I am running some experiments on MATLAB, and I have noticed that, keeping the period fixed, increasing the sampling rate of a sine signal causes the different shifted waveforms in the Fourier transform to become more distinct. They get further apart, I think this makes sense because as the sampling rate increases, the difference between the Nyquist rate and the sampling rate increases too, which creates an effect opposed to aliasing. I have also noticed that the amplitude of the peaks of the transform also increase as the sampling rate increases. Even the DC component (frequency = 0) changes. It's shown as being 0 at some sampling rate, but when increasing the sampling rate it's not 0 anymore.
All the sampling rates are above the Nyquist rate. It seems odd to me that the Fourier transform changes its shape, since according to the sampling theorem, the original signal can be recovered if the sampling rate is above the Nyquist rate, no matter if it's 2 times the nyquist rate or 20 times. Wouldn't a different Fourier waveform mean a different recovered signal?
I am wondering, formally, what's the impact of the sampling rate
Thank you.
You're conflating conversion between time-discrete and time-continuous forms of a signal with reversibility of a transform.
The only guarantee is: For a given transform of some discrete signal, its inverse transform will yield the "same" discrete signal back. The discrete signal is abstract from any frequencies. All that the transform does is take some vector of complex values, and give the dimensionally matching vector of complex values back. You can then take this vector, run an inverse transform on it, and get the "original" vector. I use quotes since there may be some numerical errors that depend on the implementation. As you can see, nowhere does the word frequency appear because it's irrelevant.
So, your real question is then, how to get an FFT with values that are useful for something besides getting the original discrete signal back through an inverse transform. Say, how to get an FFT that will tell a human something nice about the frequency content of a signal. A transform "tweaked" for human usefulness, or for use in further signal processing such as automated music transcription, can't reproduce the original signal anymore after inversion. We're trading off veracity for usefulness. Detailed discussion of this can't really fit into one answer, and is off topic here anyway.
Another of your real questions is how to go between a continuous signal and a discrete signal - how to sample the continuous signal, and how to reconstruct it from its discrete representation. The reconstruction means a function (or process) that will yield the values the signal had at points in time between the samples. Again, this is a big topic.
You are seeing several things when you increase the sample rate:
most (forward) FFT implementations have an implicit scaling factor of N (sometimes sqrt(N)) - if you're increasing your FFT size as you increase the sample rate (i.e. keeping the time window constant) then the apparent magnitude of the peaks in the FFT will increase. When calculating absolute magnitude values you would normally need to take this scaling factor into account.
I'm guessing that you are not currently applying a window function prior to the FFT - this will result in "smearing" of the spectrum, due to spectral leakage, and the exact nature of this will be very dependent on the relationship between sample rate and the frequencies of the various components in your signal. Apply a window function and the spectrum should look a lot more consistent as you vary the sample rate.
Related
I'm currently working on a program in C++ in which I am computing the time varying FFT of a wav file. I have a question regarding plotting the results of an FFT.
Say for example I have a 70 Hz signal that is produced by some instrument with certain harmonics. Even though I say this signal is 70 Hz, it's a real signal and I assume will have some randomness in which that 70Hz signal varies. Say I sample it for 1 second at a sample rate of 20kHz. I realize the sample period probably doesn't need to be 1 second, but bear with me.
Because I now have 20000 samples, when I compute the FFT. I will have 20000 or (19999) frequency bins. Let's also assume that my sample rate in conjunction some windowing techniques minimize spectral leakage.
My question then: Will the FFT still produce a relatively ideal impulse at 70Hz? Or will there 'appear to be' spectral leakage which is caused by the randomness the original signal? In otherwords, what does the FFT look like of a sinusoid whose frequency is a random variable?
Some of the more common modulation schemes will add sidebands that carry the information in the modulation. Depending on the amount and type of modulation with respect to the length of the FFT, the sidebands can either appear separate from the FFT peak, or just "fatten" a single peak.
Your spectrum will appear broadened and this happens in the real world. Look e.g for the Voight profile, which is a Lorentizan (the result of an ideal exponential decay) convolved with a Gaussian of a certain width, the width being determined by stochastic fluctuations, e.g. Doppler effect on molecules in a gas that is being probed by a narrow-band laser.
You will not get an 'ideal' frequency peak either way. The limit for the resolution of the FFT is one frequency bin, (frequency resolution being given by the inverse of the time vector length), but even that (as #xvan pointed out) is in general broadened by the window function. If your window is nonexistent, i.e. it is in fact a square window of the length of the time vector, then you'll get spectral peaks that are convolved with a sinc function, and thus broadened.
The best way to visualize this is to make a long vector and plot a spectrogram (often shown for audio signals) with enough resolution so you can see the individual variation. The FFT of the overall signal is then the projection of the moving peaks onto the vertical axis of the spectrogram. The FFT of a given time vector does not have any time resolution, but sums up all frequencies that happen during the time you FFT. So the spectrogram (often people simply use the STFT, short time fourier transform) has at any given time the 'full' resolution, i.e. narrow lineshape that you expect. The FFT of the full time vector shows the algebraic sum of all your lineshapes and therefore appears broadened.
To sum it up there are two separate effects:
a) broadening from the window function (as the commenters 1 and 2 pointed out)
b) broadening from the effect of frequency fluctuation that you are trying to simulate and that happens in real life (e.g. you sitting on a swing while receiving a radio signal).
Finally, note the significance of #xvan's comment : phi= phi(t). If the phase angle is time dependent then it has a derivative that is not zero. dphi/dt is a frequency shift, so your instantaneous frequency becomes f0 + dphi/dt.
I am running FFT algorithm to detect the music note played on a guitar.
The frequencies that I am interested are in the range 65.41Hz (C2) to 1864.7Hz (A#6).
If I set the sampling frequency of the input to 16KHz, the output of FFT would yield N points from 0Hz to 16KHz linearly. All the input I am interested would be in the first N/8 points approximately. The other N*7/8 points are of no use to me. They actually are decreasing my resolution.
From Nyquist's theory (https://en.wikipedia.org/wiki/Nyquist_frequency), the sampling frequency that is needed is just twice the maximum frequency one desires. In my case, this would be about 4KHz.
Is 4KHz really the ideal sampling frequency for a guitar tuning app?
Intuitively, one would feel a better sampling frequency would give you more accurate results. However, in this case, it seems having a lesser sampling frequency is better for improving the resolution. Regards.
You are confusing the pitch of a guitar note with spectral frequency. A guitar generates lots of overtones and harmonics at a much higher frequency than the pitch of a played note. Those higher harmonics and overtones, more than the possibly weak fundamental frequency in some cases, is what the human ear hears and interprets as the lower perceived pitch.
Any of the overtones and harmonics around or above 2 kHz that are not completely low pass filtered out before sampling at 4 kHz will cause aliasing and thus corruption of your sampled data and its spectrum.
If you want to create an accurate tuner, use a pitch estimation algorithm, not an FFT peak frequency bin estimator. And depending on which pitch estimation method you choose, a higher density of samples per unit time might allow finer accuracy or greater reliability under background noise or more prompt responsiveness.
Is 4KHz really the ideal sampling frequency for a guitar tuning app?
You've been mis-reading Nyquist's theorem if you ask it like that.
States that every sampling frequency above twice your maximum signal frequency will allow you to perfectly reconstruct your original signal. So there's no "ideal" frequency. Just a set of frequencies that are sufficient. What is ideal hence depends on a lot of other things: mainly, what your digitizer really supports (hint: most sound cards can do 44.1kHz, but not 4kHz), what kind of margin you want to have for filters etc to work on, and what kind of processing power you can spend (hint: modern smart phones, PCs and even pocket calculators don't really have a hard time processing a couple hundred kHz in real time).
Also note that #hotpaw2 is right, the harmonics are important, and are multiples of the base tone frequency.
However, in this case, it seems having a lesser sampling frequency is better for improving the resolution.
no. No matter where that comes from, it's wrong. Information theory's first and foremost result is that based upon more information, you can't make worse estimates. An oversampled signal is simply more information on the same signal.
Yes, if all you are interested in is frequencies up to 2 kHz then you only need a sampling frequency of 4 kHz. This should include an anti-aliasing filter in front of the ADC or any downconverter to prevent any higher frequency components from aliasing into a lower frequency.
If all you are interested in is specific frequencies (one or two) then you may want to look at the Goertzel algorithm which is more efficient than an FFT for a single frequency. Also, the chirp-Z transform can be used to effectively get a zoomed FFT (resulting in a higher resolution over a smaller bandwidth without the computational complexity of an FFT with the same resolution). You may want to check out this CZT tutorial
I have some geographical trajectories sampled to analyze, and I calculated the histogram of data in spatial and temporal dimension, which yielded a time domain based feature for each spatial element. I want to perform a discrete FFT to transform the time domain based feature into frequency domain based feature (which I think maybe more robust), and then do some classification or clustering algorithms.
But I'm not sure using what descriptor as frequency domain based feature, since there are amplitude spectrum, power spectrum and phase spectrum of a signal and I've read some references but still got confused about the significance. And what distance (similarity) function should be used as measurement when performing learning algorithms on frequency domain based feature vector(Euclidean distance? Cosine distance? Gaussian function? Chi-kernel or something else?)
Hope someone give me a clue or some material that I can refer to, thanks~
Edit
Thanks to #DrKoch, I chose a spatial element with the largest L-1 norm and plotted its log power spectrum in python and it did show some prominent peaks, below is my code and the figure
import numpy as np
import matplotlib.pyplot as plt
sp = np.fft.fft(signal)
freq = np.fft.fftfreq(signal.shape[-1], d = 1.) # time sloth of histogram is 1 hour
plt.plot(freq, np.log10(np.abs(sp) ** 2))
plt.show()
And I have several trivial questions to ask to make sure I totally understand your suggestion:
In your second suggestion, you said "ignore all these values."
Do you mean the horizontal line represent the threshold and all values below it should be assigned to value zero?
"you may search for the two, three largest peaks and use their location and probably widths as 'Features' for further classification."
I'm a little bit confused about the meaning of "location" and "width", does "location" refer to the log value of power spectrum (y-axis) and "width" refer to the frequency (x-axis)? If so, how to combine them together as a feature vector and compare two feature vector of "a similar frequency and a similar widths" ?
Edit
I replaced np.fft.fft with np.fft.rfft to calculate the positive part and plot both power spectrum and log power spectrum.
code:
f, axarr = plt.subplot(2, sharex = True)
axarr[0].plot(freq, np.abs(sp) ** 2)
axarr[1].plot(freq, np.log10(np.abs(sp) ** 2))
plt.show()
figure:
Please correct me if I'm wrong:
I think I should keep the last four peaks in first figure with power = np.abs(sp) ** 2 and power[power < threshold] = 0 because the log power spectrum reduces the difference among each component. And then use the log spectrum of new power as feature vector to feed classifiers.
I also see some reference suggest applying a window function (e.g. Hamming window) before doing fft to avoid spectral leakage. My raw data is sampled every 5 ~ 15 seconds and I've applied a histogram on sampling time, is that method equivalent to apply a window function or I still need apply it on the histogram data?
Generally you should extract just a small number of "Features" out of the complete FFT spectrum.
First: Use the log power spec.
Complex numbers and Phase are useless in these circumstances, because they depend on where you start/stop your data acquisiton (among many other things)
Second: you will see a "Noise Level" e.g. most values are below a certain threshold, ignore all these values.
Third: If you are lucky, e.g. your data has some harmonic content (cycles, repetitions) you will see a few prominent Peaks.
If there are clear peaks, it is even easier to detect the noise: Everything between the peaks should be considered noise.
Now you may search for the two, three largest peaks and use their location and probably widths as "Features" for further classification.
Location is the x-value of the peak i.e. the 'frequency'. It says something how "fast" your cycles are in the input data.
If your cycles don't have constant frequency during the measuring intervall (or you use a window before caclculating the FFT), the peak will be broader than one bin. So this widths of the peak says something about the 'stability' of your cycles.
Based on this: Two patterns are similar if the biggest peaks of both hava a similar frequency and a similar widths, and so on.
EDIT
Very intersiting to see a logarithmic power spectrum of one of your examples.
Now its clear that your input contains a single harmonic (periodic, oscillating) component with a frequency (repetition rate, cycle-duration) of about f0=0.04.
(This is relative frquency, proprtional to the your sampling frequency, the inverse of the time beetween individual measurment points)
Its is not a pute sine-wave, but some "interesting" waveform. Such waveforms produce peaks at 1*f0, 2*f0, 3*f0 and so on.
(So using an FFT for further analysis turns out to be very good idea)
At this point you should produce spectra of several measurements and see what makes a similar measurement and how differ different measurements. What are the "important" features to distinguish your mesurements? Thinks to look out for:
Absolute amplitude: Height of the prominent (leftmost, highest) peaks.
Pitch (Main cycle rate, speed of changes): this is position of first peak, distance between consecutive peaks.
Exact Waveform: Relative amplitude of the first few peaks.
If your most important feature is absoulute amplitude, you're better off with calculating the RMS (root mean square) level of our input signal.
If pitch is important, you're better off with calculationg the ACF (auto-correlation function) of your input signal.
Don't focus on the leftmost peaks, these come from the high frequency components in your input and tend to vary as much as the noise floor.
Windows
For a high quality analyis it is importnat to apply a window to the input data before applying the FFT. This reduces the infulens of the "jump" between the end of your input vector ant the beginning of your input vector, because the FFT considers the input as a single cycle.
There are several popular windows which mark different choices of an unavoidable trade-off: Precision of a single peak vs. level of sidelobes:
You chose a "rectangular window" (equivalent to no window at all, just start/stop your measurement). This gives excellent precission of your peaks which now have a width of just one sample. Your sidelobes (the small peaks left and right of your main peaks) are at -21dB, very tolerable given your input data. In your case this is an excellent choice.
A Hanning window is a single cosine wave. It makes your peaks slightly broader but reduces side-lobe levels.
The Hammimg-Window (cosine-wave, slightly raised above 0.0) produces even broader peaks, but supresses side-lobes by -42 dB. This is a good choice if you expect further weak (but important) components between your main peaks or generally if you have complicated signals like speech, music and so on.
Edit: Scaling
Correct scaling of a spectrum is a complicated thing, because the values of the FFT lines depend on may things like sampling rate, lenght of FFT, window, and even implementation details of the FFT algorithm (there exist several different accepted conventions).
After all, the FFT should show the underlying conservation of energy. The RMS of the input signal should be the same as the RMS (Energy) of the spectrum.
On the other hand: if used for classification it is enough to maintain relative amplitudes. As long as the paramaters mentioned above do not change, the result can be used for classification without further scaling.
I have been developing a small software in .NET that takes a signal from a sensor in real time and takes the FFT of that signal which is also shown in real time.
I have used the alglib library for the FFT function. Now my purpose is to observe the intensity of some particular frequency in time.
In order to check the software, I provided a sine wave to its input having a frequency of 1 Hz. The following image shows the screen shot from the software. The upper graph shows the frequency spectrum showing the peak at 1 Hz. However, when this peak is observed in time, as shown in lower graph, the intensity behaves like a sine wave.
My sampling frequency is 30kHz. What I do not understand is how am I getting this sine signal and why is the magnitude of frequency behaving like this?
This is an example of the effects of Windowing. It derives from the fact that the FFT is not a precise operation except for when dealing with perfectly periodic signals. When you window your signal, you turn it into a smaller chunk that may not repeat perfectly. The FFT algorithm calculates the spectrum of this chunk of audio, repeated infinitely. Since it is not a perfect sine wave, you don't get an exact value for the result. Furthermore, we can see that if your window doesn't line up perfectly with a multiple of your signal frequency, then it will phase shift with respect to your signal, the window capturing a slightly different chunk of your signal, and the FFT calculating the spectrum of a different infinitely repeated signal. If you think about it, this phase difference will naturally be periodic as well, as the window catches up with the next period of your signal.
However, this would only explain smaller variations in the intensity. Assuming you used correct labels on the axes of the bottom graph (something you should double-check), something else is wrong. You're window might be too small (although I expect not, because then you would see more spectral bleeding). Another possibility that just occurred to me is that you might just be plotting the real part of the FFT, not the magnitude. As the phase changes, the real and complex parts might vary, but you'd expect the magnitude to stay roughly the same.
How do i get frequency using FFT? What's the right procedure and codes?
Pitch detection typically involves measuring the interval between harmonics in the power spectrum. The power spectrum is obtained form the FFT by taking the magnitude of the first N/2 bins (sqrt(re^2 + im^2)). However there are more sophisticated techniques for pitch detection, such as cepstral analysis, where we take the FFT of the log of the power spectrum, in order to identify periodicity in the spectral peaks.
A sustained note of a musical instrument is a periodic signal, and our friend Fourier (the second "F" in "FFT") tells us that any periodic signal can be constructed by adding a set of sine waves (generally with different amplitudes, frequencies, and phases). The fundamental is the lowest frequency component and it corresponds to pitch; the remaining components are overtones and are multiples of the fundamental's frequency. It is the relative mixture of fundamental and overtones that determines timbre, or the character of an instrument. A clarinet and a trumpet playing in unison sound "in tune" because they share the same fundamental frequency, however, they are individually identifiable because of their differing timbre (overtone mixture).
For your problem, you could sample the trumpet over a time window, calculate the FFT (which decomposes the sequence of samples into its constituent digital frequencies), and then assert that the pitch is the frequency of the bin with the greatest magnitude. If you desire, this could then be trivially quantized to the nearest musical half step, like E flat. (Lookup FFT on Wikipedia if you don't understand the relationship between the sampling frequency and the resultant frequency bins, or if you don't understand the detriment of having too low a sampling frequency.) This will probably meet your needs because the fundamental component usually has greater energy than any other component. The longer the window, the greater the pitch accuracy because the bin centers will become more closely spaced in frequency. However, if the window is so long that the trumpet is changing its pitch appreciably over the duration of the window, then the technique's effectiveness will break down considerably.
DansTuner is my open source project to solve this problem. I am in fact a trumpet player. It has pitch detection code lifted from Audacity.
ia added this org.apache.commons.math.transform.FastFourierTransforme package to the project and its works perfectly
Here is a short blog article on non-parametric techniques to estimating the PSD (power spectral density) along with some more detailed links. This might get you started in estimating the PSD - and then finding the pitch.