ECE220 Unit2 - Subjective Questions

1

Define a Linear Time-Invariant (LTI) system. Explain the fundamental reasons why LTI systems are so widely studied and applied in Signal and Systems analysis.

An LTI system is a system that satisfies two key properties:

Linearity: A system is linear if it satisfies the superposition principle. This means that if an input produces an output and an input produces an output , then an input produces an output for any constants and .
- Mathematically: If $x_1(t) \rightarrow y_1(t)$ and $x_2(t) \rightarrow y_2(t)$ , then $a x_1(t) + b x_2(t) \rightarrow a y_1(t) + b y_2(t)$ .
Time-Invariance: A system is time-invariant if a time shift in the input signal results in an identical time shift in the output signal. That is, if produces , then produces for any time shift .
- Mathematically: If $x(t) \rightarrow y(t)$ , then $x(t - t_0) \rightarrow y(t - t_0)$ .

Reasons for widespread study and application of LTI systems:

Simplicity of Analysis: LTI systems are mathematically tractable. Their behavior is entirely characterized by their impulse response ( $h(t)$ for CT, $h[n]$ for DT). The output for any arbitrary input can be determined through convolution.
Convolution Property: The convolution operation is fundamental to LTI systems, providing a direct relationship between input, impulse response, and output: $y(t) = x(t) * h(t)$ or $y[n] = x[n] * h[n]$ .
Frequency Domain Analysis: In the frequency domain, convolution becomes multiplication. This greatly simplifies analysis and design using Fourier Transforms and Laplace Transforms, making concepts like transfer functions and frequency response very powerful.
Physical Realism: Many real-world physical systems, especially those operating around an equilibrium point, can be accurately approximated as LTI systems (e.g., RLC circuits, mechanical spring-mass-damper systems, acoustic systems).
Building Blocks: Complex systems can often be decomposed into simpler LTI subsystems connected in series, parallel, or feedback configurations, allowing for modular design and analysis.
Filter Design: The entire field of digital and analog filter design is largely based on LTI system theory, enabling precise control over signal spectral content.

2

Derive the convolution sum for a discrete-time LTI system, explaining how the linearity and time-invariance properties are used in the derivation. Assume the input signal is $x[n]$ and the impulse response is $h[n]$ .

3

Consider two discrete-time sequences: $x[n] = \{1, 2, 3\}$ for $n=0, 1, 2$ (and 0 otherwise) and $h[n] = \{1, 1\}$ for $n=0, 1$ (and 0 otherwise). Compute the convolution $y[n] = x[n] * h[n]$ using the convolution sum formula.

4

Derive the convolution integral for a continuous-time LTI system, explaining how the linearity and time-invariance properties are used in the derivation. Assume the input signal is $x(t)$ and the impulse response is $h(t)$ .

5

Consider a continuous-time LTI system with impulse response $h(t) = e^{-t}u(t)$ . Find the output $y(t)$ when the input is $x(t) = u(t)$ , where $u(t)$ is the unit step function.

6

Explain the following fundamental properties of LTI systems with respect to their impulse response $h[n]$ or $h(t)$ :

Memoryless
Causality
Stability

The properties of an LTI system are directly tied to its impulse response.

1. Memoryless System:

Definition: A system is memoryless if its output at any time $t$ (or $n$ ) depends only on the input at the same time $t$ (or $n$ ). It does not depend on past or future input values.
Condition for LTI Systems: For an LTI system to be memoryless, its impulse response must be zero for all (or ).
- Continuous-Time: $h(t) = K\delta(t)$ , where $K$ is a constant. The output is $y(t) = x(t) * K\delta(t) = Kx(t)$ .
- Discrete-Time: $h[n] = K\delta[n]$ , where $K$ is a constant. The output is $y[n] = x[n] * K\delta[n] = Kx[n]$ .
Interpretation: A memoryless LTI system is essentially a simple gain/attenuation block.

2. Causal System:

Definition: A system is causal if its output at any time $t$ (or $n$ ) depends only on the present and past values of the input, and not on future values.
Condition for LTI Systems: For an LTI system to be causal, its impulse response must be zero for all (or ).
- Continuous-Time: $h(t) = 0$ for $t < 0$ .
- Discrete-Time: $h[n] = 0$ for $n < 0$ .
Interpretation: This condition ensures that the system does not "know" about future inputs. Real-time physical systems must be causal.
- For causal CT LTI systems, the convolution integral becomes $y(t) = \int_{-\infty}^{t} x(\tau)h(t - \tau) d\tau$ .
- For causal DT LTI systems, the convolution sum becomes $y[n] = \sum_{k = -\infty}^{n} x[k]h[n-k]$ .

3. Stable System (BIBO Stability):

Definition: A system is stable in the BIBO (Bounded-Input, Bounded-Output) sense if every bounded input produces a bounded output. That is, if $|x(t)| \le B_x < \infty$ (or $|x[n]| \le B_x < \infty$ ) for all $t$ (or $n$ ), then $|y(t)| \le B_y < \infty$ (or $|y[n]| \le B_y < \infty$ ) for all $t$ (or $n$ ).
Condition for LTI Systems: For an LTI system to be BIBO stable, its impulse response must be absolutely integrable (for CT) or absolutely summable (for DT).
- Continuous-Time: $\int_{-\infty}^{\infty} |h(t)| dt < \infty$ .
- Discrete-Time: $\sum_{n = -\infty}^{\infty} |h[n]| < \infty$ .
Interpretation: This condition ensures that the system's output will not "blow up" or become infinite if the input signal remains finite. This is a crucial property for practical systems to operate reliably.

7

Discuss the commutative, associative, and distributive properties of convolution. How do these properties simplify the analysis and design of complex LTI systems?

Convolution is a fundamental operation for LTI systems, and it possesses several important algebraic properties that simplify system analysis:

1. Commutative Property:

Statement: The order of convolution does not matter. $x(t) * h(t) = h(t) * x(t)$ (for CT) or $x[n] * h[n] = h[n] * x[n]$ (for DT).
Mathematical Proof (CT example):
$y(t) = \int_{-\infty}^{\infty} x(\tau)h(t - \tau) d\tau$
Let $p = t - \tau$ , so $d\tau = -dp$ . When $\tau = -\infty$ , $p = \infty$ . When $\tau = \infty$ , $p = -\infty$ . (Note: limits invert due to $d\tau = -dp$ )
$y(t) = \int_{\infty}^{-\infty} x(t - p)h(p) (-dp) = \int_{-\infty}^{\infty} h(p)x(t - p) dp$
This is the definition of $h(t) * x(t)$ .
Implication: The input and impulse response can be interchanged without changing the output. This is useful for computational efficiency (convolving a short signal with a long signal can be done by flipping the roles) and for visualizing the convolution process.

2. Associative Property:

Statement: Convolution is associative. $(x(t) * h_1(t)) * h_2(t) = x(t) * (h_1(t) * h_2(t))$ (for CT) or $(x[n] * h_1[n]) * h_2[n] = x[n] * (h_1[n] * h_2[n])$ (for DT).
Implication: When multiple LTI systems are cascaded (connected in series), the overall impulse response is the convolution of individual impulse responses: $h_{eq}(t) = h_1(t) * h_2(t)$ . The order of the systems in a cascade does not affect the overall output. This greatly simplifies the analysis of complex systems by allowing us to represent a cascade of LTI systems as a single equivalent LTI system.

3. Distributive Property:

Statement: Convolution distributes over addition. $x(t) * (h_1(t) + h_2(t)) = (x(t) * h_1(t)) + (x(t) * h_2(t))$ (for CT) or $x[n] * (h_1[n] + h_2[n]) = (x[n] * h_1[n]) + (x[n] * h_2[n])$ (for DT).
Implication: If multiple LTI systems are connected in parallel, the overall impulse response is the sum of their individual impulse responses: $h_{eq}(t) = h_1(t) + h_2(t)$ . This means that the output of a parallel combination of LTI systems is the sum of the outputs that would be obtained if the input were applied to each system individually. This simplifies the analysis of systems where signals are processed through multiple paths simultaneously.

Simplification in Analysis and Design:
These properties are crucial because they allow us to:

Simplify System Block Diagrams: Cascaded and parallel LTI systems can be reduced to a single equivalent LTI system described by a single impulse response, making complex diagrams much simpler.
Rearrange System Components: The commutative and associative properties allow for flexibility in arranging system components without changing the overall system behavior, which can be useful for optimization or implementation constraints.
Modular Design: Design complex systems by breaking them down into simpler, well-understood LTI blocks, then combining their impulse responses through convolution or summation.
Frequency Domain Insight: In the frequency domain, these properties translate to simpler algebraic operations: convolution becomes multiplication for cascade and addition for parallel, making system analysis and filter design more intuitive and efficient.

8

Explain the concept of invertibility for an LTI system. If an LTI system has an impulse response $h(t)$ , what is the condition for it to be invertible, and how is the impulse response of its inverse system, $h_{inv}(t)$ , related to $h(t)$ ?

Invertibility of an LTI System:

An LTI system is said to be invertible if there exists another LTI system, called the inverse system, which, when cascaded with the original system, produces an output that is identical to the original input. In essence, the inverse system "undoes" the operation of the original system.

If we denote the original system by $S$ and its inverse system by $S_{inv}$ , then if $y(t) = S\{x(t)\}$ , then $x(t) = S_{inv}\{y(t)\}$ . When cascaded, the overall system is the identity system.

Condition for Invertibility and Relationship to Impulse Response:

Let the impulse response of the original LTI system be $h(t)$ (or $h[n]$ for DT) and the impulse response of its inverse system be $h_{inv}(t)$ (or $h_{inv}[n]$ for DT).

When the original system and its inverse system are cascaded, their combined impulse response must be the unit impulse function.

Continuous-Time: $h(t) * h_{inv}(t) = \delta(t)$
Discrete-Time: $h[n] * h_{inv}[n] = \delta[n]$

This is the fundamental condition for invertibility. If such an $h_{inv}(t)$ or $h_{inv}[n]$ exists, then the system is invertible.

Practical Implications and Examples:

Frequency Domain: The condition $h(t) * h_{inv}(t) = \delta(t)$ translates very simply in the frequency domain. Taking the Fourier Transform (or Laplace/Z-Transform):
- For CT: $H(j\omega) H_{inv}(j\omega) = 1 \implies H_{inv}(j\omega) = \frac{1}{H(j\omega)}$
- For DT: $H(e^{j\omega}) H_{inv}(e^{j\omega}) = 1 \implies H_{inv}(e^{j\omega}) = \frac{1}{H(e^{j\omega})}$
  This means that for an LTI system to be invertible, its frequency response (or system function) must not be zero for any frequency where the inverse system is expected to operate. If $H(j\omega)$ is zero at some frequency, then $H_{inv}(j\omega)$ would need to be infinite, which is generally not physically realizable.
Example of an invertible system: A simple gain system $y(t) = Kx(t)$ (where $K \ne 0$ ) has $h(t) = K\delta(t)$ . Its inverse system would have $h_{inv}(t) = (1/K)\delta(t)$ , since $K\delta(t) * (1/K)\delta(t) = \delta(t)$ . In the frequency domain, $H(j\omega) = K$ , so $H_{inv}(j\omega) = 1/K$ .
Example of a non-invertible system: A system that outputs zero for any input ( $y(t)=0$ ) is non-invertible because information about the input is permanently lost. Another common non-invertible system is an ideal low-pass filter (LPF) that completely blocks high frequencies. Once high-frequency components are removed, they cannot be recovered by any inverse system, as $H(j\omega)$ would be zero beyond the cutoff frequency, making $1/H(j\omega)$ undefined.

9

Differentiate between continuous-time (CT) and discrete-time (DT) LTI systems in terms of their input/output signals, mathematical description, and methods of analysis. Provide an example of where each type of system would be used.

The fundamental difference between CT and DT LTI systems lies in the nature of the signals they process and the mathematical tools used to describe and analyze them.

1. Input/Output Signals:

Continuous-Time LTI (CT-LTI) Systems:
- Input/Output signals $x(t)$ and $y(t)$ are defined for all real values of time $t$ . They are functions of a continuous variable.
- Signals can take on a continuous range of values.
Discrete-Time LTI (DT-LTI) Systems:
- Input/Output signals $x[n]$ and $y[n]$ are defined only for integer values of time $n$ . They are sequences of numbers.
- Signals can take on a continuous range of values (if analog-to-digital conversion involves only sampling, not quantization) or discrete values (if quantized).

2. Mathematical Description:

CT-LTI Systems:
- Described by convolution integral: $y(t) = x(t) * h(t) = \int_{-\infty}^{\infty} x(\tau)h(t - \tau) d\tau$ .
- Can also be described by linear constant-coefficient differential equations.
- Analysis often uses Fourier Transform ( $X(j\omega)$ ) and Laplace Transform ( $X(s)$ ).
DT-LTI Systems:
- Described by convolution sum: $y[n] = x[n] * h[n] = \sum_{k = -\infty}^{\infty} x[k]h[n-k]$ .
- Can also be described by linear constant-coefficient difference equations.
- Analysis often uses Discrete-Time Fourier Transform ( $X(e^{j\omega})$ ) and Z-Transform ( $X(z)$ ).

3. Methods of Analysis:

CT-LTI Systems:
- Calculus-based methods (integration, differentiation) are central.
- Frequency domain analysis involves continuous spectra.
- System functions are rational functions of $s$ .
DT-LTI Systems:
- Summation-based methods are central.
- Frequency domain analysis involves periodic spectra.
- System functions are rational functions of $z$ .

Examples of Usage:

CT-LTI System Example: Analog Audio Amplifier. An audio amplifier takes a continuous-time analog electrical signal from a microphone (input) and produces an amplified continuous-time analog electrical signal (output) for speakers. Ideally, it's designed to operate linearly (no distortion) and without altering the timing relationships of the signal components (time-invariant) within its operational range. It's fundamentally described by differential equations relating voltage and current in its circuit components.
DT-LTI System Example: Digital Audio Equalizer. A digital equalizer takes a discrete-time sequence of audio samples (input, obtained from an ADC) and applies digital filtering operations (e.g., boosting bass or treble) to produce a modified discrete-time sequence of audio samples (output, to be sent to a DAC). This process is implemented using difference equations, and the filtering is performed via discrete convolution or equivalent frequency-domain multiplication using DFT/FFT.

10

Explain how the unit impulse response $h(t)$ (or $h[n]$ ) completely characterizes an LTI system. Why is this not generally true for non-LTI systems?

For an LTI system, the unit impulse response $h(t)$ (for continuous-time) or $h[n]$ (for discrete-time) completely characterizes the system. This means that if you know $h(t)$ or $h[n]$ , you can determine the output $y(t)$ or $y[n]$ for any arbitrary input $x(t)$ or $x[n]$ using the convolution integral or sum.

How it works for LTI Systems:

Input Representation: Any arbitrary input signal $x(t)$ can be expressed as a superposition of scaled and shifted impulse functions:
- Continuous-Time: $x(t) = \int_{-\infty}^{\infty} x(\tau)\delta(t - \tau) d\tau$
- Discrete-Time: $x[n] = \sum_{k = -\infty}^{\infty} x[k]\delta[n-k]$
Applying Linearity: Due to the linearity property, the system's response to the sum (or integral) of inputs is the sum (or integral) of its responses to each individual scaled impulse. The scaling factors ( $x(\tau)$ or $x[k]$ ) can be pulled out.
- Continuous-Time: $y(t) = \int_{-\infty}^{\infty} x(\tau) T\{ \delta(t - \tau) \} d\tau$
- Discrete-Time: $y[n] = \sum_{k = -\infty}^{\infty} x[k] T\{ \delta[n-k] \}$
Applying Time-Invariance: Due to the time-invariance property, the response to a shifted impulse $\delta(t - \tau)$ (or $\delta[n-k]$ ) is simply the shifted impulse response $h(t - \tau)$ (or $h[n-k]$ ). Recall that $h(t) = T\{ \delta(t) \}$ and $h[n] = T\{ \delta[n] \}$ .
Resulting Convolution: Substituting this back, we get the convolution relationship:
- Continuous-Time: $y(t) = \int_{-\infty}^{\infty} x(\tau)h(t - \tau) d\tau = x(t) * h(t)$
- Discrete-Time: $y[n] = \sum_{k = -\infty}^{\infty} x[k]h[n-k] = x[n] * h[n]$

This shows that knowing only $h(t)$ or $h[n]$ is sufficient to determine the output for any input, confirming that the impulse response completely characterizes the LTI system.

Why this is NOT generally true for Non-LTI Systems:

The fundamental reason this doesn't hold for non-LTI systems is that one or both of the linearity and time-invariance properties are violated. If a system is not LTI:

Non-linear Systems: If the system is non-linear, the superposition principle does not apply. The response to a sum of inputs is not the sum of individual responses. Therefore, you cannot simply sum the responses to individual impulses to get the response to a complex input. Knowing $T\{ \delta(t) \}$ (the "impulse response") tells you nothing about $T\{ 2\delta(t) \}$ or $T\{ \delta(t) + \delta(t-1) \}$ in a predictable way. For example, a squaring system $y(t) = x^2(t)$ has an "impulse response" of $\delta^2(t)$ , which is usually defined as $\delta(t)$ . But for input $2\delta(t)$ , the output is $(2\delta(t))^2 = 4\delta(t)$ , not $2\delta(t)$ .
Time-Varying Systems: If the system is time-varying, its response to an impulse changes depending on when that impulse occurs. So, $T\{ \delta(t - \tau) \}$ is not simply $h(t - \tau)$ . It might be some $h(t, \tau)$ , meaning the impulse response itself is a function of both the observation time and the impulse application time. In this case, a single $h(t)$ is insufficient to characterize the system, as the system's behavior changes over time. For example, a system $y(t) = t \cdot x(t)$ (a time-varying gain) has $T\{ \delta(t) \} = t\delta(t) = 0$ . But $T\{ \delta(t-1) \} = t\delta(t-1) = \delta(t-1)$ , not $t\delta(t-1)$ . A single impulse response cannot capture this changing behavior.

11

What are the eigenfunctions of LTI systems? Explain why they are significant in system analysis and how they relate to the frequency response.

Eigenfunctions of LTI Systems:

The eigenfunctions of an LTI system are the complex exponentials of the form $e^{st}$ for continuous-time (CT) systems and $z^n$ or $e^{j\omega n}$ for discrete-time (DT) systems. When such a signal is input to an LTI system, the output is the same complex exponential, simply scaled by a complex constant (the eigenvalue).

Continuous-Time LTI System: If $x(t) = e^{st}$ is the input to an LTI system with impulse response $h(t)$ , the output is:
$y(t) = h(t) * e^{st} = \int_{-\infty}^{\infty} h(\tau)e^{s(t - \tau)} d\tau$
$y(t) = \int_{-\infty}^{\infty} h(\tau)e^{st}e^{-s\tau} d\tau = e^{st} \int_{-\infty}^{\infty} h(\tau)e^{-s\tau} d\tau$
The integral term is the Laplace Transform of $h(t)$ , denoted as $H(s)$ .
Thus, $y(t) = H(s)e^{st}$ .
Here, $e^{st}$ is the eigenfunction and $H(s)$ is the corresponding eigenvalue.
Discrete-Time LTI System: If $x[n] = z^n$ is the input to an LTI system with impulse response $h[n]$ , the output is:
$y[n] = h[n] * z^n = \sum_{k = -\infty}^{\infty} h[k]z^{(n - k)}$
$y[n] = \sum_{k = -\infty}^{\infty} h[k]z^n z^{-k} = z^n \sum_{k = -\infty}^{\infty} h[k]z^{-k}$
The summation term is the Z-Transform of $h[n]$ , denoted as $H(z)$ .
Thus, $y[n] = H(z)z^n$ .
Here, $z^n$ is the eigenfunction and $H(z)$ is the corresponding eigenvalue.

Significance in System Analysis:

Simplification of Analysis: The most significant advantage is that LTI systems simply scale complex exponential inputs. This transforms a convolution operation (complex in the time domain) into a simple multiplication (in the frequency/transform domain). This vastly simplifies analysis, as we don't need to re-evaluate the system's effect on each frequency component; we just multiply by the system's 'gain' at that frequency.
Foundation of Transform Analysis: This property is the very basis for the Fourier Transform, Laplace Transform, and Z-Transform. These transforms decompose arbitrary signals into their complex exponential components. When these components pass through an LTI system, they are processed independently and simply scaled. The output is then the sum of these scaled exponential components.
Frequency Response: The eigenvalues $H(j\omega)$ (for $s=j\omega$ ) or $H(e^{j\omega})$ (for $z=e^{j\omega}$ ) are precisely the frequency response of the LTI system. This tells us how the system affects the magnitude and phase of different sinusoidal components present in the input signal. For example, if $x(t) = A\cos(\omega_0 t + \phi)$ , the output is a sinusoid of the same frequency $\omega_0$ , but with modified magnitude $A|H(j\omega_0)|$ and phase $\phi + \angle H(j\omega_0)$ .
Filter Design: Understanding the frequency response (eigenvalues) is critical for designing filters. We can design systems that pass certain frequencies (e.g., low-pass filters) and attenuate others by shaping $H(j\omega)$ or $H(e^{j\omega})$ .

In summary, the eigenfunction property allows us to move from the time domain (where convolution is complex) to the frequency domain (where multiplication is simple) to understand and design LTI systems more effectively.

12

Describe the relationship between the impulse response of a causal LTI system and its description by linear constant-coefficient differential equations (for CT) or difference equations (for DT).

13

Discuss the significance of initial conditions for LTI systems described by differential or difference equations. How do they impact the system's response, and under what circumstances can they be ignored when finding the impulse response?

Initial conditions are crucial for systems described by differential or difference equations because they specify the system's state before an external input signal begins. For LTI systems, they directly influence the complete response of the system.

Significance and Impact on System's Response:

Complete Response: The total response $y(t)$ (or $y[n]$ ) of a system described by differential/difference equations can be broken into two parts:
- Zero-Input Response (ZIR): This is the system's response due to the initial conditions alone, with the input $x(t)$ (or $x[n]$ ) set to zero.
- Zero-State Response (ZSR): This is the system's response due to the input $x(t)$ (or $x[n]$ ) alone, with all initial conditions set to zero. For LTI systems, the ZSR is precisely the convolution of the input with the impulse response: $y_{zs}(t) = x(t) * h(t)$ .
  The total response is the sum of these two: $y(t) = y_{zi}(t) + y_{zs}(t)$ .
Unique Solutions: Initial conditions are necessary to obtain a unique solution for the differential or difference equation. Without them, the homogeneous solution (the natural response) would have arbitrary constants. For instance, an $N$ -th order differential equation requires $N$ initial conditions (e.g., $y(0), y'(0), \dots, y^{(N-1)}(0)$ ) for a unique solution.
System Memory: Initial conditions represent the "memory" of the system from past inputs or events. Even if the current input is zero, a system with non-zero initial conditions will continue to produce an output based on its prior state.
Causality and Rest Condition: For a system to be considered at "rest" (a common assumption for LTI system analysis, especially when deriving impulse response), it implies that all initial conditions are zero at time $t=0^-$ (or $n=0^-$ ), meaning the system has no stored energy or history prior to the input's application. This is a crucial aspect for defining LTI systems purely by convolution.

When Initial Conditions Can Be Ignored When Finding the Impulse Response:

Initial conditions are inherently ignored (or set to zero) when specifically finding the impulse response $h(t)$ (or $h[n]$ ) of an LTI system. This is because:

Definition of Impulse Response: The impulse response $h(t)$ is defined as the output of the LTI system when the input is a unit impulse $\delta(t)$ (or $\delta[n]$ ), and the system is initially at rest. Being initially at rest implies that all initial conditions are zero at $t=0^-$ (or $n=0^-$ ).
Focus on System's Intrinsic Behavior: The impulse response is a fundamental characteristic of the system itself, reflecting its intrinsic dynamic behavior, independent of any prior excitation. It captures how the system transforms an input, assuming a clean slate beforehand.
Derivation of ZSR: The primary utility of the impulse response is to compute the zero-state response (ZSR) via convolution. If initial conditions were included in $h(t)$ , the convolution sum/integral would yield the total response, not just the ZSR, making it difficult to separate the effects of the input from the initial state.

Therefore, when solving for $h(t)$ or $h[n]$ for a given differential/difference equation, one always assumes zero initial conditions for the output at $t=0^-$ or $n=0^-$ . The impulse input $\delta(t)$ or $\delta[n]$ then acts as the sole excitation to determine the characteristic response.

14

Describe the general procedure for numerically simulating discrete-time convolution using a programming language like Python or MATLAB. Include considerations for input signal representation and computational efficiency.

Numerically simulating discrete-time convolution involves implementing the convolution sum formula using arrays or vectors in a programming environment.

Convolution Sum Formula:
$y[n] = \sum_{k = -\infty}^{\infty} x[k]h[n-k]$

General Procedure:

Represent Input Signals:
- Input signal $x[n]$ and impulse response $h[n]$ are represented as one-dimensional arrays (e.g., NumPy arrays in Python, vectors in MATLAB).
- It's crucial to correctly handle the starting indices. If the arrays start at index 0, they implicitly assume $n=0$ for the first element. If signals are non-causal or start at other indices, this needs careful management (e.g., by keeping track of the starting index of each array or zero-padding).
Determine Output Length and Index Range:
- If $x[n]$ has length $L_x$ and $h[n]$ has length $L_h$ , the resulting convolved signal $y[n]$ will have length $L_y = L_x + L_h - 1$ .
- If $x[n]$ starts at $n_{x,start}$ and $h[n]$ starts at $n_{h,start}$ , then $y[n]$ will start at $n_{y,start} = n_{x,start} + n_{h,start}$ . This is important for correctly aligning the output indices.
Initialize Output Array:
- Create an array y of length $L_y$ , initialized with zeros.
Implement the Convolution Loop:
The convolution sum can be implemented using nested loops:
python

Assuming x and h are 0-indexed arrays

Lx = len(x)
Lh = len(h)
Ly = Lx + Lh - 1
y = [0.0] * Ly # Initialize with zeros

for n in range(Ly): # Loop through each output sample index 'n'
for k in range(Lx): # Loop through each input sample index 'k'
if 0 <= (n - k) < Lh: # Check if h[n-k] is within bounds of h
y[n] += x[k] * h[n - k]

Considerations for Computational Efficiency:

Direct Convolution Complexity: The direct implementation described above has a time complexity of $O(L_x \cdot L_h)$ . For very long signals, this can be computationally expensive.
Fast Fourier Transform (FFT) based Convolution: For long sequences, convolution can be performed much more efficiently in the frequency domain using the Fast Fourier Transform (FFT).
- Steps for FFT-based Convolution:
  a. Pad both $x[n]$ and $h[n]$ with zeros to a length $N \ge L_x + L_h - 1$ , typically the next power of 2 for FFT efficiency.
  b. Compute the DFT (using FFT) of the zero-padded $x[n]$ to get $X[k]$ .
  c. Compute the DFT (using FFT) of the zero-padded $h[n]$ to get $H[k]$ .
  d. Multiply the frequency domain representations element-wise: $Y[k] = X[k] \cdot H[k]$ .
  e. Compute the Inverse DFT (using IFFT) of $Y[k]$ to get $y[n]$ .
- Complexity: FFT-based convolution has a complexity of $O(N \log N)$ , which is significantly faster than $O(L_x L_h)$ for large $L_x, L_h$ .
- Most signal processing libraries (like scipy.signal.convolve in Python or conv in MATLAB) use FFT-based methods internally for efficiency when applicable.
Overlap-Add/Overlap-Save Methods: For extremely long signals that cannot fit into memory for a single FFT, these block-processing techniques are used. They break the input into smaller segments, convolve each segment, and then combine the results, often using FFT for the segment convolutions.

By understanding these considerations, one can choose the most appropriate method for simulating convolution based on signal length and performance requirements.

15

Explain the concept of correlation between two signals, $x[n]$ and $y[n]$ . How is it related to convolution, and what are its primary applications?

Correlation (Cross-correlation):

Correlation is a measure of similarity between two signals as a function of the time shift (or lag) applied to one of them. It tells us how much two signals match at different offsets. High correlation at a specific lag indicates strong similarity between the signals when one is shifted by that lag relative to the other.

Discrete-Time Cross-correlation: The cross-correlation of $x[n]$ and $y[n]$ is defined as:
$r_{xy}[l] = \sum_{n = -\infty}^{\infty} x[n]y[n-l]$
where $l$ is the lag or time shift.
Continuous-Time Cross-correlation: The cross-correlation of $x(t)$ and $y(t)$ is defined as:
$r_{xy}(\tau) = \int_{-\infty}^{\infty} x(t)y(t-\tau) dt$
where $\tau$ is the lag or time shift.

Autocorrelation:
If $x[n]$ is correlated with itself, it's called autocorrelation:
$r_{xx}[l] = \sum_{n = -\infty}^{\infty} x[n]x[n-l]$
Autocorrelation measures the similarity of a signal with a delayed version of itself. It is often used to find repeating patterns or periodicities in a signal.

Relationship to Convolution:

Correlation is very closely related to convolution. Specifically, cross-correlation can be expressed in terms of convolution:

Discrete-Time: $r_{xy}[l] = \sum_{n = -\infty}^{\infty} x[n]y[n-l]$
Recall convolution: $x[n] * h[n] = \sum_{k = -\infty}^{\infty} x[k]h[n-k]$
If we let $h[n] = y[-n]$ (time-reversed version of $y[n]$ ), then $h[n-k] = y[-(n-k)] = y[k-n]$ . This is not quite it.
However, $r_{xy}[l]$ is equivalent to the convolution of $x[n]$ with the time-reversed version of $y[n]$ :
$r_{xy}[l] = x[l] * y[-l]$
And similarly, if we define $y^*[n]$ as the complex conjugate of $y[n]$ :
$r_{xy}[l] = x[l] * y^*[-l]$ (for complex signals, but often $y[n]$ is real, so $y^*[n]=y[n]$ )
Continuous-Time: Similarly, for continuous-time signals:
$r_{xy}(\tau) = x(\tau) * y(-\tau)$

Essentially, correlation is a convolution where one of the signals has been time-reversed.

Primary Applications:

Time Delay Estimation: By finding the lag $l$ (or $\tau$ ) at which the cross-correlation peaks, one can estimate the time difference between two signals, useful in radar, sonar, geophysics, and audio processing (e.g., source localization).
Pattern Recognition/Template Matching: Identifying the occurrence of a known pattern (template) within a longer signal. For example, detecting a specific word in a speech recording, or a known image within a larger image.
Signal Detection: Detecting a weak signal embedded in noise. The correlation with a replica of the desired signal can enhance the signal while suppressing uncorrelated noise.
System Identification: Determining the impulse response of an unknown LTI system by correlating its input and output.
Periodicity Detection (Autocorrelation): Autocorrelation can reveal periodic components in a signal, even if they are obscured by noise. The peaks in the autocorrelation function correspond to the periods of the signal (e.g., finding the pitch of a speech signal).
Deconvolution: In some applications, if you know the characteristics of a distorting channel (e.g., its impulse response), you can use a form of inverse filtering (often related to correlation) to recover the original signal.

16

Discuss the challenges and practical considerations when simulating continuous-time convolution in a discrete software environment. How are these challenges typically addressed?

Simulating continuous-time (CT) convolution, which involves integrals, in a discrete software environment (like Python or MATLAB) presents several challenges because computers inherently deal with discrete data.

Challenges:

Discretization/Sampling: CT signals must be sampled to convert them into discrete-time sequences. This introduces sampling artifacts (e.g., aliasing) if the sampling rate is not high enough (Nyquist rate).
Approximation of Integration: The convolution integral $\int x(\tau)h(t-\tau)d\tau$ must be approximated by a sum. This effectively turns CT convolution into DT convolution, often using numerical integration methods (e.g., Riemann sums, trapezoidal rule).
Infinite Limits: CT convolution integral has infinite limits. Real-world signals and impulse responses are usually finite in duration or decay to near zero, but conceptually, this needs to be handled by defining a practical finite duration for the simulation.
Computational Resources: While DT convolution can be performed efficiently using FFT, ensuring that the discrete approximation accurately reflects the continuous counterpart requires a high sampling rate, leading to large arrays and increased computational time and memory.
Numerical Stability and Accuracy: The choice of sampling rate and numerical integration method impacts the accuracy of the simulated output. Too low a sampling rate can lead to significant errors.

How Challenges are Typically Addressed:

Sufficient Sampling Rate: The most critical step is to sample the CT signals $x(t)$ and $h(t)$ at a rate significantly higher than twice their highest frequency component (Nyquist rate). A common practice is to use an oversampling factor to minimize aliasing and improve approximation accuracy.
- $x_s[n] = x(n T_s)$ and $h_s[n] = h(n T_s)$ , where $T_s = 1/f_s$ is the sampling period.
Conversion to Discrete Convolution: Once sampled, the CT convolution integral is approximated by a DT convolution sum. The scaling factor $T_s$ from the integral approximation is often absorbed or applied:
$y(t) = \int_{-\infty}^{\infty} x(\tau)h(t - \tau) d\tau \approx \sum_{k=-\infty}^{\infty} x(k T_s)h(t - k T_s) T_s$
If we let $t = n T_s$ , then $y[n] = y(n T_s) \approx T_s \sum_{k} x[k]h[n-k]$ .
So, the discrete convolution $y_{DT}[n] = x_{DT}[n] * h_{DT}[n]$ (computed by conv function) will be scaled by $T_s$ (or equivalent) to approximate the CT output.
Zero-Padding and Time Alignment: Practical signals have finite duration. Signals might be padded with zeros to ensure that the entire convolution result is captured and to facilitate FFT-based convolution. Proper alignment of time indices must be maintained if the original CT signals are not causal or start at arbitrary times.
FFT-based Convolution: For computational efficiency, the resulting discrete convolution is almost always performed using the FFT algorithm, especially for longer signals. This involves padding the discrete sequences to a power-of-two length before performing FFT, element-wise multiplication in the frequency domain, and then IFFT.
Anti-Aliasing Filters: Before sampling, a low-pass anti-aliasing filter is often applied to the continuous-time signals to remove frequency components above half the sampling rate, preventing aliasing distortion in the discrete representation.

In essence, continuous-time convolution simulation boils down to a carefully handled discretization of the signals, approximation of the integral using a sum, and then performing the discrete convolution efficiently, usually via FFT, while being mindful of sampling effects.

17

What are the key differences between the step response and the impulse response of an LTI system? How can one be derived from the other?

18

Explain what is meant by a "memoryless" LTI system. Give an example of such a system and discuss the implications for its impulse response.

19

What does it mean for an LTI system to be causal? Why is causality a crucial property for real-time physical systems, and how is it reflected in the convolution sum/integral limits?

Causality in LTI Systems:

An LTI system is said to be causal if its output at any given time depends only on the present and past values of the input. It does not depend on future values of the input. This means the system cannot predict the future.

Crucial for Real-Time Physical Systems:

Causality is a fundamental requirement for any real-time physical system (e.g., electronic circuits, mechanical systems, biological processes) for the following reasons:

Physical Realizability: In the real world, cause must precede effect. A system cannot react to an event that has not yet occurred. If a system were non-causal, it would imply it could predict future inputs, which is impossible in real-time.
Stability and Predictability: Non-causal systems are often physically unstable or unrealizable. Causality ensures that the system's response is governed by events that have already happened, leading to predictable and controllable behavior.

Reflection in Impulse Response:

For an LTI system to be causal, its impulse response $h(t)$ (or $h[n]$ ) must be zero for all negative time values. That is:

Continuous-Time: $h(t) = 0$ for $t < 0$ .
Discrete-Time: $h[n] = 0$ for $n < 0$ .

Reflection in Convolution Sum/Integral Limits:

This condition on the impulse response directly modifies the limits of the convolution operation:

Continuous-Time Convolution Integral: The general integral is $y(t) = \int_{-\infty}^{\infty} x(\tau)h(t - \tau) d\tau$ .
If $h(t - \tau) = 0$ for $t - \tau < 0$ (i.e., for $\tau > t$ ), the upper limit of the integral changes from $\infty$ to $t$ :
$y(t) = \int_{-\infty}^{t} x(\tau)h(t - \tau) d\tau$
This form clearly shows that the output at time $t$ depends only on input values $x(\tau)$ for $\tau \le t$ (present and past inputs).
Discrete-Time Convolution Sum: The general sum is $y[n] = \sum_{k = -\infty}^{\infty} x[k]h[n-k]$ .
If $h[n-k] = 0$ for $n-k < 0$ (i.e., for $k > n$ ), the upper limit of the sum changes from $\infty$ to $n$ :
$y[n] = \sum_{k = -\infty}^{n} x[k]h[n-k]$
This form clearly shows that the output at time $n$ depends only on input values $x[k]$ for $k \le n$ (present and past inputs).

In both cases, if the input $x(t)$ (or $x[n]$ ) is also causal (i.e., $x(t)=0$ for $t<0$ or $x[n]=0$ for $n<0$ ), then the lower limit of the integral/sum can further be changed from $-\infty$ to $0$.

20

Explain the concept of BIBO (Bounded-Input Bounded-Output) stability for LTI systems. What is the necessary and sufficient condition for BIBO stability in terms of the system's impulse response for both CT and DT cases?

21

What is the relationship between the poles of the system function $H(s)$ (for CT) or $H(z)$ (for DT) and the stability of an LTI system? Be specific about the location of poles for a stable system.

The locations of the poles of an LTI system's transfer function (system function) are fundamental to determining its stability.

1. Continuous-Time (CT) LTI Systems and $H(s)$ :

For a CT LTI system described by a rational system function $H(s) = \frac{P(s)}{Q(s)}$ (where $P(s)$ and $Q(s)$ are polynomials in $s$ ), the poles are the roots of the denominator polynomial $Q(s) = 0$ . The impulse response $h(t)$ is the inverse Laplace Transform of $H(s)$ .

Condition for BIBO Stability: A CT LTI system is BIBO stable if and only if all its poles lie in the open left-half of the s-plane. That is, if $s_k$ is a pole, then $Re\{s_k\} < 0$ .
Region of Convergence (ROC): For a causal LTI system, its ROC for $H(s)$ must be to the right of the rightmost pole. For a stable causal LTI system, the ROC must include the $j\omega$ -axis (i.e., $Re\{s\} = 0$ ). This is only possible if the rightmost pole has $Re\{s_k\} < 0$ . If there is a pole on the $j\omega$ -axis or in the right-half plane, the ROC cannot include the $j\omega$ -axis, and the system is unstable.
Implication: Poles in the right-half plane lead to exponential growth in $h(t)$ , making it non-absolutely integrable. Poles on the $j\omega$ -axis (e.g., $s=j\omega_0$ ) lead to sustained oscillations (like $\cos(\omega_0 t)$ ) in $h(t)$ , which are not absolutely integrable, thus rendering the system unstable.

2. Discrete-Time (DT) LTI Systems and $H(z)$ :

For a DT LTI system described by a rational system function $H(z) = \frac{P(z)}{Q(z)}$ (where $P(z)$ and $Q(z)$ are polynomials in $z$ ), the poles are the roots of the denominator polynomial $Q(z) = 0$ . The impulse response $h[n]$ is the inverse Z-Transform of $H(z)$ .

Condition for BIBO Stability: A DT LTI system is BIBO stable if and only if all its poles lie strictly inside the unit circle in the z-plane. That is, if $z_k$ is a pole, then $|z_k| < 1$ .
Region of Convergence (ROC): For a causal LTI system, its ROC for $H(z)$ must be the exterior of a circle, outside the outermost pole. For a stable causal LTI system, the ROC must include the unit circle (i.e., $|z|=1$ ). This is only possible if the outermost pole has magnitude $|z_k| < 1$ . If there is a pole on or outside the unit circle, the ROC cannot include the unit circle, and the system is unstable.
Implication: Poles outside the unit circle lead to exponentially growing terms in $h[n]$ , making it non-absolutely summable. Poles on the unit circle (e.g., $z=e^{j\theta}$ ) lead to sustained oscillations in $h[n]$ (like $\cos(\theta n)$ ), which are not absolutely summable, thus rendering the system unstable.

In summary, the location of poles is a direct indicator of stability. For both CT and DT LTI systems, a bounded impulse response (absolutely integrable/summable) is required for BIBO stability, and this condition is met only when all poles are in the designated stable region (left-half plane for CT, inside the unit circle for DT).

22

Compare and contrast the behavior of LTI systems described by differential equations with those described by difference equations. Highlight similarities and differences in their analysis approaches.

LTI systems described by differential equations (Continuous-Time) and difference equations (Discrete-Time) share fundamental principles due to linearity and time-invariance, but differ in their domain and the mathematical tools used for their analysis.

Similarities:

LTI Properties: Both types of systems exhibit linearity and time-invariance. This means their behavior is completely characterized by an impulse response ( $h(t)$ or $h[n]$ ) and their output for any input is found via convolution ( $x(t)*h(t)$ or $x[n]*h[n]$ ).
Order of System: Both are classified by the order of the equation, corresponding to the number of energy storage elements or memory units. An $N$ -th order differential/difference equation typically implies $N$ poles in the system function.
Homogeneous and Particular Solutions: The general solution to both types of equations consists of a homogeneous solution (natural response, determined by system poles and initial conditions) and a particular solution (forced response, determined by the input).
Transform Domain Analysis: Both benefit immensely from transform-domain analysis (Laplace Transform for CT, Z-Transform for DT), where differential/difference equations convert to algebraic equations, and convolution becomes multiplication.
Causality and Stability: The concepts of causality (impulse response is zero for negative time) and BIBO stability (poles in specific regions) apply to both, although the specific regions differ.

Differences:

Feature	Continuous-Time LTI (Differential Equations)	Discrete-Time LTI (Difference Equations)
Signals	Continuous variables $x(t), y(t)$ , defined for all real $t$ .	Discrete sequences $x[n], y[n]$ , defined for integer $n$ .
Description	Linear Constant-Coefficient Differential Equations: $\sum a_k \frac{d^k y(t)}{dt^k} = \sum b_k \frac{d^k x(t)}{dt^k}$	Linear Constant-Coefficient Difference Equations: $\sum a_k y[n-k] = \sum b_k x[n-k]$
Fundamental Operation	Convolution Integral: $y(t) = \int x(\tau)h(t-\tau) d\tau$	Convolution Sum: $y[n] = \sum x[k]h[n-k]$
Transform Domain	Laplace Transform ( $s$ -domain) and Fourier Transform ( $j\omega$ -domain). $H(s)$	Z-Transform ( $z$ -domain) and DTFT ( $e^{j\omega}$ -domain). $H(z)$
Differentiation/Delay	Operators are $\frac{d}{dt}$ (in time domain) or $s$ (in Laplace domain).	Operators are time shifts $y[n-k]$ (in time domain) or $z^{-k}$ (in Z-domain).
Stability Region	All poles must be in the open left-half of the $s$ -plane ( $Re\{s\} < 0$ ).	All poles must be strictly inside the unit circle in the $z$ -plane ( $\|z\| < 1$ ).
Initial Conditions	Required for derivatives $y^{(k)}(0^-)$ .	Required for past output values $y[-1], y[-2], \dots$ .
Implementation	Realized by analog circuits (e.g., op-amps, RLC) for CT systems.	Realized by digital processors (e.g., microcontrollers, FPGAs) for DT systems.

Analysis Approaches:

Time Domain:
- CT: Solving differential equations can involve finding characteristic roots, particular solutions, and applying initial conditions. Direct convolution can be complex due to integration.
- DT: Solving difference equations often involves iteration or a recursive approach. Direct convolution is a summation that can be done graphically or analytically.
Transform Domain:
- CT: Converting differential equations to algebraic equations in the $s$ -domain simplifies finding the system function $H(s)$ . Inverse Laplace Transform yields $h(t)$ or $y(t)$ .
- DT: Converting difference equations to algebraic equations in the $z$ -domain simplifies finding the system function $H(z)$ . Inverse Z-Transform yields $h[n]$ or $y[n]$ .

Both approaches provide powerful tools, but the transform domain is particularly favored for LTI systems due to the simplification of convolution to multiplication and ease of analyzing frequency response and stability.

23

Explain the concept of superposition and how it applies to linear systems. Provide an example of a system that is linear and one that is not, justifying your answer mathematically.

24

Explain the concept of time-invariance for a system. Provide an example of a system that is time-invariant and one that is not, justifying your answer mathematically.

25

For a discrete-time LTI system, explain how the output $y[n]$ can be found using graphical convolution. Illustrate with a simple conceptual example.

Graphical convolution is a visual method for computing the convolution sum $y[n] = \sum_{k = -\infty}^{\infty} x[k]h[n-k]$ . It involves four main steps:

Steps for Graphical Convolution:

Flip One Signal: Choose one of the signals, say $h[k]$ , and flip it in time to get $h[-k]$ . (The choice doesn't matter due to commutativity: $x[n]*h[n] = h[n]*x[n]$ ).
Shift the Flipped Signal: Shift the flipped signal by units to get . This effectively slides along the -axis.
- For positive $n$ , $h[n-k]$ shifts to the right.
- For negative $n$ , $h[n-k]$ shifts to the left.
Multiply: For each specific value of $n$ , multiply $x[k]$ by $h[n-k]$ point-by-point (i.e., for each value of $k$ ). This results in a new sequence, $w_n[k] = x[k]h[n-k]$ .
Sum: Sum all the values of the resulting product sequence $w_n[k]$ over all $k$ . This sum gives the value of the output $y[n]$ at that specific time instant $n$ .
$y[n] = \sum_{k = -\infty}^{\infty} x[k]h[n-k]$
Repeat steps 2-4 for all relevant values of $n$ to obtain the complete output sequence $y[n]$ .

Conceptual Example:

Let $x[n] = \{1, 2, 1\}$ for $n=0,1,2$ and $h[n] = \{1, 1\}$ for $n=0,1$ . (Using start index at $n=0$ )

Original Signals:
- $x[k]$ : $1$ at $k=0$ , $2$ at $k=1$ , $1$ at $k=2$ .
- $h[k]$ : $1$ at $k=0$ , $1$ at $k=1$ .
Step 1: Flip $h[k]$ : $h[-k]$ : $1$ at $k=0$ , $1$ at $k=-1$ .
Step 2, 3, 4: Shift, Multiply, Sum (for various $n$ ):
- For $n=0$ : Shift $h[-k]$ by $0$ to get $h[0-k]$ .
  $x[k]$ : $\dots, 0, \underline{1}, 2, 1, 0, \dots$
  $h[0-k]$ : $\dots, 1, \underline{1}, 0, 0, 0, \dots$ (flipped $h[k]$ at $k=-1, 0$ )
  Product $x[k]h[0-k]$ : $1 \cdot 1 = 1$ at $k=0$ . All other products are 0.
  $y[0] = 1$ .
- For $n=1$ : Shift $h[-k]$ by $1$ to get $h[1-k]$ .
  $x[k]$ : $\dots, 0, \underline{1}, 2, 1, 0, \dots$
  $h[1-k]$ : $\dots, 0, 1, \underline{1}, 0, 0, \dots$ (flipped $h[k]$ at $k=0, 1$ )
  Product $x[k]h[1-k]$ : $(1 \cdot 1) + (2 \cdot 1) = 1 + 2 = 3$ at $k=0, 1$ .
  $y[1] = 3$ .
- For $n=2$ : Shift $h[-k]$ by $2$ to get $h[2-k]$ .
  $x[k]$ : $\dots, 0, \underline{1}, 2, 1, 0, \dots$
  $h[2-k]$ : $\dots, 0, 0, 1, \underline{1}, 0, \dots$ (flipped $h[k]$ at $k=1, 2$ )
  Product $x[k]h[2-k]$ : $(2 \cdot 1) + (1 \cdot 1) = 2 + 1 = 3$ at $k=1, 2$ .
  $y[2] = 3$ .
- For $n=3$ : Shift $h[-k]$ by $3$ to get $h[3-k]$ .
  $x[k]$ : $\dots, 0, \underline{1}, 2, 1, 0, \dots$
  $h[3-k]$ : $\dots, 0, 0, 0, 1, \underline{1}, \dots$ (flipped $h[k]$ at $k=2, 3$ )
  Product $x[k]h[3-k]$ : $(1 \cdot 1) = 1$ at $k=2$ .
  $y[3] = 1$ .

The final output sequence is $y[n] = \{1, 3, 3, 1\}$ for $n=0, 1, 2, 3$ . (Note: My manual calculation for Q3 gave $1,3,5,3$. This means the example must be simpler and/or my calculation there was wrong for this specific problem type, or I misinterpreted the indices here. Let's re-do the example from Q3 here with graphical steps for consistency and ensure it's simple enough). Okay, the $x[n] = \{1, 2, 3\}$ and $h[n] = \{1, 1\}$ from Q3 is better for example.

Let's re-do the example for $x[n] = \{1, 2, 3\}$ (for $n=0,1,2$ ) and $h[n] = \{1, 1\}$ (for $n=0,1$ ).

Step 1: Flip $h[k]$ : $h[-k]$ : $1$ at $k=0$ , $1$ at $k=-1$ .
Step 2, 3, 4: Shift, Multiply, Sum
- $n=0$ : $h[0-k]$ : $1$ at $k=0$ , $1$ at $k=-1$ .
  $x[k]h[0-k]$ : $x[0]h[0] = (1)(1) = 1$ .
  $y[0] = 1$ .
- $n=1$ : $h[1-k]$ : $1$ at $k=1$ , $1$ at $k=0$ .
  $x[k]h[1-k]$ : $x[0]h[1] + x[1]h[0] = (1)(1) + (2)(1) = 1 + 2 = 3$ .
  $y[1] = 3$ .
- $n=2$ : $h[2-k]$ : $1$ at $k=2$ , $1$ at $k=1$ .
  $x[k]h[2-k]$ : $x[1]h[1] + x[2]h[0] = (2)(1) + (3)(1) = 2 + 3 = 5$ .
  $y[2] = 5$ .
- $n=3$ : $h[3-k]$ : $1$ at $k=3$ , $1$ at $k=2$ .
  $x[k]h[3-k]$ : $x[2]h[1] = (3)(1) = 3$ .
  $y[3] = 3$ .

This yields $y[n] = \{1, 3, 5, 3\}$ for $n=0, 1, 2, 3$ , which matches the calculation in Q3. The conceptual illustration is correct.

26

For a continuous-time LTI system, explain how the output $y(t)$ can be found using graphical convolution. Illustrate with a simple conceptual example of two rectangular pulses.

27

What is the importance of the unit impulse function $\delta(t)$ and $\delta[n]$ in the context of LTI system analysis? Explain its role in determining system response.

The unit impulse function (Dirac delta function $\delta(t)$ for CT, unit impulse sequence $\delta[n]$ for DT) is of paramount importance in LTI system analysis because it serves as the fundamental building block for any arbitrary signal and, consequently, its response defines the system's complete behavior.

Importance and Role:

Defining the Impulse Response: The most crucial role is that the output of an LTI system to a unit impulse input (with zero initial conditions) is defined as the system's impulse response, $h(t)$ or $h[n]$ .
- $T\{ \delta(t) \} = h(t)$
- $T\{ \delta[n] \} = h[n]$
  The impulse response completely characterizes the LTI system. If you know $h(t)$ or $h[n]$ , you know everything about the LTI system.
Sifting Property: The impulse function has the unique "sifting property," which allows any signal to be represented as a sum (for DT) or integral (for CT) of scaled and shifted impulses:
- CT: $x(t) = \int_{-\infty}^{\infty} x(\tau)\delta(t - \tau) d\tau$
- DT: $x[n] = \sum_{k = -\infty}^{\infty} x[k]\delta[n-k]$
  This property is critical because it decomposes a complex signal into its simplest components.
Derivation of Convolution: By combining the sifting property with the linearity and time-invariance properties of LTI systems, the convolution integral/sum is derived:
- The linearity property allows the system operator to be passed through the sum/integral and for scalar multipliers ( $x(\tau)$ or $x[k]$ ) to be factored out.
- The time-invariance property allows the response to a shifted impulse $\delta(t - \tau)$ (or $\delta[n-k]$ ) to be simply a shifted impulse response $h(t - \tau)$ (or $h[n-k]$ ).
  This leads directly to $y(t) = x(t) * h(t)$ or $y[n] = x[n] * h[n]$ .
Universal Response Calculation: Since any input $x(t)$ (or $x[n]$ ) can be seen as a collection of impulses, and the system's response to each impulse is just a scaled and shifted version of $h(t)$ (or $h[n]$ ), the total output $y(t)$ (or $y[n]$ ) is the superposition (convolution) of these individual responses. This means the impulse response is the key to finding the output for any input.
Characterization of System Properties: Key system properties like memory, causality, and stability are directly determined by the shape and extent of the impulse response:
- Memoryless: $h(t) = K\delta(t)$ or $h[n] = K\delta[n]$ .
- Causal: $h(t) = 0$ for $t < 0$ or $h[n] = 0$ for $n < 0$ .
- Stable: $\int_{-\infty}^{\infty} |h(t)| dt < \infty$ or $\sum_{n = -\infty}^{\infty} |h[n]| < \infty$ .

In essence, the unit impulse function acts as a probe that, when applied to an LTI system, reveals its complete functional behavior through its impulse response, which then enables the analysis and prediction of the system's behavior for any other input.

28

Describe the differences between an LTI system and a static (memoryless) system. Provide an example of a static system that is not LTI, and justify your answer.

Differences between LTI and Static (Memoryless) Systems:

Feature	LTI System	Static (Memoryless) System
Definition	Linear AND Time-Invariant. Output via convolution.	Output depends ONLY on present input. No memory.
Memory	Can have memory (e.g., integrators, filters). Output depends on past/present/future input.	No memory. Output depends ONLY on present input.
Impulse Response	Can be of any duration. $h(t)$ or $h[n]$ can be non-zero for $t \ne 0$ or $n \ne 0$ .	Must be an impulse. $h(t) = K\delta(t)$ or $h[n] = K\delta[n]$ .
Characterization	Fully characterized by its impulse response.	Output given by a static function $y(t)=f(x(t))$ .
Convolution	The fundamental input-output relationship.	Convolution simplifies to simple multiplication/scaling.

Relationship: A memoryless LTI system is a specific subset of LTI systems. It's an LTI system whose impulse response is an impulse function itself ( $K\delta(t)$ or $K\delta[n]$ ).

Example of a Static (Memoryless) System that is NOT LTI:

Consider the system described by $y(t) = x^2(t)$ .

1. It is Static (Memoryless):

The output $y(t)$ at any time $t$ depends only on the input $x(t)$ at the same time $t$ . It doesn't depend on $x(t-1)$ or $x(t+1)$ , nor does it involve any integration or differentiation that would imply memory.
Therefore, it is a static (memoryless) system.

2. It is NOT LTI:

For a system to be LTI, it must satisfy both Linearity and Time-Invariance.
- Test for Linearity (fails):
  Let $x_1(t) \rightarrow y_1(t) = x_1^2(t)$ .
  Let $x_2(t) \rightarrow y_2(t) = x_2^2(t)$ .
  Consider the input $a x_1(t) + b x_2(t)$ . The output is $y_{test}(t) = (a x_1(t) + b x_2(t))^2 = a^2 x_1^2(t) + 2ab x_1(t)x_2(t) + b^2 x_2^2(t)$ .
  For linearity, we would expect $a y_1(t) + b y_2(t) = a x_1^2(t) + b x_2^2(t)$ .
  Clearly, $y_{test}(t) \ne a y_1(t) + b y_2(t)$ due to the $a^2, b^2$ terms and the cross-product $2ab x_1(t)x_2(t)$ .
  Thus, the system is non-linear.
- Test for Time-Invariance (passes):
  Let $x(t) \rightarrow y(t) = x^2(t)$ .
  Consider shifted input $x(t - t_0)$ . Output is $y_{shifted}(t) = (x(t - t_0))^2$ .
  Now, shift the original output: $y(t - t_0) = (x(t - t_0))^2$ .
  Since $y_{shifted}(t) = y(t - t_0)$ , the system is time-invariant.

Since the system $y(t) = x^2(t)$ is time-invariant but not linear, it is a static, non-linear system, and therefore not an LTI system.

29

Consider two LTI systems cascaded, with impulse responses $h_1(t)$ and $h_2(t)$ . What is the overall impulse response of the cascaded system? Use the properties of LTI systems to justify your answer.

When two LTI systems are cascaded (connected in series), the output of the first system becomes the input to the second system. Let the first system have impulse response $h_1(t)$ and the second system have impulse response $h_2(t)$ .

Let the overall input be $x(t)$ and the overall output be $y(t)$ .
Let the output of the first system be $y_1(t)$ .

Step-by-step Derivation:

First System Output: The output of the first LTI system, $y_1(t)$ , due to input $x(t)$ is given by the convolution integral:
$y_1(t) = x(t) * h_1(t)$
Second System Output: The input to the second LTI system is $y_1(t)$ . The output of the second system, $y(t)$ , is:
$y(t) = y_1(t) * h_2(t)$
Substitute $y_1(t)$ : Substitute the expression for $y_1(t)$ into the equation for $y(t)$ :
$y(t) = (x(t) * h_1(t)) * h_2(t)$
Apply Associative Property of Convolution: Convolution is associative. This means that for three signals $f, g, k$ , $(f*g)*k = f*(g*k)$ .
Applying this property:
$y(t) = x(t) * (h_1(t) * h_2(t))$

Overall Impulse Response:

From the final expression, $y(t) = x(t) * (h_1(t) * h_2(t))$ , we can identify the term $(h_1(t) * h_2(t))$ as the overall or equivalent impulse response of the cascaded system, because the output of an LTI system is always the convolution of the input with the system's impulse response.

Let $h_{eq}(t)$ be the overall impulse response. Then:
$h_{eq}(t) = h_1(t) * h_2(t)$

Justification using Properties:

Associativity of Convolution: This property is explicitly used in step 4 to group $h_1(t)$ and $h_2(t)$ together, showing that the overall behavior is determined by the convolution of the individual impulse responses.
Commutativity of Convolution: Although not strictly necessary for the derivation of $h_{eq}(t)$ , the commutative property ( $h_1(t) * h_2(t) = h_2(t) * h_1(t)$ ) implies that the order in which the LTI systems are cascaded does not affect the overall equivalent impulse response or the final output. This gives flexibility in system design and analysis.

In summary, the associativity of convolution guarantees that cascading LTI systems results in a single equivalent LTI system whose impulse response is the convolution of the individual impulse responses.

30

Consider two LTI systems connected in parallel, with impulse responses $h_1(t)$ and $h_2(t)$ . What is the overall impulse response of the parallel system? Use the properties of LTI systems to justify your answer.

When two LTI systems are connected in parallel, they share the same input, and their individual outputs are summed to form the overall output. Let the first system have impulse response $h_1(t)$ and the second system have impulse response $h_2(t)$ .

Let the overall input be $x(t)$ and the overall output be $y(t)$ .
Let the output of the first system be $y_1(t)$ and the output of the second system be $y_2(t)$ .

Step-by-step Derivation:

First System Output: The output of the first LTI system, $y_1(t)$ , due to input $x(t)$ is:
$y_1(t) = x(t) * h_1(t)$
Second System Output: The output of the second LTI system, $y_2(t)$ , due to input $x(t)$ is:
$y_2(t) = x(t) * h_2(t)$
Overall Output: In a parallel connection, the overall output $y(t)$ is the sum of the individual outputs:
$y(t) = y_1(t) + y_2(t)$
Substitute $y_1(t)$ and $y_2(t)$ : Substitute the expressions for $y_1(t)$ and $y_2(t)$ into the equation for $y(t)$ :
$y(t) = (x(t) * h_1(t)) + (x(t) * h_2(t))$
Apply Distributive Property of Convolution: Convolution is distributive over addition. This means that for three signals $f, g, k$ , $f * (g + k) = (f * g) + (f * k)$ .
Applying this property in reverse:
$y(t) = x(t) * (h_1(t) + h_2(t))$

Overall Impulse Response:

From the final expression, $y(t) = x(t) * (h_1(t) + h_2(t))$ , we can identify the term $(h_1(t) + h_2(t))$ as the overall or equivalent impulse response of the parallel system.

Let $h_{eq}(t)$ be the overall impulse response. Then:
$h_{eq}(t) = h_1(t) + h_2(t)$

Justification using Properties:

Distributive Property of Convolution: This property is explicitly used in step 5 to combine the two convolution terms into a single convolution with a summed impulse response. This shows that a parallel combination of LTI systems can be replaced by a single LTI system whose impulse response is the sum of the individual impulse responses.
Linearity of the System: The ability to sum the outputs from individual branches is a direct consequence of the linearity property of LTI systems. Each parallel branch itself is linear, and the summation of linear responses also results in a linear overall response.

In summary, the distributive property of convolution, supported by the linearity of the individual LTI systems, ensures that the overall impulse response of LTI systems connected in parallel is simply the sum of their individual impulse responses.

31

An LTI system is described by the difference equation $y[n] - 0.5y[n-1] = x[n]$ . Determine the impulse response $h[n]$ for this system, assuming it is causal.

32

An LTI system is described by the differential equation $\frac{{dy(t)}}{{dt}} + 2y(t) = x(t)$ . Determine the impulse response $h(t)$ for this system, assuming it is causal.

33

Explain the concept of software simulation of correlation. How would you modify the discrete-time convolution simulation procedure to implement discrete-time cross-correlation $r_{xy}[l]$ ?

Software Simulation of Correlation:

Software simulation of correlation involves numerically computing the cross-correlation sum between two discrete-time signals, $x[n]$ and $y[n]$ . It typically leverages the efficiency of convolution algorithms or direct summation.

Cross-correlation Sum Formula:
$r_{xy}[l] = \sum_{n = -\infty}^{\infty} x[n]y[n-l]$

Modification of Discrete-Time Convolution Simulation for Cross-Correlation:

The key insight is that cross-correlation is essentially convolution with one of the signals time-reversed. Specifically, $r_{xy}[l] = x[l] * y^*[-l]$ (for complex signals, or $x[l] * y[-l]$ for real signals).

Here's how to adapt a discrete-time convolution simulation procedure:

Represent Input Signals:
- x: Array representing $x[n]$ .
- y: Array representing $y[n]$ .
- Assume these are 0-indexed for simplicity, or handle starting indices explicitly.
Determine Output Length and Lag Range:
- If x has length $L_x$ (from $n=0$ to $L_x-1$ ) and y has length $L_y$ (from $n=0$ to $L_y-1$ ).
- The cross-correlation $r_{xy}[l]$ will have length $L_x + L_y - 1$ . Its minimum lag will be $-(L_y-1)$ and maximum lag will be $L_x-1$ . For software implementation, it's often common to shift the output such that $l=0$ corresponds to the middle index, or similar, depending on the library's convention.
Time-Reverse y[n]: Create a new signal, y_reversed, which is the time-reversed version of y[n]. For a 0-indexed array y of length Ly:
y_reversed[k] = y[Ly - 1 - k] (This flips and implicitly shifts origin to $k=0$ for the end of y)
A simpler conceptual way is to just use y[-k] in the sum.
Perform Convolution: Now, convolve $x[n]$ with this time-reversed signal y_reversed[n] (or just apply the convolution sum with $y[-(k-l)]$ ).
- Using the standard convolution operator (e.g., scipy.signal.convolve in Python, conv in MATLAB):
  r_xy = np.convolve(x, y_reversed)
  (Note: the time-reversal needs to be careful about indices)
- Direct Implementation for $r_{xy}[l]$ : This might be more straightforward given the formula directly:
  python
  
  Assuming x and y are 0-indexed arrays
  
  Lx = len(x)
  Ly = len(y)
  L_r = Lx + Ly - 1 # Length of correlation result
  r_xy = [0.0] * L_r # Initialize with zeros
  
  Correlation result indices typically range from -(Ly-1) to (Lx-1)
  
  Map this to 0-indexed array: offset = Ly - 1
  
  for l_idx in range(L_r): # Loop through each output index
  l = l_idx - (Ly - 1) # Calculate actual lag 'l'
  for n_val in range(Lx): # Loop through x[n_val]
  
  Compute index for y[n_val - l]
```
    y_index = n_val - l
    if 0 <= y_index < Ly: # Check bounds for y
        r_xy[l_idx] += x[n_val] * y[y_index]
```
  This direct loop explicitly implements $r_{xy}[l] = \sum_n x[n] y[n-l]$ .

Using FFT for Correlation (Efficient Method):

Just as with convolution, FFT can be used for efficient correlation:

Zero-Padding: Pad both x and y to length $N \ge L_x + L_y - 1$ , typically the next power of 2.
FFT: Compute $X[k] = \text{FFT}(x)$ and $Y[k] = \text{FFT}(y)$ .
Complex Conjugate and Multiplication: For cross-correlation $r_{xy}[l]$ , multiply $X[k]$ by the complex conjugate of $Y[k]$ : $R_{xy}[k] = X[k] \cdot Y^*[k]$ . (Note: For $r_{yx}[l]$ , it would be $Y[k] \cdot X^*[k]$ ).
IFFT: Compute the inverse FFT of $R_{xy}[k]$ to get $r_{xy}[l]$ .
r_xy = ifft(fft(x_padded) * conj(fft(y_padded)))

Libraries like scipy.signal.correlate (Python) and xcorr (MATLAB) use these efficient FFT-based methods internally.

Unit2 - Subjective Questions

Assuming x and h are 0-indexed arrays

Assuming x and y are 0-indexed arrays

Correlation result indices typically range from -(Ly-1) to (Lx-1)

Map this to 0-indexed array: offset = Ly - 1

Compute index for y[n_val - l]