FFT and frequency analysis in MATLAB - Deep Dive

Overview - FFT and frequency analysis

What is it?

FFT stands for Fast Fourier Transform. It is a method to break down a signal into its basic frequency parts. Frequency analysis helps us understand what tones or cycles make up a signal. This is useful in many fields like sound, images, and science.

Why it matters

Without FFT and frequency analysis, we would only see signals as jumbled waves. We couldn't easily find patterns like musical notes or repeated events. This would make tasks like noise removal, signal compression, or medical diagnosis much harder. FFT makes these tasks fast and practical.

Where it fits

Before learning FFT, you should know about basic signals and waves, like sine and cosine. After FFT, you can explore filtering, signal processing, and spectral analysis. It fits in the journey after learning about time-domain signals and before advanced signal processing techniques.

Mental Model

Core Idea

FFT quickly transforms a signal from time view into frequency view, showing what frequencies build the signal.

Think of it like...

Imagine a smoothie made of different fruits. FFT is like a magic filter that tells you how much of each fruit is inside the smoothie without tasting it.

Time Domain Signal ──▶ [FFT] ──▶ Frequency Spectrum

┌───────────────┐       ┌───────────────┐       ┌───────────────┐
│ Time Signal   │──────▶│ FFT Algorithm │──────▶│ Frequency     │
│ (wave over    │       │ (fast math)   │       │ Spectrum      │
│ time)         │       │               │       │ (amplitude vs │
│               │       │               │       │ frequency)    │
└───────────────┘       └───────────────┘       └───────────────┘

Build-Up - 7 Steps

1

Foundation

Understanding Signals and Frequencies

Concept: Signals can be seen as waves that change over time, and frequencies describe how fast these waves repeat.

A signal is like a sound wave or a heartbeat. It changes over time. Frequency tells us how many times the wave repeats in one second. For example, a 5 Hz wave repeats 5 times per second. We can draw signals as lines that go up and down.

Result

You can recognize that signals have repeating patterns and that frequency measures these repeats.

Understanding signals as waves and frequency as repeats per second is the base for all frequency analysis.

2

Foundation

What is Fourier Transform?

Concept: Fourier Transform changes a signal from time view to frequency view, showing the signal's frequency parts.

Imagine a signal made by mixing different sine waves. Fourier Transform finds out how much of each sine wave (frequency) is inside the signal. It uses math to do this change from time to frequency.

Result

You get a new view of the signal showing which frequencies are present and their strengths.

Seeing signals in frequency helps us find hidden patterns and understand the signal's nature.

3

Intermediate

Why Fast Fourier Transform (FFT)?

🤔Before reading on: do you think FFT is a different transform or a faster way to do Fourier Transform? Commit to your answer.

Concept: FFT is a fast way to calculate the Fourier Transform, making it practical for computers.

The normal Fourier Transform takes a long time for many data points. FFT uses clever math tricks to reduce the work from N² steps to N log N steps, where N is the number of points. This speed lets us analyze signals quickly.

Result

FFT produces the same frequency information as Fourier Transform but much faster.

Knowing FFT is an algorithmic speed-up explains why frequency analysis is widely used in real-time systems.

4

Intermediate

Interpreting FFT Output in MATLAB

🤔Before reading on: do you think FFT output directly shows frequencies or needs extra steps to map to real frequencies? Commit to your answer.

Concept: FFT output is complex numbers representing amplitude and phase for frequency bins; we must map these bins to actual frequencies.

In MATLAB, fft(signal) returns complex values. The magnitude (abs) shows strength, and angle shows phase. Frequencies correspond to indices scaled by sampling rate. Usually, we plot magnitude vs frequency to see the spectrum.

Result

You can create a frequency spectrum plot showing which frequencies are strong in the signal.

Understanding how to convert FFT output to real frequencies is key to making sense of frequency analysis.

5

Intermediate

Windowing and Its Effect on FFT

🤔Before reading on: do you think applying a window before FFT changes the frequency content or just smooths the output? Commit to your answer.

Concept: Windowing shapes the signal to reduce edge effects and improve frequency resolution in FFT.

Signals are often cut into chunks before FFT. This can cause sharp edges that create false frequencies (spectral leakage). Applying a window function (like Hamming) smooths edges, reducing leakage and making frequency peaks clearer.

Result

Frequency spectrum becomes cleaner and more accurate with windowing.

Knowing windowing prevents common FFT artifacts helps produce reliable frequency analysis.

6

Advanced

FFT Resolution and Zero Padding

🤔Before reading on: does adding zeros to a signal before FFT increase frequency detail or just interpolate the spectrum? Commit to your answer.

Concept: Zero padding adds extra points to improve frequency display resolution without adding new information.

Adding zeros to the end of a signal before FFT increases the number of frequency bins. This makes the frequency plot smoother and easier to read but does not increase actual frequency resolution. It helps in visualizing frequencies more clearly.

Result

Frequency spectrum plots look smoother and more detailed but represent the same data.

Understanding zero padding as interpolation avoids overestimating frequency precision.

7

Expert

FFT Limitations and Spectral Leakage Surprises

🤔Before reading on: do you think FFT always perfectly separates close frequencies if the signal is long enough? Commit to your answer.

Concept: FFT has limits in frequency resolution and can mix close frequencies due to spectral leakage and window choice.

Even with long signals, FFT cannot perfectly separate frequencies closer than the resolution limit (sampling rate / signal length). Spectral leakage from windowing and signal discontinuities can blur peaks. Experts use advanced windows or methods like multitaper or wavelets to improve analysis.

Result

You learn that FFT results can be misleading without careful interpretation and advanced techniques.

Knowing FFT's fundamental limits prevents misinterpretation and guides better frequency analysis methods.

Under the Hood

FFT works by recursively breaking down a signal into smaller parts, exploiting symmetries in sine and cosine calculations. It uses a divide-and-conquer approach to reduce repeated calculations. Internally, it rearranges data and combines results in stages, producing complex numbers representing amplitude and phase for each frequency bin.

Why designed this way?

The original Fourier Transform was too slow for large data sets. FFT was designed to speed up computation by reducing redundant calculations. This design balances speed and accuracy, making frequency analysis practical for real-time and large-scale data.

Signal (N points)
   │
   ▼
Split into even and odd indices
   │           │
   ▼           ▼
FFT on even   FFT on odd
   │           │
   └────Combine with twiddle factors─────▶ Final FFT Output

Each stage halves data size, combining results efficiently.

Myth Busters - 4 Common Misconceptions

Quick: Does FFT output directly give real frequencies without scaling? Commit yes or no.

Common Belief: FFT output directly shows frequencies as indices without any conversion.

Reality: FFT output indices correspond to frequency bins that must be scaled by sampling rate and signal length to get real frequencies.

Why it matters: Misreading FFT indices as frequencies leads to wrong conclusions about signal content.

Quick: Does zero padding increase actual frequency resolution? Commit yes or no.

Common Belief: Adding zeros to a signal before FFT increases the true frequency resolution.

Reality: Zero padding only interpolates the frequency spectrum; it does not add new information or improve true resolution.

Why it matters: Believing zero padding improves resolution can cause overconfidence in frequency precision.

Quick: Does FFT perfectly separate two close frequencies if the signal is long? Commit yes or no.

Common Belief: Longer signals always allow FFT to perfectly separate close frequencies.

Reality: FFT resolution depends on signal length and sampling rate, but spectral leakage and windowing limit perfect separation.

Why it matters: Ignoring these limits can cause misinterpretation of frequency components and false analysis.

Quick: Does windowing change the frequencies present in the signal? Commit yes or no.

Common Belief: Applying a window before FFT changes the actual frequencies in the signal.

Reality: Windowing shapes the signal to reduce artifacts but does not add or remove true frequencies.

Why it matters: Misunderstanding windowing effects can lead to wrong assumptions about signal content.

Expert Zone

1

FFT output is complex; phase information is often ignored but crucial for signal reconstruction and advanced analysis.

2

Choice of window affects trade-off between frequency resolution and spectral leakage; no one window fits all cases.

3

FFT assumes signal is periodic and continuous; real signals often violate this, causing artifacts that experts must correct.

When NOT to use

FFT is not ideal for signals that change frequency over time (non-stationary). Alternatives like Short-Time Fourier Transform (STFT) or wavelet transforms are better for time-frequency analysis.

Production Patterns

In real systems, FFT is used for audio equalizers, vibration analysis, wireless communication signal decoding, and medical imaging. Professionals combine FFT with filtering, windowing, and averaging to improve reliability.

Connections

Digital Signal Processing (DSP)

FFT is a core tool within DSP for analyzing and manipulating signals.

Understanding FFT deepens grasp of DSP concepts like filtering, modulation, and sampling.

Music Theory

Frequency analysis reveals musical notes and harmonics in sound signals.

Knowing FFT helps connect physical sound waves to musical pitch and timbre.

Quantum Mechanics

Fourier transforms relate position and momentum representations of particles.

Recognizing FFT as a mathematical bridge between dual views shows its fundamental role beyond engineering.

Common Pitfalls

#1 Misinterpreting FFT output indices as direct frequencies.

Wrong approach:

freqs = 0:length(signal)-1; plot(freqs, abs(fft(signal)));

Correct approach:

Fs = 1000; N = length(signal); freqs = (0:N-1)*(Fs/N); plot(freqs, abs(fft(signal)));

Root cause:

Not scaling FFT bins by sampling rate and signal length causes wrong frequency axis.

#2 Using raw signal without windowing causing spectral leakage.

Wrong approach:

Y = fft(signal); plot(abs(Y));

Correct approach:

w = hamming(length(signal)); Y = fft(signal .* w); plot(abs(Y));

Root cause:

Ignoring edge effects leads to false frequency components.

#3 Assuming zero padding increases true frequency resolution.

Wrong approach:

padded_signal = [signal zeros(1, 1000)]; Y = fft(padded_signal);

Correct approach:

Use longer actual signal or advanced methods; zero padding only interpolates.

Root cause:

Confusing interpolation with real resolution improvement.

Key Takeaways

FFT transforms signals from time to frequency domain quickly, revealing frequency components.

Interpreting FFT output requires understanding sampling rate and frequency bin mapping.

Windowing before FFT reduces artifacts and improves frequency analysis accuracy.

Zero padding smooths frequency plots but does not increase true frequency resolution.

FFT has limits in separating close frequencies; advanced methods may be needed for precise analysis.

Practice

1. What does the MATLAB function fft() do to a signal?

easy

A. It amplifies the signal amplitude.

B. It converts the signal from time domain to frequency domain.

C. It smooths the signal to remove noise.

D. It shifts the signal in time.

Solution

1. Step 1: Understand the purpose of FFT

   FFT stands for Fast Fourier Transform, which changes a signal from time domain to frequency domain.

2. Step 2: Identify what fft() does in MATLAB

   In MATLAB, fft() computes the discrete Fourier transform, showing frequency components of the signal.

3. Final Answer:

   It converts the signal from time domain to frequency domain. -> Option B

4. Quick Check:

   FFT = frequency transform [OK]

Hint: FFT always changes time signals to frequency signals [OK]

Common Mistakes:

• Thinking FFT smooths or filters signals
• Confusing FFT with time shifting
• Assuming FFT amplifies signal amplitude

2. Which of the following is the correct syntax to compute the FFT of a vector x in MATLAB?

easy

A. Y = fft x;

B. Y = fft[x];

C. Y = fft(x);

D. Y = fft{x};

Solution

1. Step 1: Recall MATLAB function call syntax

   In MATLAB, functions are called with parentheses enclosing arguments, e.g., func(arg).

2. Step 2: Check each option's syntax

   Y = fft(x); uses correct syntax: fft(x). Options A, C, and D use invalid syntax for function calls in MATLAB.

3. Final Answer:

   Y = fft(x); -> Option C

4. Quick Check:

   Function calls use parentheses () [OK]

Hint: Use parentheses () for function calls in MATLAB [OK]

Common Mistakes:

• Using square brackets [] instead of parentheses
• Omitting parentheses around arguments
• Using curly braces {} which are for cell arrays

3. Given the MATLAB code:

x = [1 1 0 0];
Y = fft(x);
absY = abs(Y);
disp(absY);

What is the output displayed?

medium

A. [1 1 1 1]

B. [2 0 2 0]

C. [4 0 0 0]

D. [2 1.4142 0 1.4142]

Solution

1. Step 1: Compute FFT of x = [1 1 0 0]

   The FFT of this sequence results in complex values whose magnitudes are [2, sqrt(2), 0, sqrt(2)] approximately.

2. Step 2: Calculate absolute values

   Taking absolute values gives approximately [2, 1.4142, 0, 1.4142].

3. Final Answer:

   [2 1.4142 0 1.4142] -> Option D

4. Quick Check:

   FFT magnitude matches [2 1.4142 0 1.4142] [OK]

Hint: FFT magnitude shows signal strength at each frequency [OK]

Common Mistakes:

• Confusing FFT output with original signal
• Not taking absolute value before display
• Assuming FFT output is all real numbers

4. Consider this MATLAB code snippet:

Fs = 1000;
t = 0:1/Fs:1-1/Fs;
x = cos(2*pi*50*t) + cos(2*pi*120*t);
Y = fft(x);
P2 = abs(Y/length(x));
P1 = P2(1:length(x)/2+1);
P1(2:end-1) = 2*P1(2:end-1);
plot(P1);

What is the likely error in this code?

medium

A. The frequency axis is missing for the plot, so frequencies are unclear.

B. The length of x is not used correctly in dividing Y.

C. The FFT function is called incorrectly without arguments.

D. The signal x is not defined properly.

Solution

1. Step 1: Analyze the FFT and amplitude spectrum calculation

   The code correctly computes the single-sided amplitude spectrum P1 from Y.

2. Step 2: Check the plot command

   The plot shows amplitude values but does not label or define the frequency axis, making interpretation difficult.

3. Final Answer:

   The frequency axis is missing for the plot, so frequencies are unclear. -> Option A

4. Quick Check:

   Plot frequency axis needed for clarity [OK]

Hint: Always plot frequency axis with amplitude spectrum [OK]

Common Mistakes:

• Forgetting to create frequency vector for x-axis
• Plotting amplitude without frequency labels
• Misusing FFT input or output sizes

5. You have a noisy signal sampled at 1000 Hz stored in vector data. You want to find the dominant frequency component using FFT. Which MATLAB code snippet correctly computes and plots the single-sided amplitude spectrum with frequency axis?

hard

A. L = length(data); Fs = 1000; Y = fft(data); P2 = abs(Y/L); f = Fs*(0:(L/2))/L; P1 = P2(1:L/2+1); P1(2:end-1) = 2*P1(2:end-1); plot(f,P1); xlabel('Frequency (Hz)'); ylabel('Amplitude');

B. Fs = 1000; Y = fft(data); plot(abs(Y)); xlabel('Frequency'); ylabel('Amplitude');

C. L = length(data); Y = fft(data); P1 = abs(Y); plot(P1);

D. Fs = 1000; t = 0:1/Fs:1; Y = fft(data); plot(Y);

Solution

1. Step 1: Understand the steps to compute single-sided amplitude spectrum

   Calculate FFT, normalize by length, create frequency vector, extract half spectrum, double amplitudes except DC and Nyquist.

2. Step 2: Check which option follows these steps correctly

   L = length(data); Fs = 1000; Y = fft(data); P2 = abs(Y/L); f = Fs*(0:(L/2))/L; P1 = P2(1:L/2+1); P1(2:end-1) = 2*P1(2:end-1); plot(f,P1); xlabel('Frequency (Hz)'); ylabel('Amplitude'); correctly computes length, frequency vector, normalizes, adjusts amplitudes, and plots with labels. Others miss frequency axis or normalization.

3. Final Answer:

   Option A code snippet -> Option A

4. Quick Check:

   Proper FFT amplitude spectrum needs frequency axis and normalization [OK]

Hint: Normalize FFT and plot with frequency vector for correct spectrum [OK]

Common Mistakes:

• Not normalizing FFT by signal length
• Plotting full FFT without single-sided spectrum
• Missing frequency axis in plot