Trimmed robust fusion under non-gaussian channel noise

In distributed multisensor fusion, local estimates may have to be communicated to a distant central processor. Hence, the communication channel noise is an important factor to the fusion algorithm. Optimal linear methods can be applied when channel noise is supposed to be Gaussian. In practice, the channel noise is not Gaussian and usually modeled by a contaminated Gaussian distribution. A two-stage trimmed robust fusion (TRIMRF) algorithm is proposed to adapt an optimal method to this case. The advantages of this method are its simplicity and capability of working effectively with little assumption about the underlying channel noise distribution.