Wireless Sensor Networks, State-of-the-Art
Overview of wireless sensor networks
Recent advances in micro-sensor technology has enabled the development of wireless sen-sor networks (WSN) in a wide range of applications [1, 8, 9, 10]. For example, we can talk about monitoring patients or assisting disabled persons in the health area ; surveillance, targeting, and reconnaissance systems in military ; product quality control in industry ; and wildlife tracking and forest monitoring .
A WSN consists of a large number of sensor nodes that are deployed around or inside an en-vironment [3, 4]. A typical sensor node is a battery-operated device, which is able to establish a relationship between the digital and the real world through a number of sensors. In order to mo-nitor an environment, usually a large number of sensor nodes are required. These nodes exchange information or transmit the collected data to a sink or destination node through wireless links.
The general scheme of a WSN is depicted in Figure 2.1. A number of sensor nodes are distri-buted inside the field, and each node transmits the collected information to the destination node (user) while probably cooperating with other nodes. Sensor nodes usually have a relatively simple function and have a low energy capacity available. To establish a communication link between the sensor field and the user, one or several sink nodes are needed to be deployed, as can be seen in Figure 2.1. The collected information in the sensor field can be transmitted to the sink nodes via the communication links among nodes and according to a routing protocol. Then, sink nodes may retransmit the received information to the user through Internet or by means of a satellite link.
Depending on the mode of deployment, WSNs can be classified into two categories : structu-red and unstructured WSNs . In a structured WSN, sensor nodes are deployed in a prearranged mode and on fixed positions. In this type of WSN, the number of sensor nodes is relatively small and the distribution of nodes needs to be carefully designed. Moreover, each node has usually ex-pensive units for reliable communication [11, 12, 13]. Generally, due to the small number of nodes, the implementation complexity of a structured WSN is relatively low. On the other hand, in an un-structured (also called ad hoc) WSN, a large number of sensor nodes are randomly distributed in the field [14, 15]. Compared with the structured case, an unstructured WSN can easily be deployed in remote geographic areas. Such a random deployment can be realized by dropping the sensor nodes from an airplane, for example. Sometimes, they are deployed in inaccessible areas or on relief for studying the meteorological conditions.
Due to this simple node deployment, unstructured networks have a wider application area than the structured ones. Meanwhile, their management is much more complex. In particular, such networks should have the possibility of automatic reconfigurability in the case of any change in the network topology due to a failure or run-out of the battery of some nodes . In summary, some of the main practical limitations of WSNs are listed below.
– large number of sensor nodes ;
– dense deployment of sensor nodes that are prone to failure ;
– possibility of frequent change of network topology ;
– stringent constraints on the transmit power, computational capacity, and memory of sensor nodes
– absence of global identification of sensor nodes to avoid large amount of overload ;
– need to work in extreme weather conditions.
Usually, each node incorporates a transceiver and there are strict constraints on the weight and the size of each node, as well as on its computation, memory, and energy resources. Special attention should hence be devoted in the design of each node to the signal processing tasks and energy consumption.
Classification of sensor nodes
Depending on the application, sensor nodes can be classified into four categories : submote, mote, supermote, and gateway devices. A submote sensor node is a basic sensing device in its lo-west and simplest form and has a very limited energy capacity. Typically, it just transmits certain information in the case of sensing an event. A mote sensor node has a slightly more complex struc-ture, compared with the submote case. This device is the essential part in multihop communica-tion, which means that, in addition to sensing, the mote node can be used as relay for receiving information from other nodes and retransmitting it to another. However, due to its small structure, its energy capacity is very limited. A supermote, on the other hand, has a much larger capacity for data communication due to employing advanced signal processing equipments and a battery of larger capacity. The most powerful node of a network is the gateway node that is usually used as a sink node (see Figure 2.1). The gateway node firstly receives information from different paths by lower level sensor devices and then retransmits it to the user after some pre-processing.
Generally, except submote devices, the other three kinds of sensor nodes should include the four units of sensing, processing, transceiver, and power. Owing to the recent advances in microe-lectronics, micromechanics, and wireless communications, it is now possible to deploy multifunc-tional, low cost, low power consumption, and small size sensors. These sensor nodes are capable to communicate between them in a cooperative manner and over relatively short distances. A ge-neral configuration of such nodes is shown in Figure 2.2. We briefly describe these four units in the following.
– The sensing unit : This is the functional part by which we establish a relationship between analog and digital worlds. Generally, it consists of two functional modules : sensing and analog-to-digital convertor (ADC). The analog information obtained from the environment is passed to the processing unit after being converted to a digital signal by the ADC.
– The processing unit : The main functions of this unit are data storage and data processing. In order to optimize the node’s power consumption, this unit usually works in one of four modes of off, sleep, idle, and active.
– The transceiver unit : To reduce the implementation complexity, this unit is usually subject to a half-duplex constraint. Moreover, considering the limitations on the node’s size and power consumption, the transceiver unit is usually equipped to one single antenna for both transmitting and receiving. When the node works as a transmitter, the output of processing unit is transmitted to other nodes according to a routing protocol. On the other hand, when it is used as a receiver, the unit stores the received data from other nodes. Again for the purposes of optimizing energy consumption, there are four operation modes for this unit : transmit, receive, idle, and sleep .
– The power unit : This is the most important part of the node that is generally made up of bat-teries of limited capacity. Obviously, the accomplishment of all of above-mentioned func-tions depends on this unit. Note that when nodes are deployed in an inaccessible region, recharging or replacing the battery is impractical. Consequently, energy efficient sensing, processing, and transceiver design are of critical importance in order to extend the network lifetime.
Energy consumption issues
Energy consumption is an important factor in the design of a WSN . In fact, sensor nodes rely on the capacity of their battery, and recharging batteries can be quite difficult, especially in ad hoc networks. Therefore, maximizing the overall energy efficiency of a WSN is a critical challenge.
Depending on the rate of energy expenditure, energy consumption in a network can be clas-sified into two factors : continuous and reporting energy consumption . Continuous energy consumption refers to the minimum energy required for maintaining a network during its lifetime without any data transmission and reception. In general, it incorporates battery leakage and the energy expended in the phases of sleep, sensing, and signal processing. On the other hand, re-porting energy consumption is the amount of energy used for data collection and transmission and depends on channel characteristics and network protocols. To reduce the reporting energy consumption, we should improve the efficiency of transmission and reception.
The network lifetime is defined as the duration from the moment of network distribution to the moment it becomes out of work. Due to the network complexity, numerous possible events can result in the network paralysis. For instance, certain sensor nodes may exhaust their energy, or some nodes may be destroyed because of lightning, torrent, etc. The lifetime of a network is mainly related to the expected wasted energy and the expected reporting energy. The expected wasted energy is that consumed in a nonfunctional network, and the expected reporting energy is that consumed by all nodes in a randomly chosen data collection scenario .
Energy consumption and protocol layers
As Figure 2.3 shows, like in any other communication system, the sensor network protocol stack contains the following five layers : physical, data link, network, transport, and application [1, 8, 11]. The design of each layer and the interaction between them affect the network energy consumption.
The first layer is known as the physical layer, which is used to design an adequate and robust modulation, as well as transmission and reception schemes, while taking power consumption is-sues into account. Notice that the transceiver power consumption also depends on the distance from the source to the destination node. In the case of long distances between these nodes, the physical layer should propose an energy efficient multihop strategy for saving energy.
The data link layer provides functional means to data transmission between nodes, and possi-bly the correction of the errors that may occur in the physical layer. A sublayer of data link layer is the medium access control (MAC) layer that provides addressing and channel access control me-chanisms. Under the conditions of noisy environment and mobile nodes, the MAC layer should minimize the energy waste originating from packet collisions, overhearing, excessive retransmis-sions, control overheads, etc. [8, 11].
The network layer helps to design the data routing provided by the transport layer (see below). To maximize the network power efficiency, sensor nodes should employ an optimal transmission routing protocol . In particular, when there are a large number of nodes located between the source and the destination nodes, there are lots of possible routes can be chosen for communi-cation. Energy efficient routes should then be chosen depending on the available energy of the nodes and the required energy for data transmission over these routes.
The transport layer is responsible for maintaining the flow of data when the WSN application requires it. Generally, a WSN is capable of tolerating a certain degree of packet loss which results from packet collision, node failure, low quality communication link, etc. . In order to preserve the quality of service (QoS), data retransmission should be done for the lost packets. This, in turn, requires more energy expenditure. So, the use of an efficient transport layer protocol is important for energy saving.
Lastly, the application layer concerns the different applications that are set up according to the sensing tasks.
We notice from Figure 2.3 that in addition to the power management plane, the design of the different protocol layers should take into consideration the mobility and task management. In other words, we should reduce the overall network power consumption while satisfying the mobi-lity and task requirements.
Design challenges of WSNs
There remain still many challenges ahead for WSNs, especially concerning the implementa-tion aspects. A WSN should satisfy different design criteria depending on the specific application. However, there are some design challenges that concern most of the WSNs . We briefly intro-duce two examples in the following.
Trade-off between communication and computation
Prior to data transmission, we could perform some pre-processing to reduce the volume of collected data, e.g. by compressing the measurement information. In a larger scale, a so-called cluster node could be in charge of the compression of data collected from different nodes in a cluster, before transferring it to the sink node. This can reduce considerably the energy expended for data transmission. However, some energy is required for performing the pre-processing and data compression. The larger the number of nodes in a cluster is, the more considerable will be the required energy for pre-processing. Therefore, to optimize power consumption, a tradeoff should be considered between data transmission and pre-processing.
Connectivity and coverage in hostile environments
In some applications, the battery of some nodes may end up when they are deployed in cer-tain wireless-unfriendly locations. Also, some nodes could experience temporary or permanent hardware failure when the environmental conditions are changed, e.g. due to torrential rains, fire hazards, etc. In order to eliminate the influence of such failures on the entire WSN function, it is necessary to dimension the number of nodes and their communication ranges, and also to employ an efficient routing protocol to ensure node connectivity and the coverage of the entire region.
Like in most wireless communication systems, a sink node in a WSN receives signals arriving from different propagation paths due to the existing scatterers/reflectors in the environment. The received signals may add up constructively or destructively at the destination, which causes signal fading. Multipath fading can considerably deteriorate the quality of data transmission. Depending on the channel coherence interval and the transmission rate, slow or fast fading conditions may hold. In particular, quasi-static fading conditions hold when the channel coherence time is larger than the frame size.
Cooperative diversity in WSNs
Fading has an important impact on the quality of data transmission and efficient diversity techniques should be employed to reduce its destructive effect. By exploiting some kind of diver-sity, e.g. in time, frequency, space, or polarization, a significant improvement in the system per-formance can be obtained conditioned to the independence of the fading on the different signal copies. Usually, the most efficient solution is to use multiple antennas at the transmitter and/or at the receiver to average over channel fading [17, 18]. However, the classical diversity techniques cannot be employed in a WSN due to the stringent constraints on the nodes’ size and cost. In par-ticular, it is impractical to use multiple antennas at each sensor node. As a result, cooperative or distributed diversity techniques are mostly employed in WSNs where nodes cooperate among each other to exploit some amount of diversity [19, 20, 21, 22]. In other words, by cooperative diversity, a virtual multi-antenna communication link is established between the source and the destination nodes. Several cooperating schemes have been proposed in the literature so far. In , repetition-coding is proposed that has a low implementation complexity at the expense of spectral efficiency. A better spectral efficiency is obtained through the use of channel coding to obtain cooperation diversity as proposed in [19, 24]. Most of the proposed techniques, however, consider the use of some relay nodes to provide cooperative diversity. In these so-called wireless relay networks, the relays cooperate among them in order to provide some distributed spatial diversity, as it is explai-ned in the following section.
Wireless relay networks
A simple form of cooperative diversity networks is a wireless relay network (WRN) where some nodes have the role of relaying the signal transmitted from a source node towards a destination node [25, 26, 27, 28, 29]. In such networks, data transmission usually takes place in a multi-hop manner. When more than one relay node participates in signal transmission, the relay nodes co-operate with each other through the use of some kind of space-time coding in order to benefit from distributed diversity. This is called distributed space-time coding.
Figure 2.4 shows the general scheme of a WRN . It includes four kinds of sensor nodes, i.e., source nodes, cluster nodes, relay (cooperative) nodes, and a sink node. The sensor field is seg-mented into several clusters, where each of them includes one cluster head node and a number of source and relay nodes. Assume that the event happens in the field of the first cluster. At the begin-ning, the source nodes firstly report the information to their cluster head node. Then, the cluster head node transmits the information to the next cluster head node while using the relay nodes to exploit some cooperative diversity. After multiple hops between the clusters, the information is received at the last cluster head node which is able to communicate it to the sink node.
As we will see, a much simpler network is considered in this thesis, comprised of one source node, one destination node, and a number of relay nodes. For such a network, we consider dual-hop data transmission, where in a first hop, data is transmitted from the source to the relays, and then, in a second hop, the received signals are processed at the relays and retransmitted to the destination. We will focus on this simple network configuration in the sequel.
There are three main cooperation strategies regarding the data processing done at the relays in the second hop : amplify-and-forward (AF) [30, 31, 32, 33], decode-and-forward (DF) [20, 21, 34, 35] and demodulate-and-forward (DemAF) [36, 37].
By the AF mode, the relay nodes just amplify the (noisy) received signal from the source node (or a superior level cluster head) and then simply convey it to the destination node (or to a lower level cluster head) [22, 30, 38]. The destination node firstly combines the received signal trans-mitted from the relays and then makes a final decision on the transmitted data. Although noise is amplified at the relays, the destination also receives several independently faded versions of the signal and can make more reliable decisions on them. The advantage of the AF strategy is that the relays have no requirement to the channel state information (CSI). So, this scheme is interesting regarding power consumption and transmission delay considerations.
On the other hand, by the DF mode, the relay nodes firstly decode the received signal and then re-encode and forward it to the destination. This way, the noise can be greatly reduced at the expense of complexity and power consumption at the relays. By the DF scheme, CSI is required for signal detection at the relays, which increases their complexity and energy consumption, as compared with the AF scheme. Also, we may suffer from erroneous data transmission from the relays. It is shown in [39, 40] that the performance of AF outperforms that of DF when a direct communication link between the source and the destination can be supported.
The so-called demodulate-and-forward (DemAF) is the another cooperation strategy which has been proposed in [37, 38, 41]. By DemAF, the received signals at the relays are first demodu-lated, and then remodulated to reconstruct the transmitted symbols before being sent to the des-tination. This way, we can effectively eliminate the influence of noise amplification at the relays. However, this comes at the expense of increased network’s energy consumption.
Throughout this thesis, we consider the amplify-and-forward cooperation strategy at the re-lays.
Table of contents :
1 General Introduction
1.1 Wireless sensor networks :merits and challenges
1.2 Thesis objective
1.3 Thesis overview
1.4 Thesis contributions
1.5 Author’s publications
1.5.1 Journal papers
1.5.2 Conference papers
2 Wireless Sensor Networks, State-of-the-Art
2.1 Overview of wireless sensor networks
2.2 Sensor nodes
2.2.1 Classification of sensor nodes
2.2.2 Sensor configuration
2.3 Energy consumption issues
2.3.1 Network lifetime
2.3.2 Energy consumption and protocol layers
2.4 Design challenges ofWSNs
2.4.1 Trade-off between communication and computation
2.4.2 Connectivity and coverage in hostile environments
2.5 Cooperative communication
2.5.1 Multipath fading
2.5.2 Cooperative diversity inWSNs
2.6 Wireless relay networks
2.6.1 Cooperation strategies
2.6.2 Distributed space-time coding
2.7 Chapter summary
3 Signal Transmission in aWRN Case Study
3.2 Assumptions and system model
3.2.1 Network structure and definitions
3.2.2 Power distribution over network nodes
3.3 Data transmission formulation
3.4 Node cooperation and DSTBC at the relays
3.4.1 Case of two relay nodes
3.4.2 Case of four relay nodes
3.5 Signal detection at the destination
3.5.1 MMSE detection under full CSI
3.5.2 MMSE detection under partial CSI
3.5.3 LLR calculation
3.5.4 ML signal detection
3.6 Channel estimation
3.6.1 LS channel estimation
3.6.2 LMMSE channel estimation
3.7 Performance study through numerical results
3.7.1 Simulation parameters
3.7.2 ML versusMMSE detection under full CSI
3.7.3 Performance comparison under perfect full or partial CSI
3.7.4 Gaussian versus enhanced Gaussian approximation under perfect partial CSI
3.7.5 LS versus LMMSE channel estimation
3.7.6 LS estimation for different numbers of pilot blocks
3.8 Chapter conclusion
4 EM-Based Semi-Blind Channel Estimation
4.2 Classical EM-based channel estimation
4.3 Improving the classical EM-based estimator
4.4 Performance comparison between CB-EMand UL-EMmethods
4.4.1 Simulation parameters
4.4.2 Case of R = 2
4.4.3 Case of R = 4
4.4.4 Estimation error variance
4.5 Chapter conclusion and discussions
5 Iterative Data Detection and Channel Estimation for Non-Orthogonal DSTBC at Relays
5.2 Reformulation of data transmission
5.3 Signal detection and channel estimation at the destination
5.3.1 First iteration
5.3.2 Succeeding iterations
18.104.22.168 Channel estimation
22.214.171.124 PIC detection for non-orthogonal DSTBC
5.4 EXIT chart analysis
5.5 Simulation results
5.5.1 BER performance
5.5.2 EXIT Chart performance
5.6 Chapter conclusion and discussions
6 Improved Detection for AFWRN with Imperfect Channel Estimation
8 TABLE OF CONTENTS
6.2 Data transmission formulation
6.3 Signal detection under perfect partial CSI
6.4 Signal detection under imperfect partial CSI
6.4.1 Mismatched signal detection
6.4.2 Improved signal detection
6.5 Numerical results
6.5.1 BER performance
6.5.2 Convergence analysis using EXIT charts
6.6 Chapter conclusion and discussions
7 Conclusions and Perspective