Signal conditioning and data acquisition

In short

Signal conditioning is everything done to a raw sensor output - amplify, filter, linearise, isolate, convert, level-shift - so that it becomes a clean, standard signal an ADC or controller can read correctly. Data acquisition (DAQ) is the chain that then samples, digitises and stores it: transducer → conditioner → multiplexer → sample-and-hold → ADC → processor/SCADA. For an Instrumentation and Control student this is the spine of the discipline: it ties sensors, op-amps, ADCs and control systems together, and it is where GATE (IN and EC/EE) sets numerical questions on bridge sensitivity, aliasing, quantisation error and ADC conversion time. It is asked in every interview at L&T, Honeywell, Yokogawa, Emerson, ABB, Siemens, Endress+Hauser, BHEL, NTPC, ISRO/DRDO and National Instruments-style roles, because a plant engineer spends most working hours on loops from the field to the DCS.

Explained

Why conditioning is needed, and the signal chain

• A raw sensor output is usually tiny (thermocouple ~40 µV/°C, strain gauge bridge a few mV full scale), high- or low-impedance, nonlinear, riding on a common-mode voltage, and buried in noise - none of which an ADC can handle directly.
• Standard chain: transducer → signal conditioning (amplify/filter/isolate/linearise) → analog multiplexer → sample-and-hold (S/H) → ADC → digital processor / memory / display → SCADA or DCS.
• Functions of signal conditioning: amplification, attenuation, filtering, impedance matching (buffering), isolation (galvanic), excitation supply (bridge, RTD current source), linearisation, cold-junction compensation, current-to-voltage or voltage-to-frequency conversion, and level shifting to the ADC's input span.
• Single-ended vs differential: differential inputs reject the voltage common to both wires (ground loops, 50 Hz pickup); almost all field measurement is differential.
• Order matters: amplify before long cable runs and before the ADC (to use the full ADC span), filter before sampling (anti-aliasing filter must be analog, ahead of the S/H), isolate where a ground loop or hazardous voltage exists.

Amplifiers and bridges

• Instrumentation amplifier (3 op-amp): two buffer op-amps with a single gain resistor R_G, then a unity difference amplifier. Very high input impedance, high CMRR (80-120 dB), gain set by one resistor - the default front end for bridges, thermocouples and shunts.
• Difference (subtractor) amplifier alone loads the source and needs precisely matched resistors; it is used only after buffering.
• CMRR = differential gain / common-mode gain, in dB. A 100 dB CMRR amplifier with 10 V of 50 Hz common mode passes only 100 µV of error.
• Chopper-stabilised / auto-zero amplifiers for DC microvolt signals, where op-amp offset drift (µV/°C) would otherwise swamp the signal.
• Wheatstone bridge: converts resistance change into voltage. Quarter bridge (one active gauge), half bridge (two, gives temperature compensation and twice the output), full bridge (four, four times the output and best linearity) - used in load cells, pressure transmitters, torque sensors.
• Strain gauge gauge factor GF = (ΔR/R)/ε, about 2.0 for metal foil, 50-200 for semiconductor (piezoresistive) gauges.
• RTD wiring: 2-wire (lead resistance adds error), 3-wire (bridge cancels one lead), 4-wire Kelvin (current through two leads, voltage sensed on the other two - lead resistance error eliminated; the lab standard). PT100 = 100 Ω at 0 °C, PT1000 = 1000 Ω; α = 0.00385 Ω/Ω/°C per IEC 60751 (adopted in India as the corresponding IS/IEC standard).
• Thermocouple conditioning: needs cold-junction compensation (CJC) - measure the terminal-block temperature with a thermistor/IC sensor and add the equivalent emf in software or with a compensating bridge - plus open-thermocouple (burnout) detection and an upscale/downscale drive on break. Types K, J, T, R, S, B per IEC 60584.
• Charge amplifier for piezoelectric accelerometers and pressure sensors: converts charge to voltage, V_out = −Q/C_f, so cable capacitance no longer affects sensitivity.

Filtering, noise and isolation

• Anti-aliasing low-pass filter is mandatory before sampling; cut-off set at or below f_s/2. Common choices: Butterworth (flat passband), Chebyshev (sharper roll-off, passband ripple), Bessel (linear phase - best for transient shape).
• Roll-off = 20n dB/decade (6n dB/octave) for an n-pole filter; to reject by a given amount at f_s/2 you compute the needed order.
• Notch filter at 50 Hz (India's mains frequency) for biomedical and low-level DC measurement; dual-slope and sigma-delta ADCs reject mains by integrating over a multiple of 20 ms (NMRR - normal mode rejection).
• Noise sources: thermal (Johnson) noise v_n = √(4kTRB), shot noise, 1/f flicker noise, mains hum, EMI from VFDs and contactors, and ground loops.
• Noise remedies: twisted pair (cancels magnetically induced emf), overall and individual shield grounded at one end only, separate instrument and power cable trays, differential/4-wire measurement, star (single-point) earthing, correct cable glands, and segregation per the plant's earthing practice (IS 3043 for earthing).
• Isolation: optical (optocoupler), magnetic (transformer), capacitive, or digital isolators. Purposes - break ground loops, protect the DAQ and the operator from high common-mode or fault voltage, and satisfy intrinsic-safety requirements in hazardous areas (zener barriers / galvanic isolators, Ex 'i' per the IS/IEC 60079 series; equipment certification in India via CCOE/PESO and BIS).
• Guarding and bootstrapping of the cable screen reduces leakage and capacitive loading on very high-impedance sources (pH electrodes, 10⁹ Ω).

Industrial standard signals and transmitters

• 4-20 mA current loop is the workhorse: current is immune to series voltage drop and cable resistance, and the live zero of 4 mA distinguishes "zero measurement" from "broken wire" and powers the transmitter.
• Two-wire transmitter: the same pair carries 24 V DC power and the 4-20 mA signal; the transmitter must live on under 4 mA (<3.5 mA typically). Four-wire transmitters have separate power.
• Loop load limit: 24 V supply, 20 mA, so total loop resistance must be below roughly 600-650 Ω after the transmitter's minimum operating voltage (typically ~10-12 V) is deducted - this is the classic "how many receivers can I put in the loop" question.
• Other signals: 0-10 V and ±10 V (short runs, lab DAQ), 1-5 V (across a 250 Ω resistor in a 4-20 mA loop), 0-20 mA (no live zero - avoid), 3-15 psi pneumatic (legacy), mV/thermocouple, RTD direct, pulse/frequency.
• HART superimposes a 1200 bps FSK digital signal (1200/2200 Hz) on the 4-20 mA loop without disturbing the analog value - used for calibration and diagnostics with a handheld communicator. Foundation Fieldbus / Profibus PA are fully digital, multi-drop; Modbus RTU over RS-485 is the most common plant serial link in India; IO-Link and industrial Ethernet (PROFINET, EtherNet/IP) in factory automation.
• Smart transmitter = sensor + conditioning + ADC + microcontroller + digital communication, with built-in linearisation, re-ranging, damping and self-diagnostics.

Sampling, aliasing and the S/H

• Shannon-Nyquist sampling theorem: a signal band-limited to f_max can be reconstructed exactly if f_s > 2 f_max; 2 f_max is the Nyquist rate and f_s/2 the Nyquist (folding) frequency.
• Aliasing: a component at f > f_s/2 reappears at |f − k·f_s| and is indistinguishable from a genuine low-frequency signal - it cannot be removed afterwards by any digital filter, which is why the analog anti-aliasing filter is non-negotiable.
• In practice f_s is chosen 5-10 × the highest frequency of interest for waveform fidelity, and oversampling is used to relax the analog filter and to improve resolution.
• Sample-and-hold: switch + hold capacitor + buffer. Parameters - acquisition time (to charge C_H to within error band), aperture (sampling) time and aperture jitter, droop rate dV/dt = I_leak/C_H, pedestal/hold-step, feedthrough.
• Aperture error: for a sine of amplitude A and frequency f, ΔV = 2πfA·t_a. Aperture jitter limits the usable input frequency at high resolution, so a simultaneous-sampling DAQ (one S/H + ADC per channel) is used where inter-channel phase matters (power quality, vibration, three-phase measurement); a multiplexed DAQ introduces a skew of one conversion time between channels.
• Large C_H gives low droop but long acquisition time - the basic S/H trade-off.

ADCs, DACs and resolution

• Resolution: n-bit ADC has 2ⁿ codes; step size (LSB) = FSR/2ⁿ. Quantisation error ≤ ±½ LSB; its RMS value = q/√12.
• Ideal SQNR = 6.02n + 1.76 dB for a full-scale sine. ENOB = (SNR_measured − 1.76)/6.02 - the honest resolution after noise and distortion.
• ADC types and where each is used:
  • Flash (parallel): 2ⁿ−1 comparators, one clock cycle - fastest, costliest, low resolution (6-8 bit); digital oscilloscopes, radar.
  • Successive approximation (SAR): binary search with an internal DAC, n clock cycles, needs an S/H; 8-18 bit, µs range - the standard microcontroller/DAQ card ADC.
  • Dual-slope integrating: integrates input then references; extremely good 50 Hz noise rejection if integration time = multiple of 20 ms; slow; used in 4½-digit bench DMMs.
  • Sigma-delta (ΣΔ): 1-bit modulator at a high oversampling rate + digital decimation filter; 16-24 bit, inherent anti-aliasing and noise shaping; the choice for temperature, strain and audio. Slow settling on channel switching.
  • Counter/ramp and tracking (servo) ADC: simple, 2ⁿ worst-case cycles; tracking type is fast for slowly varying inputs.
  • Pipeline: several low-resolution stages, high speed at 12-16 bit, with latency; used in communications and imaging.
• ADC errors: offset, gain (full-scale) error, DNL (differential nonlinearity - causes missing codes if < −1 LSB), INL, monotonicity, and temperature drift. Specified in LSB or % FSR.
• DAC for output: weighted-resistor (poor scalability) and R-2R ladder (only two resistor values - the practical choice); settling time and glitch energy matter. A reconstruction (smoothing) low-pass filter follows the DAC.
• V/f converter gives an easy, cheap, isolation-friendly route: frequency is counted over a gate time, so it is inherently integrating and noise-averaging - used in isolated and remote measurement.
• Multiplexer issues: on-resistance, crosstalk, charge injection, settling and the need for a buffer between MUX and ADC; throughput per channel = total sampling rate / number of channels.

Calibration, specification and Indian practice

• Calibration = comparing the instrument against a traceable standard and recording the deviation; adjustment/trimming changes the instrument. Traceability in India runs to NPL (CSIR-National Physical Laboratory, New Delhi) through NABL-accredited labs (ISO/IEC 17025).
• Typical loop calibration: five-point check at 0, 25, 50, 75, 100% rising and falling, computing zero error, span error, linearity, hysteresis and repeatability; results on a calibration certificate with the uncertainty budget.
• Accuracy vs precision vs resolution vs sensitivity: accuracy is closeness to true value (often quoted as ±% of span or ±% of reading), precision is repeatability, resolution is the smallest discernible change, sensitivity is output per unit input.
• Static characteristics: static error, dead zone, threshold, drift, hysteresis, linearity, span, range, reproducibility. Dynamic: time constant, settling time, rise time, bandwidth, slew rate.
• Documentation: P&ID and instrument tag numbers per ISA-5.1, loop drawings, instrument index, hook-up drawings; functional safety and SIL assessment per the IEC 61508 / 61511 series (adopted as the corresponding IS/IEC standards), EMC immunity testing per IEC 61000-4 series, enclosure protection per IS/IEC 60529 (IP code).
• Legal metrology for trade measurement (weighbridges, fuel dispensers, flow meters for billing) comes under the Legal Metrology Act, 2009 and its Rules, with model approval and stamping/verification by the state Legal Metrology department.

Key points

• Signal conditioning - amplify, filter, isolate, linearise and level-shift a sensor output into the ADC's usable span.
• Instrumentation amplifier - high Z_in, one-resistor gain, CMRR 80-120 dB; the standard bridge/thermocouple front end.
• CMRR - differential gain ÷ common-mode gain in dB; the single most important front-end spec in a noisy plant.
• Wheatstone bridge - quarter/half/full; full bridge gives 4× quarter-bridge output and best linearity.
• Gauge factor - GF = (ΔR/R)/ε; ≈2 for foil gauges, 50-200 for piezoresistive.
• 4-wire RTD - eliminates lead-resistance error; 3-wire cancels one lead; 2-wire does neither.
• Cold-junction compensation - a thermocouple reads the difference of two junctions, so the reference-junction temperature must be measured and added back.
• 4-20 mA with live zero - immune to cable resistance and lets a broken wire (0 mA) be told apart from a true zero.
• Two-wire transmitter - power and signal on one pair; loop resistance limited to roughly 600 Ω on 24 V.
• HART - 1200 bps FSK digital data riding on the 4-20 mA analog loop; the analog reading is untouched.
• Nyquist criterion - f_s > 2 f_max; the anti-aliasing filter must be analog and precede the S/H.
• Aliasing is irreversible - no digital filter can undo it once the sample is taken.
• Quantisation error - ±½ LSB maximum, RMS q/√12; SQNR = 6.02n + 1.76 dB; ENOB is the real-world figure.
• SAR ADC - n clock cycles, needs an S/H, 8-18 bit; flash is fastest; dual-slope best rejects 50 Hz; ΣΔ gives 24-bit resolution by oversampling.
• S/H droop - dV/dt = I_leak/C_H; bigger C_H lowers droop but lengthens acquisition time.
• Multiplexed vs simultaneous sampling - multiplexing costs inter-channel skew; simultaneous sampling is needed for phase-critical measurement.
• Isolation - optical, magnetic or capacitive; breaks ground loops and is mandatory for hazardous-area and high-voltage interfacing.
• Traceability - calibration chains to NPL New Delhi via NABL-accredited (ISO/IEC 17025) laboratories.

Formulas and examples

Example 1 - ADC resolution and quantisation error. A 12-bit ADC has a 0-10 V unipolar input range.

• Codes = 2¹² = 4096; q = 10/4096 = 2.441 mV.
• Maximum quantisation error = ±q/2 = ±1.221 mV; RMS = 2.441/√12 = 0.705 mV.
• Ideal SQNR = 6.02 × 12 + 1.76 = 73.98 dB.
• If a 0-50 mV thermocouple signal were fed directly, it would occupy only 50/2.441 ≈ 20 codes out of 4096 - hence the need for a gain of about 200 ahead of the ADC, which maps 50 mV to the full 10 V span.

Example 2 - strain gauge quarter bridge and required gain. Constantan foil gauge GF = 2.0, V_ex = 5 V, strain = 500 µε, quarter bridge; ADC span 0-2.5 V, 16-bit.

• ΔR/R = GF × ε = 2.0 × 500×10⁻⁶ = 1.0×10⁻³ (0.1%).
• V_o = 5 × (1.0×10⁻³)/4 = 1.25×10⁻³ V = 1.25 mV.
• For 1.25 mV to reach 2.5 V full scale, gain = 2.5/1.25×10⁻³ = 2000.
• ADC q = 2.5/65536 = 38.1 µV; referred to the bridge that is 38.1/2000 = 19.1 nV, equal to 500 µε × 19.1 nV/1.25 mV ≈ 0.0076 µε - so resolution is limited by amplifier noise, not by the ADC.
• A full bridge would give 5 × 2.0 × 500×10⁻⁶ = 5 mV, four times the output, needing only a gain of 500.

Example 3 - aliasing. A vibration signal containing 3.8 kHz is sampled at 5 kHz with no anti-aliasing filter.

• f_s/2 = 2.5 kHz, and 3.8 kHz > 2.5 kHz, so it aliases.
• f_a = |3800 − 5000| = 1200 Hz - the 3.8 kHz component is recorded as a false 1.2 kHz tone.
• Fix: a low-pass filter with f_c = 2 kHz. To attenuate 3.8 kHz by 60 dB, need 20n log₁₀(3800/2000) = 60 → n = 60/(20 × 0.2788) = 10.8, so an 11-pole filter - impractical, which is why f_s is instead raised (oversample at 20 kHz, where a 4-pole filter suffices) or a ΣΔ ADC is used.

Example 4 - 4-20 mA loop loading. A two-wire transmitter needs minimum 12 V at its terminals; supply is 24 V DC. Receivers: a 250 Ω input at the PLC, a 250 Ω chart recorder, cable resistance 40 Ω.

• Available for the loop = 24 − 12 = 12 V; R_loop,max = 12/0.020 = 600 Ω.
• Present load = 250 + 250 + 40 = 540 Ω → voltage at transmitter at 20 mA = 24 − 540×0.020 = 13.2 V > 12 V, so it works, with 60 Ω (1.2 V) of margin.
• Adding one more 250 Ω receiver gives 790 Ω → 24 − 15.8 = 8.2 V, below 12 V: the loop saturates and the reading collapses at high current.
• At 12 mA the voltage across a 250 Ω receiver = 3.0 V, which is the mid-point of the 1-5 V scale - consistent with 4 mA → 1 V and 20 mA → 5 V.

Example 5 - Pt100 and self-heating, plus S/H droop. Pt100, α = 0.00385, excited with 1 mA.

• At 100 °C, R = 100(1 + 0.00385 × 100) = 138.5 Ω; sensitivity ≈ 0.385 Ω/°C → 0.385 mV/°C at 1 mA - a small signal needing gain ~100.
• Self-heating power = I²R = (1×10⁻³)² × 138.5 = 138.5 µW; with a dissipation constant of 2 mW/°C in still air the error is 0.1385/2 ≈ 0.07 °C - acceptable; at 10 mA it would be 100× larger (≈6.9 °C) and unacceptable.
• S/H: C_H = 1 nF, leakage 10 nA → droop = 10×10⁻⁹/1×10⁻⁹ = 10 V/s. For a 12-bit, 10 V ADC (q/2 = 1.22 mV) the hold must finish within 1.22×10⁻³/10 = 122 µs - which comfortably covers a 12-cycle SAR conversion at 1 MHz (12 µs).

Where it is used

• Process plants - refineries (Reliance Jamnagar, IOCL, BPCL), fertiliser and cement plants: 4-20 mA transmitters for pressure, flow, level and temperature wired to a DCS (Honeywell Experion, Yokogawa Centum, ABB 800xA, Emerson DeltaV).
• Power generation and transmission - NTPC, BHEL and state utilities: thermocouple/RTD scanning in boilers, CT/PT conditioning, 50 Hz-rejecting integrating ADCs in energy meters and power-quality analysers; SCADA/RTU telemetry in substations.
• Test, R&D and aerospace - ISRO, DRDO, HAL, automotive NVH and engine test beds: multi-channel simultaneous-sampling DAQ with strain, accelerometer (charge amplifier) and thermocouple conditioning; LabVIEW/NI, Dewesoft, HBM, Keysight hardware.
• Medical electronics - ECG/EEG front ends: instrumentation amplifier with very high CMRR, 50 Hz notch, isolation for patient safety; built by GE Healthcare Bengaluru, Philips, Siemens Healthineers and Indian OEMs like BPL and Trivitron.
• Weighing, metrology and calibration services - load-cell bridge conditioners and weighbridges under the Legal Metrology Act, 2009; NABL-accredited calibration labs working to traceability from NPL.
• Job roles - instrumentation/C&I engineer, commissioning engineer, loop-check and calibration technician, embedded/mixed-signal hardware engineer, test and validation engineer, DCS/SCADA configuration engineer.

Interview questions

• Q: Why is 4-20 mA preferred over 0-10 V for field signals? A: A current loop is unaffected by cable resistance and series voltage drops, is less sensitive to induced noise, can power the transmitter, and the 4 mA live zero distinguishes a true zero from a broken wire.
• Q: Why must the anti-aliasing filter be analog and not digital? A: Aliasing happens at the instant of sampling; once a high-frequency component has folded into the baseband, no digital filter can tell it from a genuine low-frequency signal.
• Q: A 3.8 kHz tone is sampled at 5 kHz - what do you see? A: An alias at |3800 − 5000| = 1200 Hz, which looks like a real 1.2 kHz signal.
• Q: Why is an instrumentation amplifier used instead of a plain difference amplifier? A: Its input buffers give very high input impedance so the source is not loaded, gain is set by one resistor without disturbing resistor matching, and CMRR stays high.
• Q: Why does a thermocouple need cold-junction compensation but an RTD does not? A: A thermocouple generates emf from the temperature *difference* between measuring and reference junctions, so the reference temperature must be known; an RTD's resistance is an absolute function of its own temperature.
• Q: What does a 4-wire RTD connection achieve? A: Excitation current flows in one pair and voltage is sensed on the other pair, which carries no current, so lead resistance drops out of the measurement entirely.
• Q: Define ENOB and why it differs from the nameplate resolution. A: ENOB = (SNR − 1.76)/6.02 bits; it is lower than the stated bit count because amplifier and reference noise, distortion, jitter and nonlinearity add error beyond quantisation.
• Q: Which ADC would you choose for a 24-bit temperature logger, and why? A: A sigma-delta ADC - oversampling and noise shaping give very high resolution, it has inherent anti-aliasing, and a 20 ms integration window rejects 50 Hz mains.
• Q: What is a ground loop and how do you kill it? A: Two earth points at different potentials drive current through the signal cable's return; cure it with differential or isolated inputs, single-point (star) earthing, and shields grounded at one end only.
• Q: Why is a sample-and-hold needed before a SAR ADC but not before a flash ADC? A: A SAR takes n clock cycles, during which the input must not move by more than ½ LSB; a flash ADC decides in one clock, so a changing input matters far less.
• Q: In a 16-channel multiplexed DAQ sampling at 100 kS/s total, what is the per-channel rate and what is the catch? A: 6.25 kS/s per channel, and the channels are not sampled at the same instant - there is a skew of one conversion period between them, which corrupts phase measurements.
• Q: How much error does 10 V of 50 Hz common mode cause in an amplifier with 100 dB CMRR? A: 100 dB is a ratio of 10⁵, so the input-referred error is 10/10⁵ = 100 µV.
• Q: Why is excitation current for a Pt100 kept around 1 mA? A: To limit I²R self-heating; at 1 mA the dissipation is ~0.14 mW and the error is well under 0.1 °C, while higher currents heat the sensor and bias the reading high.

Common mistakes

• Putting the gain stage after the ADC, or forgetting gain altogether, so a few-millivolt sensor uses a handful of ADC codes out of thousands.
• Believing a digital filter or a higher sampling rate applied later can remove aliasing already present in the stored data.
• Using f_s = 2 f_max exactly instead of strictly greater, and ignoring filter transition width when picking f_s.
• Quoting quantisation error as ±1 LSB instead of ±½ LSB, or forgetting the q/√12 RMS value in SNR calculations.
• Dividing by 4 for a full bridge (the /4 factor belongs to the quarter bridge only), or omitting V_ex from the bridge output.
• Grounding a cable shield at both ends, creating the very ground loop the shield was meant to prevent.
• Exceeding the loop resistance budget by adding receivers, then blaming the transmitter when readings clip at high current.
• Forgetting cold-junction compensation, lead-resistance correction for 2-wire RTDs, or sensor self-heating - each a silent systematic error, not noise.

Related topics

Operational amplifiers · Digital signal processing: filters and FFT · Sensors and transducers · Process control and PID tuning · Industrial automation: PLC, DCS and SCADA · Measurement, error analysis and calibration