3. Signal conditioning fundamentals

Crash pulse measurements begin with a transducer that converts mechanical motion into an electrical signal, but that raw output is rarely ready for a data acquisition system. A piezoelectric accelerometer may deliver only a few millivolts per g, ride on a large DC bias, and carry high-frequency noise from structural ringing, cable triboelectrics, and electromagnetic interference. Signal conditioning is the stage that turns this fragile analog output into a clean, scaled, band-limited voltage that an ADC can digitize without losing the physical information the test exists to capture. In a crash test, that information is the acceleration–time history whose double integration yields velocity and displacement, and whose peak and duration drive injury criteria such as HIC and chest deflection thresholds.

Three operations dominate this stage: amplification, filtering, and anti-aliasing. They are not independent. The gain you choose interacts with the noise floor and the ADC full-scale range; the filter corner you set determines how much of the pulse's high-frequency content survives; and the anti-aliasing filter must be matched to the sampling rate before the ADC, not after it. Getting any one of these wrong produces data that looks plausible on a plot but is physically wrong.

Amplification and Signal Scaling

The first job is to raise the transducer output to a level that uses a useful fraction of the ADC's input range. Consider a typical IEPE (Integrated Electronics PiezoElectric) accelerometer with a sensitivity of 10 mV/g and a measurement range of ±500 g. At full scale it outputs ±5 V, which is convenient. But a low-level accelerometer with 1 mV/g and a ±100 g range outputs only ±100 mV at full scale. If the ADC has a ±10 V range, the signal occupies just 1% of the available codes, and quantization noise becomes significant relative to the signal.

An instrumentation amplifier with a gain of 100 brings that ±100 mV to ±10 V, matching the ADC range. The gain is set by a single resistor in most instrumentation amplifier topologies:

G = 1 + (49.4 kΩ / R_gain)

For G = 100:  R_gain = 49.4 kΩ / 99 ≈ 499 Ω

Two practical constraints govern gain selection. First, the amplified signal plus any DC offset must not exceed the ADC input range, or the signal clips and the peak acceleration is lost. Second, the amplifier's own input-referred noise must be well below the transducer noise floor; otherwise the gain stage adds noise rather than improving SNR. A good rule is to set gain so the expected peak signal reaches 70–80% of ADC full scale, leaving headroom for transients and offset drift.

Filtering: Removing What Is Not the Crash Pulse

Crash pulses contain energy from DC to several kilohertz. The rigid-body deceleration of the vehicle occupies roughly 0–100 Hz for a full frontal impact, while structural ringing and sensor resonance can extend to 10 kHz or more. Filtering separates the physically meaningful band from the rest.

Two filter types are standard in crash signal conditioning. A low-pass filter removes high-frequency noise and sensor resonance above the band of interest. A high-pass filter removes DC offset and low-frequency drift, particularly important when the signal will be integrated, because any DC offset integrates into a ramp that corrupts velocity and displacement. Many conditioning modules combine both into a band-pass response.

Filter order matters. A first-order Butterworth low-pass at 1 kHz rolls off at 6 dB/octave, which may not attenuate a 5 kHz resonance enough. A fourth-order Butterworth at the same corner rolls off at 24 dB/octave and provides much stronger rejection while maintaining a maximally flat passband. The trade-off is phase distortion near the corner frequency, which can shift the apparent timing of the pulse peak. For injury metric calculations where timing matters, a linear-phase (Bessel or FIR) filter is often preferred.

SAE J211 specifies channel filter classes for crash testing. A Class 1000 filter has a corner frequency of 1000 Hz and is used for accelerometer channels where the pulse content is below a few hundred hertz. A Class 60 filter (60 Hz corner) is used for displacement or chest deflection channels. The class number refers to the filter's −3 dB cutoff, and the standard defines the required attenuation at specified multiples of that frequency.

Anti-Aliasing: The Non-Negotiable Step

An ADC samples the conditioned signal at discrete intervals. If the signal contains frequency components above half the sampling rate — the Nyquist frequency — those components fold back into the digitized data as false lower-frequency signals. This is aliasing, and it is irreversible: once aliased, the false content cannot be distinguished from real data.

For a crash pulse sampled at 10 kHz, the Nyquist frequency is 5 kHz. Any energy above 5 kHz must be attenuated before the ADC. The anti-aliasing filter is a low-pass filter placed immediately before the ADC, with its corner frequency set below Nyquist and its stopband attenuation sufficient to push aliased components below the noise floor.

A practical design rule: set the anti-aliasing corner at one-fifth to one-half of the sampling rate, and choose the filter order so that the response is at least 60–80 dB down at the Nyquist frequency. For a 10 kHz sample rate with a fourth-order Butterworth at 2 kHz, the attenuation at 5 kHz is approximately 48 dB — often adequate. For more demanding measurements, an eighth-order filter or a higher sampling rate is used.

Putting the Chain Together

The complete conditioning chain for a crash pulse measurement is: transducer → constant-current source (for IEPE) → AC coupling → amplification → anti-aliasing low-pass filter → ADC. Each stage has a defined role, and the parameters of each stage must be chosen with the others in mind.

The frequency-domain view makes the division of labor clear. The passband, extending from the AC coupling corner up to the anti-aliasing corner, is where the crash pulse lives. The stopband above Nyquist is where noise and resonance must be suppressed before sampling. The transition band between them is governed by the filter order: a steeper roll-off allows a higher corner frequency and preserves more signal bandwidth, at the cost of more phase distortion and circuit complexity.

Practical Parameter Selection

For a full frontal crash test with a 10 kHz sampling rate and a 16-bit ADC, a defensible starting configuration is:

These values are a starting point, not a universal prescription. A side impact with a shorter, sharper pulse may require a higher anti-aliasing corner and a higher sampling rate. A chest deflection measurement using a different sensor type may need a different gain and coupling network. The principle remains: match the conditioning chain to the physical bandwidth of the event, and always place the anti-aliasing filter before the ADC.

Verification and Troubleshooting

After configuring the chain, verify it before the test. Inject a known calibration signal — a 100 mV, 100 Hz sine wave is standard — and confirm that the digitized amplitude matches the expected value after gain and filtering. Check the noise floor with the sensor at rest: a typical IEPE channel should show less than 10 mV RMS of noise after conditioning. If the noise floor is higher, suspect cable shielding, ground loops, or a damaged constant-current source.

Common failure symptoms and their causes:

• Clipped peaks: gain too high for the expected acceleration; reduce gain or use a sensor with a wider range.
• Ramp in integrated velocity: DC offset present; check AC coupling capacitor and amplifier offset voltage.
• High-frequency hash on the pulse: anti-aliasing filter corner too high or filter bypassed; verify filter is in circuit and corner is below Nyquist.
• False low-frequency oscillation: aliasing from an unfiltered high-frequency component; increase filter order or lower the corner.
• Phase shift in peak timing: filter order too high or corner too close to the pulse bandwidth; use a linear-phase filter or increase the corner frequency.

Signal conditioning is where the physical fidelity of a crash pulse measurement is won or lost. Amplification sets the resolution, filtering defines the bandwidth, and anti-aliasing protects the digital record from irreversible corruption. Each parameter is a deliberate choice, and each choice has a consequence that shows up in the injury metrics calculated from the data.