3. Dynamic range fundamentals

Every preimpact sensing channel — whether it is a MEMS accelerometer watching for a crash pulse or a pressure transducer monitoring a hydraulic brake circuit — must faithfully represent a physical quantity across a wide span of magnitudes. That span, and the quality of the representation within it, is what we mean by dynamic range. In the context of zeroing and range verification, dynamic range is not an abstract datasheet number; it is the set of conditions under which the sensor output still carries usable information about the input. Once the input leaves that set, the channel either saturates, clips, or drowns in noise, and any downstream zeroing or threshold logic built on that channel becomes unreliable.

This section builds the vocabulary and the quantitative relationships you need to read a datasheet critically, to predict where a channel will fail, and to design a verification test that actually exercises the limits rather than merely confirming nominal operation.

Full-scale range, saturation, and clipping

An accelerometer's full-scale range (FSR) is the maximum input magnitude the signal chain can represent before the output stops tracking the input. For a ±50 g part, the specified measurement range is −50 g to +50 g. Beyond that, the output does not simply stop at a clean rail; the internal amplifier or ADC reaches its limit and the transfer function becomes nonlinear. Saturation is the condition in which the output no longer responds proportionally to further increases in input. Clipping is the visible consequence: the waveform is flattened at the limit, and the peaks are truncated.

The distinction matters because clipping is not a harmless cosmetic effect. A clipped acceleration pulse loses its peak amplitude, and any integration performed on it (for velocity or crash severity estimation) will underestimate the true impulse. Worse, the flat top introduces high-frequency harmonics that were not present in the original signal, which can excite resonances in downstream filters and produce spurious detections. For pressure sensors, clipping during a brake-pressure spike can hide the true peak and cause a controller to under-command or over-command the actuator on the next cycle.

Recovery from saturation is also a specification worth checking. Some signal chains exhibit overload recovery time — a period after the input returns to range during which the output remains invalid. In a preimpact timeline measured in tens of milliseconds, a 2 ms recovery tail can consume a meaningful fraction of the available decision window.

Resolution, LSB, and quantization

Resolution is the smallest change in input that produces a distinguishable change in output. For a digital sensor, this is set by the ADC bit depth and the FSR. The relationship is straightforward:

LSB = FSR / 2^N

Example: ±50 g accelerometer, 16-bit ADC
  Full span = 100 g
  LSB = 100 g / 65536 ≈ 1.53 mg per count

This is the quantization resolution — the theoretical floor. The effective resolution is almost always worse, because noise, offset drift, and nonlinearity blur the decision boundaries. A part that advertises 16 bits but delivers 12 effective bits has an effective LSB of roughly 24 mg, not 1.53 mg. When you zero a channel, the residual uncertainty in the zero point is bounded below by this effective resolution, not by the datasheet bit count.

Resolution also interacts with range in a way that is easy to overlook. Selecting a wider FSR to avoid clipping directly coarsens the LSB. A ±200 g part with the same 16-bit ADC has an LSB of about 6.1 mg — four times coarser than the ±50 g part. Range and resolution trade against each other, and the correct choice depends on whether the application is peak-limited or sensitivity-limited.

Signal-to-noise ratio and noise floor

Signal-to-noise ratio (SNR) expresses how far the signal of interest stands above the channel's own noise. For a sensor, noise is typically specified as a noise spectral density in µg/√Hz (acceleration) or Pa/√Hz (pressure), which must be integrated over the measurement bandwidth to obtain an RMS noise figure:

n_rms = noise_density × sqrt(BW × noise_bandwidth_factor)

Example: 100 µg/√Hz accelerometer, 400 Hz bandwidth, factor ≈ 1.22
  n_rms ≈ 100 × sqrt(400 × 1.22) ≈ 100 × 22.1 ≈ 2.21 mg RMS

With a 1 g signal, that yields an SNR of roughly 20·log10(1000 mg / 2.21 mg) ≈ 53 dB. Halving the bandwidth to 100 Hz reduces the RMS noise to about 1.1 mg and improves SNR by 6 dB — a useful lever when the application can tolerate a slower channel. This is why bandwidth selection is a dynamic-range decision, not merely a filtering preference.

SNR and resolution are related but not identical. Resolution describes the smallest representable step; SNR describes the smallest detectable signal above the noise. A channel can have fine quantization and poor SNR, in which case the low-order bits are pure noise and contribute nothing to measurement fidelity.

Putting the numbers to work: a range-check example

Consider a preimpact channel built around a ±50 g accelerometer with 16-bit output, 100 µg/√Hz noise density, and a 400 Hz anti-alias bandwidth. The nominal zero offset is specified as ±30 mg. A verification plan should establish four things: the zero point, the LSB, the noise floor, and the saturation boundary.

First, compute the LSB: 100 g / 65536 ≈ 1.53 mg. Second, compute the RMS noise: 100 µg/√Hz × √(400 × 1.22) ≈ 2.21 mg. Note that the noise floor is larger than one LSB — the low-order bit is not meaningful, and the effective resolution is closer to 2–3 mg. Third, the zero offset tolerance of ±30 mg is roughly 14 times the RMS noise, so a zeroing routine that averages over a few hundred samples can resolve the offset well below the tolerance. Fourth, the saturation boundary is at ±50 g; any test stimulus must stay below that, and the verification should confirm that a deliberately over-range stimulus produces the expected clipped signature rather than a plausible-looking but wrong value.

A practical check uses a known static orientation. With the sensor axis vertical, the expected input is +1 g; inverted, −1 g. The measured difference between the two orientations should be 2 g within the combined tolerance of sensitivity and offset. If the difference is short, suspect sensitivity error or partial saturation; if it is noisy across repeated trials, suspect the noise floor or a loose mechanical mount. A third orientation at 90° should read approximately zero, providing a direct zero-point check independent of the ±1 g measurements.

Pressure sensors: the same physics, different units

Pressure transducers follow the same framework. A 0–250 bar brake-pressure sensor with a 12-bit ADC has an LSB of about 61 mbar; its noise floor might be specified as 0.5 mbar/√Hz, which over a 1 kHz bandwidth integrates to roughly 16 mbar RMS. The ratio of FSR to noise floor defines the usable dynamic range in decibels: 20·log10(250 bar / 0.016 bar) ≈ 84 dB. Clipping at the top end during a pressure spike is common in brake systems, and the recovery behavior of the transducer diaphragm and its signal conditioning must be verified, because a slow recovery can corrupt the next pressure-hold phase.

For both acceleration and pressure, the verification logic is the same: establish the zero, quantify the LSB and the noise floor, confirm linearity across the usable span, and deliberately probe the saturation boundary to confirm that the channel fails in a recognizable way rather than silently reporting a plausible value. That last point is the heart of dynamic range verification — the dangerous failure is not the obvious clip, but the channel that has left its linear region while still producing a number that looks reasonable.