Quality metrics to accompany spike sorting of extracellular signals - PubMed (original) (raw)

Figure 1.

Signals and signal conditioning steps in the acquisition of extracellular data. A, The extracellular signal from one microwire of a tetrode, positioned near a hippocampal CA1 neuron in vivo (Henze et al., 2000), followed by a schematic of the signal flow. Each box refers to a specific process that is implemented by an electronic or computational unit; see Ganguly and Kleinfeld (2004), their supplemental material, for details of the circuit. All extracellular signals are referenced to an unrelated region of cortex and impedance-buffered at the head of the animal. We use N-channel field effect transistors (SST4118; Vishay Siliconix) with matched transconductances that were configured as common source transconductance amplifiers; the input noise of these transistors does not exceed the noise generated by the microelectrode. The transistors provide noise immunity as well as current to drive flexible cables that connect the head stage to precision load resistors, followed by 1 Hz high-pass single-pole filters. We measured the difference between each filtered channel and the filtered reference signals with a field effect transistor-input instrumentation amplifier with a fixed gain of 100 (v/v) (INA101, Burr Brown, Texas Instruments). The amplified signal then passes through a commutator, if applicable, and is high-pass filtered (4-pole Bessel filter configured around a no. UAF42, Burr Brown), further amplified (no. OPA101, Burr Brown), anti-alias filtered (8-pole constant-phase low-pass filter; no. D68L8D—10.0 kHz, Frequency Devices), and finally digitized (16 bit digitizer, NI-6251, National Instruments). All numerical operations were performed with programs written in MATLAB (The MathWorks). B, The complex impedance of a 25-μm-diameter microwire electrode, cut at 45° and beveled on a diamond grinding wheel, as a function of frequency, together with the measured electrode noise. The measured curves have a phenomenological falloff as |Impedance| ∝ f−0.78, where f is the frequency. The measured noise is close to the expected value for thermal Johnson noise, which varies as 4kBTŹResistive for a 1 Hz bandwidth, where k_B_T ≅ 100 meV is the thermal energy. _Z_Resistive, The spectral composition of the spike signal averaged over 151 waveforms from the same single unit. The bands have 0.95 confidence limits. Part of the decrement near 10 kHz is caused by the low-pass filter. Also shown are the fit to the electrode noise and the specified amplifier noise.