Visualize the frequency response of a first-order RC high-pass filter set by R and C. Inspect cutoff, gain, phase and time constant with a live Bode plot and AC-coupled waveforms.
At $f = f_c$, $|H| = 1/\sqrt{2}\approx -3.01\,\text{dB}$ and the phase is $+45°$. For $f \ll f_c$ the magnitude rolls off at +20 dB/decade and for $f \gg f_c$ it asymptotes to 0 dB.
What is a first-order RC high-pass filter?
🙋
Low-pass made sense to me — kill the high frequencies. But why would I ever want a high-pass?
🎓
Classic example: blocking DC. A microphone capsule carries a DC bias from its power rail. If you feed that straight into an amplifier it saturates. Put a series capacitor in front (= RC high-pass) and the DC is blocked while the audio AC passes through. We call this "AC coupling".
🙋
How does the circuit differ from a low-pass?
🎓
You just swap the parts. Low-pass has R in series and C to ground; high-pass has C in series and R to ground. At DC the capacitor impedance 1/(jωC) is infinite so the output is zero. At high frequency C is essentially a wire and the input voltage shows up across R. That asymmetry kills low frequencies.
🙋
The phase is +45° at f_c — that surprised me. Low-pass was -45°, right?
🎓
Right, a high-pass introduces phase lead. At f_c it is +45° and at very low frequencies it approaches +90°. The circuit also acts as a differentiator for τ short compared to the signal period, which is how PWM edge detectors and ECG QRS picks work.
Physical model and key equations
Apply Kirchhoff's voltage law to the series RC, with the output taken across R. From $i_C = C\,d(V_{in}-V_{out})/dt = V_{out}/R$ we obtain a first-order linear ODE.
In decibel form, the response is 0 dB for $f \gg f_c$, exactly -3.01 dB at $f = f_c$, and asymptotes to +20 dB/decade for $f \ll f_c$. The phase moves monotonically from +90° down to 0°.
Real-world applications
Audio AC coupling: Place a series capacitor between amplifier stages to remove DC offset. Setting f_c below 20 Hz keeps the audio band intact while protecting downstream gain stages from rail-to-rail clipping.
ECG baseline removal: Strip out 0.5 Hz drifts from breathing and posture changes while keeping the P, QRS and T waves. A cutoff around 0.05 to 0.5 Hz is typical for diagnostic ECG.
Accelerometer gravity removal: Subtract the DC gravity vector so that only the motion-induced acceleration remains. Step counters and gesture recognizers rely on this preprocessing.
Differentiator (PWM edge detection): With τ much shorter than the input period, the high-pass output becomes a series of sharp spikes at every rising and falling edge — perfect for timing or trigger generation.
Common misconceptions and caveats
Myth 1: below f_c the signal is fully blocked. A first-order high-pass only attenuates at 20 dB/decade. At f = 0.1 f_c you get -20 dB, at 0.01 f_c only -40 dB. Truly removing DC requires accounting for capacitor leakage and any downstream DC gain.
Myth 2: a high-pass leaves the phase untouched. It introduces a +45° lead at f_c. In control loops this is useful as a lead compensator, but for waveform-shaping applications the phase distortion can be a problem.
Myth 3: just pick the biggest capacitor. Electrolytics introduce leakage current, temperature drift and polarity constraints. For low-frequency high-pass stages, film capacitors, tantalum or large MLCC arrays are usually safer choices.
FAQ
For a step input the high-pass output jumps to the input height and decays to 36.8% (= 1/e) in one τ. This is the basic property exploited by edge-detecting differentiator circuits.
Cascade a high-pass and a low-pass with f_HP < f_LP. The combined response passes a band between the two cutoffs, with the centre frequency f_0 = √(f_HP·f_LP) carrying the maximum gain.
CR is smaller, cheaper and much more common. LR (R in series with L to ground) shows up in RF and on power lines, but inductor size, core loss and self-resonance usually rule it out for audio or low-frequency work.
Set R from 10 Ω to 100 kΩ and C from 1 nF to 100 µF (100000 nF in the input). Both sliders and numeric inputs are available.
Read τ = RC and f_c = 1/(2πτ). The time-constant display selects ns, µs, ms or s to suit its value.
Vary the observation frequency from 1 Hz to 1 MHz. The Bode chart shows magnitude; the numeric result and input/output waveforms show phase lead. Click frequency sweep again to stop it.
Compare ideal AC-coupling behavior. Quantization bits N (4–16) affect only the ideal full-scale-sine ADC SNR result, not the RC response. Waveform quantization and actual noise are not simulated.
Worked Example
For an AC-coupling check, enter R = 10000 Ω, C = 100 nF and f = 1000 Hz. Then τ = 10000 × 100 × 10⁻⁹ = 1 ms and f_c ≈ 159.15 Hz. At 1 kHz the amplitude ratio is approximately 0.98757 (−0.11 dB), with +9.04° phase lead. Changing f to 100 Hz gives approximately 0.53202 (−5.48 dB) and +57.86°. The ideal half-power point is f = f_c, with −3.01 dB and +45°; it is not 100 Hz for these components. The ideal DC limit has zero output, but 0 Hz is outside the input range. These values provide a check for microphone or sensor AC-coupling calculations.
Practical Notes
Choose the cutoff from allowed passband attenuation and phase shift. For a chosen passband starting at 80 Hz, 10 kΩ and 0.8 µF place f_c near 19.89 Hz; whether this is adequate depends on the required response, not a universal audio rule.
Increasing R or C lowers f_c but does not change the first-order asymptotic slope of 20 dB/decade. A tenfold RC increase moves the cutoff down one decade.
Phase is +45° at f_c, approaches +90° at low frequency and 0° at high frequency. This ideal model excludes capacitor ESR, leakage and parasitics; evaluate tolerances and temperature coefficients from the chosen component specifications.
The model assumes an ideal voltage source and no additional output load. Real source resistance and loading change both gain and cutoff. Include them in a circuit analysis and consider buffering when loading matters; no fixed 100 kΩ threshold is implied.