# Frequency analysis basics: FFT, PSD and the sampling theorem

> Suspension behaviour is easier to judge in the frequency domain: an FFT or PSD of damper, ride-height and acceleration signals reveals body and wheel modes, and transfer functions quantify platform control and tyre-load fluctuation.

Area: Data Acquisition & Analysis · Starting level: L2 Practitioner · Updated 2026-11-16
Source page: https://race-engineer-academy.vercel.app/lesson/daq-frequency-analysis

## L1 · Foundation

**Any signal is a sum of sine waves.** A Fourier transform breaks a time signal into the amplitude (and phase) of each frequency it contains. For suspension work this is powerful because the road input is essentially random noise – looking at a single bump in the time domain tells you about *that* bump, while the spectrum characterises the suspension against *any* input.

What you typically see in a spectrum of damper or vertical acceleration data:

| Band | What it is | Typical race-car values |
|---|---|---|
| < 1 Hz | slow aero load and fuel/pitch changes, driver inputs | – |
| 1.5–6 Hz | body (sprung mass) heave, pitch, roll modes | GT3 ~2.5–3.5 Hz; LMP2/formula ~3.5–6 Hz |
| 10–25 Hz | wheel hop (unsprung mass on tyre stiffness) | ~12–20 Hz |
| > 25 Hz | kerbs, structural modes, tyre cavity, engine | – |

**Sampling theorem.** To capture a frequency you must sample at more than twice that frequency (Nyquist). Wheel hop at 18 Hz needs > 36 Hz – in practice 200 Hz or more for a clean spectrum; damper velocity work 500 Hz+. Content above half the sampling rate doesn't disappear – it folds back as a false low frequency (aliasing) at
$$f_{alias} = |f - k\,f_s|, \quad k = \text{round}(f/f_s)$$
Example: 185 Hz engine or structural vibration reaching a damper pot logged at 200 Hz appears at 15 Hz – right in the wheel-hop band, indistinguishable from real wheel hop. **Aliased content can never be removed afterwards by digital filtering**; it must be stopped *before* the ADC by an analogue anti-alias low-pass filter (cut-off typically $f_s/4$ to $f_s/2.5$). Know which of your logger inputs have one.

## L2 · Practitioner

**Practical FFT/PSD workflow.**

1. **Pick a segment** with stationary conditions – e.g., a straight or a family of medium-speed corners. Aero load changes stiffness and ride height with speed, so mixing 80 and 280 km/h data blurs modes. Segment by speed band.
2. **Remove mean/trend** (static ride height, slow aero compression).
3. **Window** each block (Hann is the usual choice) to reduce leakage from the block ends.
4. **Average** many overlapping blocks (Welch's method) – a single FFT of random data has huge variance.
5. Plot the **power spectral density** (PSD, units²/Hz) on log axes.

**Frequency resolution** is set by the block length:

$$\Delta f = \frac{f_s}{N} = \frac{1}{T_{block}}$$

At 500 Hz with 2048-point blocks, $T = 4.1$ s and $\Delta f \approx 0.24$ Hz – fine enough to separate a 3 Hz body mode from a 4 Hz pitch mode.

**Reading it.** Peaks at body modes indicate where the platform moves most; a sharp, tall peak suggests low damping, a broad low hump higher damping. Comparing damper settings: a well-damped setup lowers the body-mode peak without raising the 5–10 Hz range too much (where over-stiff damping transmits road input to the body and loads to the tyre).

**Natural frequency link.** A single-degree-of-freedom estimate is $f_n = \frac{1}{2\pi}\sqrt{k/m}$, with $k$ the **ride rate** $K_R = K_w K_t/(K_w+K_t)$ (wheel rate in series with tyre rate) and $m$ the sprung corner mass. Do not expect the PSD peak to sit exactly on $f_n$: on a well-damped race car ($\zeta \approx 0.4$–0.8) the peak location depends on the damping ratio, on which signal you analyse (body acceleration, damper displacement or velocity) and on the slope of the road-input spectrum, so 10–20 % differences are normal – and at $\zeta \gtrsim 0.7$ the body-mode peak almost disappears. Only larger discrepancies, or a peak that moves with speed, point to aero load, bump rubbers/packers engaged, tyre stiffness or a wrong motion ratio in your calculation. With light damping you can estimate $\zeta$ from the half-power bandwidth, $\zeta \approx (f_2 - f_1)/(2 f_{peak})$.

**Four-post rig vs track.** Rigs give clean swept-sine inputs and measure contact-patch load directly (not possible on a rolling tyre) – ideal for damper comparisons, but no aero, no rolling tyre and no balance information. On track you get reality but noisy, non-stationary data.

## L3 · Advanced

**Transfer functions (FRFs).** Instead of single spectra, characterise the suspension by output/input ratios:

$$H_{xy}(f) = \frac{S_{xy}(f)}{S_{xx}(f)}$$

where $S_{xy}$ is the cross-spectrum and $S_{xx}$ the input auto-spectrum. Check the **coherence** $\gamma^2 = |S_{xy}|^2/(S_{xx}S_{yy})$; below ~0.8 the FRF at that frequency is unreliable.

Two FRFs matter (Segers ch. 12):
- **Road → body motion** – platform stability (aero needs a stable ride height).
- **Road → tyre deflection / contact-patch load** – grip. Load fluctuation costs grip because tyre force is non-linear (concave) in vertical load: the average force of a fluctuating load is lower than the force at the average load.

The damping that minimises the body-motion peak and the damping that minimises tyre-load fluctuation are **different** – setup is a compromise chosen per track (bumpy street circuit favours grip; smooth high-downforce track favours platform).

**Track "virtual rig".** With wheel position, wheel load (pushrod) and body vertical acceleration you can estimate hub acceleration $a_{hub} = G_{vert} + \ddot{x}_{susp}$ and contact-patch force $F_{cp} = F_{susp} + m_{wheel}\,a_{hub}$ (plus static), then compute FRFs from on-track data. Conditions: $G_{vert}$ must be measured at that corner (or transformed from the CG: $\ddot z_{corner} = \ddot z_{CG} \pm \ddot\theta\,a \pm \ddot\phi\,t/2$) in m/s² with gravity removed; $\ddot{x}_{susp}$ is the second derivative of *wheel* travel (pot travel / MR), bump positive; $F_{susp}$ is the suspension force at the wheel (pushrod force × installation ratio). Limitations: double-differentiating pot data amplifies noise at high frequency, so resolution and sampling rate of the pots set the useful upper frequency.

**Pitfalls:** non-stationarity (aero), non-linearity (bump stops, digressive dampers – the "natural frequency" shifts with amplitude), and leakage from short blocks. Segment by speed, check coherence, and compare like with like.

## L4 · Expert

**Beyond quarter-car thinking.** A real car couples heave, pitch, roll and warp; the modal shapes depend on front/rear stiffness and mass distribution, and on aero stiffness (downforce sensitivity to ride height acts like an extra, sometimes negative, spring). On ground-effect cars, aero stiffness (negative near the floor, where downforce rises as ride height falls) lowers the effective heave/pitch frequency; if it exceeded the mechanical stiffness the result would be static divergence onto the skids, not an oscillation. Porpoising additionally needs energy input: floor stall below a critical ride height, with a lag before the flow re-attaches, acts as negative aero damping and drives a limit cycle that a mechanical-only – or even a quasi-static aero-map – model will not predict.

**Modal identification.** With four corner accelerometers and four pots you can decompose motion into modal coordinates (heave = mean, pitch = front−rear, roll = left−right, warp = diagonal) and compute FRFs per mode. Heave/pitch platform control is usually the aero priority; roll and warp affect mechanical balance and tyre loading.

**Optimisation metrics.** Common objective functions combine RMS contact-patch load variation (grip), RMS of front/rear ride height (aero), and peak pitch/heave transmissibility, weighted per track from a road-profile PSD measured on that track. Teams then sweep damper curves in a validated 7-post/sim model and verify on track with the PSD of ride height and the FRF of tyre load. Always remember the driver: an analytically optimal low-damping setup may give poor transient response and confidence.

## Key formulas

**Frequency resolution**

$$\Delta f = \frac{f_s}{N} = \frac{1}{T_{block}}$$

- f_s: sampling rate (Hz)
- N: samples per FFT block
- T_{block}: block duration (s)

**Nyquist frequency**

$$f_{Nyq} = \frac{f_s}{2}$$


**Alias frequency**

$$f_{alias} = \left|f - \text{round}\!\left(\frac{f}{f_s}\right) f_s\right|$$

- f: true signal frequency (Hz)
- f_s: sampling rate (Hz)
- f_{alias}: frequency at which it appears in the logged data (Hz)

**Natural frequency (single DOF)**

$$f_n = \frac{1}{2\pi}\sqrt{\frac{k}{m}}$$

- k: effective stiffness (N/m)
- m: mass (kg)

**Transfer function estimate**

$$H_{xy}(f) = \frac{S_{xy}(f)}{S_{xx}(f)}, \quad \gamma^2 = \frac{|S_{xy}|^2}{S_{xx}S_{yy}}$$

- S_{xy}: cross-spectral density
- S_{xx}, S_{yy}: auto-spectral densities
- \gamma^2: coherence (0–1)

**Contact patch force from on-car sensors**

$$F_{cp} = F_{susp} + m_{wheel}\,a_{hub}$$

- F_{susp}: suspension force at the wheel (N)
- m_{wheel}: unsprung mass (kg)
- a_{hub}: hub vertical acceleration (m/s²)

## Trackside takeaways

- Body modes sit around 2–6 Hz, wheel hop around 12–20 Hz; log damper pots fast enough to see both.
- Aliasing cannot be filtered out after logging – it needs an analogue anti-alias filter before the ADC.
- Use windowing and averaged PSDs (Welch); single FFTs of random data are too noisy.
- Segment by speed – aero changes stiffness, so mixed-speed spectra blur modes.
- Platform (body FRF) and grip (tyre-load FRF) are optimised by different damping – choose the compromise per track.
- Check coherence before trusting a transfer function.

## By car category

### GT3

Body modes typically ~2.5–3.5 Hz; bump-stop engagement at high speed shifts the effective frequency – analyse high-speed segments separately.

### LMP2

Stiff platform and floor-dominated aero: ride-height PSD at high speed is the key metric for damper and heave-element choices.

### Hypercar (LMH/LMDh)

Teams run 7-post rigs and full modal analysis; on-track FRFs from load cells validate the rig model.

### FIA F2

High-downforce single-seater with stiff springs – modes in the 4–6 Hz range; kerb strikes dominate high-frequency content, so gate them out.

### Super Formula

Very high aero loads and ride-height sensitivity – heave/pitch platform control is paramount; spectral analysis of front ride height at speed guides heave damper settings.

### TCR

Softer road-car-derived suspension with body modes nearer 2 Hz; wheel hop under braking on bumpy tracks is visible as 12–15 Hz content.

## References

- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): Ch. 12 Suspension Analysis in the Frequency Domain (§12.1–12.4, §12.8)
- Segers, Analysis Techniques for Racecar Data Acquisition, 2nd ed. (SAE 2014): §19.2 Analog-Digital Conversion (Nyquist and aliasing)
- Milliken & Milliken, Race Car Vehicle Dynamics (SAE 1995): Ch. 16 Ride and Roll Rates; Ch. 22 Dampers
