Signals and Systems

Convolution

Convolution is the time-domain language of system memory. It explains how every small part of an input signal excites an LTI system, and how those delayed responses add up to the waveform we finally observe at the output.

Exam focus

Graphical convolution, impulse response, finite-sequence output length.

Engineering use

Filtering, channel modeling, echo, image kernels, control response.

Core idea

Break input into impulses, delay responses, accumulate the result.

Introduction

Most students first meet convolution as a formula with a reversed signal, a dummy variable, and uncomfortable limits. That first impression is unfair to the concept. Convolution is really the answer to a practical engineering question: when a system has memory, how do earlier input values continue to shape the present output?

The impulse response is the system's fingerprint. Once that fingerprint is known, convolution uses it to predict the response to any input, from a short pulse to a complicated communication waveform. This is why the topic sits quietly underneath DSP, communication channels, image processing, control systems, and filter design.

So do not begin by memorizing the integral. Begin with the story: the input arrives piece by piece, the system reacts to each piece, and the output is the accumulated effect of all those reactions.

Why This Topic Matters

Convolution is a high-leverage topic. Once it is clear, several later ideas stop feeling like separate chapters and start looking like the same story told in different domains.

Industry relevance

  • Digital filters use convolution to smooth, sharpen, or remove unwanted components.
  • Communication receivers model channel distortion as a transmitted signal convolved with a channel impulse response.
  • Image processing uses 2D convolution for blur, sharpening, edge detection, and feature extraction.
  • Audio systems use convolution to model echo, reverberation, and room response.

Exam relevance

  • GATE frequently tests graphical convolution, overlap limits, and LTI output.
  • University exams ask derivations of continuous and discrete convolution.
  • Interviews often check whether impulse response feels physical to you, not just symbolic.

Prerequisites

  • Input-output idea of a system.
  • Impulse response $$h(t)$$ or $$h[n]$$.
  • Time shifting and reversal.
  • Integration and summation basics.
  • Linearity and time invariance.

If any of these feel weak, revise Signals Basics and LTI Systems first. Convolution becomes much easier when shifting, impulse response, and linearity are already familiar.

Basic Intuition

Imagine the input signal as a long row of tiny impulses. Each impulse taps the system at a particular instant. The system answers each tap with a copy of its impulse response, shifted to the tap time and scaled by the tap strength.

The output is what remains after all those small answers are added. Graphically, this becomes an overlap story: as one signal slides across another, only the overlapping portions interact. More useful overlap gives a stronger contribution; no overlap gives nothing.

A good mental line: convolution is the sum of all delayed system replies caused by the input.

Interactive visualization

Convolution as sliding, multiplying, and accumulating

Watch the impulse response slide across the input. Wherever the two signals overlap, their product contributes area. Those accumulated areas become the output samples of \(y(t)\).

Input x(t) and sliding h(t)Output y(t)x(t)h(t - tau)accumulatemultiply only where x(t) and shifted h(t) overlapeach accumulated area plots one point of y(t)
1

Slide

The orange impulse response h(t - tau) moves across the blue input x(t).

2

Multiply

The green region marks the overlap where point-by-point multiplication matters.

3

Accumulate

The overlap product is summed as area, giving one value of the output.

4

Build y(t)

Repeating the process for every shift draws the purple output waveform.

Mathematical Definition

Mathematically, convolution is a weighted accumulation. One signal supplies the weights, the other supplies the shifted response shape, and the integral or summation collects the total contribution at one output time.

Continuous-time definition

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

Here \(x(\tau)\) selects the input strength at an earlier time, and \(h(t-\tau)\) tells how that earlier event affects the present time \(t\).

Discrete-time definition

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

Each sample \(x[k]\) contributes a shifted copy of \(h[n]\), and the summation adds the aligned products.

Convolution in LTI Systems

Impulse response as system identity

The impulse response \(h(t)\) tells how the system reacts to the smallest ideal input event: a unit impulse. For an LTI system, this single response is enough to predict the output for any input.

That is a remarkable shortcut. Instead of testing the system with every possible waveform, we study its impulse response once and use convolution to assemble the response to everything else.

Superposition of shifted responses

Each small input piece at time $$\tau$$ produces a response shaped like $$h(t-\tau)$$. Multiplying by input strength $$x(\tau)$$ and adding all pieces gives output.

System memory hidden inside the formula

If \(h(t)\) lasts for a long time, one input event keeps influencing the output for a long time. If \(h(t)\) dies quickly, the system forgets quickly. The width and shape of the impulse response are therefore a picture of system memory.

Continuous-Time Convolution

1. Represent input using impulses

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

This identity says a signal can be rebuilt from infinitely many shifted impulses, each weighted by the signal value at that time.

2. System response to one shifted impulse

$$\delta(t-\tau) \to h(t-\tau)$$

Because the system is time invariant, delaying the impulse simply delays the impulse response; it does not change the response shape.

3. Add every scaled response

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

The integral adds all tiny response contributions. The variable $$\tau$$ is only a bookkeeping variable that lets us scan through the input history.

Physically, the integral measures accumulated influence. Same-sign overlap increases the output; opposite-sign overlap can cancel. That is why convolution is sensitive not only to signal size, but also to alignment and shape.

Discrete-Time Convolution

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

For sampled systems, integration becomes summation over sample index \(k\). This is the form implemented directly in FIR filters, DSP processors, and many software signal-processing routines.

Finite sequence length

If \(x[n]\) has length \(M\) and \(h[n]\) has length \(N\), then the convolution output length is \(M+N-1\).

Implementation view

Each output sample is a multiply-accumulate operation: multiply aligned samples, add them, shift, and repeat.

Graphical Convolution and Step-by-Step Procedure

  • Flip one signal in time.
  • Shift it across the other signal.
  • Multiply the overlapping portions.
  • Integrate or sum the overlap area.
  • Repeat for every shift to form the full output waveform.

The reversal step often feels unnatural at first. It comes from the term \(h(t-\tau)\). Once the signal is reversed, shifting lets us ask a precise question: at this output time, which past input values are still being remembered by the system?

Animated working: flip, shift, multiply overlap, integrate areainput x(tau)h(t - tau)overlap area becomes one output value y(t)integrateoutput y(t)=x(t)*h(t)

Convolution Theorem and Fourier Transform Relation

Time-domain convolution has a very useful frequency-domain interpretation. When two signals are convolved in time, their Fourier transforms multiply in frequency:

$$x(t)*h(t) \;\Longleftrightarrow\; X(j\omega)H(j\omega)$$

This is the reason filters are so easy to understand in the frequency domain. The input spectrum $$X(j\omega)$$ contains all frequency components present in the signal. The system frequency response $$H(j\omega)$$ decides which components are passed, weakened, delayed, or phase shifted.

In time domain, a filter is an impulse response convolved with the input. In frequency domain, that same filter multiplies the input spectrum. This is one of the most useful bridges between convolution and Fourier Transform.

Low-pass behavior

A smooth impulse response often suppresses rapid changes, so high-frequency components reduce at the output.

Channel behavior

A communication channel spreads pulses in time. Convolution explains intersymbol interference and multipath delay.

Implementation view

DSP hardware computes convolution using multiply-accumulate operations, especially in FIR filters.

How to Read a Convolution Diagram

A convolution diagram is not decoration; it is a problem-solving tool. Read it from left to right: fix one signal, flip the other, slide it, identify overlap, and then calculate the product area or product sum.

Before overlap

The shifted signal has not reached the fixed signal. The product is zero, so the output is zero.

Growing overlap

The overlap region increases. For positive signals, the output usually rises.

Maximum overlap

The aligned portions are largest. This often gives the peak output value.

Leaving overlap

The overlap shrinks, so the accumulated product decreases until the output returns to zero.

Important Formulas

Continuous convolution

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

Adds all delayed impulse-response contributions.

Discrete convolution

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

Used directly in FIR filters and DSP.

Commutative property

$$x*h=h*x$$

Either signal can be flipped and shifted.

Impulse property

$$x(t)*\delta(t)=x(t)$$

Impulse acts like the identity element of convolution.

Real-World Applications

In practice, convolution appears whenever a system spreads, delays, smooths, or remembers an input.

  • FIR digital filtering in DSP processors.
  • Wireless channel modeling where the received signal is the transmitted signal convolved with the channel response.
  • Image kernels for blur, sharpening, and edge detection.
  • Audio reverb using measured room impulse response.
  • Control-system response from input and impulse response.
  • Probability, where sums of independent random variables use convolution of distributions.

Solved Examples

Beginner example: impulse input

If input is $$x(t)=\delta(t)$$, output is:

$$y(t)=\delta(t)*h(t)=h(t)$$

This is why impulse response is so important: the impulse directly reveals the system's natural reply.

Intermediate numerical: discrete convolution

Let $$x[n]=[1,2]$$ and $$h[n]=[1,1]$$.

$$y[0]=1,\quad y[1]=1+2=3,\quad y[2]=2$$

So $$y[n]=[1,3,2]$$.

Advanced problem: rectangular pulses

Convolving two equal rectangular pulses produces a triangular waveform because overlap area first increases linearly, then decreases linearly.

Exam method: choose the easiest domain

If signals are short sequences, use direct summation. If signals are simple rectangles or triangles, use graphical overlap. If the question gives transforms or system frequency response, use the convolution theorem and work in the transform domain.

Common Mistakes

  • Forgetting to flip one signal before shifting.
  • Using $$h(t+\tau)$$ instead of $$h(t-\tau)$$ without tracking variables.
  • Confusing multiplication with convolution.
  • Ignoring overlap limits in graphical convolution.
  • Forgetting that convolution length in discrete finite sequences is $$L_x+L_h-1$$.

In exams, the most reliable habit is to mark the overlap interval first, then write limits. Most wrong convolution answers are not caused by difficult integration; they are caused by incorrect overlap limits.

Comparison Tables

OperationMeaningResult
MultiplicationPoint-by-point productInstant interaction only
ConvolutionFlip, shift, overlap, addFull system response
CorrelationSimilarity versus shiftDetection and matching

Interview Questions

  • What is convolution physically?
  • Why does LTI system output use convolution?
  • What is impulse response?
  • Why do we flip one signal in graphical convolution?
  • How is convolution used in filtering?
  • What is the difference between convolution and multiplication?
  • How would you explain convolution to someone who knows only impulse response?

Exam-Oriented Notes

  • For finite discrete sequences, output length is $$L_x+L_h-1$$.
  • Impulse convolved with any signal gives the same signal.
  • A shifted impulse shifts the signal: $$x(t)*\delta(t-t_0)=x(t-t_0)$$.
  • Convolution is commutative, associative, and distributive.
  • Graphical convolution requires careful overlap limits.
  • For LTI systems, output equals input convolved with impulse response.

Frequently Asked Questions

What is convolution in signals and systems?

Convolution is the operation used to find how one signal combines with another through shifting, multiplication, and accumulation. In Signals and Systems, it most often gives the output of an LTI system from the input and impulse response.

Why is convolution important?

It connects input signals, impulse response, filtering, memory, and system output in one framework. Once convolution is clear, LTI systems, Fourier Transform properties, communication channels, and DSP filters become much easier to understand.

What is the difference between convolution and correlation?

Convolution predicts system response by flipping, shifting, multiplying, and accumulating. Correlation measures similarity between signals as one is shifted, and is widely used for detection, synchronization, and pattern matching.

Why is one signal reversed in convolution?

The reversal appears because the impulse response term is h(t - tau) or h[n-k]. For a fixed output time, the integration or summation variable moves opposite inside that expression. The flip is the geometry of the formula, not an arbitrary rule.

What is the convolution theorem?

The convolution theorem says convolution in time domain becomes multiplication in frequency domain. For an LTI system, y(t)=x(t)*h(t) becomes Y(omega)=X(omega)H(omega).

How is convolution related to Fourier Transform?

Fourier Transform turns convolution into multiplication, so filtering and LTI system analysis become simpler in frequency domain. The impulse response h(t) becomes frequency response H(omega).

What is the use of convolution in engineering?

Convolution is used in FIR filters, communication channel modeling, audio reverberation, image processing kernels, radar echo analysis, control-system response, and probability distribution calculations.

Summary

  • Convolution gives the output of an LTI system when input and impulse response are known.
  • Continuous-time convolution uses integration over overlap.
  • Discrete-time convolution uses summation and appears directly in FIR filters.
  • Graphical convolution is mainly about flip, shift, multiply, and accumulate.
  • The Fourier Transform relation turns convolution into multiplication, which is why transform methods are powerful.
If you remember only one line, remember this: convolution is how a system combines the present input with the remembered effects of past input.

Practice Questions

Conceptual

  • Explain convolution using impulse response.
  • Why does convolution involve shifting?
  • What happens when a signal is convolved with impulse?

Numerical

  • Find convolution of $$[1,1,1]$$ and $$[1,2]$$.
  • Find output if $$x(t)=\delta(t-2)$$ enters an LTI system with impulse response $$h(t)$$.
  • Sketch convolution of two unit rectangular pulses.

MCQs

  • Convolution of a signal with impulse gives: zero / same signal / derivative / integral.
  • Discrete convolution of lengths 3 and 4 has length: 4 / 5 / 6 / 7.
  • For LTI systems, output is: $$x+h$$ / $$xh$$ / $$x*h$$ / $$x-h$$.