Consider any sensor x that is in the ES mode and located at the event boundary. When x transmits a packet to a neighboring sensor that is in the

NES mode, we say a flow is generated.

In the proposed approach, the slot size is set to ` × d, where ` is a real number larger than or equal to 1. Note that ` is usually set to be 1 in most TDMA-based protocols. Fig. 6.5 shows an example of the advantage of ` > 1. In Fig. 6.5, E and F are in the ES mode and their assigned slots are i and i + 1 respectively. Suppose that in some cycle E generates a flow in slot i and F generates a flow in slot i + 1. When C receives the packet sent by E, it will run a CSMA-based protocol immediately to forward the packet because it is in the NES mode. Fig. 6.5(b) shows the case of ` = 1, where the transmission of F could easily collide with the transmission of C at D due to the hidden terminal problem. However, as Fig. 6.5(c) shows, if we set ` = 2, the transmission of C will occur within slot i, thus avoiding the hidden terminal problem.

The purpose of prolonging the slot size is to separate flows in the time domain.

The advantage can be illustrated by Fig. 6.2, where we assume that Report 1, Report 3, and Report 5 belong to flow 1, and Report 2, Report 4, and Report 6 belong to flow 2. We can see that the hidden terminal problem in the non-event area can be avoided when these two flows are separated. Fig. 6.6 shows a more

Figure 6.5: The impact of slot size beyond event areas.

general example, where the pipeline effect will be formed when the ` is large enough. (As we can see that in Fig. 6.6(a), when ` is small, packets will move sequentially; on the other hand, in Fig. 6.6(b), when ` is large, the pipeline effect will be formed.) However, a larger slot size also incurs longer delay in the event area. Therefore, determining a proper value of ` is an important question. We will investigate how to choose a proper ` by simulation.

Synchronization

Time synchronization should be done in a strict way in conventional TDMA-based protocols. However, tight clock synchronization is not required in our protocol.

There are two major reasons. First, our TDMA-like protocol is built on top of a CSMA-based MAC protocol. This means that the backoff scheme and the CCA (Clear Channel Assessment) scheme can remove most of the collisions caused by synchronization error. Second, we use longer slot size to separate flows. Thus, we can tolerate a certain degree of synchronization error. For example, in Fig. 6.7(a), where ` = 3 and two sensors do not synchronize with each other, we can see that

Sink

b c d e

a

Sink

b c d e

a

Slot 1 Slot 2

Slot 1 Slot 5

Slot 3 Slot 2Slot 1

Slot 1 Slot 5 Slot 3

Flow_c

Flow_b

Flow_a

Flow_e

Flow_b

Flow_a

Flow_e

(a) (b)

Figure 6.6: The advantage of separating flows in time.

no matter which slots are used by A and B, there is no collision between A and B.

In our scheme, sensors are assumed to be synchronized by the occurrence of events, which trigger them to enter the ES mode. This scheme has two problems that may lead to synchronization error.

• Sensors may not enter ES-Mode simultaneously due to the event propaga-tion delay.

• When multiple events occur close in time and space, some sensors may de-tect multiple events. In our scheme, when a sensor in the ES mode dede-tects another event, we will allow it to continue its slot counting, instead of reset-ting to slot 1. On the contrary, some sensor may only detect one event and enter slot 1. This will also lead to synchronization error.

In order to solve these two problems, a simple adjustment scheme is proposed.

We assume that each sensor will count how many slots have passed after it entered the ES mode. This counter is denoted by s. When a sensor x transmits a packet in the ES mode, the counter x.s will be carried in the packet. Each of x’s neighbor sensors, say y, that overhears the packet will react as follows. (Note that when

y receives the packet, it can easily know the packet is transmitted by x in the ((x.s − 1)(mod MAXSLOT) + 1)-th slot of a cycle.)

• If y.s > x.s, then y will do nothing.

• If y.s = x.s, then y will fine-tune itself as follows. First, y will estimate the start time of x’s slot. If x is slower than itself, then nothing will be done.

Otherwise, y will shorten its current slot to synchronize with x. An example is shown in Fig. 6.7(b).

• If y.s < x.s, then y will estimate the start time of x’s slot, adjust its current slot to ((x.s − 1)(mod MAXSLOT) + 1), set y.s to x.s, and fine-tune itself.

An example is shown in Fig. 6.7(c).

With our adjustment scheme, when multiple events occur close in time and space, the sensors that detect the earliest event will dominate the clock in the ES mode. Although collisions could occur during the adjusting, the backoff and CCA mechanisms and the design of longer slot size can alleviate the collision problem.

Finally, one should note that we cannot use the slot number assigned to sen-sors to correct the synchronization error. For example, in Fig. 6.7(d), when B overhears the packet transmitted by A, it will switch to slot 10. Later on, when B overhears the packet transmitted by C, it will switch back to slot 9. Thus, the counters rather than the slot number should be used in the adjustment scheme.

Exploiting the Spatial Correlation of Sensor Data

So far, we mainly focus on the medium access issue. Next, we discuss how to exploit the spatial correlation of sensor data. We assume that a correlation radius Rcorr is given by applications and our goal is to minimize redundant reports un-der distortion constraints. We propose a report reduction scheme that provides two advantages. First, by exploiting TDMA’s features, we adopt a probability to reduce redundant reports without the aid of overhearing. Second, compared to

CC-MAC, when a sensor overhears a packet, the area of the overlap of correlation regions will be taken into consideration.

Our report reduction scheme consists of three steps. The first step is executed when a sensor detects an event. The sensor will use a probability to determine whether it should report or not (note that if the sensor decides not to report, it still needs to enter the ES mode). Then, it will enter the second step, during which it will try to overhear others’ packets before its slot arrives. With overhearing, some reports can be further discarded in this step. Finally, when the sensor’s slot arrives, it will enter the third step in which it will transmit one of the packets in its buffer, if any. The details of this scheme are described as follows.

Step 1: Because overhearing is opportunistic or even impossible sometimes, it is