JID: ADHOC

ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

Ad Hoc Networks xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

Ad Hoc Networks
journal homepage: www.elsevier.com/locate/adhoc

Construction of minimum connected dominating set in

wireless sensor networks using pseudo dominating set
Q1 Jasaswi Prasad Mohanty a,∗, Chittaranjan Mandal a, Chris Reade b, Ariyam Das c
a
Indian Institute of Technology Kharagpur, School of Information Technology, India
b
Kingston University, London, UK
c
Jadavpur University, Department of Computer Science & Engineering, Calcutta, India

a r t i c l e i n f o a b s t r a c t

Article history: In a wireless network, messages need to be sent on in an optimized way to preserve the
Received 11 June 2015 energy of the network. A minimum connected dominating set (MCDS) offers an optimized
Revised 31 January 2016
way of sending messages. However, MCDS construction is a NP-Hard problem. In this pa-
Accepted 1 February 2016
per, we propose a new degree-based greedy approximation algorithm named as Connected
Available online xxx
Pseudo Dominating Set Using 2 Hop Information (CPDS2HI), which reduces the CDS size
Keywords: as much as possible. Our method first constructs the CDS and then reduces its size further
Connected Dominating Set (CDS) by excluding some of the CDS nodes cleverly without any loss in coverage or connectivity.
Maximal independent set (MIS) The simulation results show that our method outperforms existing CDS construction algo-
Wireless sensor networks rithms in terms of both the CDS size and construction cost. CPDS2HI retains the current
Unit disk graph (UDG) best performance ratio of (4.8 + ln 5 )|opt | + 1.2, |opt| being the size of an optimal CDS of
Steiner Tree the network, and has the best time complexity of O(D), where D is the network diameter.
To the best of our knowledge this is the most time efficient and size-optimal CDS con-
struction algorithm. It has a linear message complexity of O(n), where n is the network
size and  is the maximum degree of all the nodes.
© 2016 Elsevier B.V. All rights reserved.

1 1. Introduction nodes or a mixture of both. Each node, contains an omni- 12

directional antenna, which broadcasts messages to all the 13
2 Wireless ad hoc and sensor networks are popularly nodes within its transmission range. Therefore, through 14
3 used in the health-care industry, food industry, agricul- broadcasting, a node can reach all of its nearby nodes with 15
4 ture and also in a wide range of military applications such one emission. If communicating nodes are not within the 16
5 as search and rescue, disaster control [1], etc. They form single hop radio transmission range of each other, then a 17
6 an important part of the next generation network in pro- communicating session is established through multi-hop 18
7 viding flexible deployment and mobile connectivity. Un- links by some intermediate nodes for relaying messages 19
8 like wired networks or cellular networks, no physical back- (multi-hop routing). One simple and intuitive method for 20
9 bone infrastructure is required in wireless ad hoc and sen- multi-hop routing between non-adjacent nodes in wire- 21
10 sor networks, thus offering new paradigms for routing. less networks is flooding, in which each node retransmits a 22
11 Wireless networks consist of either static nodes or mobile packet only once after receiving it. However, owing to the 23
low available bandwidth of the wireless channels and the 24
redundant retransmissions generated through pure flood- 25
∗
Corresponding author. Tel.: +91 9437564198. ing, the latter is not used as a communication mechanism 26
E-mail address: jasaswiprasad@gmail.com, jasaswiprasad@rediffmail. in wireless networks. The most popular means of multi- 27
com (J.P. Mohanty). hop routing in wireless networks is through the use of a 28

http://dx.doi.org/10.1016/j.adhoc.2016.02.003
1570-8705/© 2016 Elsevier B.V. All rights reserved.

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

2 J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx

29 virtual backbone. In this article, we mainly focus on con- as Connected Pseudo Dominating Set Using 2-Hop 90
30 structing a size optimal Connected Dominating Set (CDS) Information (CPDS2HI) to construct smaller CDSs. The pre- 91
31 as a virtual backbone of the network. A CDS is more par- liminary version of our technique is described in [7]. Our 92
32 ticularly used as a data aggregation backbone in remote scheme CPDS2HI works in three phases. In the first phase, 93
33 data gathering applications to optimize network commu- a smaller maximal independent set (MIS), designated as a 94
34 nication, which in turn saves communication energy and pseudo-dominating set (PDS) is constructed. The dominat- 95
35 extends the network lifetime [2]. ing set is pseudo dominating set because some of the ele- 96
36 A Dominating Set of a network is a subset of nodes ments may be omitted in the final dominating set. The sec- 97
37 such that any node not in the subset is a neighbour of ond phase of our algorithm constructs an improved Steiner 98
38 some element of that subset. It forms a Connected Dom- Tree which interconnects the PDS nodes in an improved 99
39 inating Set, if the subgraph induced by this set is con- way. In the final phase, some of the selected PDS nodes are 100
40 nected. In a wireless network, as there is no fixed infras- excluded cleverly to reduce the CDS size further without 101
41 tructure and centralized management, a CDS can be used any connectivity or coverage loss. Through simulation we 102
42 as a virtual backbone or spine for efficient routing and also show that our proposed algorithm CPDS2HI, outper- 103
43 connectivity management in such networks [3]. The CDS forms all the existing CDS construction algorithms in terms 104
44 can receive a packet from any node in the network and of CDS size and construction costs. CPDS2HI retains the 105
45 can retransmit it to any other remote node. A node, which current best performance ratio of (4.8 + ln 5 )|opt | + 1.2, 106
46 is not in the CDS can send a message to any other node |opt| being the size of an optimal CDS of the network, and 107
47 through the CDS nodes. It first sends its message to one has the best time complexity of O(D), where D is the net- 108
48 of its neighbouring CDS nodes. Now, the search space for work diameter. To the best of our knowledge this is the 109
49 any route is reduced to the CDS. If the destination node most time efficient and size-optimal CDS construction al- 110
50 is within the CDS it can get the message directly, oth- gorithm. It also has a linear message complexity of O(n), 111
51 erwise it gets the message from one of its neighbouring where n is the network size and  is the maximum degree 112
52 CDS nodes. Thus, during routing, broadcasting responsibil- of all the nodes. 113
53 ity lies only with the CDS nodes, instead of all the nodes One concern of using CDS as virtual backbone is the 114
54 in the network. As only the CDS nodes maintain routing in- nodes present in the CDS face the problem of early exhaus- 115
55 formation, we can save the storage space by reducing the tion of their battery life. To avoid this situation we need to 116
56 CDS size. A small sized CDS makes the routing easier, re- rotate the CDS periodically. So the proposed CDS construc- 117
57 duces the communication overhead, increases the conver- tion algorithm can be used in conjunction with CDS rota- 118
58 gence speed and simplifies connectivity management. So, it tion algorithms [5]. A customised rotation algorithm might 119
59 is desirable to construct a minimum connected dominating work better than a general one. However, evaluation of ro- 120
60 set (MCDS) of the network. However, computing MCDS is tation schemes and possible development of a customised 121
61 an NP-complete problem [4]. So, only polynomial time ap- rotation scheme is beyond the scope of the current work. 122
62 proximation algorithms are practical for finding out MCDS The rest of the article is organized as follows. Section 2 123
63 in wireless networks. For energy constrained wireless net- provides some preliminary definitions. In Section 3, we 124
64 works, an approximation algorithm should not only con- discuss related works on CDS construction. Section 4 lists 125
65 struct thinner CDSs, but also construct CDSs with low com- the motivation behind this work and its contributions. 126
66 putation and communication costs. Generally, the quality Section 5 explains our algorithm in detail. Section 6 pro- 127
67 of the CDS is measured by its approximation factor, which vides the analysis of our algorithm. Supporting simulation 128
68 is the ratio of its size to that of the MCDS. The construc- results are given in Section 7. Our conclusions are finally 129
69 tion cost is measured by overall message and time com- presented in Section 8. 130
70 plexities. The computation time of the CDS should also be
71 appreciably small in order to schedule speedy switches be-
72 tween disjoint CDSs to extend battery lifetime and opti- 2. Background 131

73 mize power consumption [5,6].

74 Although theoretically any centralized algorithm can be In this section, we discuss the preliminary concepts that 132

75 implemented in a distributed fashion, with the tradeoff of are relevant to our work. 133

76 higher protocol overhead, distributed algorithms are very

Definition 2.1 (DOMINATING SET). A dominating set (DS) 134
77 important for sensor networks and MANETs. CDS must be
for a graph G(V, E) is a subset V ⊆V such that each node in 135
78 constructed efficiently to be applicable in a mobile or large
V − V  is adjacent to at least one node in V . The nodes in 136
79 scale network. Due to the dynamism of wireless links and
the dominating set are called dominators. 137
80 nodal mobility, algorithms should rely on limited knowl-
81 edge of the current network topology. However, for appli- Definition 2.2 (CONNECTED DOMINATING SET). A 138
82 cations where node mobility is rare, centralized CDS con- Connected Dominating Set (CDS) is a dominating set 139
83 struction algorithms are very useful. In those applications, that induces a connected subgraph. The non-CDS nodes in 140
84 once the topology of the entire network is known to a par- the graph are called dominatees. 141
85 ticular node (may be the sink node), it can construct the
86 CDS without waiting for the information from the other Definition 2.3 (MAXIMAL INDEPENDENT SET). An inde- 142
87 nodes in every step. pendent set of nodes in a graph is a subset of nodes such 143
88 In this paper, we propose a new centralized degree- that no two nodes in the subset are adjacent. A maximal 144
89 based greedy approximation algorithm which we name independent set is an independent set which cannot be 145

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx 3

146 made larger (i.e adding any other node would result in the the spanning tree, each node in the tree is labelled as ei- 203
147 set containing some adjacent nodes). ther a dominator or a dominatee. For UDGs, the algorithm 204
has an approximation factor of 8|opt | + 1, time complexity 205
148 Definition 2.4 (UNIT DISK GRAPH). A unit disk graph of O(n) and message complexity of O(nlog n), where |opt| 206
149 (UDG) is the intersection graph of a family of unit disks being the size of an optimal CDS in the network. Later, 207
150 (unit radii) in the Euclidean plane. It provides a graph the- the approximation factor was improved to 8|opt| by Cardei 208
151 oretic model for broadcast networks. Nodes in the network in [15]. Cardei’s distributed algorithm grows from a sin- 209
152 form the vertices of graph. Each vertex is at the centre of a gle leader and uses 1-hop connectivity information with 210
153 unit disk. An edge exists between two vertices u and v, if degree-based heuristics and degree aware optimization for 211
154 and only if the euclidean distance between u and v is less identifying Steiner nodes as connectors in the CDS con- 212
155 than one unit. The unit is chosen to represent the maxi- struction. The algorithm has O(n) message complexity and 213
156 mum communication distance of a node. O(n) time complexity. Alzoubi’s multiple leaders based 214
distributed approach [12] first constructs an MIS by com- 215
157 Definition 2.5 (STEINER TREE). A Steiner Tree in a graph
paring node IDs within a 1-hop neighbourhood without 216
158 G(V, E) for a given subset of vertices (terminals) I⊆V, is a
spanning a tree or selecting a leader. In the next phase, 217
159 tree interconnecting (known as tapping) all the terminals
the MIS nodes are interconnected to form a CDS. The al- 218
160 in I using a set of Steiner nodes in {V − I}.
gorithm has an approximation factor of 192|opt | + 48. In a 219
similar work [16], the authors reported a distributed algo- 220
161 3. Related work rithm with a performance ratio of 172. 221
Among all the approximation algorithms for distributed 222
162 The idea to use the CDS as a virtual backbone for rout- CDS construction in UDGs, Li’s S-MIS algorithm [17] and 223
163 ing was first proposed by Ephermides in 1987 [8]. Since Misra’s collaborative cover heuristic [18] achieve the best 224
164 then many algorithms using CDS construction have been approximation factor of (4.8 + ln5 )|opt | + 1.2. Both these 225
165 reported. The various approaches found in the literature on algorithms, first construct an MIS and then tap the MIS 226
166 the CDS construction problem can be classified as central- nodes through a Steiner Tree construction. The collabora- 227
167 ized algorithms, localized algorithms and distributed algo- tive cover heuristic [18] constructs the MIS using effective 228
168 rithms based on the network information they use. coverage as a metric. However, it has a high message com- 229
169 In 1998, Guha and Khullar [9] first proposed two cen- plexity of O(n2 ) and time complexity of O(n). Du et al. in 230
170 tralized greedy algorithms for CDS construction in general [19] proposed a polynomial-time constant approximation 231
171 graphs having approximation ratio of O(ln ),  being the algorithm that leads to a CDS with bounded CDS size and 232
172 maximum degree of a node in the graph. The centralized guaranteed routing cost in terms of routing path length. 233
173 algorithms reported in [3,9] need global information of the In the CDS constructed by their algorithm, for each pair of 234
174 complete network, which makes them unsuitable for wire- vertices u and v there exists a routing path of maximum 235
175 less networks as there is no centralized control in a wire- length of seven times the shortest path between u and v. 236
176 less network. The algorithm produces the CDS of size less than or equal 237
177 Adjih [10] reported a localized algorithm for CDS con- to 148|opt | + 208 which is very high in comparison to col- 238
178 struction based on multipoint relays (MPR). However, to laborative cover [18]. 239
179 date no approximation analysis of that algorithm is known. Neiberg and Hurink [20] developed a local algorithm 240
180 Based on the MPR approach, several extensions have been where any vertex in the network itself decides whether or 241
181 reported, leading to localized MPR based CDS construction. not it is part of the dominating set depending only on the 242
182 However, the localized approaches provided in [10] are not vertices which are a constant number of hops away from 243
183 highly effective without an approximation factor to ensure it. Their polynomial time approximation scheme (PTAS) 244
184 an upper bound on CDS size. presents a theoretical method for computing a dominat- 245
185 The distributed approach is most widely used for CDS ing set of a connected UDG without the geometric em- 246
186 construction. Wu and Li [11] proposed a distributed CDS bedding of the graph in polynomial time with 1 +  ap- 247
187 construction algorithm in which they first construct a triv- proximation ( > 0). The processing time of a vertex 248
188 ial CDS and then delete the redundant nodes based on two is bounded by a polynomial in the number of vertices 249
189 sets of pruning rules. The algorithm requires each node to present in its locality (i.e. the radius of the area that needs 250
190 have knowledge of its 2-hop neighbourhood. The perfor- to be explored). The better the approximation (smaller 251
191 mance ratio of Wu and Li’s algorithm as reported in [12]  ) required, the larger will be the locality distance. This 252
192 is O(n), n being the network size. The performance ratio of in turn implies that the processing time for each ver- 253
193 distributed algorithms reported by Stojmenovic and Zunic tex increases and that the network messages may prop- 254
194 in [13] is also O(n), while that of Das and Bharghavan [3] agate uncontrollably far. Furthermore, the work also in- 255
195 is O(log n). None of these algorithms guarantee to gener- cludes the concept of a 2-separated collection. It empha- 256
196 ate a CDS of small size and also incur high message and sizes the fact that cardinality of the dominating set can be 257
197 time complexities. Most of the recent distributed CDS con- reduced if the topology is divided into local 2-separated 258
198 struction algorithms construct the CDS by first selecting collections, each separated from its nearest collection by 259
199 an MIS and then connecting the nodes in the MIS. Wan two intermediate hops. Our approach develops this con- 260
200 in [14] gave an ID based two phase single leader initiated cept and also investigates the trade-off required between 261
201 distributed algorithm to construct a CDS tree rooted at the the approximation ratio and the locality that needs to be 262
202 leader. They use a spanning tree. After the construction of surveyed. 263

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

4 J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx

264 In the recent literature Liu et al. [21] proposed a

265 biology-inspired algorithm named as “Physarum Optimiza-
266 tion” for Steiner Tree problem. The algorithm is with
267 lower approximation ratio in comparison to other clas-
268 sical algorithms. However, the running time is quadratic
269 and no message complexity of the algorithm is given.
270 Zhang et al. [22] proposed a CDS construction algorithm
271 based on the induced tree of the crossed cube with lower
272 approximation ratio. However, they have neither calcu-
Fig. 2. An example network to illustrate how PDS can improve CDS size
273 lated the time nor the space complexities of their algo-
over MIS.
274 rithm. Yu et al. [23] proposed a MCDS technique in beep-
275 ing model. The proposed algorithm constructs the CDS in
276 O(log3 N) rounds with an approximation ratio of atmost 2.
higher number of message exchanges. Consequently, the 309
277 However, collision problem may arise (when two or more
communication overhead will be higher. 310
278 nodes emit beep signals at the same time) in this beeping
We also note that in post Steiner Tree construction, 311
279 model. Hedar et al. [24] proposed two methods to solve
some dominators from the MIS can be downgraded to 312
280 MCDS problem. These two methods are based on genetic
dominatees without any loss in connectivity or coverage 313
281 algorithms and simulated annealing. As the complexities of
of the CDS. This is intuitively true when the neighbours 314
282 these soft computing techniques are high, these two meth-
(if any) of the MIS node to be downgraded are covered ei- 315
283 ods are not suitable for sensor networks.
ther by some Steiner nodes or by some other MIS nodes. 316
We illustrate this through an example. In the graph given 317
284 4. Motivations and contributions
in Fig. 2 the MIS with minimum size consists of nodes 5, 318
6 and 7. We will essentially require nodes 2 and 10 to 319
285 The recent competitive distributed algorithms in
connect the MIS. Thus, nodes 7, 2, 6, 10 and 5 form the 320
286 [14,15,17,18] to achieve constant satisfactory performance
CDS. But after CDS construction, node 6 in the CDS be- 321
287 ratio, construct MISs with a specific property stated in
comes redundant. Nodes 7, 2, 10 and 5 alone can also form 322
288 [17]. The characteristic property of such MISs is that any
a CDS. Thus, ideally an MIS should be treated as a pseudo- 323
289 pair of complementary subsets of the MIS must have a
dominating set (PDS) since post-CDS construction some of 324
290 distance of exactly two hops between them. This under-
the MIS nodes can be removed from the CDS to reduce 325
291 lying property assists in interconnecting the MIS nodes in
CDS size further. We are motivated towards dealing with 326
292 the next phase. However, intuitively in an MIS, a node can
the issues cited above to reduce the CDS size further at an 327
293 be separated from its nearest node by at most three hops.
optimal trade-off in the number of messages exchanged. 328
294 These MISs with lower cardinalities can effectively reduce
Thus, in this paper we report a three-phase greedy approx- 329
295 the CDS size further and improve the ratio of number of
imation scheme called CPDS2HI which contributes towards 330
296 connectors to number of independent nodes. The ratio has
improving the CDS size further than previous approxima- 331
297 significant impact on the lifetime of the network. We cite
tion algorithms. 332
298 an example to demonstrate this point. For the graph given
The major contributions of this paper are summarized 333
299 in Fig. 1, these MIS selection schemes will select an MIS of
below: 334
300 size 5 comprising nodes 1, 3, 5, 8 and 10. Each MIS node
301 is separated from its nearest neighbour in the MIS by
• A pseudo-dominating set (PDS) is constructed which 335
302 exactly two hops. However, nodes 1, 4 and 9 alone can
helps in identifying MISs of smaller sizes. 336
303 also form an MIS of the same graph. In the latter case,
• An improved Steiner Tree is constructed to connect the 337
304 each MIS node is separated from its nearest MIS neighbour
PDS nodes and some of them are selectively removed 338
305 by three hops. However, to construct such MISs, at least
in the later stage to build the CDS of smaller size. 339
306 2-hop neighbourhood connectivity information of each
• The centralized algorithm CPDS2HI has the best time 340
307 node is required. Furthermore, Steiner Tree construction
complexity of O(D), D being the network diameter and 341
308 [17,25] in the post-MIS selection phase will also incur a
also has linear message complexity. 342
• CPDS2HI identifies non-trivial CDSs with smaller sizes 343
even when the distribution of sensor nodes in the net- 344
work is uniform or random. 345

5. CDS construction by CPDS2HI 346

Our CDS construction scheme CPDS2HI works in the 347

following three phases: 348

A. Pseudo-dominating set construction 349

Fig. 1. An example network to illustrate an alternative MIS construction B. Improved Steiner Tree construction 350
technique. C. Removal of redundant dominators 351

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx 5

352 5.1. PDS construction Algorithm 1 Determination of PDS of a graph.

Input: A connected graph G(V, E ) in which color of each
353 In this phase of our proposed algorithm, we greedily
node is white.
354 construct a PDS as an MIS with lower cardinality as com-
Output: PDS of the graph G(V, E ) formed by black and
355 pared to other MIS algorithms. Any pair of complemen-
grey nodes.
356 tary subsets of our PDS can have a distance of either two
357 or three hops. Some of the nodes in the PDS may not 1: P DS ← φ , D ← φ , V D ← φ , i ← 1
358 be included in the CDS after the second phase depend- 2: while there is a white node in G with degree > 0 do
359 ing on coverage of the connectors connecting them to the 3: Roundi ← φ
360 rest of the CDS. We construct the PDS through a simple 4: for all white nodes u ∈ V do
361 degree-based algorithm in linear message complexity. The 5: if u satisfies any one of the following conditions:
362 algorithm uses 1-hop and 2-hop neighbours information of
(i) u has a degree strictly higher than each of its 1-hop
363 each node. As an MIS node can be separated from its near-
and 2-hop white neighbours
364 est MIS nodes by at most three hops, it is sufficient for
(ii) Each of the white nodes in the 1-hop and 2-hop
365 a node to examine only its 1-hop and 2-hop neighbours
neighbourhood of u have a degree lower than or
366 during this phase. In each round of the algorithm, several
same as that of u, but u has higher original degree
367 nodes may get elected as dominators from their respective
than its white 1-hop and 2-hop neighbours with the
368 local 2-hop neighbourhood pockets; this in turn yields an
same current degree as that of u
369 excellent linear time complexity. Thus, the first phase of
(iii) None of the white nodes in the 1-hop and 2-hop
370 our approach for an optimal CDS construction maintains
neighbourhood of u has a degreehigher than that
371 overall low time and message complexities retaining the
of u and there exists 1-hop or 2-hop white neigh-
372 basic characteristic of degree-based algorithms and at the
bour(s) of u that has the same current and original
373 same time successfully reduces the CDS size by reducing
degrees as that of u, but u has the least node ID
374 the cardinality of MIS. Our greedy algorithm, for deter-
among them
375 mining the PDS of a graph is given in Algorithm 1. The
376 algorithm starts with all nodes coloured white and pro- then
377 duces subsets of dominators(D) coloured black and virtual- 6: Add u to Roundi
378 dominators(VD) coloured grey. The PDS is the union of {u is selected as a dominator in the ith round}
379 black and grey nodes. Some white nodes are temporarily 7: end if
380 coloured yellow during the process to indicate that they 8: end for
381 have been removed from consideration in identifying the {Roundi contains dominator(s) selected at the ith round
382 dominators and virtual dominators. where no two dominators are within a 2-hop neighbour-
383 Next, we compare our PDS selection approach with a hood of each another}
384 popular MIS construction algorithm reported in [14] for 9: for all white nodes v ∈ Roundi do
385 the same example cited there. Their dominating tree con- 10: Change color of v from white to black
386 struction mechanism forms an MIS of size 4, as shown by 11: E ← E − ((N1 (v ) × V ) ∪ (V × N1 (v )))
387 the black nodes in Fig. 3(a). The black nodes in Fig. 3(b) 12: V ← V − N1 (v )
388 presents the PDS constructed from our first phase. In the {Delete all the adjacent nodes of v by changing its color
389 figure the blue nodes are coloured later and should be re- to yellow and edges incident on them from G}
390 garded as white at this stage. The cardinality of our PDS is 13: Update degree of the remaining white nodes in G
391 3. In our PDS, first node 5 is selected as a dominator over 14: end for
392 all its rivals in its 1-hop and 2-hop neighbourhood as node 15: D ← D ∪ Roundi
393 5 has a higher initial degree and a lower node ID (when 16: i←i+1
394 compared with node 6 which also has the same initial de- 17: end while
395 gree). In the next turn, node 6, with an effective degree 18: VD ← V − D
{Each of the undeleted white nodes (with degree 0)
are considered as virtual-dominators}
19: All the nodes in V D are marked grey
20: P DS ← D ∪ V D
{Dominators & Virtual-Dominators both form the PDS}

of 3, is chosen as a dominator. Although node 3 has the 396

same degree as that of node 6 in this turn, but it lost to 397
node 6 as the latter has a higher original degree. In the 398
next round, 0 is selected as the dominator. 399
We illustrate the PDS selection process with one more 400
example. For a distribution of nodes shown in Fig. 4, nodes 401
4, 6 and 10 form the PDS. In the first round, both the 402
Fig. 3. Example showing CDS construction. nodes 4 and 10 are selected as dominators. In the next 403

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

6 J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx

then a higher number of message exchanges will be re- 434

quired. Hence, this agreement of exploring up to 2-hop 435
neighbourhood only in our algorithm can significantly re- 436
duce CDS size without any considerable compromise in 437
number of messages exchanged as is evident from our 438

Fig. 4. CDS formed by the black dominators and blue connectors selec-
simulation results. Besides this locality distance of two 439
tively discarding grey virtual-dominators. hops in our algorithm is sufficient to divide the connected 440
graph into 2-separated independent dominating sets (i.e. 441
neighbouring local smaller MISs separated from each other 442
12
by two intermediate hops). If we reduce the locality dis- 443
MIS by Collaborative Cover
tance further, then this will result in 1-separated local 444
PDS by CPDS2HI
11 MISs that will increase the approximation ratio. 445

5.2. Improved Steiner Tree construction 446

10
Cardinality

In this phase, we tap all the dominators and virtual- 447

9 dominators in the PDS by selecting Steiner nodes from the 448

dominatees greedily to construct a Steiner Tree spanning 449
all the nodes in the PDS. For a graph G(V, E), our objective 450
8 is to find a Steiner Tree with minimum number of Steiner 451
nodes from {V − P DS}, thereby reducing the CDS size. At 452

7 the beginning of this phase, each of the dominators and 453

virtual-dominators forms separate components. The domi- 454
natees are selected as Steiner node based on their connec- 455
6 tion with separate components. Based on this idea, we for- 456
50 100 150 200
mulate our approach for the second phase in Algorithm 2. 457
Network Size
The algorithm starts with a mixture of black (dominator), 458
Fig. 5. Performance comparison of PDS construction phase with MIS se- grey (virtual-dominator), and white (other) nodes. It pro- 459
lection scheme. ceeds to convert some white nodes to blue (Steiner con- 460
nector) nodes. The resulting CDS consists of the subgraph 461
formed by the union of the grey, black, and blue nodes. 462
404 round, the remaining white node is 6 with effective degree Post-PDS construction in the graph shown in Fig. 4, only 463
405 zero. Hence, node 6 is chosen as a virtual-dominator and nodes 7 and 8 are adjacent to two independent compo- 464
406 its color is changed to grey. The recent collaborative cover nents. All the remaining dominatees are adjacent to exactly 465
407 heuristic [18], which produces smaller MISs than previ- one component (either dominator 4 or 10). Node 7 has a 466
408 ous MIS selection techniques [14,15] and gives marginally connection-load of 2 higher than that of dominatees 2, 3 467
409 better bound when the distribution of nodes is uniform, and 5 and a lower node id than dominatee 8 sharing same 468
410 yields an MIS size of 3 or 5 for the same graph in Fig. 4 connection-load and nodal degree. So, in the first round of 469
411 depending on whether node 6 is selected as the initiator the improved Steiner Tree construction, node 7 is chosen 470
412 or not. Selecting a leader to initiate the MIS construction as a Steiner node over rival dominatees 2, 3, 5 (adjacent 471
413 can also add to time and message complexities. The best to dominator 4) and 8 (adjacent to virtual-dominator 6) 472
414 distributed leader-election algorithm takes time O(n) and to tap dominator 4 and virtual-dominator 6. In the sub- 473
415 message O(nlog n) [26]. In addition, the border effect that sequent round, only dominatee 8 connects two separate 474
416 arises when dealing with the border nodes such as 3, 5, 9 components (dominator 10 and component 6-7-4). There- 475
417 and 11 is also responsible for a 67% increment in MIS size fore, in the next round, node 8 is selected as a connec- 476
418 as compared to our PDS size. Our scheme does not suffer tor to connect dominator 10 to the component 6-7-4. So, 477
419 from leader selection problems or border effects. We also we have the component 10-8-6-7-4. For the network in 478
420 perform a simple experiment to substantiate that our PDS Fig. 3(b), CPDS2HI will choose node 8 as the first Steiner 479
421 has smaller cardinality than the MIS selected from other node followed by the selection of node 12 as a connec- 480
422 CDS construction schemes. For varying sizes of connected tor to produce a CDS size of 5; whereas the dominating 481
423 networks in UDGs, we compare the cardinality of our PDSs tree construction [14] produces a CDS of size 6 as shown 482
424 with that of the MISs obtained from collaborative cover. in Fig. 3(a). 483
425 Each of the approaches was run 100 times. The averaged
426 results are reported in Fig. 5. Fig. 5 illustrates that our PDS 5.3. Removal of redundant dominators 484
427 has smaller sizes compared to the MIS selected from col-
428 laborative cover. This phase of CPDS2HI helps to reduce the CDS size as 485
429 It is worth mentioning that unlike PTAS [20], which much as possible. In this phase, we downgrade all the re- 486
430 first selects the approximation 1 +  and then deter- dundant (black) dominators and (grey) virtual-dominators 487
431 mines the locality around a node that needs to be ex- to (white) dominatees. The remaining virtual-dominators 488
432 plored, in our algorithm a node surveys only up to its are upgraded to (black) dominators. A dominating node is 489
433 2-hop neighbourhood. If we expand this locality further, redundant if by removing it from the CDS, the resultant 490

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx 7

Algorithm 2 Improved Steiner Tree Construction. dominatees covered by it can be covered by some other 496
dominators/connectors and its removal does not discon- 497
Input: A connected graph G(V, E ) with its PDS formed by
nect the CDS nodes. The CDS after discarding the redun- 498
black and grey nodes.
dant dominator(s) / virtual-dominator(s) still covers the 499
Output: CDS of the graph G(V, E ) formed by black, grey
entire network, as the downgraded nodes are covered by 500
and blue nodes.
their adjacent connectors. Based on this, we formulate our 501
1: All (black) dominators and (grey) virtual-dominators approach for the third phase in Algorithm 3. At the end of
form separate components (as isolated vertices).
2: Dominatees which are adjacent to the same component Algorithm 3 Removal of redundant dominators.
are rival dominatees. Input: A connected graph G(V, E ) with its CDS formed by
3: for all white dominatees u ∈ V − P DS do black, grey and blue nodes.
4: connection-load of u ← Number of independent com- Output: A probable smaller CDS of the graph G(V, E )
ponents adjacent to u. formed by black and blue nodes.
5: end for
6: i←1 1: for all (grey) virtual-dominators w ∈ V D do
7: repeat 2: if w is connected to the CDS by one connector or by
8: Roundi ← φ exactly two connectors that areadjacent to each other
9: for all white dominatees u ∈ V − P DS do then
10: if u satisfies any one of the following three condi- 3: C DS ← C DS − {w} {Omit virtual-dominator}
tions: 4: Change the color of w from grey to white
5: else
(i) Vertex u has a connection-load strictly higher than 6: Change the color of w from grey to black
each of its rival dominatees. 7: end if
(ii) The white rival dominatees of u has a connection- 8: end for
load lower than or sameas that of u, but u has 9: for all (black) dominators w ∈ D do
higher nodal degree than the rival dominatees with- 10: if All the dominatees of w are connected to some
the same connection-load other dominators/connectors then
(iii) None of the white rival dominatees of u has a 11: if w is connected to the CDS by one connector
connection-load higher thanthat of u and there ex- or by exactly two connectors that areadjacent to each
ists rival dominatees that have the same connection- other then
load and nodal degree as that of u, but u has the 12: C DS ← C DS − {w} {Omit dominator}
least node ID among them. 13: Change the color of w from black to white
then 14: end if
11: Add u to Roundi 15: end if
{u is selected as a connector from its white rival domina- 16: end for
tees in the ith round} 502
12: end if this phase, the CDS is the subgraph formed by the union 503
13: end for of the remaining black and blue nodes. All the other nodes 504
{Roundi contains connector(s) selected as Steiner nodes at are white. 505
the ith round} After getting the initial CDS shown in Fig. 4, we can re- 506
14: for all connectors v ∈ Roundi do duce the size of it by downgrading the virtual-dominator 6 507
15: Make connector v and the separate components, as it is connected to two connectors 7 & 8 which are ad- 508
that v connects, a single component. jacent. Therefore, node 6 is discarded from the CDS. The 509
16: Change color of v from white to blue nodes 4, 7, 8 and 10 form a CDS of size 4 even without 510
17: end for virtual-dominator 6. The CDS size obtained from collabora- 511
18: i←i+1 tive cover heuristic [18] for the same graph shown in Fig. 4 512
19: Update the connection-load of each of the remaining will be 5 or 7 depending on the choice of the leader to ini- 513
white nodes (if any). tiate the construction. 514
20: until all dominators and virtual-dominators in PDS are
in the same component 5.4. Working example 515
21: CDS ← connected component so formed with black,
grey and blue nodes In this subsection, we illustrate the complete working 516
{PDS nodes are joined to form a single component} procedure of our centralized algorithm CPDS2HI through a 517
network shown in Fig. 6(a). The nodes in the network are 518
randomly positioned. An edge between a pair of node indi- 519
cates that the nodes are within their communication range. 520
491 CDS is still connected and covers all the nodes of the net- Assume that the color of each node is white, each node’s 521
492 work. We know already a virtual-dominator does not cover 1-hop and 2-hop neighbours information is known. The 522
493 any uncovered node, so we can omit virtual-dominators step wise construction of PDS is shown in the Fig. 6(b)– 523
494 from the CDS if its removal does not disconnect the CDS (e). In the first round of this phase, node 1 becomes the 524
495 nodes. We can also omit a dominator from the CDS if the dominator and 23, 24 becomes virtual-dominators. Nodes 525

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

8 J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx

coloured yellow were neighbours of a dominator and have 526

been disconnected (removed from consideration) in identi- 527
fying further dominators. In the second round node 5, 19 528
and in the third round node 4, 15 become dominators. The 529
step wise construction of Steiner Tree is shown in Fig. 6(f)– 530
(k). At the start of this phase the dominators and virtual- 531
dominators construct the isolated components (marked by 532
the red line) shown in Fig. 6(f). In the first round of this 533
phase the dominatees 16 and 17 becomes the connectors. 534
They form new components by merging some earlier com- 535
ponents. In the subsequent rounds node 14, 11 and 8 be- 536
come connectors. The CDS after this phase is shown in 537
Fig. 6(k). In the last phase of CPDS2HI the redundant domi- 538
nating nodes 19, 23 and 24 are removed from the CDS. The 539
virtual-dominators 23 and 24 are connected to the CDS by 540
one connection. Similarly, the dominator 19 is connected 541
to the CDS by one connector and its dominatees are adja- 542
cent to some other connectors. So these redundant domi- 543
nating nodes are downgraded to dominatees. The final CDS 544
is shown in Fig. 6(l). 545

6. Algorithm analysis 546

In this section, we find the performance ratio of our 547

proposed algorithm with its time and message complexity. 548

Lemma 6.1. All black and grey nodes resulting from PDS con- 549
struction (steps of Algorithm 1) forms an MIS. 550

Proof. The while loop (step 2) of Algorithm 1 will con- 551

tinue as long as there are some white nodes in the graph 552
with degree greater than zero. When a node is selected as 553
dominator its color is changed to black and all its adja- 554
cent nodes are deleted. At the end of the while loop each 555
node is either selected as a dominator (black) or deleted 556
(as it is adjacent to some dominator) or its degree has be- 557
come zero. The white nodes with degree zero are changed 558
to color grey and also included in the PDS. In the PDS the 559
color of the nodes are either black or grey. The neighbours 560
of black nodes (yellow nodes) are not there in the PDS and 561
the grey nodes are not the neighbour of any black nodes 562
(in that case we would have deleted it before). Hence the 563
PDS is independent. The PDS is maximal because we can’t 564
add any more to the list because the other nodes (yellow 565
nodes) are dominated by black nodes.  566

Lemma 6.2. Improved Steiner Tree construction on a PDS 567

(Algorithm 2) forms a single connected component. 568

Proof. The algorithm for Steiner Tree construction 569

(Algorithm 2) first forms separate components by us- 570
ing the black and grey nodes. After that out of the 571
remaining nodes it chooses the nodes in each round which 572
are connected to maximum number of components. So 573
the connector either connects two or more components or 574
increases the size of a component. So after a finite number 575
of iterations all the components will be connected and a 576
single component will be formed.  577

Lemma 6.3. Removal of redundant dominators (by 578

Algorithm 3) does not disconnect the CDS nodes. 579
Fig. 6. Example showing CDS Construction by CPDS2HI.
Proof. The algorithm removes a virtual dominator w from 580
the CDS in the following situations: (1) w is connected to 581

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx 9

582 the CDS by one connector. In this case w can be a domi- Since PDS is an MIS, it follows from Lemmas 6.5 and 6.6 635
583 natee because the connector can dominate it. (2) w is con- that: 636
584 nected to the CDS by two connectors and they are adja- |CDS| ≤ 3.8|opt | + 1.2 + (1 + ln 5 )|opt |
585 cent. In this case by removing w from the CDS does not
586 disconnect the CDS.
= (4.8 + ln 5 )|opt | + 1.2
587 The algorithm also removes some of the dominators Thus we conclude that the performance ratio of CPDS2HI 637
588 whose dominatees are connected to some other domina- is (4.8 + ln 5 )|opt | + 1.2.  638
589 tors or connectors. Let s be a dominator whose dominatees
Theorem 6.8. CPDS2HI has time complexity of O(D) time 639
590 are connected to some other dominators or connectors. So
and O(D) rounds, where D is the network diameter. 640
591 if s is connected to the CDS by one connector or s is con-
592 nected to the CDS by two connectors and they are adjacent Proof. In our PDS construction algorithm, multiple dom- 641
593 then similar to the above argument (for virtual dominator inators may be selected simultaneously. Each dominator 642
594 w) we know that s can be removed from the CDS without is chosen from its local 2-hop neighbourhood and all its 643
595 disconnecting the CDS.  adjacent nodes become dominatees. Then the search for 644
next set of dominators, from the remaining white nodes 645
596 Theorem 6.4. Given a network, CPDS2HI determines the cor- in the locality, commences in the subsequent rounds. The 646
597 responding CDS in finite time. worst case time for the PDS construction is the maximum 647
time required to select the largest stretch of dominators 648
598 Proof. In Lemma 6.1 we proved that Algorithm 1 con-
and virtual-dominators, where each member is selected 649
599 structs an MIS. In each round of Algorithm 1, a posi-
one after another because the former choice of the dom- 650
600 tive number of nodes are removed from consideration, so
inator influences the choice of the next dominator. In the 651
601 the number of rounds is bounded. The process terminates
worst case, this longest stretch of dominators and virtual- 652
602 when all the remaining nodes have degree zero (and be-
dominators can exist along the network diameter, which is 653
603 come virtual dominators).
largest of all the shortest distances of the farthest nodes 654
604 Algorithm 2 starts with a PDS which is a dominat-
from the first set of chosen dominators. In the worst case 655
605 ing set of nodes (each considered as a separate compo-
as discussed, the number of rounds is at most O(D). There- 656
606 nent). Adding a positive number of connector nodes in
fore, the time complexity for PDS construction is O(D) time 657
607 each round preserves the dominating set property, strictly
and O(D) rounds. However, as multiple dominators can be 658
608 increases the size of one or more components so the num-
selected in each round, the average time complexity of this 659
609 ber of rounds is bounded. The connectors added may also
phase is much lower than O(D). 660
610 reduce the number of components, and since the origi-
In the Steiner Tree construction, multiple Steiner nodes 661
611 nal graph is connected, the process must terminate with a
are selected simultaneously. Each connector is selected to 662
612 dominating set forming a single component (hence a CDS).
connect the components surrounding it. By similar argu- 663
613 In phase 3, any virtual dominator or dominator re-
ment, Steiner Tree construction will also require O(D) time 664
614 moved from the CDS is chosen in such a way that the re-
and O(D) rounds. Finally, to remove the redundant dom- 665
615 duced CDS remains both connected and a dominating set.
inating nodes, each PDS nodes is checked once whether 666
616 All the virtual dominators and the dominators are consid-
to be removed or not from the PDS. So only one round 667
617 ered for downgradation to dominatee one at a time. As
is needed for this. Hence, the proposed algorithm has an 668
618 there are finite number of virtual dominators and domi-
overall time complexity of O(D) time and O(D) rounds.  669
619 nators, so it will take finite time. 
Theorem 6.9. CPDS2HI has message complexity of O(n), 670
620 Lemma 6.5. In any unit disk graph, the size of every MIS is where n is the network size and  is the maximum degree 671
621 upper-bounded by 3.8|opt | + 1.2, where |opt| is the size of of all the nodes. 672
622 the MCDS in this unit disk graph.
Proof. As CPDS2HI is a centralized algorithm it needs the 673
623 Proof. From the result reported in [27].  neighbourhood information of each node. At the sink node 674
(one node is chosen as the sink node) the consolidated 675
624 Lemma 6.6. The size of Steiner nodes obtained from CPDS2HI neighbourhood information of all the nodes is required. To 676
625 is at most (1 + ln 5 )|opt |, where |opt| is the size of any opti- construct the CDS of the entire network we run the Algo- 677
626 mal CDS. rithms 1–3 at the sink node. To find the neighbourhood 678
information of the whole network, each node should send 679
627 Proof. The proof follows directly from Theorem 2 of
a message (let it is Hello message) to its neighbours to no- 680
628 [17]. 
tify its presence. After getting the Hello messages from its 681

629 Theorem 6.7. CPDS2HI produces a CDS with size bounded by neighbours each node can construct the adjacency matrix. 682

630 (4.8 + ln 5 )|opt | + 1.2, where |opt| is the size of the MCDS. and send it to its neighbours. When a node gets the Adja- 683
cency matrix from its neighbour it updates the Adjacency 684
631 Proof. CPDS2HI in its first phase constructs the PDS as an matrix with it. If it finds any changes in it then it should 685
632 MIS. In the second phase it obtains the Steiner nodes. In send the updated Adjacency Matrix to its neighbours again. 686
633 the last phase it removes the redundant dominating nodes. By following this technique the sink node will get the fi- 687
634 Therefore, we have, nal Adjacency Matrix. In this process each node sends one 688
Hello message and a maximum of O() Adjacency Matri- 689
|CDS| ≤ |PDS| + |Steiner nodes| ces. So the total message complexity is O(n).  690

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

10 J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx

1 tation, we consider only connected networks. In the fol- 707

Collaborative Cover
No of Steiner nodes/No of independent nodes

lowing simulations, we have considered R = 25. We run 708

CPDS2HI
the algorithm 100 times for each of the different network 709
sizes. The average results are reported in the accompany- 710
0.8 ing figures and table. The complete simulation is carried 711
out in NS-2, a network simulator for wireless networks. 712
In the first experiment, we compare the Steiner nodes 713
required to connect the independent set nodes (i.e. the 714
0.6 dominators and virtual-dominators) as a function of net- 715
work size. We use a metric, which is the ratio of num- 716
ber of Steiner nodes to the number of independent set 717
nodes. We compare the results with the best existing al- 718
0.4
gorithm, collaborative cover heuristic [18], for the network 719
sizes varying from 25 to 225 as given there. The results 720
are reported in Fig. 7. The ratio provides a good measure 721
for the average effective connector degree. From the figure, 722
50 100 150 200
one can observe that for large size networks, the ratio for 723
Network Size collaborative cover becomes less than 0.3, which indicates 724

Fig. 7. Performance comparison of number of Steiner nodes and number

that on average a Steiner node connects more than three 725
of independent nodes. independent set nodes. However, the same ratio for our 726
CPDS2HI scheme in large network sizes tends to be around 727
0.5, which implies that one Steiner node often connects 728
691 7. Simulation results nearly two PDS elements. So, the average effective connec- 729
tor degree resulting from CPDS2HI scheme is less than that 730
692 In this section, we present the simulation results to of collaborative cover approach. Less effective degree of a 731
693 show the accuracy and performance of our CPDS2HI connector indicates that the connection load of the con- 732
694 scheme and compare it with the existing approaches. In nector is less which helps in enhancing the battery life and 733
695 the experimentation, the deployment area (M) is of dimen- hence the network lifetime. 734
696 sion 100 × 00 square units. We generate N number of hosts In the next experiment, we analyse the significance of 735
697 in the area M randomly, by choosing each of their abscissa ignoring virtual-dominators from PDS in the post-CDS con- 736
698 and ordinate using a uniform random number generator. struction. The network size is varied from 25 to 250 to de- 737
699 We used randomly generated instances because although termine the fraction of total virtual-dominators being dis- 738
700 deterministic node deployment has many advantages, but carded and its impact on the reduction of CDS size. Fig. 8 739
701 in large WSNs we deploy the nodes randomly to reduce shows that post-Steiner Tree construction, nearly all of the 740
702 the installation cost. The transmission range of each node virtual-dominators are discarded, occasionally at most one 741
703 is considered as R. In the network, two nodes are con- virtual-dominator is retained as a connector for bridging 742
704 nected through an edge, if the distance between them is two disjoint components of the CDS. We also find that 743
705 at most R. As the transmission range of all the nodes is post-Steiner Tree construction some of the dominators are 744
706 same, the underlying network is a UDG. In the experimen- also discarded. Results reported in Fig. 9 describe that for 745
networks of large sizes, ignoring virtual-dominators and 746
dominators according to the specified criteria, results in re- 747
Q2
duction of around 1/10th of CDS size (Fig. 10). 748
CPDS2HI Next, we compare the performance of our CPDS2HI 749
1 scheme by comparing its CDS size with the CDS sizes of 750
the existing techniques reported in [14], [15], [17], [18], 751
No. of ignoreed virtual-dominators / N

[19]. In the comparison, we considered both random and 752

uniform distribution of nodes. Firstly, we determine the
of virtual-dominators in PDS

0.95 753
CDS size for the networks of sizes 20, 40, 60, 80 and 100. 754
The result shown in Fig. 11 demonstrates that CPDS2HI 755
outperforms the dominating tree construction technique 756
0.9 [14], 8-approximate CDS algorithm [15], S-MIS approach 757
[17], collaborative cover heuristic [18] and GOC-MCDS-D 758
[19] in identifying a smaller size CDS. The result shows 759

0.85 that our approach reduces the CDS size by 16% compared 760
to the previous best collaborative cover heuristic. Fur- 761
thermore, the S-MIS [17] algorithm involving Steiner Tree 762
construction and the degree-based 8-approximate CDS 763
50 100 150 200 250
algorithm [15] result in 29% and 34% higher CDS sizes re- 764
Network Size
spectively, when compared with CPDS2HI. Our proposed 765
Fig. 8. Ratio of ignored virtual-dominators to total virtual-dominators in algorithm is able to construct the smaller CDS because of 766
pseudo-dominating set for different network sizes. the tricks in Phases 2 and 3. In Phase 2, it constructs the 767

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx 11

CPDS2HI 8-approximate CDS Algorithm

18 CPDS2HI
25
Reduction of CDS in Percentage

16

No. of rounds
20
14

12 15

10 10
50 100 150 200 250 50 100 150 200 250
Network Size No. of nodes

Fig. 9. Reduction in CDS size after discarding redundant dominators and Fig. 12. Performance comparison of number of rounds in CDS construc-
virtual-dominators post-Steiner Tree construction for different network tion.
sizes.

1
Collaborative Cover Steiner Tree by including the nodes which can connect 768
No of Steiner nodes/No of independent nodes

CPDS2HI maximum number of components. In Phase 3, it omits the 769

redundant nodes cleverly to reduce the CDS size further 770
without any coverage or connection loss. 771
0.8
One of the most important aspect of any CDS con- 772
struction technique is its time complexity. In the next 773
experiment, we compare the number of rounds required 774
0.6 by CPDS2HI and 8-approximate CDS algorithm [15] to 775
successfully construct a CDS. Here, the notion of a round 776
implies the time consumed by the sink node to complete 777
one iteration either in PDS construction phase or in Steiner 778
0.4 Tree construction phase. Therefore, the number of rounds 779
can be a useful metric to gauge the actual time expended 780
in executing a CDS construction algorithm to obtain a 781
CDS. In this experiment, 8-approximate CDS algorithm 782
50 100 150 200 [15] actually represents a class of CDS construction al- 783

Network Size gorithms that first forms an MIS by different techniques 784
with the same trait that each MIS node is separated from 785
Fig. 10. Performance comparison of CDS construction algorithms. its nearest MIS node by exactly two hops, followed by 786
greedily interconnecting the MIS nodes. The dominating 787
Collaborative Cover tree construction technique [14], S-MIS approach [17] and 788

35 Y. Li collaborative cover heuristic [18] – all fall in this class. 789

Cardei These CDS construction techniques, belonging to the same 790
Alzoubi class, will yield varying CDS sizes, but the number of 791
30 CPDS2HI rounds required by them is hardly any different. Therefore, 792
GOC-MCDS-D we have compared our CPDS2HI with only 8-approximate 793
25
CDS Size

CDS technique as the other approaches in this class 794

produce similar results. For this comparative study, we 795
20 vary the network size from 25 to 200 and determine the 796
number of rounds required for executing CPDS2HI and 797
15 8-approximate CDS algorithm. The comparison is shown in 798
Fig. 12. Although both the algorithms have the same the- 799
10 oretical upper bound of O(D) rounds, D being the network 800
diameter, it is apparent from the results that CPDS2HI 801

5 actually requires fewer rounds than 8-approximate CDS. It 802

20 40 60 80 100 is also evident from Fig. 12 that for large network sizes the 803
Network Size number of rounds required by CPDS2HI is nearly 23 rd of 804
that required by 8-approximate CDS. This directly implies 805
Fig. 11. Performance comparison of CDS construction algorithms. a near 33% reduction in execution time. We can derive 806

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

12 J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx

807 another important conclusion from this figure. The slope [9] S. Guha, S. Khuller, Approximation algorithms for connected domi- 866
808 of the curve representing CPDS2HI is significantly smaller nating sets, Algorithmica 20(4) (1998) 374–387. 867
[10] P.J.C. Adjih, L. Viennot, Computing connected dominated sets with 868
809 than that for 8-approximate CDS. This implies that as the multipoint relays, Ad Hoc Sens. Wirel. Netw. 9 (2005) 43. 869
810 network size increases, the corresponding increment in [11] J. Wu, H. Li, On calculating connected dominating set for efficient 870
811 the number of rounds is appreciably smaller for CPDS2HI routing in ad hoc wireless networks, in: Proceedings of the 3rd In- 871
ternational Workshop on Discrete Algorithms and Methods for Mo- 872
812 than other CDS construction techniques. bile Computing and Communications, ACM, 1999, pp. 7–14. 873
[12] P.J.W.K.M. Alzoubi, O. Frieder, Message-optimal connected dominat- 874
813 8. Conclusion ing sets in mobile ad hoc networks, in: Proceedings of the 3rd ACM 875
International Symposium on Mobile Ad Hoc Networking & Comput- 876
ing, ACM, 2002, pp. 157–164. 877
814 In this work, we study different approaches of con- [13] M.S.I. Stojmenovic, J. Zunic, Dominating sets and neighbour 878
815 structing a CDS in the wireless networks. We propose a elimination-based broadcasting algorithms in wireless networks, 879
IEEE Trans. Parallel Distrib. Syst. 13(1) (2002) 14–25. 880
816 new centralized degree-based greedy approximation algo-
[14] K.M.A.P.J. Wan, O. Frieder., Distributed construction of connected 881
817 rithm CPDS2HI. The CDS is constructed in three steps, dominating set in wireless ad hoc networks, in: Proceedings of 882
818 firstly by pseudo-dominating set (PDS) selection the PDS Twenty-First Annual Joint Conference of the IEEE Computer and 883
819 nodes are selected, then through an improved Steiner Tree Communications Societies, INFOCOM 2002, IEEE, 2002, pp. 1597– 884
1604. 885
820 construction technique the PDS nodes are connected and [15] X.C.M. Cardei, M.X. Cheng, D.Z. Du, Connected domination in multi- 886
821 finally, the redundant virtual-dominators and dominators hop ad hoc wireless networks, JCIS, Citeseer (2002) 251–255. 887
822 from the CDS are discarded. The simulation results show [16] M.T.T.Y. Li, S. Zhu, D.-Z. Du., Localized construction of connected 888
dominating set in wireless networks., in: Proceedings of NSF Inter- 889
823 that our algorithm constructs smaller sized CDSs in com- national Workshop on Thoretical Aspects of Wireless Ad Hoc, Sensor 890
824 parison to all other existing schemes for a wide variety and Peer-to-Peer Networks, 2004. 891
825 of distributions of sensor nodes. We have shown that the [17] F.W.C.-W.Y.P.-J.W.Y. Li, M.T. Thai, D.Z. Du, On greedy construction of 892
connected dominating sets in wireless networks., Wirel. Commun. 893
826 message and time complexity of our algorithm is O(n) Mobile Comput. 5(8) (2005) 927–932. 894
827 and O(D) respectively, where n is the network size, D is [18] R. Misra, C. Mandal, Minimum connected dominating set using a col- 895
828 the network diameter and  is the maximum degree of all laborative cover heuristic for ad hoc sensor networks, IEEE Trans. 896
Parallel Distrib. Syst. 292–302 (2010) 292–302. 897
829 the nodes. The performance ratio of our reported algorithm
[19] H. Du, W. Wu, Q. Ye, D. Li, W. Lee, X. Xu, Cds-based virtual back- 898
830 is (4.8 + ln 5 )|opt | + 1.2, where |opt| is the size of any op- bone construction with guaranteed routing cost in wireless sensor 899
831 timal CDS. The results indicate that our algorithm out- networks, IEEE Trans. Parallel Distrib. Syst. 24 (4) (2013) 652–661. 900
[20] T. Nieberg, J. Hurink., A Ptas for the Minimum Dominating Set Prob- 901
832 performs all the existing approaches that were surveyed,
lem in Unit Disk Graphs, Springer (2006) 296–306. 902
833 in terms of CDS size and message complexity. Our algo- [21] L. Liu, Y. Song, H. Zhang, H. Ma, A.V. Vasilakos, Physarum optimiza- 903
834 rithm constructs the CDS once the topology information is tion: a biology-inspired algorithm for the Steiner tree problem in 904
835 known. It does not wait for the messages from other nodes networks, IEEE Trans. Comput. 64 (3) (2015) 819–832. 905
[22] J. Zhang, S.-M. Zhou, L. Xu, W. Wu, X. Ye, An efficient connected 906
836 in each stage of construction. Although the network model dominating set algorithm in WSNS based on the induced tree of the 907
837 used in the algorithm is UDG, our reported algorithm is crossed cube, Int. J. Appl. Math. Comput. Sci. 25 (2) (2015) 295–309. 908
838 applicable for CDS construction when hosts in a network [23] J. Yu, L. Jia, D. Yu, G. Li, X. Cheng, Minimum connected dominat- 909 Q3
ing set construction in wireless networks under the beeping model, 910
839 have different transmission ranges. 2015, pp. 972–980. 911
[24] A.-R. Hedar, R. Ismail, G.A. El Sayed, K.M.J. Khayyat, Two meta- 912
840 References heuristics for the minimum connected dominating set problem with 913
an application in wireless networks, in: Proceedings of 2015 3rd In- 914
841 [1] M.K.A. Salhieh, J. Weinmann, L. Schwiebert., Power efficient topolo- ternational Conference on Applied Computing and Information Tech- 915
842 gies for wireless sensor networks., in: Proceedings of International nology/2nd International Conference on Computational Science and 916
843 Conference on Parallel Processing, IEEE, 2001, pp. 156–163. Intelligence (ACIT-CSI), IEEE, 2015, pp. 355–362. 917
844 [2] R. Misra, C. Mandal, Ant-aggregation: ant colony algorithm for op- [25] L.W.D. Du, B. Xu, The euclidean bottleneck Steiner tree and Steiner 918
845 timal data aggregation in wireless sensor networks., in: Proceedings tree with minimum number of Steiner points, in: Computing and 919
846 of International Conference on Wireless and Optical Communications Combinatorics, Springer, 2001, pp. 509–518. 920
847 Networks, 2006 IFIP, IEEE, 2006, p. 5. [26] B. Awerbuch., Optimal distributed algorithms for minimum weight 921
848 [3] R.S.B. Das, V. Bharghavan, Routing in ad hoc networks using a spine, spanning tree, counting, leader election, and related problems, in: 922
849 in: Proceedings of Sixth International Conference on Computer Com- Proceedings of the Nineteenth Annual ACM Symposium on Theory 923
850 munications and Networks, 1997. Proceedings, IEEE, 1997, pp. 34–39. of Computing, ACM, 1987, pp. 230–240. 924
851 [4] C.J.C.B.N. Clark, D.S. Johnson, Unit disk graphs, Ann. Discret. Math. 48 [27] X.J.Y.L.W. Wu, H. Du, S.C.-H. Huang, Minimum connected dominat- 925
852 (1991) 165–177. ing sets and maximal independent sets in unit disk graphs, Theor. 926
853 [5] R. Misra, C. Mandal, Rotation of CDS via connected domatic partition Comput. Sci. 352(1) (2006) 1–7. 927
854 in ad hoc sensor networks., IEEE Trans. Mobile Comput. 8(4) (2009a)
855 488–499. Jasaswi Prasad Mohanty is a Ph.D. candidate 928
856 [6] R. Misra, C. Mandal, Efficient clusterhead rotation via domatic par- in the School of Information Technology, IIT 929
857 tition in self organizing sensor networks, Wirel. Commun. Mobile Kharagpur, India. He received his Master De- 930
858 Comput. 9(8) (2009b) 1040–1058. gree in Computer Science from Utkal University, 931
859 [7] A. Das, C. Mandal, C. Reade, M. Aasawat, An improved greedy con- Bhubaneswar, India in 2006. He is working as 932
860 struction of minimum connected dominating sets in wireless net- an Assistant Professor with the department of 933
861 works, in: Proceedings of Wireless Communications and Networking Computer Science and Engineering, Silicon In- 934
862 Conference (WCNC), 2011 IEEE, IEEE, 2011, pp. 790–795. stitute of Technology, Bhubaneswar, India. His 935
863 [8] J.E.W.A. Ephremides, D.J. Baker, A design concept for reliable mobile current research interests includes wireless net- 936
864 radio networks with frequency hopping signaling, in: Proceedings of works and algorithms. 937
865 the IEEE, 75(1), 1987, pp. 56–73.

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003
 JID: ADHOC
ARTICLE IN PRESS [m3Gdc;February 22, 2016;13:24]

J.P. Mohanty et al. / Ad Hoc Networks xxx (2016) xxx–xxx 13

938 Chittaranjan Mandal received his Ph.D. from Ariyam Das received his B.E. degree (Hons) in 960
939 the Indian Institute of Technology (IIT), Kharag- Computer Science and Engineering from the Ja- 961
940 pur in 1997. He is currently working as a Pro- davpur University, India, in 2011. He held sev- 962
941 fessor with the department of Computer Sci- eral technical positions at Yahoo Research and 963
942 ence and Engineering and School of Informa- Development, Bangalore during 2011–2014. Cur- 964
943 tion Technology, IIT Kharagpur. Prior to joining rently he is working as an Assistant Professor 965
944 IIT, he served as a reader with Jadavpur Uni- in the Department of Computer Science & En- 966
945 versity, Calcutta. His research interests include gineering, Jadavpur University, Kolkata. His re- 967
946 networked systems, formal modeling and de- search interests include data mining, machine 968
947 sign and web technologies. Professor Mandal learning, information retrieval and wireless sen- 969
948 has been an Industrial Fellow of Kingston Uni- sor networks. 970
949 versity, London, U.K., since 20 0 0. He was a re-
950 cipient of a Royal Society Fellowship in 2006.

951 Chris Reade received his Ph.D. in 1984 and lec-

952 tured in Computer Science until 1999. He be-
953 came head of department of Informatics and
954 Operations Management at Kingston University
955 from 20 0 0 to 2013 and is now a consultant in
956 maths and computing. His interests are in for-
957 mal modelling, functional programming, finite
958 methods, geometric modelling and networked
959 systems.

Please cite this article as: J.P. Mohanty et al., Construction of minimum connected dominating set in wireless sensor net-
works using pseudo dominating set, Ad Hoc Networks (2016), http://dx.doi.org/10.1016/j.adhoc.2016.02.003