In many op-amp applications, wide closed-loop bandwidth isn't as important as low phase shift. Yet in a typical case of an amplifier with a gain of −20, the conventional LM741 has about 0.6° of phase shift at 300 Hz and 6° at 3 kHz. Further, many op amps with enhanced slew rate don't have much better phase lag than standard units. Even a high-speed amplifier like an LM118 will exhibit 0.6° of phase lag at 20 kHz and 6° at 200 kHz, as you can easily observe using a test circuit like that shown in Fig. 1
In Fig. 1's circuit, you apply a test signal through a 20k:1.05k attenuator, feeding the op amp an input signal amplitude of VIN/20 from a Thevenin-equivalent source resistance of 1k. At low frequencies where the amplifier's open-loop gain is very high, the circuit acts as a balanced bridge, and you can observe the null ("PHASE ERROR TEST OUTPUT") on an X-Y oscilloscope display, crossplotted vs. the test input. Normally, you use R1, to trim this low-frequency error to zero. Then, as you increase the frequency and the amplifier begins to contribute phase error, the output will normally lag behind the input. Output phase error, Φ, can be represented as Φ = arctan(2VTEST(p-p)/VIN(p-p)).
For example, with a 20V p-p sine VIN
, the output lag is 5.7° when VTEST
equals 1V p-p and 0.57° when VTEST
equals 0.1V p-p.
If you wish to maintain low phase shift out to 200 kHz, don't try to find an op amp with an enormous gain-bandwidth product. Instead, consider using a phase-compensation circuit to obtain low phase error. Fig. 2's design accomplishes this and boasts minimal peaking and overshoot as well. Here R2 provides an easy trim for low phase error; you don't need variable capacitors. Moreover, the feedforward network connected to pin 8 affords good stability, making a feedback capacitor also unnecessary.
Tests revealed that the amplifier's phase error remained well below 1° from dc to 200 kHz. In-phase error (due to gain peaking) was also low. With VIN=VOUT=20V p-p, observed total error at the null point was <0.1V p-p to 200 kHz. The step response had about 30% overshoot, while the sine response showed about +1 dB of peaking and was 3 dB down at approximately 2 MHz.