There are many comparators that you can buy to provide quick (sub-microsecond) response when a large signal changes and crosses a threshold voltage, such as a reference voltage. Unfortunately, comparators don't work well when the input signals are very small. The ability to respond correctly, without offset, drift, or noise, is normally impossible to do with a comparator unless the signal is moving more than a millivolt beyond the reference voltage. And with a comparator, you can't add a chopper-stabiliser because the comparator's offset voltage is unknowable when the input signal applied to the inputs may be many millivolts or volts.
Here's an application where the LMP2011 can act as a precision comparator with better than 10µV of resolution and precision. Yet by closing its feedback loop continuously, the op amp maintains full dc precision, low offset, and negligible drift.
In Figure 1, the signal is brought in to the summing point through R1. If the desired threshold point is zero, R2 isn't needed. (But if a particular reference voltage VREF is needed, R1 and R2 must be closely matched so the output will trip when +VIN crosses -(-VREF).)
Alternatively, if the reference voltage is small (less than 100mV) and at low impedance, it could be connected to the positive input of the LMP2011 through R4.
The summing point voltage of the LMP2011 is maintained within a few microvolts of its positive input, basically all the time. When the input crosses zero millivolts in a positive direction, the output starts to move down from its limit value, such as +1.4 V. Positive feedback is applied through the R5-R4 divider to drive the positive input negative, increase the response speed, and supply dc hysteresis. Further, ac hysteresis is applied through the R6-R4 divider. So for a short time, additional positive feedback voltage is applied to the positive input—but only for a short time. When R6-C1 has discharged, the hysteresis will be restored to the dc value of about 15µV p-p, or ±7.5µV. This is important to get freedom from noise or oscillation when the input signals are very small.
Using 2N3904s as diodes is very important for full accuracy because most ordinary diodes are much too leaky around ±60 mV to work well. Ordinary gold-doped 1N914s or 1N4148s are quite unsuitable due to their high leakage and conductance, even at room temperature.
The observed delay time for signals as small as ±10µV is about 5 ms, decreasing to 0.8µs for large signals. See the chart of delay time versus signal amplitude (Figure 2). In some cases with small input signals, less than 1 mV, the delay time can be reduced by a factor of three or four by using lower values for R1 and/or R6, such as 2 kΩ. The improvement in delay time isn't quite as big a factor for large signals. Using low values for R6 may cause dynamic errors or delays if the input waveform is asymmetrical in amplitude or in duty cycle.
or: Mail Stop D2597A, National Semiconductor
PO Box 58090, Santa Clara, CA 95052-8090.