One-Bit Robust Parameter Estimation For Sensor Networks Under Non-Ideal Channels

Wireless Sensor Networks(WSNs)are usually composed of small,integrated and communication-capable nodes randomly distributed in a certain space.These nodes can collect and process the physical parameter information of the target object.As an important information acquisition technology,WSNs enables the objective physical world to be perceived.With the development of wireless communications,microprocessors,and electronic information technologies,WSNs have a wide range of applications in the fields of military defense,industry,agriculture,and medical.In the parameter estimation problem based on sensor network,each sensor node has limited energy storage,computing power and bandwidth resources.Direct transmission of the sampled signal value of the node will lead to large energy consumption and communication load.The common solution is to compress the sampling signal of each node into 1-bit information value before transmission.Therefore,many algorithms based on 1-bit quantization are proposed,such as 1-bit compression sensing and 1-bit tracking location algorithm.Although these methods have their own advantages,they usually only consider additive noise and seldom consider multiplicative noise.In fact,multiplicative noise often occurs,such as multipath channel.Moreover,most of algorithms assume that the channel between sensor nodes is ideal.However,the channel is also susceptible to random noise interference.Therefore,the problem of 1-bit parameter estimation over non-ideal channel and multiplicative noise is studied in this thesis.For the parameter estimation over centralized sensor networks,the 1-bit compressed value by the node is transmitted to the Fusion Center for centralized processing,and the maximum likelihood is used for parameter estimation.Since the likelihood function contains hidden variables,the algorithm considers the combination of the EM framework to solve this problem,and continuously updates the parameter estimates through on iteration method to obtain the optimal solution.In addition,since adaptive algorithm has lower computation and storage requirements,and has better performance when tracking time-varying parameters,a centralized adaptive parameter estimation algorithm is proposed according to Bayesian criterion.Finally,note that a centralized network is vulnerable to external attacks or node failures,and this may cause the entire network to be paralyzed.In contrary,the distributed sensor networks have better fault tolerance.Therefore,a distributed 1-bit parameter estimation algorithm based on Bayesian criterion is proposed in ATC and CTA modes respectively in distributed networks.Through a series of simulation experiments,this thesis compares the proposed algorithm with other algorithms,and verifies the effectiveness and robustness of the proposed algorithm under the condition that the quantization threshold deviates from the optimal quantization threshold.At the same time,the experimental results also show that the proposed adaptive algorithm has good adaptability to time-varying parameters.The estimation accuracy of the distributed parameter estimation algorithm also has better estimation accuracy,and the distributed network model does not depend on a single node and has stronger robustness.