Engineering Blog
Why Oscillators Start at the Wrong Frequency or Fail to Start
The frequency equation describes the intended network. Startup and accuracy depend on loop gain, phase, loss, loading, nonlinear amplitude control, parasitics, device limits, tolerance, and the measurement method.
- Reading Time
- 17 min
- Difficulty
- Intermediate
- Last Updated
- October 3, 2026
Frequency Condition and Startup Condition
An oscillator needs the correct loop phase and enough small-signal loop gain for disturbances to grow. Nonlinearity then reduces effective gain to establish steady amplitude.
Formula reference
Useful oscillator relationships
|Aβ| > 1 at startup∠Aβ ≈ 0° mod 360°f_LC = 1 / (2π√(LC))Δf_ppm = 10⁶ · Δf/f_nomVariable definitions
- Aβ
- frequency-dependent loop gain
- L and C
- effective tank values including loading
- Δf
- frequency error
- f_nom
- nominal frequency
Use the Oscillator Types and Frequency Networks Reference to keep topology and adopted frequency model explicit.
Ten Failure Causes
1. Resonance is mistaken for guaranteed oscillation
A frequency-setting network can resonate while the active circuit fails to supply enough loop gain or negative resistance to overcome loss at the required phase.
2. Component tolerance is evaluated one part at a time
Frequency depends on ratios and products of several components. Worst-case corners, temperature coefficients, aging, and correlated drift must be propagated through the actual equation.
3. Parasitics are omitted
Device capacitance, PCB capacitance, inductor self-capacitance, probe loading, package inductance, and leakage shift the effective network, especially at high impedance or high frequency.
4. Tank Q and loss are ignored
Inductor DCR, capacitor ESR, core loss, loading, and active-device input/output resistance reduce Q and increase the startup gain required.
5. The feedback ratio is copied from the wrong topology
Colpitts, Clapp, Hartley, Wien bridge, phase-shift, and relaxation oscillators define feedback around different nodes. A component ratio is not a universal loop-gain formula.
6. The active device operates outside its linear or switching range
Limited gain-bandwidth, slew rate, common-mode range, output swing, comparator recovery, threshold error, or drive current can stop startup or distort timing.
7. Crystal load capacitance and drive level are wrong
PCB and pin capacitance participate in effective load. Excess drive can age or damage a crystal, while excessive ESR or insufficient negative resistance prevents reliable startup.
8. Relaxation thresholds are assumed ideal
Schmitt and comparator oscillators depend on actual thresholds, hysteresis, propagation delay, output saturation, charging current, leakage, and RC tolerances.
9. Delay-based oscillators use nominal gate delay
Ring frequency varies strongly with supply, temperature, process, loading, edge direction, stage count, and routing. Datasheet typical delay is not a precision timebase.
10. Frequency is measured in a way that changes it
Probe capacitance, cable termination, counter trigger level, grounding, buffer loading, supply injection, and startup timing can pull or interrupt the oscillator.
Diagnostic Scenarios
LC oscillator starts only when probed
The probe changed tank capacitance, loss, or bias. Measure through a buffer, inspect startup margin, and include device and PCB parasitics in the tank model.
Crystal frequency is consistently high
Verify effective load capacitance including both capacitors, pin capacitance, and PCB stray capacitance. Also check the crystal's specified load and frequency reference condition.
555 duty cycle misses the estimate
Measure threshold levels, discharge saturation, supply, capacitor leakage, resistor tolerance, and propagation delay rather than assuming exact one-third and two-thirds thresholds.
VCO gain changes across range
Kvco is often local slope, not one constant over the tuning span. Calculate and measure piecewise sensitivity, monotonicity, headroom, and temperature behavior.
Oscillator Debugging Workflow
- 1. Identify the exact topology, active-device nodes, feedback path, and frequency-setting network.
- 2. Calculate nominal frequency with loaded component values and explicit waveform assumptions.
- 3. Estimate tolerance, temperature, aging, control sensitivity, and parasitic frequency shift.
- 4. Check active-device gain, phase, bandwidth, output swing, bias, and startup margin.
- 5. Include tank Q, ESR, DCR, core loss, crystal ESR, load, and buffer impedance.
- 6. Verify amplitude limiting and component stress, including crystal drive and comparator recovery.
- 7. Use startup-capable simulation with realistic nonlinear models, parasitics, and initial conditions.
- 8. Measure through a low-loading buffer across supply, temperature, production, and startup corners.
Summary
A nominal frequency calculation is necessary but not sufficient. Verify startup loop gain and phase, loss, loading, amplitude control, parasitics, tolerance, active-device limits, and measurement loading across the intended operating range.
Support reference
FAQ
Why can an LC circuit resonate but an oscillator fail to start?
Resonance defines a preferred frequency. Startup additionally requires the active loop to provide sufficient gain at the correct phase to overcome tank and load losses.
Why is measured oscillator frequency different from the formula?
Tolerance, temperature, parasitic capacitance or inductance, loading, finite Q, active-device delay, amplitude limiting, and measurement equipment can shift it.
How much loop gain is needed for startup?
More than the steady-state unity-loop condition, with margin for process, temperature, supply, loading, and loss. The required margin depends on topology and startup-time target.
Why does touching or probing an oscillator change frequency?
The probe or body adds capacitance, resistance, inductance, and coupling at a sensitive node. Buffer the output and use an appropriately low-loading measurement method.
What causes crystal oscillator startup failure?
Common causes include excessive crystal ESR, insufficient negative resistance, wrong load capacitors, excessive parasitics, low amplifier gain, poor bias, bad layout, or unsuitable crystal drive.
Why does a relaxation oscillator duty cycle differ from calculation?
Real thresholds, asymmetric output swing, propagation delay, leakage, saturation recovery, source resistance, and capacitor tolerance alter charge and discharge intervals.
Can oscillator tolerances be added directly?
Only for a justified worst-case model. Recalculate the frequency equation at relevant corners; statistical combination requires credible distributions and correlations.
Does simulation guarantee oscillator startup?
No. Startup can depend on model noise, initial conditions, nonlinear device behavior, parasitics, loss, and solver settings. Verify with realistic models and hardware.