3 Aliasing in CCCSA

In theory, chopping amplifiers operate in a continuous-time manner. Input and output are connected at all times, aliasing does not occur. In reality, design constraints impose physical limitations on these assumptions. Risk of aliasing is minimized through proper design techniques. However, aliasing is still checked for during the silicon verification phase. Time domain and spectrum analysis can be used for this purpose.

In CCCSA, the resistor gain network is replaced by capacitors. To optimize for power and accuracy, the CCCSA takes advantage of a combination of chopping and auto-zeroing. During the zeroing phases, input is effectively disconnected from output. Due to this sample and hold behavior, aliasing can happen when input signal frequency is higher than half the chopping frequency according to Nyquist theorem.

A good application example is switching power supplies, where average output current is to be measured. High frequency switching ripples can fold into signal band when aliasing happens. As a result, the output typically exhibits elevated offset, or low frequency oscillation. Performing FFT on the output waveform also reveals the unwanted tones.

For demonstration purposes, two units of INA190 were tested side by side in this experiment. Both units are driven with the same 20mVpp, 800kHz sinusoidal waveform riding on a DC input to simulate the output of a switching regulator.

Figure 3-1 shows the response of the two units. The top half shows the time domain response (Yellow=Unit 1; Green=Unit 2); the bottom half shows the frequency domain response (Yellow=Unit 1; Purple=Unit 2).

The frequency of the harmonics is a function of both the chopping frequency and the input signal frequency. The harmonics are symmetrically located around the chopping frequency and multiples. From this property the chopping frequency can be estimated. Due to device variations, the harmonics frequency varies accordingly. This is why Unit 1 and Unit 2 have very different responses even though the input is identical. Such variation becomes harder to predict when temperature drift is taken into consideration.

Figure 3-2 shows the same setup, where the only difference is to reduce the input signal frequency to 790kHz. The tones shifted by 10kHz. In Unit 1, the shift resulted in an increase in the magnitude of DC output. Such an increase becomes indistinguishable from the true DC output.