Neural encoding is the complement of neural decoding. While decoding asks "given this neural activity, what was the person doing or intending?", encoding asks "given this stimulus or behavior, what neural activity pattern does the brain produce?" Understanding neural encoding is foundational for BCI design — it reveals what information is available in neural signals and how best to extract it.
Encoding Models
Rate Coding
The simplest encoding model: information is represented by the firing rate of individual neurons. A motor cortex neuron that fires at 50 spikes/second during rightward arm movement and 10 spikes/second during leftward movement encodes movement direction via its firing rate. Rate coding is the basis for population vector decoders in motor BCIs.
Temporal Coding
Information is encoded in the precise timing of spikes rather than (or in addition to) firing rate. Temporal coding can carry more information per spike than rate coding and is important in auditory and somatosensory processing. However, temporal codes are harder to decode and more sensitive to noise.
Population Coding
Information is distributed across populations of neurons rather than carried by individual cells. The population code — the collective pattern of firing rates across many neurons — represents high-dimensional variables like hand position, velocity, and intended speech phonemes. Modern BCI decoders operate on population-level representations.
Encoding for Sensory BCIs
Understanding sensory encoding is critical for the "write" side of bidirectional BCIs. To deliver artificial touch sensation via intracortical microstimulation (ICMS), researchers must understand how somatosensory cortex naturally encodes touch pressure, texture, and location — then design stimulation patterns that mimic these natural encoding patterns. The Bensmaia lab at the University of Chicago has mapped somatosensory encoding in detail to guide biomimetic ICMS design.
Encoding Models in BCI Development
Encoding models inform decoder design. If motor cortex neurons encode velocity (not position) of intended movement, then decoders should estimate velocity from neural activity and integrate to get position. If speech cortex encodes articulatory gestures (tongue, jaw, lip movements) rather than abstract phonemes, then speech BCI decoders should target articulatory features. The accuracy of the underlying encoding model directly affects BCI decoding performance.