I’ve recently found myself in a scenario where we needed to measure extracellular neural and muscle potentials. Oftentimes, the amplifiers are integrated into an ASIC (for example Intan chips, Neuropixels, Cirtec chips, or others). However, sometimes it’s useful to be able to build a discrete-component amplifier for prototyping purposes. Here’s such a design.
The basic design is passive high-pass filters on both inputs to an instrumentation amplifier, with the amplifier inputs DC-biased to half the rail voltage. Each input is capacitively coupled, which achieves objectives 2 and 3 above. To present a path for input bias current, and to ensure the input voltages stay within the rails of the amplifier, we use a split-resistor design with 100k resistors. This biases the inputs to mid-range, which is necessary because the instrumentation amplifier will not necessarily behave appropriately if inputs go outside its supply voltages. Instrumentation amplifiers are provide a straightforward way to DC bias their outputs, so that if the inputs were shorted together (exactly equal), the amplifier would produce a given DC offset (spec 4).
The instrument amplifier chosen here is a INA333 instrumentation amplifier. At a gain of 100 (spec 5), it has a bandwidth of 3.5 kHz, barely missing spec 6. This amplifier draws 50uA quiescent current (meeting spec 7 at 1.8V). Noise on the INA333 is spec’d at 50 nV/sqrt(Hz), which is about 3 uV for a 3500 Hz bandwidth, and 3.5 uV at 5 kHz, consistent with spec 8. However, total noise will also incorporate the thermal noise of the split resistor, which is ~2.9uV RMS at room temperature at 5kHz.
One shortcoming of this design is that the split resistor design presents inherent tradeoffs: larger resistor values dissipate less energy and create a larger input impedance, but increase thermal noise.