ZoomFFT Frequency Refinement

The problem it solves

The resolution of a normal FFT spectrum is the sampling rate divided by the number of points. At 5120 Hz and 4096 points it is only 1.25 Hz. When a machine has two components 1 Hz or less apart — for example a gear mesh frequency and its sidebands, or motor running speed and supply interference — the normal spectrum shows one broad peak, and neither frequency nor amplitude can be read correctly.

The traditional solution is to acquire many more points. That means longer acquisition, more memory and slower computation, and it does not help with data that has already been recorded. ZoomFFT uses complex modulation, anti-alias low-pass filtering and decimation to raise the resolution of the selected band without extra hardware or resampling — a digital magnifying glass for the spectrum.

Functions

Screen layout

1. Parameter table: per-channel peak frequency, peak amplitude, resolution and zoom factor before and after zoom.
2. Full-band FFT spectrum, with two vertical lines (LowCut and HighCut) marking the band being zoomed.
3. ZoomFFT spectrum: the high-resolution spectrum of the selected band.
4. Time waveform of the current frame, to confirm amplitude and absence of clipping.

The left panel contains channel selection, zoom factor, center frequency, window type, averaging, the test mode switch and the refresh, export and help buttons.

Gearbox diagnosis

Sidebands spaced at the shaft speed appear on both sides of the gear mesh frequency. On a low-speed shaft of a few hundred rpm, the spacing can be below 1 Hz and a normal FFT cannot resolve it. With zoom, the spacing, number and relative amplitude of the sidebands are clear, showing which shaft is affected and whether the fault is modulation-type or a local defect.

Early rolling bearing faults

Bearing characteristic frequencies and their harmonics are often weak, close together and buried in the background. Zooming into the band of interest helps reveal early faults and gives time for predictive maintenance.

Dense lines in motors and rotating machines

Running speed, supply frequency (50 Hz and harmonics), blade pass and fan frequencies are often close together. Acceptance and quality tests need the true amplitude of each, so that interference is not mistaken for a fault or the reverse. The zoom spectrum reads frequency to the order of 0.01 Hz.

Structural modes and resonance checks

When two natural frequencies are very close, one peak in a normal spectrum contains both. Zoom analysis separates them for accurate modal identification and resonance avoidance.

Dominant frequency in vibration acceptance

Zone evaluation to vibration standards such as ISO 10816 / GB/T 6075 often requires confirming whether the dominant vibration comes from running speed, electromagnetic excitation or structural resonance, because each needs a different correction. The zoom spectrum provides reliable frequency attribution.

Training and demonstration

Test mode generates two tones 1 Hz apart (for example 400 Hz and 401 Hz). The normal spectrum shows one raised peak; the zoom spectrum separates two clear peaks. It demonstrates frequency resolution without external equipment.

Why a normal FFT has limited resolution

Δf = Fs / N

Fs is the sampling rate and N the number of points. With Fs fixed, a smaller Δf needs a larger N and therefore a longer acquisition. When the signal is stationary and only a narrow band matters, a very long FFT over the full bandwidth is wasteful — most lines are not of interest.

Complex-modulation ZoomFFT

ZoomFFT analyzes only the narrow band of interest at high resolution and discards the rest.

Δf′ = Fs / (Mult × N) = Δf / Mult

The resolution becomes 1/Mult of the original. For example, 1.25 Hz becomes 0.156 Hz at 8× and 0.039 Hz at 32×.

Engineering details

• High-order FIR anti-alias filter: decimation lowers the Nyquist frequency, and any residual out-of-band signal folds back into the band as a false image peak. A linear-phase FIR low-pass filter designed by the window method is used; its order increases with the zoom factor (order = 32 × zoom factor) and image suppression reaches the order of −120 dB. Filtering uses fast FFT convolution in a dedicated DSP numerical engine, so a higher order adds almost no computing time.
• Zero-phase filtering: the signal is filtered forward, reversed, filtered again and reversed back. Amplitude and phase have no systematic distortion, so zoom amplitudes are reliable.
• Continuous data buffer: zoom analysis needs a long record spanning several frames. A rolling buffer (length = frame points × zoom factor) is updated with each new frame, so the zoom spectrum refreshes continuously and frequency readings stay stable.
• Window and correction: a window (Hanning by default) is applied in the zoom chain and the true amplitude is restored with the window correction factor, using the same scaling as the full-band spectrum so the two can be compared directly.
• Exponential averaging: new spectrum × 1/k + previous average × (1 − 1/k), where k is the set number of averages; the first frame is used directly. Averaging affects only the magnitude display and can be reset at any time.

Performance

Frequency-domain computation and filter convolution run in a built-in compiled numerical library. The filter frequency response is calculated only when parameters change and is reused for all channels and frames. With 8× zoom on several channels, real-time mode refreshes smoothly; for factors above 20, offline replay mode is recommended.

• Zoom recorded data without resampling, so no new test is needed for high-resolution analysis.
• Resolution raised by dozens of times: peaks 1 Hz apart or closer are separated, and sideband structure and modulation become visible.
• Eight channels zoomed simultaneously, with results compared in one table.
• Zero-phase processing: filtering does not distort the waveform, and amplitudes before and after zoom use the same scaling.
• High-order anti-alias design: image suppression to the order of −120 dB prevents misleading false peaks.
• Real-time operation at common zoom factors, accelerated by a compiled math library.
• The zoom band is marked on the full-band spectrum, with peaks of both spectra in one table.
• Exponential averaging and reset for stable spectra in noisy conditions.
• Built-in test signals demonstrate 1 Hz peak separation without sensors.
• Excel export of all zoom data and the parameter table.

Preparation

Confirm that sampling rate, analysis points and engineering units are set in the main acquisition window and that acquisition or replay is running. For real signals, confirm that sensors are connected and channels enabled. For a demonstration, enable test mode.

Basic procedure

1. Select channels. Disabled channels are grayed out; the analysis refreshes automatically.
2. Set the zoom factor (for example 2, 4, 8, 16). The original resolution, the zoomed resolution and the equivalent total points are shown. Use up to 20× in real-time acquisition; higher factors are for offline replay.
3. Set the center frequency: type the value and press Enter, drag the slider, or check the LowCut and HighCut lines in the full-band spectrum. If the band exceeds the analysis range, the field turns red and the band is limited to the valid range.
4. Read the results: the full-band spectrum with band markers, the zoom spectrum with high-resolution peak frequencies, and the table comparing peaks before and after zoom.
5. Optional: enable exponential averaging and set the count (default 10); restart averaging to clear the history.
6. Optional: keep the window on top.
7. Double-click a plot to enlarge it; double-click again to restore the layout.
8. Export all data: an Excel file with sampling parameters, peak table and all zoom spectra opens automatically.
9. Open the built-in PDF reference if needed.

Recommended settings

In test mode, the built-in signal contains two equal tones at 400 Hz and 401 Hz.

• Full-band FFT spectrum: one peak near 400 Hz; it cannot be seen whether it is one peak or two.
• Zoom spectrum (8×, center 400 Hz): at 0.156 Hz resolution, the peaks at 400.0 Hz and 401.0 Hz are fully separated.
• Parameter table: before zoom, the peak reading falls on a line between the two; after zoom, the peak frequencies read 400.0 Hz and 401.0 Hz.