Module 9: Electronics · Lesson 9.7 · about 40 min
ADCs and DACs
Data converters at GATE depth: weighted-resistor and R-2R ladder DACs, resolution, LSB and full scale, DAC errors (offset, gain, DNL, INL, monotonicity), sampling and the Nyquist rate, aliasing and sample-and-hold, quantisation error and the 6.02n + 1.76 dB rule, and the main ADC architectures: flash, successive approximation, counter ramp, dual slope and sigma-delta, with their conversion times.
Module 9 · about 40 minutes. A digital system meets the physical world through converters. A DAC is a weighted sum of reference currents, which uses the op-amp summer of lesson 9.3; most ADCs are a comparator, a DAC and the registers and counters of lesson 9.6 in a loop.
1. Digital-to-analogue converters
For an -bit code and reference :
Weighted-resistor DAC. Bit switches a resistor into the virtual ground of an inverting summer (the MSB gets ). Simple, but an 8-bit design needs resistors spanning a ratio of 128, each accurate to better than half an LSB, which is impractical.
R-2R ladder. Only two values. At every node the current meets two paths of each: the switched leg, and the rest of the ladder, which is a series followed by an equivalent (two in parallel). So the current halves at every node: the MSB branch carries , the next , and so on. Into a summer with feedback , the output is .
Worked example. An 8-bit DAC with V: 1 LSB = 19.5 mV, full scale 4.980 V. The code 10110100 = 180 gives V.
- resolution
- 1 LSB, often quoted as a percentage of full scale: `$1/(2^n - 1)$`, 0.39% for 8 bits
- offset and gain errors
- a shift and a slope error of the whole transfer line; trimmed out by calibration
- DNL
- differential non-linearity: how far each step differs from 1 LSB. If any step is negative (DNL < −1 LSB) the DAC is **non-monotonic**: a larger code gives a smaller output, which is fatal inside a feedback loop
- INL
- integral non-linearity: the largest distance of the transfer curve from a straight line
- settling time
- time to come within ±½ LSB of the final value after a code change; the worst case is the mid-scale transition 0111… → 1000…, where every bit switches
2. Sampling and quantisation
Sampling. A signal with no frequency content above is fully determined by samples taken at (Nyquist-Shannon). A frequency above produces exactly the same samples as its alias at , folded into to . Once sampled, an alias cannot be told apart from a real signal, so the analogue anti-alias filter must remove it before the converter. A sample-and-hold freezes the input while the conversion runs; without it, a signal that moves by more than 1 LSB during the conversion is converted wrongly.
Quantisation. Rounding to the nearest level leaves an error between and . For a busy signal this error is roughly uniform, with mean-square value
for a full-scale sine of amplitude . A 12-bit converter gives at most 74.0 dB, a 16-bit one 98.1 dB. A signal at one tenth of full scale loses 20 dB. Real converters add noise and distortion, and their effective number of bits is ENOB = (SINAD − 1.76)/6.02 from the measured signal-to-noise-and-distortion ratio.
Worked example. A 7 kHz tone sampled at 10 kHz aliases to 3 kHz; sampled at 6 kHz it aliases to kHz. To digitise audio up to 20 kHz at 44.1 kHz, the anti-alias filter has only the band from 20 to 22.05 kHz to fall by roughly 96 dB, which is why audio converters oversample and filter digitally instead.
3. ADC architectures
| Type | How it works | Conversion time | Typical use |
|---|---|---|---|
| flash | comparators against a resistor ladder, then an encoder | one clock | GHz rates, up to ~8 bits |
| successive approximation (SAR) | binary search with a DAC and one comparator | clocks | 8 to 18 bits, up to MHz |
| counter ramp | counter drives a DAC until it passes the input | up to clocks | simple, slow |
| tracking | up/down counter follows the input | one step per clock once locked | slowly varying inputs |
| dual slope (integrating) | integrate the input for a fixed time, then a reference down to zero | clocks or more | multimeters, high accuracy |
| sigma-delta | 1-bit converter at a very high rate in a feedback loop, then digital filtering | many clocks per output | audio, sensors, 16 to 24 bits |
Successive approximation. Try the MSB: set it to 1, compare the DAC output with the input, keep it if the DAC is below, clear it otherwise; then the next bit. A 4-bit SAR with a 0 to 8 V range (LSB = 0.5 V) converting 5.3 V tries 1000 (4 V, keep), 1100 (6 V, clear), 1010 (5 V, keep), 1011 (5.5 V, clear): the result is 1010 = 5.0 V, the input truncated to a whole LSB. (Adding half an LSB of offset to the DAC makes it round instead.)
Dual slope. Integrate for a fixed time (a fixed count), then integrate until the integrator returns to zero, which takes :
, and the clock frequency cancel, so only must be accurate. With equal to a whole number of mains periods (20 ms at 50 Hz), hum integrates to zero. With a 100 kHz clock, counts and V, an input of 1.234 V takes 1234 counts to discharge: the count reads directly in millivolts.
Sigma-delta. Oversampling by a ratio spreads the quantisation noise over a band times wider than the signal, and filtering keeps only the signal band: plain oversampling gains half a bit per doubling of . The feedback loop also pushes the noise towards high frequencies (noise shaping), so a first-order modulator gains 1.5 bits per doubling, which is how a 1-bit comparator yields 24-bit audio.
“Sampling at twice the highest frequency of interest is enough.” The Nyquist condition concerns the highest frequency present at the converter input, not the highest one you care about. Noise or interference above folds down on top of the wanted band and cannot be removed afterwards. And exactly is not enough: a sine sampled at exactly twice its frequency can land on its zero crossings every time.
- 1 LSB = ; ; full scale is one LSB below .
- R-2R ladder: two resistor values, current halves at each node; weighted resistors need a spread.
- Non-monotonic if DNL < −1 LSB; settling is worst at mid-scale.
- Nyquist ; aliases at ; filter before sampling.
- Quantisation noise ; SQNR = 6.02n + 1.76 dB (full-scale sine).
- Flash comparators, one clock; SAR clocks; ramp up to ; dual slope .
Practise: exam questions on this lesson
Next: that completes the electronics module. Lesson 10.1 explains the GATE PH paper itself: sections, marking and timing.
Going on to CSIR NET? Noise, filtering and lock-in detection covers aliasing and sizing an ADC against the noise.
Check yourself
Sources
- W. Kester (ed.), Data Conversion Handbook, Analog Devices (2005), chapters 2, 3 and 6
- P. Horowitz and W. Hill, The Art of Electronics, 3rd ed. (2015), chapter 13
- R. J. Tocci, N. S. Widmer and G. L. Moss, Digital Systems, 12th ed. (2017), chapter 11
- GATE Physics (PH) syllabus, Electronics section, published by the organising IIT each year
Your ticks stay in this browser. Save them as a file on the course home.
Updated 2026-11-22. Not affiliated with the IITs, IISc or the GATE organising institute. The general aptitude part of GATE is not covered here. Check the official GATE website for the current pattern. All questions are original.