Simulation-Based Performance Evaluation

right after the node becomes orphan and just before the device starts listening, only in the case of Lazy-energy-scan scheme the oset calculation is made after the sensor has gathered RSSI measurements from all 16 channels. At time t0

when a device becomes an orphan, it computes a sequence of channels on which it looks for the coordinator. Depending on the enabled discovery scheme, the sensor nodes arrange all available 16 channels to be scanned, so that the rst channel in this sequence is considered rst, the second channel is considered next and so forth. The average channel oset species the position (0-based) of the actual channel of the coordinator at time t0 on this list smaller values

indicate that the coordinator is potentially found earlier. This value is averaged over the number of orphan states occurred in all simulation runs. This metric illustrates how ecient the proposed schemes are in providing the sensor nodes with information in ideal scenarios where no external interference is available. The considered simulation scenario followed by the simulation results are pre- sented.

5.3. Simulation-Based Performance Evaluation

The results shown here are obtained using the same simulation scenario as in Section 3.2. The schemes described in Section 5.1 are evaluated for varying values of the interference intensity λ and for an average number of WiFi interferers of ∆ = 300.

The trends observed below are also true for smaller average numbers of interferer densities. However, because in scenarios with lower densities of WiFi interferer nodes the number of occurred orphan states is lower and the percentage of time spent in orphan state is not substantial, therefore the eciency of the dierent schemes is hard to measure foe lower values of∆ (see Figure 4.6d). For this study,

5. Orphan Recovery

(a) Energy consumed by sensor nodes (b) Success rate

(c) Percentage of time without PAN (d) Average channel oset

Figure 5.1.: Comparison of the no-adaptation scheme with the dierent proposed coordinator discovery schemes added to the Lazy scheme where∆ = 300

95% condence level for the success rate has been conducted. The results are shown in Figure 5.1.

Looking at Figures 5.1b, 5.1a and 5.1c, the only noticeable dierence is between the no-adaptation scheme, which has the lowest performance gains, and all the other schemes which have very similar performance. There are slight variations between the dierent Lazy schemes, but nothing that is statistically signicant.

Nonetheless, Figure 5.1d sets apart these schemes from each other. The no- adaptation scheme does not need to switch channels, therefore this scheme has a constant trend set to zero. In other words, this scheme never needs to switch channels since the coordinator is zero hops away. In scenarios whereλ is roughly between 10

5.3. Simulation-Based Performance Evaluation and 30%, the Lazy-ordered scheme nds its corresponding coordinator after almost four hops, after which the probability of nding the coordinator decreases to almost 2 hops. One explanation is that usually a WiFi interference source creates interference on four WBSN channels, and the coordinator selects the next best channel with the lowest noise. This is due to an implementation choice: when a coordinator node nds several channels of the same minimal energy level, it selects the channel closest to its current operating channel in ascending order. However, as the interference grows, the options for the next best channel in terms of noise level reduces.

In the Lazy-random scheme, the sensor node sorts the sequence of channels to be scanned randomly, therefore it is expected that the average channel oset remains constant at around seven hops (counting from zero). Nevertheless, for scenarios with

λ≤ 20% this is not true. This is mainly because for such scenarios, an insucient

number of orphaning states occurs, and the results are not statistically signicant2_.

This is the same reason for lower values of ∆ no substantial dierence was found.

Interestingly, the Lazy-energy-scan scheme showed the worst performance in terms of average channel oset. Although it is assumed that channel measurements are not noisy and the coordinator node selects the channel with the minimal energy level, the sensor and the coordinator node measurements are taken at dierent times and the coordinator selects between channels with minimal energy level the channel numerically closest to its current operating channel. This is the main reason why, as the intensity of the external interference increases, the chance of dierentiating between noisy channels reduces.

The Lazy-sequence-scan and the Lazy-heuristic-sequence-scan scheme were able to nd the coordinator with in zero to one hop, dramatically reducing the average channel oset. However, due to high-interference conditions the overall performance of these schemes does not change. The Lazy-heuristic-sequence-scan manages to slightly reduce its energy consumption and time without PAN by spending double

2_{As mentioned earlier in this section, sucient number of replications are considered to reach a}

5. Orphan Recovery

the time for scanning on the rst two channels in the sequence. However, the performance of Lazy-heuristic-sequence-scan is the same as the Lazy-ordered scheme in terms of the energy consumed by the sensor node, the success rate and the percentage of time without PAN.

5.4. Discussion

The main nding is that, although the proposed Lazy-heuristic-sequence-scan and Lazy-sequence-scan scheme are very eective in predicting the channel where the coordinator could be found (due to the good average channel oset performance, see Figure 5.1d), the overall time spent in the orphan state is not aected substan- tially by this, and all the Lazy schemes show approximately the same performance (Figures 5.1c, 5.1b and 5.1a). One likely explanation for this is that the orphan, after switching to the right channel, still fails to receive the beacons because of high- interference levels after all, the average number ∆ = 300 of interferers chosen is

relatively high. In other words: these two approaches do the right thing, but it does not help, at least not in high-interference situations. While this appears to be a negative result, it is nonetheless useful for designers as it might save implementation eorts. The other schemes do much less well in terms of channel oset performance but still spend on average the same time in the orphan state (with exception of the no-adaptation scheme).