Electronics Guide

Wheatstone Bridge Fundamentals

A Wheatstone bridge consists of four resistive elements arranged in a diamond pattern with excitation voltage applied across one diagonal and the output signal measured across the other. When all four resistors are equal, the bridge is balanced and produces zero output voltage. Changing one or more resistor values unbalances the bridge, producing an output voltage proportional to the resistance change.

With the excitation voltage V_EX across one diagonal, R₁ and R₂ forming one branch and R₃ and R₄ the other, the output is the difference between the two midpoint voltages:

V_out = V_EX [R₁/(R₁ + R₂) − R₃/(R₃ + R₄)]

For a single active element of nominal resistance R that changes by ΔR, and for changes small enough that ΔR stays a tiny fraction of R, this reduces to the familiar small-signal result:

V_out ≈ V_EX (ΔR/4R)

The scale of that result governs everything downstream. A strain gauge worked to 1000 microstrain changes by about 0.2 percent, so a quarter bridge excited at 10 volts delivers roughly 5 millivolts on a 5 volt common-mode pedestal.

The bridge output is proportional to the excitation rather than fixed by it, and that proportionality is what makes ratiometric measurement possible. If the analog-to-digital converter derives its reference from the same source that excites the bridge, a change in excitation scales the signal and the reference together and divides out of the result, so excitation tolerance, drift, and low-frequency noise cancel to first order. The cancellation reaches no further. It does nothing about the bridge nonlinearity, the resistance of the excitation leads, the sensor's own temperature coefficient, or amplifier offset, and it holds only while the converter reference is the voltage actually present at the bridge terminals, which is the argument for sensing the excitation at the bridge rather than at the supply.

Bridge Configurations

Bridges can employ one, two, or four active sensing elements, each configuration offering different sensitivity and compensation characteristics:

• Quarter bridge: One active element and three fixed resistors. The simplest arrangement, but it carries the full bridge nonlinearity, offers no inherent temperature compensation, and gives half the output of a half bridge.
• Half bridge: Two active elements. Placed in adjacent arms, meaning the two resistors of one branch, and arranged so their resistances move in opposite directions, they double the output to V_EX(ΔR/2R) and cancel the nonlinearity exactly; any resistance change common to both, including the thermal drift they share, cancels as well. Two elements that move in the same direction, placed in opposite arms, also double the output, but there neither the nonlinearity nor the common temperature drift cancels. An unstrained dummy gauge in the arm adjacent to the active gauge adds no sensitivity at all and exists solely to cancel the temperature coefficient.
• Full bridge: Four active elements, two rising and two falling, which gives V_out = V_EX(ΔR/R) exactly: four times the quarter-bridge output, with both the nonlinearity and the common temperature response cancelled. This is the standard construction of load cells and pressure transducers.

Excitation Methods

Bridge excitation can use a voltage source or a current source, and the choice decides which errors the interface inherits:

Voltage excitation is simpler to implement, and because the output depends only on ratios of resistances, the sensitivity expressed in millivolts per volt is independent of the absolute bridge resistance and of its temperature coefficient. The weakness lies in the excitation cable: its resistance forms a divider with the bridge, so part of the source voltage never reaches the bridge terminals, and the shortfall itself drifts as the cable warms.

Current excitation holds the total current through the bridge constant, so voltage dropped in the excitation leads no longer subtracts from the excitation and that gain error disappears. In exchange, the voltage across the bridge now follows the absolute bridge resistance, so unit-to-unit resistance tolerance and the bridge temperature coefficient act directly on the reading.

Remote sensing recovers the accuracy of voltage excitation in the presence of lead resistance. Two extra sense wires, which carry no appreciable current and therefore drop no voltage, report the excitation actually present at the bridge terminals to the amplifier or to the converter reference input. This six-wire connection is the usual partner of ratiometric measurement.

Bridge Linearization

The bridge equation contains an inherent nonlinearity: the output is not strictly proportional to resistance change, because the active element sits inside the divider whose ratio it is changing. For a voltage-excited quarter bridge the exact relation is:

V_out/V_EX = (ΔR/R) / (4 + 2ΔR/R)

Set against the small-signal form, this falls short of the straight line by a fraction of about ΔR/2R. An element worked to 1 percent resistance change therefore departs from linearity by roughly 0.5 percent of reading, while the 0.2 percent change of a foil gauge at 1000 microstrain departs by only about 0.1 percent. Several techniques address the error, and they are not interchangeable:

• Keep the change small: The ΔR/2R rule makes the nonlinearity negligible for strain gauges and for narrow-range pressure sensors, which is why many designs simply live with it.
• Use complementary elements: A half bridge whose two elements move in opposite directions, or a full bridge with two of each, removes the nonlinearity exactly rather than merely reducing it, because the offending terms cancel between the arms. This is the real fix, and it is why precision load cells are full bridges.
• Constant-current excitation: Driving the whole bridge from a current source changes the exact relation to ΔR/(4R + ΔR), which halves the nonlinearity but does not abolish it. The output is exactly linear only where the current passes through the sensing element alone, as in the four-wire measurement of a single RTD.
• Feedback linearization: Enclosing the active arm in an op-amp loop that holds the voltage across it constant and measuring the current the loop supplies yields an output proportional to conductance; driving the element with a known current and measuring its voltage yields one proportional to resistance. Either way, no divider term remains to curve the response.
• Digital correction: Characterizing the response and applying a polynomial or lookup correction in firmware reaches whatever accuracy the repeatability of the sensor allows, at the cost of a calibration step for each unit.

Lead-resistance compensation solves a different problem and should not be confused with any of these. Three-wire and four-wire connections, remote sense leads, and current excitation all attack the error caused by resistance in the wiring; none of them alters the shape of the bridge transfer curve.

Bridge Amplifier Design

Instrumentation amplifiers are the standard interface for bridge outputs, offering high input impedance, excellent common-mode rejection, and precise differential gain. Key design considerations include:

• Common-mode voltage: The bridge output rides on a common-mode level of approximately half the excitation. The amplifier must accept that level within its input range while amplifying a differential signal thousands of times smaller. Because the common-mode level is static, the error it produces through finite common-mode rejection is a fixed offset that calibration removes; what survives calibration is the drift of that rejection with temperature and its decline with frequency, which is what admits mains hum and similar interference.
• Gain selection: Choose gain to utilize the full input range of the following ADC while maintaining adequate noise margin.
• Offset adjustment: Bridge imbalance produces offset voltage that may require nulling before or after amplification.
• Filtering: Low-pass filtering after amplification rejects high-frequency noise while preserving measurement bandwidth, and it must roll off below half the sampling rate to keep out-of-band noise from aliasing into the reading.
• Input protection and matching: Series resistors and clamp diodes guard the inputs against fault voltages, and a differential capacitor with matched capacitors to ground suppresses radio-frequency energy. The two input paths must match, because any imbalance between them converts common-mode voltage into differential error.

Thermocouple Principles

The Seebeck effect is a property of a conductor carrying a temperature gradient, not of the junction itself. The junction merely forces the two wires to a common potential at the hot end; the voltage that appears at the cold end is the difference between the contributions the two dissimilar wires accumulate along their length. Two consequences shape every thermocouple interface. The sensor reports a temperature difference and never an absolute temperature, so the temperature of its cold end must be established by other means. And the voltage is small, tens of microvolts per degree Celsius, giving a full-scale signal of a few tens of millivolts, which demands high gain together with microvolt-class offset and drift.

The standard types differ in range and sensitivity. The sensitivities below are representative values near the middle of each range, since the Seebeck coefficient of every type varies with temperature and the reference relation is a polynomial rather than a straight line:

• Type K (Chromel-Alumel): The general-purpose workhorse, −200 to 1260 degrees Celsius, roughly 41 microvolts per degree Celsius.
• Type J (Iron-Constantan): Common at moderate temperatures, −40 to 750 degrees Celsius, roughly 52 microvolts per degree Celsius. The iron leg rusts in humid service, which limits its life.
• Type T (Copper-Constantan): Excellent from cryogenic to moderate temperatures, −200 to 350 degrees Celsius, roughly 43 microvolts per degree Celsius. Its copper leg makes an unusually well-behaved cold junction.
• Type E (Chromel-Constantan): The highest sensitivity of the standard types, −40 to 900 degrees Celsius, roughly 68 microvolts per degree Celsius.
• Types R, S, and B (platinum and platinum-rhodium): For furnace and process temperatures beyond the reach of base-metal types; the reference tables extend to about 1768 degrees Celsius for types R and S and about 1820 degrees Celsius for type B, with continuous service ratings lower. Sensitivity is only around 10 microvolts per degree Celsius, so these types trade signal for range and stability, and type B produces almost nothing near room temperature.

The Cold Junction Problem

The measured voltage depends on both junction temperatures, so a reading means nothing until the cold junction temperature is known. With that junction held at 0 degrees Celsius, the traditional ice bath, the standard reference tables convert voltage to temperature directly. In a real instrument the cold junction is the terminal block where the thermocouple wires meet copper, sitting at whatever the ambient happens to be, and both wires must meet copper at the same temperature for a single correction to apply.

The remedy is to measure the cold junction with a separate sensor and either add an equivalent voltage in hardware or apply the correction in software:

T_sensing = f(V_measured + V_compensation)

Here V_compensation is the voltage the same thermocouple type would produce with its measuring junction at the cold junction's actual temperature and its reference at 0 degrees Celsius, and f is that type's voltage-to-temperature polynomial. Because the correction enters the sum directly, an error in the cold-junction measurement passes almost one for one into the result: one degree Celsius of error in the ambient sensor costs very nearly one degree Celsius in the reported temperature. The accuracy of a thermocouple channel is therefore usually set by that sensor and by the thermal gradient across the terminal block, not by the thermocouple.

Hardware Compensation Techniques

Hardware cold junction compensation adds a temperature-dependent voltage to the thermocouple signal before amplification:

• Thermistor-based compensation: A thermistor in a bridge or voltage divider network generates a voltage that varies with cold junction temperature. Component values are chosen to match the thermocouple's Seebeck coefficient over the expected ambient temperature range.
• Semiconductor temperature sensor: Linear sensors placed at the terminal block provide a signal proportional to its temperature, which is scaled and added to the thermocouple output. The LM35 gives a voltage scaled to degrees Celsius, while the AD590 gives a current proportional to absolute temperature, so the two demand different scaling and different handling of the zero.
• Integrated compensation ICs: Specialized chips combine temperature sensing with precision voltage generation, optimized for specific thermocouple types. Examples include the AD594/AD595 series for types J and K.

Software Compensation

Digital cold junction compensation measures both the thermocouple voltage and the cold junction temperature independently, then combines them mathematically:

1. Measure cold junction temperature with a precision sensor (RTD, thermistor, or IC sensor)
2. Convert cold junction temperature to equivalent thermocouple voltage using polynomial or lookup table
3. Add this voltage to the measured thermocouple voltage
4. Convert the total voltage to sensing junction temperature

Software compensation offers flexibility (any thermocouple type with appropriate tables), accuracy (can use high-order polynomials for linearization), and the ability to handle nonstandard reference temperatures.

Thermocouple Amplifier Design

Thermocouple amplifiers face demanding requirements:

• Low offset voltage: With signals as low as tens of microvolts, input offset voltage directly affects accuracy. Chopper-stabilized or auto-zero amplifiers achieve sub-microvolt offsets.
• Low drift: Offset voltage drift with temperature must be small compared to the thermocouple sensitivity, typically below 1 microvolt per degree Celsius.
• High common-mode rejection: Ground loops and electromagnetic interference can create large common-mode voltages that must not appear as signal errors.
• Input protection: Thermocouples may connect to grounded objects at different potentials, requiring protection against large common-mode voltages.
• Open thermocouple detection: When a thermocouple fails open, the amplifier input floats, potentially giving false readings. A small bias current or detection circuit should identify this condition.

RTD Characteristics

RTDs use the predictable resistance-temperature relationship of pure metals, most commonly platinum. The platinum RTD, called a PRT or named for its nominal resistance as a Pt100 or Pt1000, is the industrial standard for precision temperature measurement. IEC 60751 defines the grade in general use, whose mean temperature coefficient between 0 and 100 degrees Celsius is 0.00385 per degree Celsius:

• Pt100: 100 Ω at 0 degrees Celsius, rising to 138.51 Ω at 100 degrees Celsius and 390.48 Ω at 850 degrees Celsius. The slope is 0.391 Ω per degree Celsius at 0 degrees Celsius and averages 0.385 Ω per degree Celsius over the first hundred degrees, which is the figure usually quoted.
• Pt1000: Ten times the resistance and ten times the slope, which lifts the signal an order of magnitude above lead resistance and amplifier offset. It is the usual choice for two-wire and three-wire wiring and for battery-powered instruments, where it also cuts the excitation current needed for a given signal.
• Temperature range: −200 to 850 degrees Celsius for wire-wound industrial elements; thin-film elements are commonly specified over a narrower span.
• Tolerance classes: IEC 60751 states tolerance as a constant plus a term proportional to temperature: Class AA is ±(0.1 + 0.0017|T|) degrees Celsius, Class A ±(0.15 + 0.002|T|), Class B ±(0.3 + 0.005|T|), and Class C ±(0.6 + 0.01|T|). A Class A element is therefore good to 0.15 degrees Celsius at 0 degrees Celsius but only 0.35 degrees Celsius at 100 degrees Celsius, and each class is defined over its own restricted range. The American equivalent is ASTM E1137, whose Grade A and Grade B carry different formulas.

The Callendar-Van Dusen equation describes RTD resistance against temperature, and it takes two forms. At and above 0 degrees Celsius it is a simple quadratic:

R(T) = R₀(1 + AT + BT²)

Below 0 degrees Celsius a cubic term is added:

R(T) = R₀[1 + AT + BT² + C(T − 100)T³]

For the IEC 60751 platinum grade the coefficients are A = 3.9083 × 10⁻³ per degree Celsius, B = −5.775 × 10⁻⁷ per degree Celsius squared, and C = −4.183 × 10⁻¹² per degree Celsius to the fourth power. The condition on C is not a footnote: C is defined as zero at and above 0 degrees Celsius, and carrying the cubic term into the positive range is a common and expensive mistake. At 500 degrees Celsius it subtracts about 21 Ω from a Pt100 that should read 280.98 Ω, roughly 60 degrees Celsius of error across the range where most industrial measurements are made.

Recovering temperature from resistance is straightforward above 0 degrees Celsius, where the quadratic has a closed-form solution, and awkward below it, where the cubic is normally handled by iteration or by a lookup table with interpolation. Both are cheap beside the measurement itself, so there is no excuse for a linear approximation outside a narrow span.

RTD Excitation and Self-Heating

An RTD is measured by passing a current through it, and that current heats it. The dissipated power raises the element above the temperature it is supposed to report, an error no calibration removes, because it depends on the thermal path to the medium. Manufacturers characterize it with a dissipation constant in milliwatts per degree Celsius, and the error is simply the dissipated power divided by that constant. Published values run from roughly 2.5 milliwatts per degree Celsius for small thin-film elements to roughly 65 milliwatts per degree Celsius for larger wire-wound ones, and any given element performs far better in flowing water than in still air.

The arithmetic is unforgiving because power rises as the square of the current. A Pt100 carrying 1 mA at 0 degrees Celsius dissipates 0.1 mW, which a thin-film element at 2.5 milliwatts per degree Celsius turns into about 0.04 degrees Celsius of error; the same sensor at 850 degrees Celsius has risen to 390 Ω and dissipates 0.39 mW. Doubling the current to 2 mA quadruples both figures. Excitation is therefore usually held at or below 1 mA, and where a larger signal is needed the current is pulsed, the converter sampling during the pulse so that average power stays low. Thermistors deserve the same calculation and are more easily caught by it, since their resistance is high and a plain divider across a supply rail can dissipate more than their dissipation constant allows.

Lead Resistance Compensation

The low resistance and modest sensitivity of an RTD make lead resistance a first-order error rather than a detail. Three wiring configurations address it:

• Two-wire connection: The simplest, and it adds the whole lead resistance to the measurement. With a Pt100 at 0.391 Ω per degree Celsius, 1 Ω of loop resistance is about 2.6 degrees Celsius of error, and copper itself drifts by roughly 0.4 percent per degree Celsius, so the error moves with the weather. Acceptable only for short leads, for a Pt1000, or where the lead resistance is measured and stable.
• Three-wire connection: A third wire lets the interface measure one lead and subtract it, on the assumption that all three leads are identical. What remains is the mismatch between leads rather than their resistance, typically an order of magnitude smaller. This is the common industrial configuration.
• Four-wire (Kelvin) connection: One pair carries the excitation current while the other senses the voltage across the element. The sense pair carries no appreciable current and therefore drops no voltage, so lead resistance leaves the result entirely, mismatched or not. This is the reference method for calibration-grade work.

Thermistor Characteristics

Thermistors are ceramic semiconductor devices exhibiting large resistance changes with temperature. NTC (negative temperature coefficient) types decrease resistance with increasing temperature and are most common for measurement applications:

• High sensitivity: About 4 percent change in resistance per degree Celsius near 25 degrees Celsius for a B constant of 3400 kelvins, roughly ten times the 0.39 percent per degree Celsius of a platinum RTD. The coefficient is not fixed: it varies as the reciprocal of the square of absolute temperature, so it falls steadily as the sensor warms.
• Nonlinear response: Resistance falls close to exponentially with temperature, so the raw output needs linearization or a table before it can be read as a temperature.
• Fast response: Small thermal mass allows rapid tracking, which is why thermistors dominate in battery packs, inrush limiting, and anywhere a thermal runaway must be caught quickly.
• Limited range: Most useful over a span of 50 to 100 degrees within their operating limits; outside it the resistance either swamps the interface or falls into the noise.
• Interchangeability: Resistance tolerance and spread in the B constant mean that precision work requires calibration of each part, unless an interchangeable grade is bought at a premium.

Thermistor Linearization

The highly nonlinear thermistor response requires linearization for practical use. Several techniques apply:

• Parallel resistance: Adding a fixed resistor in parallel with the thermistor straightens the response curve over a limited range. The optimal parallel resistance equals the thermistor resistance at the center of the desired range.
• Voltage divider optimization: Choosing the series resistor in a voltage divider to equal the thermistor resistance at mid-range provides approximately linear voltage output over that range.
• Steinhart-Hart equation: The relation 1/T = A + B ln R + C(ln R)³, with T in kelvins, models a thermistor to within a few thousandths of a degree over a wide span once its three coefficients are fitted to three measured points. The simpler B-parameter form, which uses a single constant and a reference point, is adequate over narrow ranges.
• Lookup tables: For digital systems, storing measured resistance-temperature pairs and interpolating provides arbitrary accuracy without complex calculations.

RTD and Thermistor Interface Circuits

Common interface approaches include:

• Bridge circuits: Place the sensor in a Wheatstone bridge for differential output and reduced sensitivity to excitation variations. Works well with a four-wire RTD connection using remote sensing.
• Current source excitation: Drive the sensor with a precision current source and measure the resulting voltage. This is the natural partner of the four-wire connection, and it gives a voltage directly proportional to resistance across the whole temperature range.
• Ratiometric measurement: Use the same reference for both sensor excitation and ADC reference, canceling effects of reference drift.
• Sigma-delta ADCs: Modern sigma-delta converters with integrated current sources and multiplexed inputs provide complete RTD interface solutions with minimal external components.

Strain Gauge Fundamentals

A strain gauge consists of a resistive element (typically metal foil or semiconductor) that changes resistance when mechanically stretched or compressed. The gauge factor relates resistance change to applied strain:

GF = (ΔR/R) / ε

Here ε is the mechanical strain, the change in length divided by the original length. It is dimensionless and conventionally quoted in microstrain, parts per million. The same symbol denotes permittivity in the capacitive section below, where it therefore appears only with subscripts. Constantan foil, the material of most general-purpose gauges, has a gauge factor close to 2, and platinum-tungsten alloys reach about 4. Semiconductor gauges reach 50 to 200 but drift far more with temperature and are themselves markedly nonlinear.

The numbers that follow are small. At a gauge factor of 2, a strain of 1000 microstrain, a tenth of a percent of elongation and comparable with the elastic limit of mild steel, changes the gauge by 0.2 percent. In a quarter bridge excited at 10 volts that is 5 millivolts; a full bridge of four gauges yields 20 millivolts. Every requirement placed on the amplifier follows from the size of that signal.

Bridge Configuration for Strain Measurement

Strain gauges almost universally operate in bridge configurations to maximize sensitivity and provide temperature compensation:

• Quarter bridge: Single active gauge with three completion resistors. Requires external temperature compensation and offers lowest sensitivity.
• Half bridge (bending): Two gauges on opposite faces of a beam, wired into adjacent arms, so that tension on one face and compression on the other add. The output doubles, the nonlinearity cancels, temperature effects common to both gauges cancel, and axial strain, which stretches both faces alike, is rejected.
• Half bridge (Poisson): One gauge aligned with the strain and one transverse to it in the adjacent arm. The transverse gauge sees the Poisson contraction, about −0.3 times the axial strain in steel, so the output is roughly 1.3 times that of a quarter bridge while temperature effects still cancel.
• Full bridge: Four active gauges, two in tension and two in compression. Maximum sensitivity, exact cancellation of the nonlinearity and of common temperature effects, and rejection of the strain components the transducer is not meant to measure. This is the standard configuration for load cells and force transducers.

Whatever the configuration, a shunt calibration resistor switched across one arm checks the whole chain end to end. A 100 kilohm resistor across one 350 Ω arm lowers it by about 0.35 percent, which at a gauge factor of 2 imitates roughly 1750 microstrain of accurately known size. Comparing the reading against the calculated value verifies excitation, gain, and wiring without touching the specimen.

Excitation Considerations

Strain gauge excitation involves trade-offs between signal level, self-heating, and stability:

• Voltage versus current excitation: Under voltage excitation the output in millivolts per volt depends only on ratios of resistance, so a gauge whose resistance drifts with temperature does not change the sensitivity; what the excitation cable's resistance steals from the bridge terminals does. Current excitation removes that cable error but makes the output proportional to the absolute bridge resistance, so the gauge's own temperature coefficient becomes a gain error. Voltage excitation with remote sense leads captures both advantages at the cost of two more wires.
• AC versus DC excitation: DC excitation is standard for most applications. AC excitation at a carrier of 1 to 10 kHz moves the measurement away from the amplifier's own offset drift and from low-frequency noise, and it permits transformer coupling for isolation, at the cost of a demodulator and a carrier-rate bandwidth limit.
• Excitation level: Higher excitation raises the signal in proportion but heats the gauge as the square, and that heat reaches the specimen through the adhesive and the backing. A 350 Ω full bridge excited at 5 volts dissipates about 18 mW in each gauge; the same bridge at 10 volts dissipates about 71 mW, which a plastic or composite specimen cannot carry away. Practical excitation runs from 1 to 10 volts, chosen with the specimen material and the gauge area in mind.

Amplifier Requirements

Strain gauge amplifiers must meet demanding specifications:

• High gain: Full-scale bridge outputs run from about 2 to 20 mV, so gains of 100 to 1000 are needed to fill the input range of a converter.
• Low noise: Noise referred to the input must be small beside the resolution sought. Sixteen bits across a 10 mV full scale puts one code at 153 nV, and an input-referred noise of 150 nV RMS swings about 1 microvolt peak to peak, some six and a half codes, leaving roughly 13 noise-free bits. Resolving the last code demands either a far quieter front end or averaging, which buys resolution as the square root of the number of samples.
• Common-mode rejection: The bridge output rides on half the excitation. Holding the resulting error below 0.01 percent of a 10 mV full scale, that is 1 microvolt referred to the input, would require about 128 dB of rejection against a 2.5 V common-mode level. Only the best instrumentation amplifiers reach that at DC and high gain, and none hold it at mains frequency, so practical designs treat the static part as an offset and calibrate it out, spending the amplifier's rejection on what calibration cannot reach: the drift of that offset with temperature and the interference that arrives as a changing common-mode voltage.
• Low drift: Offset and gain drift with temperature should not exceed the desired measurement stability.
• Wide bandwidth: Static strain measurements require only DC response, but dynamic strain measurement may need bandwidths from Hz to kHz depending on the application.

Practical Amplifier Topologies

Several amplifier configurations suit strain gauge applications:

• Discrete instrumentation amplifier: Three op-amps in the classic in-amp topology provide flexibility in gain selection and the ability to optimize each stage independently.
• Integrated instrumentation amplifier: Devices like the INA114, AD620, or AD8421 offer excellent performance with minimal external components. Internal laser-trimmed resistors ensure high CMRR.
• Programmable gain amplifiers: PGAs with multiple gain settings enable auto-ranging to maximize resolution across varying signal levels.
• Integrated signal conditioners: Complete strain gauge conditioner ICs include excitation, amplification, filtering, and sometimes digitization in a single device.

Noise Reduction Techniques

Achieving the signal-to-noise ratios required for precision strain measurement requires attention to multiple noise sources:

• Shielding: Shielded cables and shielded bridge enclosures reduce electromagnetic interference pickup.
• Twisted pairs: Twisting the excitation and signal wires minimizes magnetically induced noise.
• Filtering: Low-pass filters at the amplifier input and output reject high-frequency noise. The bandwidth should match the signal requirements.
• Synchronous detection: With AC excitation, synchronous demodulation rejects noise at frequencies other than the carrier.
• Averaging: Digital oversampling and averaging improve effective resolution beyond the ADC's native capability.

Capacitive Sensing Principles

Capacitance depends on geometry and material properties:

C = ε_rε₀A/d

Here A is the plate area, d the separation, ε₀ the permittivity of free space, 8.854 picofarads per meter, and ε_r the relative permittivity of the material between the plates. Permittivity is written as this product throughout rather than as a bare ε, to keep it distinct from the strain ε of the strain-gauge section above. Capacitive sensors exploit a change in any of the three quantities:

• Variable gap: Position and pressure sensors vary the plate separation, changing capacitance inversely with distance.
• Variable area: Linear and rotary position sensors vary the overlapping plate area, producing linear capacitance change with position.
• Variable dielectric: Level sensors, humidity sensors, and material analyzers detect changes in the dielectric material between plates.

The variable-gap form is inherently nonlinear, since capacitance varies as the reciprocal of the separation. Differential construction removes that. With a moving plate between two fixed ones, the ratio (C₁ − C₂)/(C₁ + C₂) reduces to the displacement divided by the nominal gap, exactly linear, and independent of area and of permittivity and therefore of the humidity and temperature that move them. Precision capacitive accelerometers and pressure sensors are built this way.

Excitation Methods

Capacitive sensors require AC excitation, typically at frequencies from 1 kHz to 1 MHz:

• Sine wave excitation: Provides clean, well-defined excitation but requires linear amplifiers and precision rectification for demodulation.
• Square wave excitation: Easier to generate and enables simple switching circuits for charge-based measurement.
• Pulse excitation: Charging a capacitor through a known resistance and measuring charge time provides a simple frequency or time-domain output.

Higher excitation frequencies improve signal levels (capacitive reactance decreases with frequency) and reduce noise susceptibility, but increase power consumption and may cause electromagnetic compatibility issues.

Impedance Measurement Approaches

Several circuit topologies measure capacitance or capacitance changes:

• AC bridge: Similar to resistive bridges but using capacitors as bridge elements. The output voltage varies with capacitance imbalance and excitation frequency. Requires phase-sensitive detection to separate reactive and resistive components.
• Charge amplifier: Measures the charge transferred to a capacitor when switched between voltage levels. Output voltage is proportional to the ratio of sensor and feedback capacitance.
• Oscillator frequency: Using the sensor capacitance as part of an LC or RC oscillator produces frequency output proportional to capacitance. Simple and noise-resistant but requires frequency measurement.
• Sigma-delta modulator: The capacitor-switching action in sigma-delta converters naturally measures capacitance ratio, making them well-suited for capacitive sensor interfaces.

Synchronous Demodulation

Synchronous (or phase-sensitive) demodulation extracts the capacitance-related signal while rejecting noise at other frequencies:

1. The sensor is excited with a sinusoidal or square wave at a known frequency
2. The resulting signal is amplified to useful levels
3. A synchronous demodulator multiplies the signal by a reference at the excitation frequency
4. Low-pass filtering extracts the DC component, which is proportional to capacitance

This technique provides excellent noise rejection because only signals at exactly the excitation frequency and in phase with it produce DC output. Noise at other frequencies averages to zero.

Parasitic Capacitance Handling

Capacitive sensors typically produce very small signals (femtofarads to picofarads of change), making them susceptible to parasitic capacitance from cables, PCB traces, and sensor construction. Several techniques minimize these effects:

• Driven shields: Surrounding the sensitive node with a conductor driven to the same potential eliminates the capacitance between them. The shield "follows" the sensor signal, preventing current flow.
• Differential measurement: Using two sensors in a differential configuration cancels common-mode parasitic effects.
• Ratiometric techniques: Measuring the ratio of two capacitances rather than absolute values cancels many error sources.
• Short connections: Mounting the interface electronics close to the sensor minimizes cable capacitance.

Integrated Capacitive Interface ICs

Several integrated circuits provide complete capacitive sensor interfaces:

• Capacitance-to-digital converters: Parts such as the AD7745 and AD7746 from Analog Devices and the FDC1004 from Texas Instruments digitize capacitance directly, resolving changes well below a femtofarad, and carry a programmable offset that subtracts the sensor's standing capacitance so the converter range covers only the change of interest.
• Capacitive touch controllers: Specialized ICs for proximity sensing and touch interfaces handle multiple channels with automatic calibration.
• Programmable analog front-ends: General-purpose analog front-ends can be configured for capacitive measurement with appropriate firmware.

LVDT Operating Principles

An LVDT consists of a primary winding and two secondary windings wound on a hollow cylindrical form. A ferromagnetic core slides within the bore, coupling the primary to the secondaries. The position of the core determines the differential voltage between the two secondaries:

• At center position, equal coupling to both secondaries produces zero differential output
• Moving the core toward one secondary increases its coupling while decreasing the other's
• The magnitude of the differential output indicates displacement from center
• The phase of the output (0 or 180 degrees relative to excitation) indicates direction

LVDT Excitation

LVDTs require sinusoidal excitation, typically between 50 Hz and 20 kHz, with 1 to 10 kHz the most common choice; aircraft systems traditionally drive LVDTs, synchros, and resolvers at the 400 Hz of their AC power. Three considerations set the choice:

• Frequency selection: Higher frequencies provide faster response but increase core losses. The frequency should be at least 10 times higher than the maximum rate of position change.
• Voltage level: Typical excitation is 1 to 10 V RMS. Higher voltage increases output signal level but also increases power dissipation.
• Stability requirements: Because LVDT sensitivity depends on excitation voltage, ratiometric measurement (comparing output to excitation) provides best accuracy.

LVDT Signal Processing

Processing the LVDT output involves several stages:

1. Differential amplification: Amplify the difference between the two secondary outputs, rejecting common-mode noise.
2. Synchronous demodulation: Multiply by the excitation signal to extract magnitude and sign. This converts the AC signal to DC proportional to position.
3. Low-pass filtering: Remove the residual carrier frequency and its harmonics, leaving only the position-dependent DC component.
4. Ratiometric processing: Dividing by the excitation amplitude or summed secondary outputs improves accuracy when excitation varies.

Resolver Principles

A resolver is a rotary transformer that provides absolute angular position measurement. It contains one rotor winding (excited with AC) and two stator windings oriented 90 degrees apart:

• The rotor winding couples to both stator windings, with coupling depending on rotor angle
• One stator output varies as sin θ times the excitation amplitude, where θ is the rotor angle
• The other stator output varies as cos θ times the excitation amplitude
• Together, sine and cosine signals provide unambiguous angle measurement over a full rotation

Resolver-to-Digital Conversion

Converting resolver outputs to digital angle requires extracting the angle from sine and cosine signals:

• Tracking converter: A feedback system adjusts a digital angle estimate until a resolver pattern generator produces signals matching the actual resolver outputs. Provides continuous tracking with update rates in the tens of kHz and resolutions to 16 bits or higher.
• Direct conversion: Sample both sine and cosine channels and compute the arctangent digitally. Modern converters reach 14 to 16 bits of resolution with update rates above 100 kHz, and the approach suits systems that already digitize the carrier for other reasons.
• Integrated converters: Single-chip resolver-to-digital converters combine the excitation oscillator, the analog input conditioning, and the tracking loop in one device, and they usually add fault detection for a broken or shorted winding. Examples appear below.

Synchro and LVDT Signal Conditioning ICs

Specialized ICs simplify LVDT and resolver interfaces:

• AD598: A complete LVDT signal conditioner with on-chip oscillator and synchronous demodulator, forming (A − B)/(A + B) from the two secondary amplitudes. The ratio makes the output independent of excitation amplitude, but it assumes the sum of the two secondary voltages stays constant over the stroke, which suits five-wire LVDTs.
• AD698: A universal LVDT signal conditioner that instead forms the ratio A/B, dividing the output by a reference taken from the excitation. It does not require the sum of the secondaries to be constant, so it serves four-wire and three-wire LVDTs, half-bridge types, and RVDTs, and drift in the drive oscillator divides out.
• Resolver-to-digital converters: Parts such as the AD2S1210 combine excitation generation, sine and cosine input conditioning, and a tracking loop in one device, with resolution programmable from 10 to 16 bits. The maximum tracking rate falls as the resolution setting rises, so the two must be traded against the fastest shaft speed the system will see.

Photodiode Operating Modes

Photodiodes can operate in two distinct modes:

• Photovoltaic (zero-bias) mode: No external bias, the photodiode generates its own voltage. Offers lowest noise but highest junction capacitance and slowest response. Best for precision low-speed measurements.
• Photoconductive (reverse-bias) mode: External reverse bias reduces junction capacitance and improves speed, but increases dark current (leakage current) and associated noise. Preferred for high-speed applications.

Transimpedance Amplifier Design

The basic transimpedance amplifier connects the photodiode to the op-amp's inverting input, with a feedback resistor setting the gain (transimpedance):

V_out = −I_photoR_f

The op-amp maintains the inverting input at virtual ground, so the photodiode operates at essentially zero bias (or constant bias if a reference voltage is applied to the non-inverting input).

Key design parameters include:

• Feedback resistance: Sets the sensitivity in volts per ampere. A larger R_f improves the signal-to-noise ratio, because its own noise grows only as the square root of its value while the signal grows in proportion, but the available bandwidth falls in inverse proportion to that same square root. Gain and speed are traded here and nowhere else.
• Feedback capacitance: Required for stability wherever the photodiode capacitance is significant, which is to say almost always. Its value must be chosen deliberately rather than inherited from the layout.
• Op-amp selection: Low input bias current keeps the offset across R_f small, low current noise preserves the resolution at high gain, and a high gain-bandwidth product buys both stability margin and signal bandwidth.

Stability Considerations

The photodiode's junction capacitance, in parallel with the op-amp's input capacitance and the stray capacitance of the board, faces the feedback resistor and places a zero in the noise gain. The noise gain then rises toward the falling open-loop response, the two meet with the loop gain dropping at 40 dB per decade, and the result is peaking, ringing, or outright oscillation. Compensation is not optional in any transimpedance amplifier of useful gain.

A capacitor C_f across R_f flattens the noise gain again. Writing C_in for the total capacitance at the inverting input and GBW for the amplifier's gain-bandwidth product, the closed loop is second order, and the bandwidth available with a maximally flat response is:

f_−3dB = √(GBW / (2πR_fC_in))

The feedback capacitor that achieves it is:

C_f = √(2C_in / (2πR_fGBW))

Dropping the factor of 2 inside the radical is a common error. The smaller value leaves a quality factor of 1 rather than 0.707, which is about 45 degrees of phase margin, some 1.25 dB of peaking, and the ringing that goes with it. Take a 1 megohm transimpedance with 20 pF of total input capacitance and a 10 MHz amplifier: the bandwidth is about 280 kHz and C_f is about 0.8 pF. That is comparable with the stray capacitance between two adjacent pads, which is why the feedback resistor is mounted with its ends well separated, why the feedback path is kept short, and why a guard ring around the summing node earns its place.

The bandwidth expression also fixes the terms of the gain trade. Bandwidth falls as the square root of the transimpedance and as the square root of the detector capacitance, so a photodiode of ten times the area, bringing roughly ten times the capacitance, costs a factor of about three in bandwidth. Reverse bias, which thins the depletion region and shrinks its capacitance, is the usual escape wherever the added dark current can be tolerated. Amplifiers intended for this service help as well: decompensated types, stable only above some minimum gain, offer more gain-bandwidth and lower voltage noise than their unity-gain-stable relatives, and monolithic transimpedance amplifiers place the feedback network on the die where stray capacitance is small and predictable.

Noise Analysis

Four noise sources compete in a photodiode amplifier, and which of them dominates depends on the gain, the detector, and the band:

• Thermal noise of R_f: The feedback resistor contributes a noise current of √(4kT/R_f) in each root hertz of bandwidth, about 0.13 picoamperes per root hertz for 1 megohm at room temperature. That current falls as the square root of R_f while the signal it is measured against rises in proportion to R_f, so the signal-to-noise ratio improves as the square root of R_f. Gain taken in the feedback resistor is therefore cheaper than the same gain taken in a following stage, which adds noise without adding signal.
• Amplifier current noise: The input current noise flows in the feedback resistor exactly as the photocurrent does, so it adds to the signal without any dilution. This is why FET and CMOS input stages dominate low-light work, their low current noise and femtoampere-class bias currents mattering more than their higher voltage noise.
• Amplifier voltage noise: The input voltage noise reaches the output multiplied by the noise gain, which climbs from unity at DC toward 1 + C_in/C_f at high frequency. With a large detector and the small feedback capacitor that wide bandwidth demands, that factor reaches tens, and because it applies across the widest part of the band it usually dominates the total noise of a fast transimpedance stage even when the resistor looks worse at DC.
• Shot noise: Both the signal photocurrent and the dark current of a reverse-biased diode carry shot noise of √(2qI) per root hertz. At high light levels this is a floor no circuit technique can go below, and it is one reason photovoltaic operation, with essentially no dark current, is preferred for the smallest signals.

Optimizing the total means balancing these against the bandwidth and the stability the application requires, and the balance usually resolves into a single question: how much bandwidth the measurement can give up in exchange for gain in the feedback resistor.

Wide Dynamic Range Considerations

Light intensity can vary over many orders of magnitude, requiring special techniques for wide dynamic range:

• Switchable gain: Multiple feedback resistors with analog switches enable gain ranging. Care is needed to minimize switch leakage and on-resistance effects.
• Logarithmic amplifiers: Replace the feedback resistor with a diode or transistor junction to produce output proportional to the logarithm of photocurrent, compressing dynamic range.
• Integration mode: Integrate photocurrent on a capacitor for a fixed time, then measure the accumulated voltage. Dynamic range is limited by capacitor size and ADC resolution but can be extended by varying integration time.

Application-Specific Considerations

Different applications emphasize different performance aspects:

• Optical communications: Highest bandwidth (GHz range) with moderate noise performance. Uses reverse-biased PIN or avalanche photodiodes with specialized wideband amplifiers.
• Spectroscopy: Low noise and excellent linearity over wide dynamic range. Often uses zero-bias operation and lock-in detection with chopped light sources.
• Position sensing: Multiple photodiodes in arrays or quadrant configurations. Matching and ratiometric measurement minimize errors from intensity variations.
• Power measurement: Precision current measurement with calibrated responsivity. Often includes temperature compensation for the photodiode's temperature-dependent sensitivity.

Piezoelectric Sensor Characteristics

Piezoelectric sensors behave electrically as a charge source in parallel with a capacitance (the sensor's own dielectric):

• Charge sensitivity: Expressed in picocoulombs per unit input (pC/g for accelerometers, pC/N for force sensors). Typical values range from 1 to 100 pC per unit.
• Sensor capacitance: Typically 100 to 2000 pF, affecting cable and amplifier design.
• Frequency response: Extends from very low frequencies (limited by amplifier time constant) to the sensor's resonance frequency, which can exceed 100 kHz.
• Temperature effects: Sensitivity and capacitance vary with temperature, requiring compensation for precise measurements.

Charge Amplifier Operating Principle

A charge amplifier places a capacitor in the feedback path of an inverting op-amp configuration. The op-amp maintains virtual ground at its inverting input, forcing all sensor charge to transfer to the feedback capacitor:

V_out = −Q/C_f

Here Q is the charge delivered by the sensor and C_f is the feedback capacitance. Because the summing node is held at virtual ground, no voltage develops across the sensor or its cable, so their capacitance draws no charge and the sensitivity is independent of cable length. That independence is the whole reason for the topology, and it is lost the moment the amplifier's open-loop gain falls low enough for the summing node to move.

Low-Frequency Response and Bias

A charge amplifier cannot work without a DC path around its feedback capacitor. The op-amp's input bias current has nowhere else to go, and integrated on C_f it walks the output into a rail; the capacitor alone also leaves the inverting node without a defined operating point. A high-value resistor R_f across C_f supplies both: it drains the bias current and sets the DC gain, at the cost of turning the integrator into a high-pass filter with its corner at:

f_c = 1 / (2πR_fC_f)

The same time constant R_fC_f governs how quickly the response to a step decays back toward zero, so a long event needs a corner well below its own frequency content. A 1 Hz corner with 100 pF of feedback capacitance calls for 1.6 gigohms, which demands a specialist resistor, a guarded layout, and an amplifier whose bias current is measured in picoamperes: 1 picoampere across 1.6 gigohms is already 1.6 mV of output offset, and a nanoampere-class part would saturate. Active bias arrangements, which present a high incremental resistance while offering a low-resistance DC path, are the alternative where the resistor itself becomes impractical.

This corner is also the formal reason a piezoelectric sensor cannot measure a static quantity. The charge produced by a constant force bleeds away through the feedback resistor, through the sensor's own insulation resistance, and through the cable, so the reading decays to zero however carefully the circuit is built. Piezoelectric sensors measure change; a genuinely static force calls for a strain gauge, a capacitive sensor, or a piezoresistive one.

Practical Charge Amplifier Design

Key design considerations include:

• Input protection: Large transients can generate damaging charge levels. Input clamping diodes or resistors limit voltage excursions.
• Cable considerations: Special low-noise cables minimize triboelectric noise (charge generated by cable flexing). Keep cable runs short when possible.
• Shielding: High-impedance charge inputs are susceptible to electromagnetic interference. Use shielded cables with the shield connected to signal ground.
• Feedback capacitor selection: Use low-leakage capacitors (polystyrene, polypropylene, or C0G ceramic) to minimize drift and maintain low-frequency response.
• Op-amp selection: Choose op-amps with low input bias current (FET input or electrometer types), low noise, and adequate bandwidth for the application.

IEPE (ICP) Sensors

Integrated Electronics Piezoelectric (IEPE), also known as ICP (Integrated Circuit Piezoelectric, a PCB Piezotronics trademark), sensors incorporate a charge amplifier within the sensor housing, providing a voltage output that is easier to interface:

• Constant current excitation: IEPE sensors require a constant current bias, typically 2 to 20 mA, delivered along the same coaxial cable that carries the signal. The internal amplifier modulates that current so the signal rides on the bias voltage, and the current available sets how much cable the sensor can drive at high frequency.
• AC coupling: The output sits on a DC bias, typically around 8 to 14 volts, which must be removed by AC coupling before the signal can be amplified. The bias voltage doubles as a health check, since an open or shorted sensor drives it to one extreme or the other.
• Simplified interface: Standard coaxial cable can be used over long distances without the noise sensitivity of charge-mode operation.
• Limited temperature range: The internal electronics limit operating temperature compared to charge-mode sensors.

Applications

Piezoelectric sensors and charge amplifiers find use in:

• Vibration measurement: Accelerometers for machinery monitoring, modal analysis, and seismic sensing
• Dynamic force measurement: Force sensors for impact testing, cutting force measurement, and ballistic studies
• Dynamic pressure: High-frequency pressure transducers for combustion analysis and acoustic measurements
• Acoustic emission: Ultrasonic sensors for non-destructive testing and condition monitoring
• Ultrasonic applications: Transducers for medical imaging, NDT, and sonar systems

System-Level Considerations

Successful sensor interface design requires attention to the complete signal path:

• Grounding: Establish a clear grounding strategy early in design. Separate analog and digital grounds, connect at a single point, and use ground planes to minimize impedance.
• Shielding: Shield sensitive signals from interference sources. Connect shields at one end only (usually the amplifier end) to prevent ground loop currents.
• Power supply considerations: Provide clean, stable power to analog circuits. Use separate regulators for analog and digital sections and include adequate decoupling.
• Temperature management: Consider heat sources in the system and their effect on precision components. Thermal gradients across matched components degrade matching.

Component Selection

Component quality significantly affects interface performance:

• Resistors: Use metal film or thin film for low noise and tight tolerance. Consider temperature coefficient for precision applications.
• Capacitors: Film capacitors (polypropylene, polystyrene) for precision timing and filtering. Avoid general-purpose ceramic types in signal paths.
• Connectors: Choose connectors rated for low-level signals. Gold plating minimizes contact resistance and oxidation.
• PCB material: FR-4 suffices for most applications, but high-frequency or very low leakage designs may require specialized materials.

Calibration and Testing

Proper calibration ensures accuracy:

• End-to-end calibration: Calibrate the complete system including sensor, wiring, and electronics rather than individual components.
• Known reference inputs: Use traceable standards appropriate for the measurement type (calibrated weights, temperature baths, precision resistors).
• Multi-point calibration: Characterize nonlinearity with measurements at multiple points across the operating range.
• Temperature characterization: Measure performance across the expected operating temperature range to enable compensation.