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, part 3
Part 3 of 4 of the Signals and Systems hub: 9 more official GATE EC questions on Fourier transform, DTFT and frequency response, each solved step by step and checked against the official answer key. Part 1 covers signals and system properties, impulse response and convolution, Fourier series and DTFS, Laplace and z-transform and sampling, DFT and digital filters.
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
Fourier transform X(jw) of x(t) = t / (1 + t^2)^2
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- Start from a known pair: (the two-sided exponential and its transform swapped by duality).
- Differentiate in time: . So .
- The differentiation property says . Therefore .
- Since , this is : option (A). It is purely imaginary and odd in , as it must be for a real, odd signal.
Trap: option (B) has the right size but no . A real odd signal can never have a real transform, so (B) is ruled out at once. The code integrated numerically over and got , against . The ratio to option (B) was at every test frequency (, , , ), and only (A) matched at all four. Options (C) and (D) lack the factor , so they are not zero at , but for an odd signal.
Differentiating a signal multiplies its spectrum by jw (j times the angular frequency), which is why derivative-based edge detection in image processing and sensor differentiators boost high frequencies, and why they amplify noise too.
LTI system with |H(f)| = 1 for |f| up to alpha: is it a pure delay, and does S_Y(f) = S_X(f) in band?
Statement I: The system is necessarily a pure delay system for inputs which are bandlimited to .
Statement II: For any wide-sense stationary input process with power spectral density , the output power spectral density obeys for .
Which one of the following combinations is true?
- (A)Statement I is correct, Statement II is correct
- (B)Statement I is correct, Statement II is incorrect
- (C)Statement I is incorrect, Statement II is correct
- (D)Statement I is incorrect, Statement II is incorrect
Answer: (C) Official key: C
- A pure delay by has : magnitude and a phase that is a straight line in with slope . So a delay needs two things: flat magnitude and linear phase.
- The figure gives only . The phase can be anything. For example with inside the band has the same flat magnitude, but its phase is curved, so different frequencies are shifted by different times (the group delay changes with ) and the output has a different shape. Statement I says the system is necessarily a pure delay, which is false. Statement I is incorrect.
- For a wide-sense stationary input, the output power spectral density is . The phase drops out. Inside we have , so there. Statement II is correct.
- Statement I incorrect and Statement II correct: option (C).
Trap: reading a flat magnitude as \"distortionless\" and choosing (A) or (B). The code built two systems with on and fed them a sum of three tones at , and Hz. With linear phase (a s delay) the output matched a delayed copy of the input to within 0.10% rms. With phase , the best match over every delay from to s was still off by 28% rms. For Statement II it filtered white noise through that curved-phase filter, averaged periodograms over 256 segments, and found in band and 0.0001 outside.
In audio and RF filter design a flat magnitude is never the whole story: an all-pass section keeps the magnitude at one but bends the phase, and designers check group delay so that a pulse or a steep edge is not smeared. Linear-phase FIR filters are chosen when waveform shape must be kept.
Energy of m(t) cos(w0 t) when m(t) is band-limited to B with energy E and w0 = 10B
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- Multiplying by splits the spectrum into two copies, each scaled by : .
- Because , the two copies (occupying ) do not overlap. So the energies simply add, with no cross term.
- Parseval: each copy has energy , since shifting a spectrum does not change its energy and scaling by scales energy by .
- Total: , option (B).
- Time-domain check: . The fast term averages out against the slowly varying because is far above , leaving .
Trap: option (A), , is the energy of only one of the two spectral lobes. The code used the band-limited pulse ( rad/s, rad/s), summed and with a dense Riemann sum over s, and found the energy ratio . It then split the FFT of the modulated signal into its positive-frequency and negative-frequency lobes: each holds of the modulated energy, that is each against , so one lobe alone gives option (A). Option (C) comes from treating as .
This is the power budget of a double-sideband transmitter: the carrier multiplication moves the baseband spectrum to a high frequency without changing its shape, and half of the baseband power ends up in each sideband. The same idea tells an RF engineer how much signal energy a mixer delivers around each of the two images.
Fourier transform X(w) of x(t) = e^(-t^2), given the integral of e^(-y^2) is sqrt(pi)
Note:
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- . Complete the square in the exponent: .
- So . The integral of a Gaussian along a line shifted by a constant imaginary amount is the same as along the real line, so it equals by the given note.
- , option (C).
- Quick checks: rules out (B), whose value at is ; must be bounded by , which rules out (A) because it grows with .
Trap: option (D), , is the width you get by mixing this pair with the pair for (whose transform is ). The code computed as a dense Fourier sum of for and got . At option (C) gives but option (D) gives , so only (C) matches all five points.
The Gaussian is its own Fourier transform up to scale, which is why Gaussian pulse shaping in GSM-style GMSK radios and Gaussian blur kernels in image processing both give smooth, ringing-free responses. A narrower pulse in time always means a wider spectrum, so the choice of width trades occupied bandwidth against symbol or edge sharpness.
Match continuous/discrete and periodic/aperiodic signals with the type of their spectrum
| Signal type | Spectral characteristics |
|---|---|
| (i) Continuous, aperiodic | (a) Continuous, aperiodic |
| (ii) Continuous, periodic | (b) Continuous, periodic |
| (iii) Discrete, aperiodic | (c) Discrete, aperiodic |
| (iv) Discrete, periodic | (d) Discrete, periodic |
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- Rule of thumb: periodic in one domain means discrete (lines) in the other, and discrete in one domain means periodic in the other. Aperiodic means continuous in the other, and continuous means aperiodic in the other.
- (i) Continuous, aperiodic signal: the continuous-time Fourier transform is continuous and aperiodic, so (i) (a).
- (ii) Continuous, periodic signal: the Fourier series has coefficients only at integer harmonics (discrete) and they do not repeat in (aperiodic), so (ii) (c).
- (iii) Discrete, aperiodic signal: the DTFT is a continuous function of that repeats every (periodic), so (iii) (b).
- (iv) Discrete, periodic signal: the DTFS has coefficients for integer that repeat with period , so (iv) (d).
- Result: (i) (a), (ii) (c), (iii) (b), (iv) (d). This is option (B).
Trap: option (A) is the 'periodic gives periodic' idea, which only holds for (iv), not for (ii) and (iii). The code computed one spectrum of each kind: the transform of never repeats under any shift tested, the sawtooth Fourier coefficients decay and do not repeat in , the DTFT of a 5-point sequence repeats exactly every and varies smoothly, and the DTFS of a period-8 signal repeats every 8 coefficients. Classifying them this way reproduced exactly the mapping in (B) and no other option.
A sampled signal (discrete in time) has a spectrum that repeats every sampling frequency, which is exactly why an ADC needs an anti-aliasing filter. In the other direction, a finite FFT record treats the data as periodic, which is why spectra from an FFT show discrete bins and why windowing is used to tame leakage.
X = 1 - e^(-jW) + 2e^(-3jW) through H = 1 - 0.5 e^(-j2W): output y[n]?
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- A term in a DTFT is the sample . So and .
- Output spectrum: .
- Multiply term by term: .
- Collect: , with .
- Inverse DTFT: , option (C). The same result is the convolution : coefficients .
Trap: the other options are sign slips. Option (B) matches the true output except for the signs of the and terms (the code compared coefficient by coefficient and found they differ only at and ): the term is and the term is , and signs are easy to lose when the meets a negative or positive sample. The code computed with np.convolve and again from the product of -point DFT samples with an inverse FFT: both gave . With in the coefficient would be , not an option.
Multiplying frequency responses is how a cascade of FIR stages, such as a comb filter followed by a differentiator in firmware, is collapsed into one filter whose taps are the convolution of the individual tap lists. The same polynomial multiplication is what a DSP does when it merges two filters to save cycles.
x(t) = 10cos(10.5Wt) through h(t) = pi (sin(Wt)/(pi t))^2 cos(10Wt): output?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- Pair: of height on (width ).
- The square in time is times the convolution of this rectangle with itself, a triangle on with peak . Multiplying by : the transform of is a triangle with peak at and zero at .
- Multiplying by in time splits the spectrum into two copies scaled by at . So is a triangle of peak centred at and extending from to , and a mirror image at .
- Input frequency is from the centre, so . This is real and positive, so no phase shift.
- , option (A).
Trap: option (B), , comes from forgetting that halves the triangle: with peak the gain at the same offset would be and the output . The code took and evaluated numerically with a dense sum over : , (triangle values and ), , and the output amplitude against .
A band-pass filter built from a low-pass prototype by multiplying the impulse response by a cosine is the standard way FIR band-pass filters are designed in DSP firmware, and it is how a channel-select filter is shifted to an intermediate frequency in a receiver. The halving of the passband gain by the cosine is the reason such designs are usually rescaled by a factor of 2.
Real signal f(t) with Fourier transform F(w): which bound on |F(w)| is always true?
Consider a continuous-time, real-valued signal whose Fourier transform exists.
Which one of the following statements is always TRUE?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- The magnitude of an integral is at most the integral of the magnitude: .
- The complex exponential has magnitude for every real and , so and .
- Equality holds at when for all , so the bound is tight and the strict inequality in (B) is not always true.
- Answer: (A).
Trap: option (C) looks the same as (A) but drops the absolute value inside the integral. Take : then while , so (C) fails. The code evaluated the Fourier integral on a dense grid for five signals (positive, negative, sign-changing pulse, damped cosine, shifted Gaussian burst) at 481 frequencies: (A) held every time (largest ratio was 1.000), (B) never held, and for the code found against , breaking (C). Option (D) fails too, for example at where .
This bound says a spectrum can never be larger than the total absolute area of the waveform, so a short, small pulse can only have a small spectrum. EMC engineers use it as a quick upper bound on the spectral content of a transient before measuring it with a spectrum analyser.
x(t) = e^(-2t) for t >= 0: bandwidth B in Hz holding 99% of the energy?
If of energy of lies within Hz, then which of the following options is TRUE for the value of ?
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- The Fourier transform is , so the energy spectral density is .
- Total energy (time domain): . (Parseval gives the same: .)
- Energy in : , because . Dividing by the total , the fraction is .
- Set it to : , so .
- Convert to hertz: . And , , so : option (B).
Trap: converting rad/s to Hz by dividing by instead of gives , which falls inside option (C), to , i.e. to . The code checked the transform numerically against at , found the point by a dense energy integral and bisection, , and tested the four option intervals: only (B) contains , and the wrong conversion lands in (C).
Defining the bandwidth as the band that holds a given fraction of the signal energy is how engineers size the filter or channel for a pulse that has no sharp edge in frequency, such as an exponentially decaying sensor or radar return. The wrong radian-to-hertz conversion is a classic spec error.