Multi-unit activity (MUA) is the combined spiking output of all neurons detectable on a single electrode channel, measured by counting threshold crossings of the extracellular voltage signal without attempting to assign spikes to individual neurons. MUA provides a simpler and more robust signal than isolated single units, and it has become increasingly important in clinical BCI systems.
Extraction
MUA is extracted by:
- Bandpass filtering the raw neural signal to the spike band (typically 300-5000 Hz)
- Setting a voltage threshold (commonly -3.5 to -4.5 times the RMS noise level)
- Counting threshold crossings per time bin (e.g., per 20 ms)
The resulting threshold crossing rate reflects the collective firing activity of all neurons within the electrode's detection radius (roughly 100-150 micrometers).
MUA vs. Single Units in BCI
A key insight from BCI research is that MUA often provides decoding performance comparable to carefully sorted single units for many tasks. The Kalman filter decoders used in BrainGate and Neuralink systems typically operate on threshold crossing rates rather than sorted single-unit spike trains. This has important practical implications:
- No spike sorting required: Eliminates a computationally expensive and error-prone processing step
- More robust to electrode drift: As electrodes shift relative to neurons, the set of detected neurons changes, but the aggregate MUA remains informative
- Easier to maintain long-term: Even as individual neurons become undetectable due to glial scarring, MUA from remaining nearby neurons still drives effective decoding
Limitations
MUA sacrifices the information carried by individual neuron identities and precise spike timing. For tasks requiring fine-grained neural population analysis — such as studying neural population dynamics or performing high-resolution sensory decoding — single-unit isolation remains preferable. However, for the practical goal of real-time BCI control, MUA represents an excellent tradeoff between signal quality and robustness.