Also known as: analog-to-digital converter, ADC
An analog-to-digital converter (ADC) measures a continuous signal many times per second, turning it into a stream of numbers.1 In an SDR it produces the IQ samples that software works on, and its two headline specs — how fast it samples and how finely it resolves amplitude — bound almost everything downstream.
How it works
Conversion has two axes. Along time, the ADC samples at a fixed sample rate, which sets how much bandwidth can be captured per the Nyquist theorem and requires an anti-alias filter to keep out-of-band energy from folding into the band. Along amplitude, each sample is rounded to the nearest of a finite set of levels — this is quantization, and the tiny rounding error it introduces is quantization noise that sets a hard noise floor. An N-bit converter offers 2ᴺ levels; each extra bit adds about 6 dB of ideal dynamic range, so an 8-bit RTL-SDR resolves roughly 48 dB while a 12-bit or 14-bit radio resolves far more.
The converter’s input range defines full scale. Drive it past that ceiling and the signal clips, producing harmonics and intermodulation that no software can undo — the peak is measured in dBFS, with 0 dBFS being the ceiling. Real ADCs never reach the ideal bit count; their true resolution is the ENOB (effective number of bits), which folds in thermal noise, distortion, and clock jitter.2
Variants
- Successive-approximation (SAR) — a binary search per sample; common at moderate speeds.
- Pipeline — staged conversion for the tens-to-hundreds of MSa/s that direct-sampling SDRs need.
- Delta-sigma (ΔΣ) — heavily oversamples a coarse quantiser and shapes the noise out of band, trading speed for resolution; the basis of many HF/audio front-ends.
- Flash — a bank of comparators for the fastest, lowest-resolution conversion.
In practice
Two techniques squeeze more out of a given converter. Oversampling runs the ADC faster than Nyquist strictly requires, spreading the fixed quantization noise over a wider band so that after filtering and decimation less of it lands in the channel — each 4× of oversampling buys about one extra effective bit. Adding a small amount of dither — random noise — before quantization decorrelates the rounding error, trading a slightly higher noise floor for the removal of ugly quantization tones.
Relevance to SDR
Setting gain so strong signals stay just below the ADC’s ceiling, without burying weak ones in the quantization and thermal noise floor, is central to clean reception — the essence of gain staging. GopherTrunk assumes the incoming IQ has been digitised sanely; a capture that clipped in the ADC (peaks at 0 dBFS) is unrecoverable regardless of DSP.
Sources
-
Analog-to-digital converter — Wikipedia, on sampling a continuous signal into discrete digital values. ↩
-
Effective number of bits — Wikipedia, on the real-world resolution of an ADC after noise and distortion. ↩