How do I design a feedback amplifier for wideband flat gain?
Feedback Amplifier Design
Without feedback, a transistor's gain decreases at approximately 6 dB/octave (20 dB/decade) with increasing frequency. This natural rolloff makes wideband amplifier design challenging. Negative feedback creates a flat gain region where the loop gain (open-loop gain minus feedback) is large enough to force the closed-loop gain to follow the feedback ratio, independent of the transistor's exact gain.
Shunt-shunt feedback (a resistor Rf from drain to gate of a FET, with a DC blocking capacitor) is the most common topology. It reduces the gain from the transistor's S21 to approximately Gv ≈ -Rf/Z0 at frequencies where the loop gain is high. The input and output impedances are also modified by the feedback, typically providing a reasonable match to 50 Ω across the bandwidth.
The noise figure of a feedback amplifier is higher than the transistor's NFmin because the feedback resistor contributes thermal noise. For a shunt feedback amplifier: NF ≈ NFmin + (4·Rs)/Rf + terms. With Rf = 500 Ω and Rs = 50 Ω: the feedback adds approximately 0.4 dB to the noise figure. Using a higher Rf reduces noise but limits the gain flatness bandwidth.
Frequently Asked Questions
How flat can I make the gain?
With properly designed feedback: ±0.5 dB over 3-4 octaves (e.g., 0.5-8 GHz). Multi-section feedback networks or reactive element tuning can extend flatness further. Commercial MMIC feedback amplifiers (like ERA series from Mini-Circuits) achieve ±0.5 dB over 5+ octaves.
What is the maximum bandwidth?
Limited by the transistor's fT. The feedback amplifier bandwidth can approach 50-80% of fT with aggressive feedback. For a transistor with fT = 30 GHz, a feedback amplifier can be designed for 0.1-20 GHz bandwidth. Beyond 80% of fT, the loop gain is too low for effective feedback.
Is the feedback amplifier unconditionally stable?
Usually yes. Resistive feedback inherently degrades the S12 isolation less than it degrades S21 gain, so the K factor typically improves with feedback. However, at very high frequencies (near fT), the phase shift of the feedback path can cause conditionally stable behavior. Always verify K across the full frequency range.