Sensor Signal Conditioning Basics

Detailed Explanation

Every sensor produces a raw electrical output: a resistance that changes with temperature, a voltage that changes with pressure, a current proportional to light intensity, or a differential voltage from a bridge circuit. Almost none of these outputs are directly suitable for an MCU's ADC — they may be too small (millivolts), at the wrong common-mode level, riding on noise, or capable of exceeding the ADC's maximum input voltage. Signal conditioning transforms the sensor output into a clean, properly scaled voltage that the ADC can digitise accurately.

Sensor Output Types

Understanding the sensor's output type is the first step in designing the signal conditioning chain:

Sensors with digital output (covered in how to interface a digital temperature sensor and accelerometer and IMU) require no analog signal conditioning — the signal processing is handled in the sensor IC itself.

---

Stage 1: Gain (Amplification)

Sensors often produce signals much smaller than the ADC's full-scale input range. An op-amp gain stage scales the signal up.

Non-inverting amplifier (for positive signals):

         Rf
Vin ──┤ +  ├──── Vout = Vin × (1 + Rf/Rin)
     │ OA  │
     │  −  │
     Rin   │
     GND   ─── Feedback

Gain = 1 + Rf/Rin (always ≥ 1, non-inverting, input impedance = op-amp input impedance which is typically > 1 MΩ for FET input types).

Example: A pressure sensor outputs 0–50 mV for 0–100 kPa. ADC input range = 0–3.3 V. Required gain = 3.3 V / 0.050 V = 66 V/V.

Non-inverting amplifier: Gain = 1 + Rf/Rin = 66. Choose Rin = 1 kΩ, Rf = 65 kΩ (use 66.5 kΩ from E96 series for closest match). Actual gain = 1 + 66.5/1 = 67.5 (2.3% high — acceptable, calibrate out in firmware).

For the relationship between gain and bandwidth, see what is an op-amp? — the gain-bandwidth product limits how much gain is achievable at a given signal frequency.

For differential sensors, use an instrumentation amplifier in place of a simple non-inverting amplifier.

---

Stage 2: Offset Shifting

Many sensors do not have a zero output at zero stimulus, or the amplified signal's minimum value may not be 0 V. An offset circuit shifts the output range.

Summing amplifier for offset shift:

Vsensor ──── Rin1 ──┐
                    │ Rf
Voffset ──── Rin2 ──┤ ──── OA −  ──── Vout = −(Vsensor × Rf/Rin1 + Voffset × Rf/Rin2)
                    │
                    GND

The offset voltage Voffset can come from a voltage divider on the supply, a precision voltage reference IC, or a DAC output.

Practical alternative: Use Texas Instruments' free "Single-Supply Op Amp Design Tool" (SLOA097) to automatically calculate resistor values for any gain and offset combination with a single op-amp stage.

Example (4-20 mA current loop to 0–3.3 V):

A 4-20 mA loop with a 250 Ω shunt produces 1.0–5.0 V. To map this to 0–3.3 V:

• Required gain = 3.3/(5.0−1.0) = 0.825 (attenuation)
• Required offset shift = −0.825 × 1.0 V = −0.825 V at the output

A single op-amp stage with: inverting gain = −0.825, then invert (or use a two-stage design) will produce the correct result. Alternatively, use a resistor voltage divider (R1 = 15 kΩ, R2 = 100 kΩ) to bring 5.0 V down to ~3.3 V and the 1.0 V minimum down to ~0.65 V, then use a comparator-style clamping for the 0–3.3 V full-scale mapping.

---

Stage 3: Anti-Aliasing Filter

Before the conditioned signal reaches the ADC, a low-pass filter removes frequency content above the Nyquist frequency (half the sample rate) to prevent aliasing. This is not optional — without it, high-frequency noise folds back into the measurement band and corrupts every sample.

First-order passive RC filter:

Vin ──── R ────┬──── Vout (to ADC)
               │
               C
               │
              GND

Cutoff frequency: fc = 1 / (2π × R × C)

This is the simplest anti-aliasing filter and is adequate for slow sampling rates (< 1 kHz). The RC network has 1 pole → −20 dB/decade rolloff beyond fc. For a 100 Hz sample rate with fc = 20 Hz: the filter attenuates the Nyquist frequency (50 Hz) by 20 × log10(50/20) = 8 dB — not enough for a 12-bit ADC. To achieve adequate attenuation (> 72 dB at Nyquist), use a higher-order filter.

Second-order active Sallen-Key low-pass filter:

A two-pole (−40 dB/decade) Sallen-Key filter provides much better attenuation. For detailed filter design including component selection and Butterworth vs Bessel approximations, see how to design active filters.

Practical rule for RC filter sizing:

For a 12-bit ADC sampled at fs:

• Set first-order RC filter cutoff at fc = fs/5 (adequate for simple monitoring)
• Use a second-order Sallen-Key at fc = fs/3 for better transition band if noise is a concern

For a 1 kHz sample rate, fc = 200–333 Hz. R = 10 kΩ, C = 47 nF → fc = 338 Hz. Adjust component values to target.

ADC source impedance constraint: The RC filter's output impedance at the ADC input is R (at low frequency). For SAR ADCs (typical in MCUs), the input must settle to < 0.5 LSB within the acquisition time. If R is too large (> 5–10 kΩ for a fast SAR), the RC charging time is too slow. Use a buffer op-amp (unity-gain follower) after the filter to drive the ADC input from a low-impedance source. See what is an ADC? for SAR ADC acquisition time details.

---

Stage 4: Input Protection

The ADC's input pins can be damaged if the voltage at the pin exceeds the supply voltage (Vdd) or goes below GND. This can happen from:

• Sensor powering on before the MCU
• Cable disconnection with the sensor held high
• Inductive transients from long sensor cables

Clamping diodes: Most MCUs have internal protection diodes from every I/O and ADC pin to Vdd and GND. These clamp inputs to approximately Vdd + 0.3 V and GND − 0.3 V. However, they can only absorb a few milliamps before damage — they are not designed to handle sustained out-of-range voltages.

Series resistor for protection: Adding a 1–10 kΩ series resistor at the ADC input limits the current through the internal clamping diodes: I = (Voverrange − Vdd) / R_series. With R = 10 kΩ and a 5 V sensor connected to a 3.3 V ADC pin: I = (5 − 3.3) / 10,000 = 170 µA — safe for most MCU internal diodes (typically rated 10 mA continuous).

External TVS or clamping diodes: For interfaces exposed to ESD or large transients (external connectors, sensors on long cables), add dedicated TVS diodes or BAT54 Schottky diode clamps to limit peak voltage at the ADC input. Place these as close as possible to the PCB connector where the cable attaches.

---

Stage 5: Calibration

Even with a well-designed signal conditioning circuit, manufacturing tolerances in resistors, op-amps, and sensor elements introduce gain and offset errors. Two-point calibration corrects these:

1. Apply a known low stimulus (e.g. 0°C for a temperature sensor) and record the ADC output: ADC_low
2. Apply a known high stimulus (e.g. 100°C) and record: ADC_high
3. Compute calibration coefficients:
   • Gain correction: k = (Stim_high − Stim_low) / (ADC_high − ADC_low)
   • Offset correction: b = Stim_low − k × ADC_low
4. At runtime: Measurement = k × ADC_reading + b

Store k and b in MCU flash or EEPROM. For the highest accuracy, perform this calibration at the operating temperature (not just room temperature), since op-amp offset and resistor values drift with temperature.

For a complete sensor product development engagement including signal conditioning circuit design, PCB layout, and calibration firmware, Zeus Design's engineering team supports all stages from sensor selection through to production.

Design Considerations

• Choose rail-to-rail op-amps for single-supply designs: When the signal conditioning circuit runs from a single 3.3 V supply and the sensor signal may be close to 0 V or 3.3 V, the op-amp must swing its output all the way to the supply rails. Standard op-amps (LM741, TL071) cannot swing within 1–2 V of the supply rail on a single supply. Use a rail-to-rail output op-amp (OPA314, MCP6001, LMV321) to maintain linearity across the full ADC input range.
• Buffer the voltage reference: If the signal conditioning circuit uses the MCU's AVCC or an on-chip voltage reference as an excitation source for the sensor (e.g. thermistor voltage divider), decoupling AVCC adequately is critical. Switching regulator noise on AVCC appears directly as ratiometric error in the measurement. Add a ferrite bead + 100 nF + 10 µF between the main supply and the sensor excitation line, or use a dedicated voltage reference IC.
• Limit the gain bandwidth product requirement: A non-inverting op-amp with gain 100 at signal frequencies up to 1 kHz needs a GBW of at least 100 kHz. Most general-purpose op-amps (OPA2134, TL072, MCP6002) have GBW of 1–10 MHz, far exceeding this. The GBW constraint becomes relevant only at high gains (> 100 V/V) combined with high signal frequencies (> 10 kHz) — rare in sensor measurement applications.

Common Mistakes

• Placing the anti-aliasing filter after the ADC: Some designs omit the anti-aliasing filter and apply a digital filter (FIR or IIR) in firmware. A digital filter can remove noise that is already within the measurement band, but it cannot undo aliasing — once a 60 Hz signal has been aliased down to 5 Hz by a 65 Hz sample rate, no digital filter can distinguish it from a real 5 Hz signal. The anti-aliasing filter must always be analog, before the ADC.
• Ignoring the op-amp's output current limitation: Op-amps are not power devices — a general-purpose op-amp sources only 10–40 mA. If the signal conditioning output drives a low-impedance load (< 1 kΩ), the op-amp may enter current limiting and distort the output. Check the datasheet for short-circuit current and design the load impedance (including the ADC input network) to stay well within limits.
• Forgetting that the RC filter forms a loaded voltage divider: The series resistor R in the ADC input RC filter forms a voltage divider with the ADC's internal sampling capacitor. If R = 10 kΩ and the ADC's input capacitance is 20 pF, the time constant for the ADC to settle to < 0.5 LSB at 12 bits is τ = R × C × ln(2^13) ≈ 10,000 × 20×10⁻¹² × 9.01 ≈ 1.8 µs. If the ADC acquisition time is only 1 µs, the reading will be settled less than the required 0.5 LSB — increase acquisition time in the MCU ADC configuration or reduce R.