2. Signal conditioning and filtering for crash data

Raw crash signals arriving from accelerometers, load cells, and potentiometers are never ready for direct injury-criterion computation. They carry sensor noise, structural ringing, electrical interference, and — if digitized carelessly — aliased high-frequency content folded back into the passband. Signal conditioning is the disciplined sequence that turns these raw voltages into defensible, standards-compliant time histories. This chapter covers the three pillars of that sequence: CFC channel frequency class filtering, anti-aliasing before sampling, and data quality validation that proves the conditioned signal is trustworthy before it feeds HIC, chest deflection, or femur-load calculations.

Why Filtering Is Mandated, Not Optional

Occupant injury criteria are defined against filtered data. SAE J211/1 and ISO 6487 specify channel frequency classes (CFC) that bound the frequency content of a measurement channel so that results are comparable across laboratories, sleds, and test campaigns. A HIC computed from unfiltered head acceleration will be inflated by high-frequency structural ringing that the human head cannot physically transmit; a chest deflection filtered too aggressively will under-report peak compression. The filter is therefore part of the measurement definition, not a cosmetic cleanup step.

Each CFC specifies a low-pass characteristic with a defined −3 dB cutoff and a minimum attenuation at a stopband frequency. The commonly used classes are:

The CFC filter is a fourth-order Butterworth low-pass applied in both forward and reverse directions (zero-phase, filtfilt-style), so the effective magnitude response is squared (eighth-order) while phase distortion is cancelled. This bidirectional application is why the −3 dB point of the single-pass prototype is chosen so that the two-pass result meets the class definition. Implementations must respect the standard's tolerance band on the transition region, not merely hit the cutoff.

Anti-Aliasing: The Step That Cannot Be Undone

Digitization samples the continuous signal at rate f_s. Any energy above the Nyquist frequency f_s/2 is folded back into the baseband and becomes indistinguishable from genuine low-frequency content. No post-hoc digital filter can remove it. The defense is an analog anti-aliasing low-pass filter placed before the ADC, with its corner set below Nyquist and its stopband attenuation sufficient to push aliased energy below the measurement noise floor.

A practical rule: sample at least 10× the highest frequency of interest. For a CFC 1000 channel (content to ~1650 Hz), that implies f_s ≥ 16.5 kHz; crash DAQ systems commonly run 20–100 kHz per channel. The analog front-end filter is typically a 2nd- or 4th-order Butterworth with corner at roughly 0.4–0.45 f_s, giving ≥ 40 dB attenuation at Nyquist. The digital CFC filter is then applied after sampling and is a separate, standards-mandated operation — it does not substitute for the analog anti-aliasing filter.

Data Quality Validation Before Injury Computation

Conditioned data must pass a validation gate before it is used. The checks below are the minimum defensible set; each has a concrete pass/fail symptom.

• Pre-trigger baseline and offset: the pre-event window (typically 20–50 ms before T₀) should sit at zero mean with low variance. A non-zero mean indicates DC offset or sensor drift; subtract the pre-trigger mean before filtering.
• Clipping and saturation: check for flat-topped peaks at the ADC full-scale rail. Clipped peaks invalidate peak acceleration and any criterion derived from them.
• Dropout and spike detection: single-sample spikes (often from cable flex or EMI) survive a CFC filter as ringing. Flag samples exceeding a robust threshold (e.g., median + 8×MAD) and inspect the raw trace.
• Filter verification: confirm the applied CFC class matches the channel's test-plan requirement, and confirm the filter was applied once, bidirectionally, at the correct sample rate.
• Integration sanity: for acceleration channels, integrate to velocity and check that the post-impact velocity returns toward zero (or matches the known sled Δv). A large residual indicates offset or filter-induced drift.

Worked Example: Conditioning a Head Acceleration Channel

Consider a driver head accelerometer channel recorded at f_s = 20 kHz, test plan requiring CFC 60 for HIC. The conditioning sequence and expected observations are:

1. Verify the analog front end. Confirm the DAQ anti-aliasing filter is enabled with corner ≈ 9 kHz (0.45 × 20 kHz) and ≥ 40 dB attenuation at 10 kHz. If the channel was recorded with the AA filter bypassed, the data is suspect regardless of downstream processing.
2. Remove DC offset. Compute the mean over the 30 ms pre-trigger window and subtract it. Expected: post-subtraction baseline mean ≈ 0, standard deviation at the noise floor (e.g., < 0.5 m/s² for a 500 g sensor).
3. Apply the CFC 60 filter. Use a 4th-order Butterworth with the standard's −3 dB point (≈ 100 Hz for the single-pass prototype), applied forward and reverse. Verify the effective response meets the CFC 60 tolerance band.
4. Check for clipping and spikes on the raw (pre-filter) trace. A flat top at ±full-scale or isolated single-sample excursions are grounds for rejection or re-test.
5. Integrate to velocity. Integrate the filtered acceleration; the post-impact velocity should return near zero (or match the sled Δv within a few percent). A large residual points to residual offset or an incorrect filter corner.

Only after all five checks pass should the channel be released to the HIC integrator. The validation gate is deliberately placed after filtering but before criterion computation, because filtering can mask or reshape the very defects the gate is designed to catch.

Filter Class Selection and Consistency

The CFC class is dictated by the injury criterion and the test protocol, not by analyst preference. Head acceleration for HIC uses CFC 60; chest acceleration for the chest injury criterion uses CFC 60; femur load uses CFC 180; belt force uses CFC 600. Mixing classes across channels that feed a single criterion, or re-filtering an already-filtered channel, produces results that are not comparable to the standard and cannot be defended in a test report. Document the applied class, filter order, direction (zero-phase), and sample rate for every channel in the data-reduction log — this metadata is as important as the numbers themselves.

In summary, signal conditioning is a chain with an irreversible step (anti-aliasing) and a standards-defined step (CFC filtering), bracketed by validation that proves the chain did what it claims. Treat the analog anti-aliasing filter as a hardware prerequisite, the digital CFC filter as a protocol-mandated transformation, and the validation gate as the non-negotiable checkpoint before any injury criterion is computed.