Bode Plot

What a Bode plot is, and why two curves

Feed a stable linear system a sine wave and, after the transients die away, the output is a sine wave at the same frequency — but scaled in amplitude and shifted in time. Sweep the input frequency from very slow to very fast and record those two numbers — the amplitude ratio and the phase shift — at every frequency, and you have completely characterised the system. A Bode plot is simply the graph of those two numbers. The top panel plots magnitude in decibels; the bottom panel plots phase in degrees; the shared horizontal axis is frequency on a logarithmic scale, usually spanning four or five decades.

Two design choices make the Bode plot the workhorse of classical control rather than a mere data display. First, magnitude is expressed in decibels, defined as 20·log₁₀ of the gain ratio. Second, frequency runs logarithmically. Together these turn multiplication into addition and curves into straight lines. A real plant cascaded with a controller has an open-loop transfer function that is a product of first- and second-order factors. Taking the logarithm of a product gives a sum, so each pole and each zero contributes its own straight-line asymptote that you literally add to the others. Phase adds directly in degrees with no logarithm needed. The result: a fourth-order loop that would be a hopeless tangle in linear coordinates collapses into five or six line segments you can sketch on the back of an envelope.

The asymptote rules — building the curve from poles and zeros

Every Bode plot is assembled from a handful of canonical building blocks. Write the open-loop transfer function in time-constant (Bode) form, then add up the pieces:

Factor	Magnitude asymptote	Phase contribution	Corner
Constant gain K	Flat at 20·log₁₀K dB	0° (or 180° if K<0)	—
Integrator 1/s	−20 dB/decade, all ω	−90° constant	—
Differentiator s	+20 dB/decade, all ω	+90° constant	—
Real pole 1/(1+s/ωₚ)	0 then −20 dB/dec	0° → −90°, −45° at ωₚ	ω = ωₚ
Real zero (1+s/ω_z)	0 then +20 dB/dec	0° → +90°, +45° at ω_z	ω = ω_z
Complex pole pair	0 then −40 dB/dec	0° → −180°, −90° at ωₙ	ω = ωₙ (peak if ζ<0.5)

The phase transition of a single real pole spreads across roughly two decades — from one decade below the corner frequency to one decade above it — passing through exactly −45° at the corner. That decade-wide phase smear is the single most important fact in the whole subject, because it is the phase lag piling up from your plant's poles that eventually pushes the loop toward the −180° danger line and eats your phase margin. A simple integrator-plus-pole plant already starts at −90° from the integrator and adds another −90° from the pole, so it asymptotes to −180° all by itself before any controller dynamics are even added.

Reading the margins — the two crossover frequencies

Stability of a unity-feedback loop is read off the open-loop Bode plot at two specific frequencies. The gain-crossover frequency ω_gc is where the magnitude curve passes through 0 dB — unity gain, where the loop neither amplifies nor attenuates. The phase-crossover frequency ω_pc is where the phase curve passes through −180°.

Phase margin  PM = 180° + ∠G(jω_gc)        ← measured at the 0 dB crossing
Gain margin   GM = −20·log₁₀|G(jω_pc)| dB   ← measured at the −180° crossing

Intuitively: at ω_gc the loop gain is already unity, so the only thing standing between you and the −1 point (gain 1, phase −180°, the onset of sustained oscillation) is how far the phase still sits above −180°. That gap is the phase margin. At ω_pc the phase has already reached −180°, so the only thing protecting you is how far the gain still sits below unity. That gap, in decibels, is the gain margin. For a textbook minimum-phase loop, both margins positive means stable; either one negative means the closed loop has a right-half-plane pole and will oscillate or run away.

Worked example — a motor position loop

Take a DC servo positioning a load. The plant is an integrator (position is the integral of velocity) cascaded with a motor electrical/mechanical pole at 20 rad/s, and we add a proportional gain K:

Plant:   G(s) = K / [ s · (1 + s/20) ]
Choose:  K = 50

Magnitude pieces:
  Gain 50            → +34 dB flat
  Integrator 1/s     → −20 dB/decade everywhere
  Pole at 20 rad/s   → extra −20 dB/decade above ω = 20

Find gain crossover (|G| = 1, i.e. 0 dB):
  Below the pole, |G(jω)| ≈ 50/ω = 1  →  ω ≈ 50 rad/s  ... but that's past the pole,
  so include it: |G| = 50 / (ω·√(1+(ω/20)²)) = 1
  Solving numerically →  ω_gc ≈ 29 rad/s

Phase at ω_gc:
  ∠G = −90° (integrator) − atan(29/20)   = −90° − 55°  = −145°
  Phase margin PM = 180° − 145° = 35°      ← marginal, ~20% overshoot expected

Phase crossover (∠G = −180°):
  −90° − atan(ω/20) = −180°  →  atan(ω/20) = 90°  →  ω_pc → ∞
  So |G| never reaches 0 dB at −180°  →  Gain margin = ∞ (infinite)

This is the canonical "type-1 system with one extra pole" result: an infinite gain margin (the phase only reaches −180° asymptotically) but a finite, and here marginal, phase margin of about 35°. The design fix is a lead compensator — a zero/pole pair that injects up to about +60° of phase right at ω_gc to push the phase margin up to a healthier 50–55°, trading a little extra high-frequency gain for damping. That is the everyday job a Bode plot is built for: see the deficit, add phase exactly where the gain crosses over.

Loop shaping — bending the curves on purpose

Once you can read a Bode plot you can design with it. The art is called loop shaping, and the rules are physical:

• Low-frequency gain buys accuracy. High loop gain at low frequency drives steady-state error toward zero and rejects slow disturbances. An integrator (1/s) gives infinite DC gain and zero steady-state error to a step — at the cost of −90° of permanent phase lag.
• The crossover region sets stability and speed. Aim to cross 0 dB at a slope of −20 dB/decade. A single-pole-like −20 dB/dec slope at crossover corresponds to about +90° of asymptotic phase headroom and a healthy phase margin; crossing at −40 dB/dec leaves you near −180° and almost no margin. Closed-loop bandwidth is roughly ω_gc, so where you put crossover sets the response speed.
• High-frequency roll-off buys noise rejection. Above crossover you want the gain to fall steeply — −40 dB/dec or more — so sensor noise and unmodelled high-frequency resonances are attenuated. A lag compensator or a low-pass filter does this, but each added pole costs phase near crossover, so it is a balancing act.

A lead compensator (1+s/ω_z)/(1+s/ωₚ) with ω_z < ωₚ raises gain and adds phase around crossover — used to recover phase margin and increase bandwidth. A lag compensator (the same form with ω_z > ωₚ) raises low-frequency gain for accuracy while keeping the crossover region nearly untouched. A PID controller is just a lag-lead pair plus an integrator, and its three gains map onto exactly these Bode-plot moves.

Bode plot versus the alternatives

Property	Bode plot	Nyquist plot	Root locus
Domain	Frequency (jω)	Frequency (jω)	Complex s-plane (gain sweep)
What you read	Gain margin, phase margin, bandwidth, roll-off	Encirclements of −1, absolute stability	Closed-loop pole locations vs gain
Best for	Lead/lag & PID design, measured plants	Open-loop-unstable, conditionally stable loops	Choosing gain, pole placement
Handles RHP poles correctly	No (margins can mislead)	Yes (rigorous count)	Yes
Works from measured data	Yes — sweep a real plant	Yes — same data, replotted	No — needs a model
Reads from the curve	Margins, ω_gc, ω_pc directly	Stability yes/no, less quantitative	Damping & speed vs gain

The three are complementary. Bode and Nyquist plot identical information — both are G(jω) — but Bode separates magnitude and phase so you can read the margins and shape compensation, while Nyquist condenses them into one trajectory whose encirclements of −1 give the rigorous answer for awkward plants. Root locus answers a different question entirely: how the closed-loop poles migrate as you turn the gain knob. In practice you design on the Bode plot, confirm on the Nyquist plot when the loop is conditionally stable, and use root locus to pick the final gain.

Failure modes — where the simple reading lies

• Conditionally stable loops. If the open-loop phase dips below −180° and then climbs back above it, there are two phase-crossover frequencies and the loop is unstable for a band of gains in between. The headline margins can both read positive while the loop is on the edge. Worse, reducing the gain — normally the safe move — can push you into the unstable band. Always sanity-check a multi-resonance plant on Nyquist.
• Non-minimum-phase (right-half-plane) zeros. A boost converter, a tail-rotor-controlled helicopter yaw, or a flexible-arm robot all have RHP zeros. These add phase lag like a pole but bend magnitude up like a zero, so they devour phase margin while masking the loss in the gain plot. They impose a hard upper limit on achievable bandwidth: you cannot cross over much above the RHP-zero frequency without going unstable.
• Time delay. A pure transport delay e^(−sT) is flat at 0 dB (no magnitude effect at all) but contributes phase lag −ωT that grows without bound with frequency. It is invisible on the magnitude plot and lethal on the phase plot — a 10 ms delay costs 57° of phase at 100 rad/s. Networked and sampled-data loops live or die on this term.
• Lightly damped resonances. A complex pole pair with ζ < 0.5 produces a sharp magnitude peak and a fast −180° phase drop. If that peak pokes above 0 dB near crossover, the loop can oscillate at the resonant frequency — the classic source of "singing" in disk-drive head loops and gantry stages.
• Gain-only margin checks. Reporting just the gain margin or just the phase margin can hide trouble. A loop can have a comfortable 12 dB gain margin and a dangerous 15° phase margin simultaneously. Both numbers are needed; the conservative single figure is the vector margin — the shortest distance from the Nyquist curve to −1.

Bode plots in real engineering practice

• Measured frequency response. The greatest practical strength: you do not need a model. A network/dynamic-signal analyzer or a swept-sine on a servo drive measures gain and phase at hundreds of frequencies on the real hardware, plots them as an empirical Bode plot, and reads the margins of the actual loop — bearings, cables, flex, and all. This is how disk drives, hard-disk head positioners, machine-tool axes, and switch-mode power supplies are tuned on the bench.
• Switch-mode power supplies. A buck or boost converter's control-to-output Bode plot, measured with an injection transformer, is the standard acceptance test. Designers target ~60° phase margin and crossover at one-fifth to one-tenth of the switching frequency; the RHP zero of the boost topology shows up as a phase plunge that caps the bandwidth.
• Op-amp stability. The open-loop gain curve of an operational amplifier, rolling off at −20 dB/decade through its unity-gain frequency, is a Bode plot; the rate-of-closure between it and the feedback network's 1/β curve predicts whether the amplifier rings or oscillates.
• Flight control and aeroservoelasticity. Margins are mandated — MIL-F-9490D requires 6 dB gain margin and 45° phase margin across the flight envelope, with structural-mode notch filters shaped directly on the Bode plot to keep wing-bending resonances below 0 dB.
• Audio and instrumentation. A loudspeaker's or amplifier's magnitude-and-phase response is a Bode plot by another name; crossover-network and equalizer design is loop shaping on those curves.

Common pitfalls when using Bode plots

• Plotting the closed loop instead of the open loop. Margins are read off the open-loop transfer function L(s) = G(s)·C(s). Sketch the closed-loop response and the crossover frequencies mean nothing for stability.
• Trusting straight-line asymptotes near a corner. The exact magnitude is 3 dB below the asymptote intersection at a single real pole's corner, and a lightly damped complex pair peaks far above its asymptote. The asymptotes are for sketching; verify the margins on the exact curve.
• Forgetting the delay term. Any sampling, transport lag, or computation delay adds −ωT of phase that the rational transfer function never shows. Tune a sampled loop ignoring it and the real hardware will have far less phase margin than the model promised.
• Crossing over at −40 dB/decade. A steep slope at gain crossover means the phase is already near −180°, leaving almost no phase margin. Aim to cross at −20 dB/decade.
• Reading margins on a conditionally stable or non-minimum-phase loop. When the phase wanders across −180° more than once, or the plant has RHP zeros, the simple two-margin reading can be flatly wrong. Confirm with Nyquist.