Research Of Clustering Algorithms For Radar Sensor Networks

In order to improve the target detection performances while extending networks’ lifetime, new clustering algorithms and decision fusion approaches using constant false alarm rate(CFAR) for clustered radar sensor networks(RSN) are proposed in this paper.By analyzing the advantage of the four clustering algorithms, we finally design an optimal clustering scheme for RSN. Compared with the classical fuzzy-logic-based clustering algorithm, namely the cluster-head election mechanism using fuzzy logic(CHEF), the new proposed clustering scheme not only extends 40.94% networks’ lifetime, but also provides robust detection performances in terms of different signal-to-noise ratios(SNRs)and various false alarm rates. The novelty and contributions of this paper are four-fold:1. We propose four new clustering algorithms using fuzzy logic systems under fading environment for both single-hop and multi-hop data transmission. First of all, fuzzy c-means clustering approach is applied to cluster the RSN based on radar sensors’ locations. Then as for cluster-head selection, two fuzzy logic systems, namely two-antecedent fuzzy logic system(F2)(including radar sensors’ energy, and distance between radar sensors and the base station) and three-antecedent fuzzy logic system(F3)(including the two antecedents of F2 and the fading channel of the signal transmitted by a radar sensor to the base station), are designed to compute the likelihood of radar sensors to be elected as a cluster head at the first stage. In case of single-hop routing, the cluster-head election approach using fuzzy c-means and singular value decomposition-QR(FCMSVDQR) is proposed to decide the final cluster head. As for multi-hop routing, firstly the radar sensor with the highest fuzzy-logic-system likelihood will be elected as a cluster head.Secondly, the graph-based optimal routing selection(GORS) algorithm is proposed for intra-cluster multi-hop data transmission. In GORS, each radar sensor selects the optimal multi-hop path to the cluster head within a cluster with the highest fading coefficient. Thus, four clustering algorithms, namely the clustering algorithm using F2 and FCMSVDQR(F2&FCMSVDQR), the clustering algorithm using F3 and FCMSVDQR(F3&FCMSVDQR), the clustering algorithm using F2 and GORS(F2&GORS) and the clustering algorithm using F3 and GORS(F3&GORS) are proposed. The fading channel,to the best of our knowledge, is investigated for the first time in the clustered RSN for target detection.2. We also investigate the CFAR decision fusion approaches for RSN. In case of single-hop routing, we propose two CFAR decision fusion approaches, namely the lowSNR and likelihood-ratio-based decision fusion in the central limit theory(LLDFCLT)and the high-SNR and likelihood-ratio-based decision fusion in Kaplan-Meier estimator(HLDFKE), respectively. Their CFAR detection performances combined with two classical single-hop clustering approaches, the low energy adaptive clustering hierarchy(LEACH) and the hybrid energy efficient distributed clustering approach(HEED) are analyzed and compared in RSN. As for multi-hop routing, we propose the absolute CFAR decision fusion approach(ACFARDF) and the relative CFAR decision fusion approach(RCFARDF). In ACFARDF, all radar sensors including relay nodes make CFAR decisions. Different from ACFARDF, relay nodes make the maximum-likelihood-estimate decision in RCFARDF. The innovation of our work is that all detection radar sensors,cluster heads and the base station make CFAR decisions by using adaptive decision thresholds.Monte Carlo simulations show that: 1) compared with HLDFKE, LLDFCLT provides better target detection performances; 2) the probability of detection in RCFARDF is higher than that of ACFARDF, but the false alarm rates of relay nodes in RCFARDF are higher than those of ACFARDF at low SNRs.3. By using LLDFCLT and ACFARDF fusion approaches, we compare our four clustering algorithms with CHEF in terms of target detection and the lifetime of RSN.Our analyses are verified through extensive simulations in various false alarm rates and different cluster numbers in RSN.4. By analyzing the advantage of our four clustering algorithms, we design the optimal clustering scheme for RSN: 1) the best ratio of cluster heads to the total radar sensors in fuzzy c-means clustering approach is 0.08; 2) when the number of residual alive radar sensors(NRARS) in RSN is more than the 50% of initial radar sensors, F3&FCMSVDQR should be applied; 3) when NRARS is between 10% and 50% of initial radar sensors,F3&GORS should be used; 4) when NRARS is less than the 10% of initial radar sensors,F2&GORS should be deployed.The new clustering algorithms and the CFAR decision fusion approaches proposed in our paper not only provide great advantages in energy efficiency and target detection,but also serve as the theoretical foundation for RSN’s clustering theories and their applications.