GATE Electronics (EC): Signals and Systems (LTI systems and convolution, Fourier series and transform, DTFT, Laplace and z-transform, sampling, DFT and digital filters), previous-year questions solved
Every signals and systems question from the last five GATE Electronics and Communication papers, solved: signal operations and system properties, continuous and discrete convolution, Fourier series, Fourier transform and DTFT, Laplace and z-transform with regions of convergence, sampling and aliasing, the DFT and digital filters. Each answer matches the official key and was recomputed by code: numeric convolution, numeric transforms, an FFT or a test of the system on real inputs.
Questions from official GATE EC papers (IIT Kharagpur 2022, IIT Kanpur 2023, IISc Bengaluru 2024, IIT Roorkee 2025, IIT Guwahati 2026); solutions by Skill-Lync. · Updated 17 Nov 2026 · 9 of 31 questions on this page
Outputs of four systems for input sin(t): sin(-t), sin(t+1), sin(2t), sin^2(t). Which are definitely not LTI?
- (A)
- (B)
- (C)
- (D)
Answer: (C) and (D) Official key: C,D
- Key fact: a complex exponential is an eigenfunction of every LTI system, so a sinusoid at frequency comes out as a sinusoid at the same , only scaled in amplitude and shifted in phase. An LTI system cannot create a new frequency.
- : . This is just a gain of (amplitude change of , a phase shift). The system is LTI and gives exactly this output. So is not definitely non-LTI. (It could also be a time reversal , which is not time-invariant, but the data do not force that.)
- : has the same frequency with a phase shift of rad. The time advance is LTI and gives this output. Not definitely non-LTI.
- : has frequency rad/s while the input has rad/s. No LTI system can do this. is not LTI.
- : contains a DC term and a rad/s term and no rad/s term at all. Again new frequencies appear, so is not LTI (it looks like squaring, which is nonlinear).
- Answer: (C) and (D).
Trap: choosing (A) because looks like time reversal, or (B) because the shift looks like a change of the time axis. The question asks for systems that are definitely not LTI, and one input-output pair cannot prove that when a valid LTI system (gain , or an advance of s) produces the same output. The code confirmed that and reproduce the and outputs, then took an FFT of each output: and contain only the line at 1 rad/s, only at 2 rad/s, and has lines at 0 and 2 rad/s but none at rad/s.
The same frequency-in, same-frequency-out test is how engineers separate linear distortion from nonlinear distortion: a linear amplifier or filter only changes size and phase of a test tone, while harmonics or a DC shift in the output (as in a clipped or squared signal) show up on a spectrum analyser as a sign of nonlinearity.
System y(t) = x(e^t): is it causal and is it time invariant?
- (A)Causal and time invariant.
- (B)Non-causal and time varying.
- (C)Causal and time varying.
- (D)Non-causal and time invariant.
Answer: (B) Official key: B
- Causality: the output at time must depend only on inputs at times up to . Here uses at time . For every real we have (the smallest gap, , is at ). So always looks at a future value of the input. The system is non-causal.
- Example: at the output is , the input one second ahead.
- Time invariance: delay the input by to get . The new output is .
- Delaying the old output by gives .
- The two are different functions: in general (take , : against ). The system is time varying. Answer: (B), non-causal and time varying.
Trap: option (C), causal and time varying, comes from thinking the argument is just some positive time and so cannot be in the future. The code tested a pair of inputs that are both zero up to but differ after : the outputs already differ at because reaches there, so causality fails. It also checked that on a grid of points, that the shifted-input output does not equal the shifted output (time varying), and that linearity holds, so the only pair that fits is (B). Options (A) and (D) fail the time-shift test.
Time-warping systems like x(e^t) appear when audio or sensor data is resampled onto a non-uniform clock, for example when correcting for drifting sample clocks. Because the mapping is not a pure delay, such a stage cannot be described by a single impulse response, which is why it is analysed separately from the LTI part of a chain.
Which radian frequencies give x[n] = A cos(Omega n + pi/3) a period of 40?
- (A)
- (B)
- (C)
- (D)
Answer: (A) and (D), or (A) and (C) and (D): both accepted in the official key Official key: A;D OR A;C;D
- A discrete-time sinusoid repeats after samples when , which needs for some integer . For : .
- Test each option by computing and checking that it is a multiple of . (A) : yes. (B) , an odd multiple of , so : no. (C) : yes. (D) : yes.
- Fundamental period. The smallest period is where . (A) , , so . (D) , , so . (C) , , so the fundamental period is .
- If the question means the fundamental period equals 40, the answer is (A) and (D). If it means that 40 is a period, then (C) also qualifies, since two periods of 20 make 40. The official key accepts both, A;D or A;C;D.
Trap: option (B), . Its ratio is rational, so the signal is periodic, but the fundamental period is , not . After samples the signal comes back inverted. The code built for all four options, tested over 360 samples (true for A, C, D only) and found the smallest period: , , , for A, B, C, D.
Whether a sampled tone repeats exactly after a given number of samples decides whether an FFT of that length shows a clean line or leaks into neighbouring bins. Engineers choose the test-tone frequency as an integer number of cycles in the record, coherent sampling, for exactly this reason when measuring ADC performance.
Two LTI subsystems with b delta[n] added before and minus b delta[n] added after: linear? time invariant?
Consider the discrete-time system below with input and output . In the figure, and denote the impulse responses of LTI Subsystems 1 and 2, respectively. Also, is the unit impulse, and .
Assuming , the overall system (denoted by the dashed box) is _______.
- (A)linear and time invariant
- (B)linear and time variant
- (C)nonlinear and time invariant
- (D)nonlinear and time variant
Answer: (D) Official key: D
- Write the output. After the first adder the signal is . Subsystem 2 convolves it with , and the last adder subtracts : .
- The first term is an LTI operation on . The rest is a fixed sequence, , that does not depend on at all.
- Because , this fixed sequence is not zero. So the zero input does not give the zero output. A linear system must map to , so the system is nonlinear (it is only affine).
- Time invariance: shift the input by . The LTI part shifts by , but the fixed sequence stays where it is. The new output is not the old output shifted by , so the system is time variant.
- Answer: nonlinear and time variant, option (D).
Trap: it is tempting to say the and cancel, giving an LTI system (A). They would cancel only if , because the first impulse is filtered by and the second is not. The code built the system from random FIR : the zero-input response was with a peak magnitude of 3.39 (not zero), superposition failed, and a 4-sample input shift did not shift the output. With the same code gave a time-invariant system, so the constant term is what breaks it. Option (C) fails for the same reason: the offset does not move when the input moves.
A fixed offset or injected term added inside a signal chain, such as a DC offset in an ADC front end or a calibration tone added in firmware, makes a processing block affine rather than linear. Engineers remove or subtract it before using LTI tools like convolution, transfer functions or frequency response, because those tools assume zero in gives zero out.
x(t) = 2 on 0 to 1, else 0: ratio E[x(t)] / E[3x(-3t+5)]?
Let .
Which of the following options correctly represents the ratio, ?
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- Energy of : a pulse of height and width , so .
- Now . It is nonzero when , that is . The width is : the time scale by compresses the pulse, and the minus sign (a flip) and the shift by do not change the width.
- The height is . So .
- Ratio : option (C).
- General rule: for , . Here , so and the ratio is .
Trap: the amplitude and the time scale act in opposite directions. Forgetting the compression, , gives the ratio , which is option (D). Forgetting the amplitude , , gives the ratio , which is option (A). The code integrated and with dense Riemann sums: , , ratio .
Signal energy under gain and time-scaling is what decides whether a stretched or compressed waveform still fits a transmitter's power or energy budget: a pulse sent three times faster carries one third of the energy at the same height, and a three times bigger amplitude brings back nine times the energy.
Block diagram with LTI blocks h1[n], h2[n], two delays and two adders: which statement is necessarily true?
Which of the following statements is necessarily TRUE?
- (A) is causal.
- (B) is linear and time invariant.
- (C) is linear and time varying.
- (D) is non-linear.
Answer: (B) Official key: B
- Follow the signals. . The lower adder forms . Then , and .
- Every block is linear and time invariant: the two given sub-systems, the unit delays , and the adders and subtractors. A cascade or a parallel connection of linear time-invariant systems is again linear and time invariant, so is linear and time invariant: option (B).
- In the -domain: , , , . So .
- Why not (A): causality depends on and , and the question only says they are LTI. If (a one-step advance) the output responds before the input, so is not necessarily causal.
- Why not (C) and (D): a delay does not make a system time varying and no block is non-linear.
Trap: picking (A) because the figure shows delays . The delays are causal, but the blocks and are not stated to be. The code drew random non-causal impulse responses (taps from and ), simulated the diagram, and found a nonzero output before the impulse arrives, so (A) is refuted; for the same blocks the linearity error and the time-shift error are both below , so (B) holds. It also checked that the closed form above equals the DTFT of the simulated impulse response at three frequencies.
Composite DSP blocks are built exactly like this: filters, delays and adders combined into a larger structure, such as comb filters or lattice sections. Because each piece is LTI, the whole is LTI and a single transfer function describes it. Whether the whole is causal, and so implementable in real time, depends on whether each filter is.
x1(t) = e^-t u(t) convolved with x2(t) = u(t) - u(t-2): limit of y(t) as t goes to infinity
Answer: Official key: 0.0 to 0.0
- . The pulse is for and elsewhere, and is nonzero only for .
- For the window is : . This rises to at .
- For the window is : .
- As , , so .
- A quicker check: both signals are absolutely integrable, so the output is an energy-type signal that must die out. A pulse that has passed completely through a decaying system leaves nothing behind.
Trap: assuming the answer is the area (the DC gain times the pulse length). That is the area under , not its final value. Also the peak is , not . The code convolved the two signals with np.convolve on a ms grid up to s. It matched the closed form to within 1.0e-03, the values at , , were 0.0430, 0.00029 and 1.3e-08, and the last point was 2.7e-17.
This is the response of a first-order low-pass (an RC stage or a moving-average style sensor filter) to a finite-length pulse: the output rises while the pulse lasts and then decays back to zero, and the tail is what limits how closely pulses can be spaced in a data link or a sensor readout.
x1(t) = u(t+1.5) - u(t-1.5), x2(t) pulse plus two impulses: integral of y = x1 * x2
Answer: Official key: 15 to 15
- The area under a convolution is the product of the areas: . (This is the Fourier transform at : .)
- is a rectangle of height from to , so .
- has three parts: a unit impulse at (area ), a rectangle of height from to (area ) and an impulse of weight at (area ). So .
- .
Trap: the usual error is to read the impulse at as having area instead of its weight , or the rectangle as height and area . Either one gives , which is outside the key (15). The code built on a ms grid (an impulse of weight is one sample of height ), convolved it with using np.convolve and summed the result: , with and .
The same area rule is the DC-gain check for a filter: the sum of an FIR filter's taps is its gain at zero frequency, so a filter that should pass DC without change must have taps that add to 1. Image-blur kernels are normalised for the same reason, so that overall brightness is preserved.
Input x(0.5t) applied to a system with impulse response h(0.5t): output in terms of y(t)?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- The output of an LTI system is the convolution .
- For the new system the input is and the impulse response is , so the output is . Note that evaluated at the shifted argument is .
- Substitute . Then and , and the limits stay because the scale factor is positive: .
- The integral is the convolution evaluated at the time , which is . So : option (A).
- In general, scaling both signals by gives . Here , so the factor is and the time argument is .
Trap: option (B), , counts the factor twice, once for each scaled signal. Only the change of variable in the integral produces a factor, and it produces it once. The code tested this on and with a dense Riemann-sum convolution: the scaled system output equals at and is half of . Options (C) and (D) also have the wrong argument and size.
This is the rule behind resampling and time-stretching: if you play a recording and a filter's impulse response at a different speed together, the output is the original output at the new time, scaled in amplitude by the speed ratio. The same amplitude factor shows up when an analogue prototype filter is frequency-scaled to a new bandwidth.