GATE Electrical: Signals and Systems (Fourier series and transform, Laplace and Z-transform, sampling), previous-year questions solved
Part 2 of the GATE EE Signals and Systems questions from the official 2022-2026 papers: discrete and continuous Fourier series, Parseval's power theorem, ideal low-pass and band-pass filters, Fourier-transform properties, the region of convergence of Laplace and Z-transforms, decimation, sampling, aliasing and when a sampled signal is periodic. Every solution matches the official answer key and points out the common traps. Signal operations, system properties and convolution are on part 1.
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 · 15 questions
N = 3 with a₋₃ = 2 and a₄ = 1: which signal is it?
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- DTFS coefficients repeat with period : and (and ).
- .
- Rewrite with : , so . Option (B).
The question hides a simple signal behind an unfamiliar form. Two quick tests: at , , which rules out (A) and (D) (they give 4); (C) gives at . All four options were evaluated for to ; only (B) matches.
Periodicity of DTFS coefficients is the reason an -point FFT only has distinct bins: frequencies and are indistinguishable after sampling. Every spectrum analyser and FFT-based harmonic meter is built on this.
X(Ω) = (1 + cos Ω)e^(−jΩ); x_p[n] repeats x[0..2] with period 5: what is |a₃|?
Answer: Official key: 0.037 to 0.039
- Expand: .
- Read off : , , , zero elsewhere. So one period of is , which is all of .
- Because one period of holds the whole of , .
- , i.e. .
Trap: forgetting the , which gives 0.191. The factor only changes the phase, not the magnitude. We also inverted numerically and took a direct 5-point DFT: same .
Sampling a spectrum at points is what an -point FFT does. If the signal is longer than samples, the copies overlap in time (time-domain aliasing). Engineers zero-pad short records, as in this question, so one period holds the whole signal.
x(t) = 1 + 2cos 2πt + 2cos 4πt + 2cos 6πt: what is its average power?
Answer: Official key: 7 to 7
- The period is s (fundamental 1 Hz; the other terms are its 2nd and 3rd harmonics).
- Parseval: average power = sum of the powers of the components, because cross terms between different harmonics average to zero.
- DC term: . Each cosine of amplitude 2: , three of them: 6.
- Total: .
Trap: adding amplitudes squared without the for sinusoids, which gives 13. Numerical integration over one period gives 7.000.
This is how total harmonic distortion and RMS current are computed in power systems: the RMS of a distorted current is the square root of the sum of the squared RMS values of its harmonics. Power-quality meters and inverter test reports rely on it.
Square wave of height ±k with period 2π: the third-harmonic coefficients add up to 2/(3π). What is k?
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- is a square wave of period that switches between and . It is odd, so every cosine coefficient is zero: .
- Sine coefficients: , which is for odd .
- , so : option (B).
Trap: forgetting that the cosine part vanishes and halving the sum between them. Numerically, the third-harmonic sine coefficient per unit is and the cosine coefficient is 0.
A square wave is what an inverter leg actually produces. Its harmonics fall as (third harmonic at one third of the fundamental), which is why six-step motor drives cause torque ripple and why PWM and output filters exist.
H(s) = (As + B)/(s² + Cs + D) with positive constants: what can it not be?
- (A)low pass filter.
- (B)high pass filter.
- (C)band pass filter.
- (D)an integrator.
Answer: (B) or (D): both are accepted Official key: B OR D
- High frequency: the numerator has a lower degree than the denominator, so . A high-pass filter needs a non-zero gain as , so it is impossible.
- Low frequency: , finite. An integrator needs like as (a pole at ), which needs . Also impossible with .
- Low-pass is possible (make small: ), and band-pass is possible (make small: ).
- So two options are correct, and the official answer key accepts either (B) or (D).
We swept 3,000 random sets of positive : low-pass and band-pass shapes both appeared; a flat high-frequency gain and a rise at low frequency never did.
Reading a filter's type straight off the degrees of is a daily check when designing anti-aliasing filters for ADCs, EMI filters for converters, or the low-pass on a current sensor: the relative degree sets the high-frequency roll-off.
Four tones through h(t) = 2(sin 10πt/πt) cos 40πt: which survive?
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- is an ideal low-pass filter: gain 1 for , zero outside.
- Multiplying by shifts that band to with half the height; the factor 2 restores it. So for , zero elsewhere: an ideal band-pass filter.
- Input frequencies: and are blocked; and pass with gain 1 and no phase shift.
- Output: : option (C).
Trap: forgetting the modulation and treating as a low-pass filter, which gives (A). We evaluated the Fourier integral of numerically: gain 1.00 at and , 0.00 at and .
Modulating a low-pass prototype to build a band-pass filter is exactly how digital filter designers work: design one good low-pass, then shift it. Radio receivers, vibration-band monitoring and power-line carrier systems all use filters made this way.
X(ω) = 1 for |ω| < W₀: what happens to x(t) as W₀ grows?
- (A) tends to be an impulse as .
- (B) decreases as increases.
- (C)At ,
- (D)At ,
Answer: (A) Official key: A
- Inverse transform: .
- grows with , so (B) is false.
- At : , which depends on and is never negative, so (C) is false and (D) holds only for the single value .
- As the spectrum becomes flat (= 1 everywhere), which is the transform of : the peak grows, the main lobe of width shrinks, and the area stays . Option (A).
Numerically inverting the spectrum for gives : exactly , never in general.
The wider a system's bandwidth, the sharper its response to a pulse. That trade-off sets how fast an oscilloscope, a current probe or a data-acquisition channel can follow a fast edge, which is why bandwidth is the headline spec on every scope.
H(jω) = 1 for |ω| ≤ 200π: what is h(0)?
Answer: Official key: 200 to 200
- At the inverse transform is just the area under the spectrum: .
- .
Equivalently , whose limit at is . Trap: forgetting the and answering , or using the one-sided band and getting 100.
The peak of a filter's impulse response grows with its bandwidth. That is why wideband measurement channels pick up more noise spikes, and why anti-aliasing filters are set only as wide as the signal of interest needs.
x(t) = e^(−t⁴) cos t: what is dX/dω at ω = 0?
Answer: Official key: 0.0 to 0.0
- is real and even (both and are even), so is real and even.
- The derivative of a smooth even function at is zero.
- Directly: , and is odd, so the integral is .
No need to attempt the transform of (it has no simple closed form): symmetry answers it in one line. Numerical integration gives .
Symmetry arguments save real computation: an even (zero-phase) filter kernel has a real, even spectrum, which is why image-processing and offline signal-cleaning filters are designed symmetric, to avoid phase distortion.
Poles at 1/5, 2/3 and 3: for which ROC does the DTFT converge?
- (A)Discrete-time Fourier transform of x[n] converges if R is
- (B)Discrete-time Fourier transform of x[n] converges if R is
- (C)Discrete-time Fourier transform of x[n] converges if R is such that x[n] is a left-sided sequence
- (D)Discrete-time Fourier transform of x[n] converges if R is such that x[n] is a right-sided sequence
Answer: (B) Official key: B
- The DTFT is the Z-transform on the unit circle, , so it converges only if the ROC contains .
- Poles at . Possible ROCs are rings between poles: (left-sided), , , (right-sided).
- Only contains the unit circle: option (B) (a two-sided sequence).
(A) and (D) are the same ring, , which excludes ; the left-sided ROC in (C) is , which also excludes it.
The unit-circle test is the stability test for digital filters and digital controllers: a causal design is stable only if all its poles sit inside . Every IIR filter and every discretised PID loop on a microcontroller is checked this way.
x(t) = e^(t²) between t = 1 and t = 10: what is the ROC of its Laplace transform?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- The step bracket keeps only on : a finite-duration signal.
- On that interval is large ( at the end) but finite, so is the integral of a bounded function over a finite interval, finite for every .
- ROC: the entire -plane, option (A).
Trap: grows faster than any , so its Laplace transform over all would not exist at all. The window is what saves it. We evaluated the integral for to : finite every time.
Every real recording is finite in length, which is why transforms of measured data always exist. Engineers rely on this when they take the FFT or Laplace-domain fit of a test record, even of a signal that would blow up if it went on forever.
y[n] = x[2n] (keep every second sample): what is Y(z) in terms of X(z)?
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- : the odd-index terms cancel.
- Put : the even terms are exactly .
- Option (C).
(A) is the transform of the up-sampled signal (zeros inserted between samples), the opposite operation. (B) has the exponent sign flipped. We tested all four on a random 10-sample signal at 12 random complex : only (C) matches.
Keeping every second sample is decimation, used to cut the data rate of oversampled ADCs and in multirate filter banks. The term is the aliased copy, which is why a low-pass filter must come before the decimator.
x[n] = 0.9^|n| for all integers n: what is the ROC of X(z)?
- (A)
- (B)
- (C)
- (D)
Answer: (C) Official key: C
- Split the two-sided sequence: .
- The right-sided part converges for ; the left-sided part for , i.e. .
- Both must converge: , option (C).
A two-sided sequence always has a ring-shaped ROC (or none at all). (D) is the union of the two regions instead of the intersection. Partial sums settle at and blow up at , as expected.
Two-sided responses like are the zero-phase smoothing kernels used offline on recorded data. The ROC containing is what guarantees the filter has a well-defined frequency response.
x(t) = 2cos(180πt)cos(60πt) sampled at 200 Hz, then a 100 Hz low-pass: highest frequency left?
Answer: 80 Hz Official key: 80 to 80
- Product to sum: , i.e. tones at 120 Hz and 60 Hz.
- Sampling at Hz puts copies at . The 120 Hz tone is above Hz, so it aliases to Hz. The 60 Hz tone stays at 60 Hz.
- The 100 Hz low-pass keeps 60 Hz and 80 Hz. Maximum: 80 Hz.
Trap: answering 120 Hz by forgetting aliasing. The Nyquist rate for this signal is 240 Hz, and 200 Hz is below it. A 200-point DFT of one second of samples shows energy only in the 60 Hz and 80 Hz bins.
This is exactly how aliasing corrupts real data: an inverter's switching harmonics or a motor's vibration above half the sampling rate shows up as a false low-frequency tone. Data-acquisition systems put an analogue anti-aliasing filter before the ADC for this reason.
Sampling a periodic x_c(t) (period T) every Tₛ: when is x[n] periodic?
- (A) will always be periodic with period for all values of
- (B) will always be periodic with period for all values of
- (C) will never be periodic
- (D) will be periodic if and only if is a rational number
Answer: (D) Official key: D
- for all needs to be a whole number of periods: , i.e. , a ratio of integers.
- If in lowest terms, the period is samples, which need not equal (for the period is 5). So (A) is false.
- If is irrational (say ), the sample instants never line up with the waveform again and never repeats. So (B) and (C) are false.
- Option (D).
(Strictly, a constant signal is periodic under any sampling; the statement is meant for genuinely varying signals.) Our period search found for , for , and no period up to 5,000 samples for .
When a controller samples a 50 Hz grid waveform, the sampling rate is chosen as a whole multiple of 50 Hz (for example 10 kHz = 200 samples per cycle) so every cycle samples the same points. That is what lets a power meter or PLL compute clean per-cycle RMS values.