Forwarding Zone enabled PSO routing with Network lifetime maximization in MANET

Abstract

Network lifetime maximization is one of the most sought after issues in Mobile Adhoc Networks (MANETs). Whereas, due to geographical routing based approaches, the packet transmission becomes more suitable in dynamic environment such as MANETs. Direct heuristics are not suitable in such scenarios to provide desired solution as the problem becomes NP-hard in dense networks, thus researchers focused to utilize meta-heuristic techniques. Particle Swarm Optimization (PSO) is one of the most effective meta-heuristic techniques to solve such problems with near optimal solution. However, meta-heuristic techniques (PSO) become slow in convergence and require more computational time when network size increases. Therefore, in this work, PSO is adaptively modified (APSO) to best fit in our scenario, and re-enforced using Forwarding Search Space (FSS) heuristic technique to overcome the PSO’s convergence and computational time related issues, significantly improves the performance of PSO. In FSS, a Forwarding Zone (FZ) is selected between source and destination such that the optimal solution lies in that area and APSO is applied for an effective routing in FZ area instead of complete network. To utilize the complementary characteristics of both (APSO and FSS), a hybrid FZ-APSO is proposed for routing in dense network with minimum delay and energy consumption in order to increase the lifetime of the network. Comparative simulation results evidenced that the proposed FZ-APSO routing algorithm significantly improved the performance of the routing in terms of energy consumption, end to end delay, computational time and network lifetime.

Appendices

Appendix A

When minimum and maximum requirements of data (energy and delay) are collected, respective current normalized value of energy ($E_{cons_{i}}$) and delay (Del_i) are calculated as $\frac {E_{i} - E_{min}}{E_{max}-E_{min}}$ and $\frac {D_{i} - D_{min}}{D_{max}-D_{min}}$ respectively. Function F(Del_i,E_i) is given by, $d_{i}^{min},d_{i}^{max}$ and $E_{i}^{min},E_{i}^{max}$ denotes the minimum and maximum ranges of delay and energy consumption per node. This function F helps the particles to find the balancing point in normalized data values ($E_{cons_{i}}$, Del_i) by adjusting their positions such that the fitness is minimized in the result over different course of iterations. To clearly see the affect of the proposed function, a random set of of energy consumption (0.5011, 2.5005, 4.0002, 0.001, 1.5007, 3.5003, 5.0, 4.5001, 2.0006, 3.0004, 1.0008) and delay (3.07, 5.05, 10, 2.08, 1.09, 4.06, 0.1, 9.01, 7.03, 8.02, 6.04) is taken where $E_{i}^{min},E_{i}^{max}$ and $d_{i}^{min},d_{i}^{max}$ are 0.001, 5 and 0.1, 10 respectively. After applying min-max normalization [52] and arranging the normalized values in ascending order gives the set [0.0, 0.1, 0.2, 0.3, 0.4 0.5, 0.6, 0.7, 0.8, 0.9, 1.0] and the resultant fitness is obtained with it. Figure 13 shows the fitness value for these normalized valued of energy and delay. It is clear from Fig. 13 that value of fitness function is minimum when both input variables are almost equal (0.5 and 0.5).

Fig. 13

Fitness function for normalized energy consumption and delay

Appendix B

For mathematical analysis of the proposed method, 10 nodes are considered where each has 8 dimensions. dimension(k) may vary for different S and D pairs with different path lengths. Number of dimensions for all particles is same which is equal to the maximum number of available paths between source and destination. It is illustrated clearly in Fig. 3 where random nodes are randomly generated in the FZ region of network. Let us assume that after calculating FZ between S and D, there are 100 intermediate nodes between them. Particles are initialized in that region by assigning the coordinates of nodes to the particles with a uniform random number in [0, 100] interval. Lets assume, in initial iteration position and velocity of particle 1 (P₁) is generated with the help of (4) and (5) which is shown in Table 1. Before applying APSO on the generated particle, SPV and VPG operators are applied to avoid the processing on invalid particles. It is given in Table 2.

Initially, each particle itself is local best (p_best) and let us assume that particle 4 (P₄) is global best g_best. If position of P₄ is (5.26, 1.58, 4.62, 2.74, 5.12, 4.65, 1.75, 1.4), then using (9) and (10), updated position of particle 1 in each dimension will be according to (16–17) where c₁ = c₂ = 2 and w = 0.6

For dimension 1,

$$\begin{array}{@{}rcl@{}} {V_{1}^{1}}(t + 1) &=& 0.6\times0.12 + 2\times0.4\times(4.12 - 4.12) \\&&+ 2\times0.7\times(5.26 - 4.12) = 1.668 \end{array} $$

(16)

$$ {X_{1}^{1}}(t + 1) = 4.12 + 1.668 = 5.788 $$

(17)

In the same way, velocity and position updation will be performed for all dimensions of particle. During iterations, these position and velocity of all particles will be performed similarly. For simplicity, updated position of particle have almost same location of a MANET node and resultant g_best’s position will be assumed to be the path between S and D.

Appendix C

In VPG operator invalid path is converted into valid one via swapping suspecting elements of path between S and D. During iterative updatios of position and velocity vectors of particles, there is a high probability that the sequence of the paths can be invalid. Thus VPG operator is again applied to convert invalid path into valid path. A representation of the resultant path of after 1st iteration of the above discussed example with SPV and VPG calculation is shown in Tables 15 and 16 respectively.

Table 15 Output of SPV operator in iteration 1

Table 16 Output of VPG operator in iteration 1