How Frequency Resolution Actually Works Beyond Δf
Most people learn about frequency resolution as follows.
“Δf = Sampling rate / N”
However, this is only part of the story.
True frequency resolution depends on several factors in addition to Δf.
Basic formula (Δf)
meaning
- F s → Sampling frequency
- N → Number of samples
intuition
Spacing between FFT bins
Why Δf is not everything
Even if Δf is small, it may not be possible to separate two frequencies that are close to each other.
example
- The signal includes 50Hz and 52Hz.
- Δf = 1 Hz
It might still look like a single spectral peak!
The reason? → Spectral leakage and window effect
Effective frequency resolution = ability to separate peaks
Actual resolution is "Can you distinguish between two nearby frequencies?"
Key practical insights
- Resolution ≠ Spacing
- Resolution = Peak separation
The role of signal length (time duration)
The actual frequency resolution is improved as follows.
Longer observation time
relationship
where,
- T = Total signal duration (measurement time)
intuition
The longer the listening time, the better the frequency clarity.
Windowing effect (very important!)
Adjust spectrum peak width and leakage using the window.
Key Idea
- Rectangular window → Sharp but noisy
- Hann/Hamming Window → Wider but cleaner
compromise
- Narrow peak → Excellent resolution
- Low leakage → Clear spectrum
Spectrum leakage (reason why peaks spread)
If the signal is not perfectly periodic, the energy is spread across the entire frequency band.
result
- The spectrum peaks become wider.
- When multiple frequencies exist, it becomes more difficult to distinguish between them.
When truncation matches an integer number of periods (Coherent Sampling),
DTFT displays a Sinc-shaped spectrum due to the window effect.
On the other hand, FFT concentrates energy only on the discrete frequency band corresponding to the main lobe and prevents leakage.
When truncation does not match the number of integer periods (non-coherent sampling),
Due to the windowing effect , the DTFT maintains a Sinc-shaped spectrum.
On the other hand, FFT disperses energy beyond the main lobe into adjacent frequency bands, causing spectral leakage.
Misconceptions about Zero Padding
Important misunderstanding
- Zero padding does not increase resolution.
What is the actual work?
- Add more points between the sections.
- Makes the graph denser.
- Zero padding increases the sampling density of the DTFT, allowing the FFT to appear closer to a continuous spectrum.
Key Insights
It is an interpolation method, not a resolution improvement.
When comparing the FFT with zero padding applied and the FFT without it, the graph density is higher
What actually improves resolution?
Increase the signal duration
- The more data there is, the better the frequency separation. I will explain this using the MALMIJAL example below.
Select the appropriate window
- Balance of leakage and clarity
Higher SNR (Signal-to-Noise Ratio)
- Noise reduction → Clearer peaks
Average (PSD)
MALMIJAL Workflow
Cannot distinguish adjacent frequencies
- Fs = 1000Hz, T = 0.1 sec , 50Hz sine wave and 52Hz sine wave
- FFT iterations = 100, Δf = Fs/NFFT = 10Hz
- Apply FFT
- 50Hz and 52Hz cannot be distinguished due to spectrum leakage and blurring
Unable to distinguish between 50Hz and 52Hz due to insufficient frequency resolution
Frequency resolution improvement
- Fs = 1000Hz, T = 1 sec , 50Hz sine wave and 52Hz sine wave
- FFT iterations = 1000, Δf = Fs/NFFT = 1Hz
- Apply FFT
- Improves frequency resolution to distinguish between 50Hz and 52Hz
Distinguishing between 50Hz and 52Hz through improved frequency resolution (long recording length)
Cases where adjacent frequencies cannot be distinguished by windowing
- Fs = 1000Hz, T = 1 sec , 50Hz sine wave and 52Hz sine wave
- FFT iterations = 1000, Δf = Fs/NFFT = 1Hz
- The power spectrum is applied using a Blackman window , where the width of the main lobe is approximately 6Δf.
- Even if the same frequency resolution as before is used, the two frequencies cannot be distinguished due to the window effect.
Unable to distinguish between 50Hz and 52Hz due to the window effect (blurring phenomenon)
Key Summary
- Δf is the bin interval, not the actual resolution.
- Effective resolution is the ability to separate frequencies, and it is controlled by time length, window, and leakage.
- Zero padding does not mean actual resolution improvement.
conclusion
Frequency resolution is often mistakenly understood as Δf = F s / N, but actual resolution means more than just the frequency bin interval.
- Δf represents only the FFT bin interval and does not represent the ability to actually distinguish adjacent frequencies.
- True frequency resolution depends on the ability to separate adjacent spectral peaks , which is affected by signal duration, window, and spectral leakage.
- As the observation time increases, the resolution improves, and the window selection and noise level affect how clearly peaks can be identified.
- Zero padding only interpolates the spectrum and does not improve actual resolution.
In summary,
actual frequency resolution is determined not only by Δf but also by peak separation, and signal length, windowing, and noise conditions must be carefully controlled.
Additional references
##The following topics may also be helpful.
How Frequency Resolution Actually Works Beyond Δf
Most people learn about frequency resolution as follows.
“Δf = Sampling rate / N”
However, this is only part of the story.
True frequency resolution depends on several factors in addition to Δf.
Basic formula (Δf)
meaning
intuition
Spacing between FFT bins
Why Δf is not everything
Even if Δf is small, it may not be possible to separate two frequencies that are close to each other.
example
It might still look like a single spectral peak!
The reason? → Spectral leakage and window effect
Effective frequency resolution = ability to separate peaks
Actual resolution is "Can you distinguish between two nearby frequencies?"
Key practical insights
The role of signal length (time duration)
The actual frequency resolution is improved as follows.
Longer observation time
relationship
where,
intuition
The longer the listening time, the better the frequency clarity.
Windowing effect (very important!)
Adjust spectrum peak width and leakage using the window.
Key Idea
compromise
Spectrum leakage (reason why peaks spread)
If the signal is not perfectly periodic, the energy is spread across the entire frequency band.
result
When truncation matches an integer number of periods (Coherent Sampling),
DTFT displays a Sinc-shaped spectrum due to the window effect.
On the other hand, FFT concentrates energy only on the discrete frequency band corresponding to the main lobe and prevents leakage.
When truncation does not match the number of integer periods (non-coherent sampling),
Due to the windowing effect , the DTFT maintains a Sinc-shaped spectrum.
On the other hand, FFT disperses energy beyond the main lobe into adjacent frequency bands, causing spectral leakage.
Misconceptions about Zero Padding
Important misunderstanding
What is the actual work?
Key Insights
It is an interpolation method, not a resolution improvement.
When comparing the FFT with zero padding applied and the FFT without it, the graph density is higher
What actually improves resolution?
Increase the signal duration
Select the appropriate window
Higher SNR (Signal-to-Noise Ratio)
Average (PSD)
MALMIJAL Workflow
Cannot distinguish adjacent frequencies
Unable to distinguish between 50Hz and 52Hz due to insufficient frequency resolution
Frequency resolution improvement
Distinguishing between 50Hz and 52Hz through improved frequency resolution (long recording length)
Cases where adjacent frequencies cannot be distinguished by windowing
Unable to distinguish between 50Hz and 52Hz due to the window effect (blurring phenomenon)
Key Summary
conclusion
Frequency resolution is often mistakenly understood as Δf = F s / N, but actual resolution means more than just the frequency bin interval.
In summary,
actual frequency resolution is determined not only by Δf but also by peak separation, and signal length, windowing, and noise conditions must be carefully controlled.
Additional references
##The following topics may also be helpful.