H2-1 · What a DC servo is (and what it is not)

Quick answer (definition): A DC servo (offset servo) is a very-slow integrator loop that senses DC or ultra-low-frequency error at a chosen node and injects a small correction so the output returns to a reference level. When its corner frequency is set far below the passband, it removes drift without changing in-band gain, phase, or distortion.

Why it matters in practice: In audio and precision instruments, DC offset consumes headroom and can force early clipping or degrade distortion performance. Drift over minutes/hours can also bias measurement results and complicate production verification. A well-designed servo prevents these issues while preserving the intended low-frequency behavior of the main signal path.

When it is typically used: audio preamps that must keep bass response intact; instrumentation chains that run for long durations; systems where DC error appears at a mid/output node (not only at the input) and where “just add a coupling capacitor” would introduce unwanted phase/group-delay tradeoffs.

H2-2 · Where DC offset & drift come from in real chains

Core idea: DC offset and minute-scale drift are not “one number.” They are the visible result of multiple slow mechanisms—temperature, leakage, reference movement, dielectric memory—being converted into a DC/ultra-low-frequency component at a critical node. A servo works because it targets that node and forces the DC component back to a safe baseline.

Practical source layers (from common to sneaky):

• Input offset (Vos) & temperature drift: device drift and thermal gradients translate into a changing output baseline over minutes.
• Bias current × impedance: tiny input currents multiplied by high source/feedback impedance become large DC errors, especially in high-value resistor networks.
• Leakage currents: PCB contamination, humidity, flux residue, and protection devices can create unintended DC paths (nA–pA) that look like “mysterious drift.”
• Capacitor dielectric absorption (DA): stored charge slowly releases, creating a slow “memory” that mimics drift even after a step settles.
• Reference / mid-rail movement: single-supply midpoints and references can wander with load, temperature, or supply noise, shifting the baseline.
• Sensor intrinsic DC: some sensors/chains naturally carry a DC component that must be managed without sacrificing low-frequency fidelity.

Field symptoms engineers actually see:

• Output baseline creeps over minutes: often points to temperature drift, leakage, or bias×impedance effects.
• Low-end “swelling” or “thinning”: typically a sign that the servo corner is too high or interacting with another low-frequency pole.
• ADC headroom is permanently reduced: DC consumes swing, lowering effective dynamic range and increasing distortion risk.
• Turn-on pop / slow return after overload: initial conditions or integrator wind-up dominate (handled later in the anti-windup chapter).

Why drift is a specs problem (not just an aesthetic problem): DC offset steals output swing, can push stages closer to clipping, and can elevate distortion under large signals. In precision measurement, slow baseline movement can bias averages and long integrations. In production, drift is slow—without a plan to detect it efficiently, yield and consistency suffer.

H2-3 · Three practical servo architectures (sense & inject)

Choose a DC servo by two coordinates: where DC is sensed and where correction is injected. The same integrator behaves very differently when the injection point changes impedance, noise coupling, or saturation recovery.

Architecture A — Output-sense + summing-node injection (most common): DC is sensed at the output (captures the chain’s total DC error), then a very-slow correction is injected through a resistor into a summing node (often the inverting summing point). In-band behavior stays dominated by the main loop, while DC is “trimmed” back to the reference.

Architecture B — Output-sense + feedback-leg modulation: The servo still senses output DC, but the correction is applied by slowly modulating a feedback element (or an equivalent “DC-only” feedback leg). This can reduce direct noise injection into the highest-gain summing point, at the cost of stronger interaction with closed-loop parameters if implemented carelessly.

Architecture C — Mid-node sense + correction (single-supply midpoint management): Instead of only targeting output DC, the servo controls a mid-rail or “virtual ground” node so the whole chain’s DC reference remains stable. This is common in single-supply systems where midpoint movement consumes headroom and manifests as slow baseline drift across multiple stages.

H2-4 · How to choose f_cservo and settling time (a step-by-step recipe)

Servo parameters are defined by two numbers: the lowest frequency that must remain untouched (f_{passband_min}) and the allowed baseline recovery time (T_settle). Lower f_cservo improves transparency but slows recovery; higher values speed recovery but can distort low-frequency amplitude/phase.

Step 1 — Set an upper bound for f_cservo based on passband transparency: a practical starting point is f_cservo ≤ f_{passband_min}/20 for general instrumentation and ≤ f_{passband_min}/50…/100 when low-frequency phase/group delay must remain exceptionally clean. If low-end tone or LF waveform shape changes, the servo is too fast or interacting with another low-frequency pole.

Step 2 — Convert f_cservo into a time constant: choose a first-cut integrator time constant using τ ≈ 1/(2π·f_cservo). This yields an initial RC value for the servo integrator. This is only a starting point—node impedance and injection ratio will shift the effective correction strength and the observed settling time.

Step 3 — Account for injection ratio and DC loop gain: the observable recovery speed depends on how strongly the servo correction couples into the main loop at DC. A “quiet” servo often uses a small injection ratio to avoid noise coupling; that improves transparency but increases the time needed to correct baseline. Conversely, a strong injection ratio corrects faster but increases risk of LF interaction and wind-up during clipping.

Step 4 — Verify against real failure modes (must pass all):

• Startup pop: if pop is unacceptable, add anti-windup limiting and/or controlled startup (preset or mute/ramp) rather than raising f_cservo into the passband.
• Overload recovery: after clipping, ensure the integrator does not saturate and “stick.” Long recovery is usually wind-up, not “normal settling.”
• LF amplitude/phase integrity: confirm no low-frequency swelling/thinning or phase anomalies near f_{passband_min}.
• Noise migration: ensure servo 1/f and resistor noise remain confined below the usable band edge.

H2-5 · Stability: making the servo “invisible” to the main loop

A stable servo is not enough. The design goal is non-interaction: the servo must correct DC/ultra-low-frequency baseline while remaining effectively invisible across the intended passband.

Core principle: enforce frequency-domain separation. The servo loop must be much slower than the main loop’s low-frequency dynamics, so the two loops do not compete for phase margin. When the servo is too fast (or too strongly injected), it can create low-frequency gain ripple, phase warping, or even slow oscillation.

Make the servo invisible (action recipe): keep the servo crossover at least an order of magnitude below the passband minimum; start with conservative injection (small coupling) to protect passband integrity; solve startup/overload behavior with limiting and controlled initial conditions rather than pushing the servo faster into the band edge.

Acceptance tests (prove non-interaction):

• Low-frequency sweep or step response: no LF hump and no phase “kink” near the passband minimum.
• Clip-and-release test: recovery meets the target settle time and shows no slow oscillation.
• Mode/gain switching: baseline returns cleanly without “breathing” or repeated overshoot.
• Environmental drift: baseline correction remains monotonic and stable across temperature/humidity stress.

H2-6 · Noise, distortion, and dynamic-range budget (what the servo secretly breaks)

A DC servo is not free. It can introduce low-frequency noise, create distortion modulation through operating-point movement, and reduce dynamic headroom through saturation and recovery behavior.

Hidden cost #1 — Noise: servo resistors add thermal noise and the servo amplifier adds 1/f noise. Depending on injection point sensitivity, this noise can leak toward the band edge and thicken the low-frequency noise floor. High-value injection networks also become more sensitive to leakage and environmental variation, which can turn into slow baseline wander.

Hidden cost #2 — Distortion: the servo can create a slow “moving operating point.” If the injection path is nonlinear, or the servo output hits a limit during large-signal events, the baseline correction becomes a slow control signal that modulates gain or bias. The result can be time-varying THD/IMD, especially after overload or during mode switching.

Hidden cost #3 — Dynamic range: baseline correction requires real output swing somewhere in the chain. Strong injection ratios can demand larger servo swing, increasing the chance of saturation and long recovery. Dynamic performance is often dominated by how the servo behaves during saturation, not by the steady-state drift spec.

H2-7 · Startup pop, overload recovery, and anti-windup techniques

This chapter focuses on the most field-visible problems: turn-on pop, slow recovery after clipping, and integrator wind-up. The goal is to keep baseline correction effective without letting the servo create audible or measurable transients.

Three practical techniques (and when to use them):

Acceptance tests (prove it is fixed):

• Turn-on: baseline enters the allowed window in the specified time without a visible output step or audible pop.
• Clip-and-release: recovery meets the target settle time and does not show a second jump caused by a railed integrator.
• Mode/gain switching: baseline returns monotonically (no “breathing” or repeated overshoot).
• Repeatability: behavior remains consistent across boards and environmental stress (humidity/contamination sensitivity is a red flag).

H2-8 · Component choices that actually matter (R/C leakage, DA, bias currents)

Many “correct on paper” DC servos fail in real hardware because of component physics: leakage, dielectric absorption (DA), and bias-current induced errors. This chapter provides selection principles that preserve long-term stability and repeatability.

Resistor realities (R): very large resistors reduce loading but increase thermal noise and make the servo more vulnerable to board leakage and humidity. Bias currents flowing through large impedances can directly create a DC error term that the servo then “chases,” reducing the net drift suppression. Practical designs choose R values that balance injection strength, noise, and environmental robustness rather than maximizing resistance by default.

Capacitor realities (C): dielectric absorption behaves like a slow memory path—after a transient, the integrator can show a long “return tail” or subtle baseline rebound. Leakage turns an ideal integrator into a leaky integrator, adding unintended poles/zeros and shifting the apparent corner frequency. Temperature drift changes τ and can make startup/recovery behave differently across conditions.

Servo amplifier realities: the servo path usually does not need high bandwidth; it needs low-frequency correctness. Prioritize low 1/f noise, low input bias current (especially with high impedances), and fast recovery from output limiting. Ensure the amplifier’s input common-mode range supports the servo node voltage under all operating states.