I. Neural recording is a mixture

When a microelectrode is inserted into brain tissue, the voltage trace it records contains multiple frequencies simultaneously. These components can be separated through signal-processing techniques. This concept is analogous to a radio: while many stations broadcast simultaneously, they can be isolated by tuning into the specific frequency bands in which they transmit information.

II. Signal Separation: Individual vs. Population Activity

Unlike EEG, which captures distant signals from the scalp, a microelectrode inserted directly into brain tissue sits in close proximity to neurons. As a result, the recorded signal represents a mixture of activity at two distinct scales: single-unit spikes from individual neurons and aggregate population activity.

To make sense of this, consider the electrode as a radio receiver. Just as a radio picks up a chaotic mix of signals where each station occupies a distinct frequency band, the raw neural recording contains multiple physiological phenomena operating at distinct frequencies. By applying digital filters, we first separate this raw trace into two fundamental components:

  1. Low-Frequency (<300 Hz): Represents LFP (Local Field Potential).
  2. High-Frequency (>300 Hz): Represents Spikes (Single-Neuron Action Potentials).

The following table highlights the fundamental differences between these two signal types.

Feature Local Field Potential (LFP) Single-Neuron Spikes
What it Measures Summed electrical activity from many nearby neurons. Rapid action potentials from individual neurons.
Physiological Origin Synaptic Inputs: Current flowing into/out of dendrites (subthreshold activity). Neuronal Outputs: Voltage-gated sodium channel firing (action potentials).
Frequency Range Slow: 1 – 300 Hz Fast: 300 – 5000 Hz (Events ≈ 1 ms)
Spatial Range Wide: ~0.5 – 1 mm radius (hundreds of neurons). Narrow: ~50 – 100 $\mu$m radius (very close to tip).
Information Content Network-level dynamics, oscillations (Theta, Gamma), population coordination. Precise firing rates, spike timing, stimulus selectivity, neural coding.
Analysis Methods Power spectra, Coherence, Phase synchronization. Spike sorting, Tuning curves, Temporal coding.

A Note on Frequency Scales: To intuitively grasp these timescales, consider that even 50 Hz is perceived as continuous. A four-cylinder car engine running at 3,000 RPM cycles at exactly 50 Hz (Gamma band). At this speed, you hear a continuous “hum” rather than discrete engine strokes. The 300 Hz cutoff is six times faster than this engine hum, effectively separating the slow, rhythmic background (LFP) from the extremely rapid, discrete events of neuronal firing (Spikes).

III. High-Frequency Component (Spikes)

  • Source: When a specific neuron close to the electrode fires, it creates a brief, high-amplitude voltage deflection.
  • Processing: Extracted using a High-Pass Filter (e.g., cut-off at 300 Hz).
  • Analogy: The voice of the person sitting right next to you. You hear their specific words and timing distinct from the background roar.

IV. Low-Frequency Component (LFP)

  • Source: As neurons receive input, currents flow into dendrites. These currents overlap and add up spatially, producing a smooth, slow-changing signal.
  • Processing: Extracted using a Low-Pass Filter (e.g., cut-off at 300 Hz).
  • Analogy: The roar of the crowd in a stadium. You hear the collective intensity and mood (cheering vs. booing), but not individual conversations.

1. Brain Rhythms: The Radio Analogy

Within the LFP signal, distinct cognitive processes utilize different frequency ranges. This is analogous to a radio: multiple stations broadcast simultaneously, but you can isolate specific information (news, jazz, rock) by tuning into specific frequency bands.

These bands are not arbitrary; they result from specific neurobiological mechanisms (such as synaptic decay times and signal transmission delays). Note that the bands increase logarithmically in width.

Band Frequency Associated States (General)
Delta 2 – 4 Hz Deep sleep, anesthesia.
Theta 4 – 8 Hz Memory processing, navigation (Hippocampus), drowsiness.
Alpha 8 – 12 Hz Visual idling, inhibition of irrelevant regions.
Beta 15 – 30 Hz Motor control, maintenance of cognitive state.
Lower Gamma 30 – 80 Hz Feature binding, active information processing.
Upper Gamma 80 – 150 Hz High-level cognitive functioning.

Note: Boundaries are not precise (e.g., Theta can vary from 3–9 Hz) and depend on individual differences like age and brain structure.

VI. The Time-Frequency Uncertainty Principle

A major challenge in analyzing these rhythms is the trade-off between knowing when something happened and what frequency it was.

1. The Heartbeat Analogy

Consider measuring a heartbeat.

  • To measure a heart rate (Frequency), you need to count beats over time (e.g., 5 seconds). You cannot measure a “rate” in a 1-millisecond instant.
  • If you shorten the window to capture a “transient” change, you lose precision on the exact rate.

2. The Trade-off

  • High Temporal Precision: Allows you to identify exactly when an event occurred, but makes it difficult to distinguish 20 Hz from 20.5 Hz.
  • High Frequency Precision: Allows you to distinguish exact frequencies, but smears the event over time (you can’t tell if it started at 200ms or 230ms).

3. Interpretation Warning

When a paper reports “Activity at 15 Hz at 346 ms,” it is a statistical abstraction.

  • “15 Hz” actually means a weighted range where 15 Hz contributed maximally.
  • “346 ms” actually means a weighted sum of activity from preceding and succeeding time points.