Introduction

An increasing number of sensor network applications share a common pattern: mobile sensor and actuator nodes move freely in an area that can be covered by a backbone of stationary nodes. In these applications, sensor nodes are deployed on mobile objects such as animals, goods, or humans, to track their activities.

The common setting of these applications is micro-mobility, i.e., the movement of nodes is confined to a limited area such as a floor or a building. Due to their mobility, sensor nodes need to frequently select a new routing parent from the set of stationary backbone nodes (concurrent joins and failures).

Requirements

Architecture overview

MobiSense is a hybrid architecture combining a fixed infrastructure network and mobile sensor nodes (see Figure above). The system automatically organizes nodes into a cluster-tree topology and exploits multi-channel communication to ensure contentionfree clusters.

Our design has been driven by the following concerns:

• rapidly changing link dynamics associated with mobile node movement cause the network to experiment a high degree of handoffs (join/leave requests) of mobile nodes.

• central management of a contention-free network schedule for every node that hands off or propagating every join request to the manager is energy prohibitive.

• enabling high throughput or maintaining a network schedule conflict-free in a large high churn network under single-channel communication is impractical.

• many systems also need downlink communication, be it for dissemination of actuator commands, remote reconfiguration of mobile nodes, or simply to enable IP-based communication from mobiles to gateway.

An energy-efficient micro-mobility architecture

Minimizing handoff latency and overhearing

An example of the Rapid Network Information Gathering scheme used in MobiSense. During the Network Discovery Slots, each cluster-heads broadcast small probe messages on a common channel. Nodes in the DISCOVERY state listen to those probe messages for a rapid network information gathering.

Rapid network (re)admission and convergence

To save energy at backbone nodes, MobiSense dynamically adjusts the access window size (AW), i.e. the number of mini-slots in the access period.

An example of a cluster-head sub-frame that shows the synchronization slot, the downlink slots, the scheduled slots, and the access mini-slots. A mobile node attempts to join that cluster after selecting a random access mini-slot

As the cluster size grows (i.e. the amount of remaining resources decreases) we decrease the access window to allow the backbone node to enter sleep mode earlier. Specifically, let Cmax and C be the maximum and current cluster size, respectively. Then, we allocate

time slots for the access window. Each time-slot in AW is further subdivided into mini-slots, creating a total of

mini-slots.

Handoff

MobiSense handoff scheme: a mobile node constantly monitors the received power from its cluster-head. If the received power is below the sensitivity threshold, it first announces its departure from that cluster, and secondly, it listens on a common channel during a constant number of slots to search for a new cluster. Excessive overhearing on multiple channels is avoided, while the discovery delay is constant.

• In MobiSense mobile nodes themselves decide if and when to hand off, and the ND messages allows it to gather sufficient information to make an informed handover decision.

• This technique eliminates signaling overhead between the backbone nodes involved in the handover.

• A node triggers a handover decision when the RSSI of packets from the current cluster-head drops below a power threshold of −75dBm.

Results

MobiSense achieves a low handoff-latency and high throughput while maintaining high reliability and energy-efficiency.

Handoff Latency

The handoff latency is divided into two parts: (i) discovery, and (ii) (re)admission latencies. The discovery latency is a constant delay, because nodes use network discovery slots. However, the (re)admission delay is not deterministic and depends on two factors: first, cluster heads synchronize their children on different phases, adding random offsets. Second, when multiple nodes are requesting admission the same cluster, collisions might occur, increasing the admission delay.

We evaluate the worst case scenarios, i.e., a number of nodes move from one cluster to another at the same time and trigger the handoff requests simultaneously.

Convergence Latency

We place mobile nodes in random starting positions and vary both the number of mobile nodes and the number of cluster-heads in the network.

Throughput, Reliability & Duty-Cycle

Throughput

Multi-channel communications and clustering allows to segment the network into small contention-free and dynamically scheduled clusters. By minimizing the broadcast domain of nodes we achieved high throughput, as neighboring clusters do not need to content for medium access.

Reliability

Duty-Cycle

We evaluate (1) the duty cycle of MobiSense under low and high traffic scenarios, and (2) the overhead of control traffic, i.e., time synchronization, scheduling.

Limitations

There are two main limitations in MobiSense:

1. Scalability: the current system is designed to test micro-mobility in WSNs. Hence, our implementation only allows for one level of cluster heads.

2. Synchronization overhead: changes in traffic demands of mobile nodes inside a cluster or the arrival or departure of a mobile node trigger changes on the schedule of all nodes in a cluster. These changes are propagated to child nodes through the synchronization packet, allowing a node to react immediately to changes of its transmission schedule. This requires every node to listen to synchronization packets transmitted by its cluster-head in each super-frame. However, this causes extra energy consumption in low traffic scenarios when a node does not have any data to send to its clusterhead during consecutive super-frames.

[Code]

References