Signal Decomposition

Introduction

Signal decomposition is the art of breaking a complex signal into simpler, more manageable components. Just as white light passes through a prism and separates into distinct colors, any signal can be decomposed into fundamental building blocks that are easier to analyze individually. The specific decomposition you choose depends entirely on your analysis goal — if you want to understand Fourier symmetry properties, use even-odd decomposition; if you need to separate a constant bias from fluctuations, use DC-AC decomposition; if you want an energy-preserving representation for compression, use orthogonal basis decomposition.

The power of decomposition lies in the fact that many analysis techniques become dramatically simpler when applied to the individual components rather than to the original composite signal. This principle underlies virtually all transform-based signal processing.

Even-Odd Decomposition

Any real signal can be uniquely written as the sum of an even function and an odd function:

$$x(t) = x_e(t) + x_o(t)$$

where: $$x_e(t) = \frac{x(t) + x(-t)}{2} \qquad x_o(t) = \frac{x(t) - x(-t)}{2}$$

Verification of Properties

The even part satisfies $x_e(-t) = x_e(t)$ (symmetric about $t = 0$): $$x_e(-t) = \frac{x(-t) + x(t)}{2} = x_e(t) \quad \checkmark$$

The odd part satisfies $x_o(-t) = -x_o(t)$ (anti-symmetric): $$x_o(-t) = \frac{x(-t) - x(t)}{2} = -x_o(t) \quad \checkmark$$

The sum reconstructs the original: $x_e(t) + x_o(t) = \frac{x(t) + x(-t)}{2} + \frac{x(t) - x(-t)}{2} = x(t) \quad \checkmark$

Significance in Fourier Analysis

This decomposition is crucial because:

• The Fourier Transform of an even signal is purely real
• The Fourier Transform of an odd signal is purely imaginary
• For a real signal: $\text{Re}\{X(j\omega)\} = \mathcal{F}\{x_e(t)\}$ and $j\cdot\text{Im}\{X(j\omega)\} = \mathcal{F}\{x_o(t)\}$

This means if you know a signal is even, you only need to compute cosine coefficients in its Fourier representation, cutting the computational work in half.

Worked Example

Decompose $x(t) = e^{-2t}u(t)$ (one-sided decaying exponential):

Even part: $x_e(t) = \frac{e^{-2t}u(t) + e^{2t}u(-t)}{2} = \frac{1}{2}e^{-2|t|}$

This is a symmetric double-sided exponential with its peak at the origin.

Odd part: $x_o(t) = \frac{e^{-2t}u(t) - e^{2t}u(-t)}{2} = \frac{1}{2}\text{sgn}(t) \cdot e^{-2|t|}$

This is anti-symmetric: positive for $t > 0$ and negative for $t < 0$.

DC-AC Decomposition

Every signal can be split into a constant (DC) component and a time-varying (AC) component:

$$x(t) = x_{DC} + x_{AC}(t)$$

For periodic signals with period $T$: $$x_{DC} = \frac{1}{T}\int_0^T x(t) \, dt$$

The AC component is the remainder: $x_{AC}(t) = x(t) - x_{DC}$, which always has zero mean.

Physical Meaning

In electrical engineering, the DC component represents the average voltage or current, while the AC component represents the fluctuations. A 5V power supply with 50mV ripple has $x_{DC} = 5$ V and $x_{AC}(t)$ is the 50mV oscillation.

Power Decomposition

For signals with both DC and AC content: $$P_{total} = P_{DC} + P_{AC} = x_{DC}^2 + \frac{1}{T}\int_0^T x_{AC}^2(t) dt$$

The total power splits cleanly into DC power and AC power because $x_{AC}$ has zero mean (the cross-term vanishes).

Causal/Anti-Causal Decomposition

A signal can be separated into its causal part (existing for $t \geq 0$) and anti-causal part (existing for $t < 0$):

$$x(t) = x_+(t) + x_-(t)$$

where $x_+(t) = x(t) \cdot u(t)$ is the causal part and $x_-(t) = x(t) \cdot u(-t)$ is the anti-causal part (with appropriate handling of $t = 0$).

This decomposition is important in filter design: a physically realizable (causal) filter can only use the causal part of the ideal impulse response.

Orthogonal Basis Decomposition

The most general and powerful form of decomposition represents a signal as a linear combination of orthogonal basis functions $\{\phi_k(t)\}$:

$$x(t) = \sum_{k} c_k \phi_k(t)$$

The basis functions satisfy orthogonality: $\langle \phi_m, \phi_n \rangle = 0$ for $m \neq n$.

The expansion coefficients are computed as: $$c_k = \frac{\langle x, \phi_k \rangle}{\|\phi_k\|^2} = \frac{\int x(t)\phi_k^*(t) dt}{\int |\phi_k(t)|^2 dt}$$

Common Orthogonal Bases

Parseval's Theorem (Generalized)

For orthogonal decompositions, signal energy distributes across components: $$E_x = \sum_k |c_k|^2 \|\phi_k\|^2$$

No energy is lost or created in the decomposition — it simply redistributes among the basis components. This energy conservation makes orthogonal decompositions ideal for compression: discard components with small $|c_k|$ and reconstruct with minimal error.

Impulse Decomposition

Perhaps the most fundamental decomposition represents any signal as a continuum of weighted, shifted impulses:

$$x(t) = \int_{-\infty}^{\infty} x(\tau)\delta(t - \tau) d\tau$$

This is the sifting property of the delta function, and it is the mathematical foundation of convolution. It says: "a signal is the superposition of impulses at every time instant, each weighted by the signal's value at that instant."

For discrete-time: $$x[n] = \sum_{k=-\infty}^{\infty} x[k]\delta[n-k]$$

When this decomposed signal passes through an LTI system, each impulse produces a shifted impulse response, and the output is the sum (convolution): $$y[n] = \sum_k x[k] h[n-k] = x[n] * h[n]$$

Step Decomposition

A signal can also be represented as a sum of shifted step functions:

$$x(t) = \int_{-\infty}^{\infty} x'(\tau) u(t-\tau) d\tau$$

This shows that any signal can be built from infinitesimally small step changes, each weighted by the derivative $x'(\tau)$.

Practical Applications

• JPEG compression: Decomposes image blocks using DCT, discards small coefficients
• MP3 audio: Modified DCT separates audio into frequency bands for perceptual coding
• Power system analysis: DC-AC separation identifies harmonic distortion
• Causal filter design: Only the causal part of an ideal filter is implementable
• Fourier analysis: Even-odd decomposition simplifies coefficient computation
• LTI system theory: Impulse decomposition leads directly to convolution

Key Takeaways

• Even-odd decomposition is unique and simplifies Fourier analysis (cosine vs sine components)
• DC-AC decomposition separates the average from fluctuations; powers add independently
• Orthogonal bases enable energy-preserving, invertible representations
• Parseval's theorem guarantees energy conservation across orthogonal decompositions
• Impulse decomposition is the foundation of convolution and all LTI system theory
• The choice of decomposition depends entirely on your analysis objective