Measurement methods · Signal analysis
Frequency spectra and FFT resolution
Does the tallest spectrum peak always identify the mode?
A spectrum separates a recording into frequency components. Understanding sampling, record length and leakage is essential before reading a resonant frequency from a peak.
Try the interactive demoStart with something familiar
Hearing two people speak at once gives one mixed sound at your ear. A frequency spectrum asks which repeating components are present in a mixed recording; it does not automatically identify which object produced each component.
Time: components mixed
One synthetic signal, two frequencies
Frequency: components separated
Schematic peaks, not a measured spectrum
Notice what changes
The time trace shows how pressure varies as time passes. The spectrum reorganizes the same samples by frequency: horizontal position means oscillations per second, not time.
Understand why
Listening for longer can help distinguish nearby frequencies. A shorter record has broader features and more widely spaced FFT bins. Drawing more points by zero-padding does not add missing observations.
Connect it to wood and making
A wood tap contains several modes and room sound. A strong peak is a candidate, not proof of the intended bending mode. Move the tap, repeat and compare the spatial pattern.
Put names to the ideas
Bin spacing and duration
For N samples at sampling rate f_s, duration T = N/f_s and FFT bin spacing is f_s/N = 1/T. Longer useful recordings provide finer spacing, but a decaying signal eventually reaches noise.
Leakage and windows
A finite record can spread a component over nearby bins. A window reduces some leakage while changing peak width and amplitude. State the window when comparing spectra or using a half-power width.
A strong peak is not an identity
Tap and microphone positions weight the modes differently. Room sound, mains interference or an unwanted mode may produce the strongest peak. Use repeat taps and modal evidence.
T is the useful observation duration. The spacing 1/T means 0.5 seconds gives 2 Hz bins. Peak width also depends on windowing and damping, so bin spacing alone is not a guarantee that two peaks are resolved.
A worked example
A 0.5 s useful record has 2 Hz bin spacing. Zero-padding can make the displayed curve smoother, but cannot create the resolving power of an actually longer recording.
Explore it yourself
Record unclipped taps, avoid automatic voice processing where possible, compare repeated spectra, and save the chosen peak with the original recording.
- Observation duration
- 0.125 s
- Bin spacing
- 8.0 Hz
- Known frequencies
- 200 / 206 Hz