SCP-MAC : Reaching Ultra-Low Duty Cycles

Energy consumption is a critical factor in sensor networks. Since radio costs remain a large part of the energy costs in sensor network hardware, there has been much focus on minimizing energy consumption in radio medium access control (MAC) protocols. Scheduled protocols such as S-MAC [6], T-MAC [4], and TRAMA reduce energy consumption by coordinating nodes into periodic sleep/wakeup schedules. Their premise is that the cost of coordination is minimal compared to the savings in coordinated access. Recently a class of low-power listening (LPL) protocols, such as WiseMAC [1] and B-MAC [3], reduce listen overhead by replacing polling in contention periods with very low power “channel active” probes, replacing explicit coordination with per-message coordination via long pre-message preambles. However, both of scheduled and LPL-based MAC protocols are limited to duty cycles of 1–2%: scheduled protocols are limited by the delay one can tolerate between schedules, and LPL-based protocols are limited by the increasing transmit costs due to longer preambles. We explore a new approach that can achieve ultra-low duty cycles of 0.01–0.1%, potentially reducing energy consumption by a factor of 10–100. This poster describes two novel results. First, we examine the the fundamental question of the relative benefits of coordinated network access compared to unsynchronized polling. We argue that the use of LPL-like channel probing is necessary, but it must be combined with scheduled access in order to operate in ultra-low duty cycles. Second, we propose a new MAC protocol based on scheduled channel polling (SCP-MAC). The novelty of SCP-MAC is the combination of scheduling and polling; we also describe novel additions including split contention windows and piggybacked synchronization with zero overhead. We use theoretical analysis to find the best possible operating points for LPL and SCP. We demonstrate that SCP-MAC can operate for 2–3 times longer than to LPL-based MACs for the same energy budget when each is tuned for a completely periodic workload. Scheduled polling as a better match for unpredictable traffic when tuned for low-duty cycle operation. LPL suffers when mismatched to changing traffic loads because of preamble length. By contrast, SCP only pays penalty in latency, not in energy, and even the latency penalty can be eliminated with algorithms such as adaptive listen. We show in testbed experiments that LPL consumes 8 times more energy than SCP when presented with short-term bursty traffic.