

Distributed Reduced Complexity O(

(n/k)

) 2(n+ k·hops

avg

) O(1)

n = Number of transmissions

k = Number of clusters

Figure 8: Complexity of different scheduling approaches compared.

Parameter Value

Communication range: 500m

Data rate: 1000bps

Propagation speed: 1500 m/s

Node placement: random / uniform

(a) General parameters

Parameter Small Medium Large

Clusters: 4 x 3 7 x 7 14 x 14

Area size: 3.2 x 3.1km 5.5 x 6.6km 11 x 13km

Nodes: 500 2000 8000

(b) Different deployment sizes

Figure 9: Simulation parameters.

The evaluation results are shown in Figure 10(a).

We see that the centralized approach performs the

best, which is expected. This is due to the fact that

the centralized approach has all link and deployment

information of the network during the scheduling,

while the distributed approach splits up the schedul-

ing in sub-problems and uses local information only.

The centralized approach places a lower bound on the

achievable schedule length.

The reduced complexity centralized algorithm

performs only slightly worse, the difference in sched-

ule lengths is only marginal. Therefore the reduced

complexity centralized algorithm is a good alterna-

tive to the full complexity centralized algorithm. In

Section 4 and Section 6 we have already shown that

the reduced complexity algorithm has large beneﬁts in

terms of computation and memory complexity. From

the results of the simulation, we can conclude these

beneﬁts come at almost no cost in terms of schedule

efﬁciency.

Among the distributed approaches, the distributed

approach which minimizes schedule length and uses

3 timeslots, performs about twice as worse as the cen-

tralized approach. The approach that orders the trans-

missions based on distance of the transmission per-

forms worse. The fact that the distributed approach

performs worse when the network size increases is

because for every timeslot the maximum schedule

length from all clusters using that timeslot is used.

If more clusters use the same timeslot, the maximum

schedule length over all these clusters will go up.

In Figure 10(b) the amount of communications cy-























  

















(a) Schedule length of different scheduling ap-

proaches























  









Scheduling approach

Network size Centralized Distributed

500 2813 1040

2000 21147 4418

8000 160878 19540

(b) Number of communication cycles required

to setup schedule

Figure 10: Results of simulation for different deployments

and scheduling approaches.

cles required to set up the network are shown. The

difference between the centralized and distributed ap-

proach can be seen quite clearly. The centralized ap-

proach does not scale very well to larger sizes and

requires large number of communication cycles. The

distributed approach grows almost linearly with the

size of the network, the number of communication

cycles required is a little over 2 times the number of

nodes in the network. The packet size of the messages

is independant of the number of nodes in the network

as has been noted before and contains only position

and transmission information, or total schedule length

for the cluster heads.

MAC SCHEDULING IN LARGE-SCALE UNDERWATER ACOUSTIC NETWORKS

8 CONCLUSIONS

Scheduling algorithms for underwater communica-

tion allows mitigating the effects of the long propaga-

tion delay of the acoustic signal. Scheduling has sig-

niﬁcant beneﬁts in terms of throughput, energy con-

sumption, and reliability.

In this paper we extended the set of simpliﬁed

scheduling constraints of (van Kleunen et al., 2011)

to allow scheduling of multi-hop networks. We have

introduced a centralized and a distributed schedul-

ing techniques for underwater acoustic communica-

tion systems.

The reduced complexity centralized approach has

) computational complexity but introduces sig-

niﬁcant communication overhead.

The distributed approach groups all transmissions

together in clusters from which they originate. Nodes

within a cluster communicate with the cluster-head

only for scheduling their link. Our approach does

not place any restrictions on the communication pat-

terns. It does not restrict communication between

base-station and node and nodes can communicate di-

rectly with other nodes within communication range.

Each cluster-head will calculate a schedule for its

cluster and will forward the total schedule length of its

cluster to a central scheduler. The central scheduler

will schedule the timeslots and assign a timeslot to

each cluster. Compared to the centralized approach,

the distributed approach has a much lower communi-

cation and computational overhead.

We evaluated the communication and computa-

tional complexity and showed that the distributed ap-

proach is much more scalable towards larger net-

works. We also evaluated the schedule lengths of dif-

ferent scheduling approaches. The reduced complex-

ity centralized approach calculates only marginally

less efﬁcient schedules, and is therefore a good re-

placement for the full complexity approach. The

schedule lengths of the distributed approach are on

average 270% of the centralized approach when 3

timeslots are used, and 580% when 7 timeslots are

used. This shows that when the scalability, compu-

tational and communication beneﬁts are irrelevant a

centralized approach is still much preferred.

Future work includes considering the effects of

acoustic signal such as refraction, multipath and prop-

agation speed variability on performance. Other ef-

fects that will be considered are node dynamics, po-

sition estimation errors and time-synchronisation er-

rors.

ACKNOWLEDGEMENTS

This work is supported by the SeaSTAR project

funded by the Dutch Technology Foundation (STW).

REFERENCES

Hsu, C.-C., Lai, K.-F., Chou, C.-F., and Lin, K. C.-J. (2009).

St-mac: Spatial-temporal mac scheduling for under-

water sensor networks. In INFOCOM, pages 1827–

1835. IEEE.

Kredo II, K. and Mohapatra, P. (2010). Distributed schedul-

ing and routing in underwater wireless networks.

Globecom 2010.

Molins, M. (2006). Slotted fama: a mac protocol for un-

derwater acoustic networks. In In IEEE OCEANS06,

Singapore, pages 16–19.

Peleato, B. and Stojanovic, M. (2007). A channel sharing

scheme for underwater cellular networks. In OCEANS

2007 - Europe, pages 1–5.

Stojanovic, M. (2008). Design and capacity analysis of

cellular-type underwater acoustic networks. In IEEE

Journal of Oceanic Engineering, vol.33, No.2, pages

171–181.

Syed, A. A., Ye, W., and Heidemann, J. (2007). T-lohi:

A new class of mac protocols for underwater acoustic

sensor networks. Technical report.

van Kleunen, W., Meratnia, N., and Havinga, P. J. (2011). A

set of simpliﬁed scheduling constraints for underwater

acoustic mac scheduling. AINA 2011.

WINSYS 2011 - International Conference on Wireless Information Networks and Systems