Online Spike Sorting

The voltage threshold determines which extracellular events enter the sorting pipeline. A threshold set too low can admit noise and increase incorrect assignments, whereas a threshold set too high may exclude genuine spikes. Because online analysis guides decisions during acquisition, threshold selection directly affects the reliability of estimated firing activity and the quality of subsequent closed-loop responses.

Waveform features provide measurable characteristics that help distinguish spikes produced by different neurons. After an event is detected, these features support clustering or classification, allowing the system to associate similar waveforms with a putative neuronal source. When waveforms overlap or differ only subtly, feature quality can strongly influence whether population activity is separated accurately.

Both approaches assign detected spikes to neuronal groups, but they organize that assignment differently. Clustering groups events according to similarities in extracted waveform features, while classification assigns events using established categories. In online spike sorting, either strategy must operate rapidly during acquisition, since delayed processing would limit its usefulness for adaptive stimulation, brain-computer interfaces, or other immediate experimental decisions.

Noise, waveform overlap, and changes in signal quality can make spike assignments less stable during an experiment. These conditions may alter the features used to distinguish neuronal sources or introduce events that resemble genuine spikes. Continuous validation is therefore important: it helps identify declining sorting quality before inaccurate firing estimates influence closed-loop control or interpretation of population activity.

The workflow begins with extracellular voltage acquisition, followed by detection of events that cross a selected threshold. The system then extracts waveform features and assigns each event to a neuronal unit through clustering or classification. Ongoing validation checks whether assignments remain reliable as noise, waveform overlap, or signal-quality changes affect the incoming recording.

Researchers need this approach when neuronal activity must influence an experiment while recording continues. It can provide rapid firing estimates for closed-loop experiments, brain-computer interfaces, neuroprosthetic control, and adaptive stimulation. The immediate output allows researchers or devices to respond to detected activity, rather than postponing decisions until recordings have been processed after acquisition.

Sorted output provides estimates of firing from individual neuronal sources and can be combined to examine population activity. In neuroscience experiments, these estimates help reveal how groups of neurons behave during ongoing acquisition. Their value depends on assignment reliability, so interpretation should account for noise, overlapping waveforms, and changes in signal quality that may affect unit separation.