Distinguishing Switched Capacitor Filter Clock Feedthrough From Noise

A switched capacitor filter can produce a small clock related output even with no applied signal. That observation does not mean every visible output fluctuation has the same cause. Separate discrete clock components, fast switching transients, and random passband noise so the correction addresses the actual error.

Read the measurement definition

The Analog Devices LTC1068 data sheet distinguishes clock feedthrough from wideband noise and from switching transients whose observed amplitude depends strongly on probing and bypassing. Use the selected device’s definitions when comparing a bench result with a specified value.

Record supply voltage, clock frequency, filter configuration, input termination, and output load. A specification measured with the input grounded should not be compared directly with an unconnected high impedance input.

Classify the observed spectrum

Inspect the output with the input terminated as intended for a no-signal measurement. Identify narrow components synchronous with the clock and compare them with the surrounding noise floor. Change the clock deliberately within supported conditions to see which features move with it.

Use time domain measurements to inspect edge related transients, but keep the probe return short and its bandwidth documented. A long ground connection can turn a small switching event into an impressive ringing waveform that is partly a measurement artifact.

Choose the correction for the error type

A suitable output low pass network can reduce out-of-band clock components. Calculate its effect on the wanted response and output loading before adding it. It cannot selectively remove random noise already inside the required signal band.

Review clock routing, supply bypassing, and coupling into nearby high impedance nodes. Keep the filter’s clock path identifiable so a layout change can be evaluated against the same measurement setup.

Check the receiving circuit

Determine whether the downstream stage samples, rectifies, or overloads on the remaining clock component. An out-of-band tone can still matter if a following circuit translates it into the measurement band. Evaluate the actual receiving chain rather than only a bandwidth limited display of the filter output.

Save separate results for clock related amplitude, in-band noise, and any relevant transient peak, with their measurement bandwidths. Repeat after the chosen correction and confirm that the passband remains acceptable. A useful conclusion identifies which contribution was reduced and why. Calling every output disturbance noise makes it difficult to decide whether additional filtering, better layout, or a different operating configuration is the appropriate next step. Record whether the clock remains active during each no-signal capture.