2026-05-21
A frequency-to-voltage converter (FVC) produces an output voltage proportional to the frequency of an input pulse train. It's the inverse of a VCO, and it's how you read tachometers, flow meters, anemometers, and turbine sensors with an analog input or a slow ADC.
The classic approach uses a charge-pump architecture, built around a chip like the LM2907/LM2917 or assembled from discrete parts. The principle is elegant: every input pulse dumps a fixed packet of charge (Q = C·V) onto an integrating capacitor, and a resistor bleeds that charge off continuously. At equilibrium, the average current pumped in equals the current bled out, so the output voltage settles proportional to pulse frequency.
The core equation is beautifully simple:
where C1 is the charge-pump capacitor, R1 is the output resistor, and V_cc is the regulated supply that defines the charge-packet voltage swing.
Real-world example: automotive tachometer. An engine ignition coil produces one pulse per cylinder firing. For a 4-cylinder engine at 6000 RPM, that's 200 Hz. You want 0–5 V output for 0–8000 RPM (267 Hz max). Pick C1 = 10 nF, V_cc = 8 V, and solve for R1:
Round to 220 kΩ and trim with the gain network.
The critical tradeoff is ripple versus response time. The output capacitor C2 (across R1) smooths the staircase into a clean DC voltage. Bigger C2 means less ripple but slower response to frequency changes. Rule of thumb: ripple is approximately
If you can tolerate 50 mV ripple with V_cc = 8 V and C1 = 10 nF, you need C2 ≥ 0.8 µF. Settling time is roughly 3·R1·C2 — so with 220 kΩ and 1 µF, expect ~660 ms to settle. For fast-changing signals, you face a fundamental noise-bandwidth tradeoff.
Practical gotchas: input signals from coils or sensors are noisy and bipolar, so put a Schmitt-trigger comparator (or use the LM2907's built-in one) ahead of the charge pump. The hysteresis kills spurious double-triggering. Also, the charge-pump capacitor must be a low-leakage type — ceramic C0G/NP0 or film, never X7R, because dielectric absorption causes nonlinearity. Finally, watch your minimum frequency: at very low f_in, ripple becomes a large fraction of V_out and the converter looks broken.