Fake-Source Based Techniques

Algorithms utilise dummy messages sent by afake source to provide SLP. Some

nodes are chosen as fake sources and periodically send dummy messages to obfuscate the real traffic. In the early stages of this technique, Ozturk et al. [115] introduce the concept of fake sources and propose a theoretical algorithm called short-lived fake source routing (SLFSR). The solution works as follows. If a

node receives a real message, it generates a probabilitypto decide whether to

send a dummy message. Ifpis below a thresholdP, then the node broadcasts

a dummy message to all its neighbours. SLFSR consumes more energy but it could improve the safety period. However, both Kamat et al. [74] and Ozturk et al. [115] state that only one fake source at the time for only one dummy packet is not enough to distract an adversary.

Chen and Lou [23] provide two solutions called dynamic bidirectional tree (DBT) and zigzag bidirectional tree (ZBT). They both use the fake-source

technique to confuse attackers, in such ways that the attackers are not sure if they are tracking real traffic from the source, or following dummy traffic. In DBT, each node knows its distance to the sink and of its neighbours to the sink. The source randomly sends messages to neighbours with a shorter or equal hop-distance, which works similar to the first stage of phantom routing. Then

intermediary nodes use a probability pto randomly select a neighbour to create

a branch and forward dummy traffic for h hops. The second solution (ZBT)

makes messages walk zigzags in the network. Firstly the sink generates one proxy sink with each of its sides. Then the source randomly selects a node as a

proxy source which isihops away from itself. The real traffic is following the

route that messages are from the source, to the proxy source, to the proxy sink, and finally to the sink.

Jhumka et al. [71] propose another algorithm. Jhumka et al. first prove the fake sources selection problem to be NP-complete and the algorithm works as follows. The source node sends a normal message to the sink. When the sink

receives it, it waits a short period and broadcastshawayimessages that floods

the network. When a 1-hop neighbour of the sink receives thehawayimessage

it becomes a temporary fake source (TFS) and broadcasts hf akei messages

for a period. Before the TFS becomes a normal node they broadcast hchoosei

messages. When a normal node receives the hchoosei message it becomes a

permanent fake source (PFS) if the node believes itself to be the furthest node in the network from the sink, otherwise it will become a TFS.

Bradbury et al. [17] improve the algorithm in [71] through the online esti- mation of its parameters. As the consequence, the improved algorithm provides a better SLP level than [71] without requiring prior network knowledge. Then Bradbury et al. propose DynamicSPR which is an extended version of the dy- namic fake source technique [17]. DynamicSPR optimises the way fake sources are allocated, in such a way that fake sources perform a directed random walk away from the sink. This algorithm reduces the number of fake sources present in the network and also the number of messages the technique sends (thus reducing

energy usage).

Many other algorithms have been proposed with state-of-the-art fake-source techniques [69, 70, 102]. However, these algorithms based on the fake-source techniques mentioned so far can only provide SLP against a local attacker. For the scope of the global attacker, a global protection scheme called Periodic is developed in which every node sends a message after a fixed period [105]. This provides perfect protection against an attacker with a global view of the network. The authors [105] create a model involving traces of source detection, which is used to measure the privacy of those traces as well as the energy cost of providing SLP. In addition, a different approach where statistical techniques are used to show that their global protection scheme provides high level of SLP [126]. Nodes use slotted transmission and send a packet at each interval. If a node does not have a real message to send at a slot, it sends a dummy message. This approach does not provide perfect global SLP as [105] does, but instead provides statistically strong SLP. Their model and solution aim to make the distribution of message broadcasts from nodes indistinguishable from a certain statistical distribution.

Other techniques consist of a hybrid between generating fake messages and having messages modify their routing path. Tree-based diversionary routing [94] imposes a tree structure on the network using fake sources at leaf nodes, with a focus on using the minimal energy possible at nodes 1-hop from the sink node to lengthen the networks lifetime. Similarly, fog or cloud techniques [36, 100] have been proposed to provide SLP where a normal message is routed through a group of nodes called a fog and then onwards to other fogs.

Perhaps the most significant disadvantage of the described fake-source based techniques is the volume of messages broadcast to provide SLP. This leads to increased energy consumption and an increased number of collisions, both of which result in a decreased message receive ratio. Another issue is that fake-source based techniques can perform poorly with multiple sources due to collisions between fake messages [82]. This means that a trade-off between energy

expenditure and privacy must be made [69], making dummy message schemes challenging for many large-scale networks.