Time synchronization is important in many systems, both wired and wireless, and a large number of time synchronization schemes exist. Well-known synchronization schemes include GPS [2] and the Network Time Protocol (NTP) [1]. There are many insecure and secure time synchronization protocols for WSNs, but they are still insufficient in some aspects.
2.1. Conventional Schemes
Wireless Sensor Networks pose a number of challenges beyond traditional network systems. Elson and Romer [32] describe the differences quite exhaustively and the other detailed design principles are described by Santashil et al.[30] Network Time Protocol (NTP) is the current standard for synchronizing clock on Internet, but it is not suitable for Wireless Sensor Networks because of some constraints. One is that WSN applications need higher precision such as the order of microsecond because of their close coupling with the physical world, but the precision of NTP is about the order of milliseconds. The other reason is that sensor node is usually resource-constrained and NTP is not energy efficient although it has been widely deployed and proved to be effective and robust on the Internet. Global Position System (GPS) devices could synchronize the time to external timescale and its accuracy could be about 200 nanoseconds relative to UTC. However GPS is not available in places such as inside building and underwater and it might also be power-consumed and expensive relative to resource constraint sensor nodes. Thus, both protocols are not suitable for Wireless Sensor Networks.
2.2. Insecure Time Synchronization Protocols for WSNs
These insecure protocols could be categorized as three parts. The protocols in first category are precision-driven and focus on maximizing the clock precision [5][6][7][9]. The reference broadcast synchronization (RBS) scheme [5] is the first work addressing the time synchronization issue in sensor networks; however it could only synchronize multiple receivers in a local region. Later, another scheme, called timing-sync protocol for sensor networks (TPSN) [7], was proposed and it could achieve network-wide time synchronization. The most precise scheme, called Flooding Time Synchronization Protocol (FTSP) [9], is the one with many advantages comparing to others. It decomposes the transmission delay exactly and adjusts sending timestamp in advance. The techniques FTSP used are the mac layer time-stamping which eliminates the most uncertain delay and linear regression to predict the relation between local clock and global clock, thus, FTSP could be more precise than others. Another advantage of FTSP is that it supports dynamic network topology and this makes FTSP more complete. Unfortunately, FTSP is an insecure scheme because it was designed without security concern. Two types of attacks have been proposed, namely pulse-delay attack and insider attack introduced by compromised nodes. Both of these attacks form the outliers of collected syncMsg, but FTSP cannot filter them. In FTSP, when attacks occur, some of the reference points whichcontain a pair of global and local timestamps referring to the same time instant become outliers and introduce serious error. Another problem is that FTSP propagate timestamp by broadcast, but it does not support message authentication; so external attackers could forge the packets with an arbitrary syncMsg. Our goal is to filter the outliers which are further away from their expected values than what are deemed reasonable.
Second category is lightweight driven; their concentration is not to maximize accuracy, but to minimize the complexity to achieve a given precision [26][27][28].
Thus, the needed synchronization accuracy is assumed to be given as a constraint, and the target is to devise a synchronization algorithm with minimal complexity to achieve a given precision. This approach is supported by the claim of the authors that the maximum time accuracy needed in sensor networks is relatively low (within fractions of a second), so it is sufficient to use a relaxed, or lightweight, synchronization scheme in sensor networks.
Another category is scalability driven [27][29][30]. These protocols consider clock synchronization might not be necessary at all times, except during sensor reading integration. Providing clock synchronization all the time will be a waste on the limited resources of sensors. For saving resources, the nodes re-synchronize only when there is a need for synchronization.
2.3. Secure Time Synchronization Protocols for WSNs
There are some studies for secure time synchronization in sensor network proposed recently [20][21][22][23][31], but there are some insufficiencies among them. The insufficiency of the Secure Pairwise Synchronization (SPS) proposed by Ganeriwal et al. [20] is that it just aborts the action when detecting the attacks and it cannot achieve the goal of time synchronization. Manzo et al. discussed the attacks against time synchronization protocols and proposed some countermeasures [21]. But it still suffers from pulse-delay attack and it doesn‟t resolve the conflict when using the μ TESLA-based broadcast authentication which requires loose time synchronization.
Sun et al. proposed a resilient time synchronization protocol whose focus is on the defense of compromised nodes [23], but it also suffers from pulse-delay attack. The
problem of Song et al. proposed two methods for detecting and tolerating delay attacks [22], but the insufficiency is that it doesn‟t support global time synchronization in multi-hop sensor networks. The latest scheme, called TinySeRSync [31], has solved all existing attacks. The concept of TinySeRSync for solving insider attack is to choose the median from 2t+1 data. It adopts the Secure Pairwise Synchronization (SPS) [20] with a slight modification to deal with pulse-delay attack and wormhole attacks. To avoid substantial communication overhead as well as frequent message collisions in dense sensor networks, it designs a local authenticated broadcast for the propagation of global synchronization messages, effectively harnessing the broadcast nature of wireless communication. The insufficiency of TinySeRSync is that the sensor cannot get the clock skew between its local clock and the global clock to compensate the constant clock drifts. Due to the effect of clock skew, the nodes need to resynchronize frequently to maintain certain precision. The more messages a sensor transmits, the more power it consumes. For saving power, maximizing the interval of resynchronization period by compensating the constant clock drifts is necessity.
Finally, the summaries of these insufficiencies of secure protocols are in Table 1.
Table 1: The insufficiencies of existing secure time synchronization protocols
Paper Existing problems
Secure time synchronization service for sensor networks [20]
Only abort the action when detecting the attacks Time synchronization attacks in sensor networks [21] Cannot defend against
pulse-delay attack
Secure and resilient clock synchronization in wireless Cannot defend against
sensor networks [23] pulse-delay attack TinySeRSync: Secure and resilient time
synchronization in wireless sensor networks [31]
Only synchronize the initial offset instead of clock skew