Time Constants and Doublets
Estimate both ends of an amplifier's band from one time constant per capacitor. Then find the pole-zero pair that a Bode plot hides.
You Will Be Able To
- Estimate the upper −3 dB frequency from open-circuit time constants and name the capacitor that limits it.
- Estimate the low corner of an AC-coupled stage from short-circuit time constants.
- Find a pole-zero doublet from the settling error of a step.
- Check a gain, a resistance or a sensitivity with one
tf,sensorpzline.
- Lesson
- E9.7
- Time
- 35 min
- Runs on
- ngspice
- Check
- 5 questions
A buffer passed review with 84° of phase margin and settled to 1 % in 0.65 µs. The converter behind it needed 0.1 %, and that took 3.79 µs where the design promised 1.00 µs.
- buffer A
- buffer B
- Time 660 nsbuffer A = -1.08% · buffer B = -0.9278%
Both errors pass −1 % near 0.66 µs. A 1 % test sees no difference.
- Time 996.6 nsbuffer A = -0.1001% · buffer B = 0.4899%
A is inside 0.1 % at 1.00 µs and stays there.
- Time 1.18 µsbuffer A = -0.02826% · buffer B = 0.5544%
B overshoots to +0.55 %, then comes back slowly.
- Time 3.786 µsbuffer A = -0.001% · buffer B = 0.1%
B enters the 0.1 % band at 3.79 µs, 3.8 times later than A.
Read the plot as a table
| Time | buffer A | buffer B |
|---|---|---|
| 660 ns | -1.08% | -0.9278% |
| 996.6 ns | -0.1001% | 0.4899% |
| 1 µs | -0.09778% | 0.4929% |
| 1.18 µs | -0.02826% | 0.5544% |
| 3.786 µs | -0.001% | 0.1% |
| 10 µs | -0.001% | 0.0003051% |
One run, time on a log axis.
- |T| of A
- |T| of B
- phase of A
- phase of B
- Frequency 104.7 kHz|T| of A = 19.6 dB · |T| of B = 19.19 dB · phase of A = -90.59° · phase of B = -93.32°
Near the pair, the two phase curves are 2.7° apart at most.
- Frequency 1 MHz|T| of A = -0.04322 dB · |T| of B = -0.8621 dB · phase of A = -95.71° · phase of B = -96.28°
Above the pair, B has 0.83 dB less gain. Both loops cross 0 dB near 1 MHz.
Read the plot as a table
| Frequency | |T| of A | |T| of B | phase of A | phase of B |
|---|---|---|---|---|
| 1 kHz | 60 dB | 60 dB | -89.43° | -89.48° |
| 10 kHz | 40 dB | 39.99 dB | -90° | -90.52° |
| 100 kHz | 20 dB | 19.61 dB | -90.57° | -93.29° |
| 104.7 kHz | 19.6 dB | 19.19 dB | -90.59° | -93.32° |
| 1 MHz | -0.04322 dB | -0.8621 dB | -95.71° | -96.28° |
| 10 MHz | -23.01 dB | -23.84 dB | -135° | -135.1° |
| 100 MHz | -60.04 dB | -60.87 dB | -174.3° | -174.3° |
Phase margin: 84.3° for A and 84.2° for B.
The step shows a slow tail that the Bode plot hides. This lesson gives quick estimates for the band, then finds that tail.
Open-Circuit Time Constants
- C_1 alone
Remove CF and CL. C1 sees only RS.
- C_L alone
CL sees the resistance at the output: RD beside the transistor's ro.
- C_F alone
CF sits across the gain. It sees RS + Rout + |A|·RS, the Miller effect of E6.1.
Read the circuit as text
- v_in voltage source: + →
n1, − →0 - R_S resistor (100 kΩ): pin 1 →
g, pin 2 →n1 - C_1 capacitor (50 fF): pin 1 →
g, pin 2 →0 - C_F capacitor (20 fF): pin 1 →
n2, pin 2 →g - M1 NMOS transistor (2/0.15): gate →
g, drain →n2, source →0 - R_D resistor (36 kΩ): pin 1 →
VDD, pin 2 →n2 - C_L capacitor (200 fF): pin 1 →
n2, pin 2 →0 - out port: terminal →
n2
Three capacitors, two nodes.
An open-circuit time constant is one capacitor times the resistance it sees with every other capacitor removed. Add them all. The upper −3 dB frequency is close to 1/(2π·Στ).
- Step
- Measure
- Structure
- Analysis
- One capacitor alone
With the others open, C1 sees only RS. CF sees 854.5 kΩ.
- One time constant each
Each capacitor gives one term. The largest term names the limit.
- Add, then invert
The three terms add to 26.69 ns. That gives 5.964 MHz.
Read the diagram as text
- Pick one capacitor: one at a time
- Open the others: remove every other capacitor
- Find the resistance: what this capacitor sees
- Multiply: time constant = R times C
- Add them all: one term per capacitor
- Estimate the corner: 1 over 2π times the sum
- Pick one capacitor → Open the others
- Open the others → Find the resistance
- Find the resistance → Multiply: R
- Multiply → Add them all: τ
- Add them all → Estimate the corner: Στ
The table below fills in each step.
| Capacitor | Resistance it sees | Time constant |
|---|---|---|
| C1 = 50 fF | RS = 100 kΩ | 5.000 ns |
| CL = 200 fF | Rout = 22.98 kΩ | 4.596 ns |
| CF = 20 fF | 100 k + 22.98 k + 7.315 × 100 k = 854.5 kΩ | 17.09 ns |
tf printed the gain, −7.315, and Rout.
Why does CF see so much? Push a test current i into its gate end and take it out at its drain end. The gate rises by i·RS. The drain falls by i·Rout, and by the gain times the gate rise. The voltage across CF is the sum of the three.
The sum is 26.69 ns, so the estimate is 5.964 MHz. The AC run measures 6.122 MHz.
The smallest capacitor holds 64 % of the sum. The table tells you which part to change first.
pz finds the true poles: 6.139 MHz and 115.4 MHz. The sum of time constants also counts the second pole, so the estimate is 2.6 % low.
Short-Circuit Time Constants
Read the circuit as text
- v_in voltage source: + →
n1, − →0 - C_1 capacitor (10 pF): pin 1 →
n2, pin 2 →n1 - R_B resistor (1 MΩ): pin 1 →
n3, pin 2 →n2 - VB port: terminal →
n3 - M1 NMOS transistor (2/0.15): gate →
n2, drain →n4, source →0 - R_D resistor (36 kΩ): pin 1 →
VDD, pin 2 →n4 - C_2 capacitor (330 pF): pin 1 →
n5, pin 2 →n4 - R_L resistor (100 kΩ): pin 1 →
n5, pin 2 →0 - out port: terminal →
n5
VB = 0.7815 V sets the gate.
At low frequency a coupling capacitor blocks the signal. For each one, short the others and find the resistance it sees. Each gives a corner at 1/(2π·R·C), and the corners add.
C1 sees RB. C2 sees Rout + RL.
- Frequency 15.92 kHzgain = 12.22 dB
C1 alone gives a corner at 15.9 kHz.
- Frequency 16.8 kHzgain = 12.47 dB
The gain is 3 dB below its mid-band 15.48 dB at 16.80 kHz.
- Frequency 19.84 kHzgain = 13.16 dB
The two corners add to 19.84 kHz. The estimate is 18 % high.
Read the plot as a table
| Frequency | gain |
|---|---|
| 100 Hz | -60.42 dB |
| 1 kHz | -20.71 dB |
| 10 kHz | 9.384 dB |
| 15.92 kHz | 12.22 dB |
| 16.8 kHz | 12.47 dB |
| 19.84 kHz | 13.16 dB |
| 100 kHz | 15.37 dB |
| 1 MHz | 15.48 dB |
| 10 MHz | 15.48 dB |
The capacitor with the highest corner sets most of the result.
Both estimates miss toward the safe side. The real band is a little wider than the estimate at each end.
A Doublet Hides in the Bode Plot
A doublet is a pole and a zero close together. Amplifier B has a pole at 100 kHz and a zero at 110 kHz, near a tenth of its unity-gain frequency.
The pole and the zero undo each other, except for 0.83 dB of gain above them. Closing the loop moves the pole to 111.4 kHz. It now sits beside the zero and still does not cancel it exactly.
What is left is a small term that decays with the time constant of that pole, 1.43 µs. The main response has 0.16 µs.
| Buffer | Inside 1 % | Inside 0.1 % | Inside 0.01 % |
|---|---|---|---|
| A, no doublet | 0.670 µs | 0.997 µs | 1.335 µs |
| B, pair at 100 and 110 kHz | 0.653 µs | 3.786 µs | 6.954 µs |
| C, pair at 10 and 11 kHz | 0.729 µs | 0.995 µs | 33.62 µs |
Last time each error enters the band, same run as the first figure.
Each doublet fails only the accuracy level near its own tail size. Measure settling to the accuracy the system needs, in the time domain.
In a real op-amp, a doublet appears where a zero is meant to cancel a pole and misses it.
One-Line Checks: Tf, Sens and Pz
| Line | What it prints | On the 30 kΩ and 10 kΩ divider |
|---|---|---|
tf v(mid) VS | DC gain, input resistance, output resistance | 0.25, 40 kΩ, 7.5 kΩ |
sens v(mid) | Change of the output per unit change of each part | RA: −11.25 µV/Ω, RB: +33.75 µV/Ω |
pz in 0 out 0 vol pz | Poles and zeros, in rad/s | None: the divider has no capacitor |
Each line needs no stimulus beyond the operating point.
- Resistor RA 30 kΩv(mid) = 450 mV
At 30 kΩ the output is 450.0 mV.
sensprints the slope here. - Resistor RA 31 kΩv(mid) = 439 mV
One kilohm more gives 439.0 mV: 11.0 mV lower, close to the slope times 1 kΩ.
Read the plot as a table
| Resistor RA | v(mid) |
|---|---|
| 20 kΩ | 600 mV |
| 25 kΩ | 514.3 mV |
| 30 kΩ | 450 mV |
| 31 kΩ | 439 mV |
| 35 kΩ | 400 mV |
| 40 kΩ | 360 mV |
A sensitivity is exact only for small changes.
Asked in interviews
How do you estimate the bandwidth of an amplifier by hand?
Take each capacitor alone, with the others open, and find the resistance it sees. Add the time constants. The −3 dB frequency is about 1/(2π·Στ). The largest term names the capacitor to fix.
Why is the open-circuit estimate a little low?
The sum of time constants equals the sum of all pole time constants. It charges the dominant pole for the others. Here it gave 5.964 MHz against a measured 6.122 MHz.
What is a pole-zero doublet and why does it matter?
A pole and a zero close together. The Bode plot and the phase margin hardly change. The step response gains a small slow tail, which can multiply the settling time to high accuracy.
Try It
Part 1 runs tf on the amplifier. It prints the DC gain and the resistance at out, which the next lines reuse.
Part 2 writes one time constant per capacitor and adds them. Parts 3 and 4 measure the −3 dB frequency and the poles to compare.
The last lines copy each scalar into the AC plot, because meas reads only vectors of an analysis.
Predict before each change. Halve CF to 10 fF: which time constant falls, and by how much does f3 rise? Set RS to 10k: which capacitor now holds the largest share?
Run the netlist to draw its plots here.
a0 = 1.728449e+01 f3 = 6.121989e+06 gain = 7.315174e+00 r_out = 2.297865e+04 tau_c1 = 5.000000e-09 tau_cf = 1.708992e-08 tau_cl = 4.595730e-09 f_octc = 5.964065e+06 f_pole1 = 6.139193e+06 f_pole2 = 1.154327e+08 div_gain = 2.500000e-01 div_rin = 4.000000e+04 div_rout = 7.500000e+03Check Yourself
Common Mistakes
Wrong Sign off a 0.1 % settling spec from the phase margin, or from a 1 % settling run.
The doublet here moved the phase margin by 0.1° and still made the 0.1 % time 3.8 times longer.
Fix Run the step and measure the last time the error enters the band the system needs.
meas tran ts WHEN v(out)=0.99 RISE=1let e=abs(v(out)-1)
meas tran ts WHEN e=0.001 CROSS=LASTWrong Use the open-circuit rule at the low end: look for the largest time constant.
At the low end the corner frequencies add, so the capacitor with the smallest time constant dominates.
Fix For a high corner add time constants. For a low corner add the corner frequencies, each found with the other capacitors shorted.
Wrong Run pz from a source that has no ac value.
ngspice 42 then aborts with "The input signal is shorted on the way to the output".
Fix Give the input source an ac value, as an AC run needs.
VIN in 0 DC 0
pz in 0 out 0 vol pzVIN in 0 DC 0 AC 1
pz in 0 out 0 vol pzRemember
- Upper corner: one capacitor at a time, the others open, add the time constants. The largest term names the limit.
- Lower corner: one capacitor at a time, the others shorted, add the corner frequencies. The highest one dominates.
- Both estimates are a little pessimistic: 2.6 % low at the top and 18 % high at the bottom in this lesson.
- A doublet near 0.1·fu moved the phase margin by 0.1° and made 0.1 % settling 3.8 times slower.
tf,sensandpzcheck a gain, a resistance, a slope or a pole in one line.