3. Anti-aliasing filter fundamentals
Every crash data acquisition chain begins with a sensor, passes through some conditioning electronics, and ends at an analog-to-digital converter (ADC). The anti-aliasing filter sits at the boundary between the continuous-time world and the sampled-data world, and its job is deceptively simple: remove energy above the Nyquist frequency before sampling so that out-of-band content cannot fold back into the band of interest. In crash testing this matters more than in most measurement domains, because the signals of interest — acceleration pulses, occupant kinematics, belt loads — are broadband transients whose spectra extend well beyond the frequencies engineers care about, and because the post-processing (CFC filtering, injury criteria computation) assumes the sampled record is free of aliased artifacts.
This section explains how anti-aliasing filters are built, why the analog/digital split exists, how filter order shapes the transition band, and what practical trade-offs govern the choice in a crash acquisition system.
Why Aliasing Is Irreversible
Sampling at rate fs maps any input frequency f to a discrete-time frequency that is indistinguishable from f ± k·fs. A 1,200 Hz component sampled at 10 kHz appears at 1,200 Hz; the same component sampled at 2 kHz appears at 800 Hz — inside the passband, and impossible to separate from genuine 800 Hz content afterward. No digital post-processing can undo this, because the information has been destroyed at the sampling instant. This is the single most important reason the anti-aliasing filter must be analog and must precede the ADC.
Analog and Digital Stages Working Together
Real crash DAQ front-ends almost never rely on a single filter. A typical architecture uses a cascade: a fixed analog low-pass at the input for anti-alias protection and noise limiting, followed by the ADC, followed by a digital decimation/filtering stage that sets the final usable bandwidth and sample rate. The analog stage must be flat enough in-band and steep enough out-of-band to keep aliased energy below the resolution of the measurement; the digital stage provides the precise, drift-free, linear-phase response that is difficult to achieve with op-amps and capacitors alone.
The figure shows the two-stage concept and the corresponding magnitude responses. Note that the analog stage alone does not need to be brick-wall: it only has to push energy above fs/2 low enough that the residual aliased components fall below the noise floor or the required measurement resolution. The digital stage then sharpens the final band edge without adding analog component tolerance, temperature drift, or phase distortion.
Filter Order and the Transition Band
Filter order determines how fast the response rolls off. For a Butterworth prototype, the asymptotic slope is −20·n dB/decade, where n is the order. A 2nd-order filter gives −40 dB/decade; a 4th-order gives −80 dB/decade; an 8th-order gives −160 dB/decade. The practical consequence is the width of the transition band between the highest frequency of interest and the Nyquist frequency.
Consider a concrete design example. Suppose a crash channel must preserve content to 1,000 Hz (CFC 1000 class), the ADC samples at 10 kHz, and the analog anti-aliasing filter is a 4th-order Butterworth with a −3 dB corner at 1,500 Hz. At the Nyquist frequency of 5,000 Hz, the normalized frequency is 5,000/1,500 ≈ 3.33, so the attenuation is approximately 80·log10(3.33) ≈ 42 dB. If the input signal contains a strong 5 kHz component at, say, 100 mV, the aliased residue is about 0.8 mV — potentially significant against a 10 mV full-scale signal. Raising the order to 8 (or lowering the corner to 1,200 Hz) would push that residue below 0.1 mV. These numbers are illustrative design estimates, not measured results, but they show the arithmetic an engineer actually performs when sizing the filter.
Analog vs. Digital: The Real Trade-offs
| Property | Analog anti-aliasing filter | Digital filter |
|---|---|---|
| Position in chain | Before the ADC — mandatory for alias protection | After the ADC — cannot remove aliasing already folded in |
| Component tolerance / drift | Affected by resistor/capacitor tolerance and temperature | Coefficients are exact and drift-free |
| Phase response | Nonlinear unless Bessel or all-pass corrected | Can be made exactly linear-phase (FIR) |
| Order achievable | 2nd–8th practical; higher orders costly and noisy | Hundreds of taps feasible |
| Latency | Low, continuous-time | Group delay proportional to tap count |
| Noise contribution | Adds thermal noise and possible distortion | Quantization-limited only |
The key insight is that the two stages are not interchangeable. The analog filter is a protection device; the digital filter is a precision device. A common mistake in crash DAQ configuration is to set the digital decimation filter aggressively while leaving the analog front-end wide open, assuming the digital stage will clean things up. It will not — any energy above Nyquist that reaches the ADC is already aliased.
Practical Selection Guidance for Crash Channels
For a typical vehicle crash channel, the following reasoning sequence works well:
- Determine the highest frequency of interest. For CFC 1000 channels this is 1,000 Hz; for CFC 600 it is 600 Hz. Add margin for the filter transition band.
- Choose the sample rate. A common rule is fs ≥ 5–10× the highest frequency of interest, giving room for the analog roll-off before Nyquist.
- Size the analog filter. Pick the lowest order that keeps aliased energy below the required resolution at fs/2, accounting for worst-case component tolerance. A 4th-order Butterworth at 1.2–1.5× the band edge is a frequent starting point.
- Select the digital stage. Use a linear-phase FIR decimator to set the final band edge and sample rate; verify group delay is acceptable for the event duration.
- Verify with a known input. Inject a sine sweep or a step and confirm the composite response meets the CFC channel class tolerance before the test, not after.
In summary, the anti-aliasing filter is a two-stage contract: the analog stage guarantees that nothing above Nyquist survives to the ADC, and the digital stage delivers the precise, linear-phase band limiting that the measurement standard requires. Filter order is the lever that trades transition-band width against complexity, noise, and cost — and in crash acquisition, where the signals are transient and the post-processing assumptions are strict, that lever deserves deliberate attention rather than a default setting.