See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/317184559

A Novel Scalable Key Pre-distribution Scheme for

Wireless Sensor Networks Based on Residual
Design

Article in Wireless Personal Communications May 2017

DOI: 10.1007/s11277-017-4326-9

CITATIONS READS

0 39

3 authors, including:

Hamid Haj Seyyed Javadi Mohaddese Anzani

Shahed University Shahed University
88 PUBLICATIONS 161 CITATIONS 8 PUBLICATIONS 3 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Key Predistribution Scheme for Clustered Hierarchical Wireless Sensor Networks based on Combinatorial
Designs View project

All content following this page was uploaded by Mohaddese Anzani on 23 July 2017.

The user has requested enhancement of the downloaded file.

 A Novel Scalable Key Pre-distribution Scheme for
Wireless Sensor Networks based on Residual Design

Vahid Modiria , Hamid Haj Seyyed Javadib and Mohaddese Anzanib1

a Department of Computer Engineering, Islamic Azad University, Science and Research
Branch, Tehran, Iran
b Department of Mathematics and Computer Science, Shahed University, Tehran, Iran

Email: va modiri@srbiauacir, hsjavadi@shahedacir, anzani@shahedacir

Abstract

Key distribution is one of the most challenging security issues in wireless sensor
networks. To achieve a high level of security, each pair of nodes must share a
secret key in order to communicate with each other. Recently, many researchers
have used combinatorial designs as key pre-distribution scheme in wireless sen-
sor networks. In this paper, we describe a new construction of a design in
combinatorial algebraic called residual design and use it for key establishment.
This is the first time for application of residual design. Our approach is a highly
scalable key management scheme for wireless sensor networks which provides a
good secure connectivity. We show that the basic mapping from residual de-
sign to key pre-distribution has an extremely high network scalability while this
mapping does not have high resilience. Therefore, we present a new approach
for key pre-distribution based on residual design that improves the resilience of
the network while maintaining connectivity and high scalability. Performance
and security properties of the proposed scheme are studied both analytically
and computationally to compare our scheme to main existing schemes. The
obtained results show that at equal key-ring size, our scheme provides better
scalability with high connectivity and resilience.
Keywords: Wireless sensor networks, Key pre-distribution, Security, Residual
design.

Corresponding author: Vahid Modiri

Preprint submitted to Journal of LATEX Templates October 25, 2016

 1. Introduction

A wireless sensor network (WSN) is a collection of sensor nodes that have

very limited storage capacity, energy and computational capabilities. WSNs
have many applications such as military, environmental and habitat monitor-
5 ing, vehicle monitoring, and industrial process control. Nodes in WSNs sense,
process and communicate with each other over wireless links to relay the moni-
tored data to a base station [1]. In such applications, security becomes essential
since the WSN is prone to malicious attackers.
Secure communication between a pair of sensor nodes requires privacy, in-
10 tegrity and authentication. Therefore, any two nodes should share a common
secret key. There are several key distribution and agreement approaches, known
as key agreement approaches problem, which can be used to set-up secret keys
between communicating nodes. The key agreement has been studied in general
network structures [2].
15 Based on cryptosystems, there are three different types of key management
in wireless sensor networks. Asymmetric, symmetric and hybrid. Public-key
algorithms such as Elliptic Curve Cryptography (ECC) and RSA have high
computation overhead and consume high energy in each sensor node. Power
resources and limited computation of sensor nodes, makes it undesirable to
20 use these types of algorithms. Sensor nodes may need a lot of seconds up to
minutes to execute all steps of public-key algorithms and this time may increase
the vulnerability of Denial of Service (DoS) attacks.
In addition, wireless sensor networks may not have good public-key infras-
tructure (PKI) for key distribution. Public keys should be distributed into nodes
25 via the base station, which produces high communication overhead.
Three types of key agreement schemes have been classified as trusted server,
self-enforcing, and key distribution. Since there is no fixed infrastructure and
network configuration is unknown prior to deployment, key pre-distribution

2
 scheme (KPS) is the best solution which is used in several research studies
30 [3][4].
In pre-distribution schemes, a list of keys (key-ring) is stored into sensors
before the deployment, which keys are randomly drawn from a key-pool.
KPSs can be random, deterministic and hybrid. In random schemes, key-
rings are chosen from a key-pool in a random manner and are assigned to sensor
35 nodes. In deterministic schemes, key-rings are chosen based on deterministic
methods to provide better key connectivity between nodes. Hybrid schemes are
the combination of both deterministic and random approaches to inherit best
of both worlds.
A few approaches have been proposed to provide pairwise security in WSNs.
40 A simple way is to store N 1 secret pairwise keys in each node where N is
network size. This scheme has full resilience against node capture, but it is
impractical for networks with extremely large amount of nodes.
Eschenauer and Gligor [5] proposed a randomized key pre-distribution scheme
for distributed WSNs. This approach consists of three phases: key pre-distribution,
45 shared-key discovery, and path-key establishment. In the key distribution phase,
a large key pool is generated and every sensor node is loaded with a fixed num-
ber of keys chosen randomly from this key pool along with their key identifiers.
Then the shared-key discovery phase takes place, where two nodes in their wire-
less communication range, exchange the list of key identifiers from their own
50 key-rings to find a common key. The path-key establishment phase takes place
if there is no common key between a pair of nodes in a wireless communication
range. Then two nodes try to communicate with each other through a multi-hop
path.
Based on this scheme, Chan et al. [6] proposed a q-composite random key
55 pre-distribution scheme, which increases the security of communication between
two nodes. In this scheme, two nodes can establish a connection only if they
share at least q keys.
In [7], Qian proposed a key pre-distribution scheme in which a hash func-
tion is used to improve resilience against node capture attack. Li et al. in [8]

3
60 proposed a threshold for random key pre-distribution schemes by which they
guarantee that each node in the network can establish a secure path with its l-
hop neighbours.
In [9], Blom proposed a -secure key pre-distribution scheme where each
node stores a row of a secret matrix and a column of a public matrix. Bloms
65 scheme is a deterministic scheme where any pair of nodes can share a common
secret key.
In [10], a polynomial-based key pre-distribution scheme is proposed for group
key pre-distribution. Blundo et al. use a bivariate t-degree symmetric polyno-
mial to establish secure connection.
70 Combinatorial designs are the other methods used to design a deterministic
key pre-distribution. In [11], Camtepe and Yener proposed a deterministic key
pre-distribution scheme by using the finite Generalized Quadrangles (GQ) and
Symmetric Balanced Incomplete Block Design (SBIBD). The main advantage
of this scheme is full connectivity coverage. However, the SBIBD scheme does
75 not scale to very large networks. Lee and Stinson used transversal designs
(TD) which provide better resilience [12]. Other works like are those of Ruj
and Roy who used Partially Balanced Incomplete Block Design (PBIBD) [13]
and of Chakrabarty et al. who used merging blocks [14] for deterministic key
pre-distribution.
80 In [15], Ruj et al. proposed a trade-based key management scheme. In
combinatorial trade, there exist unique secret pairwise keys between connected
nodes.
Bechkit et al. in [16], proposed another key pre-distribution approach based
on Unital design to improve scalability while providing good connectivity. Also,
85 they proposed a modified unital-based key pre-distribution scheme in order to
improve the network scalability.
The hybrid schemes which inherit benefits of both probabilistic and deter-
ministic schemes have been studied by several researchers. In [17], Liu et al.
proposed an approach where nodes are pre-loaded with bivariate polynomials
90 instead of keys. This approach is a combination of the random scheme pro-

4
 posed by Eschenauer et al. [5] with the Blundos scheme [10]. Camtepe and
Yener [3] and Kavitha and Sridharan [18] proposed hybrid designs for key pre-
distribution in sensor networks which employ combinatorial designs. In [19],
Dargahi et al. proposed a key pre-distribution scheme based on combinatorial
95 and hybrid designs. In this scheme, key-rings are selected from two key pools
and are assigned to sensor nodes before the deployment of sensor network. [20]
Presented key management in wireless sensor networks based on multiple key
encryption methods.

1.1. Our Contributions and Organization of This Work

100 As pointed out, energy resources and limited computation of sensor nodes
usually make it impractical to use public key algorithms, such as Diffie-Hellman
and elliptic curve key agreement.
In this paper, we focus on combinatorial constructions for deterministic key
pre-distribution schemes. For comparison of different schemes, three important
105 criteria scalability, connectivity and resilience is considered. Scalability
is the maximum size that the network can support. In wireless sensor networks
the connectivity is the probability that two nodes share at least a common key.
Network resilience is the ability to adapt correctly in the face of nodes attacks
and stress issues.
110 The contributions of our work can be summarized in the following points:

We introduce, for the first time, the use of residual design theory as an
important algebraic architecture, in key pre-distribution for WSNs. We
show that the basic mapping from residual design to key pre-distribution
is an extremely scalable scheme while providing good connectivity.

115 We propose a modified residual based key pre-distribution scheme in order

to increase the network resilience while maintaining a good scalability.

We evaluate and compare our new scheme against main existing approaches
considering different criteria. The simulated results and comparison with

5
 the mentioned approaches show that, our novelty provides better scalabil-
120 ity with high resilience and connectivity.

This paper is organized as follows. In Section II, we provide a brief overview to

combinatorial design theory. In Section III, we introduce our key distribution
scheme and show how to map the residual design to key distribution. We present
our analysis and computational results in Section IV. Finally in Section V, we
125 conclude.

2. Background on Combinatorial Designs

2.1. Preliminaries

A set system or design is a pair (X, A) where X is a set of v elements (points)

and A is a finite set of subsets of X called blocks. The degree of a point x X
130 is the number of blocks containing x. The rank of a set system is the size of
the largest block and (X, A) is said to be uniform of rank k if all blocks have
the same size k. A Balanced Incomplete Block Design (BIBD) is one of such
designs. In the following we provide some properties of BIBD.

Definition 1. A (v, k, )-BIBD or equivalently (v, b, r, k, )-BIBD is an ar-

135 rangement of v distinct objects into b blocks where each block contains exactly
k distinct objects and each object occurs in exactly r different blocks such that
each pair occurs together in exactly blocks.

Also in a (v, k, )-BIBD, we have: (v 1) = r(k 1) and bk = vr. In partic-

ular, a BIBD is called symmetric BIBD when b = v and therefore r = k [21].
140 Symmetric (v, k, )-BIBD is denoted by (v, k, )-SBIBD.
A (q 2 + q + 1, q + 1, 1)-BIBD with q 2 is called a projective plane of order
q and a (q 2 , q 2 + q, q + 1, q, 1)-BIBD is called an affine plane of order q where
q 2. For every prime power q 2 there exists a (q 2 + q + 1, q + 1, 1)-SBIBD
(i.e., a projective plane of order q) and also a (q 2 , q, 1)-BIBD (i.e., an affine
145 plane of order q) [21].

6
 2.2. Related Works
There are several key pre-distribution schemes which use combinatorial tech-
niques. In [3], Camptepe and Yener proposed symmetric key pre-distribution
designs based on (v, k, )-SBIBD with parameters (q 2 + q + 1, q + 1, 1) in which q
150 is a prime power such that q 2 +q +1 N where N is the number of nodes in the
network. The proposed mapping from symmetric BIBD to key pre-distribution
allows to construct q 2 + q + 1 key-rings from key pool of q 2 + q + 1 keys such
that each key-ring contains k = q + 1 keys and each two key-rings shares ex-
actly one common key. The main advantage of this scheme is that it provides
155 full connectivity between any pair of nodes in the network. Since providing
higher connectivity leads to lower resilience, symmetric design does not ensure
a perfect network resilience. Therefore, they proposed a hybrid design accord-
ing to which the complement of each block is used in order to provide key-rings
for additional nodes. In hybrid design, the large prime power q is considered
160 in a way that q + 1 < k. Then b of N blocks of size q + 1 are generated by
base symmetric design. N b blocks are randomly selected among k-subset of
the complementary design blocks. The hybrid design improves scalability and
resilience of underlying symmetric design.
Ruj et al. in [15] proposed a trade-based key pre-distribution scheme denoted
165 Trade-KP. A t (v, k) Steiner trade (also called combinatorial trade) consists
of collections T = {T1 , T2 } where Ti , (i = 1, 2) is a collection of k-elements
blocks chosen from a finite set X such that each t-set chosen from X occurs in
precisely the same number of blocks of T1 , as those of T2 . A 2 (v, k) Steiner
trade T = {T1 , T2 } is said to be strong, if any block in T1 intersects any block
170 of T2 in at most two elements. They proposed a new construction of strong
steiner trade to establish pairwise keys in sensor networks. For a prime power
q, they construct sets of k-subsets (k 4), T1 and T2 from X. A k-subset of
T1 is represented by t1i,j , such that, t1i,j = {(x, (xi + j) mod q) : 0 x < k},
where 0 i, j < q. A k-subset of T2 is represented by t2i,j , such that, t2i,j =
175 {(x, (x2 + xi + j) mod q) : 0 x < k}, where 0 i, j < q. This construction
results in a 2 (qk, k) strong Steiner trade. For mapping, each block of the set

7
 of blocks T1 T2 selected for a key-ring is such that the size of key-ring is k
(0 < k q) and the scalability of the scheme is 2q 2 .
In [16], Bechkit et al., proposed a unital-based key pre-distribution scheme
180 denoted NU-KP. Given a finite set X of v points, a Unital design is a 2-design
(q 3 + 1, q 2 (q 2 q + 1), q 2 , q + 1, 1). They proposed a basic mapping from unital
design to key pre-distribution in which b = q 2 (q 2 q + 1) key-rings of size
k = q + 1 are generated from a key pool of q 3 + 1 keys. In order to enhance
the key sharing probability while maintaining high network scalability, they
185 presented a new unital-based key pre-distribution scheme denoted t-UKP. In
this scheme, unital design blocks are generated and each node is preloaded with
t disjoint blocks. In their combinatorial approach, choosing the t parameter may
produce different results. In order to maintain a high key sharing probability

and high scalability, they proposed to choose t = q. The t-UKP scheme with

190 t = q is denoted UKP*.

2.3. Residual Design

There are a number of easy constructions which can create new BIBDs once
given the existence of a symmetric design. The relation between affine and
projective planes can be generalized to other block designs. If B0 is any block
195 of (v, b, r, k, )-BIBD, then any two elements that do not belong to B0 must
occur together in of the remaining blocks, while any two elements of B0 must
be together in 1 of the remaining blocks. It follows that the blocks B \ B0
form a BIBD when B ranges through the remaining blocks. We shall refer to
these as the residual design of the original with respect to the block B0 . We can
200 start with a symmetric balanced incomplete block design. The design resulting
from the following theorem is known as a residual design [22].

Theorem 2. Let (V, B) be a symmetric BIBD with B = {B1 , B2 , . . . , Bv } and

V = {x1 , . . . , xv }. Then for any i,

B1 \Bi , B2 \Bi , . . . , Bi1 \Bi , Bi+1 \Bi , ..., Bv \Bi

are the blocks of a (v k, v 1, k, k , )-BIBD of the point set X\Bi .

8
 Example 3. Consider (7, 3, 1)-SBIBD with the blocks
B1 = {1, 2, 3}, B2 = {1, 4, 5}, B3 = {1, 6, 7}, B4 = {2, 4, 6}, B5 = {2, 5, 7},
205 B6 = {3, 4, 7}, B7 = {3, 5, 6}.
Then,
B2 \ B1 = {4, 5}, B3 \ B1 = {6, 7}, B4 \ B1 = {4, 6}, B5 \ B1 = {5, 7},
B6 \ B1 = {4, 7}, B7 \ B1 = {5, 6}. Clearly Bi \ B1 for i = 2, ..., 7, are the blocks
of a (4, 6, 3, 2, 1)-BIBD over the point set {4, 5, 6, 7}.

210 Suppose that a design D has the parameters (v k, v 1, k, k , ) for some

v, k and . One can ask whether there exists a (v, k, )-design of which D is a
residual? The following theorem would answer this question [22].

Theorem 4. Suppose that D is a (v k, v 1, k, k , )-BIBD. If = 1 or

= 2, there is a (v, k, )-design of which D is the residual.

215 2.4. Network Model

In this work, we construct the residual design foundation by the symmetric

BIBD with parameters (q 2 + q + 1, q + 1, 1). Let the ith class of residual design
constructed from the point set X \ Bi be denoted by Ci and Bj \ Bi be denoted
by Bij for i = 1, ..., v and j = 1, ..., i 1, i + 1, ..., v respectively. Since in this
220 work, key-rings are preloaded before nodes deployment, it is necessary that
classes be specified.

2.5. Construction

Using symmetric design with parameters (q 2 + q + 1, q + 1, 1) to construct

residual design, we consider some following properties.

225 Property 1. The point set of each class in our approach forms a BIBD with
parameters (v, b, r, k, ) = (q 2 , q 2 + q, q + 1, q, 1).

proof. This property results of Theorem 2.

Property 2. Given the key-ring size k = q + 1 and the key pool size v = q 2 +
q +1, residual design can support the network size up to N = (q 2 +q +1)(q 2 +q).

9
230 proof. Since each of classes in residual design forms (q 2 , q 2 + q, q + 1, q, 1)-
BIBD and the number of classes is exactly q 2 + q + 1, we can support totally
(q 2 + q + 1)(q 2 + q) sensor nodes.

Property 3. The common point set of two classes has q 2 q elements.

proof. This can be shown by proving that any class Ci formed by remaining
235 elements of X \ Bi . Consider blocks Bi and Bj in symmetric design which have
q + 1 keys in key-ring. Then X \ (Bi Bj ) has (q 2 + q + 1) (2q + 1) elements
and therefore two classes Ci and Cj have q 2 q common elements.

Property 4. Each block constructed in residual design is repeated q times.

proof. Since each element in symmetric BIBD occurs in exactly q + 1 different

240 blocks and each block in residual design is constructed from Bi \ Bj , each block
in residual design is repeated q times.

Property 5. In residual design, each key appears in exactly q 2 (q + 1) blocks.

proof. According to residual design, each key j appears in all but q + 1 classes.
This means that key j is coming in the point set of q 2 classes ((q 2 + q + 1)
245 (q + 1) = q 2 ). Since each object in point set of any class appears in q + 1 blocks
of this class, there exist q 2 (q + 1) blocks containing key j.

3. Combinatorial Design to Key Distribution

In this section we explain how residual designs are used to distribute pairwise
keys to sensor nodes.

250 3.1. Mapping from Residual Design to Key Distribution

Consider a distributed wireless sensor network of N sensor nodes, where each

sensor has a key-ring of K keys coming from a key pool P . As expressed before,
we focus on an especial type of symmetric BIBD with parameters (q 2 + q + 1, q +
1, 1) in which q is a prime power. Camtepe and Yener [11] proposed a symmetric
255 design to key distribution based on (q 2 + q + 1, q + 1, 1)-SBIBD. In this scheme,

10
 for a prime power q, b = q 2 + q + 1 N blocks are constructed. Then each
generated block is assigned to each sensor node as a key-ring. Since symmetric
designs guarantee that any pair of blocks have = 1 object in common, their
scheme has full connectivity between any pair of nodes in the network. Contrary
260 of full connectivity, adding more nodes to the network reduces the scalability
of the scheme. Our basic approach is highly scalable since a great number of
blocks can be generated with parameter q.

Table 1: MAPPING FROM RESIDUAL DESIGN TO KEY PRE-DISTRIBUTION

Residual design Key pre-distribution

Point set (|S|) Key pool(KP )

2
Object set size (|S| = q + q + 1) Key pool size (|KP | = q 2 + q + 1)
Blocks key-rings
2

2


Number of blocks ( q + q + 1 q +q ) Number of key-rings
Size of a block (k = q) Size of a key-ring

In this work, we start by developing a simple scalable key pre-distribution

scheme based on residual design. We propose a basic mapping to decide on
265 choosing key-rings from a key set of key-pool and assign them to sensor nodes
before deployment (see table 1).
First, we consider the minimum prime power q in a way that (q 2 +q +1)(q 2 +
q) N to construct (q 2 + q + 1, q + 1, 1)-SBIBD. This symmetric design can
be used to construct residual designs (Theorem 2). We use residual designs
270 which has q 2 + q + 1 classes where each class has q 2 elements to generate totally
b = (q 2 + q + 1)(q 2 + q) blocks of size q. Then the generated blocks are assigned
to N sensor nodes as key-rings, where b N . The construction algorithm is
summarized in Algorithm I.

3.2. Analysis of the Proposed Approach

275 In this section, we provide an analysis of storage overhead, computation

overhead, communication overhead, connectivity, scalability and resilience of

11
 Algorithm I: Residual Design
Require: N {Total number of nodes}
1. Find the minimum prime power q such that (q 2 + q + 1)(q 2 + q) N .
2. Generate the base Symmetric Design with parameters (q 2 + q + 1, q + 1, 1).
v objects P = {a1 ,a2 ,...,av }.
b blocks B = {B1 ,B2 ,...,Bb } of size q + 1.
3. Generate b = (q 2 + q + 1)(q 2 + q) blocks for constructing Residual Design
from the base Symmetric Design:
Blocks Bij = Bi \ Bj where i, j = 1, ..., q 2 + q + 1.
4. Assign blocks to specified nodes.

the proposed scheme that we denote by RD-KP. The notations used in present
paper are summarized in table 2.

Table 2: List of used notation

Notation Definition

N total number of nodes in the network

NRD number of supported sensor nodes in RD-KP
NRD number of supported sensor nodes in RD*-KP
q A prime number which satisfies certain conditions
Bi i-th block of symmetric design
Ci ith class of residual design constructed from point set of X \ Bi
Bij the resulting blocks of Bj \ Bi
xij the j-th element of class i

PSC Probability that a pair of selected blocks is from same class

PDC Probability that a pair of selected blocks is from different classes
PRD Probability that a node have a common key with other nodes in RD-kP

12
 Table 3: Key pre-distribution schemes in terms of Storage, Computation and Communication
Schemes Storage Overhead Computation Overhead Communication Overhead

Fully pairwise key scheme n1 0 0

EG method O(q) O(q) O(q)
q-Composite scheme O(q) O(q) O(q)
BIBD-based Design O(q) O(q) 0
Hybrid symmetric design O(q) O(q) O(q)
RD O(q) O(q) O(q)

3.2.1. Communication Overhead

280 Communication overhead or multi-hop communication is the maximum in-
direct path length between 2 nodes. In our approach, this path length depends
on the length of key-ring of the nodes that is q + 1. The average key path length
or communication overhead is O(q).

3.2.2. Computation Overhead

285 Computation overhead is mostly used in public key cryptosystems during
encryption and decryption phases inside nodes in real environment. In pre-
distribution mechanisms, base stations execute most of computations of algo-
rithm. In real environment, each node need maximum time of q + 1 for compu-
tation. It means the average computation overhead of RD would be estimated
290 as O(q).

3.2.3. Storage Overhead

Following from the property 1, the key-ring size of a node with key length of l
is q + 1 . Therefore, the storage overhead of each sensor node is l(q + 1). In table
3, we compared storage overhead, computation overhead and communication
295 overhead of different key pre-distribution schemes rather than RD.

13
 3.2.4. Scalability
Following from the property 2, the scalability of the wireless network would
be N = (q 2 + q + 1)(q 2 + q).

3.2.5. Connectivity
300 In this section, we consider blocks Bij and Bi0 j 0 of residual design. Any pair
of selected blocks can be either one of the following two types:
1) Type SC: i = i0 , that is both of the blocks belong same class (e.g. Ci ). In
this case, each block has common keys with q 2 other blocks.
2) Type DC: i 6= i0 , that is two blocks belong to different classes Ci and Ci0 .

305 Proposition 5. The probability PSC that any pair of blocks from same class
q2
has at least a common key is q 2 +q .

proof. Since each class has q 2 + q blocks and following from the definition of
q2
Type 1, we have PSC = q 2 +q .

Proposition 6. The probability QSC that any pair of blocks is Type-SC is

q2 + q

2
QSC = .
(q 2 + q)(q 2 + q + 1)

2

Proposition 7. The probability QDC that any pair of blocks is Type-DC is

q2 + q q2 + q

1 1
QDC = .
(q 2 + q)(q 2 + q + 1)

2

The blocks distribution state of type DC can also be categorized in 3 cases:

310 1) In each class Ci there exists one block that has no common key with blocks
of class Cj (Clearly this block is the result of Bj \ Bi ).
2) There exist q 1 blocks in class Ci which all q objects are in common point

14
 set of classes Ci and Cj .
3) There exist q 2 blocks in class Ci which q 1 objects are in common point set
315 of classes Ci and Cj .

Property 6. There exist exactly q 1 blocks of class Ci in class Cj .

proof. Since the common point set of every two classes in our approach is
q 2 q, the number of blocks consisting q elements of these q 2 q elements is
q 2 q
q = q 1.

320 Property 7. The number of blocks of any class Cj that are in common with
any of the q 1 blocks of case 2 is q 2 + 1.

proof. Since each class contains q(q + 1) blocks and also according to the case 2
of common objects, we have

q(q + 1) (q 1) = q 2 + 1.

Proposition 8. Let Bij be one of the q 1 blocks of class Ci satisfying case 2

and block Bi0 j 0 of different class, then the probability that Bij and Bi0 j 0 have at
least one common key is
q1 q2 + 1
.
q2 + q q2 + q
proof. It follows from Property 6 and Property 7.

Property 8. The number of blocks of any class Cj that are in common with
any of the q 2 blocks of case 3 is q 2 q + 1.

proof. We have easily

(q 1)(q + 1) (q 1) + 1 = q 2 q + 1.

Proposition 9. Let Bij be one of the q 2 blocks of class Ci satisfying case 3 and
block Bi0 j 0 of different class, then the probability that Bij and Bi0 j 0 have at least
one common key is
q2 q2 q + 1
.
q2 +q q2 + q

15
325 proof. It follows from Property 8.

Proposition 10. The probability PDC that any pair of blocks from different
class has at least a common key is

q1 q2 + 1 q2 q2 q + 1
PDC = + .
q2 + q q2 + q q2 + q q2 + q

proof. It follows from Proposition 8 and Proposition 9.

Theorem 11. The probability PRD that any pair of blocks shares one or more
objects in residual design is expressed as follows:

q2 q2 + 1 q2 q2 q + 1
 
q1
PRD = 2 QSC + + QDC .
q +q q2 + q q2 + q q2 + q q2 + q

proof. It follows from Proposition 5, Proposition 6, Proposition 7 and Propo-

sition 10.

3.2.6. Resilience
330 In this section we consider the probability that a link is compromised when
an attacker captures x nodes. For simplicity we define some notations as follows:

Cx : event that x nodes (key-rings) are captured;

Dj : event that a block including key j is compromised;
lj : event that a given link is secured with key j;
335 l : event that a given link is secured;
Lj : event that a link secured with key j is compromised;
L : event that a link is compromised.

According to the definition of resilience and the above notations we are inter-
ested in finding the value of P (L | Cx ).
The probability that a given link is secured with key j is expressed as:

q 2 (q + 1) q 2 (q + 1)

2 2
P (lj | l) = = .
NRD (q 2 + q + 1)(q 2 + q)

2 2

16
340 Also, the probability that x compromised blocks include key j can be defined
as:

NRD q 2 (q + 1) q4 + q3 + q2 + q

x x
P (Dj | Cx ) = 1 =1 .
NRD (q 2 + q + 1)(q 2 + q)

x x

Clearly, the probability that a given link secured with j is compromised

when x nodes are captured is:

P (Lj | Cx ) = P (lj | l)P (Dj | Cx ).

Finally, in residual design, the probability that a link is compromised when an

attacker captures x nodes can be written as:

q 2 (q + 1) q4 + q3 + q2 + q

q 2X
+q+1
2 x


P (L | Cx ) = 1
.

j=1 (q 2 + q + 1)(q + 1) (q 2 + q + 1)(q + 1)

2 x

4. Modified Residual Design

In this section we propose modified residual design in order to improve re-

silience while maintaining high network scalability. According to the last section,
345 any block constructed from residual design is repeated totally q times. Therefore
two nodes may have same key-ring. In this new approach, we remove repeated
blocks. In other words, each pair of nodes is assigned with distinct key-rings.
Therefore, we need a greater q rather than the basic approach to generate the
same amount of blocks. We denote the modified residual design by RD*-KP.
350 For any prime power of q, the parameters of different schemes are summarized
in table 4.

17
 Table 4: Parameters of SBIBD, UKP, RD-KP, RD*-KP
Design v b r k

Symmetric BIBD q2 + q + 1 q2 + q + 1 q+1 q+1

3 2 2 2
NU-KP q +1 q (q q + 1) q q+1
RD-KP q2 + q + 1 (q 2 + q + 1)(q 2 + q) q 2 (q + 1) q
RD*-KP q2 + q + 1 (q 2 + q + 1)(q + 1) q(q + 1) q

4.1. Key Pre-Distribution

Before the deployment step, we generate blocks of residual design, where

each block corresponds to a key-ring. Contrary to the basic approach that two
355 nodes can be preloaded with one block, we preload each node with disjoint
blocks. It means that each two nodes share between zero and q 1 keys.
After the deployment step, each two neighbours exchange the identifiers of
their keys in order to determine the common keys. Otherwise, when neighbours
do not share any key, the path key establishment phase takes place.
360 The RD*-KP, enhances the network resilience since the attacker has to com-
promise more keys to break a secure link. Moreover, the connectivity in this
approach is similar to basic approach. Also we show that this approach main-
tains a high network scalability although it remains lower than that of the basic
approach.

365 4.2. Theoretical Analysis

In this section, we analyze some important properties of the modified resid-

ual design. The connectivity, storage overhead, computation overhead and com-
munication overhead is similar to residual design. We evaluate scalability and
resilience against node capture.

370 4.2.1. Scalability

Since each block of residual design is repeated q times and each node is



preloaded with disjoint blocks from the q 2 + q + 1 q 2 + q possible blocks of

18
 the residual design, it is obvious that the maximum number of nodes that we
can support is equal to



q2 + q + 1 q2 + q
= q 2 + q + 1 (q + 1) .


NRD =
q

4.2.2. Resilience
In our new proposed approach, each key exists in q(q + 1) key-rings and two
communicating nodes must have a common key i in their key-rings. With the
same method of residual approach, we can evaluate the resilience of the new
approach. The probability that a link between two nodes is secured using key i
is:

q(q + 1) q(q + 1)

2 2
P (li | l) = = .
NRD (q 2 + q + 1)(q + 1)

2 2
Moreover, the probability that the key i appears in one or more of x com-
promised key-rings is:
   
NRD q(q + 1) (q + 1)(q 2 + 1)
x x
P (Di | Cx ) = 1   =1  2
.
NRD (q + q + 1)(q + 1)
x x

Therefore, the probability that a link is compromised when x key-rings are

captured by an attacker is computed as:

   
2
q(q + 1) (q + 1)(q 2 + 1)
(q +q+1)
X 2 x
P (L|Cx ) =  2
(q + q + 1)(q + 1) 1 
 2
(q + q + 1)(q + 1) .

j=1
2 x

We can state that our proposed approach improves the resilience compared
375 with the residual design since the probability P (L|Cx ) obtained by our proposed
approach is obviously smaller than that of residual design which is demonstrated
in section III.

19
 Figure 1: Network Scalability and Required key-ring size. (a)Scalability of RD-KP and RD*-
KP is compared with SBIBD, Hybrid Symmetric, Trade and UKP*-KP key pre-distribution
schemes. RD-KP achieves a high network scalability and RD*-KP would provide better net-
work scalability to UKP*-KP, Trade and SBIBD key pre-distribution schemes. (b)In reverse,
the required key-ring size of RD-KP and RD*-KP is compared with SBIBD, Hybrid Symmet-
ric, Trade and UKP*-KP key pre-distribution schemes at equal network size.

4.3. Performance Comparison

In this section, we compare the two proposed approaches to existing schemes
380 considering different criteria.

20
Figure 2: Connectivity Comparison. Direct secure connectivity of RD*-KP is compared with
SBIBD, Hybrid Symmetric, 3-Composite, Trade and UKP*-KP key pre-distribution schemes.
The figure shows that the 3-Composite and Trade schemes provide a bad secure connectivity
coverage compared to other schemes. In addition the figure shows that RD*-KP scheme gives
very good probability of connectivity.

4.3.1. Scalability
In figure 1 (a) we compare the scalability of the proposed schemes against
SBIBD-KP, UKP* and Trade-KP methods. In the SBIBD scheme [11] with
order q, the key-ring size is k = q + 1. This scheme is used for generating
the maximum number of q 2 + q + 1 key-rings. Combinatorial trade [15] con-
sists of collection union T = T1 T2 , where Ti is a collection of q 2 blocks of
size k. The key-ring size is 4 k q and the number of supported sensors
is exactly N = 2q 2 . A proper choice for k in Trade-KP can be k = q 1
where q is a prime power. In unital-based key pre-distribution [16] of order m,
where each node is preloaded with t(m+1) distinct keys, there would be at least

m2 (m2 m + 1) ((t 1)(m2 1)(m + 1) + t)

t

key-rings, where t is the t-UKP parameter.

21

In UKP* scheme the parameter t is selected as t = m. The figure shows
385 that at equal key-ring size, the RD-KP scheme greatly enhances the scalability
compared to the other schemes. Additionally, RD*-KP scheme has a higher
network scalability than UKP*, Trade-KP and SBIBD-KP. We then plot in
figure 1 (b) the required key-ring size when using the same schemes. The figure
shows that at equal network size, the RD-KP and RD*-KP schemes allow to
390 reduce the key-ring size rather than the other schemes.

4.3.2. Connectivity
In figure 2 we compare the network security coverage of different 6 schemes.
SBIBD scheme ensures a perfect key sharing probability. In Trade-KP scheme
k(k1)
[15], the fraction of nodes directly connected is 2(2q 2 1) where 4 k q and
q is a prime power. The same as the last subsection, we choose the key-ring
size k = q 1. In q-composite scheme [6], two nodes must share at least q
common keys to be able to establish a secure link. The connectivity probability
of q-composite scheme [23] is calculated as:

nk k

q1
X ki i
1 .
i=0 k

n

In t-UKP the secure connectivity coverage [16] is given by:

(m + 1)2 2
1 (1 )t .
(m3 + m + 1)

In Hybrid Symmetric Design, the key-ring size is k = q + 1 and we simulated

this scheme for a network with N = (q 2 + q + 1) + 0.2(q 2 + q + 1) nodes.
395 Indeed, 0.2(q 2 + q + 1) nodes are loaded from complementary blocks. As the
connectivity coverage of both RD-KP and RD*-KP approaches are the same,
weve just drawn the connectivity of RD*-KP. The obtained results show that

22
 RD*-KP scheme gives a better connectivity coverage rather than UKP* and
also rather than Hybrid Symmetric Design for the key-ring sizes greater than
400 17. Additionally, our proposed schemes are much better than 3-composite and
Trade-KP. Although the RD*-KP connectivity coverage is greater than 0.851 for
the key-ring sizes greater than 10, this metric is lower compared to SBIBD-KP.

4.3.3. Resilience
We compare the proposed RD* scheme against those of the Trade-KP,
SBIBD and Hybrid Symmetric ones. In figure 3 (a) all four schemes are com-
pared at equal number of compromised nodes for key-ring size of k = 24. In
figure 3 (b) we plot the network resilience for key-ring size of k = 42. Here, we
provide deeper details for k = 42. In [16] the resilience of Trade-KP is expressed
as
2q 2 4q + 2 2q 2 4q + 2
+ 4(q 1)
x x1
1 .
2q 2

x
405 Here, choosing k = q 1, q would be 43.
In [3], author proved that the resilience of SBIBD-KP scheme is

q2

x
P (L|Cx ) = P (Dj |Cx ) = 1
q2 + q + 1

x

where q is a prime power. For key-ring size of 42, q would be 41. We simulated
the resilience of Hybrid Symmetric Design scheme for q = 41 and N = 1.2(q 2 +
q + 1) = 2070 nodes. The analysis shows that for the equal key-ring size, the
resilience of RD*-KP is better than SBIBD-KP and Hybrid Symmetric Design.
410 It also gives a better resilience than Trade-KP for compromised nodes number
(CNN) greater than 50.

23
Figure 3: Resilience Comparison. (a)Resilience of RD*-KP is compared with SBIBD, Hybrid
Symmetric and Trade key pre-distribution schemes. They are all compared with the same key-
ring size k = 24. RD*-KP provides a good resilience compared to SBIBD, Hybrid Symmetric
and Trade schemes for compromised nodes number greater than 27. (b)Resilience of RD*-KP
is compared with different schemes with the same key-ring size k = 42. For compromised
nodes number greater than 50, RD*-KP has advantage in terms of resilience.

4.4. Discussion
Assess our work, in this section we compare our approach against other
schemes. In tables 5, 6 we provide numerical results comparing scalability,

24
 Table 5: Comparison of different schemes in terms of Connectivity and scalability
key-ring size SBIBD-KP Hybrid Symmetric-KP
Number of nodes Pc Number of nodes Pc
30 931 1 931 0.8929
42 1807 1 1807 0.8902
68 4557 1 4557 0.9010
80 6481 1 6481 0.8961

key-ring size Trade-KP UKP*

Number of nodes Pc Number of nodes Pc
30 1922 0.226 16471 0.727
42 3698 0.232 72010 0.709
68 9522 0.239 557039 0.687
80 13122 0.240 1097421 0.681

key-ring size RD-KP RD*-KP

Number of nodes Pc Number of nodes Pc
30 865830 0.943 28861 0.943
42 3263442 0.963 77701 0.963
68 22019556 0.972 323817 0.972
80 41996880 0.980 524961 0.980

415 connectivity coverage and resilience of the six schemes (SBIBD-KP, Hybrid
Symmetric-KP, Trade-KP, UKP*, RD-KP and RD*-KP) at equal key-ring size.
Using RD-KP and RD*-KP schemes, we have the maximum number of sup-
ported nodes in network scalability. For Example, Give KRS = 80, RD-KP
would generate nodes more than 6000 times the SBIBD-KP and more than
420 35 times the UKP*. We can observe that our plan obviously better than the
other three schemes SBIBD-KP, Hybrid Symmetric-KP and Trade-KP in terms
of network resilience. As an example, for KRS = 42 and CNN = 90, the re-
silience of RD*-KP = 0.8542, SBIBD-KP = 0.8978, Trade-KP = 0.9259 and
Hybrid Symmetric-KP = 0.8856. Additionally, our proposed schemes increase

25
 Table 6: Comparison of different schemes in terms of Resilience
SBIBD-KP
KRS-CNN 10 30 50 70 90 110
24 0.3607 0.7454 0.9022 0.9639 0.9872 0.9954
42 0.2191 0.5260 0.7140 0.8285 0.8978 0.9395

Hybrid Symmetic-KP
KRS-CNN 10 30 50 70 90 110
24 0.4026 0.7493 0.8862 0.9517 0.9854 0.9935
42 0.2270 0.5223 0.7262 0.8350 0.8856 0.9341

Trade-KP
KRS-CNN 10 30 50 70 90 110
24 0.1892 0.7013 0.9171 0.9799 0.9955 0.9990
42 0.0777 0.4086 0.6805 0.8419 0.9259 0.9666

RD*-KP
KRS-CNN 10 30 50 70 90 110
24 0.3200 0.6748 0.8396 0.9097 0.9406 0.9542
42 0.2088 0.4919 0.6829 0.7844 0.8542 0.9066

425 the probability of network connectivity over three methods Hybrid Symmetric-
KP, Trade-KP and UKP*. For instance, in both RD-KP and RD*-KP, we
maintain a high connectivity coverage over 0.943.

5. Conclusion

In this work, we designed a scalable key pre-distribution scheme in WSNs

430 using a combinatorial design. We used for the first time the residual design
theory as an important combinatorial algebraic architecture. We provided a
mapping from residual design to key pre-distribution in WSNs. We achieved
an extremely high network scalability and have advantage over other key pre-
distribution schemes in terms of connectivity. The computational cost of RD

26
435 is less than or equal the similar symmetric key management algorithms. to We
then proposed modified residual design key pre-distribution scheme. Although
the modified approach allows to reach same connectivity with the first scheme,
analysis and numerical results in our simulations show that the optimized ap-
proach provides a better network resilience while giving lower network scalability
440 against residual design key pre-distribution scheme at equal key-ring size.

27
 References

[1] A. J. D. Rathnayaka and V. M. Potdar, Wireless Sensor Network transport

Protocol: A critical review, Journal of Network and Computer Applications
ELSEVIER, 1-13, 2011.

445 [2] A. J. Menezes, P. C. van Oorschot, and S. A. Vanstone, Handbook of Applied

Cryptography. NY: CRC Press, 1990.

[3] S. A. Camtepe, and B. Yener. Combinatorial Design of Key Distribution

Mechanisms for Wireless Sensor Networks. IEEE/ACM Transactions on
Networking, 15: 346-358, 2007.

450 [4] CY. Ghen, and HC. Chao, A Survey of Key Distribution in Wireless Sensor
Networks. Security and Communication Networks, 2011.

[5] L. Eschenauer and V. D. Gligor. A key-management scheme for distributed

sensor networks. In Proceeding of the 9th ACM Conference on Computer
and Communications security, 41-47, 2002.

455 [6] H. Chan, A. Perring, and D. Song, Random Key pre-distribution Schemes
for Sensore Networks. In Proceeding of IEEE Symposium on Security and
Privacy, 197-213, 2003.

[7] S. Qian, A novel key pre-distribution for wireless sensor networks. Physics
Procedia, 25: 2183-2189, 2012.

460 [8] WS. Li, CW. Tsai, M. Chen, WS. Hsieh, and CS. Yang, Threshold be-
havior of multi-path random key pre-distribution for sparse wireless sensor
networks. Mathematical and Computer Modelling, 57 (11): 2776-2787, 2013.

[9] R. Blom, An Optimal Class of Symmetric Key Generation Systems. In Pro-

ceeding of Eurocrypt, Advanceds in Cryptology, Springer, 335-338, 1985.

465 [10] C. Blundo, A. D. Santis, and A. Herzberg, U. Vaccaro, and M. Yung,

Perfectly-Secure Key Distribution for Dynamic Conferences. In Proceeding,
Advances in Cryptology (CRYPTO 92), 471-486, 1992.

28
 [11] S. A. Camtepe, and B. Yener. Combinatorial design of key distribution
mechanisms for wireless sensor networks. In P. Samarati, P. Y. A. Ryan, D.
470 Gollmann, and R. Molva, editors, ESORICS, volume 3193 of Lecture Notes
in Computer Science, Springer, 293-308, 2004.

[12] J. Lee, and D. Stinson, A Combinatorial Approach to Key pre-distribution

for Distributed Sensor Networks. IEEE Wireless Communications and Net-
working Conference (WCN 05), IEEE Communication Society, 1200-1205,
475 2005.

[13] S. Ruj, and B. Roy, Key pre-distribution Using Partially Balanced Designs
in Wireless Sensor Networks. 5th International Symposium (ISPA), Springer,
431-445, 2007.

[14] D. Chakrabarti, S. Maitra, and B. K. Roy, A Key pre-distribution Scheme

480 for Wireless Sensor Networks: Merging Blocks in Combinatorial Design. Int.
J. Inf. Sec., 105-114, 2006.

[15] S. Ruj, A. Nayak and I. Stojmenovic: Pairwise and Triple Key Distribu-
tion in Wireless Sensor Networks with Applications. IEEE Transactions on
computers, 62 (11): 2224-2237, 2013.

485 [16] W. Bechkit, Y. Challal, A. Bouabdallah, and V. Tarokh. A highly scalable

key pre-distribution scheme for wireless sensor networks. IEEE Transactions
on Wireless Communications, 12 (2): 948-959, 2013.

[17] D. Liu, P. Ning, and R. Li. Establishing pairwise keys in distributed sensor
networks. ACM Transactions on Information and System Security (TIS-
490 SEC), 8 (1): 41-77, 2005.

[18] T. Kavitha, and D. Sridharan. Hybrid design of scalable key distribution

for wireless sensor networks. IACSIT International Journal of Engineering
and Technology, 2 (2): 136-141, 2010.

29
 [19] T. Dargahi, H. H. S. Javadi, and M. Hosseinzade. Application-specific hy-
495 brid symmetric design of key pre-distribution for wireless sensor networks.
Wiley Journal of Security and Communication Networks, 1561-1574, 2015.

[20] J. Zhang, V. Varadharajan, Wireless sensor network key management sur-

vey and taxonomy, Journal of Network and Computer Applications ELSE-
VIER, 63-75, 2010.

500 [21] D. Stinson, Combinatorial designs: Construction and Analysis. Springer,

2004.

[22] W. D. Wallis, Combinatorial designs. Pure and Applied Mathematics Ser,

New York, 1988.

[23] E. Kendall, M. Kendall, and W. S. Kendall. A Generalised Formula for

505 Calculating the Resilience of Random Key Predistribution Schemes. IACR
Cryptology ePrint Archive, 426, 2012.

30

View publication stats