Skip to content

Measurement, Signal Conditioning, Impedance Matching and Feedback

CSIR NET Physical Sciences: measurement chain, bridges, instrumentation amplifiers, impedance matching, loading, feedback topologies, ADCs, PID control.

  • 19 min read
  • Updated
On this page · 16 sections
  1. Prerequisites
  2. The Hook: The Voltmeter That Changed the Voltage
  3. The Measurement Chain
  4. Bridges
  5. Impedance: Loading and Matching
  6. Amplifiers for Measurement
  7. Feedback
  8. Measurement and Control
  9. Digitisation
  10. Memory Tricks
  11. Don’t Confuse
  12. Common Mistakes
  13. Practice
  14. Quick Revision Table
  15. Key Takeaways
  16. What’s Next

Prerequisites

The Hook: The Voltmeter That Changed the Voltage

The Measurement Chain

graph LR
    A[Measurand] --> B[Transducer]
    B --> C[Excitation and bridge]
    C --> D[Amplifier: in-amp or buffer]
    D --> E[Filter: anti-alias, notch]
    E --> F[Sample and hold, ADC]
    F --> G[Processing, display]
    G -. control action .-> H[Actuator: heater, valve, magnet supply]
    H -. acts on .-> A

Signal conditioning is everything between B and F:

FunctionPurposeTypical circuit
Excitationpower a passive sensorconstant-current source (RTD), constant-voltage bridge supply
Conversionturn \( \Delta R \), \( \Delta C \), \( Q \) or \( I \) into a voltageWheatstone bridge, charge amplifier, transimpedance amplifier
Amplificationraise millivolts to the ADC range; reject common-mode signalinstrumentation amplifier
Isolationbreak ground loops, protect from high voltageisolation amplifier, optocoupler, transformer
Linearisationcorrect a nonlinear lawlook-up table, log amplifier, bridge with feedback
Filteringremove noise, prevent aliasingactive low-pass, notch at the mains frequency
Compensationremove a known interferencethermocouple cold-junction compensation, dummy gauge
Transmissionsend the signal over distance without loss4–20 mA current loop, differential lines

Signal recovery, the extraction of a signal buried in noise, continues the chain with averaging, lock-in detection and box-car integration: see Fourier Transforms, Lock-in Detection, Box-car Integration and Modulation.

Static and dynamic characteristics

  • Static: accuracy (closeness to the true value), precision (repeatability), sensitivity, linearity, resolution, hysteresis, zero and span drift.
  • Dynamic: a first-order sensor (thermometer in a bath) responds to a step as \( 1 - e^{-t/\tau} \), reaching 63% in \( \tau \) and 99% in about \( 4.6\tau \); a second-order sensor (accelerometer, galvanometer) has a natural frequency and damping ratio, and is usually designed near critical damping (damping ratio about 0.7) for flat response.

Current loops

Industrial sensors often send their reading as a current from 4 to 20 mA. Current is the same everywhere in a series loop, so cable resistance does not change it. The “live zero” at 4 mA distinguishes a zero reading from a broken wire (0 mA) and leaves enough current to power the sensor. A transmitter spanning 0–100 °C sends \( 4 + 16 \times (T/100) \) mA, so 10 mA means 37.5 °C.

Bridges

The Wheatstone bridge converts a small \( \Delta R \) into a voltage near zero rather than a small change on a large voltage, so the amplifier can apply high gain without saturating. At balance \( R_1/R_2 = R_3/R_4 \), independent of the excitation. For strain gauges the quarter-, half- and full-bridge outputs are \( V_{ex}G\varepsilon \times \frac{1}{4}, \frac{1}{2}, 1 \) (see Transducers and Detectors).

AC bridges compare impedances: the general balance condition \( Z_1Z_4 = Z_2Z_3 \) must hold in both magnitude and phase, so two components must be adjusted. Standard forms: Maxwell (inductance against a capacitance standard), Schering (capacitance and loss angle of dielectrics), Wien (frequency, also the feedback network of the Wien-bridge oscillator), Hay (high-Q inductors). The Wien bridge balances only at \( \omega = 1/RC \) for equal arms, which is why it is also a notch filter.

Ratiometric measurement: if the ADC reference is derived from the same excitation as the bridge, drift in the excitation cancels in the ratio.

Impedance: Loading and Matching

A real source is a Thevenin voltage \( V_s \) behind \( Z_s \); a real load is \( Z_L \). What the load receives depends on what you want.

GoalConditionWhy
Measure a voltage accurately\( Z_L \gg Z_s \) (voltage bridging)\( V_L = V_sZ_L/(Z_s + Z_L) \to V_s \)
Measure a current accurately\( Z_L \ll Z_s \)ammeter must not limit the current
Transfer maximum power\( Z_L = Z_s^* \) (resistive: \( R_L = R_s \))maximum of \( P = V_s^2R_L/(R_s + R_L)^2 \)
Avoid reflections on a line\( Z_L = Z_0 \) of the linesee High-Frequency Devices
Minimise noise contribution of an amplifiersource resistance near \( e_n/i_n \)noise matching; see the noise page

At maximum power transfer only half the source power reaches the load: efficiency 50%. Power systems never operate there; signal systems (antennas, RF stages, weak-sensor preamplifiers) often do.

Transformer matching

An ideal transformer with turns ratio \( n = N_p/N_s \) reflects a load \( Z_L \) on the secondary to

$$\boxed{Z_{\text{in}} = n^2Z_L}$$

To drive an 8 Ω speaker from a 50 Ω source at maximum power, \( n = \sqrt{50/8} = 2.5 \). Matching raises the delivered power from 0.24 W to 0.50 W for a 10 V source in this case. At radio frequencies the same job is done by L-networks, stubs and quarter-wave sections.

Buffers

The voltage follower (gain 1, non-inverting) has very high input impedance and very low output impedance: it transfers voltage from a high-impedance source to a low-impedance load without loading. It has no voltage gain, but it has large power and current gain.

Amplifiers for Measurement

Instrumentation amplifier

The three-op-amp in-amp (derivation on the op-amp page) gives

$$\boxed{V_o = \left(1 + \frac{2R}{R_G}\right)\frac{R_3}{R_2}(V_2 - V_1)}$$

Its virtues for sensors: both inputs are high impedance (no loading of the bridge), gain is set by one resistor \( R_G \) without disturbing the matched resistors that set CMRR, and the first stage provides all its gain to the differential signal while passing common-mode signal at unity gain, so CMRR rises with gain.

Why CMRR matters: a bridge excited at 10 V has both outputs sitting near 5 V (common mode) while the signal is millivolts. The output error from common mode is \( A_{cm}V_{cm} = A_dV_{cm}/\text{CMRR} \). Referred to the input it is \( V_{cm}/\text{CMRR} \). With CMRR = 100 dB (\( 10^5 \)) and \( V_{cm} = 2 \) V, the input-referred error is 20 µV, comparable to a microstrain-level signal.

Isolation amplifiers

Transfer the signal across an insulation barrier (transformer, optical or capacitive coupling) so that input and output grounds are separate. Used for patient-connected and high-voltage measurements and to break ground loops.

Feedback

The basic relation

For a forward gain \( A \) and feedback fraction \( \beta \) (negative feedback),

$$\boxed{A_f = \frac{A}{1 + A\beta}}, \qquad \frac{dA_f}{A_f} = \frac{1}{1 + A\beta}\frac{dA}{A}$$

With \( A = 10^5 \) and \( \beta = 0.01 \), \( A_f = 99.90 \). Halving \( A \) to \( 5 \times 10^4 \) gives \( A_f = 99.80 \): a 0.1% change for a 50% change in the amplifier. That desensitivity, plus bandwidth extension (by \( 1 + A\beta \) for a single-pole amplifier) and reduction of distortion and of disturbances generated inside the loop, is what feedback buys. It does not reduce noise injected at the input along with the signal.

The four topologies

Feedback can sample the output voltage (shunt connection at the output) or the output current (series connection), and mix at the input in series (as a voltage) or in shunt (as a current).

Topology (mixing–sampling)Also calledStabilises\( Z_{\text{in}} \)\( Z_{\text{out}} \)Ideal amplifier type
Series–shuntvoltage-seriesvoltage gain\( \times(1 + A\beta) \)\( \div(1 + A\beta) \)voltage amplifier
Shunt–shuntvoltage-shunttransresistance\( \div(1 + A\beta) \)\( \div(1 + A\beta) \)current-to-voltage converter
Series–seriescurrent-seriestransconductance\( \times(1 + A\beta) \)\( \times(1 + A\beta) \)voltage-to-current converter
Shunt–seriescurrent-shuntcurrent gain\( \div(1 + A\beta) \)\( \times(1 + A\beta) \)current amplifier

Rule: series raises impedance, shunt lowers it, at whichever port. The op-amp non-inverting amplifier is series–shunt (huge input, tiny output impedance); the inverting amplifier and transimpedance amplifier are shunt–shunt (virtual-ground input, low output impedance); an emitter resistor in a common-emitter stage gives series–series feedback.

Stability: when feedback turns positive

The loop gain \( A\beta \) is frequency dependent. If at some frequency its phase reaches \( -180^\circ \) (so the feedback becomes positive) while \( \lvert A\beta\rvert \geq 1 \), the circuit oscillates. That is the Barkhausen criterion, wanted in an oscillator and fatal in an amplifier. Measures of safety: phase margin (\( 180^\circ \) minus the loop phase lag at \( \lvert A\beta\rvert = 1 \); about 45° or more for a well-behaved step response) and gain margin. Each amplifier pole contributes up to 90° of lag, so a loop with three or more poles can always be driven unstable by enough gain. Op-amps are internally compensated with one dominant pole so that they are stable at unity gain, the worst case.

Measurement and Control

Open and closed loop

An open-loop controller applies a preset action (a heater at fixed power) and cannot correct for disturbances. A closed-loop controller measures the output, compares it with the set point and acts on the error \( e = r - y \). For a plant \( G(s) \) and controller \( C(s) \) with unity feedback:

$$\frac{Y}{R} = \frac{CG}{1 + CG}$$

the same form as the amplifier feedback equation.

On–off control

The simplest controller switches the actuator fully on below the set point and off above it, usually with hysteresis (a dead band) to stop rapid chattering. The controlled variable then oscillates about the set point with an amplitude set by the dead band and the system lags. A domestic thermostat works this way.

PID control

$$\boxed{u(t) = K_pe(t) + K_i\int_0^te\,dt' + K_d\frac{de}{dt}}$$
TermEffectSide effect
Proportional \( K_p \)faster response, smaller errorleaves a steady-state offset; too large gives overshoot or oscillation
Integral \( K_i \)removes steady-state error (acts until error is zero)slows settling, can overshoot; integral wind-up when the actuator saturates
Derivative \( K_d \)anticipates, adds damping, reduces overshootamplifies high-frequency noise; usually filtered

For a plant with DC gain \( G_0 \) under proportional control, a step set point leaves a steady-state error fraction

$$e_{ss} = \frac{1}{1 + K_pG_0}$$

so \( K_pG_0 = 18 \) leaves 5.3% and \( K_pG_0 = 99 \) leaves 1%. For a first-order plant of time constant \( \tau \), proportional feedback also speeds the closed loop to \( \tau/(1 + K_pG_0) \). Integral action drives \( e_{ss} \) to zero for a step because a non-zero error would keep the integral growing.

A cryostat temperature controller is the standard laboratory case: sensor (diode or RTD), PID loop, heater. Tuning trades speed against overshoot, and the thermal lag between heater and sensor is what limits how hard the loop can be pushed.

Digitisation

ADC resolution and quantisation noise

An \( n \)-bit converter over full scale \( V_{FS} \) has

$$\text{LSB} = \frac{V_{FS}}{2^n}, \qquad \sigma_q = \frac{\text{LSB}}{\sqrt{12}}, \qquad \text{SNR}_{\text{ideal}} \approx 6.02n + 1.76\ \text{dB}$$

(the SNR formula is for a full-scale sine). A 12-bit, 5 V converter has LSB = 1.22 mV and SNR about 74 dB. Gain before the ADC should make the largest expected signal nearly fill the range; otherwise bits are wasted. A sample-and-hold freezes the input during conversion, and an anti-aliasing low-pass filter removes components above half the sampling rate first (sampling theory is on the Fourier page). Converter architectures (flash, successive approximation, dual slope, sigma–delta) are core-syllabus material.

Memory Tricks

Don’t Confuse

ThisvsThat
Matching for maximum power (\( R_L = R_s \), 50% efficiency)Bridging for voltage measurement (\( R_L \gg R_s \))
CMRR limits error from voltage common to both inputsOffset voltage: error present even with both inputs grounded
Feedback reduces distortion generated inside the loopFeedback cannot remove noise that enters with the signal
Proportional control: steady-state offset \( 1/(1 + K_pG_0) \)Integral control: zero offset for a step
Barkhausen (\( A\beta = 1 \), phase 0 around the loop) for oscillationNegative feedback with phase margin for stability
Accuracy: closeness to the true valueResolution: smallest change that can be seen

Common Mistakes

Practice

All questions below are original practice questions written for this page. They are not from any CSIR NET paper.

Level 1: Foundation

  1. For maximum power transfer from a source of internal resistance \( R_s \), what should the load resistance be, and what is the efficiency?
  2. Which feedback topology raises input impedance and lowers output impedance?
  3. What is the LSB of a 10-bit ADC with a 5 V full scale?
Answers
  1. \( R_L = R_s \); 50%.
  2. Series–shunt (voltage-series), as in the non-inverting amplifier.
  3. \( 5/1024 \approx 4.9 \) mV.

Level 2: Part B Style

Practice Q1. A source of internal resistance 50 Ω drives an 8 Ω load through an ideal transformer. For maximum power transfer the turns ratio \( N_p : N_s \) should be:

(a) 1 : 2.5 (b) 2.5 : 1 (c) 6.25 : 1 (d) 1 : 6.25

Answer: (b) 2.5 : 1

\( n^2 = 50/8 = 6.25 \), \( n = 2.5 \). Option (c) is the impedance ratio, not the turns ratio.

Practice Q2. An amplifier of open-loop gain \( 10^5 \) has negative feedback with \( \beta = 0.01 \). If the open-loop gain falls by 50%, the closed-loop gain changes by about:

(a) 50% (b) 5% (c) 1% (d) 0.1%

Answer: (d) 0.1%

\( A_f = 10^5/1001 = 99.90 \); with \( A = 5 \times 10^4 \), \( A_f = 5 \times 10^4/501 = 99.80 \). The change is 0.1%. (The small-change formula \( (1/(1 + A\beta))\,dA/A \) gives 0.05% for an infinitesimal change; for a 50% drop the exact answer is 0.1%.)

Practice Q3. A 0–100 °C temperature transmitter uses a 4–20 mA loop. A reading of 10 mA corresponds to:

(a) 37.5 °C (b) 50 °C (c) 60 °C (d) 62.5 °C

Answer: (a) 37.5 °C

\( T = 100 \times (10 - 4)/16 = 37.5 \) °C. Option (b) forgets the 4 mA live zero (10/20).

Practice Q4. A first-order plant of DC gain 2 is under proportional control with \( K_p = 9 \). For a unit step in the set point, the steady-state error is:

(a) 0 (b) 5.3% (c) 10% (d) 11%

Answer: (b) 5.3%

\( e_{ss} = 1/(1 + K_pG_0) = 1/19 = 0.053 \). Option (d) is \( 1/K_p \); option (a) would need integral action.

Level 3: Part C Style

Practice Q5. A full strain-gauge bridge (gauge factor 2) is excited at 10 V and must measure strains up to \( 1000 \times 10^{-6} \). Its output feeds a three-op-amp instrumentation amplifier (difference stage at unity gain, first-stage resistors \( R = 24.9 \) kΩ) and then a 12-bit ADC with a 0–5 V range. Consider:

  1. The bridge output at full strain is 20 mV.
  2. To make full strain fill the ADC range, the gain resistor must be \( R_G = 200 \) Ω.
  3. The strain resolution (one LSB) is about 0.24 microstrain.
  4. A quarter bridge with the same gain would give the same resolution.

Which are correct?

(a) 1 and 2 only (b) 1, 2 and 3 only (c) 2 and 4 only (d) 1, 2, 3 and 4

Answer: (b) 1, 2 and 3 only

Statement 1: full bridge \( V = V_{ex}G\varepsilon = 10 \times 2 \times 10^{-3} = 20 \) mV. True.

Statement 2: required gain \( 5\ \text{V}/20\ \text{mV} = 250 \). \( 1 + 2R/R_G = 250 \Rightarrow R_G = 2(24.9\ \text{k}\Omega)/249 = 200 \) Ω. True.

Statement 3: full strain spans 4096 codes, so one LSB is \( 1000/4096 \approx 0.244 \) microstrain. True.

Statement 4: a quarter bridge gives a quarter of the output, so at the same gain it uses only a quarter of the range; resolution is 4 times coarser (about 0.98 microstrain). False.

Practice Q6. An instrumentation amplifier with differential gain 100 and CMRR 100 dB amplifies a bridge signal that sits on a common-mode voltage of 2 V. Consider:

  1. The common-mode gain is \( 10^{-3} \).
  2. The output error due to common mode is 2 mV.
  3. Referred to the input, this error is equivalent to a 20 µV differential signal.
  4. Doubling the differential gain at the same CMRR doubles the input-referred common-mode error.

Which are correct?

(a) 1, 2 and 3 only (b) 1 and 2 only (c) 3 and 4 only (d) 1, 2, 3 and 4

Answer: (a) 1, 2 and 3 only

Statement 1: \( A_{cm} = A_d/\text{CMRR} = 100/10^5 = 10^{-3} \). True.

Statement 2: \( A_{cm}V_{cm} = 10^{-3} \times 2 = 2 \) mV. True.

Statement 3: \( 2\ \text{mV}/100 = 20 \) µV, which is simply \( V_{cm}/\text{CMRR} \). True.

Statement 4: the input-referred error \( V_{cm}/\text{CMRR} \) does not depend on \( A_d \) at fixed CMRR. False. (In a real in-amp the CMRR itself improves with gain, which makes it better still.)

Practice Q7. Consider these statements about feedback in measurement circuits:

  1. A transimpedance amplifier for a photodiode uses shunt–shunt feedback, giving low input impedance.
  2. Series–series feedback produces an amplifier whose output current is set by the input voltage, with high output impedance.
  3. Negative feedback with loop gain 1000 reduces noise picked up by the sensor cable by a factor of about 1000.
  4. A three-pole amplifier can oscillate if enough negative feedback is applied.

Which are correct?

(a) 1, 2 and 4 only (b) 1 and 3 only (c) 2, 3 and 4 only (d) 1, 2, 3 and 4

Answer: (a) 1, 2 and 4 only

Statement 1: current is mixed in shunt at a virtual ground and output voltage is sampled in shunt. True.

Statement 2: series mixing (voltage input), series sampling (current output): a transconductance amplifier with output impedance raised by \( 1 + A\beta \). True.

Statement 3: cable pickup enters together with the signal at the input; feedback treats it as signal. False.

Statement 4: three poles give up to 270° of lag, so the loop phase passes \( -180^\circ \) at some frequency; if \( \lvert A\beta\rvert \geq 1 \) there, Barkhausen is met. True.

Quick Revision Table

ItemResult
Loading\( V_L = V_sR_L/(R_s + R_L) \); want \( R_L \gg R_s \) for voltage
Max power\( R_L = R_s \) (\( Z_L = Z_s^* \)), efficiency 50%
Transformer\( Z_{\text{in}} = (N_p/N_s)^2Z_L \)
In-amp gain\( (1 + 2R/R_G)(R_3/R_2) \)
CMRR error (input-referred)\( V_{cm}/\text{CMRR} \)
Feedback gain\( A/(1 + A\beta) \to 1/\beta \)
Desensitivity\( dA_f/A_f = (dA/A)/(1 + A\beta) \)
Topologiesseries ×\( (1 + A\beta) \), shunt ÷\( (1 + A\beta) \)
Oscillation\( A\beta = 1 \) with zero net loop phase (Barkhausen)
P-control offset\( 1/(1 + K_pG_0) \)
PIDP speed, I removes offset, D damps
ADCLSB \( = V_{FS}/2^n \); SNR \( \approx 6.02n + 1.76 \) dB
4–20 mA\( I = 4 + 16x \) mA, live zero
First-order sensor63% in \( \tau \), 99% in \( 4.6\tau \)

Key Takeaways

What’s Next