ABSTRACT
Neuroscience stands at the threshold of a significant expansion owing to the clarity of propitious avenues and the availability of adequate approaches for understanding brain function. Indeed, guiding concepts, experimental techniques, and analytical tools are available for progress in many fields ranging from ion-channels to neural systems studies. In the investigation of sensory and cognitive processes, one of the prerequisites is instrumentation for obtaining reliable recordings of the physiological activity of neuronal assemblies. The concurrent activity of several neurons can be observed with an extracellular electrode that records the action potentials of neurons in the vicinity of the tip. The main advantage over an intracellular electrode is the ability to record from more than one neuron at the same time. The extracellular electrode also allows recording without damaging the neurons and enables longer recording periods. The cost of these benefits is the requirement for sorting the interleaved neural spike trains of several neurons to determine the firing instants of individual neurons. Due to differences in their geometry and the impedances connecting them to the electrode, the depolarization of different neurons is manifested with different transient waveforms in the recording. However, typically, the waveform of a given neuron preserves its general shape during a recording period. Therefore, the activity of individual neurons can be determined by sorting the different types of neural waveforms. An additional challenge in extracellular recordings is the relatively low signal-to-noise ratios (SNR) that can occur in many cases. The background noise is mainly due to the activity of a large number of distant neurons resulting in a considerable overlap between the spectra of waveforms of interest and noise. The neural waveform recognition problem is a typical example of detection and classification of transient patterns embedded in colored noise.
