LSI Estimation Systems and Asymptotic Behavior

April 10, 2026 · Luciano Muratore

Having explored transforms like the DFT and DTFT, we now shift our focus to the systems that process these signals. This analysis centers on Linear Shift-Invariant (LSI) estimation systems and the rigorous mathematical conditions that govern their long-term behavior.

1. What is an LSI Estimation System?

A Linear Shift-Invariant (LSI) system is defined by two fundamental properties:

• Linearity: The principle of superposition holds: $T(au + bv) = aT(u) + bT(v)$.
• Shift-Invariance: A delay in the input results in an equal delay in the output. If $y[n] = T(u[n])$, then $y[n-k] = T(u[n-k])$.

These systems are entirely characterized by their impulse response $h[n]$. For any given input $u[n]$, the output is determined by the convolution of the input and the impulse response: $(T_{h}u)[n] = (h * u)[n] = \sum_{k=-\infty}^{\infty} u[n-k]h[k]$ Convolution essentially represents a weighted memory of all past inputs.

2. Stability and Norm Conditions

To analyze these systems, we categorize signals using $l^p$ spaces:

• $u \in l^p$: Signals with finite energy or magnitude.
• $h \in l^1$: An absolutely summable impulse response.

A key result for any LSI operator $T_h$ is that if $h \in l^1$ and $u \in l^p$ (for $1 \le p < \infty$), the operator is bounded on $l^p$: $\|T_{h}u\|_p \le \|h\|_1 \|u\|_p$ This is the mathematical definition of Bounded-Input Bounded-Output (BIBO) stability.

3. Asymptotic Behavior of the Output

The structure of stable systems dictates their behavior as time progresses. If an input signal $u$ belongs to an $l^p$ space, it must eventually vanish: $u[n] \to 0$ as $n \to \infty$.

For a stable LSI system ($h \in l^1$), the output will also vanish under these conditions: $(T_{h}u)[n] \to 0 \text{ as } n \to \infty$ This occurs because of the system’s finite memory: older inputs are increasingly suppressed by the summable impulse response $h[n]$.

4. Estimation Error Convergence

In estimation contexts, we define the estimation error $e[n]$ as the difference between the system’s output and a desired target signal $d[n]$: $e[n] = (T_{h}u)[n] - d[n]$

Convergence to zero ($e[n] \to 0$) occurs when:

• The target signal settles ($d \in l^p$ and $d[n] \to 0$).
• The estimator successfully approximates the mapping $u \to d$.

decaying to zero as n increases The estimation error typically fluctuates initially before converging to zero over time[cite: 706, 717].

5. Real-World Interpretation

These mathematical conditions translate directly into physical phenomena:

These principles form the foundation for several applications:

• Digital filters for smoothing time-series data.
• Wiener and Kalman filters for tracking stabilizing states.
• RC circuit responses to transients.
• Control systems with decaying dynamics.

Key Takeaways

• Stable LSI systems preserve the relationships between boundedness and finite norms.
• If the input signal dies out, the output must also eventually vanish.
• If both the input and the desired target signal vanish, the estimation error is guaranteed to converge to zero.
• These proofs provide the rigorous mathematical framework necessary for modern filtering and control algorithms.