Extracellular recordings have been used as a tool to study brain activity. The signals recorded, however, are contaminated by noise from the recording equipment as well as the activity of non-targeted biological structures, calling for robust spike detection algorithms. Such algorithms are even more critical for the analysis of the ever more popular multi-sensor e.g. tetrode recordings.;The generalized likelihood ratio test (GLRT) is the theoretical optimal detector for deterministic signals submersed in random noise of known distribution. While the multi-linear expansion of the GLRT detector may be applied for spike detection, it may suffer from poor estimation accuracy due to under-sampling with respect to the increasing number of the parameters, as well as high computation times unsuitable for real-time implementation. Special cases of the GLRT detector utilizing a fewer number of parameters are considered along with the general form. |