Dobie R A, Wilson M J
Department of Otolaryngology-Head and Neck Surgery, University of Texas Health Science Center at San Antonio.
Electroencephalogr Clin Neurophysiol. 1993 Nov-Dec;88(6):516-24. doi: 10.1016/0168-5597(93)90040-v.
Several different and related measures have been proposed for objective response detection in the frequency domain. We compared magnitude-squared coherence (MSC) to phase coherence (PC) using simulations with specified signal-to-noise ratios (SNRs) and varying numbers of subaverages; the performance measure was area under a receiver operating characteristic (ROC) curve. MSC was superior to PC; test time required for equivalent performance is about 3 times greater for PC than for MSC. MSC performance for a given final SNR increased with the number of subaverages, but reached a plateau at 16 subaverages. Simulations of noise non-stationarity (high-amplitude noise in some subaverages compared to the others) led to decreased performance advantage for MSC over PC. However, weighted averaging restored this advantage. MSC is shown to be a simple algebraic transform of Victor and Mast's (1991) "circular T2" statistic and of two earlier statistics; all have identical statistical power.
针对频域中的客观反应检测，已经提出了几种不同但相关的测量方法。我们通过使用具有指定信噪比（SNR）和不同子平均数量的模拟，将幅度平方相干（MSC）与相位相干（PC）进行了比较；性能指标是接收器操作特性（ROC）曲线下的面积。MSC优于PC；对于等效性能，PC所需的测试时间大约是MSC的3倍。给定最终SNR时，MSC的性能随着子平均数量的增加而提高，但在16个子平均时达到平稳状态。噪声非平稳性的模拟（某些子平均中的高幅度噪声与其他子平均相比）导致MSC相对于PC的性能优势降低。然而，加权平均恢复了这种优势。结果表明，MSC是Victor和Mast（1991）的“圆形T2”统计量以及另外两个早期统计量的简单代数变换；所有这些统计量都具有相同的统计功效。
