In a discrete audio power amplifier, frequency compensation shapes both the amplifier’s transient response and its ability to correct distortion through feedback. Double-pole compensation offers a way to retain more forward-path gain across the audio band than conventional single-pole Miller compensation, but it also changes the loop’s phase and closed-loop response. Engineers evaluating this approach need to consider the input stage, transimpedance stage, output-stage behavior and the full feedback loop together.
How the gain block and transimpedance stage work
A common discrete voltage-gain block begins with a differential input pair operating as a transadmittance stage: its output current responds to the difference between the input and feedback voltages. Emitter-degeneration resistors apply local negative feedback. This reduces transconductance but improves linearity and can make the stage easier to stabilize.
A current mirror converts the differential pair’s output to a single-ended signal and helps balance the pair’s collector currents. The following stage is better understood as a transimpedance stage (TIS) than simply a voltage-amplifier stage. It converts input current into output voltage, while its compensation capacitor provides local feedback. An output buffer then supplies the current needed to drive the load.
In the conventional arrangement, a Miller capacitor across the second stage establishes a dominant pole. Local feedback through that capacitor can reduce the TIS input impedance and improve its linearity. The stage’s local gain also matters: greater open-loop gain around the compensation capacitor can increase separation between the dominant and first non-dominant poles, supporting stability.
What double-pole compensation changes
Single-pole Miller compensation rolls off forward-path gain at about 20 dB per decade. A double-pole network, formed by two capacitors and a resistor, replaces the dominant single pole with a complex-conjugate pole pair. Together, the poles produce an initial roll-off of about 40 dB per decade. A later zero returns the response to a single-pole roll-off.
The practical attraction is that the pole pair can be placed at a higher frequency than a single dominant pole, while maintaining the intended forward-path unity-gain frequency. This can preserve more forward-path gain through the audio band. More gain is available to the global feedback loop, which can reduce output-stage error. It also reduces the current the transadmittance input stage must supply to the compensation network, potentially limiting input-stage distortion associated with that load.
These benefits depend on choosing the components as a network, not treating the capacitors as interchangeable with a Miller capacitor. The forward-path zero must be sufficiently below unity-gain frequency for the phase shift from the initial double-pole roll-off to be limited. The relationship between the two capacitors also affects the unity-gain frequency; for an equivalent comparison with a single-pole design, their series combination is relevant.
Account for peaking, phase margin and transients
Higher forward-path gain does not automatically mean a better-behaved amplifier. The double-pole response can be underdamped and exhibit a resonant peak. In the global feedback loop, the forward-path zero may sit close enough to unity loop-gain frequency to reduce phase margin compared with a single-pole response. Actual stability also depends on poles outside a simplified model, including those associated with the output stage.
A purely resistive global feedback network can also leave an ultrasonic peak in the closed-loop response and produce an underdamped step response. A feedback capacitor across the feedback resistor provides phase-lead compensation: it increases loop transmission at higher frequencies and reduces closed-loop bandwidth. This can suppress the ultrasonic peak while improving phase margin, without lowering gain across the entire frequency range.
That adjustment still requires verification. Phase lead can raise unity loop-gain frequency, so the apparent improvement in phase margin does not by itself establish adequate gain margin once non-dominant poles are considered. Increasing the feedback capacitance can reduce overshoot, but it also increases signal rise time. Component selection therefore involves a trade-off among stability, bandwidth and transient behavior.
Use SPICE loop-gain analysis with care
Forward-path response is useful, but loop transmission is more directly relevant to feedback stability. In SPICE, loop gain can be estimated by breaking the loop at a suitable voltage-transfer point, applying a test signal and measuring the returned signal. The test point should minimize loading changes. In the described approach, the amplifier’s inverting input is suitable when the feedback network has relatively low Thevenin impedance and the input stage is degenerated.
Simulation setup matters. A test source can disturb the DC operating point, so the analysis must preserve the amplifier’s quiescent conditions. A large inductor can isolate the loop at AC while remaining effectively shorted at DC; coupling capacitors can keep the test source from changing the operating point. More exact methods can account for non-ideal impedances at the loop break, but they require careful AC termination.
Small-signal AC analysis is not a complete stability check. It linearizes circuit elements around a fixed operating point and does not fully represent signal-dependent parameters, such as transistor junction capacitances that vary with voltage. Time-domain transient analysis provides a further check. Apply a fast-edged input and inspect output ringing under anticipated loads; the design guidance described here treats settling within fewer than four ringing peaks as an indication of acceptable margins. Simulations remain dependent on the accuracy of the device models.
Practical implications for amplifier design
Double-pole compensation can improve audio-band feedback conditions in a discrete power amplifier, but it is not a drop-in route to lower distortion or guaranteed stability. Its value depends on the complete gain block: input-stage degeneration, TIS gain and impedance, output-stage poles, global feedback and any phase-lead network all affect the result.
For design teams, the practical approach is to compare compensation networks at the same intended unity-gain frequency, examine loop gain and closed-loop response, and verify transient behavior in simulation. Assessing those elements together helps determine whether the added forward-path gain is useful without overlooking ultrasonic peaking, reduced phase margin or slower settling.