改进的格拉布斯准则在信号检测门限估值中的应用

观测数据中野值（信号或干扰等）的存在将会影响信号检测门限估计的准确度和可信度，直接应用格拉布斯准则仅能有效地处理观测数据中野值占有率小于１０％时的情况，本文根据门限估计的需求，对该准则作了必要的改进，使之能够适应野值可能较多时的情况，改进后的准则将野值占有率提高了近１６％，，即达到２７％，此外，在野值的除过程中，野值的位置信息也可同时获得，为门限检测提供了信号所处的位置信息。

Application of Improved Grubbs' Criterion to Estimation of Signal Detection Threshold

The wild values, such as signals or disturbances, existing in observed data would affect the accuracy and reliability of estimation of signal detection threshold. When the ratio of number of wild values to that of total data is less than 10%, using the Grubbs' criterion could effectively eliminate them. In older to meet the actual requirements for threshold estimation, this paper makes some necessary improvements on this criterion to adapt itself to the situation of more wild values. The improved criterion would achieve about 16% of increasing over the former ratio, i. e. it may be around 27%. Besides, because the position information of those eliminated wild values could be also obtained at the same time during processing them, it is very convenient to make signal detection more effective.