Low-noise neural amplifier

The low-noise amplifier is the first active stage, tasked with lifting microvolt-scale signals — tens of µV for extracellular spikes, single µV for LFP — above the transistors' own thermal and flicker noise, while rejecting a DC electrode offset that can be a thousand times larger than the signal. The canonical topology is capacitively-coupled with a resistive-feedback pseudo-resistor (the Harrison amplifier): the midband gain is set by a capacitor ratio (typically around 40 dB), a sub-hertz high-pass corner is set by an enormous pseudo-resistor, and a low-pass rolls off above the spike band near a few kHz.

Its dominant constraint is the thermal-noise floor of the input transistor, whose input-referred noise falls only as one over the square root of transconductance, and transconductance rises with bias current. So halving the input noise costs roughly four times the current — the tradeoff formalised by the noise-efficiency factor. Flicker (1/f) noise dominates at the low frequencies where LFP lives and is fought with large input devices, chopping or auto-zeroing.

Everything downstream inherits this stage's noise, so the LNA sets the achievable signal-to-noise ratio for the whole channel. In a thousand-channel implant its per-channel current, multiplied across all channels, is often the single largest contributor to the power (and therefore heat) budget.