Electronics Guide

Series-Shunt Feedback (Voltage Amplifier)

In series-shunt feedback, the feedback network samples the output voltage and subtracts a proportional voltage from the input in series with the source. This topology creates a voltage amplifier with high input impedance and low output impedance. The classic non-inverting operational amplifier configuration exemplifies series-shunt feedback, where the feedback resistor divider samples output voltage and applies it to the inverting input in series with the non-inverting input signal.

Shunt-Series Feedback (Current Amplifier)

Shunt-series feedback samples output current and injects a proportional current back to the input in parallel with the source. This configuration produces a current amplifier with low input impedance and high output impedance. The classic example is the two-transistor feedback pair, in which a resistor returns a sample of the output current from the second stage's emitter to the first stage's input node, yielding a stable current gain that is largely independent of device parameters and temperature.

Series-Series Feedback (Transconductance Amplifier)

Series-series feedback samples output current while applying the feedback signal in series with the input voltage. The result is a transconductance amplifier that converts input voltage to output current with high input impedance and high output impedance. Emitter degeneration is the simplest case: a resistor R_E in the emitter of a common-emitter stage makes the feedback factor equal to R_E, so the stage transconductance falls from g_m to g_m/(1 + g_mR_E) and approaches 1/R_E when g_mR_E ≫ 1. The transconductance then depends on a resistor rather than on a bias current and a temperature-dependent device parameter.

Shunt-Shunt Feedback (Transresistance Amplifier)

Shunt-shunt feedback samples output voltage and returns a current to the input in parallel with the source signal. This topology creates a transresistance amplifier converting input current to output voltage, exhibiting low input impedance and low output impedance. The inverting operational amplifier configuration with its feedback resistor directly connecting output to inverting input demonstrates shunt-shunt feedback, as does the transimpedance amplifier that converts photodiode current to voltage.

Return Ratio and Loop Transmission

Loop gain analysis requires careful consideration of loading effects and of the actual signal path around the feedback loop. The return ratio method breaks the loop at a convenient point, nulls the external input, injects a test signal, and measures what returns; the return ratio is the negative of the returned signal divided by the injected signal, so that positive values again indicate negative feedback. For accurate analysis, the break point must preserve the loading each block presents to the other, which is why a working loop cannot simply be cut open on the bench. The practical answer is injection of a small series voltage, through a transformer or a low-value resistor, at a node where the impedance looking forward is much larger than the impedance looking back, followed by a network analyzer sweep. R. D. Middlebrook's 1975 paper "Measurement of Loop Gain in Feedback Systems" in the International Journal of Electronics gives the dual-injection procedure that removes the residual loading error by combining a voltage injection and a current injection at the same point.

Gain Desensitivity

The most valuable property of negative feedback is reduced sensitivity to variation in the forward amplifier. Sensitivity is the ratio of the fractional change in closed-loop gain to the fractional change in open-loop gain:

S = (dA_closed/A_closed) / (dA/A) = 1/(1 + Aβ)

A fractional change in A therefore reaches the output divided by 1 + Aβ. In the example above, halving the open-loop gain from 10⁵ to 5 × 10⁴, a change of 50 percent, moves the closed-loop gain only from 9.9990 to 9.9980, a change of 0.01 percent. Transistor parameter spread, temperature drift, and aging in the active devices are suppressed in the same proportion. The protection does not extend to the feedback network: the sensitivity of the closed-loop gain to β is −Aβ/(1 + Aβ), which approaches −1, so a one percent error in the divider becomes a one percent error in the gain. Precision resistors, not a precision amplifier, set the accuracy of the circuit.

Phase Margin

Phase margin is 180 degrees minus the loop phase lag at the gain crossover frequency, the frequency where the loop gain magnitude equals unity. A phase margin of zero places the loop exactly on the Barkhausen condition, and a negative phase margin produces a growing oscillation. Practical designs typically target 45 to 60 degrees.

Those numbers are not arbitrary. They come from the second-order prototype loop Aβ = ω_n²/[s(s + 2ζω_n)], for which phase margin maps directly onto the closed-loop damping ratio ζ, and ζ in turn onto the overshoot of the step response, exp(−πζ/√(1 − ζ²)). Working the prototype through gives a phase margin of 45 degrees at ζ = 2^−5/4 ≈ 0.420, which overshoots by 23.3 percent rather than the 20 percent often quoted. The other common design points are 51.8 degrees at ζ = 0.5 with 16.3 percent overshoot, 60 degrees at ζ ≈ 0.612 with 8.8 percent, and 65.5 degrees at ζ = 0.707 with 4.3 percent. Real loops carry additional poles and zeros, and a closed-loop zero in particular raises overshoot above what the poles alone predict, so treat these figures as a calibrated guide rather than a prediction. Higher phase margins improve robustness but cost closed-loop bandwidth.

Gain Margin

Gain margin is the factor by which the loop gain may rise before the loop reaches the Barkhausen condition. It is the reciprocal of the loop gain magnitude at the phase crossover frequency, where the loop phase lag reaches 180 degrees, and it is normally quoted in decibels as −20 log₁₀|Aβ| at that frequency. A loop whose gain has fallen to 0.1 by phase crossover has a gain margin of 10, or 20 dB. Common practice asks for at least 6 dB, and for 10 to 20 dB where component tolerance, temperature range, and load variation are wide. A loop whose phase never reaches 180 degrees has, by this definition, an infinite gain margin, which is one reason the two margins are always reported together: each covers an uncertainty the other misses, and a loop can hold a comfortable figure in both while its Nyquist curve still passes close to the critical point.

Magnitude Plot Interpretation

The Bode magnitude plot shows loop gain in decibels against log frequency. Each real pole bends the asymptote down by a further −20 dB/decade, each real zero bends it up by +20 dB/decade, and a complex pole pair contributes −40 dB/decade. Gain crossover, where the curve passes through 0 dB, sets the closed-loop bandwidth of a loop that crosses only once. Operational amplifier work invites a second reading of the same axes: plot the open-loop gain curve and the 1/β line, which is the noise gain, together. Their vertical separation is the loop gain in decibels, they meet at crossover, and the rate of closure, the difference between their slopes where they meet, indicates relative stability. A closure of 20 dB/decade generally gives comfortable phase margin, while 40 dB/decade signals a loop at the edge.

Phase Plot Analysis

The phase plot displays total loop phase shift versus frequency. Each left-half-plane pole contributes up to 90 degrees of lag, passing through 45 degrees at the pole frequency and completing most of its swing within a decade on either side. A left-half-plane zero contributes the same amount of lead. A right-half-plane zero is the trap: it raises the magnitude like any zero but adds lag like a pole, which is why boost and buck-boost converters operating in continuous conduction must cross over well below their right-half-plane zero. The phase at gain crossover gives the phase margin directly. Recognizing how individual poles and zeros shape the phase response lets engineers diagnose a stability problem by inspection and place compensation where it will do the most good.

Asymptotic Approximations

Bode analysis employs straight-line asymptotic approximations that simplify hand analysis. Magnitude asymptotes break at pole and zero frequencies, with slopes changing by 20 dB/decade per singularity; the straight lines are worst at the break itself, where the true curve of a single real pole lies 3.0 dB below the corner, and they come within 1 dB an octave away. Phase asymptotes hold the phase constant below one-tenth of the pole frequency, hold it constant above ten times the pole frequency, and run a straight line between, an approximation whose largest error is about 5.7 degrees. Both errors are small enough that a hand sketch settles most compensation decisions, and large enough that a design sitting on a 45-degree phase margin deserves a numerical check before release.

Nyquist Diagram Construction

The Nyquist diagram plots the loop gain as a complex number, real part against imaginary part, as s traverses a closed contour that runs up the imaginary axis from −j∞ to +j∞ and closes through an infinite semicircle in the right half-plane. Poles sitting on the imaginary axis, such as the integrator at the origin in a type 1 loop, are excluded by indenting the contour around them with a small semicircle, which maps to a large arc on the diagram. For a physical system the plot below the real axis mirrors the plot above it, so measuring or computing magnitude and phase at discrete positive frequencies and reflecting the result suffices. The resulting curve reveals both stability and robustness through its relationship to the critical point.

Encirclement Counting

The criterion states that Z = N + P, where Z is the number of closed-loop poles in the right half-plane, P is the number of open-loop right-half-plane poles, and N is the number of clockwise encirclements of the −1 point. Stability requires Z = 0. For the common case of no open-loop right-half-plane poles, that means zero net encirclements. An open-loop unstable plant, by contrast, requires exactly P counterclockwise encirclements, which is why such loops need a minimum gain as well as a maximum. The shortest distance from the curve to the −1 point is the vector margin, and it equals the reciprocal of the peak closed-loop sensitivity, so a curve that grazes the critical point implies a large sensitivity peak and a poorly damped response even where the gain and phase margins read well.

Conditionally Stable Systems

Some feedback systems exhibit conditional stability, where stability depends on the loop gain remaining within certain bounds. These systems may be stable at nominal gain yet become unstable if the gain falls, through saturation, slew limiting, or a startup transient, as readily as if it rises. The Nyquist plot reveals the condition through multiple crossings of the negative real axis, with crossings that lie beyond −1 cancelling in the encirclement count. Such designs demand analysis at every operating point and tolerance corner, and they demand attention to the large-signal path, because a conditionally stable loop can latch into oscillation after a transient that a linear analysis never sees.

Construction Rules

Root locus plots follow systematic rules that enable hand sketching. Loci begin at the n open-loop poles, where the gain is zero, and terminate at the m open-loop zeros or at infinity, where the gain is infinite. The number of branches equals n. A point on the real axis lies on the locus only where an odd number of poles and zeros lies to its right. The n − m branches that run to infinity follow asymptotes leaving the real axis at the centroid, the sum of the poles minus the sum of the zeros divided by n − m, at angles of (2k + 1) × 180 degrees divided by n − m. Break-away and break-in points occur where branches leave or join the real axis.

Stability Assessment

Closed-loop stability requires all poles to remain in the left half of the s-plane. The root locus reveals the gain value at which poles cross the imaginary axis into the right half-plane, indicating the onset of instability. The frequency of that crossing is the frequency at which the unstable loop will oscillate. Because a ray from the origin at an angle θ from the negative real axis marks a constant damping ratio, with ζ = cos θ, reading damping off the locus is a matter of measuring an angle, and a designer can pick a gain that meets damping and natural-frequency targets directly on the plot.

Compensation Effects

Adding compensation poles and zeros reshapes the root locus, enabling designers to achieve pole locations impossible with the uncompensated system. Lead compensators add a zero that attracts loci toward the left, improving stability and speed. Lag compensators add a low-frequency pole-zero pair, the zero close to the pole, that raises the low-frequency gain and therefore reduces steady-state error without significantly disturbing the dominant poles. Root locus visualization makes the effects of different compensation strategies immediately apparent, and it shows the cost as well: the lag pair leaves a slow closed-loop pole near the origin that adds a small, long tail to the step response.

Dominant Pole Compensation

Dominant pole compensation introduces a single low-frequency pole that reduces loop gain at all frequencies above it. As the simplest form of lag-type compensation, it rolls the gain off before the problematic high-frequency poles contribute significant phase shift, ensuring adequate phase margin at crossover. The compensation pole sits well below the original crossover, which reduces closed-loop bandwidth but buys stability. The internally compensated operational amplifier is the everyday example: a capacitor around the second stage places a pole in the single-hertz range, and the open-loop gain then falls at 20 dB/decade across the remaining decades to unity, so the part is stable at any noise gain down to one.

Lead Compensation

Lead compensation adds phase lead in the crossover region by introducing a zero below crossover and a pole above it. The phase boost from the zero improves phase margin, while the higher-frequency pole limits the noise gain that the zero would otherwise leave rising. The maximum lead occurs at the geometric mean of the zero and pole frequencies, √(ω_zω_p), and its size follows from the ratio α = ω_p/ω_z as arcsin[(α − 1)/(α + 1)]: a ratio of 4 yields 36.9 degrees and a ratio of 10 yields 54.9 degrees. Ratios beyond roughly 10 return little further phase for a great deal of high-frequency gain, so designs that need more lead use two cascaded sections. Lead compensation can raise bandwidth relative to the uncompensated loop while holding adequate margins.

Lead-Lag Compensation

Complex systems often require combined lead-lag compensation to meet multiple performance objectives. The lag portion increases low-frequency gain for improved steady-state accuracy, while the lead portion provides phase boost for adequate stability margins. Proper design requires careful frequency placement to prevent interaction between the two sections: the lag zero normally sits a decade or more below crossover, so that its residual phase lag, a few degrees at most, does not consume the lead section's contribution. This technique is common in precision servo systems requiring both accuracy and speed.

Pole-Zero Cancellation

When problematic poles in the forward path cause stability or response issues, compensation zeros can be placed to approximately cancel them. Exact cancellation is impossible, because component tolerance, temperature, and aging move the plant pole while the compensating zero stays put. What remains is a closely spaced pole-zero pair, a dipole, whose closed-loop pole carries a small residue but a long time constant, producing a slow tail on the step response that no amount of loop gain removes. Pole-zero cancellation therefore requires explicit analysis of sensitivity to component variation, and it must never be applied to a right-half-plane pole, where imperfect cancellation leaves an unstable mode that is invisible in the cancelled transfer function and fatal in the circuit.

Rate of Closure

The rate of closure is the difference in slope between the loop gain magnitude and the 1/β line where the two meet, and it serves as a quick proxy for phase margin. A loop whose gain falls at 20 dB/decade through crossover, with every other singularity far away, approaches 90 degrees of phase margin; one falling at 40 dB/decade approaches zero. Both limits are asymptotic, and nearby singularities move the result by a computable amount: a second pole one decade above crossover still costs 5.7 degrees, and a pole sitting exactly at crossover costs 45 degrees. Compensation therefore aims for 20 dB/decade closure, adding zeros to cancel excess poles in the crossover region. Because the rate of closure reads off the same plot as the gain, it makes a fast check during design iterations, to be confirmed on the phase plot before the design is committed.

Bandwidth Selection

Closed-loop bandwidth selection involves trade-offs between response speed, noise immunity, and stability robustness. Higher bandwidth provides faster response but admits more high-frequency noise and leaves less margin against parasitic poles. Lower bandwidth improves noise rejection and stability but slows the response. The noise cost is quantifiable: a single-pole closed-loop response integrates white noise over an equivalent noise bandwidth of (π/2) × f_{3 dB}, about 1.57 times the nominal bandwidth, so the integrated noise grows as the square root of the bandwidth chosen. The best bandwidth depends on the signal spectrum, the noise environment, and the dynamic requirements of the specific application.

Distortion Reduction

Negative feedback reduces harmonic distortion generated inside the forward path by the factor 1 + Aβ. Two conditions limit the result, and both are routinely overlooked. First, the relevant loop gain is its value at the frequency of the distortion product, not at the fundamental: an amplifier with 60 dB of loop gain at 1 kHz but only 20 dB at 20 kHz suppresses a 20 kHz harmonic by 20 dB, not 60. Second, the derivation is a small-signal one, so it fails once the forward path clips or slew-limits, at which point the loop briefly loses control and the distortion rises abruptly rather than gracefully. Distortion introduced by the feedback network itself is not reduced at all, which is why the divider in a low-distortion amplifier uses metal-film rather than carbon-composition resistors.

Bandwidth Extension

For a forward amplifier with a single dominant pole, A(s) = A₀/(1 + s/ω_p), and a frequency-independent β, feedback divides the DC gain by 1 + A₀β and multiplies the pole frequency by the same factor. The gain-bandwidth product is therefore constant, and reducing closed-loop gain by a factor N raises bandwidth by N. An operational amplifier with a 10 MHz gain-bandwidth product delivers roughly 1 MHz of closed-loop bandwidth at a noise gain of 10 and 100 kHz at a noise gain of 100. The rule holds only for the dominant-pole case: current-feedback amplifiers, whose bandwidth is set chiefly by the feedback resistor, and multi-pole amplifiers near their secondary poles both depart from it. No amount of feedback overcomes the large-signal slew rate limit either, which a bias current and a capacitor set rather than the loop.

Impedance Modification

Negative feedback scales terminal impedances by 1 + Aβ. Series connections at the input multiply the input impedance by that factor and shunt connections divide it; voltage sampling at the output divides the output impedance and current sampling multiplies it. An operational amplifier with 75 Ω of open-loop output resistance, configured for a noise gain of 10 with an open-loop gain of 10⁵, presents 75/(1 + 10⁴) ≈ 7.5 mΩ at DC. The benefit fades with frequency exactly as the loop gain does: for a 10 MHz gain-bandwidth part the loop gain at 100 kHz is only 10, and the closed-loop output impedance has risen to about 6.8 Ω, which is why a measured operational amplifier output impedance rises with frequency and looks inductive. Within its useful band, though, feedback makes a near-ideal voltage source or current source out of an imperfect amplifier.

Noise Considerations

Feedback does not improve the noise performance of the forward amplifier. The amplifier's input-referred noise enters the loop at the same point as the signal, so both are reduced by the same factor and the signal-to-noise ratio at the output is unchanged. What feedback does change is the contribution of everything else. The feedback network itself adds thermal noise, and it is not negligible: a 1 kΩ resistor generates about 4 nV/√Hz at room temperature, comparable to the input noise of a good low-noise amplifier, so a high-value divider can dominate the noise budget of a circuit built around a quiet part. Noise and ripple that enter after the output sampling point, by contrast, are output disturbances, and the loop suppresses them by 1 + Aβ along with every other output error.

Parasitic Elements

Stray capacitance, lead inductance, and parasitic resistance introduce poles and zeros that the schematic does not show. The classic case is capacitance at the summing node. An inverting stage with 10 kΩ of feedback and 10 kΩ to ground, carrying 5 pF of combined amplifier input and board capacitance at the inverting input, places a zero in the 1/β curve at 1/(2π × 5 kΩ × 5 pF) ≈ 6.4 MHz, where 5 kΩ is the two resistors in parallel. Above that frequency the noise gain climbs at 20 dB/decade, so a loop crossing over above 6.4 MHz closes at 40 dB/decade and peaks or rings. A few picofarads across the feedback resistor restore a single-pole closure, and a capacitor chosen so that its product with the feedback resistor equals the product of the parallel resistance and the input capacitance, here 2.5 pF, flattens the noise gain entirely. Layout that keeps the summing node small is the first defense, and compensation that accounts for what remains is the second.

Power Supply Rejection

A loop that senses output voltage rejects supply disturbance automatically, because supply-induced output error is just another output disturbance for the loop to correct. The closed-loop rejection improves on the amplifier's open-loop supply sensitivity by roughly 1 + Aβ, which means rejection falls with frequency exactly as loop gain does. A regulator or amplifier that rejects 100 Hz ripple by 100 dB may reject a 500 kHz switching harmonic by only 20 dB. Sound design compares the loop gain, not merely the DC specification, against the actual spectrum of the supply disturbance, and it treats the supply bypassing and the loop as one problem rather than two.

Load Sensitivity

Changes in load impedance affect the forward gain, and in some topologies the feedback factor as well, so stability margins must be verified at the worst-case load rather than the nominal one. Capacitive loads are the common hazard: the open-loop output resistance and the load capacitance form a pole inside the loop, and 50 Ω driving 1 nF puts that pole at about 3.2 MHz, well inside the crossover region of a fast amplifier. The standard remedies are a small series isolation resistor between the output and the load, which moves the pole outside the loop at the cost of a divider with the load, and a split feedback network that senses the amplifier side of that resistor at high frequency and the load side at DC, preserving DC accuracy while keeping the capacitive pole out of the loop.