Background: Why I Calculate Parameters Before Picking Up the Soldering Iron
Recently I added a bandpass stage to a sensor signal-conditioning circuit. The requirements were a center frequency f0=1kHz and bandwidth BW=500Hz, which gives Q=f0/BW=2, on a single 5V supply with only one op-amp allowed. My first version picked the capacitor values off the top of my head—C1=C2=100nF—then worked backward to the resistors. The result: at Q=2 the resistors either dropped to a few hundred ohms or soared into the megaohms, bias current and noise crept in together, and the gain was coupled to Q—change one resistor and all three specs drifted at once.
The upside of a single-op-amp Sallen-Key bandpass is a low parts count and high non-inverting input impedance, but its parameters are tightly coupled, and hand-tuning is just too inefficient. Once I sorted out these relationships, I handed the rest of the work to the calculator.
Parameter Relationships: Pick k First, Then Solve for the Five Resistors
In this circuit, the parameter relationships revolve around an intermediate variable k. The tool automatically picks C1 to keep R1 near 33kΩ:
Then:
The gain branch is set by R4/R5, and the gain is tied to Q:
Take Q=2 as an example: , i.e., an in-band gain of about 2× (roughly 6dB), with R4 and R5 equal. The tool automatically places R5 near 10kΩ. The higher the Q, the slightly lower H becomes: at Q=10, . This gain–Q coupling is the easiest thing to overlook, and it's why many people copy feedback resistor values from a reference yet can't get the gain to line up.
Running It Through the Online Calculator
Open the BPF Op-Amp Bandpass Filter Designer (Sallen-Key, single op-amp). I entered f0=1kHz and BW=500Hz for the design conditions; it also accepts high-pass/low-pass corner frequencies fL and fH and converts them. The page returned standard values of roughly: C1=10nF, C2=4.7nF (i.e., C1/2 rounded to the nearest standard value), R1≈33kΩ, R2≈11kΩ, R3≈66kΩ, and R4 and R5 at 10kΩ each. It also gives the actual center frequency, bandwidth, and a Bode plot after rounding, so I don't have to work through tolerance numbers myself. The GBW and slew-rate checks live on the same page: this set of values doesn't demand much op-amp bandwidth, and my 10MHz rail-to-rail part is plenty. If I changed the example to f0=100kHz and Q=20, the bandwidth requirement would surge, and an ordinary low-bandwidth op-amp would fall short.
Conclusions and Common Pitfalls
Based on my measurements, here are my takeaways:
- Make C1 accommodate the resistor, not the other way around. Keeping R1 near 33kΩ gives a balanced trade-off among noise, bias current, and power consumption. Fixing C1=100nF first pushes all the resistors down by 10×.
- C2 is not arbitrary. Many low-pass tutorials reach for equal-value capacitors out of habit, but in this bandpass topology C2=C1/2—switching to equal values directly changes Q and gain.
- Don't be greedy with Q. At Q=2, ordinary 5% resistors and X7R capacitors work fine. If you push Q above 20, switch to 1% resistors and COG/NPO capacitors, or the center frequency and bandwidth will drift.
- Check the rounding error first. After rounding to standard resistor values, f0 and Q will deviate from the design targets; go by the actual curves the tool provides, and if the deviation is large, switch to E96 values or fine-tune C1.
- Check the gain before the op-amp. H is coupled to Q, so the feedback resistors must be recalculated from H. When the output swing is large, verify the slew rate: .
Related Tools
A filter never exists in isolation. If you need to suppress low-frequency drift or high-frequency glitches at the input first, you can work out a Sallen-Key low-pass/high-pass to carve out the band. If you want a steeper roll-off, cascade two bandpass stages, or switch to a multiple-feedback (MFB) bandpass topology, which has different component sensitivities. Once the values are computed, cross-check the GBW and slew-rate specs when picking the op-amp—all of these checks are available right on the bandpass calculator page, far faster than the Excel tables I used to fiddle with.