GATE Electronics (EC): Probability, Statistics and Complex Variables (Engineering Mathematics), previous-year questions solved
Every probability, statistics and complex-variable question from the last five GATE Electronics and Communication papers, 11 in all: a median from a bar graph, a Markov state diagram, a Gaussian through a nonlinearity, uniform and dice probabilities, complex roots, and contour integrals by residues. Each answer matches the official key and was recomputed by code: exact enumeration or a seeded Monte Carlo for probabilities, and a numerically integrated contour next to the residue sum for every complex integral.
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 18 Nov 2026 · 11 questions
Median number of wickets per match from a bar graph (frequencies 5, 7, 8, 25, 20, 17, 8, 4, 3, 2, 1)
Answer: Official key: 4.0 to 4.0
- Read the frequencies off the bars for wickets: . They add up to matches.
- The median is the middle of the ordered list of matches. With matches it sits between the and the match.
- Add up the frequencies from the left: matches have at most wickets, at most , at most , at most , at most .
- The first matches have wickets or fewer, so the to the matches all have wickets. The and are both .
- Median .
Trap: the most common wrong answers are the mode (, the tallest bar), the mean (, pulled up by the long right tail) and the median of the eleven bar heights () or of the labels to (). The median is of the matches, not of the bars. The code expanded the graph into the individual matches, sorted them, took the and values ( and ), and checked that each of the four wrong readings falls outside the key.
The median is the statistic engineers use when a few extreme values would spoil the mean: latency and packet-delay reports quote the median and the 99th percentile, and a median filter removes impulsive spikes from a sensor or image signal without blurring edges the way an average does.
Markov chain on states A, B, C with p1 to p7 on the arrows: which relations always hold
- (A)
- (B)
- (C)
- (D)
Answer: (A) and (C) Official key: A,C
- In a Markov chain the probabilities on the arrows leaving a state add up to , because the chain must go somewhere. This holds for every state, whatever the numbers are.
- From the arrows are (to ), (to ) and the self-loop : . Option (C) is true.
- From the arrows are the self-loop and (to ): .
- From the arrows are the self-loop and (to ): .
- Both and equal , so they equal each other. Option (A) is true.
- (B) mixes arrows that enter and leave different states, and nothing forces it. (D) adds three self-loops, which are on three different states, so no row sum produces it. Both are false in general.
Trap: (B) and (D) look like \"add up the arrows\" statements, so they are tempting. A counterexample kills both: is a valid chain (all three rows sum to ). Then (B) reads and (D) reads . The code drew random valid parameter sets (every row summing to ) and tested the four options on each. (A) and (C) never failed. (B) failed in of them and (D) in . It also checked that (B) and (D) are not even true for the symmetric chain , and all other (where (B) holds but (D) gives ).
Markov chains with self-loops are the standard model for bursty errors on a wireless channel (a good state and a bad state, as in the Gilbert-Elliott model) and for queue lengths in a router buffer. The row-sum rule, that the outgoing probabilities of every state add to one, is the first check when a transition matrix is estimated from logged data.
X ~ N(0,1) is passed through the clipped function y = h(x) of the figure: which form has the pdf of Y
(In the options given below, are non-zero constants and is piece-wise continuous function)
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- From the figure, for , for , for , for and for . So always lies between and .
- Wherever is flat, a whole interval of is squashed onto one value of . That gives a point mass, which a pdf shows as a function. There are three flat pieces, so three terms:
- , , . These are the weights , so the terms sit at .
- On the two sloping pieces ( and ) the map is one-to-one with slope . has an ordinary density there, on and on . That is the piecewise continuous part , with total area .
- The three weights and add up to . The pdf is , option (B).
Trap: option (A) forgets the point mass at , which is the biggest one ( of the probability). Options (C) and (D) put the functions at , the values of where the corners are, not the values of : never leaves . The code sent standard normal samples through (fixed seed) and found exactly three values that repeat, , and , with frequencies , and against the exact , and . The histogram of the remaining samples matched in bins, and the continuous part had mass against the exact . Only (B) has terms at exactly those three values.
A limiter or saturating amplifier does exactly this to Gaussian noise: the output has point masses at the clipping levels and a Gaussian-shaped piece in between. The same happens at the input of an ADC, where a signal beyond full scale is clipped to the end codes and shows up as spikes at the two extreme values of the histogram.
X and Y are iid uniform on [0,1]: find P(X >= Y)
Answer: Official key: 0.50 to 0.50
- The pair is uniform on the unit square , which has area , so probability equals area.
- The event is the part of the square on or below the diagonal . That is a triangle with legs and .
- Area of the triangle . As an integral, .
- Symmetry gives the same answer: and play identical roles, so , and for a continuous variable. The two equal halves add up to . .
Trap: a discrete picture gives a different number. If and were uniform on only values (), ties would have probability and . On a -value grid it is . Only for a continuous variable is the tie probability zero and the answer exactly . The code confirmed this by exact fractions on both grids, by the sympy double integral (), and by random pairs (, with in 0 of them).
Two independent noise samples with the same distribution are equally likely to be the larger one. A detector that compares two branches carrying only noise therefore picks either branch with probability 1/2, which is why the error probability of a decision with no signal information is 0.5 (the 'coin flip' floor of a bit error rate).
A pot has 2 red and 2 blue balls: probability that two balls drawn without replacement differ in colour
What is the probability that the two balls drawn have different colours?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- Label the balls . Choosing of balls can be done in equally likely ways.
- Different colours means one red and one blue: ways ().
- , option (A).
- Check by the same-colour case: and are of the pairs, so and .
- Check in sequence: after the first ball (any colour) there are balls left and of them have the other colour, so .
Trap: option (C), , is the answer if the first ball is put back (with replacement), because then each draw is a fair coin between red and blue. Option (B), , comes from counting only \"red first, then blue\": , and forgetting the \"blue first, then red\" order, which has the same probability. The code listed all pairs and all ordered draws with exact fractions (both give ) and confirmed that with replacement gives and the one-order count gives .
This is the arithmetic of sampling a small batch without putting parts back: picking two chips from a lot of four with two defective ones gives a mixed pair two times in three. Acceptance-sampling plans for incoming components use the same without-replacement counting (the hypergeometric distribution).
Two fair dice are rolled: expected value of the sum of the outcomes
The expectation of is ______ (rounded off to two decimal places).
Answer: Official key: 6.95 to 7.05
- For one fair die, each face has probability , so .
- Expectation is linear: . This holds whether or not the dice are independent.
- .
- Check from the distribution of the sum: for , which is symmetric about , so the mean is .
Trap: stopping after one die gives . Multiplying instead of adding the two means gives , which is (the product), not the sum. Both are outside the key range to . The code enumerated all outcomes with exact fractions: , and the same result came from convolving the one-die pmf with itself. The same enumeration gives .
Linearity of expectation is why the average power of a sum of two independent noise sources is the sum of their powers, and why the mean of a received signal plus noise is the mean of the signal. It is used every time a signal-to-noise ratio is built up from separate noise contributions.
Value of A in the closed-path integral of 2^z/(z^2 - 1) = -i pi A over the path C in the figure
where , then the value of is ______ (rounded off to two decimal places).
Answer: Official key: 0.50 to 0.50
- Cauchy's residue theorem: for a simple closed path run counter-clockwise, . Read the direction from the arrows on : it is counter-clockwise.
- The poles of are and . The path crosses the real axis to the left of and again between and , so is inside and is outside. Only the residue at counts.
- Simple pole at : .
- .
- Compare with : .
Trap: counting the pole at as well. Its residue is , so the sum would be and ; the figure shows the path stops short of . Forgetting the factor gives , and reading the path clockwise gives . The code integrated numerically along a counter-clockwise curve with the same crossings as the figure ( points, trapezoid rule) and got ; sympy residues give at and at , and a circle of radius (which holds both poles) gave .
Contour integrals of this kind are how the inverse z-transform is done by residues: the sequence value is the sum of residues of X(z) z^(n-1) at the poles inside the region of convergence's contour. Which poles lie inside the path decides everything, the same way pole positions decide causality and stability of a filter.
Let w^4 = 16j: which of 2e^{j2pi/8}, 2e^{jpi/8}, 2e^{j5pi/8}, 2e^{j9pi/8} cannot be a value of w?
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- Write the right side in polar form: , with modulus and angle (plus any multiple of ).
- If then . So , giving , and , giving .
- The four roots are at angles , , and , all with modulus , spaced apart.
- Options (B), (C) and (D) are the angles , , , so they are roots. Option (A) has angle , which is not in the list, so the value that cannot be is option (A).
- Direct check of (A): .
Trap: the question asks for the value that cannot be a root, so picking one of the three valid options is the usual slip. Another is to space the roots by (dividing by instead of by ), which would wrongly accept (A). The code raised all four options to the fourth power: (B), (C) and (D) give to within and (A) gives . Sympy's solve of returned angles , , and .
Roots of unity and of other complex numbers are the twiddle factors of the FFT and the equally spaced phase points of PSK constellations. Four roots spaced 90 degrees apart, as here, are exactly the four points of QPSK rotated by 22.5 degrees.
Contour integral of (z+2)/(z^2+2z+2) counter-clockwise on |z + 1 - (3/2)j| = 1
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- The contour is the circle with centre and radius .
- Poles: gives , so .
- Distances from the centre: to it is , inside. To it is , outside.
- Residue at the simple pole : .
- Counter-clockwise: , option (B).
Trap: using the wrong pole. The residue at is , and times it is , option (D); that pole is outside the circle. Option (A), , is the right residue run clockwise. The code parametrised the circle, integrated numerically with points and got , which is . The clockwise run gave , and sympy residues agree.
Taking only the poles inside a chosen contour is the rule used to invert Laplace and z-transforms by residues. A pair of complex-conjugate poles such as -1 plus or minus j, as here, is the typical signature of an underdamped second-order system, and which one lies inside the contour tells you which part of the response you are picking up.
Integral of f(z) = sin(pi z)/(z^2 (z - 2)) around the circle |z| = 3
- (A)
- (B)
- (C)
- (D)
Answer: (D) Official key: D
- The singularities of are at and . Both lie inside , so the integral is times the sum of the two residues (counter-clockwise by default).
- At : the numerator is , so the zero of cancels the pole of . The residue is (a removable singularity).
- At : near , so . The double pole is really a simple pole, and .
- , option (D).
Trap: dropping the minus sign of gives and so , option (A). Options (B) and (C) are what you get by treating as non-zero at and adding a spurious term. The code integrated numerically around with points and got , and . Sympy residues give at and at ; only (D) matches.
A zero in the numerator cancelling a pole, as at z = 2 here, is the same pole-zero cancellation that removes a mode from a transfer function. In practice it matters because a cancelled pole does not show in the response, yet the full set of poles inside the contour still decides every residue sum used for inverse transforms.
Which contour integral statements on the unit circle (counter-clockwise) are true: e^z, z^n, cos z, sec z?
- (A)
- (B), where is an even integer
- (C)
- (D)
Answer: (A) and (B) Official key: A;B
- Cauchy's theorem: if is analytic inside and on , then . Otherwise use residues for the singularities inside.
- (A) is entire (no singularity anywhere), so . True.
- (B) For an integer , if and for every other integer (with , the integral is ). The one exception, , is odd, so for every even (positive, zero or negative) the integral is . True.
- (C) is entire, so . The statement says , so it is false.
- (D) has poles where , at . The nearest is at distance from the origin, so none is inside the unit circle. is analytic inside and on , the integral is , and the statement is false.
- True statements: (A) and (B).
Trap: reading (D) as 'secant has poles, so the integral is not zero'. The poles at are outside the unit circle, so they do not count. Another slip is to assume always integrates to zero, forgetting (; the code got ). The code integrated each integrand numerically around the unit circle ( points): and, for every even from to , gave (largest size ); gave size and size , so statements (C) and (D) fail.
Cauchy's theorem is why a contour integral picks up only the poles that lie inside it, and the integral of z to the minus one around the origin, 2 pi j, is the building block of every inverse z-transform by residues. Whether a pole sits inside or outside the unit circle is also the discrete-time stability test for a causal filter.