Fourier Analysis, Frequency Response, and Spectral Views
Fourier analysis, frequency response, spectrum, convolution theorem, sinusoid
1 Role
This is the second page of the Signal Processing and Estimation module.
Its job is to make the core shift from time-domain thinking to frequency-domain thinking.
The first page said:
LTIsystems are described by convolution
This page says:
- the same
LTIsystems often become much easier to understand in the spectral domain
2 First-Pass Promise
Read this page after Signals, Convolution, and Linear Time-Invariant Systems.
If you stop here, you should still understand:
- why Fourier-style decompositions are useful
- why sinusoids are special for
LTIsystems - what frequency response means
- why convolution becomes multiplication in the spectral domain
3 Why It Matters
Convolution is conceptually clean, but it can be cumbersome to compute with directly.
A long filter applied to a complicated signal can be hard to parse in the time domain.
The frequency-domain viewpoint gives a different strategy:
- decompose the signal into oscillatory components
- understand how the system acts on each component
- recombine the result
This matters because many questions are naturally spectral:
- which frequencies pass through?
- which ones are attenuated?
- where is distortion happening?
- what does smoothing really remove?
This is the first page where Signal Processing and Estimation starts to feel like a distinct subject rather than only “convolution with extra notation.”
4 Prerequisite Recall
- an
LTIsystem is fully determined by its impulse response - convolution expresses the output from the input and impulse response
- linearity means we can decompose inputs into simple pieces and add the responses
- complex exponentials and sinusoids are the natural oscillatory building blocks of signals
5 Intuition
5.1 Sinusoids Are Special For LTI Systems
For an LTI system, a sinusoid does not usually come out as a completely different waveform.
Instead, the system typically:
- rescales it
- shifts its phase
That is why sinusoids behave like natural “test signals” for system analysis.
5.2 Frequency Is Another Coordinate System For Signals
A signal can be viewed in at least two useful ways:
- in time: how it changes from one instant to the next
- in frequency: how much oscillation at each scale is present
At first pass, the frequency view is not replacing the time view.
It is giving a second coordinate system that often makes structure simpler.
5.3 Frequency Response Is The System’s Spectral Signature
The frequency response tells us how the system acts on each sinusoidal frequency.
So instead of describing the system by an impulse response in time, we can describe it by:
- gain versus frequency
- phase shift versus frequency
That is the first-pass meaning of a spectral system description.
5.4 Convolution Becomes Multiplication
The biggest computational payoff is this:
- convolution in time is awkward
- multiplication in frequency is easier
So many filtering and distortion questions become simpler after a Fourier transform.
5.5 Spectral Views Explain Filtering
Low-pass, high-pass, and band-pass behavior are naturally frequency-domain ideas.
Even if a filter is defined in time, its practical meaning is often:
- keep slow variations
- suppress rapid oscillations
- isolate a frequency band
6 Formal Core
Definition 1 (Definition Idea: Fourier Representation) A Fourier representation rewrites a signal in terms of oscillatory building blocks such as sinusoids or complex exponentials.
At first pass, the point is not every technical version of the Fourier transform.
It is the organizing idea:
signals can be decomposed into frequency components
Theorem 1 (Theorem Idea: Sinusoids As Special Inputs For LTI Systems) For an LTI system, complex exponentials behave like eigenfunctions: the output at a given frequency has the same frequency, changed only by a complex scale factor.
That scale factor depends on frequency and is the frequency response.
Definition 2 (Definition: Frequency Response) The frequency response H(\omega) of an LTI system describes how the system scales and phase-shifts oscillatory components at frequency \omega.
At first pass:
|H(\omega)|tells us the gain\arg H(\omega)tells us the phase shift
Theorem 2 (Theorem Idea: Convolution Theorem) Convolution in the time domain becomes multiplication in the spectral domain.
In symbolic first-pass form:
\[ \mathcal{F}(x * h) = \mathcal{F}(x)\,\mathcal{F}(h). \]
This is the main computational theorem of the page.
Theorem 3 (Theorem Idea: Filtering Can Be Read Spectrally) A filter modifies a signal by amplifying some frequencies and attenuating others.
So the same system can be read:
- in time by its impulse response
- in frequency by its response curve
7 Worked Example
Take the moving-average system from the previous page:
\[ y[n]=\frac12 x[n]+\frac12 x[n-1]. \]
Its impulse response is
\[ h[0]=\frac12,\qquad h[1]=\frac12. \]
Now feed in a slowly varying signal and a rapidly alternating signal.
For a slowly varying input, adjacent samples are similar, so averaging barely changes the signal.
For an alternating input like
\[ 1,-1,1,-1,\dots \]
neighboring samples tend to cancel under averaging.
So the same filter:
- preserves low-frequency behavior
- suppresses high-frequency oscillation
That is the first-pass spectral meaning of smoothing.
8 Computation Lens
When you meet a system or filter, ask:
- what does it do in time?
- what frequencies does it preserve, suppress, or distort?
- is the time-domain formula simpler, or is the spectral-domain formula simpler?
- is the important phenomenon gain, phase shift, bandwidth, or aliasing risk?
These questions often tell you which representation is the right one for the problem.
9 Application Lens
9.1 Communication
Frequency response explains channel bandwidth, attenuation, and distortion much more directly than raw convolution formulas.
9.2 Denoising And Filtering
Many smoothers are best understood as frequency-selective operators, not only local averages.
9.3 Spectral ML And Representation Models
Transforms, convolutional models, graph spectra, and many sequence-processing pipelines reuse this same “change basis, analyze by components” viewpoint.
10 Stop Here For First Pass
If you stop here, retain these five ideas:
- Fourier analysis gives a spectral coordinate system for signals
- sinusoids are special inputs for
LTIsystems - frequency response tells us how the system acts on each frequency
- convolution in time becomes multiplication in the spectral domain
- filtering is often easiest to understand as frequency selection
11 Go Deeper
The strongest next page is:
The strongest adjacent live pages are:
12 Optional Deeper Reading After First Pass
- MIT 6.003: Signals and Systems lecture notes - official note index for the full signals-and-systems sequence. Checked
2026-04-25. - MIT 6.003 frequency response lecture - official notes page focused on frequency response. Checked
2026-04-25. - Stanford EE102A course outline - official outline emphasizing frequency-domain representation, sampling, and communication examples. Checked
2026-04-25. - Stanford EE102A notes page - official note index for the lecture sequence. Checked
2026-04-25. - Stanford EE102A bulletin - official Stanford bulletin entry describing frequency-domain representation and filtering. Checked
2026-04-25.
13 Sources and Further Reading
- MIT 6.003: Signals and Systems lecture notes -
First pass- official note index for the signals-and-systems arc. Checked2026-04-25. - MIT 6.003 frequency response lecture -
First pass- official notes page for frequency response and spectral system views. Checked2026-04-25. - Stanford EE102A course outline -
First pass- official outline covering Fourier and frequency-domain ideas. Checked2026-04-25. - Stanford EE102A notes page -
First pass- official note index for the course sequence. Checked2026-04-25. - Stanford EE102A bulletin -
Second pass- official Stanford bulletin entry summarizing the syllabus. Checked2026-04-25.