Understanding Windowing in FFT Analysis
Windowing is a signal-processing step in which a mathematical weighting function — a “window” — is applied to a block of time-waveform data before it is handed to the Fast Fourier Transform. The window’s shape smoothly tapers the amplitude of the captured signal down to zero at the start and end of the time block, so the data stitches together without abrupt jumps. This single operation is what suppresses a pervasive error called spectral leakage and is therefore essential to producing an accurate vibration spectrum. In practical vibration analysis, choosing and applying a window correctly is the difference between a clean, trustworthy spectrum and a smeared, misleading one.
1. Definition: What Is a Windowing Function?
A windowing function is a profile — a set of multiplying factors, one per sample — that is laid over the raw time block. Where the window value is 1.0 the sample passes through unchanged; where it falls toward 0.0 the sample is attenuated. Because nearly every window peaks in the middle and tapers at both ends, multiplying the time record by the window forces the captured snippet to begin and end at zero amplitude. The mathematics of the FFT are unchanged; windowing simply pre-conditions the data so the transform’s built-in assumptions are satisfied. Without it, the spectrum the analyser returns can be quantitatively wrong even when the sensor and the rest of the measurement chain are perfect.
2. The Problem: Spectral Leakage
The FFT carries an inherent assumption: it treats the finite block of time data it analyses as one complete cycle of a perfectly periodic signal that repeats forever. Real machinery signals almost never oblige. When acquisition starts and stops at arbitrary instants, the end of the captured block does not line up with its beginning, so when the FFT mentally wraps the block back on itself it sees sharp, artificial discontinuities at the boundaries.
The transform interprets those abrupt jumps as genuine high-frequency content that does not exist in the machine. Energy that truly belongs to a single, discrete frequency peak is smeared — it “leaks” — into the neighbouring frequency bins on either side. The consequences are three-fold:
- Reduced amplitude accuracy: the measured height of the peak reads lower than its true value because its energy has been spread across many bins instead of concentrated in one.
- Broadened peaks: the line appears wider and less sharply defined than the underlying physics warrants, blurring the frequency estimate.
- Loss of resolution: the spilled energy lifts the noise floor around a large peak, burying smaller adjacent peaks — exactly the small harmonics and sidebands that often carry the diagnostic story.
3. The Solution: Applying a Window
Windowing cures leakage by smoothly compelling the signal to look periodic within the block. Multiplying the raw waveform by the window tapers the amplitudes at the extreme start and end to zero, which removes the boundary discontinuities and, in effect, tricks the FFT into seeing a continuous, gap-free signal. The pay-off is a markedly cleaner spectrum:
- Significantly improved amplitude accuracy, so peak heights can be trusted against vibration-severity limits.
- Sharper, better-defined frequency peaks that pin a fault to a specific order or component.
- A lower effective noise floor, letting small signals stand out beside large ones.
There is, inevitably, a trade-off. Tapering the ends discards some of the record’s energy and widens the main spectral lobe slightly, so windowing trades a little frequency resolution for a large reduction in leakage. Every window is a different point on that compromise, which is why several shapes exist.
4. Common Types of Windows
Dozens of windowing functions have been devised, each weighting the time block a little differently. For general-purpose machinery work, one dominates.
Hanning Window
The Hanning window (a raised-cosine taper) offers an excellent compromise between frequency resolution and amplitude accuracy, and it is the recommended default for virtually all standard rotating-machinery vibration measurements. Unless there is a specific reason to do otherwise, the Hanning window should always be selected. It is the right choice for the continuous, broadly periodic signals that dominate condition monitoring.
Other Windows
- Rectangular window (also called Uniform, or “None”): equivalent to applying no window at all. It has the best frequency resolution but the worst spectral leakage, and is suitable only when the signal is known to be perfectly periodic within the block — or when capturing very sharp, fully-contained transient events such as an impact.
- Flattop window: delivers the most accurate amplitude measurement of any common window, at the cost of very poor frequency resolution (very wide peaks). It is the window of choice for calibration work and any task where the exact amplitude of a peak matters more than its exact frequency — for example, verifying a sensor against a calibration certificate on a known reference shaker.
- Hamming window: closely related to the Hanning window, with minor trade-offs in sidelobe behaviour; rarely needed in routine machinery diagnostics.
5. When to Use a Window — and How It Interacts With Resolution
For machinery condition monitoring the rule is simple: always use a Hanning window for general spectral analysis. Disabling the window — selecting Rectangular on an ordinary running signal — yields inaccurate and potentially misleading data, because leakage will distort both the peak heights and the apparent noise floor. Modern instruments apply the Hanning window by default precisely because it is essential to a reliable, accurate spectrum.
Windowing does not act alone. Because tapering broadens each spectral line, the practical frequency resolution you achieve is the combined result of the window choice and the analysis parameters — block length (number of samples), sampling rate and span. When peaks sit very close together, lengthening the time record sharpens them faster than changing the window will; you can preview that trade-off with an FFT resolution calculator before committing to a measurement setup. Windowing is also distinct from, and complementary to, signal filtering: a filter removes unwanted frequency bands from the signal, whereas a window conditions whatever band remains so the FFT can represent it faithfully.
6. Windowing in the Field
In hands-on diagnostics the window is rarely something the analyst thinks about consciously — and that is by design. When an engineer captures a spectrum or runs a balancing job with a portable two-channel instrument such as the Балансет-1А, the software applies a Hanning window automatically before computing the FFT, so the 1× running-speed peak and its harmonics appear at their true amplitude and the right frequency without any extra steps. That correctly windowed spectrum is what lets the same instrument separate a genuine unbalance peak from nearby noise and verify the result after correction. Understanding what the window is doing under the hood helps an analyst recognise when a non-default choice — Flattop for a calibration check, Rectangular for a clean transient — is genuinely warranted.
