Research On Highly-efficient Quantization For Decentralized Detection Algorithm Based On Wireless Sensor Networks

Wireless Sensor Networks(WSNs)collaboratively monitor,sense,and collect information about a target of interest with all-weather and all-time.Accurate data are transmitted to a client device after sampled information is pre-processed.Because of their huge practical application value,WSNs currently have attracted much attention from military,industrial and commercial circles.With a scarcity in the spectrum of resources and wide use of WSNs,a simple and highly-effective decentralized detection method has become increasingly significant.In this paper,in the context of WSNs,mainly two fundamental decentralized detections,location and scale parameters,are investigated.Because of stringent bandwidth/energy constraints,quantization is employed by a local sensor node.In light of their respective advantages,a Generalized Likelihood Ratio Test(GLRT)is combined with a quantization technique.Modeling transmission channel,designing detector,optimizing quantization thresholds,and solving the non-linear and non-convex optimization problems are studied.To establish a foundation for the system improvement and optimization,a decentralized GLRT detection method,based on quantized measurements,is proposed to reduce redundant data and improve system performance.Through relative theoretical study and simulation result analysis,results are summarized as follows:1.Study of GLRT detection method based on local quantized data.Because of stringent bandwidth/energy constraints,data compression is required in each sensor node.How quantization affects system performance is analyzed in detail.Analysis reveals that one-bit uniform quantization employed in local sensor nodes reduces redundant data and degrade system performance as well.Therefore,the location parameter detection problem is studied in this paper.A decentralized ’GLRT detection method,based on local multi-level quantization,is proposed.On the premise that system and computational complexity are moderately increased,the detection performance of the decentralized detection system is improved.Meanwhile,an estimation approach with higher accuracy is also obtained.The proposed method is more suitable for engineering application.2.Decentralized detection of scale parameters based on local quantized data.A detection method based on one-bit uniform quantization is invalid for analyzing binary hypothesis testing in a decentralized scale parameter detection problem.Compared with a regular quantizer,a non-regular quantizer is more suitable for a scale parameter detection problem.Therefore,a detection approach,based on local non-regular quantization,is proposed in this paper.Simulation results demonstrate that the proposed method reduces redundant data and improves system performance as well.3.Decentralized scale parameter detection based on an adaptive quantization approach.The above mentioned scale parameter detection method,based on non-regular quantization,is analyzed.Results show that this approach is prone to channel error,and the quantization threshold’s optimization relies on prior knowledge of crossover probability in distortion channels.In practice,however,prior knowledge is unknown.Therefore,non-regular quantization is employed in local sensor nodes,and a detection method,based on adaptive quantization is proposed.Simulation result demonstrate that an optimum quantization threshold can be computed and estimated by sensor nodes,based on information shared with each other,thereby improving system performance.Because prior knowledge is not required,the proposed method is more suitable for practical applications.4.Decentralized detection with a Rao test suitable for non-Gaussian noise.In the context of non-Gaussian noise,a sufficient condition for one-bit optimum quantization is theoretically explained,and the research problem is addressed by providing a counter example.Statistical characteristics of non-Gaussian noise and the influence on optimum quantization thresholds are analyzed.Because detection of a Maximum Likelihood Estimate(MLE)of an unknown parameter is difficult,a Rao detection method,based on quantized data,is proposed in this paper.Compared with the GLRT detection approach,the proposed method does not require a MLE,thereby leading to simple structure and implementation.The optimum quantization structure in Gaussian and non-Gaussian noise is analyzed.Simulation results demonstrate that,compared with other methods,the proposed method performs better.