The paper proposes a new prototype link layer protocol for low power wireless communication. The key objective of the design is to minimize power use. The important idea behind the protocol is that receivers send periodic probes periodically. When a sender has a packet for that receiver, it responds to the probe with an Auto-ACK packet, which is generated by the hardware. The sender then waits a short, random period of time and then sends over the data payload. The receiver then sends another probe once the packet has been received, indicating the previous packet has been successfully received.
The novelty of this design is that it takes advantage of the auto-ack design of IEEE 802.15.4 standard to improve the efficiency of the protocol. Using the hardware ack mechanism, the A-MAC reduces the turnaround time from approximately 3.75ms (Compare to Ri-MAC) to exactly 192us, preventing overhearing of irrelevant nodes. In addition, the authors presents their discovery that if multiple acks are sent by multiple senders, as long as the path delay difference is less than 500ns, do not causes destructive collide and can be received successfully. This second advantages improves the efficiency and interference–resistant when there are environmental wireless interference source around, such as a 802.11 AP.
They also implemented variants of the protocol for broadcast and network wakeup. For broadcast, a sender waits for and acknowledges the probes from all of its neighbors and then sends the packet respectively. For wakeup, two approaches are given. The first is to use the same protocol is used to send for wakeup messages. However, the process is much slower as nodes send probes at a much lower rate when not awake. The second approach is to have a reserved broadcast address. This however, violates the 802.15.4 standard and is not supported by all hardware. The paper also presents a pollcast implementation.
The paper then presents many results of experiment showing that A-MAC is indeed power efficient and reliable, including multiple Ack robustness experiments, 802.11 interference experiments and micro and marcro- benchmarks. Results concluded that the performance is reasonable.
Students of CS450 discussed multiple topics about this paper:
1. When multiple senders want to send multiple data packets to the same receiver, it may repeat contentions. For example, when Node S1 and Node S2 want to send packet to node R, they both wait and answer the R’s periodical probe, send auto-acks and send data packets. Data packets will certainly cause contention, and then R declares a probe with contention window.S1 and S2 respectively flip coins and choose its backoff times. Let’s assume S1 timeout first and resent data packets. When S1 finishes sending the first data packet, R sends a probe in responding to S1. It is unclear that what S2 will do at this time. If S2 here reset itself and answer to the probe with auto-act and then send its packet, it is surely to cause contention once again, which means each data packets either from S1 or S2 will cause a new contention or even worse, increasing contention window. On the other hand if S2 does not reset at the arrival of R’s probe that replies S1’s data packet, S2 will wait until S2 timeouts, which is a suboptimal solution.
2. The disadvantage of receiver-initiated protocols is that because of periodic probing, their channel utilization scales with node density and probe frequency rather than strictly with traffic. Packet delivery ratio drops noticeably with increase in density and decreasing probe intervals.