Reliable, Low-delay Communication in Wireless Sensor Networks

(1)

Degree project in

Reliable, Low-delay Communication in Wireless Sensor Networks

MAHESH BOGADI SHANKAR PRASAD

Stockholm, Sweden, Nov 2011 Automatic Control

Master's thesis

(2)

Abstract

Wireless sensor networks consist of tiny computers embedded into an environment which can monitor almost anything - such as light, motion, proximity, temperature, biometrics and chemical substances. Actuators conjoined to sensor networks can be used not only to sense the environment, but also to interact with it. Such a design is used to develop automatic control systems, ex. a production line in a factory. These systems are delay critical and demand high reliability.

Hence wireless sensor networks incorporated into such systems must provide sucient reliability as well as low delay communication. However, wireless sensors possess power-constrained radios. Furthermore, wireless communication is expensive in terms of power consumption. Wireless link conditions are often harsh, unpredictable and vary considerably in both space, and time. Wireless sensor networks are formed by multi-hop wireless meshes. Consequently, the communication in wireless sensor networks lacks the required reliability, and often exhibits long communication delays.

This Master's thesis investigates about the development of a reliable, and low end-to-end delay data collection scheme for wireless sensor networks. The approach is to decrease the number of retransmissions for a packet at the data link layer in order to decrease the end-to-end delay. However, a decrease in the number of retransmissions for a packet leads to lower reliability. In order to compensate for the reliability suered, an erasure coding scheme, and a multi- path routing paradigm are investigated. Accordingly, the thesis focuses on an implementation, and evaluation of an ecient combination of fountain coding, braided multi-path routing, and proportionally fair packet scheduling.

The thesis concludes that fountain coding in combination with braided multi- path routing and proportionally fair packet scheduling is an ecient solution for a wireless sensor network with high loss rates.

(3)

List of Tables

4.1 Experiments for the evaluation . . . 30

(6)

List of Figures

2.1 Network stack . . . 9

2.2 Collection Tree . . . 12

2.5 Neighbor discovery . . . 19

3.1 System design of the nodes . . . 21

3.2 System design of the sink . . . 22

3.3 Contiki Collect . . . 23

3.4 Overview of fountain coding . . . 24

3.5 Encoding and decoding of fountain coding . . . 25

3.6 The series of packets . . . 26

3.7 Fountain coding for two packets . . . 27

4.1 Network topology . . . 31

4.2 Reliability at dierent transmission and reception success rates . 32 4.3 Maximum retransmissions at dierent transmission and reception success rates . . . 33

4.4 Average end-to-end delay at dierent transmission and reception success rates . . . 34 4.5 Power consumption at 60% transmission and reception success rate 35 6.1 Reliability per node 60% transmission and reception success rate 40 6.2 Reliability per node 65% transmission and reception success rate 41 6.3 Reliability per node 70% transmission and reception success rate 41 6.4 Reliability per node 75% transmission and reception success rate 42 6.5 Reliability per node 80% transmission and reception success rate 42 6.6 Reliability per node 85% transmission and reception success rate 43 6.7 Reliability per node 90% transmission and reception success rate 43

(7)

6.8 Reliability per node 95% transmission and reception success rate 44 6.9 Reliability at dierent transmission and reception success rates . 45 6.10 Retransmissions 60% transmission and reception success rate . . 46 6.11 Retransmissions 65% transmission and reception success rate . . 46 6.12 Retransmissions 70% transmission and reception success rate . . 47 6.13 Retransmissions 75% transmission and reception success rate . . 47 6.14 Retransmissions 80% transmission and reception success rate . . 48 6.15 Retransmissions 85% transmission and reception success rate . . 48 6.16 Retransmissions 95% transmission and reception success rate . . 49 6.17 Retransmissions 95% transmission and reception success rate . . 49 6.18 Maximum retransmissions at dierent transmission and reception

success rates . . . 50 6.19 Average end-to-end delay per node at 60% transmission and re-

ception success rate . . . 51 6.20 Average end-to-end delay per node at 65% transmission and re-

ception success rate . . . 58 6.27 Delay distribution 60% transmission and reception success rate . 59 6.28 Delay distribution 65% transmission and reception success rate . 59 6.29 Delay distribution 70% transmission and reception success rate . 60 6.30 Delay distribution 75% transmission and reception success rate . 60 6.31 Delay distribution 80% transmission and reception success rate . 61 6.32 Delay distribution 85% transmission and reception success rate . 61 6.33 Delay distribution 90% transmission and reception success rate . 62 6.34 Delay distribution 95% transmission and reception success rate . 62 6.35 Delay range at 60% transmission and reception success rate . . . 63 6.36 Delay range at 65% transmission and reception success rate . . . 64 6.37 Delay range at 70% transmission and reception success rate . . . 64 6.38 Delay range at 75% transmission and reception success rate . . . 65

(8)

6.39 Delay range at 80% transmission and reception success rate . . . 65 6.40 Delay range at 85% transmission and reception success rate . . . 66 6.41 Delay range at 90% transmission and reception success rate . . . 66 6.42 Delay range at 95% transmission and reception success rate . . . 67 6.43 Average end-to-end delay at dierent transmission and reception

success rates . . . 68 6.44 Power consumption at 60% transmission and reception success rate 69 6.45 Power consumption at 65% transmission and reception success rate 69 6.46 Power consumption at 70% transmission and reception success rate 70 6.47 Power consumption at 75% transmission and reception success rate 70 6.48 Power consumption at 80% transmission and reception success rate 71 6.49 Power consumption at 85% transmission and reception success rate 71 6.50 Power consumption at 90% transmission and reception success rate 72 6.51 Power consumption at 95% transmission and reception success rate 72

(9)

Chapter 1

Introduction

Wireless Sensor Networks (WSN) consist of tiny computers embedded into the environment that can monitor almost anything - such as light, motion, proximity, temperature, biometrics and chemical substances. It has become a vision to deploy sensor networks not only to sense the environment, but also to interact with it, by means of networked actuators conjoined to the networked sensors.

Many such WSNs require denite reliability, and delay guarantees.

In order to provide high reliability and low delay, this thesis mainly explores incorporation of fountain codes and multi-path routing into WSNs. Fountain coding is a type of network coding. It is used to transfer broadcast or multi-cast data reliably[1]. Multi-path routing brings path-redundancy into the network and increases reliability in WSNs. Some of the multi-path routings are proven to be energy ecient, adaptive to topology changes, and disseminates the network load well [2].

Fountain coding is an an ecient erasure coding scheme[3], in which the transmitter disperses coded fragments towards the receiver [1]. Once the receiver has received any N fragments in any order, where N is greater than the original number of data fragments, the whole data can be recovered. Incorpo- ration of this coding scheme increases reliability by sending redundant encoded packets. The computational costs of the fountain codes are small, scaling linearly with the number of data fragments [1].

Multi-path routing constructs multiple paths for the transmission of sensor information in order to increase redundancy. Thus, it increases the probability of successful packet delivery to the sink [2]. In large sensor networks, it also mitigates network congestion. Thus it may reduce the end-to-end delay

(10)

1.1. PROBLEM STATEMENT CHAPTER 1. INTRODUCTION

[2]. It conserves energy, and enables spectrum reuse[2]. In braided multi-path routing, there is a primary path that is mainly used for routing, and several alternate braided paths are maintained [2]. Braided paths may share nodes. In other words, node disjointedness (no node is present in multiple paths) is not a requirement. Braided multi-path routing is an ecient localized enforcement- based scheme for collection tree [2]. Proportionally fair scheduling improves the multi-path routing [4]. It dynamically determines the number of packets to be sent to the parent nodes (primary and secondary) of the multi-path, inversely proportional to the routing metric of the parents. [4].

1.1 Problem Statement

A wireless sensor network comprises of wireless sensors as its nodes. The wireless sensors possess power-constrained radios. They are networked by form- ing multi-hop wireless meshes. Furthermore, wireless communication is expensive in terms of power consumption. Also, wireless link conditions are often harsh, unpredictable and vary considerably in both space, and time. Conse- quently, the communication in wireless sensor networks lacks the required reliability. In addition, it often exhibits long communication delays. This does not bet for applications such as a production line in a factory, which requires reliable, and ecient control. Hence, there is a need for a solution which provides low delay as well as high reliability.

1.2 Method

The method used in the thesis is that of experimentation [5]. I implemented a fountain coding scheme and a braided multi-path with proportionally fair packet scheduling combination as a proposed solution to enhance reliability and reduce end-to-end delay. Then I ran the experiments (with and without the solution) with dierent packet transmission and reception success rates using the Cooja network simulator[6]. I measured reliability (percentage of the total packet loss), end-to-end delay (average time taken by a packet to reach the sink from a source), maximum link-to-link retransmissions (average of maximum data-link layer retransmissions suered by a packet whilst traveling from the source to the sink) and power consumption of each node. Priorly, I had studied wireless sensor network architecture[7], reprogramming wireless sensor networks on Contiki platform, network coding, multi-path routing techniques and Cooja

(11)

1.3. LIMITATIONS CHAPTER 1. INTRODUCTION

simulator [6]. The results from the experiments are plotted as graphs using gnuplot [8]and compared. This comparative analysis is used in the evaluation of the results and is the basis for the conclusion of the thesis.

1.3 Limitations

Energy eciency is one of the prominent considerations in designing a WSN.

But, the thesis does not design the solution with respect to energy eciency.

However, the eorts of the thesis to reduce number of retransmissions at the link layer, and the usage of a braided multi-path routing have been proven to be energy ecient. The thesis does not explore the other considerations of WSNs, such as mobility, multi-sink, etc.

1.4 Scientic Contributions

WSNs are formed by multi-hop wireless meshes. Consequently, the communication in wireless sensor networks lacks the required reliability, and often exhibits long communication delays. The thesis contributes mainly at providing a WSN with denite delay and reliability guarantees. This is the rst known eort about evaluating fountain coding in WSN, with combination of braided multi-path, and proportionally fair packet scheduling. The thesis shows promis- ing results regarding reliability, and low-delay that can be achieved by using fountain coding and multi-path routing in a WSN. In short, the scientic con- tribution of the thesis is a solution towards building a robust WSN with respect to reliability and low-delay.

1.5 Report Structure

The layout of the succeeding parts of this report is as follows. In the next chapter, the background studies are summarized. The design, and implementation details are covered in the consecutive chapter, followed by the chapter on evaluation of the results. The last chapter provides the conclusion of the thesis study, and the future works that can be performed using it as a basis.

(12)

Chapter 2

Background

This chapter introduces the required background to the thesis. The thesis requires background knowledge of WSN architecture, network stack for WSN, collection tree paradigm, reliability factors in WSN, delay factors in WSN, network coding, multi-path routing, packet scheduling, and the Contiki operating system.

2.1 Wireless Sensor Network (WSN)

A WSN consists of tiny sensing computers called sensor nodes[7]. These sensors communicate through radio communication (wireless) to form a network.

WSNs can monitor by sensing almost anything - such as light, motion, proximity, temperature, biometrics and chemical substances. Actuators are devices, such as valves or switches, that perform actions such as turning things on or o

or making adjustments in an operational system. Actuators conjoined to sensor networks can be used not only to sense the environment, but also to interact with it. Such a design is used to develop automatic control systems, ex. a production line in a factory [9]. These systems are delay critical and demand high reliability. Hence wireless sensor networks incorporated into such systems must provide sucient reliability as well as low delay communication.

However, wireless sensors possess power-constrained radios. Furthermore, wireless communication is expensive in terms of power consumption. Wireless link conditions are often harsh, unpredictable and vary considerably in both space and time. Wireless sensor networks are formed by multi-hop wireless meshes. Consequently, the communication in wireless sensor networks lacks the required reliability, and often exhibits long communication delays.

(13)

2.2. NETWORK STACK CHAPTER 2. BACKGROUND

2.2 Network stack

The unpredictable wireless link conditions and resource constrains cause adver- sity in ensuring reliability in WSNs. The wireless communication between the sensor nodes is often prone to errors which can degrade the overall quality of a network. In addition, the multi-hop communication in WSNs amplies the unre- liability of data transmissions to the sink. Ensuring the required low delay along with sucient reliability compounds the problem. Hence, an ecient communication providing low delay and reliability in a WSN depends on the link-level transmissions between sensor nodes [10], and the end-to-end transmissions from sensors to the sink. The sensor to sensor or the node to node communication is ensured and controlled at the data link layer of the network stack, where as, end-to-end communication is facilitated by the network as well as the transport layers [11]. Therefore, the thesis explores the data-link, network, and transport layers of WSNs from the perspective of increasing reliability and lowering delay.

This section briefs the background about dierent layers of the network stack of a WSN.

Figure 2.1: Network stack

(14)

2.3. DATA-LINK LAYER CHAPTER 2. BACKGROUND

2.3 Data-link layer

The data-link layer is responsible for reliability of the packet transfer from one sensor node to another in a multi-hop data-path from the source to the sink in a WSN. Retransmissions between two nodes are used to achieve reliability at the data-link layer. Traditionally, a sensor node in a WSN transmits a data frame to the next neighbor node on the data path, and waits for an acknowledgment for that frame. If the acknowledgement does not arrive within a give timeout period, the node retransmits the frame. The node retries to send the frame to the neighbor node until it receives an acknowledgement for the frame, or the maximum number of allowed retries is reached. However, retransmissions have drawbacks too. In some cases, they increase delay, cause congestion and clog buers of the network nodes. Hence, decreasing the number of retransmissions mitigates delay, congestion and buer usage concerns. It also conserves energy.

The number of retransmissions allowed at the data-link layer may be congured to a lower value, when other reliability measures such as redundant data transmission, error correcting or/and data encoding techniques are incorporated into the system.

2.4 Network layer

The network layer is responsible for determining routes from every sensor node to the sink in a WSN for end-to-end delivery of sensor data. The routes tra- versed by the packets from the sensors to the sink are computed, and established by network layer routing algorithms. Routing in WSNs is very dierent from normal networks, because of the unique qualities and limitations of WSNs. The unpredictable wireless link dynamics causes the topology of WSNs to change continuously and erratically [11]. This necessitates that data ow routes in a WSN to be formed dynamically in order to provide reliable data delivery in WSNs. Hence, reliability should also be guaranteed at this layer in terms of dynamic route determination and management [11].

The routing scheme in a WSN plays a vital role regarding reliability, and delay of the end-to-end data transmission. An ecient routing in a network leads to smart distribution of the trac across the network. However, several ecient routing paradigms such as OSPF, require each node to gather information of the whole network topology, which can lead to high delay in network topology convergence and signicant amount of routing data exchange. Such a routing paradigm is not suited for a WSN. Therefore, a WSN necessitates

(15)

2.4. NETWORK LAYER CHAPTER 2. BACKGROUND

a routing method which can use localized information which avoids delay in topology convergence and routing data exchange expenses[11].

Routing in a WSN involves two basic activities :

Path determination - determining and designating routes from a source to the sink.

Packet switching - transporting packets through designated routes [11].

2.4.1 Routing in WSN

Routing protocols in a WSN use metrics to evaluate the best paths for a packet to travel from a source to the sink. A routing metric is a standard of measure- ment, such as path bandwidth, that is used by routing algorithms to determine the best path to a destination. The metric value of a route indicates the proximity of the node to the sink or the total cost of all links in a path from the node to the sink. Expected transmission count (ETX) metric is a metric for multi-hop wireless networks [12]. Route selection using ETX accounts for link loss ratios, the asymmetry of the loss ratios in the two directions of each link, and the reduc- tion of throughput due to interference among the successive hops of a route [12].

A basic notion for path determination at a node is to nd routes with lowest metric values from the node to the sink. These routes or shortest paths dene a connectivity graph (a tree) describing which nodes can communicate directly over a single hop. However, generally, WSNs are data-gathering networks in- volving a many-to-one communication model. Hence, in WSNs routes trac for the data are not bidirectional, they are unidirectional towards the sink. Thus the graph formed by the routes is a directed tree with the sink as its root. This tree based routing is also known as collection tree.

WSNs are multi-hop wireless meshes. Wireless communication links found in WSNs are prone to signicant loss rates, and the loss rate varies dynamically with environmental factors. Routing protocols must take into account such un- derlying factors. Hence it must encompass connectivity analysis, neighborhood management with limited storage, and routing on dense WSNs with simple, low-power radios.

(16)

Figure 2.2: Collection Tree

For these reasons, the design goals for a routing protocol in WSN include selection of reliable and low-delay routes, simplicity, low operational overhead, robustness and stability in the face of unusual and unforeseen circumstances, rapid convergence or agreement on routes, and exibility in quickly and accu- rately adapting to a wide variety of network circumstances. In addition, WSNs cannot aord to abstract all the details of lower-layer protocols and, thus, motivate cross-layer design and optimizations in routing. These factors motivate dierentiated approaches to routing in WSNs.

2.4.2 Multi-path routing

Reliability of a route in a WSN established by a routing protocol may decrease because of the wireless link conditions, network congestion and energy depletion at the sensor nodes. In such a scenario, an obvious approach to better the reliability is through redundancy. Multiple copies of a packet is sent through a single path, which increases the probability that the sink receives the packet.

However, such a single path approach does not address issues such as node failure, congestion, issues with topology changes and long delays. In certain cases, a single path approach may worsen the network with several negative eects,

(17)

such as increase in congestion, heavy usage of certain paths and fast depletion of energy of the nodes in those paths, and thus resulting in low reliability. In order to address these negative eects, routing solutions based on a multi-path approach are put forth for WSNs [11].

Multi-path routing determines and assigns multiple routes from a given sensor node to the sink. The transmission of data among the multi-path brings path redundancy and increased reliability by amplication of the probability of the sink successfully receiving the data through anyone of the multiple paths.

Multi-path routing decreases network congestion, provides proof against node failures and helps to distribute and share network trac among multiple paths which may avoid heavy usage of certain paths, hence nodes. Multi-path routing is apt when redundant data transmission is used to increase reliability.

The two important metrics in evaluating the performance of a multi-path routing scheme are resilience and maintenance overhead. There is an inherent trade-o between these two quantities. Becoming more resilient typically consumes more energy, and incurs more maintenance overhead. Typically, multi- path routing has been deployed for two reasons: load balancing and robustness.

Load-balancing is essential to conserve energy in sensor networks[2].

There are two types of paths generally constructed in multi-path routing:

A primary path which is the most optimal path with least cost or metric from a node to the sink.

A set of alternate paths. An alternate path is a sub-optimal path with cost or metric slightly higher than the primary path.

A desirable goal in multi-path routing is to deliver data along the primary (best available) path. However, for a scalable recovery from failure of the primary path as well as distributing the load, a small number of alternative paths are constructed. Maintaining the alternate paths must incur low computational overhead and memory. Also it is demonstrated that multi-path routing can be used for energy ecient recovery from failure in wireless sensor networks [2].

The multi-paths classied generally into two types based on the exclusivity of the nodes among the alternate paths :

Disjointed multi-paths - These multi-path do not share any node among dierent paths. Thus the paths are disjointed by nodes of the network.

(18)

In this formation failure of one path does not aect the performance of the other paths. However, it results in increase in the overall latency of the alternate paths. Moreover, in order to establish disjointed paths, each node of the network needs to be aware of the whole topology of network or the information about the other paths, which brings the issues of routing convergence and routing data exchange expenses. Therefore, disjointed multi-paths are computational expensive to construct and maintain without providing signicant benets of reliability and delay [2].

Braided multi-paths - In this case, paths may have overlapping portions and hence can share nodes. As a result, all the paths are not indepen- dent of each other which facilitates usage of mere localized information to construct routing information base, which results in faster convergence and lesser routing data exchange [2]. Generally the alternate paths over- lap the portions of the primary path. Hence, it appears as if the alternate paths are braided around the primary path. Thus, the latency incurred by the alternate paths may be lower than the alternate paths constructed in disjointed multi-paths. Disjoint multi-path and braided multi-path have similar patterned failure resilience. However the braided multi-paths have higher resilience to isolated failures and lesser overhead for maintenance of alternate paths. Also braided multi-path is more energy ecient [2].

(19)

Figure 2.3: Braided multi-path routing

2.4.3 Proportionally fair packet scheduling

This section investigates scheduling techniques appropriate for multi-path routing. A compromise-based solution is required in order to maintain a balance among conicting, and competing interests. For example, in order to increase reliability of end to end data transmission in a WSN, one may think about sending certain packets on non-shortest paths. Similarly, in order to reduce network congestion of routing paths, one may ooad the data trac to other high-cost paths. Proportionally fair packet scheduling is a compromise-based algorithm. It tries to maximize total wireless network throughput at the cost of scheduling packets onto sub-optimal paths. This is done by assigning each data path a scheduling priority (depending on the implementation) that is inversely

(20)

2.5. TRANSPORT LAYER CHAPTER 2. BACKGROUND

Figure 2.4: Proportionally fair packet scheduling

proportional to its anticipated resource consumption [4].

Weighted fair queuing (WFQ) is used to achieve proportionally fair scheduling. The scheduling weights for data paths i are set to wi = 1 / ci, where the cost ci is the amount of consumed resources. In case of WSN, the cost ci can be the routing metric of the path. In other words, lesser number of packets are scheduled onto paths with higher routing metric [13].

2.5 Transport layer

The objective of transport layer protocols is to provide reliability and QoS services for data transfer over inherently unreliable, and resource-constrained WSNs. The following are the interrelated guarantees and services that may be needed in WSNs at this layer [14]:

Reliable delivery guarantee - It should ensure that the data arrive from origin to destination without loss.

Priority delivery - The data generated within the WSN may be of dierent priorities; e.g., the data corresponding to an unusual event detection may have much higher priority than periodic background readings. If the network is congested, it is important to ensure that at least the high priority data get through, even if the low-priority data have to be dropped or suppressed.

(21)

2.5. TRANSPORT LAYER CHAPTER 2. BACKGROUND

Delay guarantee - In time-critical applications, particularly those where the sensor data are used to initiate some form of actuation or response, the data packets generated by sensor sources may have strict requirements for delivery to the destination within a specied time.

Energy-ecient delivery - Energy wastage during times of network congestion must be minimized, for instance by forcing any necessary packet drops to occur as close to the source as possible.

Fairness - Depending on the nature of the application, varying notions of fairness may be relevant. These notions range from ensuring that all nodes in the network provide equal amounts of data (e.g. in a simple data- gathering application), to maxmin fairness, to proportional fairness.

Application-specic, data-centric quality of service - In general, given the bandwidth/energy resource constraints, a data-centric QoS goal requires that the network as a whole provides a picture of the sensed environment as accurate as possible.

The thesis concentrates on providing certain reliable delivery guarantee, delay guarantee, and fairness in a WSN. Therefore, it investigates on fountain coding, and redundant packet transmission.

2.5.1 Fountain codes

Fountain codes are a type of network codes. Fountain codes are sparse-graph erasure codes. Erasure code is a forward error correction code [3]. They are suitable for communications with data loss and unreliable data channels, such as a WSN. Typically in a WSN, the receiver (the sink) is not aware of the pres- ence of the sender. Nor does it send any acknowledgment back to the sender notifying that the data is being received fully as well as correctly. Traditional approach to provide sucient reliability from data loss in WSN is to simply transmit each packet multiple times, anticipating that one of them will reach the receiver. In contrast, fountain codes make encoded packets that are random functions of a set of packets (data units). The transmitter sprays packets at the receiver without concerning about which packets may be successfully received.

Once the receiver has received any N packets, where N is just slightly greater than the original number of packets sent K, all the packets in the set can be recovered. The encoded packets need not be received or sent in any order. The computational costs of the best fountain codes are small and they scale linearly

(22)

2.6. CONTIKI OPERATING SYSTEM CHAPTER 2. BACKGROUND

with the number of packets in the set.

As D. J. C. MacKay describes in his paper [1]:-

The encoder of a fountain code is a metaphorical fountain that pro- duces an endless supply of water drops (encoded packets); let us say the original source le has a size of Kl bits, and each drop contains l encoded bits. Now, anyone who wishes to receive the encoded le holds a bucket under the fountain and collects drops until the number of drops in the bucket is a little larger than K. They can then recover the original le.

Fountain codes are rateless. In rateless codes, the number of encoded packets that can be generated from the plain data is potentially innite. Hence, one can increase or decrease the number of encoded packets sent, depending upon the degree of reliability required and/or the degree of erasure. In other words, if one can calculate how lossy the communication is, then the number of encoded packets to be generated and sprayed at the receiver can be determined.

If the link dynamics of the channel changes constantly and unpredictably, the determination of the number of encoded packets to be sent can be performed dynamically, and adaptively. This exibility makes fountain codes an apt choice for a WSN.

2.6 Contiki operating system

Contiki is an open source multi-tasking operating system[15]. It is designed for memory-ecient networked embedded systems and wireless sensor networks [15]. Contiki has been used is a variety of projects, such as road tunnel re monitoring, intrusion detection, wildlife monitoring, and in surveillance networks [15]. Contiki contains two communication stacks, namely uIP and Rime[15].

2.6.1 Rime

Rime is a lightweight communication stack designed for low-power radios [16].

Protocols or applications running on top of the Rime stack can implement additional protocols that are not in the Rime stack.[17].

(23)

2.6. CONTIKI OPERATING SYSTEM CHAPTER 2. BACKGROUND

2.6.2 Contiki collect

Contiki collect is a collection tree protocol (CTP) [18]-like address-free data collection protocol implemented in the Contiki operation system [15].

2.6.3 Neighbor discovery

I have used the neighbor discovery already present in the Contiki collect implementation. It discovers the neighbors surrounding a node. The neighbor discovery provides the address and the metric from the sink of a neighboring node [15]. Beacon frames are used for neighbor discovery [19].

Figure 2.5: Neighbor discovery

2.6.4 Contiki MAC

Radio Duty Cycling (RDC) and Medium Access Control (MAC) protocols are important parts of the Contiki network stack. They play vital role to determine the power consumption of the nodes. They also determine the behavior of the WSN nodes to handle network congestion [19]. .

(24)

Chapter 3

Design and

Implementation

This Master's thesis aims at development of a reliable, and low end-to-end delay communication scheme for wireless sensor networks based on a collection tree.

The thesis explores options in transport, network and data-link layers of the network stack. The approach is to decrease the number of retransmissions for a packet at the data link layer in order to decrease the end-to-end delay. However, a decrease in the number of retransmissions for a packet leads to lower reliability.

Hence the approach needs to compensate for the reliability suered. Accord- ingly, the thesis focuses on an implementation, and evaluation of an ecient combination of fountain coding, braided multi-path routing, and proportionally fair packet scheduling.

In this chapter, I describe the design and implementation of a fountain coding, braided multi-path routing, and proportionally fair packet scheduling.

3.1 System design

The Contiki operating system [15] is the implementation platform. I made modications, and additions into the Contiki collect code to achieve the solution proposed by the thesis. Diagrammatic representations of the system design for a node and the sink of the WSN are provided subsequently.

Each node of the WSN is a data source. The data generated by the application running on the node is encoded into packets using fountain coding. The

(25)

3.1. SYSTEM DESIGN CHAPTER 3. DESIGN AND IMPLEMENTATION encoded packets are queued into the packet queue. The incoming packets from the neighbors are also queued into the same packet queue. The queued packets are transmitted (sent out) to the neighboring nodes based on the braided multi-path routing and the proportionally fair packet scheduling.

Multi-path routing and proportionally fair packet scheduling decisions are made based on the routing table. The routing table is updated according to the beacons received from the neighbors by the neighbor discovery module. A beacon of a node comprises of the address of the node which is broadcasting it and the metric value of the node to the sink. The neighbor discovery module also broadcasts the beacons informing the neighbors of its own address and the metric value to the sink. The routing tables on each node dene the multi-path collection tree implemented in the thesis.

Figure 3.1: System design of the nodes

(26)

3.1. SYSTEM DESIGN CHAPTER 3. DESIGN AND IMPLEMENTATION

The following gure explains the system design of the sink of the WSN. The sink is the collector of the data from all the nodes in the WSN. The neighbor discovery module of the sink behaves similar to any other node in the WSN.

All incoming encoded packets from dierent nodes are decoded using fountain coding at the sink.

Figure 3.2: System design of the sink

(27)

3.2. CONTIKI COLLECT CHAPTER 3. DESIGN AND IMPLEMENTATION

3.2 Contiki collect

Figure 3.3: Contiki Collect

The Contiki collect is a collection tree implementation on the Contiki platform[15].

It implements neighbor discovery as well as the collection tree building with the sink as the root of the tree. The implementation by default assigns the address 01 to the sink. I have used the Contiki collect and incorporated multi-path routing and proportionally fair packet scheduling into it.

(28)

3.3. FOUNTAIN CODING CHAPTER 3. DESIGN AND IMPLEMENTATION

3.3 Fountain coding

Figure 3.4: Overview of fountain coding

The above gure provides an overview of the fountain coding implemented in the thesis. Three packets in a series are used to generate a set of seven encoded packets at each node of the WSN. The three packets involved are identied by the ag-values 1(0001), 2(0010) and 4(0100) respectively. The three packets are in a series and have consecutive sequence numbers. For example, if packet 1 has sequence number 10, then packet 2 has sequence number 11 and packet 3 has the sequence number 12. When a subset of encoded packets reaches the sink, it is used to decode the plain packets. The below diagram explains both the encoding and decoding.

(29)

Figure 3.5: Encoding and decoding of fountain coding

3.3.1 Encoding

The encoding operation is a simple XOR(⊕) of dierent combination of the three packets (P1, P2 and P3) to generate seven encoded packets. The resulting encoded packets are shown in the gure below. The series involves the three plain packets(P1, P2 and P3), three packets(P1⊕P2, P2⊕P3 and P1⊕P3) resulting from XOR of two packets, and a packet (P1⊕P2⊕P3) resulting from XOR of all the three packets. In this case, fountain coding is applied on a set of three packets.

(30)

Figure 3.6: The series of packets

3.3.2 Decoding

The decoding operation is similar to encoding. It also involves simple XOR of the packets received as shown in the gure above. Out of the seven encoded packets sent, the sink can decode all the three plain packets if it receives at least four of the encoded packets. Hence the WSN can suer up-to 42.86% ( 3 x 100

÷7 ) of encoded packet loss.

3.3.3 Variation

I also tried with another variation of the fountain code. If we know that the WSN suers lesser amount of packet loss, then we can apply fountain coding on only a set of two consecutive packets. The two packets involved are identied by the ag-values 1(001), and 2(010) respectively. This would result in a set of three encoded packets, comprising of the two (P1 and P2) plain packets and a packet (P1⊕P2) resulting from the XOR of the two plain packets. In this case the sink can decode all the two plain packets, even if an encoded packet is lost.

Hence the WSN can suer up-to 33.33% (1 x 100 ÷ 3) packet loss. The below

(31)

3.4. BRAIDED MULTI-PATH

ROUTING CHAPTER 3. DESIGN AND

IMPLEMENTATION

gure describes the variation.

Figure 3.7: Fountain coding for two packets

3.4 Braided multi-path routing

This section details about the implementation of braided multi-path routing in the thesis. The below gure describes the braided multi-path routing. Each node selects two types of routing parent nodes to forward or send the encoded packets. A primary and a secondary parent are the two types of routing parent nodes. The primary parent is the neighboring node with the lowest routing metric value. The secondary parent is a neighboring node which has routing metric value greater than the routing metric value of the primary parent and lesser than the routing metric value of the forwarding node. The dierence of the routing metric values between the primary and the secondary parent is limited to lesser than or equal to eight. If a node could not nd any secondary node, then it would not run braided multi-path routing. Multi-path routing is applied to a packet at each hop along its path from its source till the sink.

(32)

3.5. PROPORTIONALLY FAIR

PACKET SCHEDULING CHAPTER 3. DESIGN AND

IMPLEMENTATION

Figure 3.8: Braided multi-path routing

3.5 Proportionally fair packet scheduling

This section describes about the implementation of proportionally fair packet scheduling in the thesis. The below gure describes the scheduling. This scheduling determines the number of packets to be sent to the primary and the secondary parents. The number of packets send to these parent nodes is proportional to the dierence of their metric values. In this case for every two packets sent to the primary parent, a packet is sent to the secondary parent.

Thus the primary parent receives twice the number of packets compared to the secondary nodes.

(33)

3.6. DESIGN CHOICES

AND ALTERNATIVES CHAPTER 3. DESIGN AND

IMPLEMENTATION

Figure 3.9: Proportionally fair packet scheduling

3.6 Design choices and alternatives

There are two design alternatives to chosen from with respect to braided multi- path routing implementation in the thesis. One alternative is to perform the braided multi-path routing and proportionally fair packet scheduling only at the source of a packet, and the other alternative is to perform the braided multi-path routing and proportionally fair packet scheduling at each hop along the path of a packet from its source till the sink. The latter alternative is chosen since it provides lesser congestion, and more reliability, which is determined through comparing the results by running both the alternatives.

The Contiki MAC is chosen because it is designed for low power consumption [19].

(34)

Chapter 4

Evaluation

4.1 Experiments

This section evaluates the results of the experiments. The methodology of the evaluation is to compare the results from experiments at dierent transmission and reception success rates. The experiments are as shown in the table below.

Number Fountain Braided Proportionally Color Type coding multi-path fair packet of line of line

routing scheduling

1 No No No Green Dotted

2 Yes Yes Yes Red Solid

Table 4.1: Experiments for the evaluation

Experiment 1 is the setup with neither fountain coding, nor redundancy, nor braided multi-path routing. In experiment number 2, reliability of the network is increased by sending three redundant packets using fountain coding. More- over a braided multi-path routing is incorporated into this. All the experiment runs are conducted using the Cooja simulator[6]. Rime stack and the contiki MAC (CSMA) at channel check rate of 8 Hz is used [19].

The experiments are run at the transmission and reception success rates of 60%, 65%, 70%, 75%, 80%, 85%, 90% and 95%. The results are shown as graphs and analyzed. The number of packets in the set of plain packets for fountain coding in all the cases of transmission and reception success rates is three.

(35)

4.2. NETWORK TOPOLOGY CHAPTER 4. EVALUATION

In this section, I only summarizes the results and provide high level graphs.

Further, at the end of the report, I present more detailed graphs in appendix 5.

4.2 Network topology

Figure 4.1: Network topology

The network topology used for the evaluation is a mesh grid of twenty nodes.

Node 1 is the sink node of the WSN.

4.3 Reliability

The summary graph shows how reliability increases in dierent experiments at dierent transmission and reception success rates. The experiment 2 provided 100% reliability in all the test runs, whereas the experiment 1 suered a minimum of 2% total packet loss. Hence I deduce that experiment 2 with fountain coding and braided multi-path has enhanced the reliability of the WSN.

(36)

4.4. RETRANSMISSIONS CHAPTER 4. EVALUATION

0 1 2 3 4 5 6

60 65 70 75 80 85 90 95

Total packet loss per node(%)

Transmission and reception success rates(%) Packet loss versus success rates

With fountain code and multi-path routing Without fountain code and multi-path routing

Figure 4.2: Reliability at dierent transmission and reception success rates

4.4 Retransmissions

The summary graph shows how the number of retransmissions decreased in dif- ferent experiments at dierent transmission and reception success rates. The experiment 2 experienced lower number of retransmissions in all the scenarios except one, but still provided 100% reliability. Hence I infer that one can decrease the maximum number of link-to-link retransmissions to be set when measures like fountain coding (data redundancy) and braided multi-path routing (path redundancy) are incorporated into the WSN.

(37)

4.5. END-TO-END DELAY CHAPTER 4. EVALUATION

0 2 4 6 8 10 12 14 16 18

60 65 70 75 80 85 90 95

Average of maximum number of link-to-link retransmissions

Transmission and reception success rates(%) Retransmissions versus success rates

Figure 4.3: Maximum retransmissions at dierent transmission and reception success rates

4.5 End-to-end delay

The summary graph shows how the end-to-end delay decreased in dierent experiments at dierent transmission and reception success rates. I have also provided graphs showing delay distribution and delay range (maximum and minimum delay) for each transmission and reception rates in the appendix.

After analyzing all the graphs related to the end-to-end delay, I deduce that experiment 2 did not suer considerably high delay compared to experiment 1, even though it has provided 100% reliability in all scenarios.

(38)

4.6. POWER CONSUMPTION CHAPTER 4. EVALUATION

0 2000 4000 6000 8000 10000 12000 14000 16000

60 65 70 75 80 85 90 95

Average end-to-end delay in milliseconds

Transmission and reception success rates(%) End-to-end delay versus success rates

Figure 4.4: Average end-to-end delay at dierent transmission and reception success rates

4.6 Power consumption

From the below graph and the graphs in the appendix related to the power consumption, I infer that the power consumption in the experiment 2 is higher than the experiment 1. As I expected, the power consumption is more in case of the experiment 2. That is because of the redundant packet transmission. For each data packet transmitted from a source in the experiment 1, three additional redundant packets are transmitted from the source in the experiment 2, which causes the observed higher power consumption.

(39)

4.7. SUMMARY CHAPTER 4. EVALUATION

0 0.2 0.4 0.6 0.8 1 1.2 1.4

5 10 15 20

Power in mW

Node number Power consumption per node

Figure 4.5: Power consumption at 60% transmission and reception success rate

4.7 Summary

I summarize the evaluation section, that the experiment 2 performed better than the experiment 1 in all dierent transmission and reception rates. It not only provided higher reliability, but also reduced the number of retransmissions.

Also, it did not show considerably high end-to-end delay per packet than the experiment 1.

(40)

Chapter 5

Conclusions and future work

The thesis concludes that fountain coding in combination with braided multi- path routing, and proportionally fair packet scheduling is an ecient solution for a wireless sensor network with high loss rates. Hence the combination can be used to develop a reliable, and low end-to-end delay data collection scheme for wireless sensor networks. The thesis shows that end-to-end delay for a packet can be decreased by decreasing the number of retransmissions suered at the data link layer. However, a decrease in the number of retransmissions for a packet leads to lower reliability. The thesis shows that a combination of an erasure coding scheme such as fountain coding, and a multi-path routing paradigm can be used, in order to increase the reliability. Proportionally fair packet scheduling facilitates an ecient use of multi-path routing.

However, the solution is not appropriate for a wireless sensor network which have considerably very low loss rates. It also consumes more energy than the traditional approach.

Future works on the topic may involve:

1. Investigation on advanced fountain coding techniques, such as Raptor codes.

2. Investigation on multi-path routing metrics.

3. Investigation on advanced packet scheduling algorithms.

4. Investigation on the dynamic determination of the degree of reliability of fountain coding and multi-path routing.

(41)

Bibliography

[1] D.J.C. MacKay. Fountain codes. volume 152, pages 10621068. IEE Pro- ceedings online no. 20050237, May 2005. [Online; accessed 03-September- 2011]. (document), 2.5.1

[2] Deepak Ganesan, Ramesh Govindan, Scott Shenker, and Deborah Estrin.

Highly-resilient, energy-ecient multipath routing in wireless sensor networks. Mobile Computing and Communications Review, Volume 1, Number 2, October 2001. [Online; accessed 03-September-2011]. (document), 2.4.2 [3] Alexandros G. Dimakis, P. Brighten Godfrey, Yunnan Wu, Martin O. Wain- wright, and Kannan Ramchandran. Network coding for distributed storage systems. March 2008. [Online; accessed 28-October-2011]. (document), 2.5.1

[4] Harold J.Kushner and Philip A.Whiting. Convergence of proportional- fair sharing algorithms under general conditions. IEEE Communications Society and IEEE Signal Processing Society, July 2004. [Online; accessed 03-September-2011]. (document), 2.4.3

[5] R. A. Bailey. Design of Comparative Experiements. Cambridge University Press, April 2008. [Online; accessed 28-October-2011]. (document) [6] Fredrik Österlind, Adam Dunkels, Joakim Eriksson, Niclas Finne, and

Thiemo Voigt. Cross-level sensor network simulation with cooja. In Pro- ceedings of the First IEEE International Workshop on Practical Issues in Building Sensor Network Applications (SenseApp 2006), Tampa, Florida, USA, November 2006. (document), 4.1

[7] Adam Dunkels, Thiemo Voigt, and Juan Alonso. Connecting wireless sensor networks with the internet. ERCIM News, April 2004. (document), 2.1 [8] Gnuplot homepage. http://www.gnuplot.info/. [Online; accessed 28-

October-2011]. (document)

(42)

BIBLIOGRAPHY BIBLIOGRAPHY

[9] F.L. Lewis, D.J.Cook, S.K.Das, and John Wile. Smart Environments:

Technologies, Protocols, and Applications, chapter 'Wireless Sensor Net- works'. Wiley Series on Parallel and Distributed Computing, January 2005.

[Online; accessed 03-September-2011]. 2.1

[10] Jose Araujo, Euhanna Ghadimi, and O. Landsiedel. Random-access medium access control in wireless sensor networks: What's the best choice?

In ADHOC 11: Proceedings of the 10th Scandinavian Workshop on Wire- less Ad-hocNetworks, 2011. 2.2

[11] Nirupama Bulusu and Sanjay Jha. Wireless Sensor Networks: A Systems Perspective. Artech House, Inc., 2005. [Online; accessed 03-September- 2011]. 2.2, 2.4, 2.4.2

[12] Douglas S.J.De Couto, Daniel Aguayo, John Bicket, and Robert Morris. A high-throughput path metric for multi-hop wireless routing. Proceedings of the 9th ACM International Conference on Mobile Computing and Network- ing (MobiCom '03), September 2003. [Online; accessed 03-September-2011].

2.4.1

[13] Lijun Qian, Ning Song, Dhadesugoor R.Vaman, Xiangfang Li, and Zo- ran Gajic. Power control and proportional fair scheduling with minimum rate constraints in clustered multihop td/cdma wireless ad hoc networks.

proceedings of IEEE Wireless Communication and Networking Conference (WCNC), April 2006. [Online; accessed 03-September-2011]. 2.4.3

[14] Bhaskar Krishnamachari. Networking Wireless Sensors, chapter 'Transport Reliability and Congestion Control Networking Wireless Sensor'. Cam- bridge University Press, January 2006. 2.5

[15] Adam Dunkels, Björn Grönvall, and Thiemo Voigt. Contiki - a lightweight and exible operating system for tiny networked sensors. In Proceedings of the First IEEE Workshop on Embedded Networked Sensors (Emnets-I), Tampa, Florida, USA, November 2004. 2.6, 2.6.2, 2.6.3, 3.1, 3.2

[16] Adam Dunkels. Rime - a lightweight layered communication stack for sensor networks. In Proceedings of the European Conference on Wireless Sensor Networks (EWSN), Poster/Demo session, Delft, The Netherlands, January 2007. 2.6.1

[17] Adam Dunkels, Fredrik Österlind, and Zhitao He. An adaptive communication architecture for wireless sensor networks. In Proceedings of the Fifth ACM Conference on Networked Embedded Sensor Systems (SenSys 2007), Sydney, Australia, November 2007. 2.6.1

(43)

BIBLIOGRAPHY BIBLIOGRAPHY

[18] Omprakash Gnawali, Rodrigo Fonseca, Kyle Jamieson, David Moss, and Philip Levis. Collection Tree Protocol. [Online; accessed 03-September- 2011]. 2.6.2

[19] Adam Dunkels, Luca Mottola, Nicolas Tsiftes, Fredrik Österlind, Joakim Eriksson, and Niclas Finne. The Announcement Layer: Beacon Coordina- tion for the Sensornet Stack. In Proceedings of EWSN 2011, Bonn, Ger- many, February 2011. 2.6.3, 2.6.4, 3.6, 4.1

(44)

Chapter 6

Appendix

6.1 Reliability

This section provides the comparison of the reliability among the experiments at dierent transmission and reception success rate.

0 5 10 15 20 25

5 10 15 20

Total packet loss per node %

Node number Reliability versus node

Figure 6.1: Reliability per node 60% transmission and reception success rate

(45)

6.1. RELIABILITY CHAPTER 6. APPENDIX

0 5 10 15 20 25 30

5 10 15 20

0 2 4 6 8 10 12 14 16 18 20

5 10 15 20

(46)

0 2 4 6 8 10

5 10 15 20

0 2 4 6 8 10 12 14 16 18 20

5 10 15 20

(47)

0 5 10 15 20 25

5 10 15 20

0 2 4 6 8 10 12 14 16

5 10 15 20

(48)

0 2 4 6 8 10 12 14 16

5 10 15 20

(49)

6.2. RETRANSMISSIONS CHAPTER 6. APPENDIX

6.1.1 Summary graph about reliability

0 1 2 3 4 5 6

60 65 70 75 80 85 90 95

Total packet loss per node(%)

Transmission and reception success rates(%) Packet loss versus success rates

Figure 6.9: Reliability at dierent transmission and reception success rates

6.2 Retransmissions

This section provides the comparison of the retransmissions among the experiments are dierent transmission and reception success rate.

(50)

0 2 4 6 8 10 12 14 16 18

5 10 15 20

Average of maximum link-to-link retransmissions

Node number

Average of maximum link-to-link retransmissions versus node With fountain code and multi-path routing Without fountain code and multi-path routing

Figure 6.10: Retransmissions 60% transmission and reception success rate

0 2 4 6 8 10 12

5 10 15 20

Node number

(51)

0 2 4 6 8 10

5 10 15 20

Node number

0 1 2 3 4 5 6 7 8

5 10 15 20

Node number

(52)

0 1 2 3 4 5 6 7 8

5 10 15 20

Node number

0 2 4 6 8 10

5 10 15 20

Node number

(53)

0 1 2 3 4 5 6

5 10 15 20

Node number

0 1 2 3 4 5 6

5 10 15 20

Node number

(54)

6.3. END-TO-END DELAY CHAPTER 6. APPENDIX

6.2.1 Summary graph about retransmissions

0 2 4 6 8 10 12 14 16 18

60 65 70 75 80 85 90 95

Average of maximum number of link-to-link retransmissions

Transmission and reception success rates(%) Retransmissions versus success rates

Figure 6.18: Maximum retransmissions at dierent transmission and reception success rates

6.3 End-to-end delay

This section provides the comparison of the average end-to-end delays among the experiments are dierent transmission and reception success rate.

(55)

0 5000 10000 15000 20000 25000 30000

5 10 15 20

Node number Average end-to-end delay per node

Figure 6.19: Average end-to-end delay per node at 60% transmission and reception success rate

(56)

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000

5 10 15 20

(57)

0 2000 4000 6000 8000 10000 12000 14000 16000

5 10 15 20

Reliable, Low-delay Communication in Wireless Sensor Networks

Degree project in