**7. Practical implementation aspects**

18 Will-be-set-by-IN-TECH

<sup>0</sup> <sup>5</sup> <sup>10</sup> <sup>15</sup> <sup>20</sup> <sup>25</sup> <sup>30</sup> <sup>35</sup> <sup>40</sup> <sup>45</sup> <sup>50</sup> <sup>0</sup>

Number of sensor nodes

**Figure 8.** Total delay to distribute the content to all SN vs. the number of SNs for different values of *d*LR. which leads to an increase in data transmission time. In addition, the clustering approach outperforms the multihop approach by leading to shorter delays in all the investigated scenarios. Fig. 7 shows that when SNs are deployed in a confined area at a distance *d*LR from the BS, the multihop approach leads to average delays comparable to the non-cooperative scenario when *d*LR = 300m, and to better average delay performance when *d*LR increase to 500m. The trend continues with larger distances. When the BS is placed at the cell center, with the SNs deployed throughout the cell area, the non-cooperative scenario leads to better

Fig. 8 shows that the proposed cooperative methods significantly outperform the non cooperative case by leading to shorter maximum delay. Particularly, the clustering method leads to considerably shorter maximum delay compared to both the multihop approach and

Thus, the suboptimal clustering approach leads to significant energy savings that are comparable to the multihop approach as shown in Figs. 5 and 6, and it leads to much shorter delays in transmitting the measurement data as shown in Figs. 7 and 8, and thus constitutes a

> *d*LR (m) 300 500 1000 Centered BS Number of clusters 27 16 8 35

average delay than the multihop approach, but not than the clustering approach.

suitable approach leading to both energy and delay efficiency in WSNs.

2

the non-collaborative scenario.

**6.4. Bandwidth savings**

**Table 4.** Number of Collaborative Clusters for *K* = 50

4

6

8

Maximum delay (s)

10

12

14

Multihop: dLR=300m Clustering: dLR=300m No cooperation: dLR=300m Multihop: dLR=500m Clustering: dLR=500m No cooperation: dLR=500m Multihop: BS at center of 1x1km cell Clustering: BS at center of 1x1km cell No cooperation: BS at center of 1x1km cell

> In this section, we discuss some practical limitations of the proposed techniques and propose methods to overcome these limitations.

## **7.1. CSI Exchange for algorithm implementation**

In the proposed methods, the BS is assumed to be aware of the channel state information (CSI), and hence of the achievable rates *Rk*<sup>0</sup> on the LR links in addition to the CSI and rates *Rkj* (*j >* 0) on the SR links. Since the sensors considered are not assumed mobile, this can be achieved by a training phase that precedes the actual data transmission phase. The BS can know the CSI on the LR via feedback from the SNs, which is common in state-of-the-art wireless communication systems. On the SR, SNs can take turns in broadcasting pilot signals. Thus, each SN can estimate its CSI, and hence the rate *Rkj*, with every other SN within its transmission range, by measuring the received strength of the pilot signals. The SR pilot broadcasting process can be coordinated by the BS to avoid collisions. When each SN gets a CSI estimate on its SR links with the other SNs, it can feed-back this information to the BS on the LR link. After this training phase, the BS can then coordinate the data transmission process using the proposed methods. The same analysis applies in a limited mobility scenario, without necessarily having the sensors fixed. Hence, in the case of fixed SNs or in a low mobility scenario (portable SNs), the overhead due to the training phase can be considered low since a long time can elapse before the channel conditions change and the need arises to repeat the process.

In addition, it should be noted that SNs form cooperative clusters with other SNs when they can successfully hear their pilot transmission, i.e., when *Rkj* is high enough to allow efficient communication between SNs. When *Rkj* is too low between two SNs *k* and *j*, these will automatically be in different clusters. Thus, if no CSI feedback is received about the link between SNs *k* and *j*, then there will not be a possibility for direct communication between these SNs in the multihop approach of Section 5.1. Furthermore, in the clustering approach of Section 5.2, SN *k* cannot be a cluster head in a cluster of which *j* is a member and vice versa. This leads to eliminating several candidates in the search conducted in the schemes of Sections 5.1 and 5.2, and hence to a significant reduction in the complexity of the algorithms. Consequently, the results of Section 5.3 correspond to a worst-case scenario, and the complexity in practical scenarios is generally lower.
