Lab 9.3 — Multi-Format IQ Reader¶

Goal¶

Extend the IQ analysis workflow from CI16-only reading to a metadata-driven reader that supports ci16, cu8 and cf32 captures.

Engineering question¶

How can one analysis script read IQ recordings from different tools without rewriting the processing pipeline?

Supported formats¶

Format	Binary layout	Typical source
`ci16`	signed int16 interleaved I/Q	AD9363, custom SDR
`cu8`	unsigned uint8 interleaved I/Q	RTL-SDR raw captures
`cf32`	float32 interleaved I/Q	GNU Radio, MATLAB/Python exports

Executable file¶

python blocks/block_09_recording_and_analysis_tools/python/lab_9_3_multi_format_iq_reader.py

Generated artifacts:

docs/assets/lab93_multiformat_iq_metrics.json
docs/assets/lab93_multiformat_iq_spectrum_ci16.png
docs/assets/lab93_multiformat_iq_spectrum_cu8.png
docs/assets/lab93_multiformat_iq_spectrum_cf32.png

Processing chain¶

flowchart LR
    META[Metadata JSON] --> DISPATCH[Format dispatcher]
    DISPATCH --> CI16[CI16 reader]
    DISPATCH --> CU8[CU8 reader]
    DISPATCH --> CF32[CF32 reader]
    CI16 --> FFT[Common FFT analysis]
    CU8 --> FFT
    CF32 --> FFT
    FFT --> REPORT[Metrics + plots]

Why this matters¶

Real SDR experiments often mix tools. A course capture may start from HDSDR/RTL-SDR, then move to AD9363, then be exported through GNU Radio or MATLAB. The analysis stage should be metadata-driven, not hard-coded for one binary layout.

What to expect¶

ci16: peak=124987.793 Hz, SNR=73.22 dB
cu8: peak=124987.793 Hz, SNR=73.14 dB
cf32: peak=124987.793 Hz, SNR=73.23 dB

All three formats give the same peak. The 8-bit CU8 file loses only 0.1 dB of "SNR" here. That does not mean 8 bits are as good as 16: the measured figure is peak over median FFT bin, which includes the FFT's processing gain over a long record, and the synthetic tone's own noise is far above the 8-bit quantization noise spread over that many bins.

Exercises¶

Remove the synthetic noise and rerun. Which format now shows a lower floor, and by roughly how much (hint: about 6 dB per bit)?
Add a second tone 50 dB weaker than the first. In which format does it disappear first?

Report checklist¶

List all tested IQ formats.
Show metadata for each format.
Confirm sample count and scaling.
Compare measured peak frequencies.
Compare SNR/DC/clipping indicators.
Explain which format is best for the next processing step.

Engineering conclusion template¶

The same analysis pipeline successfully read formats ____ and produced consistent peak-frequency estimates.
The largest frequency error was ____ Hz. The preferred format for further processing is ____ because ______.