NETWORK PLANNING TOOL WITH TRAFFIC-ADAPTIVE

PROCESSING FOR WIRELESS SENSOR NETWORKS

José Cecílio and Pedro Furtado

University of Coimbra, Coimbra, Portugal

Keywords: WSN, Network Planning, Measurement, Performance, Reliability.

Abstract: Some applications of Wireless sensor networks (WSNs), especially in industrial sense and react scenarios,

require fairly fast sampling rates. Considering that a few sensors may share a common sink, sharing part of

their path on the way to the sink may result in undesirable message losses and delays that cannot be solved

without modifying data communication rates. Our research focuses on planning a WSN to avoid excess

traffic during sensing and acting to guarantee the minimal delay for critical scenarios. In this paper we

propose an integrated approach to plan, test and reconfigure a network. Initially, our approach gives

guidance for a base-plan for the network. With this first-cut plan we test the performance of the network and

if necessary reconfigure it. The results of tests given are followed by traffic-level adjustments of the system

by several possible techniques: adjustment of number of nodes, network partitions, reduction of the

sampling rate or in-network processing with strategies such as aggregation techniques or in-node closed

control loops. We evaluate experimentally the proposed approach with two different mechanisms of

communication, and different levels of traffic, showing that our planning and reconfiguration allows users

to make the best choices for the application context.

1 INTRODUCTION

When deployed in an industrial setting for

monitoring-and-control applications, latencies and

message losses can become paramount, as some

industrial applications need high sampling rates and

may apply closed loop control. Then the question

arises as whether one can provide guarantees to

urgent message delivery. One way to try to provide

added guarantees is to deploy a WSN network with

real-time specific algorithms that would include at

least completely pre-planned synchronous time-

division mechanisms. There is extensive research

into synchronous time-division protocols for WSN.

Among all the work, protocols based on time

division multiple access (TDMA) attract much

attention, since TDMA inherently avoids collisions,

the primary factors causing message loss. TDMA-

based protocols are very efficient when network link

traffic is kept below a certain level, resulting in

neglectable packet losses, where the whole system

will guarantee delivery for messages. However,

CSMA-based protocols are more flexible for

dynamic networks, where data traffic generated at

each node can change over time.

One issue to deal with is how to keep network

traffic below a certain limit, in order to guarantee

minimal losses and delays. We start by defining a

simple network planning approach assuming a

planned deployment of nodes. Then we provide a

tool for users to test the deployment concerning

metrics such as packet and message losses and

latencies. If the test results in non-conformance to

user-dictated application requirements, there is a

need to modify some parameter(s) and to re-test the

solution until it conforms to requirements. Perhaps it

is absolutely necessary to decrease the sensing rate,

remove nodes from the network (or to create

network partitions), or perhaps the sending rate, if

some extra delay can be tolerated.

The proposed approach consists on creating a

planned network by taking into account a set of

constraints. The approach includes: a Network

Planning algorithm to deal with high-rate sampling;

a Network Status module offering simple network

status tests that provide information to the user, such

as the message loss ratio, delays and the degree of

difficulty in delivering application-level messages; a

module that provides commanding / reconfiguring

the WSN to adjust the system until the network

Cecílio J. and Furtado P..

NETWORK PLANNING TOOL WITH TRAFFIC-ADAPTIVE PROCESSING FOR WIRELESS SENSOR NETWORKS.

DOI: 10.5220/0003815600590064

In Proceedings of the 1st International Conference on Sensor Networks (SENSORNETS-2012), pages 59-64

ISBN: 978-989-8565-01-3

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

status tests are satisfied.

The rest of the paper is organized as follows:

section 2 discusses related work; section 3 presents

the tool for planning, reconfiguring and evaluating

the network performance; section 4 shows the

experimental results obtained for TDMA (schedule-

based) and X-MAC (contention-based) protocols

and section 5 concludes the paper.

2 RELATED WORK

In this section we first review related work on MAC

protocols and its adaptation, planning and

monitorization. One important key issue in WSNs

that influences whether the deployed system will be

able to save battery power or to handle high

sampling rates gracefully is the MAC protocol and

its configurations. WSN MAC protocols can be

classified into two main families or their

combination: Carrier Sense Multiple Access

(CSMA), and Time Division Multiple Access

(TDMA). TDMA protocols will schedule the

activity of the network in a period in which all nodes

will be active. In the idle times between data

gathering sessions, nodes can turn off the radio

interface and lie in a sleep state. Thus, the main and

most important advantage of TDMA is time critical

and low power consumption. There are innumerous

works addressing TDMA protocols. Several

protocols have been designed for quick

broadcast/convergecast, others for generic

communication patterns. The greatest challenges are

the spatial-reuse of the time-slots, interference

avoidance, low-latencies, and energy-efficiency.

SS-TDMA (Kulkarni) is a TDMA protocol

designed for broadcast/convergecast in grid WSNs.

The slot allocation process tries to achieve cascading

slot assignments. Each node receives messages from

the neighbours with their assigned slots.

In RT-Link [Rowe] and PEDAMACS (Ergen)

protocols, time-slot assignment is accomplished in a

centralized way at the gateway node, based on the

global topology in the form of neighbour lists

provided by the WSN nodes.

CSMA protocols may be suited for event-driven

WSN applications with dynamic topologies. Some

protocols such as S-MAC, B-MAC, WiseMAC and

X-MAC are frequently used in WSN. S-MAC (Ye)

defines periodic frame structure divided into two

parts, with nodes being active in the first fraction of

the frame and asleep for the remaining duration. The

length of each of the frame parts is fixed according

to the desired duty-cycle.

B-MAC (Polastre) and WiseMAC (El-Hoiydi)

are based on Low-Power Listening (LPL) (Polastre)

that is a very simple mechanism designed to

minimize the energy spent in idle listening. X-MAC

(Buettner) also is based on Low-Power Listening but

reduces the overhead of receiving long preambles by

using short and strobed preambles allowing

unintended receivers to sleep after receiving only

one short preamble and the intended receiver to

interrupt the long preamble by sending an ACK

packet after receiving only one strobed preamble.

Monitorization tools can be used to evaluate

network performance. It is necessary to have

information on the fraction of packets and

application messages losses, latencies and other

simple metrics that provide enough information

about the network health. It is also necessary to have

alternative in-network processing approaches.

We have reviewed existing tools to monitor

network health (SNMS, SNIF, DiMo), and designed

our own simplified tool adapted to our planning

objectives. Our current tool does not include some

important parts that the reviewed tools do include

and which we plan to add later, such as node failure

detection mechanisms.

To avoid congestion, our work uses in-network

data processing. Data processing in sensor networks

has been studied extensively, and in-network

processing is the general term used for techniques

that process data on a node or group of nodes before

forwarding it to the user.

Our study is related to these ones in that in-

network data processing approaches are used to

decrease the amount of communication that is

needed. But in our work these approaches are part of

an integrated system, with guaranteed delivery of

messages with minimal loss, and monitorization and

configuration to reduce network traffic to acceptable

levels.

3 NETWORK PLANNING

In this section we devise the basic network planning

approach, considering schedule-based with a fixed

topology and simple slot-based planning. We also

discuss why this approach can also be a basis for a

first-cut plan on contention-based protocols,

although more testing is required in that case to

determine whether the system will behave gracefully

with strict sampling rates.

SENSORNETS 2012 - International Conference on Sensor Networks

3.1 Planning for Schedule-based

Protocols

TDMA protocols create a schedule for network

activity: each node is assigned at least one slot in a

time frame, which is considered to be the number of

slots required to get a packet from each source to the

sink. In this work we consider a simple protocol

with a global network frame where all nodes have

the same length in the frame. The time axis is

divided into fixed-length base units called epochs.

Each epoch is subdivided into k.n time slots, where k

is the number of slots required for successful

transmitting a message and n is the number of nodes

in the network. There are various techniques to

determine the value of k depending on application

requirements. We used k=3 slots to provide added

guarantee for delivery of messages. The first slot is

used to send a message, the second is reserved to

receive an acknowledgment from receiver and the

last slot is used to retransmit a message if the node

doesn’t receive an ack. The slot assignment of each

node can be pre-configured, as a planned

deployment is assumed. Each node is identified by a

unique identification number i and can only transmit

any message on the assigned ki slot. Figure 1

displays an example for two nodes (mote 1 and 2) in

a star topology.

Figure 1: Slot usage for Mote 1 and 2.

One important issue on planning a deployment to

use a TDMA protocol is related with the time slot.

The slot size should be as small as possible to reduce

the epoch size and consequently the end-to-end

delay.

To determine the slot time, we take into account

the following times:

 Time to transfer a message from the MAC layers

data FIFO buffer to buffer of radio transceiver (t

);

 Time to transmit a message (t

);

 Time a receiver needs to process the message

and initiate the transmission of an acknowledgment

message (t

);

 Time to transmit an acknowledgment (t

);

 Time to transfer and process the

acknowledgment from the radio transceiver and to

perform the associated actions for received/missed

acknowledgment (t

Also, a small guardian time is required at the

beginning and end of each slot to compensate for

clock drifts between nodes (t

). Thus, the minimum

size of a transmission slot is given as:





T  ),(),(max

(1)

The information content of messages and sampling

rate should be planned as soon as epoch and slot

assignments are known.

The epoch size (including a built-in inactivity

period) defines the minimum delay of messages, and

a maximum packet size defines the maximum

information length of a single message.

Based on a set of network constraints that can be

defined by the user, it is possible to determine

adequate values for parameters such as network size

or minimum delay, depending on which variable is

to be determined.

Minimum Sending Period: The minimum

sending period is the minimum time between two

messages are sent by a node and it depends on the

network size (n), the slot time (

time

slot

) and the

number of slots reserved for each node (k):





Node

Period

sending min

(2)

Where

Node

represents the time needed by each

node and it is given by:

timeNode

slotkt 

The maximum sending rate is the inverse of this

value. Likewise, the maximum network size is the

maximum size of the network measured in number

of nodes. Given a sending rate, this value is

determined as a function of the minimum sending

period:

kslot

iodsendingPer

time

Max





(3)

The maximum delay measures is the time taken

from the moment when a sample is collected to the

moment when it is delivered to the client. This

measure includes the time it takes for a sample to

wait for the next sending instant.



ceiveTimeonTimeTransmissi

Size

Epoch

Delay

max





(4)

Where epoch size is the time needed to do complete

a cycle in schedule time line. It is defined by:

TimeinactivityslotknEpoch

TimeSize



(5)

NETWORK PLANNING TOOL WITH TRAFFIC-ADAPTIVE PROCESSING FOR WIRELESS SENSOR NETWORKS

We can also estimate the closed-loop maximum

delay. The closed-loop maximum delay measures

the amount of time from the instant when a sample

is collected to the moment an actuation is enacted

that already carries a response to the sampled value.

The closed loop control can be applied on motes or

in client workstation, depending where it is applied,

we can define the maximum delay as:





Client

Closedloop

nTimeComputatioDelayw

nTimeComputatio







max2

max

(6)

Where w indicates the place where closed loop

control is applied. This variable is a binary variable.

If w=0, closed loop control is applied on mote,

otherwise closed loop control is applied on client

workstation.

Other parameter that can be estimated is the

radio duty cycle of each node. It is measured as:

Size

Time

Dutycyle

Epoch

Slotk

Radio



[%]

(7)

The above planning is a simple way to plan a flat

wireless sensor network. If we are interested to plan

a tree network with several levels each epoch should

be subdivided into





















Levels lLevel

Nodes

lidescendent

knk

(8)

time slots. Where:



Levels

is the number of levels used by the

topology.



jLevel

Nodes



is the number of nodes that exists in

level j.



jidescendent

is the number of nodes below level j

connected to node i.



is the number of slots required by each node.

In this case, the minimum sending period is given

by:































Levels lLevel

Nodes

Node

lidescendent

Node

Size

tnt

EpochiodsendingPer

minmin

(9)

3.2 Planning for Contention-based

Protocols

While planning for schedule-based protocols can be

based on deterministic formulas, contention-based

protocols have to deal with probabilistic collisions

and other network factors. In this work we consider

the X-MAC as an example of contention-based

protocol, and apply a planned topology similar to the

one present in the previous section. X-MAC uses a

strobe preamble that consists of a sequence of short

preambles prior to DATA transmission, as illustrated

in figure 3.

These short preambles indicate that a node has

data to transmit. Assuming a star topology and that

the sink is always active, when a receiver receives a

strobe, it sends an acknowledgment to the sender

which indicates that the transmission can start.

When the sender receives this ack, it stops to send

strobes and starts transmitting the message

immediately.

Figure 2: Operation of X-MAC.

After transmission concludes, the sender goes to

sleep until it has new messages to send or until the

next listen period. This listen period is used to check

if another node (the sink) wants to send messages to

the node. In our prototype we limit the X-MAC

period to the same slot time used on schedule-base

protocol. Since the sink is always on, we assume that

when a node wishes to communicate to send values,

it only needs to send a small strobe and gets an ack

back immediately from the sink (single hop and sink

always with power). As a first-cut approximation,

our planning assumes no collisions (best-case

scenario). With this scenario, exactly the same slot-

based logic used in schedule-based planning is

assumed. We assume that each node will transmit in

a slot, sending a small probe, getting ack, sending

the data packet, getting an ack and possible sending

retry. This planning will result in the following

timings:

 Time to transmit a strobe (

);

 Time to receive and process a strobe (

);

 Time to transmit an acknowledgment (

);

 Time to transmit a message (

);

SENSORNETS 2012 - International Conference on Sensor Networks

 Time a receiver to process the message and

initiate the transmission of an acknowledgment

message (

);

 Time to transmit an acknowledgment (

xam

);

 Time to perform the associated actions for

received/missed acknowledgment (

Thus, the minimum time per node is given as:

paxampmxmxarsxsnode

tttttttT 





(10)

The above plan is assuming no collisions, which is

an acceptable assumption when there is little

congestion in the medium. If the collision

probability is significant, due to high sampling rates,

the planned slot time per node should be multiplied

by a factor α, where α is a percentage of slot time

increase that accommodates backoff periods due to

collisions. For instance, if each node takes 21ms,

this is multiplied by 1.04 to assume an average 4%

increase in total slot time.

To determine the percentage of increase we

consider the probability of any node transmitting in

the same slot time as another node. We use the

following expression to determine that:



 1

(11)

Where t is the total time reserved for each node, n is

the number of node and T is a Period (sending

period).

For example, if each node occupies 10% of the

period, and there are 10 nodes, the overhead increase

will be 9%. So we assume a basic slot-like plan and

we add 9% to the period to better account for

collisions.

For contention-based protocols also it is possible

to determine adequate values for parameters such as:

 The maximum delay



ceiveTimeonTimeTransmissi

backoffStrobeDelay

TimeTime

max





(12)

Where

Time

Strobe

is the time needed to receive an

ack from receiver and

Time

backoff

is the time needed

to start the transmission of the message.

 The closed-loop maximum delay can also be

estimated by (6) where Delay is given by expression

(12).

3.3 Assessing and Reconfiguration

Network planning determines a fist-cut layout of the

network. After network planning and tests,

reconfiguration may be necessary to improve the

performance of the system. The user can change

configuration parameters until the desired

characteristics are obtained. In order to do this, the

reconfiguration requires a set of configurations that

should be used in the successive test procedure, until

the desired characteristics are obtained.

The reconfiguration module collects information

provided by network status, considering metrics

such as message loss ratio, delays, number of fail

ack and battery consumption. If any metrics fail to

provide desired guarantees it is necessary to

reconfigure the system. The reconfiguration module

allows changing the number of nodes and network

partitions, decouple the sending rate from the

sampling rate and join few reading into same packet

or perform aggregations (Summarize information).

4 EXPERIMENTAL

EVALUATION

In this section we report the evaluation of our

approach using schedule and contention based

protocols. The objective is threefold: to test network

planning, to characterize and compare alternative

protocols and configurations under different traffic

conditions; to show that excessive traffic intensity is

promptly characterized by the tests and test

reconfiguration works.

In our approaches we are interesting to give a

tool for planning a network to lead with high rate

and give guarantees of alarm detection. With the tool

we can plan a network with schedule-based and

contention-based protocols.

In tests we used a slot time equal to 7ms,

corresponding to 21 ms of the period reserved to a

single node. Assuming the sampling rate and

sending rate are the same, if a node needs 21ms, and

we deploy a network with 32 nodes, we conclude

through eq.(3) that the minimal guaranteed delay is

672 ms and the minimal period is equal to that too.

In figure 3 we show the message loss ratio

obtained during the tests of our network planning

tool. The results were obtained for contention-based

and schedule-based protocols with different

sampling rates.

Message loss ratio is one of our key measures

that is used as an indicator of excess traffic

conditions. In figure 3 we observe high message loss

ratio for high traffic rate (sampling rate below 800

ms). During the plan, our tool gives 672ms as

minimum sampling rate. When we can verify

through the figure 3, our planning is correct, because

NETWORK PLANNING TOOL WITH TRAFFIC-ADAPTIVE PROCESSING FOR WIRELESS SENSOR NETWORKS

if we used sampling rates less than minimum, many

messages are lost. Therefore if we consider a

sampling rate greeter than 672ms, who corresponds

to a one reading per slot time period, all messages

are successful delivery.

Figure 3: Message loss ratio with X-MAC and TDMA

protocols.

When we use a contention-based protocol (X-

MAC), the minimum sampling rate obtained to

schedule-based protocol is insufficient to guarantee

minimal losses. In our planning tool we added a

tolerant factor (more 25% of needed time) to prevent

collisions. In figure 3 we can also verify that is

factor is important when we lead with high rate. For

low sampling rates, X-MAC protocol have little

losses with good performance, but TDMA is better

with losses equal zero.

Based on these results and in the planning if a

user wants to increase the sampling rate to 500ms,

he has to create partitions on the network or reduce

the number of nodes using a single network only.

Applying the planning tool is obtained for a

schedule-based protocol 18 motes as limit. Who is

needed a tolerant factor for contention-based

protocol the limit is less.

Figure 4: Evaluation of message loss ratio versus number

of nodes.

Figure 4 shows the influence of adding nodes in

the network performance. The test was run with 500

ms of sampling rate. The test starts with a single

mote and after 15 minutes we add more one node

and store the influence of then in the message loss

ratio.

The results show an increase in the message loss

with the increase in the number of active motes.

When the number of nodes is less than 18 (number

given by eq.(4)), the message loss is zero for

schedule-based protocols, but for contention-based

protocols is closed zero until get 16 motes. If the

number of nodes is greater the message loss increase

significantly. When the number of nodes is greater

than 22, the message loss becomes critical (higher

than 5%).

5 CONCLUSIONS

In order to make WSNs more reliable in practical

contexts with constraints such as high sense and

control rates, there is a need for approaches to help a

user to correctly plan a WSN. Sensor networks tend

to have congestion problems at high sampling rates,

which may endanger the timely delivery of

messages. In this paper we have proposed a tool to

plan and test a WSN. Our approach allows to

configure the network traffic (taking into

consideration application requirements such as

sampling rate, maximum delivery delay and whether

the data should be detailed or can be summarized)

and measure their performance.

We proposed a module to plan, configure and

reconfigure the network. Our experimental results

study the traffic characteristics of the approaches

under different conditions, and to conclude on their

usefulness.

REFERENCES

Kulkarni, S. S. and Arumugam, M., “SS-TDMA: A Self-

Stabilizing MAC for Sensor Networks”, Sensor

Network Operations, IEEE Press (2005).

Rowe, A., Mangharam, R., Rajkumar, R. Rajkumar, “RT-

Link: A Global Time-Synchronized Link Protocol for

Sensor Networks”, Elsevier Ad hoc Networks, Special

Issue on Energy Efficient Design in Wireless Ad Hoc

and Sensor Networks, 2007.

Ergen, S. C. and Varaiya, P., "PEDAMACS: Power

Efficient and Delay Aware Medium Access Protocol

for Sensor Networks," IEEE Trans. Mobile Comput.,

vol. 6, July 2006.

Polastre, J., Hill, J., Culler, D., “Versatile low power

media access for Wireless Sensor Networks”, in ACM

SenSys, 2004.

Buettner, M., Yee, G. V., Anderson, E. and Han, R., “X-

mac: A short preamble mac protocol for duty-cycled

wireless sensor networks”, in ACM Sensys, 2006.

SENSORNETS 2012 - International Conference on Sensor Networks