Check out the new USENIX Web site.

Timing Guard Bands

Figure 1: Minimizing guard bands reduces wasted active time and increases channel capacity.

The single biggest impediment to a node's battery-powered lifetime is the energy spent during radio communication, and secondary to that, the time spent in its “awake” (as opposed to its low-power shutdown “sleep” state). Reduce these times, and lifetime improves substantially. However, as soon as nodes in the network begin sleeping and are correspondingly offline, other nodes in the network that are still awake can no longer use them as a communications hub to route sensor data back to a command-and-control station (referred to as a “gateway” node). To optimize sleep time and network performance simultaneously all of the nodes must synchronize their internal clocks and sleep and wake at the same time[*].

The ratio between a node's total lifetime and the time spent in its awake state is its application duty cycle. Many monitoring and control applications allow exceptionally low duty cycles. Outdoor temperature, for example, may not need to be sampled faster than every two minutes [2] and rare-event detection systems (ex. [10]) may even allow nodes near-indefinite sleep sans event occurrence.

Though the node is shut down during sleep, a timer must still remain active to maintain synchrony with the rest of the network. This maintains the synchronized wake-up capability required by the network to resume operation. Yet, as duty cycle decreases, sleep power dominates average power consumption and timer power, in turn, comprises the bulk of the sleep power. Consequently, minimizing that timer's power consumption is key to assuring low average power consumption.

Conversely, the longer a node stays asleep, the more drift accumulates in the timer from clock frequency errors. To insure that nodes attempt communication only when other nodes are available for correspondence, the node must transmit a communication-request signal long-enough to overcome the uncertainty between their local notion of time and that of their worst network peer. This extra time required to ensure that connections can be reestablished is the temporal guard band (Figure 1). An accurate clock can save energy by reducing guard times, or in other words, for equal guard times, a node with a more accurate clock can sleep longer. However, we must ensure that this accuracy does not come at a substantial energy cost or the gains made through coordinated time will disappear.

Thomas Schmid 2008-11-14