Decentralized Detection Based On Wireless Sensor Networks

Wireless sensor networks(WSNs)have received considerable attention given their wide applicability to military reconnaissance,environmental monitoring,healthcare,and etc.One of the main tasks for a WSN is distributed detection,which has the chance to detect a noncooperative target,spectrum vacant or noncooperative signal efficiently and robustly.This paper considers decentralized detection under the assumptions of parallel access channel for information transmission and no feedback from fusion center.In this scenario,decentralized detection methods are proposed under fixed-sample-size(FSS)and sequential test setup,respectively.The main contributions of this paper are summarized as follows:(1)For the decentralized(active)detection of a noncooperative target,the energy of the reflected signal from the target may be aspect and distance dependent.In this scenario,the decentralized fixed-sample-size(FSS)detection scheme based on uniform quantization and generalized probability ratio test(GLRT)is proposed.Firstly,the quantization thresholds of sensors are optimized by maximizing the Kullback-Leibler divergence between the alternative and the null hypotheses,which improves the detection performance.Then,the proposed decentralized detection scheme is extended to the scenario where sensor data are transmitted over binary symmetric channels(BSCs).Finally,an adaptive quantization method is proposed to reduce the communication overhead.(2)The decentralized sequential probability ratio test(SPRT)based on level-triggered sampling(LTS)is studied.Firstly,we derive the analytical expression of error probability,and design the SPRT deciding thresholds based on it.Compared with the existing Wald's Identity,the derived analytical expression describes the error performance more accurately.Hence,under certain error probability constraints,the designed SPRT deciding thresholds are more effective in reducing the decision delay.Then,we analyze the influence of nonideal channel on the LTS-based decentralized SPRT scheme,and provide a decentralized SPRT scheme based on the bit error rate(BER)of data transmission.(3)For the decentralized detection of noncooperative Gaussian random signal,we propose the decentralized truncated sequential(TOS)test scheme based on improved-LTS.Firstly,we design the improved-LTS method for the pre-processing of sensor observations,then we derive the generalized likelihood ratio(GLR)statistic and propose a TOS test rule for making the final decision.Compared with the FSS test rule,the proposed TOS test significantly accelerates the detection process under alternative hypothesis.Meanwhile,the TOS test avoids long decision delay.On the other hand,compared with decentralized TOS test based on conventional uniform sampling and one-bit quantization,the proposed improved-LTS-based TOS test scheme significantly lowers the communication overhead,and greatly improve the detection performance,which is owning to the adaptive sampling mechanism of the improved-LTS technique.In order to avoid the computation of complexed GLR statistic,we propose a new decentralized TOS test scheme based on the generalized locally optimum detection(G-LOD)statistic.(4)For the decentralized detection of noncooperative deterministic signal,we re-design the improved-LTS and derive the GLR statistic,then we propose the corresponding decentralized TOS test scheme.Consider that the GLR statistic is complex in computation,we further propose a decentralized TOS test scheme based on the generalized Rao(G-Rao)test statistic.