The routing layer is a subset of the network layer that sits just above the MAC layer in the network protocol stack and is responsible for building reliable and efficient communication links with other sensor nodes at the network layer. This includes choosing communication paths whose links have the lowest associated cost (i.e. minimum hops, minimum energy level, and optimum medium quality), constructing and destructing paths with neighbouring nodes, and maintaining routes. Existing routing protocols can be broadly generalized into three categories: proactive, reactive, and hybrid [14]. Proactive protocols maintain routes at all times, reactive protocols only build routes when they are needed in an on-demand basis, and hybrid protocol combines both proactive and reactive routing. In our context, reactive or hybrid routing is clearly more appropriate for two reasons. Firstly, proactive routing requires routing tables to be stored by each sensor node and this is not feasible as the sensor nodes have limited memory capacity. Secondly, routes may frequently change due to signal disintegration causing the need to frequently update routing tables; in proactive routing protocols, performing frequent route updates are expensive.
The way in which different routing protocols route data packets can also be classified into 3 general categories as mentioned in [14] namely, direct communication routing where sensor nodes directly route data packets to the sink, flat routing where sensor nodes route packets in a multihop fashion to the sink, and cluster routing where sensor nodes form network clusters, with each cluster consisting of a changeable cluster head and the cluster heads route packets in a multihop fashion to the sink. Since direct communication routing only supports one-hop transmissions and flat routing results in sensor nodes surrounding the sink to deplete their energies faster (thus decreasing network lifetime), we will adapt a cluster routing scheme as it is the most efficient in our context.
Two such reactive/hybrid cluster routing schemes are LEACH [11] and SPIN [10]. The concepts of LEACH and SPIN are almost similar in such that the network is connected by clusters and each cluster consists of several sensor nodes. One sensor node in each cluster is elected as the cluster head and has the responsibility to route all data packets from its cluster to other neighbouring cluster heads in a multihop fashion until it reaches the sink. To balance the workload of all sensor nodes, the duties of being cluster heads rotate among all sensor nodes in their respective clusters. In a manufacturing environment, there are several challenges in designing a reactive/hybrid cluster routing protocol. Firstly, we need to consider periodic signal disintegration not only in the inter-cluster level, but also in the intra-cluster level. This adds complexity to each sensor node for handling signal disintegration at two different levels, given the limited memory and processing capabilities of sensor nodes. Secondly, intra-cluster communication must be very reliable such that no data should be lost even in the presence of signal disintegration. In time-dependent applications, sensor nodes only transmit current data and do not re-transmit lost data. If packets are lost in between clusters, then we will get a `hole' in the network for that data gathering period and energies of sensor nodes within the affected clusters will be eventually wasted. Thirdly, if a schedule-driven MAC scheme is adopted, a sensor node may power down its radio transceiver to conserve energy at a particular time. As a result, this may potentially cause the network to be temporarily partitioned in both inter- and intra- cluster levels.