LTI Systems Introduction

Introduction

Linear Time-Invariant (LTI) systems occupy a privileged position in signal processing theory. They are the largest class of systems for which a complete, elegant mathematical framework exists — one that connects time-domain behavior (convolution), frequency-domain characterization (transfer functions), and stability analysis (pole-zero plots) into a unified theory. Every filter you design, every communication channel you model, and every control system you analyze will likely be treated as an LTI system, at least as a first approximation.

The power of LTI theory comes from the combination of two properties. Linearity enables the superposition principle, allowing complex signals to be decomposed into simple components. Time-invariance ensures that the system behaves consistently regardless of when it operates. Together, these properties yield the convolution integral (or sum) as the universal input-output relationship, with the impulse response as the complete system characterization.

What Makes LTI Systems Special

An LTI system has three remarkable characteristics that no other class of systems possesses simultaneously:

1. Complete Characterization by Impulse Response

For an LTI system, knowing the impulse response $h(t)$ tells you everything about the system. The response to ANY input can be computed via convolution:

$$y(t) = \int_{-\infty}^{\infty} h(\tau)x(t-\tau)d\tau = x(t) * h(t)$$

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

No other system class has this property. For nonlinear or time-varying systems, the impulse response does NOT determine the system.

2. Sinusoids are Eigenfunctions

When a complex exponential $e^{j\omega t}$ passes through an LTI system:

$$y(t) = H(j\omega) \cdot e^{j\omega t}$$

The output is the SAME complex exponential, scaled by the complex constant $H(j\omega)$. The system does not create new frequencies — it only modifies the amplitude and phase of existing frequencies. This eigenfunction property is why Fourier analysis works perfectly for LTI systems.

3. Convolution Becomes Multiplication in Transform Domain

$$Y(j\omega) = H(j\omega) \cdot X(j\omega)$$

The computationally expensive convolution integral becomes simple multiplication in the frequency domain. This enables efficient frequency-domain system analysis and FFT-based implementation.

Why the Assumptions are Reasonable

Many real systems are approximately LTI under normal operating conditions:

Linearity: Most electronic circuits operate linearly when signal levels are well within the power supply range. Amplifiers, filters, and transmission lines are linear for small signals.

Time-invariance: A resistor\' s value does not change from one moment to the next (barring temperature effects). A digital filter with fixed coefficients is time-invariant by construction.

The approximation breaks down at extremes — amplifier saturation violates linearity, component aging violates time-invariance — but within design specifications, LTI models are excellent.

The Convolution Framework

The derivation of convolution follows naturally from linearity and time-invariance:

Step 1 (Impulse Decomposition): Any input can be written as a sum of weighted, shifted impulses: $$x[n] = \sum_k x[k]\delta[n-k]$$

Step 2 (Linearity): The response to the sum is the sum of individual responses: $$y[n] = \sum_k x[k] \cdot T\{\delta[n-k]\}$$

Step 3 (Time-Invariance): The response to a shifted impulse is a shifted impulse response: $$T\{\delta[n-k]\} = h[n-k]$$

Result: $y[n] = \sum_k x[k]h[n-k] = x[n] * h[n]$

This derivation shows exactly why BOTH linearity and time-invariance are needed for convolution to apply.

System Characterization Tools

LTI systems can be characterized equivalently in multiple domains:

All representations contain the same information — they are simply different views of the same system.

Cascade, Parallel, and Feedback Connections

LTI systems can be interconnected:

Cascade (Series): $H_{total}(s) = H_1(s) \cdot H_2(s)$ — transfer functions multiply

Parallel: $H_{total}(s) = H_1(s) + H_2(s)$ — transfer functions add

Feedback: $H_{total}(s) = \frac{H_1(s)}{1 + H_1(s)H_2(s)}$ — classical feedback formula

These simple rules enable building complex systems from well-understood components.

Limitations of the LTI Model

Not everything is LTI:

• Modulators: $y(t) = x(t)\cos(\omega_c t)$ — linear but time-varying
• Compressors: Dynamic range compression is nonlinear (level-dependent gain)
• Quantizers: Inherently nonlinear (step-function mapping)
• Adaptive filters: Time-varying by design (coefficients change)

For these systems, more general analysis tools (time-varying systems theory, Volterra series for nonlinear systems) are needed.

Key Takeaways

• LTI systems satisfy both linearity (superposition) and time-invariance (shift-invariance)
• The impulse response $h(t)$ completely characterizes any LTI system
• Output is computed via convolution: $y = h * x$
• Sinusoids pass through unchanged in frequency — only amplitude and phase are modified
• Convolution in time equals multiplication in the frequency domain
• Transfer functions enable algebraic analysis of cascaded and feedback systems
• The LTI model is an excellent approximation for most engineering systems within their linear operating range