GATE Instrumentation (IN): Signals and Systems (sampling, Fourier series, Laplace and z-transforms, LTI systems, DFT and FFT), previous-year questions solved, part 2
Part 2 of 2 of the Signals and Systems hub: 16 more official GATE IN questions on LTI systems, convolution and correlation and DFT, FFT and digital filters, each solved step by step and checked against the official answer key. Part 1 covers signals, sampling and Fourier series and Laplace, Fourier and z-transforms.
Questions from official GATE IN papers (IIT Kharagpur 2022, IIT Kanpur 2023, IISc Bengaluru 2024, IIT Roorkee 2025, IIT Guwahati 2026); solutions by Skill-Lync. · Updated 19 Nov 2026 · 16 of 31 questions on this page
System dy/dt + y = 3x(t-3)u(t-3): transfer function including the time delay
Here represents a unit-step function. The transfer function of this system is ________
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- The transfer function is the ratio with zero initial conditions. Two Laplace rules do the work: differentiation turns into , and a delay of seconds turns into .
- Transform both sides: , because the delay is .
- Factor the left side: .
- . Option (B): a first-order lag with gain 3 and a 3 s pure delay.
Trap: the pole comes from the coefficient of (, a time constant of 1 s), not from the 3 in front of , which is the gain. The delay exponent is , not . We integrated the ODE for a unit step and compared it with the step response of each option: only gain 3, pole at 1 and a 3 s delay matches (final value 2.993, which is the DC gain of 3).
Process loops such as a heated pipe or a conveyor line behave like a first-order lag plus a transport delay, and this is the standard model used to tune their controllers. The delay is what limits how aggressively a loop can be tuned.
System y(t) = x(t) sin(2 pi t): is it linear and is it time-invariant
- (A)Linear and time-variant
- (B)Non-linear and time-invariant
- (C)Linear and time-invariant
- (D)Non-linear and time-variant
Answer: (A) Official key: A
- Two tests decide this. Linearity: the response to must equal . Time invariance: delaying the input by must only delay the output by .
- Linearity: the system multiplies by a fixed function of time, . Then . It is linear.
- Time invariance: feed in the delayed input . The output is . The delayed output is . These differ unless is a whole number of seconds, so the system is time-variant.
- Linear and time-variant: option (A).
Trap: calling it non-linear because the output contains a product. A product of the input with a fixed function of time is still linear; it would be non-linear only if the output multiplied the input by itself or by another signal. We tested superposition with random signals (linear: True) and a 0.3 s input delay against the delayed output (time-invariant: False), which matches only option A.
A multiplier driven by a fixed oscillator is exactly how a lock-in amplifier or a modulator works. It is linear in the signal but time-varying, and that time variation is what moves the signal to a new frequency.
Transfer function (s - pi)/(s + pi), steady-state output sin(pi t): which input produces it
Let be the unit-step function. The input that results in a steady-state output is ________
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- For a stable LTI system, a sinusoidal input at frequency gives, once the transient has died away, a sinusoid at the same frequency: its amplitude is multiplied by and its phase is shifted by . The only pole is at , in the left half plane, so the transient decays and the steady state exists.
- The frequency is . .
- So and : this is an all-pass section that leaves the amplitude alone and advances the phase by a quarter cycle.
- The output must be , so the input must be the output shifted back by that same quarter cycle: . Option (C).
Trap: adding the phase instead of removing it. Option B, , comes out as , which has the wrong sign. Option A comes out as , and D has the wrong phase as well. We integrated the state equation of for each of the four inputs and compared the output with after eight seconds: the worst error is 1.9e-11 for C and at least 1.4 for the others.
Phase shift of a sinusoid through an all-pass section is how signal conditioners and lock-in measurements align a reference with a sensor signal, and the same steady-state reasoning is used to read a frequency response straight off a transfer function. Always check that the system is stable first, because for an unstable one the steady-state sinusoid does not exist.
LTI impulse response delta(t) + 0.5 delta(t - 4), input cos(7 pi t / 4): the output
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- Convolving an input with an impulse leaves it unchanged, and convolving with a shifted impulse shifts it: . So .
- With , the delayed term is .
- A shift of is three and a half full turns, equal to a shift of , so .
- . Option (A).
- Frequency-domain check: , with zero phase, so the output is the input scaled by 0.5, a cosine again.
Trap: ignoring that the delayed copy is out of phase and adding the amplitudes to get 1.5 (option B). The delay of 4 s is not a whole number of periods, because the period is . We evaluated at 200 time points and compared it with each option: only A matches.
A direct signal plus an attenuated, delayed copy is what an echo or a reflection on a long cable does, and the sum can partly cancel or reinforce depending on the frequency. Checking the delay against the signal period shows at once whether a reflection is in phase or in antiphase.
Zero-lag autocorrelation of the discrete-time sequence x[n] = [1, 2, 3, 4]
- (A)1
- (B)10
- (C)20
- (D)30
Answer: (D) Official key: D
- The autocorrelation of a sequence at lag is : slide a copy of the signal by samples and sum the products.
- At zero lag the copy sits exactly on the original, so every sample is multiplied by itself: , the signal energy.
- . Option (D).
Trap: summing the samples instead of their squares gives , which is option (B). Zero-lag autocorrelation is always the largest value of . We computed the full correlation and checked that its centre value equals 30.
Zero-lag autocorrelation is the signal power times the record length, which is how a data logger estimates noise power on a sensor channel. Autocorrelation at other lags is used to find periodic disturbances, for example a pump pulsation hiding in a flow-transmitter signal.
Sinusoidal transfer function whose polar plot is a vertical line downward from the point 1
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- A polar plot shows in the complex plane as runs from 0 to : the real part along the horizontal axis, the imaginary part along the vertical axis.
- Read the plot. At the point is at 1, so . As grows the point moves straight down: the real part stays at 1, the imaginary part becomes more and more negative without limit.
- does exactly that: real part 1 for every , imaginary part , magnitude growing without bound. Option (A).
- (C) has the same line but runs upward. (D) is a semicircle that ends at the origin. (B) has magnitude 1 for all , so it traces the unit circle.
Trap: choosing (C) from the right line but the wrong direction. We evaluated each option for from 0 to 200 and tested the real part, the sign of the imaginary part and the final magnitude: only (A) holds a real part of 1 and heads to .
A transfer function with a numerator of the form 1 minus j omega T is a right-half-plane zero, the signature of a non-minimum-phase process such as the inverse response of a boiler drum level. Plotting the frequency response and reading its shape is how you spot it from test data.
Causal LTI system y[n] = (1/alpha) y[n-1] + x[n], initially at rest, input delta[n-p]: value of y[p+1]
- (A)0
- (B)1
- (C)
- (D)
Answer: (C) Official key: C
- The output for a unit impulse input is the impulse response. Start with : initially at rest, so for . Then , , , and so on.
- So , a decaying geometric sequence because .
- Time invariance: delaying the input by samples delays the output by samples, so for .
- . Option (C).
Trap: the exact value of does not matter (so long as the system was at rest before the impulse), and choosing (option D) is the value two samples after the impulse, . We ran the recursion for = 1.5, 2, 3 and 7 with = 11, 15 and 20: was always (for , : , , ).
This first-order recursion is the digital exponential averager used in instruments to smooth noisy readings: each new sample is blended with a fraction of the old output. Its impulse response, a geometric decay, tells you how many samples it takes for a spike or a step in the signal to fade.
Cross-correlation h(t) of f(t) and g(-t): continuous-time Fourier transform H(jw) in terms of F and G
Note: and denote the continuous-time Fourier transforms of and , respectively.
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- The cross-correlation of and is . It is a convolution with one signal reversed: .
- Time reversal turns the Fourier transform into , and convolution becomes multiplication, so . For real signals is the complex conjugate of .
- Here and , whose transform is , so .
- . Option (A).
Trap: option (C), , is the cross-correlation of and without the reversal in the question. Reversing first undoes the reversal built into correlation, so the result is the same as plain convolution of and . We tested this with 32-point random real sequences and a circular cross-correlation computed directly, then compared its DFT with the four options: only matched.
Correlation is how instruments find a time delay, for example the transit time between two sensors in a cross-correlation flow meter, and the frequency-domain product is how it is computed quickly on an FFT. Knowing which signal is conjugated tells you whether the peak appears at a positive or a negative lag.
Eigenfunctions of stable LTI continuous-time systems: e^(j2pi t/3), cos, 2^t or sin?
- (A)
- (B)
- (C)
- (D)
Answer: (A) and (C) Official key: A;C
- An eigenfunction of a system is an input that comes out as the same function multiplied by a constant: .
- For an LTI system with impulse response and input , the output is . So every complex exponential is an eigenfunction, with eigenvalue .
- (A) is with : an eigenfunction, with , the frequency response. A stable system has a frequency response that exists, so this always works.
- (C) is , so , also a complex exponential. For a stable causal system (all poles in the left half-plane) exists for every with a real part above the rightmost pole, so it exists at and the output is .
- (B) and (D) are real sinusoids, each a sum of two exponentials, and . Each term is scaled by a different factor, and , so the output is or , shifted in phase. A cosine goes in and a mix of cosine and sine comes out, which is not a constant multiple of the input.
- The eigenfunctions are (A) and (C).
Trap: expecting a real sinusoid to be an eigenfunction because it keeps its frequency. It keeps the frequency but not the shape. We pushed each option through two stable causal systems, and , by numerical convolution: for the output-to-input ratio was constant at (that is of ), for it was constant at , and for the cosine it changed with time (, , at three instants). A stable but non-causal system can have no response to at all, so the answer assumes the usual causal setting.
Eigenfunctions are why instruments are characterised with a sine sweep: a pure tone through a linear system comes out as the same tone with a new amplitude and phase. That is the basis of frequency-response testing of sensors, filters and control loops.
Which of three differential equations do not represent a linear dynamical system
- (A)(i) and (ii) only
- (B)(i) only
- (C)(ii) only
- (D)(ii) and (iii) only
Answer: (C) Official key: C
- A system is linear if it obeys superposition: the response to is times the first response plus times the second. For a differential equation this means , its derivatives and appear only to the first power, never multiplied together and never inside a function. The coefficients may depend on time.
- (i): . The unknown and its derivatives appear to the first power. The coefficient changes with time, so the system is time-varying, but it is still linear.
- (ii): . The squared derivative breaks superposition: doubling quadruples that term. The system is non-linear.
- (iii): has constant coefficients and first powers only: a linear time-invariant (LTI) system.
- Only (ii) is not linear: option (C).
Trap: reading the time-varying coefficient in (i) as non-linear. The test for linearity is about the dependence on and , and a coefficient that depends on breaks time-invariance, not linearity. We applied each left-hand side to two test functions and checked : only (ii) fails.
Linear time-varying models are common in process work, for example a tank whose area or a heat exchanger whose gain drifts with time, and the superposition tools still apply to them. Truly non-linear terms, such as a squared flow or a saturating valve, need linearisation around an operating point before you can use transfer functions.
Matching four system responses to linear/non-linear and time-invariant/time-varying properties
| List-I | List-II |
|---|---|
| (M) | (I) Linear and Time-invariant |
| (N) | (II) Non-linear and Time-varying |
| (O) | (III) Linear and Time-varying |
| (P) | (IV) Non-linear and Time-invariant |
- (A)M-IV, N-III, O-II, P-I
- (B)M-II, N-I, O-IV, P-III
- (C)M-III, N-IV, O-II, P-I
- (D)M-I, N-IV, O-III, P-II
Answer: (A) Official key: A
- Two tests settle every case. Linearity: the response to must be times the first response plus times the second. Time invariance: delaying the input by must only delay the output by , so the response to must equal .
- (M) : squaring breaks superposition (non-linear). Delaying the input delays the output, so it is time-invariant. Non-linear and time-invariant: IV.
- (N) : linear, because the output is the input read at another instant. Time-reversal is time-varying: the input gives , but . Linear and time-varying: III.
- (O) : the absolute value makes it non-linear and the reversal makes it time-varying. II.
- (P) : a pure delay of 3 s, linear and time-invariant. I.
- The matching is M-IV, N-III, O-II, P-I: option (A).
Trap: treating as time-invariant because it has no explicit factor of . Any operation that reverses time or scales it is time-varying. We tested each of the four systems on test signals for superposition and for a 0.8 s shift of the input, and compared the resulting labels with all four options: only (A) matches.
A pure delay is a good model of a transport lag in a pipe or on a conveyor, and it is linear and time-invariant. A squaring or absolute-value block appears in rectifiers and in power or RMS computation, and those need non-linear treatment, so an LTI tool such as a transfer function does not apply to them.
Bilinear transformation of 1/((s+1)(s+3)) with sampling period 0.1 s: the discrete-time transfer function
Bilinear transformation with a sampling period of 0.1 s is employed to obtain the discrete-time transfer function . Then is ________
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- The bilinear transformation maps the analogue transfer function to a digital filter by the substitution . With , .
- First factor: .
- Second factor: .
- Invert and multiply: . Option (C).
- Quick checks: the DC gain must match, . At option C gives , while options A and D give and option B gives 0. The poles at and are the images of and , and the two zeros at come from .
Trap: a wrong sign in a bracket (options A, B and D each have one). We evaluated the bilinear substitution at four points on the unit circle and compared with every option: only option C matches.
The bilinear transformation is the usual way to turn an analogue filter or controller design into code that runs on a microcontroller or a PLC at a fixed sample time. It keeps a stable analogue design stable, which is why it is the default in digital controller implementation.
Filter y[n] - 0.5 y[n-2] = a x[n-4] with magnitude 0.5 at omega = pi/2: find a
Answer: Official key: 0.70 to 0.80
- For a difference equation, take the discrete-time Fourier transform of both sides. A delay of samples becomes the factor .
- , so .
- A pure delay has magnitude 1, so the numerator contributes only . At : , so the denominator is .
- , so .
Trap: forgetting that is and not turns the denominator into , giving . We evaluated numerically at with and got a magnitude of 0.500.
This is a recursive (IIR) filter with a built-in delay, a common shape for digital smoothing in transmitters. The designer sets the gain so that the response at a chosen frequency meets a target, exactly as in this question.
Complex multiplications needed for a 16-point DFT by the decimation-in-time radix-2 FFT
Answer: Official key: 32 to 32
- A radix-2 FFT splits an -point DFT into two -point DFTs and combines them in butterflies. Each butterfly uses one complex multiplication (by a twiddle factor) and two additions.
- Repeating the split gives stages. For there are stages, each with butterflies.
- Number of complex multiplications .
Trap: using (which is the addition count) or the direct DFT count . The FFT saves most of the work: 32 multiplications against 256. We counted the radix-2 recursion down to and got 32.
The FFT is what makes frequency analysis practical on a small embedded processor, for example vibration spectra on a machine-condition monitor. Cutting 256 multiplications to 32 for 16 points, and far more for 1024 points, is the difference between real-time and not.
Value z[0] of the 4-point circular convolution of two sequences from their 4-point DFTs
Note: The DFT of a -point sequence is defined as
Answer: Official key: 2 to 2
- Circular convolution in time is multiplication in the DFT domain: . So find the product, then invert the DFT.
- Multiply term by term: ; ; ; . So .
- The inverse DFT is . At every exponential is 1, so .
Trap: leaving out the factor and answering 8, or taking (a DFT value) as the time sample. We inverted numerically and cross-checked with the direct circular sum built from the inverse DFTs of and : both give 2. Sum without the factor: 8.
Fast convolution works exactly like this: take the FFT of the signal and of the filter, multiply, and inverse-transform. It is how long FIR filters and correlators are run in real time on measurement instruments.
DTFT of the 4-point sequence 1, 2, 2, 4 sampled at 2 pi k/3: value of y[0] of the 3-point DFT inverse
Answer: Official key: 5 to 5
- An -point DFT is the DTFT sampled at equally spaced frequencies . Here , but the sequence has 4 samples, which is more than the DFT length.
- Sampling the frequency axis makes the time sequence periodic: the 3-point inverse DFT returns , the original sequence wrapped around with period 3 (time-domain aliasing).
- The fourth sample lands on index . So , and the other two stay as they are: , .
Trap: giving as if the extra sample were dropped, or as the sum of all four. We evaluated directly at and took the 3-point inverse DFT with numpy: .
Time aliasing is the reason a spectrum computed on too few frequency bins gives a wrapped-round signal when transformed back. In practice engineers choose the FFT length at least as long as the record, or zero-pad, so that nothing wraps onto the start of the record.