Chopper Stabilization

Chopper stabilization is the frequency-domain answer to the same offset-and-1/f-noise problem autozeroing solves in the time domain — instead of measuring an amplifier's own error and subtracting it, chopping moves the input signal somewhere else in the spectrum before the amplifier ever touches it, so the amplifier's own offset and flicker noise end up relocated to a different frequency than the signal rather than canceled directly.

The mechanism works by modulating the input signal up to a chopping frequency before amplification, then demodulating the output back down afterward, and the asymmetry this creates is the entire trick. A chopper switch multiplies the input by a square wave at the chopping frequency, shifting the signal's spectrum up to sit centered on that frequency rather than at DC; the amplifier then processes this shifted signal along with whatever offset and low-frequency noise it contributes internally, which remains at its own original location near DC since the amplifier's own imperfections were never modulated. A second chopper switch at the output, synchronized to the same chopping frequency, demodulates — multiplying by the same square wave again undoes the input modulation and brings the signal back down to DC, but because demodulation is itself just another frequency shift, it also shifts whatever was sitting at DC at that point — the amplifier's own offset and 1/f noise — up to the chopping frequency instead.

This is why the amplifier's offset is never actually measured or subtracted in a chopper design — it is simply relocated to a frequency the rest of the signal chain can filter away. The offset value itself is completely unaffected by any of this; a chopper amplifier's internal transistors have exactly the same mismatch-driven offset they would have without chopping. What changes is where that offset ends up sitting in the output spectrum — pushed up to the chopping frequency and its odd harmonics, cleanly separated from the signal now sitting back at DC, where a low-pass filter on the output can remove it without touching the signal at all.

The chopping frequency has to be chosen specifically above the flicker-noise corner for this technique to actually deliver its promised benefit, which is the direct link back to that corner frequency as a design parameter. If the chopping frequency sat below the device's own 1/f noise corner, the relocated offset and noise would land back in a region still dominated by flicker noise rather than the flat thermal floor above the corner — the technique would move the problem without solving it. Choosing a chopping frequency safely above the corner guarantees the relocated noise lands in thermal-noise-dominated territory, where it is flat, predictable, and genuinely removable by filtering rather than simply shifted into an equally noisy neighborhood.

The practical cost of chopper stabilization is output ripple, the direct consequence of the relocated offset never being perfectly filtered away. Real chopper switches are not ideal — charge injection and clock feedthrough from the chopping switches themselves, combined with whatever offset does get relocated to the chopping frequency, produce a residual ripple in the output at the chopping frequency and its odd harmonics, visible as a periodic disturbance riding on top of the recovered signal unless adequately filtered. A simple output low-pass filter trades ripple suppression against usable signal bandwidth, since the filter has to sit below the chopping frequency to remove the ripple but above whatever the actual signal content needs to pass; more sophisticated designs use a second, nested chopping stage or notch filtering specifically tuned to the chopping frequency to suppress ripple without sacrificing as much bandwidth.

Chopping and autozeroing solve the identical underlying problem through genuinely different mechanisms, and the choice between them is an architectural one, not a performance ranking. Autozeroing is inherently a sampled, discrete-time technique — it needs a dedicated measurement phase every cycle, which folds wideband noise down into the signal band as an unavoidable side effect of sampling. Chopper stabilization can run as a fully continuous-time technique, avoiding that folding penalty entirely, but trades it for the output ripple problem autozeroing does not have. Many precision instrumentation amplifiers combine both techniques in the same design specifically because each one's weakness is the other's non-issue — chopping handles the continuous-time case cleanly while autozeroing cleans up whatever chopper ripple and residual offset remain.

Chopper stabilization is proof that offset and 1/f noise can be fought without ever measuring either one directly — moving the signal somewhere the amplifier's own imperfections are not is just as valid a strategy as measuring those imperfections and subtracting them, and real precision analog design increasingly uses both strategies layered together rather than picking one. The flicker-noise corner this technique depends on is not incidental background information; it is the specific number that tells a designer whether a chosen chopping frequency will actually work or simply relocate the problem to an equally hostile part of the spectrum.