4. Signal conditioning fundamentals for crash DAQ

By the time a crash event begins, the data acquisition chain has already made most of its critical decisions. The accelerometers, load cells, strain gauges, and potentiometers mounted on the vehicle and dummy produce signals that are physically meaningful but electrically fragile: millivolt-level differential outputs, high source impedance, common-mode noise from the vehicle electrical system, and bandwidth content that extends well beyond the range of interest. Signal conditioning is the stage that converts these raw transducer outputs into signals the digitizer can sample accurately and unambiguously. In a crash test, where the entire event lasts 100–200 ms and there is no opportunity to repeat the run, conditioning errors are not recoverable.

This chapter covers the four conditioning functions that appear in essentially every crash DAQ channel — amplification, filtering, anti-aliasing, and bridge completion with excitation — and then addresses the noise and mechanical challenges specific to the high-g environment.

Amplification and Gain Staging

Crash transducers typically produce small signals. A piezoresistive accelerometer with a full-scale range of ±2000 g and a sensitivity of 0.05 mV/g/V excitation produces only 100 mV at full scale when driven with 1 V. A strain-gauge load cell might output 2 mV/V at rated load. If the digitizer has a ±10 V input range and a 16-bit ADC, one LSB corresponds to about 305 µV. A 100 mV full-scale signal would occupy only about 328 codes — roughly 8.4 effective bits. That is unacceptable for measuring peak deceleration and pulse shape.

Amplification solves this by scaling the signal to fill the ADC input range. The required gain is straightforward:

Gain = V_ADC_fullscale / (Sensitivity × Excitation × Full-scale measurand)

Example: accelerometer, ±2000 g, 0.05 mV/g/V, 1 V excitation
V_signal_fs = 0.05 mV/g/V × 1 V × 2000 g = 100 mV
Gain = 10 V / 0.1 V = 100 (40 dB)

In practice, gain is set slightly below the theoretical maximum to preserve headroom for overshoot and to avoid clipping. A common approach is to target 80–90% of ADC full scale at the expected peak. For the example above, a gain of 80 (38 dB) would place the expected peak at 8 V, leaving 2 V of headroom.

Two practical constraints govern gain staging. First, the amplifier's input-referred noise must be low enough that the amplified noise floor does not degrade resolution. A typical instrumentation amplifier for crash DAQ has an input noise density of 10–20 nV/√Hz; over a 10 kHz bandwidth, this contributes about 1–2 µV RMS, which is well below the transducer's own noise. Second, the amplifier must remain linear during the high-g event. Crash accelerometers can experience overloads of 2–3× their rated range during hard impacts, and the amplifier must either handle this linearly or recover quickly from saturation. An amplifier that clips and takes milliseconds to recover will corrupt the entire pulse.

Filtering and Anti-Aliasing

Every crash signal contains energy above the band of interest. The mechanical structure rings at its natural frequencies, the transducer has its own resonant frequency (often 20–50 kHz for accelerometers), and the vehicle electrical system injects noise across a broad spectrum. If these high-frequency components are not removed before sampling, they alias — folding back into the baseband and appearing as false low-frequency content that cannot be distinguished from real data.

The Nyquist criterion requires that the sampling rate be at least twice the highest frequency component in the signal. For crash data, the band of interest is typically DC to 1–3 kHz (some standards specify 1 kHz for occupant response, 3 kHz for structural response). A sampling rate of 10–20 kS/s is common, giving a Nyquist frequency of 5–10 kHz. An anti-aliasing filter must attenuate everything above the Nyquist frequency before the ADC.

A practical anti-aliasing filter is a low-pass filter with a corner frequency at or slightly above the band of interest and a steep roll-off. For a 10 kS/s sample rate with a 3 kHz band of interest, a 4th-order Butterworth filter with a 3 kHz corner provides about 24 dB/octave attenuation, reaching roughly 60 dB at 10 kHz. This is usually sufficient. The filter must be placed before the ADC — after the amplifier, before the digitizer input — and its cutoff must be stable over temperature and time.

Anti-aliasing filters are not optional. If the filter is omitted or its corner frequency is set too high, high-frequency energy from transducer resonance or electrical noise will alias into the data. The resulting error is not obvious in the time domain but appears as spurious low-frequency content in the spectrum. Once aliased, it cannot be removed by post-processing.

Bridge Completion and Excitation

Strain gauges and piezoresistive accelerometers are resistive elements that change resistance in response to strain or acceleration. They are almost always configured in a Wheatstone bridge, which converts the small resistance change into a differential voltage. The bridge requires three things: completion resistors to form the remaining legs, a stable excitation voltage, and a differential amplifier to extract the bridge output.

In a quarter-bridge configuration, one element is the active gauge and three are completion resistors. The bridge output is:

V_out = V_exc × (ΔR / (4R + 2ΔR)) ≈ V_exc × (GF × ε / 4)

where GF = gauge factor, ε = strain

For a typical strain gauge with GF = 2 and a strain of 1000 µε, the bridge output is approximately 0.5 mV/V. With 5 V excitation, that is 2.5 mV — a small signal that requires amplification and careful noise management.

Excitation must be stable because the bridge output is directly proportional to it. A 1% drift in excitation produces a 1% error in the measured strain. In crash testing, excitation is typically provided by a precision voltage reference or a constant-current source. Constant-current excitation has the advantage of automatically compensating for lead resistance changes, but it requires the bridge completion resistors to be matched to the gauge resistance.

Bridge completion resistors must be physically located close to the gauge to minimize lead resistance effects. In a quarter-bridge configuration, the lead resistance adds to the gauge resistance and introduces a gain error. Three-wire and four-wire configurations are used to compensate for this, but they require additional conductors in the cable — a significant consideration when routing cables through a vehicle that will be destroyed.

Noise Mitigation in High-g Environments

The crash environment is electrically hostile. The vehicle's electrical system, the firing of pyrotechnic devices, the mechanical shock of the impact, and the triboelectric effects of cables flexing and striking surfaces all contribute noise. Three techniques are essential.

Differential signal transmission. Every transducer signal should be routed as a twisted pair or shielded twisted pair, with the signal carried differentially. The instrumentation amplifier at the input rejects common-mode noise — the noise that appears equally on both conductors. A good instrumentation amplifier achieves a common-mode rejection ratio (CMRR) of 80–100 dB at DC and 60–80 dB at 1 kHz. This is the single most effective noise mitigation technique.

Shielding and grounding. Cable shields must be connected to the DAQ ground at one end only — typically the DAQ end — to avoid ground loops. The transducer end of the shield is left floating. If the transducer housing is grounded to the vehicle structure, the shield should be connected at the DAQ end and the transducer should be isolated from the structure, or a ground-isolation amplifier should be used. Ground loops in a crash test can inject tens of millivolts of noise, which is comparable to the signal itself.

Mechanical strain relief. Cables in a crash test experience severe mechanical stress. They must be strain-relieved at both the transducer and the DAQ connector to prevent the cable from pulling on the connector or the transducer. Cable motion during the crash generates triboelectric noise — charge generated by the cable's dielectric and conductor rubbing together. This noise is proportional to cable length and motion velocity. Keeping cables short, securing them to the vehicle structure, and using low-triboelectric-noise cable materials (such as PTFE dielectric) reduces this effect.

Crash DAQ Signal Conditioning Chain Transducer Accelerometer Load cell / Strain gauge Potentiometer Bridge Completion Quarter / Half / Full Excitation: 1–5 V Constant V or I Amplifier Instrumentation amp Gain: 10–1000 CMRR: 80–100 dB Anti-Alias Filter Low-pass, 4th order fc = 3 kHz −24 dB/oct ADC 16-bit 10–20 kS/s 0.5–100 mV mV-level diff 1–10 V band-limited digital codes Noise Sources and Mitigation in High-g Environment Common-mode noise Vehicle electrical system → Differential input + CMRR Ground loops Multiple ground paths → Single-point shield ground Triboelectric noise Cable flexing during impact → PTFE dielectric, strain relief Transducer resonance 20–50 kHz ringing → Anti-alias filter Aliasing: Why Anti-Alias Filtering Is Mandatory Frequency Amplitude Signal band 0–3 kHz Noise / resonance 8–20 kHz Nyquist (5 kHz) Aliased folds back Anti-alias filter response Frequency passband −24 dB/oct Nyquist ≥60 dB

Practical Configuration Example

Consider a typical frontal impact test with a 50th-percentile male dummy. The head accelerometer is a piezoresistive device with a ±2000 g range, 0.05 mV/g/V sensitivity, and 1 V excitation. The chest accelerometer is a similar device with a ±500 g range. The femur load cell is a strain-gauge type with 2 mV/V at 10 kN rated load and 5 V excitation.

For the head channel, the signal at 2000 g is 100 mV. With a target of 8 V at the ADC input, the required gain is 80. The anti-alias filter is set to 3 kHz with a 4th-order Butterworth response. The sampling rate is 20 kS/s, giving a Nyquist frequency of 10 kHz. At 10 kHz, the filter attenuates by approximately 60 dB, reducing the 100 mV signal to 0.1 mV — well below the ADC's noise floor.

For the femur load cell, the signal at 10 kN is 10 mV. With a target of 8 V, the required gain is 800. The same 3 kHz anti-alias filter is used. The excitation is 5 V from a precision reference with 0.05% stability. The bridge completion resistors are matched to the gauge resistance within 0.01% to minimize offset drift.

In both cases, the cables are shielded twisted pairs with the shield grounded at the DAQ end only. The cables are secured to the vehicle structure every 150 mm to minimize triboelectric noise. The transducer housings are isolated from the vehicle structure where possible; where isolation is not possible, a ground-isolation amplifier is inserted between the transducer and the DAQ input.

Do not assume that a channel is working correctly because it produces a signal. Verify the entire conditioning chain before the test: inject a known calibration signal at the transducer connector, confirm the expected output at the ADC, and check that the noise floor is below the required resolution. A channel that appears to work but has a ground loop or a misconfigured filter will produce data that looks plausible but is wrong.

Verification and Troubleshooting

Before every crash test, each channel should be verified end-to-end. The standard method is a shunt calibration: a known resistor is placed in parallel with one arm of the bridge, producing a known equivalent strain or acceleration. The DAQ should read the expected value within tolerance. If it does not, the gain, excitation, or bridge completion is incorrect.

Common failure symptoms and their causes:

  • Noisy signal with 50/60 Hz content: Ground loop or missing shield connection. Check that the shield is grounded at one end only and that the transducer is isolated from the vehicle structure.
  • Signal clipped at high g: Gain set too high or amplifier saturation. Reduce gain or use an amplifier with a higher overload recovery specification.
  • Unexplained low-frequency content in the spectrum: Aliasing. Verify that the anti-alias filter is present, correctly configured, and placed before the ADC.
  • Zero offset drift during the test: Excitation instability or bridge completion resistor mismatch. Use a precision reference and matched resistors.
  • Erratic signal during cable motion: Triboelectric noise. Replace cable with low-noise dielectric, improve strain relief, and secure the cable to the structure.

Signal conditioning is the foundation of crash data quality. The transducer may be perfect and the ADC may be flawless, but if the conditioning chain is misconfigured, the data will be unusable. The techniques in this chapter — proper gain staging, anti-alias filtering, stable bridge excitation, differential transmission, and disciplined grounding — are not optional refinements. They are the minimum requirements for producing data that can be trusted in a crash test.

← Previous 📖 Contents Next →