Technical Resources

Comparators and Threshold Detection: Hysteresis, Propagation Delay, and Output Stages

A comparator answers one question — is this voltage above or below that one — and answers it thousands to billions of times per second. Used correctly it is the cleanest decision element in analog design; used like an op-amp without feedback, it oscillates, chatters, and mis-times. Here is how the parts differ and how to design threshold detection that holds up.

Comparator vs Op-Amp: Related, Not Interchangeable

Both are high-gain differential amplifiers, but they optimize for different failures:

  • An op-amp is compensated with an internal dominant pole so it never oscillates with feedback — which makes it slow when saturated. Drive an op-amp into comparison duty and it leaves saturation at op-amp speed (microseconds), not comparator speed (nanoseconds).
  • A comparator is deliberately uncompensated: maximum speed, no stability guarantee with feedback — because its job is to saturate. Running it with linear feedback violates its design assumptions.
  • Check the datasheet category, not the schematic resemblance. "Op-amp, comparator mode" app notes exist, but dedicated comparators cost the same and behave predictably.

Hysteresis: The Difference Between a Decision and Chatter

A slow or noisy input crossing a single threshold produces multiple output transitions — contact bounce for voltages. Hysteresis (positive feedback) splits the threshold into two:

  • Inverting configuration: the reference sits on the non-inverting input through a divider from the output. Trip points become Vref ± (β × Vswing), where β is the feedback fraction — calculate both transitions explicitly for single-supply rails, where the output swing is asymmetric.
  • Rule of thumb: total hysteresis should be 2–3× your input noise peak-to-peak, and small enough not to distort the measurement's meaning. A window comparator or ADC is the answer when hysteresis would be too wide.
  • Resistor values matter: too large and input bias currents shift the thresholds; too small and the output stage loads the reference. Keep divider current 100× the bias current as a starting point.

Propagation Delay: Speed Has Fine Print

  • Response time depends on overdrive: the datasheet headline (say 10 ns) is measured with a large overdrive (100 mV); at 5 mV of overdrive the same part may take 3–10× longer. Design with the overdrive-relevant curve, especially for slow-ramping inputs.
  • Dispersion (jitter): transition-to-transition delay variation matters in timing applications — check dispersion specs, not just average delay.
  • Slow inputs need help: a comparator with a ramping millivolt input idles near its threshold and amplifies supply noise. Add hysteresis, or precede it with a preamplifier stage, or use a part with built-in input hysteresis.

Output Stages and System Consequences

Output TypeBehaviorFits
Push-pull (CMOS/TTL)Drives both rails, fastest edges, defined logic levelDirect MCU inputs, logic interfacing
Open-drain / open-collectorSinks only; pull-up sets the logic railLevel translation, wired-OR fault buses, isolation barriers
  • Open-drain outputs enable free level translation: the pull-up resistor can reference a different rail than the comparator's supply — the same trick I2C uses.
  • Ground bounce: fast output edges couple into the input through shared ground; separate the input ground return from the output return, or the comparator will retrigger itself.
  • Supply rejection at the moment of switching: the output transition disturbs the supply, which the input sees. Decouple the comparator's supply locally and heavily.

Design Checklist

  1. Write the threshold, noise floor, and required decision rate before selecting the part.
  2. Choose hysteresis from measured noise, not assumptions; calculate both trip points on your real rails.
  3. Verify response time at your actual overdrive from the curves, not the headline number.
  4. Plan the output stage against the receiver: logic rail, speed, and fault behavior (open-drain fails high-impedance, push-pull fails defined).
  5. Layout: short input traces, guarded high-impedance nodes, separated ground returns.

JTDZ Tech supplies fast, precision, and low-power comparators from TI, ADI, and onsemi — with overdrive curves and output-stage options documented. Tell us your threshold, speed, and load and we will quote the part that decides cleanly, every time.

RF Mixers and Frequency Conversion: Mixer Types, Conversion Loss, and Image Rejection
Inside IC Final Test: Wafer Sort, Burn-In, and What Yield Tolerance Means for Buyers