GATE Electrical: Signals and Systems (signal operations, system properties, convolution), previous-year questions solved
Real Signals and Systems questions from the last five GATE Electrical papers, part 1 (the time domain): the sifting property, energy and periodicity, linearity, time invariance, causality, BIBO stability, impulse responses and convolution. Each solution matches the official answer key, names the trap that costs marks, and shows where the idea is used in real engineering. Fourier series, Fourier, Laplace and Z-transforms and sampling are on part 2.
Questions from official GATE EE papers (IIT Kharagpur 2022, IIT Kanpur 2023, IISc Bengaluru 2024, IIT Roorkee 2025, IIT Guwahati 2026); solutions by Skill-Lync. · Updated 16 Nov 2026 · 13 questions
x(t) = −t² on (−4, 4): what is the integral of x(t) δ(t + 3)?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- Sifting property: . Here sits at .
- lies inside the window , so the step bracket equals 1 there.
- : option (A).
Two traps: reading as an impulse at (same answer here only by luck, because is even), and dropping the minus sign in front of , which gives (B). Checked numerically with a narrow unit-area pulse in place of .
The impulse is how instruments and models are characterised: hammer-tap tests on a structure, a step or pulse injected into a motor drive, or a unit-impulse block in Simulink. Sampling a signal at one instant, which is what an ADC does, is the sifting property in hardware.
x(t) has energy E. What multiple of E is the energy of 2x(2t − 1)?
Answer: Official key: 2.0 to 2.0
- Energy of : .
- Substitute , so : the integral becomes .
Rule to remember: amplitude multiplies energy by ; compressing time by divides it by ; a time shift changes nothing. So . We confirmed it by integrating a test signal and its transformed copy numerically (ratio 2.000).
Energy and power bookkeeping under time scaling matters whenever a recorded signal is replayed faster or resampled: vibration and audio data, test-rig recordings sped up for analysis, or sensor logs decimated before storage.
x[n] = 1 + 3 sin(15πn/8 + 3π/4) − 5 sin(πn/3 − π/4): what is the period N?
Answer: Official key: 48 to 48
- A discrete sinusoid repeats after samples when for an integer , so must be an integer.
- First term: , : smallest integer is (with ).
- Second term: , : smallest is .
- The sum repeats when both do: .
Trap: computing and treating it as the period. In discrete time the period must be a whole number of samples. A brute-force search over also stops first at 48.
When a controller or DSP samples several tones at once (mains harmonics, gear-mesh and shaft frequencies), the combined sample pattern only repeats after the LCM of the individual periods. That is why FFT frame lengths are chosen to hold whole periods, otherwise the spectrum leaks.
y(t) = x(eᵗ): is the system linear? Is it causal?
- (A)linear and causal.
- (B)linear and non-causal.
- (C)non-linear and causal.
- (D)non-linear and non-causal.
Answer: (B) Official key: B
- Linear: the input gives . The system only reads the input at another instant, so superposition holds.
- Causal? The output at time uses the input at time . Since for every real (for example reads ), the output always depends on a future input.
- So the system is linear and non-causal: option (B).
Trap: thinking that only negative times look ahead. Even at the system reads , a later instant. Our check fed two inputs that agree up to and differ after it: the outputs differed at for every tried.
Non-causal operations are fine offline: post-processing a recorded crash test or flight log can look ahead (zero-phase filtering in MATLAB's filtfilt does exactly this). A real-time controller cannot, which is why control filters always add delay.
y(t) = e^(−t) ∫ from −∞ to t of e^τ x(τ) dτ: linear, time-invariant, causal?
- (A)nonlinear.
- (B)linear and time-invariant.
- (C)linear but not time-invariant.
- (D)noncausal.
Answer: (B) Official key: B
- Move inside: .
- That is a convolution with : the weight depends only on .
- Every convolution is linear and time-invariant, and for makes it causal too. Option (B).
Trap: seeing the factor outside the integral and calling the system time-varying (option C). A time-varying gain outside is cancelled here by the inside. We checked it numerically: shifting the input by 1.5 s shifted the output by exactly 1.5 s, and the output matched a direct convolution with .
This is the first-order low-pass filter written in the time domain: an RC filter on a sensor line, or the exponential moving average used to smooth current and temperature readings in battery management and motor-control firmware.
T₁ sums x[0..n]; T₂ sums x[k]/2ᵏ. Which is BIBO stable?
- (A) and are BIBO stable.
- (B) and are not BIBO stable.
- (C) is BIBO stable but is not BIBO stable.
- (D) is not BIBO stable but is BIBO stable.
Answer: (D) Official key: D
- is an accumulator. The bounded input gives , which grows without limit, so is not BIBO stable.
- For , let . Then for every : bounded, so is BIBO stable.
- Option (D).
Note that is not time-invariant (the weight depends on the absolute index ), so you cannot use the LTI test directly. Bound the output itself, as above. With 200 random inputs bounded by 1, the output of never exceeded 2.
A pure accumulator is the integrator in every PI controller, and it is exactly why integrator wind-up happens: a sustained error makes the sum grow without bound. Drive and process controllers clamp or bleed the integrator for this reason.
An LTI system's impulse response is u(t). Is every bounded input safe?
- (A)Every bounded input signal applied to results in a bounded output signal.
- (B)It is possible to find a bounded input signal which when applied to results in an unbounded output signal.
- (C)On applying any input signal to , the output signal is always bounded.
- (D)On applying any input signal to the output signal is always unbounded.
Answer: (B) Official key: B
- The impulse response is , so the system is an integrator: .
- BIBO stability needs . Here diverges, so the system is not BIBO stable.
- A concrete bounded input that breaks it: gives , which grows without limit (our numerical convolution gives ). Option (B).
(D) is too strong: a zero input or a short pulse gives a bounded output. Unstable means some bounded input blows up, not every input.
Integrators appear wherever position is obtained from velocity or charge from current: an IMU integrating gyro rate drifts without bound from a tiny constant bias, and coulomb counting in a battery pack drifts the same way, which is why both are periodically re-referenced.
y(t) + (1/4) dy/dt = 2x(t): what is the impulse response?
- (A)
- (B)
- (C)
- (D)
Answer: (D) Official key: D
- Laplace transform with zero initial conditions: .
- .
- Causal inverse transform: : option (D).
Trap: reading the as the decay rate, which gives (A) or (C). Normalise the derivative coefficient to 1 first: . Time-stepping the equation for a unit step and differentiating the result gives , matching .
First-order models like this describe an RC sensor filter, a thermocouple's lag or a motor's electrical time constant. Engineers identify from a measured step or pulse response and then tune the controller around it.
dy/dt + 3y = 2x, initially at rest: what is the impulse response?
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- (zero initial conditions, since the system starts at rest).
- : option (C).
Option (D) has the right shape but no : it would be non-zero before the impulse arrives, which a system initially at rest cannot do. Options (A) and (B) swap the gain and the pole. Numerical check: .
An RC low-pass on an ADC input is exactly this equation. Its impulse response tells the firmware engineer how long a spike on the sensor line takes to die out, which sets how soon after a disturbance a reading can be trusted.
h[n] = 1 for n = 0, 1, 2 and 0 otherwise: what is y[n] for an input x[n]?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- For an LTI system, .
- Only are non-zero, so : option (A).
Trap: writing instead of , which gives (C), a non-causal system. Since for the system must be causal, so any option using is out. Checked by convolving a random input with .
This is a 3-tap moving-sum (moving-average) FIR filter, the simplest digital filter there is. Embedded firmware uses it to smooth ADC readings from current, temperature and position sensors.
A unit pulse on (−1, 1) convolved with a ramp from −2 to 2 on (1, 5): when is the output largest?
Answer: s Official key: 4 to 4
- is a unit pulse on , so : the area of inside a 2 s window centred at .
- On , : negative on , positive on , zero outside.
- The area is largest when the window covers only the positive part, , i.e. centred at s, where .
Check with the derivative: . Just before it is ; just after, the window's front edge passes where drops to 0, so turns negative. Our numerical convolution on a 0.01 s grid peaks at with .
A sliding window sum like this is a matched filter or moving integrator: radar and ultrasonic sensors find an echo by sliding a template over the received signal and looking for the peak of the convolution. The window position at the peak is the time of arrival.
y(t) = x(−t). What is always true of x(t) * y(t)?
- (A)It is an even signal.
- (B)It is an odd signal.
- (C)It is a causal signal.
- (D)It is an anti-causal signal.
Answer: (A) Official key: A
- . This is the autocorrelation of .
- Replace by and substitute : .
- So the result is always even: option (A).
It cannot be odd, because for any non-zero signal. Causality depends on , so (C) and (D) are not always true. We checked with a deliberately lopsided test signal: to within .
Autocorrelation is how engineers find hidden periodicity in noisy data: bearing-fault detection in vibration signals, pitch detection in audio, and GPS receivers locking onto a satellite's code all look for peaks in this even function.
x₁ has area 1.5 and x₂ has area 1: what is the area under x₁ * x₂?
- (A)1.5
- (B)2.5
- (C)3.5
- (D)4
Answer: (A) Official key: A
- Area property of convolution: . (In transform terms: .)
- Area of : a unit square plus a triangle, .
- Area of : a triangle of base 2 and height 1, .
- : option (A).
No need to compute the convolution itself, which is the trap that eats time in the exam. We computed it anyway, numerically: the area under comes out as 1.500.
The same rule says a filter's DC gain multiplies the area of whatever passes through it. Engineers use it to check a smoothing filter is normalised (taps summing to 1) so a sensor reading is not scaled up or down by the filtering.