Twin-T Notch Filter Design: Matching, Depth, Trimming, and Measurement

Specify the Interference and the Signal You Must Preserve

State the unwanted frequency range, required attenuation across that range, acceptable loss and phase shift in the wanted band, source impedance, load impedance, operating temperature, and allowed input amplitude. A 50 Hz or 60 Hz label is not a promise that interference stays at exactly one frequency. Mains-derived tones, motor harmonics, or drifting interference may require rejection across a band rather than only a very deep null at one point. Decide whether passive Twin-T is appropriate before selecting parts; the Active and Application-Specific Filter Topologies Guide compares topology families, whereas this article follows one passive implementation into matching and validation.

Preserve the Adopted Twin-T Connections

The low-pass T is two series resistors R with a 2C capacitor from their midpoint to the reference node. The high-pass T is two series capacitors C with an R/2 resistor from their midpoint to reference. The branches connect in parallel between input and output. The ideal transfer function below assumes a low-impedance source and a high-impedance load. A differently drawn or actively fed-back Twin-T may use different labels and Q; do not copy its adjustment procedure without checking the actual netlist.

Calculate a Buildable 50 Hz Example

For a 50 Hz nominal null and base C = 100 nF, R = 1/(2π × 50 Hz × 100 nF) ≈ 31.831 kΩ. Use two series resistors close to that value, two series capacitors of 100 nF, a shunt resistor near 15.915 kΩ, and a shunt capacitor near 200 nF. At 60 Hz with the same C, base R would be about 26.526 kΩ. These are nominal targets, not assertions about available standard values, component drift, or rejection depth.

The Notch Filter Calculator solves ideal base values and checks entered ratio errors. Enter the values that will actually be fitted, including measured shunt values, then use its frequency and balance outputs as screening evidence. Its imbalance status is deliberately not a dB-depth prediction.

Separate Frequency Spread From Cancellation Error

A common scaling change to both R and C moves the nominal frequency: if both are 5% high, the simple balanced estimate becomes 50/1.05² ≈ 45.35 Hz; if both are 5% low, it becomes 50/0.95² ≈ 55.40 Hz. This corner calculation says nothing by itself about the null depth, because both branches could move together and remain ratio-matched. Conversely, a small mismatch between the two nominally equal elements can leave the apparent minimum near 50 Hz while allowing substantial signal leakage. Treat absolute tolerance, pair tracking, and shunt-ratio accuracy as separate requirements.

Select components by measured value or use matched networks when depth matters. Capacitor dielectric, temperature coefficient, DC bias, aging, leakage, and parasitic capacitance must be considered for the actual signal environment. Put the network close together on the PCB, keep the two paths geometrically comparable, and avoid leakage across high-impedance nodes. Monte Carlo analysis can estimate yield, while worst-case corner sweeps provide bounded design checks when input ranges are genuinely guaranteed. Neither replaces measurement of the assembled stage.

Isolate, Trim, and Measure the Real Network

Drive the passive Twin-T from a sufficiently low source impedance and sense it with a high input impedance. A buffer may isolate a real source or load, but its output swing, bias, noise, and stability still need checking. Probe capacitance can disturb the network, particularly at high impedances; use the same loading during tuning and final verification. If an active feedback loop is introduced to raise Q, analyze that circuit separately: the passive Q = 0.25 equation no longer describes its full response.

Record voltage magnitudes at the same test frequency and conditions. Rejection is 20 log10(|Vin|/|Vout|) dB. A 40 dB reading means the output magnitude is 1% of the input; 60 dB means 0.1%, assuming both readings exceed the instrument floor. If the residual is below that floor, report a lower bound such as “at least 60 dB under the stated setup,” not an exact depth. Also record passband insertion loss: a tiny output everywhere is not a useful notch.

Common Mistakes

• Treating the ideal mathematical null as a measured attenuation guarantee.
• Matching base R and C but ignoring R/2 and 2C shunt ratios.
• Confusing absolute frequency tolerance with branch cancellation error.
• Using the calculator's generic Q/BW reference as an exact loaded passive Twin-T model.
• Adding an active feedback loop while retaining the passive Q equation.
• Ignoring source resistance or a finite load resistance.
• Trimming frequency without watching residual level and wanted-band loss.
• Claiming a depth from the calculator's mismatch warning alone.
• Choosing capacitors whose bias or temperature drift breaks matching.
• Measuring only the nominal room-temperature assembly.
• Reporting the analyzer noise floor as an exact deep null.
• Letting the measurement probe alter the very network being tuned.

Primary Engineering Sources

The passive-versus-active Twin-T Q boundary and practical tuning limitations are discussed in Analog Devices MT-225. For buffering and high-Q implementation context, see TI/National Semiconductor Linear Brief 5. Use their own schematics and labels when applying their active-circuit equations; this page's calculations follow the passive topology declared above.