Lab 6.9 — RTL-SDR vs AD936x receiver quality and ADC resolution

Goal

Compare an RTL-SDR receiver with a Pluto-compatible Zynq + AD936x receiver under the same RF stimulus and separate three effects that are often mixed together:

1. the complete receiver-path difference;
2. the influence of output sample resolution;
3. the remaining difference when the AD936x capture is requantized to 8 bits.

The lab does not claim that software requantization isolates the physical ADC perfectly. It provides a controlled estimate that is useful for engineering comparison and for understanding why nominal ADC bit depth is not the same as usable system dynamic range.

Equipment

Recommended conducted setup:

Zynq/AD936x TX
      |
30 dB fixed attenuator
      |
0...31.75 dB step attenuator
      |
2-way 50 ohm splitter
   +--+------------------+
   |                     |
RTL-SDR RX        second Zynq/AD936x RX

Required items:

• RTL-SDR with manual tuner gain;
• Zynq-7020 + AD936x board running the Pluto-compatible image;
• a second board or another stable RF source for transmission;
• 50 ohm coaxial cables;
• at least 30 dB of fixed attenuation;
• a two-way splitter/combiner rated for the selected frequency;
• optional digital step attenuator;
• optional NanoVNA for cable/splitter loss characterization;
• optional tinySA for checking transmitter harmonics and unexpected spurs.

Do not connect an enabled transmitter directly to either receiver without a verified attenuation budget.

What is being compared

Comparison	Interpretation
RTL-SDR native 8-bit vs AD936x native	Complete receiver difference: RF frontend, LO, filters, gain, ADC, clocking and sample format
AD936x native vs the same samples requantized to 8 bits	Quantization-only sensitivity using the same analog capture
RTL-SDR 8-bit vs AD936x requantized to 8 bits	Approximate same-resolution path comparison
AD936x 6/8/10/12-bit sweep	How the recorded waveform responds to reduced sample resolution

The third row is still not a pure analog-front-end measurement. Differences in gain calibration, analog bandwidth, clock phase noise, DC removal and sample-rate processing remain.

Metrics

The executable analysis reports:

• tone level in dBFS;
• RMS and peak level in dBFS;
• DC level;
• SNR in a common analysis bandwidth;
• SINAD;
• SFDR;
• noise density in dBFS/Hz;
• system ENOB estimated from SINAD;
• clipping fraction;
• detected tone frequency and frequency error.

The system ENOB is calculated as

ENOB_system = (SINAD - 1.76) / 6.02

This is an end-to-end receiver result, not a datasheet characterization of the standalone ADC.

Why equal settings matter

Before recording, document and keep fixed:

• RF input power at the splitter input;
• center frequency and tone offset;
• manual receiver gain;
• RF bandwidth;
• sample rate;
• cable and splitter losses;
• capture duration;
• input port;
• AGC state;
• DC/IQ correction state;
• analysis bandwidth.

The two receivers may use different sample rates, but the analysis bandwidth must be common and must fit inside both Nyquist bands. Never normalize each capture to its own maximum before calculating dBFS, clipping or noise-density metrics.

Measurement sequence

1. Characterize passive losses

Use the NanoVNA or a trusted reference to record:

Element	Frequency	Insertion loss	Notes
fixed attenuator	____ MHz	____ dB
step attenuator	____ MHz	____ dB	selected setting
splitter path A	____ MHz	____ dB	RTL-SDR branch
splitter path B	____ MHz	____ dB	AD936x branch
cable A	____ MHz	____ dB
cable B	____ MHz	____ dB

Swap the receiver branches once. If the result changes with the splitter port, the passive-path asymmetry must be included in the uncertainty budget.

2. Configure both receivers

Use manual gain. Choose a tone level that produces approximately -10...-20 dBFS on both receivers without clipping. Record a terminated-input capture for each receiver before enabling the source.

Suggested starting point:

Parameter	Value
RF frequency	915 MHz
tone offset	+120 kHz
RTL-SDR sample rate	2.4 MS/s
AD936x sample rate	2.4 or 3.84 MS/s
common analysis bandwidth	1.8 MHz or less
capture length	at least 65 536 complex samples

3. Record matched IQ files

Recommended formats:

• RTL-SDR: interleaved unsigned CU8;
• AD936x/libiio: interleaved signed CI16 with 12 significant bits.

Record metadata beside each file. The CI16 container width is not the ADC resolution; pass the number of significant bits explicitly to the analyzer.

4. Run the offline comparison

Synthetic CI baseline:

python blocks/block_06_rf_frontend_and_ad9363/python/lab_6_9_compare_receivers.py

Real captures:

python blocks/block_06_rf_frontend_and_ad9363/python/lab_6_9_compare_receivers.py \
  --rtl-iq captures/rtl_915m_tone.cu8 \
  --rtl-format cu8 \
  --rtl-stored-bits 8 \
  --rtl-sample-rate-hz 2400000 \
  --ad936x-iq captures/ad936x_915m_tone.ci16 \
  --ad936x-format ci16 \
  --ad936x-stored-bits 12 \
  --ad936x-sample-rate-hz 3840000 \
  --tone-offset-hz 120000 \
  --tone-search-span-hz 20000 \
  --analysis-bandwidth-hz 1800000 \
  --quantize-bits 6,8,10,12

Outputs:

• docs/assets/lab69_receiver_comparison_metrics.json;
• docs/assets/lab69_receiver_comparison_sinad.png;
• docs/assets/lab69_receiver_comparison_noise_density.png.

5. Sweep input power

Increase attenuation in 5 or 10 dB steps. At each point record:

Input level or attenuation	Receiver	Tone dBFS	SNR	SINAD	SFDR	BER/EVM if modulated	Clipping
____	RTL-SDR
____	AD936x native
____	AD936x -> 8 bit

For a QPSK extension, add BER and EVM. A receiver can show an apparently good spectral SNR while BER remains poor because of CFO, impulsive interference, compression or synchronization failures.

6. Strong-blocker extension

Add a strong adjacent carrier while keeping the wanted signal fixed. Increase the wanted-to-blocker ratio until synchronization or BER fails. This experiment is usually more informative than a single-tone noise-floor comparison because extra ADC headroom matters most when a weak wanted signal coexists with a strong interferer.

Required conclusions

The report must answer:

1. What is the full native SINAD/SFDR advantage of one receiver over the other?
2. How much does AD936x SINAD degrade when the same samples are reduced to 8 bits?
3. After equalizing resolution at 8 bits, how much difference remains?
4. At what input level does each receiver stop behaving linearly?
5. Does the result follow nominal bit depth, or is it limited by analog noise, spurs, gain distribution or clock quality?
6. Are SNR conclusions consistent with BER/EVM for the modulated-signal extension?

Limitations

Without access to internal analog nodes, this experiment cannot independently measure the RTL2832U ADC, tuner noise figure and tuner linearity. It measures the receiver as a system. A calibrated noise source and power meter are required for a traceable noise-figure measurement. Without them, report practical sensitivity and uncertainty rather than claiming a calibrated noise figure.