In short: A Wien bridge oscillator places a series RC and a parallel RC network around an op-amp's non-inverting input; with equal R and C the network passes exactly 1/3 of the signal at zero phase shift at f = 1/(2πRC), so a non-inverting gain of 3 sustains oscillation, and a slowly acting nonlinear element (lamp or diodes) pins the gain at exactly 3 to stabilise the amplitude.
Why the Wien bridge still matters
Ask any analog interview panel how to build a clean low-frequency sine source and the Wien bridge comes up quickly. It needs one op-amp, four resistors and two capacitors, yields THD of 0.01% or better when properly levelled, and tunes over a decade with a dual-gang pot. The classic audio generators of the 1940s used it, and the same topology survives in lab signal sources and in sensor-excitation circuits. If you have already studied the three-stage RC phase-shift oscillator, you will notice the Wien bridge needs no inverting stage and no 180° phase shift: the feedback path is non-inverting, so the network must give zero phase at the oscillation frequency.
The circuit and the frequency-selective network
The op-amp is connected as a non-inverting amplifier. Its gain is set by Rf and Rg: A = 1 + Rf/Rg. The output also feeds a positive-feedback network: R1 in series with C1 (the series arm Zs), and R2 in parallel with C2 (the parallel arm Zp) to ground. The node between them drives the non-inverting input.
Deriving f = 1/(2πRC)
Let Zs = R + 1/(sC) and Zp = R/(1 + sRC). The feedback factor is the divider ratio:
β(s) = Zp / (Zs + Zp)
= sRC / (1 + 3sRC + (sRC)^2)Put s = jω and define x = ωRC. Then β(jω) = jx / (1 − x² + j3x). The phase is zero only when the real part of the denominator vanishes: 1 − x² = 0, so x = 1, i.e. ω0 = 1/(RC) and f0 = 1/(2πRC). At that point β = j / (j3) = 1/3, purely real. Below f0 the network behaves as a high-pass (phase lead, +90° at DC side); above f0 it becomes a low-pass (phase lag, −90°). It is a band-pass of modest Q: the peak is only a gentle hump, which is why amplitude control matters so much.
Key fact: with R1 = R2 = R and C1 = C2 = C, the bridge attenuates by exactly 3 with exactly 0° phase at f0 = 1/(2πRC).
Concrete numbers: R = 10 kΩ and C = 15.9 nF give f0 = 1/(2π · 10⁴ · 15.9×10⁻⁹) ≈ 1.00 kHz. Replace each R with one gang of a 10 kΩ–100 kΩ dual pot and the same C gives roughly 100 Hz to 1 kHz. Switch C by decades (1.59 µF, 159 nF, 15.9 nF) to span 10 Hz to 100 kHz, which is how bench generators do it.
The gain-of-3 condition
The Barkhausen criterion demands loop gain Aβ = 1 at zero phase. Since β(f0) = 1/3, A must equal 3, so Rf = 2Rg. The closed-loop pole story is more instructive. The characteristic equation 1 − Aβ(s) = 0 becomes:
s^2 + s(3 − A)/(RC) + 1/(RC)^2 = 0The damping term is (3 − A). For A < 3 the poles are in the left half-plane and any disturbance dies out. For A = 3 the poles sit exactly on the jω axis. For A > 3 they move into the right half-plane and the amplitude grows exponentially with time constant 2RC/(A − 3). With A = 3.05 and RC = 159 µs, the envelope time constant is about 6.4 ms, so a thermal-noise seed becomes volts in tens of milliseconds. This is also why you cannot just trust resistor tolerances: 1% resistors alone would put A anywhere from 2.94 to 3.06, either dying out or clipping.
Amplitude stabilisation
A linear circuit cannot hold A at exactly 3 forever, so every practical Wien bridge uses a nonlinearity that lowers the gain as the amplitude rises. Two approaches dominate:
- Incandescent lamp as Rf (or part of Rg). A small tungsten lamp has a strongly positive temperature coefficient. Higher output current heats the filament, raising its resistance and reducing gain. Its thermal time constant is hundreds of milliseconds, far slower than the signal period, so it responds to the RMS level rather than the instantaneous waveform and distortion stays very low. This was the Hewlett-Packard 200A audio oscillator trick.
- Antiparallel diodes across part of Rf. Make Rf = Rf1 + (Rf2 in parallel with two silicon diodes). At small signal the diodes are off and A is a bit above 3; as the peaks approach about 0.6 V across Rf2 the diodes conduct and divert current, lowering effective gain toward 3. It is cheap and fast but clips the peaks, so expect 1% THD unless you soften the knee with series resistors.
Modern designs use a JFET as a voltage-variable resistor in Rg, driven by a peak detector, which gives a true automatic gain control loop with distortion in the 0.001% region.
Key fact: stabilisation works because the loop gain depends on amplitude. The oscillation amplitude settles where the amplitude-dependent gain equals exactly 3, so the amplitude is set by the limiter, not by the supply rails.
Practical design and simulation notes
| Parameter | Worked value | Why it matters |
|---|---|---|
| R, C | 10 kΩ, 15.9 nF | f0 ≈ 1 kHz |
| Rg | 10 kΩ | Sets the gain denominator |
| Rf (small signal) | about 20.5 kΩ | A ≈ 3.05, guarantees start-up |
| Op-amp GBW | at least 100 × f0 × 3 | Extra phase lag shifts the frequency and distorts |
| Supply | ±12 V | Headroom well above the limited amplitude |
Because a real amplifier has finite gain-bandwidth, it adds phase lag near f0, and the oscillation frequency drops slightly below 1/(2πRC). With a 1 MHz GBW op-amp at 100 kHz, the error is large; an op-amp such as the TL072 (3 MHz) is comfortable to about 20 kHz, while a wideband part like the OPA134 or LM6171 stretches the range. In SPICE-class tools such as Cadence Spectre or Siemens Questa ADMS, run a transient with a small initial condition or a brief current pulse; a perfectly symmetric simulation will otherwise sit at zero forever. Then use a periodic steady-state analysis (PSS) with a harmonic-balance engine such as Spectre RF to measure THD and phase noise without waiting for a long settling transient. Layout matters too: any stray capacitance at the non-inverting node adds to C2, so keep that trace short.
If you want a second comparison, our Colpitts LC oscillator walkthrough shows how the same Barkhausen reasoning applies when the frequency-selective element is a resonant tank rather than an RC bridge. Teams that need a production-quality oscillator or ADC front end can find design services at Avecas.
Common interview questions and pitfalls
- Why is gain 3 and not 1? Because the bridge attenuates by 3 at resonance.
- What happens if C1 ≠ C2 or R1 ≠ R2? Then f0 = 1/(2π√(R1R2C1C2)) and the required gain becomes 1 + R1/R2 + C2/C1.
- Why not use clipping at the rails as the limiter? The output becomes a distorted near-square wave with high THD.
- Why does a dual-gang pot cause trouble? Gang mismatch changes the gain condition as you tune, which is why good generators add an AGC loop.
Want to design, simulate and debug oscillators end to end? Join the analog track in the CourseTron courses, or explore hands-on lab training at ChipXpert.
Frequently asked questions
Why must the gain be exactly 3?
The RC network passes one third of the signal at f0 with zero phase, so the loop gain is A/3. Oscillation is sustained only if that equals 1, hence A = 3.
What happens if the gain is above 3?
The poles move into the right half-plane and the amplitude grows until the output clips or a limiter reduces the effective gain back to 3.
Why use a lamp instead of diodes?
A lamp reacts slowly to RMS amplitude, so it does not distort individual cycles. Diodes act within a cycle and flatten the peaks, adding harmonics.
How do I tune the frequency?
Vary R1 and R2 together (dual-gang pot) or switch both capacitors; f = 1/(2πRC) scales inversely with either.