ISRO Scientist/Engineer SC (Electronics) common mistakes
ISRO Scientist/Engineer SC (Electronics) · Understand the traps and how to avoid them.
Using the hybrid-pi small-signal model with g_m = I_C/V_T but then substituting I_C computed with the wrong bias point (e.g., taking the collector current as the supply voltage divided by the collector resistor, ignoring the drop across R_C and the transistor's own operating point), which shifts every gain, input resistance, and bandwidth answer by a factor of two or more.
Candidates memorize the formulas r_pi = beta/g_m and g_m = I_C/V_T but practice them on circuits where the quiescent current is obvious. In a real bias network with emitter degeneration and a divider base, the DC operating point needs a Thevenin reduction of the base divider first, and under exam time pressure people skip that step and assume a convenient current.
What to do instead
Before touching any small-signal formula, write the DC bias solution explicitly: Theveninize the base divider (V_BB, R_BB), assume V_BE = 0.7 V, solve I_E = (V_BB - V_BE)/(R_E + R_BB/(beta+1)), then set I_C ≈ I_E and g_m = I_C/V_T with V_T = 26 mV at room temperature. Only after that number is written down should you compute r_pi, r_o, and gains. Practice ten bias networks until the DC step is automatic.
Treating the input resistance of an inverting op-amp configuration as infinite (or as R_in of the op-amp) instead of recognizing it as the value of the input resistor R1, and similarly mislabeling the virtual-ground node as a true ground when computing output impedance or noise gain.
The ideal op-amp rules (infinite input impedance, zero output impedance) are learned first and then applied to the whole circuit rather than to the amplifier block. The virtual short means the inverting node sits at the same potential as the non-inverting input but is not a low-impedance node, and this distinction is exactly what the exam tests.
What to do instead
Separate the two levels: ideal op-amp terminals (infinite Z_in at the input pins, zero Z_out at the output pin) versus closed-loop circuit input impedance (R1 for inverting, infinite for non-inverting, R1+R2 to virtual ground when computing noise gain). Draw the virtual-ground node with a dashed line and label it 'virtual, not real ground' every time. Then verify with a quick sanity check: an inverting amplifier driven by a current source would have infinite gain, so its input resistance must be R1.
Reducing a Boolean expression or a Karnaugh map and stopping at a sum-of-products form that is minimal in literal count but not hazard-free, or failing to identify the essential prime implicants before writing the expression, producing an answer that differs from the intended minimal form.
Textbook practice rewards 'smallest expression', so candidates circle the largest groups and move on. Exam questions often ask for the minimal SOP with a specific number of terms, or ask which implicant is essential, and a group that looks large may not cover a minterm that only one implicant covers.
What to do instead
Follow a fixed procedure: list minterms, mark all prime implicants, identify essential ones (minterms covered by exactly one prime implicant), then cover remaining minterms with the fewest additional implicants, preferring larger groups. For static-1 hazards, add the consensus term covering the adjacent 1s. Always verify by evaluating the final expression at two or three minterms and at one don't-care.
Applying the sampling theorem as 'f_s > 2f_max' without checking whether the signal is bandpass rather than baseband, and then reporting the minimum sampling rate as 2f_max when the correct answer for a bandpass signal occupying [f_L, f_H] is 2f_H/n for the largest integer n with f_s ≥ 2(f_H - f_L).
The baseband Nyquist rule is over-learned and the bandpass version is a single formula that is easy to forget. Candidates also forget that sampling a sinusoid exactly at 2f produces zero-valued samples, so a strict inequality is required.
What to do instead
Classify the signal first: baseband (spectrum from DC) or bandpass (spectrum in a band away from DC). For bandpass, compute n = floor(f_H/(f_H - f_L)) and then f_s = 2f_H/n, checking that f_s ≥ 2B. Memorize the aliasing map: a component at f maps to |f - k f_s|, so fold every component into [0, f_s/2] before declaring no aliasing.
Computing the Fourier transform or Laplace transform of a two-sided or causal signal and getting the region of convergence wrong or omitting it entirely, then concluding two different signals have the same transform.
Most classroom problems are causal and right-sided, so the ROC is 'obviously' Re(s) > something and gets dropped. Exam questions deliberately pair a right-sided and a left-sided exponential with the same algebraic transform to test whether the candidate knows the ROC distinguishes them.
What to do instead
For every transform, write the ROC as part of the answer. Remember: right-sided → ROC to the right of the rightmost pole; left-sided → to the left of the leftmost pole; two-sided → a vertical strip between poles; finite-duration → entire s-plane. For the Fourier transform, the ROC must include the jω axis. Check stability of an LTI system by asking whether the jω axis lies in the ROC.
Calculating the free-space path loss or received power in dB without converting the transmit power and antenna gains consistently, e.g., mixing dBm and dBW, or forgetting that the Friis equation's path loss term is (4πd/λ)^2 and using d in km with λ in meters.
The link-budget equation is a chain of multiplications that becomes additions in dB, and unit slips hide inside the additions. Candidates also forget the 20 log10(4π/c) constant and try to re-derive it under time pressure.
What to do instead
Write the link budget in one canonical form: P_r(dBm) = P_t(dBm) + G_t(dBi) + G_r(dBi) - 20log10(d_m) - 20log10(f_Hz) - 20log10(4π/c), or use FSPL(dB) = 32.44 + 20log10(d_km) + 20log10(f_MHz). Convert every quantity to the canonical units before substituting, and carry the units through the arithmetic as a check.
Solving a root-locus or Routh-Hurwitz stability problem by counting right-half-plane roots of the characteristic polynomial but confusing the number of sign changes in the first column of the Routh array with the number of unstable poles, or mishandling a row of zeros by not using the auxiliary polynomial.
The Routh procedure has special cases (zero first element, entire row of zeros) that are taught briefly and rarely practiced, and the sign-change counting rule is easy to state but easy to invert under pressure.
What to do instead
Memorize the rule exactly: the number of sign changes in the first column equals the number of closed-loop poles in the right half-plane. For a zero first element, replace it with ε and take the limit; for a row of zeros, form the auxiliary polynomial from the row above, differentiate it, and substitute the derivative coefficients. After solving, cross-check with the root-locus asymptote angles and centroid, which must be consistent with the number of unstable branches.
Answering microprocessor questions about the 8085/8086 by recalling the mnemonic but not the flag effects, the addressing mode, or the exact T-state count, and then choosing an option that is 'roughly right' on timing (e.g., saying a CALL takes the same cycles as a JMP, or that MOV affects flags).
Candidates study instruction sets as vocabulary lists rather than as tables of opcode, addressing mode, flags affected, and machine cycles. The exam tests the table entries, and 'roughly right' is wrong.
What to do instead
Build a one-page table for the instructions most often tested: MOV, MVI, LDA/STA, LXI, ADD/ADC, INR/DCR, CMP, JMP/Jcc, CALL/RET, PUSH/POP, and the string instructions on 8086. For each, record bytes, T-states, flags affected, and addressing mode. Note the invariants: MOV never affects flags; INR/DCR affect all flags except CY; CMP is a subtract that discards the result. Then do timed recall drills.
Computing the directive gain or half-power beamwidth of an antenna array by adding element patterns as scalars, ignoring the array factor phase, or using the element spacing in wavelengths without converting from meters at the operating frequency.
The array factor is a phasor sum, and the pattern depends on the progressive phase shift ψ = kd cosθ + β. Candidates treat the array as a product of amplitudes and lose the nulls and the beam steering entirely.
What to do instead
Always write AF = Σ exp(j n ψ) with ψ = kd cosθ + β, then use the closed form |AF| = |sin(Nψ/2)/sin(ψ/2)|. Convert spacing to wavelengths (d/λ) before substituting, and remember the principal maximum occurs at ψ = 0, giving the steering angle cosθ_0 = -β/(kd). Check the result: N elements with d = λ/2 should give a beamwidth narrower than a single element and N-1 nulls between grating lobes.
On spacecraft electronics and instrumentation questions, quoting a component or bus specification (e.g., total ionizing dose tolerance, single-event latchup immunity, or a bus data rate) from a datasheet memory without checking whether the question is about the commercial, industrial, or space-qualified grade, and ignoring derating and radiation-latchup mitigation requirements.
The same part number family spans grades, and the exam phrases questions around the qualification level rather than the part. Candidates also conflate TID (cumulative dose in krad) with SEE (single-event effects, which are instantaneous and not dose-dependent).
What to do instead
Separate the two radiation failure classes explicitly: TID causes parametric drift and is specified in krad(Si) with a dose-rate dependence; SEE causes transient upsets or destructive latchup and is specified by LET threshold in MeV·cm²/mg. For any part question, first identify the grade (commercial 0 to 70 °C, industrial -40 to 85 °C, military -55 to 125 °C, space-qualified per the relevant screening standard) and apply the derating margin before quoting a number. When a specific figure is not certain, state the qualification class and the mitigation (shielding, watchdog, EDAC, latchup-current limiting) rather than a guessed number.