GATE Electronics (EC): Communications (random processes, AM and FM, information theory, digital modulation, PCM and coding), previous-year questions solved, part 3
Part 3 of 3 of the Communications hub: 13 more official GATE EC questions on digital modulation and detection and pulse code modulation and error-control coding, each solved step by step and checked against the official answer key. Part 1 covers random processes and noise, analog modulation and receivers and information theory and channel capacity.
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 · 13 of 31 questions on this page
Alternate QPSK and 16-QAM symbols at 1 mega-symbol/s: raw data rate in Mb/s
Answer: Official key: 2.99 to 3.01
- A symbol from an -point constellation carries bits. QPSK has , so bits per symbol. 16-QAM has , so bits per symbol.
- The stream alternates, so half the symbols are QPSK and half are 16-QAM. Per symbol the average is bits.
- The symbol rate is mega-symbol per second, counting both kinds. Raw rate .
Trap: adding the two bit counts to get treats a QPSK-plus-16-QAM pair as one symbol, but the mega-symbols per second counts each symbol separately. Taking the whole stream as 16-QAM gives , also wrong. The code built an alternating stream of symbols, counted the bits from the constellation sizes and got at , inside the official range to .
Wi-Fi, LTE and 5G NR all switch the modulation order (QPSK, 16-QAM, 64-QAM, 256-QAM) from one resource block or frame to the next as the link quality changes. The raw rate is always symbol rate times the bits per symbol of the order in use, before the code rate is applied.
ML detector, f(y|xA) = e^-(y+1) u(y+1) and f(y|xB) = e^(y-1)(1-u(y-1)): symbol error probability
Answer: Official key: 0.22 to 0.25
- ML rule: decide when , else . ML means equal priors, so the two symbols are equally likely.
- Supports: is non-zero only for , and only for . For only can have been sent, and for only . The densities overlap on .
- On the overlap, means , so . Decision regions: for and ; for and .
- Given the detector errs when :
- Given the detector errs when :
- With equal priors, .
Trap: adding the two conditional errors and forgetting the priors gives , which is outside the official range. The code integrated both error regions on a million point grid (, , ) and ran a Monte Carlo ML detector on million samples of each symbol with a fixed seed, which gave , matching .
Maximum likelihood detection is what every digital receiver does at the slicer: compare the likelihood of the received sample under each hypothesis and pick the larger. Here the densities have bounded support, so errors come only from the overlap region, which is the same picture as the overlapping Gaussian tails in a real slicer.
Optimum symbol error probability for symbols {-4, 0, 4} equiprobable in AWGN of variance 4
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- Equal priors and Gaussian noise, so the optimum (MAP, equal to ML) detector picks the nearest symbol. The decision thresholds are the midpoints, and .
- The noise standard deviation is . A threshold at distance from a symbol is standard deviation away.
- Symbol is wrong only if (it crosses upward): . Symbol is wrong only if : .
- Symbol is wrong if or : .
- Average over the three equal probabilities: , so option (B).
- Numerically , so .
Trap: giving the middle symbol one error edge instead of two gives , option (C). Using or the threshold distance of gives options (A) and (D). The code ran a Monte Carlo of 4 million symbols with a nearest-symbol detector ( against the closed form ), searched thresholds by brute force (best pair at ) and confirmed that only matches.
Multilevel signalling such as PAM-4 on 400G Ethernet links and ASK in optical links is analysed this way: inner levels have two error edges and outer levels one, so the symbol error probability is for equally spaced -PAM.
Symbol S uniform on {-2, 0, 2}, Y = S + N, ML decoder: Pe = alpha P(N greater than 1), find alpha
- (A)
- (B)
- (C)
- (D)
Answer: (D) Official key: D
- With equal priors and Gaussian noise of the same variance for every symbol, the ML decoder is the minimum-distance decoder. It picks the nearest symbol, so the decision thresholds are the mid-points and .
- : an error occurs if or , that is or . By symmetry, probability .
- : the only way to be wrong is , that is (the region above is decoded as , correct). Probability . The same holds for .
- Average over the three equal-probability symbols: the two outer symbols each err with probability and the middle one with , so .
- So , option (D).
Trap: option (B), , comes from giving the middle symbol only one decision boundary, so every symbol seems to contribute . Option (C), , comes from counting only the two outer symbols. The code simulated the ML decoder (it maximises the Gaussian likelihood over the three symbols) with equiprobable symbols for noise standard deviations , and and divided the measured error rate by : the ratios were 1.338, 1.336 and 1.334, all close to . The one-boundary path gives and the outer-only path gives , so options (B) and (C) fail the simulation.
Multi-level signalling such as PAM-4 on 100G and 400G Ethernet electrical links has exactly this feature: inner levels have two neighbours and outer levels only one, so the average symbol error is set by the count of neighbours, here of the single-boundary tail. Receivers put decision thresholds midway between levels, as in this question, when the levels are equally likely.
Equivalent Nyquist bandwidth of a four-level (-1, 0, +1, +2) signal at 10^4 symbols per second
Which of the following options is the correct value of equivalent Nyquist bandwidth of ?
- (A)10 kHz
- (B)64 kHz
- (C)5 kHz
- (D)20 kHz
Answer: (C) Official key: C
- The bandwidth of a pulse train is set by how often pulses arrive, not by how many levels each pulse has. The symbol rate is symbols per second, so one symbol every .
- Nyquist's zero-ISI criterion: pulses sent at can be separated at the receiver with no inter-symbol interference only if the signal occupies at least . The ideal pulse is with a flat spectrum on .
- Equivalent Nyquist bandwidth , option (C).
- The four levels carry at most bits per symbol, so the bit rate is at most kbit/s. That changes the bit rate, not the symbol rate, so it does not change this bandwidth.
Trap: option (D), kHz, is the bit rate ( bit/s) taken as a bandwidth, and option (A), kHz, is the symbol rate itself (the baseband bandwidth is half of it). The code built a random sequence of the four levels at symbols per second with ideal sinc pulses, averaged the FFT power spectrum of records of symbols, and found a flat spectrum that ends at 5.00 kHz, with of the power below kHz and below kHz. It also checked that a band limited to kHz cannot be zero-ISI (the aliased spectrum sum is zero at ), and that the sinc pulse at exactly kHz is zero at every other symbol instant (largest value ).
This is the baseband bandwidth rule behind line codes and pulse shaping in modems: at a symbol rate of R, an ideal channel of R/2 is the narrowest that still allows zero ISI, and real raised-cosine filters use R(1+alpha)/2. Multi-level signalling such as PAM-4 in high-speed links raises the bit rate for a given bandwidth by sending more bits per symbol, but the symbol rate sets the bandwidth.
Minimum bandwidth of a 16-QAM signal over AWGN with an ML receiver at 4 x 10^6 bits per second
The minimum transmission bandwidth (in MHz) of the modulated signal necessary for optimum recovery of information at the receiver is ______. (rounded off to two decimal places)
Answer: Official key: 0.95 to 1.05
- 16-QAM has points, so each symbol carries bits.
- Symbol rate: symbols per second.
- Zero-ISI (Nyquist) limit: the baseband pulses need at least .
- QAM puts those two baseband pulse streams (in-phase and quadrature) on a carrier. The spectrum is shifted to , so each side of the carrier gets , and the modulated (passband) signal needs .
- Minimum transmission bandwidth .
Trap: MHz is the baseband Nyquist bandwidth, not the bandwidth of the modulated signal; MHz comes from forgetting that a symbol carries bits. The code sent random 16-QAM symbols at Msymbol/s with ideal sinc pulses on a MHz carrier, averaged the FFT power spectrum of records, and measured the spectral lobe at half height from MHz to MHz, a width of MHz. The band that holds of the power is MHz wide. Both are inside the key range to .
Wi-Fi, LTE and 5G NR radios pack bits into QAM symbols so a fixed channel carries more bits per second: 16-QAM gives 4 bits per symbol per hertz of symbol rate, 64-QAM gives 6. The symbol rate times (1 + roll-off) is the number a spectrum planner uses to fit a carrier into its allotted channel.
Nearest BER of coherent ML QPSK demodulation over AWGN at Eb/N0 = 8.4 dB, using the erfc approximation
Assume the complimentary error function .
Which is the nearest bit error rate (BER) at the output of the demodulator?
- (A)
- (B)
- (C)
- (D)
Answer: (B) Official key: B
- Convert from dB: .
- QPSK is two independent BPSK streams on the in-phase and quadrature carriers. Each bit sees against noise density , so the bit error rate is the same as BPSK: .
- Put . With the given approximation, .
- , which is closest to , option (B).
Trap: option (A), , comes from using the BPSK-per-symbol thinking with the wrong energy: instead of gives , nearest to . The code evaluated the exact function (), the question's approximation (), and a Monte Carlo of bits with a fixed seed ( errors, BER ) and a separate run of complete QPSK symbols (BER ). All of them land nearest .
QPSK is the modulation used for the robust, low-rate modes of LTE and 5G NR and for many satellite links, and 'about 8 to 10 dB of Eb/N0 for a bit error rate near 10^-4 to 10^-5' is the rule of thumb a link-budget engineer carries around. Because QPSK halves the bit error in each quadrature independently, it reaches the same bit error rate as BPSK for the same Eb/N0 while sending twice the bits per hertz.
(7,4) Hamming code over a BSC with crossover 0.1: probability of correct decoding
For , the probability that a transmitted codeword is decoded correctly is _________ (rounded off to two decimal places).
Answer: Official key: 0.84 to 0.86
- The code word has bits and the channel flips each bit independently with probability . By the statement, decoding is correct exactly when the number of bit errors is or .
- No error: .
- Exactly one error, in any of positions: .
- Sum: . Probability of correct decoding .
Trap: counting only the error-free case, , forgets that the code corrects one error. Adding two-error patterns as well ( more) gives , but with minimum distance a two-error pattern is always decoded to a wrong codeword. The code listed all codewords of the Hamming code (), ran minimum distance decoding on all error patterns, found exactly pattern of weight and of weight decode correctly and none of any larger weight, and summed over them to get .
Hamming codes of this family are the single-error-correcting codes behind many memory ECC schemes (extended with one more parity bit to detect double errors, SECDED). The calculation shows why such codes only pay off when the raw bit error rate is low: at about of blocks still fail.
1-bit quantiser for a density on [-2, 1]: difference of the MSE-optimal output levels alpha* - beta*
Consider a 1 bit quantizer that maps positive samples to value and others to value . If and are the respective choices for and that minimize the mean square quantization error, then = _________ (rounded off to two decimal places).
Answer: Official key: 1.15 to 1.18
- Read the density from the figure: below , a straight rise from at to a peak at , a flat top of height from to , and after .
- Total area , so . On : . On : .
- For a fixed decision region, the output level that minimises the mean square error is the centroid, the conditional mean of the samples that fall in that region. The regions are fixed by the question: positive samples go to , the others to .
- Positive samples (): the density is flat, so the conditional mean is the midpoint, .
- Other samples (): the mass is and
- .
Trap: taking the midpoint of each interval, and , gives , which is outside the range because the density is not flat on the negative side: it is heavier near , which pulls the centroid to . The code integrated the density on a million point grid to get the centroids (, , ), searched a grid of pairs for the lowest mean square error (minimum at , ), and checked on million Monte Carlo samples drawn from the density.
Setting each quantiser output to the centroid of its cell is the central step of the Lloyd-Max quantiser design and of -means clustering. Real ADCs and speech or image coders use it on non-uniform signal statistics so that the levels sit where the probability mass is.
CRC-4 with generator x^4 + x + 1 applied to the bit sequence 1 1 1 0 1 0 1 0 1
- (A)
- (B)
- (C)
- (D)
Answer: (A) Official key: A
- The generator is the bit pattern (coefficients of ). It has degree , so the CRC is bits.
- Append four zeros to the message: (this multiplies the message polynomial by ).
- Divide by using XOR (no borrows). Each step takes the leading five bits that start with 1: ; then ; then ; then ; then ; then .
- The remainder is the last four bits, . The transmitted word is the message followed by the remainder: .
- Check: dividing this -bit word by leaves remainder . This is option (A).
Trap: dividing the message by without appending the four zeros gives remainder , and the word , which is option (D). The code did the GF(2) division, checked that option (A) is the only option that starts with the message and is divisible by , and encoded all nine-bit messages to confirm every codeword is divisible and the minimum distance is .
CRCs of exactly this kind protect Ethernet frames (a 32-bit CRC), USB packets, and storage and wireless link layers: the sender appends the remainder, and the receiver divides again and discards the frame if the remainder is not zero. A CRC-4 with this generator, , is also used in short-frame protocols such as ITU-T G.704.
SQNR of an 8-bit mid-rise quantizer for A cos(2 pi f0 t + theta), A in [-2,2], R_X = (2/3) cos
Answer: Official key: 45.00 to 45.30
- Signal power is the autocorrelation at zero lag: . (Check: for , and , so .)
- The signal never exceeds , so the quantizer range is to , a span of V. An -bit quantizer has levels, so the step is .
- Quantization noise power for a uniform error in : .
- .
- In dB: .
Trap: using the full-scale-sine rule . That rule assumes a sinusoid of fixed amplitude equal to the full-scale value; here the amplitude is random and the average power is only , not . A halved step size would give . The code sampled the process with random and ( samples), quantized with an -bit mid-rise quantizer over , and measured 45.15 against the closed form . Both wrong values were asserted to be outside to .
This is the standard PCM/ADC sizing step: signal power divided by gives SQNR, and it rises per bit only when the input fills the converter range. Telephone and audio codecs lose several dB of SQNR when the real signal power sits below full scale, which is why companding and gain control are used.
Minimum Hamming distance of the (6,3) binary linear block code with the given generator matrix G
Answer: Official key: 3 to 3
- For a linear code, the distance between two codewords is the weight of their sum, which is another codeword. So is the smallest weight among the non-zero codewords.
- The codewords are (mod 2). Rows: (weight ), (weight ), (weight ).
- Sums of two rows: (weight ), (weight ), (weight ). Sum of all three rows: (weight ).
- Weights of the seven non-zero codewords: . The smallest is .
Trap: the Singleton bound gives , and an answer of comes from treating that upper bound as the value. The code enumerated all 8 codewords (all distinct), listed the non-zero weights 3, 3, 3, 3, 4, 4, 4, and found the smallest is 3. A weight- word exists (), so is not reached. It checked the answer two more ways: the smallest distance over all pairs of codewords is , and the smallest set of linearly dependent columns of the parity-check matrix has columns.
Minimum distance decides how many errors a block code can detect () and correct (). A code with corrects any single-bit error, which is exactly what Hamming codes do in memory ECC and what a short control-channel code needs for a single flipped bit.
Probability that a (7, 4, 1) Hamming decoder fails to decode correctly at input bit error rate 0.10
The probability that the decoder will fail to decode a received word correctly is ______. (rounded off to two decimal places)
Answer: Official key: 0.10 to 0.20
- The Hamming code has codewords of length with minimum distance , so it corrects any single bit error (). It is a perfect code: the words within one bit of a codeword cover every received word.
- The decoder is right when the word has or bit errors. With or more errors the syndrome points to the wrong place or to a different codeword, so the decoded word is wrong.
- With bit error probability on each of the bits: and .
- .
Trap: is the probability of at least one bit error; it forgets that a single error is corrected, and lands outside the key range to . The code built the code from its parity-check matrix, found codewords with minimum distance and checked it is perfect, decoded all error patterns by syndrome and confirmed that every pattern with or more errors fails and every pattern with or error is corrected (the weighted sum is , equal to the closed form ). A Monte Carlo of 400,000 words (random codeword, bit flips, syndrome decoding) gave 0.1496. Counting only exactly errors gives , which is inside the key range by chance but ignores the -to--error patterns.
Single-error-correcting Hamming codes protect memory words in ECC RAM (extended with one more parity bit to detect double errors, SECDED) and short control fields in radio links. The calculation of a residual word-error probability from the raw bit error rate is the first step in deciding whether a code is strong enough or whether a longer code is needed.