A boring lecture about signals and systems, and its digital implementation in DSP.
Borderline 'song' less 'poetic'.
The actual material (text that is read out by a vocal synthesiser) is in Japanese.
## Simple notes
For things that are not easily pronounced (e.g. formulae, expressions, symbols), inclusive either:
- Mark how the thing should be pronounced in a pair of adjacent parentheses
- e.g. $H(\omega)$ (エッチオメガ)
- (inclusive) Or don't embed these elements too deep into the flow of sentences; use common, laymen's language instead
- e.g. not `Fs/2(サンプリング周波数の半分)までの帯域を忠実に再構成できる` but `サンプリング周波数の半分までの帯域を忠実に再構成できる`
The general direction is to stay boring, but the deadpan kind of boring. Keep technical language appearing but explain things in more laymen's language.
## Outline
### 1. Signals and systems
- What is a signal
- A function of one or more variables carrying information; here, amplitude over time
- What is a system
- A system maps an input signal to an output signal
- Characterized entirely by its behavior on inputs — the internals are not required
- Most systems are nonlinear — linearity is a special property, not the default
- Linearity: superposition holds — scaling and additivity
- Most physical systems violate this under sufficient conditions (e.g. overdrive: linear below clip threshold, nonlinear at and above it)
- Time-invariance
- A time-invariant system responds identically regardless of when the input arrives
- Violation: any system with time-varying parameters (e.g. modulated effects)
- LTI as the restricted, studied case
- Satisfies both simultaneously — analytically tractable
- A deliberate simplification; not a description of typical reality
- The impulse signal; impulse response
- The impulse: unit energy concentrated at a single point in time
- The impulse response: what an LTI system outputs when fed an impulse
- Completely characterizes the system
- The impulse response is a function — it exists in the same space as signals; systems are absorbed into signal space
### 2. LTI and convolution
- An LTI system is fully characterized by its impulse response
- Any input decomposes into scaled, shifted impulses
- By linearity and time-invariance, the output is the corresponding sum of scaled, shifted impulse responses
- Convolution as the operation
- That sum is convolution: $y = x * h$
- In continuous time, this is an integral: $y(t) = \int x(\tau) \, h(t - \tau) \, d\tau$
- Symmetric in its arguments: $x * h = h * x$
- Signal and system are interchangeable operands — both are functions; convolution makes no distinction between them
- Commutativity: for LTI systems, order of composition is irrelevant
- Two LTI systems in series have a combined impulse response equal to the convolution of each
- $h_1 * h_2 = h_2 * h_1$ — order does not affect the result
### 3. Signal transformation
- A signal can be represented equivalently in the frequency domain
- The Fourier transform: $s(t) \leftrightarrow S(\omega)$; any signal is a sum of sinusoids at different frequencies
- Each sinusoid has a frequency, amplitude, and phase
- $S(\omega)$ is the spectrum: how much of each frequency is present
- Analog signals carry components from DC to arbitrarily high frequencies
### 4. Filters
- A filter is an LTI system — characterized by its frequency response $H(\omega)$
- $H(\omega)$ is the Fourier transform of the impulse response $h(t)$
- Filtering: shaping a signal's spectrum by selectively attenuating or preserving frequency bands
- Basic types: lowpass, highpass, bandpass
- Lowpass: passes low frequencies, attenuates high
- Highpass: passes high frequencies, attenuates low
- Bandpass: passes a target band, attenuates above and below
- In discrete time: finite impulse response (FIR) and infinite impulse response (IIR) filters
### 5. Discretization
- Moving from continuous to discrete: two axes require discretization
- Amplitude: $N$ bits → $2^N$ discrete levels
- Each sample is mapped to the nearest representable level; the error is quantization noise
- The representable range is fixed by design and hardware (ADC/DAC reference voltage)
- Hard clip at the boundary — analog degrades gradually toward its physical limits; digital does not
- Time axis: sampling captures only finitely many values per second
- Analog signals contain frequency components up to infinity — a discrete system cannot represent all of them
- Discrete time is viable when the high-frequency components that would be lost are not needed
- Nyquist: given sampling rate $F_s$, faithful reconstruction is guaranteed up to $F_s/2$
- What $F_s/2$ should be is an applied engineering question — Nyquist does not answer it
- Components above $F_s/2$ are lost: aliasing
- LTI and convolution hold in discrete time — commutativity survives
- Convolution becomes a sum (finite for FIR, infinite for IIR)
### 6. Nonlinearity and consequences
- Real systems violate LTI: time-varying parameters, feedback, saturation
- Each breaks a different condition: time-variance breaks TI, feedback and saturation break linearity
- Examples
- Modulated delay (chorus, flanger): delay time varies with time → time-invariance breaks; same input at different times yields different output
- Overdrive: saturation above threshold → linearity breaks; EQ before overdrive ≠ overdrive before EQ
- Commutativity breaks — order of composition is no longer irrelevant
- $h_1$ followed by $h_2$ ≠ $h_2$ followed by $h_1$ in general
- No universal rule; behavior must be examined per case
### 7. Summary
- Most of the framework rests on the LTI assumption
- Signal–system equivalence, convolution, commutativity, frequency-domain analysis — all require linearity and time-invariance
- Real systems break the assumption routinely
- Nonlinearity is not binary — it is a spectrum
- Some systems are close enough to linear that the theory applies as a practical approximation
- Others are not — linear theory offers no useful prediction
- Judging where a system falls requires knowing what linearity is in the first place
- The framework is studied not as a description of reality, but as a baseline against which deviations are measured
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.