Broadcasting, Data Aggregation, and Opportunistic Forwarding

Abstract

A large number of SNs are randomly deployed in a WSN, and the question that comes is how each SN route data to BS? In order to conserve energy, usually multi-hop solution is adopted.

1 Introduction

A large number of SNs are randomly deployed in a WSN, and the question that comes is how each SN route data to BS? In order to conserve energy, usually multi-hop solution is adopted. But, how is the path determined in a WSN? There is no other choice but to broadcast request for route to BS. Broadcasting is a common operation in many applications , e.g., graph-related and distributed computing problems, and is also widely used to resolve many network layer problems.

2 Broadcasting

The broadcast in a WSN is spontaneous as little or no local information about neighboring nodes or location of BS may be collected in advance. The broadcast is frequently unreliable as you shoot in the dark, and a broadcast message is distributed to as many neighboring SNs as possible without putting too much effort. A SN may miss a broadcast message because it is off-line, or it is temporarily isolated from the network, or it experiences repetitive collisions. Broadcast acknowledgements may cause serious medium contention. In many applications, 100% reliable broadcast is unnecessary. Moreover, SN can easily detect duplicate broadcast messages. If flooding is used blindly, many redundant messages will be sent and serious contention/collision will be incurred in a WSN. When a SN decides to rebroadcast, all its neighbors may already have the message and this is called redundant rebroadcasts . Transmissions from neighbors may severely contend with each other, and due to the absence of any collision detection mechanism, collisions are more likely to occur and cause more damage. For example, in WSN of Fig. 11.1a, transmission and blind retransmission to look for BS can be defined as follows: Step 1, SN 1; Step 2, SNs 2, 6, 7, and 9; Step 3, SNs 4, 3, 8, and 10; and Step 4, SN 5. This makes many SNs to receive multiple copies of the query, and if controlling is done, this can be done only in 4 rebroadcast steps: Step 1, SN 1; Step 2, SN 2, 6, and 9; and Step 3, SN 3, and is shown in Fig. 11.1b.

Fig. 11.1

a Blind rebroadcast in a WSN with each SN doing this and b controlled rebroadcast in a WSN with only 4 rebroadcasts

Step for rebroadcast in a 13-node WSN is illustrated in Fig. 11.2a. It may be noted that there could be many different collisions present as two close-by SNs rebroadcast to a single neighbor. Many possible collisions are also shown. Figure 11.2b, c shows optimal rebroadcast schemes in two different WSNs. Assume that the total area covered by the radio signal transmitted by a transceiver is a circle of radius r _c. When two SNs are deployed close to each other, then an intersection area of two circles of radio transmission range of r _c whose centers are d apart is defined as INTC(d) (Fig. 11.3a). An additional coverage provided by a SN that rebroadcasts the packet is equal to πr ²_c  − INTC(d). When d = r _c, the additional coverage is approximately 0.61πr ²_c and is the maximum value as shown in Fig. 11.3b. On the other hand, the average additional coverage by rebroadcasting from randomly located neighboring SN is 0.41πr ²_c . An expected additional coverage provided by a host’s rebroadcast after the same broadcast packet received by host k times is denoted by EAC(k) and is given in Fig. 11.3c. Every attempt should be made to minimize the number of retransmissions of a broadcast message, and attempt is made to deliver a broadcast packet to each and every SN in the WSN. Jitter allows one neighbor SN to acquire the channel first, while other neighbor SNs detect that the channel is busy. A random delay time (RDT) allows a SN to keep track of redundant packets received over a short time interval. By keeping track, SN rebroadcasts a given packet no more than one time by caching original source SN ID of the packet and the packet ID.

Fig. 11.2

a Blind rebroadcast in a WSN and b, c optimal broadcast steps in 2 WSNs

Fig. 11.3

a Coverage by second SN in a WSN, b maximum possible coverage, and c additional area covered as a function of number of transmissions

Broadcasting can be categorized as either simple flooding or probability-based methods. In simple flooding, a source SN broadcasting a packet to all neighbors, the neighbors, upon receiving the broadcast packet, rebroadcast the packet exactly once. Probability-based methods are similar to ordinary flooding except SNs only rebroadcast with a predetermined probability. In dense WSNs, multiple SNs share similar transmission coverage area. In a Counter-Based Scheme, a relationship between the numbers of times a packet is received by the SN and the probability of a SN’s transmission to cover additional area on a rebroadcast. One special case is dominating set of SN group is defined such that all remaining SNs are at the most 1-hop away from this group and is called dominating set (Fig. 11.4). Thus, if all members receive at least one copy of the message, then rebroadcasting by dominating set will cover all SNs of the whole WSN. This is true as all remaining SNs are one hop away from a SN of the dominating set . A dominating set of a graph G = (V, E) is a vertex subset S ⊆ V, such that every vertex v ∈ V is either in S or adjacent to a vertex of S and is applicable to any WSN, both randomly deployed and regular WSN. Now, the next question is how the SNs of a dominating set get a copy of the query about route to BS. A simple scheme is to determine a connected dominating set (CDS) of a graph G whose induced graph is connected, and this enables any SN to be within communication range of another member SN so that each member gets a copy of query for broadcast to rest of SNs in WSN (Fig. 11.5).

Fig. 11.4

a Dominating set group of a WSN and b another example

Fig. 11.5

a Dominating set group of SNs in a WSN and b connected dominating set in a WSN

Achieving broadcast in a large WSN is quite involved as shown in Fig. 11.6. A simple observation is that some SNs are highly connected than others, and similarly, some SNs are sparsely connected than others, while some SNs are in between (Fig. 11.7). Determining dominating set is a complex process and needs to have knowledge of global connectivity . A simple scheme has been devised in [1] where local knowledge of 2-hop neighboring SNs is analyzed to determine connectivity of each SN with respect to adjacent SNs (Table 11.1).

Fig. 11.6

Dominating set of a large WSN

Fig. 11.7

Dominating set and SN connectivity in a WSN

Table 11.1 Connectivity of SNs in a WSN

If A be the area of the WSN, N be the number of SNs, and r _c be the communication range of each mobile node with a being the fraction of the area covered and a being the fraction of area covered,

$$ \alpha = \frac{{\pi r_{\text{c}}^{2} }}{A} $$

(11.1)

then, the average number of neighboring nodes N _neighbors can be given by:

$$ N_{\text{neighbors}} = (N - 1)\alpha . $$

(11.2)

The probability p _i that a mobile node has i neighbors can be given by:

$$ p_{i} = \left( {\begin{array}{*{20}c} {N - 1} \\ i \\ \end{array} } \right)\left( {1 - \alpha } \right)^{N - 1 - i} \alpha^{i} . $$

(11.3)

The probability p ₁, p ₂, p ₃, and p ₄ can be given by:

$$ P_{1} = \sum\limits_{i = 1}^{N - 1} {p_{i} } \left( {\begin{array}{*{20}c} {\begin{array}{*{20}c} {N - 1} \\ \end{array} } \\ i \\ \end{array} } \right)\left( {\sum\limits_{j = 1}^{i - 1} {p_{j} } } \right)^{i} \left( {1 - \sum\limits_{j = 1}^{i - 1} {p_{j} } } \right)^{N - 1 - i} \quad {\text{Largest}} . $$

(11.4)

$$ P_{2} = \sum\limits_{i = 1}^{N - 1} {p_{i} } \left( {\sum\limits_{k = 1}^{i/2} {\left( {\left( {\begin{array}{*{20}c} i \\ k \\ \end{array} } \right)\left( {\sum\limits_{j = 1}^{N - 1} {p_{j} } } \right)^{k} \left( {\sum\limits_{j = 1}^{i - 1} {p_{j} } } \right)^{i - k} } \right)} } \right)\quad {\text{in}}\;{\text{between}} . $$

(11.5)

$$ P_{3} = \sum\limits_{i = 1}^{N - 1} {p_{i} } \left( {\sum\limits_{k = i/2}^{i} {\left( {\left( {\begin{array}{*{20}c} i \\ k \\ \end{array} } \right)\left( {\sum\limits_{j = 1}^{N - 1} {p_{j} } } \right)^{k} \left( {\sum\limits_{j = 1}^{i - 1} {p_{j} } } \right)^{i - k} } \right)} } \right)\quad {\text{in}}\;{\text{between}} . $$

(11.6)

$$ P_{4} = \sum\limits_{i = 1}^{N - 1} {p_{i} } \left( {\begin{array}{*{20}c} {N - 1} \\ i \\ \end{array} } \right)\left( {\sum\limits_{j = i + 1}^{N - 1} {p_{j} } } \right)^{k} \left( {1 - \sum\limits_{j = i + 1}^{N - 1} {p_{j} } } \right)^{N - 1 - i} \quad {\text{smallest}} . $$

(11.7)

Relations 11.4–11.7 indicate probability that different SNs will rebroadcast the query for locating BS and are mapped in Fig. 11.8. The total number of forwarding steps is given by:

$$ N_{F} = N\left( {\tau_{1} P_{1} + \tau_{2} P_{2} + \tau_{3} P_{3} + \tau_{4} P_{4} } \right), $$

(11.8)

where t _i is the forwarding probability for the SN in group i. Even though such a scheme does not provide 100% coverage of all SNs in a WSN, a good coverage and excellent saving are achieved under mobility and about 20% higher goodput is obtained than the conventional AODV protocol [2]. Lightweight and efficient network-wide broadcast relies on two-hop neighbor knowledge obtained from hello packets (beacon signals ); however, instead of a SN explicitly choosing other SNs to rebroadcast, the decision is implicit due to connectivity property.

Fig. 11.8

Forwarding probability of SNs

3 Lifetime of a WSN

Data is transmitted by SNs to BS in a multi-hop fashion in response to query issued by the BS to all SNs. The types of queries include one-time queries, persistent queries, and historical queries and need to be routed to all SNs from the BS. There is no well-accepted definition of lifetime of a WSN. Some researcher consider when one of the SNs runs out of energy is considered lifetime. A better definition is if 95% of monitored area is covered.

In responding to a query, a shortest path is first determined using broadcasting of needed path from a SN to BS. Such a route is used in forwarding data, and this increases disparity in energy levels of SNs, limiting the life of the network. So, enough energy is consumed by SNs involved in receiving data and forwarding toward BS by retransmission and could run out of energy at a faster rate than remaining part of the WSN (Fig. 11.9).

Fig. 11.9

Forwarding along shortest path from a SN to the BS

Better utilization of network resources is feasible if diverse service requirements for different type of data can be taken into account in selecting the route. In a real application, only few real-time data are generated that create alarms or warning, and most of the time, data for periodic monitoring are generated. So, non-critical data can be forwarded from SNs to BS along longer paths, and shorter paths are reserved only for time-critical data. This is illustrated for 2-D mesh-connected SNs in Fig. 11.10, and a quick situational awareness is needed for timely decision making. SNs are yet to become inexpensive to be deployed with any redundancy. A denser deployment of sensor and transmission of sensed data may cause more energy consumption and increased delay due to collisions. However, transmitting data between two far apart SNs may cause increased energy consumption. However, data reduction by aggregation plays a vital role in minimizing the energy consumption .

Fig. 11.10

a Multi-path routing balances energy consumption in a mesh-connected WSN and b multi-path routing balances energy consumption in a randomly deployed WSN

Data aggregation is needed as individual SN readings are of limited use while delivering large amount of data from all SNs to a central BS consumes lot of energy. Some of the operations that are adopted for aggregation are average, sum, min–max, count, and variance. Such a scheme provides approximate contours of SNs’ residual energy assuming Gaussian distribution, and the boundary line between sensing and no-sensing is approximated. This approach conserves limited energy and bandwidth and increases system lifetime . Data aggregation scheme is expected to process data as it flows from SNs to BS (sink node), and there is a trade-off between energy efficiency and data quality. An initial tree can be established using sink as the root and SNs as nodes. Nodes aggregate data from their children and forward result to BS as data are generated periodically (Fig. 11.11a). A tree is established, and BS broadcasts request for data. Nodes send reply to parent, and it is important to discover how many children are there at each level and establish reverse paths to BS.

Fig. 11.11

a Tree structure with the BS as the root and SNs as sources and b tree structure in a WSN with CHs

The nodes set time-outs based on position in the tree, and data wave reaches BS in one period. Optimal data aggregation is known to be NP-hard. In a practical WSN, instead of tree-based approach (e.g., shortest path), it is advantageous to use cluster-based approach as shown in Fig. 11.11b. There is trade-offs between data accuracy versus lossy or loss-free versus delay in computation. In LEACH [3], the role of CH keeps on changing to preserve energy and the network lifetime can be increased further if only one CH sends packet directly to BS. The role of CH can be changed randomly per round based on residual energy . In a distributed kernel regression, even though any node has all kernel regression coefficients, it cannot answer queries involving an entire region. The message size is variable vector corresponding to a square region and BS may need data values for the entire network. Another approach is rather than each SN forwarding data along selected path to BS in multi-hop fashion, and BS moves around randomly and distributes the query when it is one hop away from a SN. We consider an equivalent reverse problem of a mobile BS or a relay node (RN) pick up data from a SN just one hop away, and we call this random mobility of a BS or RS as shown in Fig. 11.12.

Fig. 11.12

Mobile BS or RN collecting data from SN when one hop away

If total m SNs are used in the network of diameter L, BS with communication range l _i , each BS can collect info from N _A SNs as given in Fig. 11.13a. Let N _D be the number of data-centric sensors within area of diameter L, and the question is how many such transmissions are required? For fixed m and L, as l _i is reduced = min(l _i ), then asymptotically,

$$ \mathop {\lim }\limits_{{I_{i} \to \infty }} \frac{{N_{D} }}{{N_{A} }} = \frac{1}{m}. $$

(11.9)

If m and L are fixed and l _i is increased, BS can aggregate their information and send just a single copy, thereby achieving m-fold saving.

Fig. 11.13

a Aggregation with clustering in a WSN and b TREG-based aggregation in a WSN

PEDAP (Power-efficient Data gathering and Aggregation Protocol) [4] employs MST ( minimum spanning tree )-based routing using energy as the metric. It optimizes energy only locally, and end-to-end latency is increased. TREG Scheme for data aggregation shown in Fig. 11.13b [5] assumes that NT nodes report their data to the tree node closest to them. Measured attribute varies smoothly in a continuous manner with respect to space. Values (z) stored at each tree node can be considered as function values having x − y inputs and values are stored as tuples $\left( {z_{m} ,x_{m} ,y_{m} } \right)$, where z _m is the attribute value sensed by a node at location (x _m , y _m ) (Fig. 11.13b). A polynomial equation is generated through function approximation with three input variables (z, x, y) for all the data points in one particular node of the query tree. The goal is to compress redundant data into few β coefficients and transmitted as follows:

$$ p(x,y) = \beta_{0} + \beta_{1} y + \beta_{2} y^{2} + \beta_{3} x + \beta_{4} xy + \beta_{5} xy^{2} + \beta_{6} x^{2} + \beta_{7} x^{2} y + \beta_{8} x^{2} y^{2} . $$

(11.10)

Parent node regenerates values of attribute by using coefficients obtained from their children, and values are recomputed by generating random x’s and y’s in the range $\left\{ {x_{\hbox{min} }, y_{\hbox{min} } ,x_{\hbox{max} } ,y_{\hbox{max} } } \right\}$ and substitute them in p(x, y) for temperature distribution on rooftop (Fig. 11.14a). Recomputing data values in entire range helps increase the accuracy of the function approximation process and substituting them in the relation:

Fig. 11.14

a Rooftop temperature distribution using a WSN and b temperature distribution using regression polynomial in a WSN

$$ \begin{aligned} p\left( {x,y} \right) & = 26.1429 + 0.0427163y - 0.000167934y^{2} \\ & \quad + 0.014x + 0.000249xy - 0.00000009231xy^{2} \\ & \quad - 0.0000181258x^{2} - 0.000000860054x^{2} y + 0.00000000116143x^{2} y^{2} \\ \end{aligned} $$

(11.11)

This leads Fig. 11.14b as values obtained using regression polynomial. Percentage error $\varepsilon$ is calculated as the absolute deviation from the true value and percentage error :

$$ E = \left( {\frac{{\left| {z - \bar{z}} \right|}}{z} \times 100} \right) \le \varepsilon_{\text{Th}} , $$

(11.12)

where $\varepsilon_{\text{Th}} = 6\%$ is error threshold.

Maximum error of 5.64% is observed with tree of depth 4, while error is mostly in 0–1.68% range. Compression ratio of 50% is obtained at each level of the tree and is independent of depth of aggregation tree. The data field will contain just the 9 coefficients that lead to fixed data packet size, giving substantial energy savings. A general multi-linear regression model is as follows:

$$ z = f\left( {x_{1} ,x_{2} , \ldots ,x_{m} } \right) = \beta_{0} + \sum\limits_{k = 1}^{m} {\beta_{k} x_{k} } . $$

(11.13)

Squared error needs to be minimized by applying least square criteria given by:

$$ F\left( {\mathop \beta \limits^{ \to } } \right) = \left( {X\mathop \beta \limits^{ \to } - \mathop z\limits^{ \to } } \right)^{\text{T}} \left( {X\mathop \beta \limits^{ \to } - \mathop z\limits^{ \to } } \right). $$

(11.14)

$$ X = \left( {\begin{array}{*{20}c} 1 & {y_{1} } & {y_{1}^{2} } & {x_{1} } & {x_{1} y_{1} } & {x_{1} y_{1}^{2} } & {x_{1}^{2} } & {x_{1}^{2} y_{1} } & {x_{1}^{2} y_{1}^{2} } \\ 1 & {y_{2} } & {y_{2}^{2} } & {x_{2} } & {x_{2} y_{2} } & {x_{1} y_{2}^{2} } & {x_{2}^{2} } & {x_{2}^{2} y_{2} } & {x_{2}^{2} y_{2}^{2} } \\ : & : & : & : & : & : & : & : & : \\ 1 & {y_{n} } & {y_{n}^{2} } & {x_{n} } & {x_{n} y_{n} } & {x_{1} y_{n}^{2} } & {x_{n}^{2} } & {x_{n}^{2} y_{n} } & {x_{n}^{2} y_{n}^{2} } \\ \end{array} } \right)\quad \beta = \left( {\begin{array}{*{20}c} {\beta_{0} } \\ {\beta_{1} } \\ : \\ {\beta_{8} } \\ \end{array} } \right)\quad \vec{\beta } = \left( {X^{T} X} \right)^{ - 1} \quad X^{T} \vec{z}. $$

(11.15)

For estimating β, a unique inverse of X should exist. Attribute values generated at a particular node is sent to its parent along with $\left\{ {x_{\hbox{min} }, y_{\hbox{min} } ,x_{\hbox{max} } ,y_{\hbox{max} } } \right\}$, where the minimum and maximum of x’s and y’s are taken over all the SNs in the subtree under the current parent that report to tree nodes. % error is plotted in Fig. 11.15a for tree of depth 4, while its variation as a function of depth is given in Fig. 11.15b.

Fig. 11.15

a % error with depth 4 using regression polynomial and b % error as a function of depth in regression polynomial

A large number of SNs produce data periodically and internal nodes in the data collection tree average data received from downstream nodes and forward the result toward sink. Aggregator concatenates multiple data items and transmits as a single packet. So, the question is how long should a node wait to receive data from its children before forwarding data already received? There is a trade-off between data accuracy and freshness. It is quite possible that the coefficients need not be updated every time new data are received. When coefficients are updated every two hours over a 24-h period, % error strictly increases with time due to increased number of approximations and is shown in Fig. 11.16a. The first five updating steps (till 10th hour of observation) has error 5.3% (<6) while approximated maximum and minimum temperature for a 12 h period closely follows the actual data (Fig. 11.16b).

Fig. 11.16

a % error when coefficients updated every 2 h and b temperature variation when coefficients updated every 2 h

A similar analysis has been done for 3-D mesh-connected WSN (Fig. 11.17a) using the following regression polynomial using 8 coefficients:

Fig. 11.17

a 3-D mesh deployment of SNs and b % error in 3-D mesh deployment of SNs

$$ t = F(x,y,z) = a_{0} + a_{1} z + a_{2} y + a_{3} yz + a_{4} x + a_{5} xz + a_{6} xy + a_{7} xyz. $$

(11.16)

The error rate as a function of tree height is given in Fig. 11.17b.

4 Query Processing and Data Collection

As discussed earlier, routing from each SN to BS is required to respond to a query from BS and could be application specific and data centric while it is desirable to incorporate data aggregation capability at CHs while minimizing energy consumption. Users can issue declarative queries through BS without having to worry about how the data are generated, processed, and transferred within the network from SNs to BS via CHs, and how SNs/CHs are programmed to satisfy changing user interest to maintain transparency. The BS sends an inquiry packet to CH having data as well as all CHs located between BS and CH to request an update on average of SNs in their clusters. Up on receiving the request from the BS, CHs sends back a reply packet carrying the aggregated value of sensed parameter to the BS. These values then can be used as indexes into table to lookup for the optimal value of value that should be used in response to the query . The BS then sends this value as a query packet to the CH. The WSN should be able to concurrently handle several user query requests through running multiple queries. Given a set of queries Q that have been submitted to the base station, rewrite them into a new query set Q′. The optimal situation is that data requested by queries in Q′ will be just enough to answer queries in Q, and the same data needed for various queries in Q will be acquired only once by queries in Q′. Whenever a query is updated, it is checked against the synthetic query list to see whether it is beneficial to other synthetic queries; if so, the most beneficial pairs are rewritten, and the newly updated synthetic query will be checked against the synthetic query list; this process terminates when there is no further beneficial rewriting (Fig. 11.18a). Long running queries to monitor events occurring in several target regions geographically were separated from each other. Communication architecture supports continuous in-network query processing (Fig. 11.18b). The presence of CH introduces heterogeneity in the network, and few CHs could be used as additional query processor rather than a central storage at the BS. So, the question is which of the CHs could be selected as a query processor in a WSN. In-network query processing could lead to possible reduction in volume of data. One example of query tree is illustrated in Fig. 11.19a. If data tuples of size x is generated by each of n target regions, then data are aggregated at each higher level and are reduced by a factor φ, and then the corresponding data reduction tree is shown in Fig. 11.18b.

Fig. 11.18

a External storage and processing (BS) in a WSN and b in-network processing (at CHs) in a WSN

Fig. 11.19

a Routing tree corresponding to a query and b query tree from different regions in a WSN

Thus, a query is processed in a distributed manner by few CHs acting as aggregator operators. The CHs acting as query operators have additional storage, processing capability, and power to perform effectively. The cost of data transferring data is proportional to distance to be transferred and the cost from L and R to P through X of Fig. 11.20a given by cost function f(X, Y) is given by:

$$ f(X,Y) = \left| {\left| {LX} \right|} \right|d_{l} + \left| {\left| {RX} \right|} \right|d_{r} + \left| {\left| {XP} \right|} \right|d_{p} , $$

(11.17)

where ||LX|| is the distance between L and X and d _l is the volume of data from L. Optimal placement the operator X requires minimization of function f(X). A simple nonlinear optimization method of steepest descent can be used to find X such that f′(X) = 0. The issue that needs to be addressed is how to translate query tree to energy-aware routing tree. This requires adapting operator placement in a decentralized manner and providing robustness and scalability in a decentralized manner so that right set of CHs can be selected as a query operator. Initially, data rates and reduction factor are not known and is determined iteratively using bottom-up starting from leaf CHs and top-down from BS approaches as shown in Fig. 11.20b. After few iterations, results are obtained which are closer to optimal solution as shown in Fig. 11.20b.

Fig. 11.20

a Translating query tree to energy-aware routing in a WSN and b top-down and bottom-up iteration to lead close to optimal solution

5 Mobility as an Enabler in WSNs

In an ad hoc network, distributions of node, speed, position, distances, etc., change with time, while in a WSN, the number of SNs are assumed fixed and are static with no mobility . But, in order to minimize energy consumption in SNs, rather than forwarding data to BS in multi-hop fashion, BS itself moves around and collects data from those SNs which are within communication range of BS. This can also be achieved by special nodes called relay nodes (RNs) which move around the field of WSN, collect data from different SNs, and ultimately deliver to the BS. This enables SNs to transmit data that can be collected by RN in one hop and conserve energy. This makes the WSN a delay-tolerant system where latency is average time taken for an event to be detected at the BS and total delay E is given by:

$$ E ( {\text{Total}}\;{\text{Delay)}} = E ( {\text{Forward)}} + E ( {\text{Relay)}}, $$

(11.18)

where E(Forward) is time taken to forward the sensed data to RNs and E(Relay) is time taken for the RNs to deliver the message to the BS. SN can forward data to RN. If multiple RNs are utilized, a SN can forward data through multiple copies via several RN that helps minimize total delay for data to be delivered at BS. This delay with multiple copies is usually smaller than single copy forwarding. Such a scheme is based on an opportunistic connection for a message forwarding until it reaches the destination (BS). Opportunistic connection also implies that another mobile RN is encountered as RNs are randomly moving and the network does not have any control over the mobility of RNs while the network connectivity to SNs and BS is intermittent. The latency of such networks is such that a message may take very long time to reach the destination.

6 Conclusions

It is important to find WSN topology of WSNs, and some way of broadcasting is needed. It is desirable to have an effective broadcasting mechanism of beacon message cs for this purpose. Energy needs to be conserved as SNs are battery powered with limited resources while they still have to perform basic functions such as routing. The method of dominating set and connected dominating set schemes do offer many advantages, but are themselves difficult to implement for a large WSN. Data aggregation allows drastic reduction in volume of data, and such effect ought to be explored as much as possible. In-network processing inside WSN offers many advantages and allows to incorporate effective query processing. Mobility of Relay node makes a WSN delay-tolerant, and its usefulness in conserving energy by opportunistic routing needs to be examined carefully.

7 Questions

• Q.11.1. What is meant by delay-tolerant WSN?

• Q.11.2. What are the important factors in selecting a CH?

• Q.11.3. What are the advantages of a dominating set?

• Q.11.4. How do you differentiate connected dominating set with a regular dominating set?

• Q.11.5. A topology is shown in the following diagram. Can you determine the dominating set for the network? Can you also determine connected dominating set?

  

• Q.11.6. The following topologies use two or more tiles as base in defining the coverage of a given area. Draw Voronoi diagram for each of them and determine the following:

  • (a) The number of each type of tile needed in covering an area of size 1000 × 1000.

  • (b) What is the maximum sensing radius needed for each type of tile?

  • (c) Draw the Delaunay Triangles and determine the communication requirements.

• Q.11.7. What is meant by data aggregation?

• Q.11.8. Why so much emphasis is given on data aggregation in a WSN?

• Q.11.9. What are the limitations of regression polynomials in representing data of a WSN?

• Q.11.10. Can you try other polynomials and determine percentage error for a given application?