Probability Based Adaptive Invoked Clustering Algorithm in MANETs

A mobile ad hoc network (MANET), is a self-configuring network of mobile devices connected by wireless links. In order to achieve stable clusters, the cluster-heads maintaining the cluster should be stable with minimum overhead of cluster re-elections. In this paper we propose a Probability Based Adaptive Invoked Weighted Clustering Algorithm (PAIWCA) which can enhance the stability of the clusters by taking battery power of the nodes into considerations for the clustering formation and electing stable cluster-heads using cluster head probability of a node. In this simulation study a comparison was conducted to measure the performance of our algorithm with maximal weighted independent set (MWIS) in terms of the number of clusters formed, the connectivity of the network, dominant set updates,throughput of the overall network and packet delivery ratio. The result shows that our algorithm performs better than existing one and is also tunable to different kinds of network conditions.

💡 Research Summary

The paper introduces a novel clustering scheme for mobile ad‑hoc networks (MANETs) called the Probability Based Adaptive Invoked Weighted Clustering Algorithm (PAIWCA). The authors begin by reviewing existing clustering approaches—Highest‑Degree, Lowest‑ID, Node‑Weight, Weighted Clustering Algorithm (WCA), and Maximal Weighted Independent Set (MWIS)—and point out their shortcomings, especially the high overhead of recomputing node weights during clustering and the rapid depletion of cluster‑head (CH) batteries. WCA, while more sophisticated, still requires the weight of every node to be known only after the clustering process has started, leading to frequent re‑elections when new nodes appear or when a CH’s energy falls low.

PAIWCA addresses these issues in two phases: (1) clustering setup and (2) clustering maintenance. In the setup phase each node independently calculates a weight Wv before any clustering begins. The weight is a linear combination of five parameters: transmission range (Tr), transmission rate (Tx), average mobility (Mv), power consumed (Pv), and a newly introduced Cluster‑Head Probability (ChProb). The formula used is

 Wv = w1·Tr + w2·Tx + w3·Mv + w4·Pv – ChProb

with experimentally chosen constants w1 = 0.2, w2 = 0.2, w3 = 0.05, w4 = 0.05. ChProb itself is defined as

 ChProb = Cprob·(Eresidual / Emax) + Tr

where Eresidual is the node’s current residual energy, Emax is the maximum possible energy (identical for all nodes), and Cprob is a scaling factor. A minimum threshold Pmin (e.g., 10⁻⁴) prevents a node with too low energy from becoming a CH. Because all nodes already know their Wv, the algorithm simply selects the node with the smallest weight as the CH; its neighbors are then excluded from further elections. This process repeats until every node belongs to a cluster.

The maintenance phase handles mobility‑induced changes. When a regular node moves out of its cluster’s boundary, it searches for a neighboring CH. If it finds one, it joins that cluster. If the node cannot hear any CH, it declares itself a CH. When a node arrives in a region already occupied by a cluster, its ChProb is compared with the current CH’s ChProb. If the newcomer’s probability is higher, it instantly replaces the existing CH without triggering a full re‑clustering. Only when a CH’s ChProb falls below Pmin does the algorithm invoke a full re‑election using the same weight‑based selection.

The authors evaluate PAIWCA using NS‑2 simulations with 10–300 nodes placed in a 500 × 500 m area, transmission ranges from 5 to 200 m, node speeds from 10 to 100 m/s, and random waypoint mobility. They compare three metrics against WCA and MWIS: (i) number of clusters formed, (ii) network connectivity (probability that any node can reach any other), (iii) dominant‑set updates (frequency of CH changes), (iv) overall throughput, and (v) packet delivery ratio (PDR). Results show that PAIWCA consistently forms fewer clusters, achieves near‑100 % connectivity as transmission range grows, reduces dominant‑set updates dramatically (indicating fewer CH changes), and yields higher, more stable throughput. Moreover, PAIWCA maintains a higher PDR under varying pause times (i.e., different mobility levels) than the baseline algorithms, demonstrating robustness to node movement.

Despite these promising results, the paper has several limitations. First, it assumes a static topology during the clustering setup, which may not hold in highly dynamic MANETs where partitions can appear abruptly. Second, the choice of weight coefficients (w1–w4) and the threshold Pmin is not justified beyond empirical tuning; a sensitivity analysis is missing, leaving open the question of how the algorithm behaves under different parameter settings. Third, the simulation environment does not model realistic wireless channel effects such as interference, fading, or variable packet loss, which could affect the reliability of the reported throughput and PDR gains. Fourth, the computational complexity of PAIWCA is described qualitatively (fewer re‑elections) but not quantified; analytical bounds on time and space complexity would be valuable for assessing scalability to very large networks. Finally, practical deployment would require each node to continuously monitor its transmission range, rate, mobility, and residual energy, which may introduce measurement overhead not accounted for in the study.

In summary, PAIWCA contributes a thoughtful extension to weighted clustering by pre‑computing node weights and leveraging a probabilistic CH selection metric that incorporates residual energy. This design reduces the frequency of costly re‑clustering, balances energy consumption among CHs, and improves key performance indicators in simulated MANET scenarios. Future work should explore adaptive tuning of the weight parameters, robustness under rapid topology changes, and validation on real hardware or more sophisticated network simulators that capture physical‑layer dynamics.