09. Analysis of the Recorded Signal in Python¶

Purpose of the section¶

To learn how to analyze a recorded IQ signal in Python and use a scripting approach to automate measurements.

1. Why Python is needed¶

Python is useful in the course as a fast universal tool. It allows the student to:

read IQ files;
plot graphs;
compute FFT;
automate repetitive measurements;
prepare small utilities for processing recordings.

For engineering practice this is especially important because Python is well suited for quickly testing hypotheses.

2. Tasks of the analysis¶

Within the first block it is necessary to:

read the IQ data file;
convert the data to a complex signal;
plot the time waveform;
plot the spectrum;
find the main tone frequency;
save the results.

3. Recommended libraries¶

For the first analysis, the following are enough:

numpy
matplotlib

If needed, one can also use:

scipy

4. General work sequence¶

Open the file.
Read the raw array.
Separate I and Q.
Form the complex array.
Plot the time waveform.
Compute FFT.
Find the spectral peak.
Output numerical results.

5. Example Python script¶

import numpy as np
import matplotlib.pyplot as plt

filename = "tone_capture_iq.bin"
fs = 2.4e6

raw = np.fromfile(filename, dtype=np.int16)

i_data = raw[0::2]
q_data = raw[1::2]

x = i_data.astype(np.float64) + 1j * q_data.astype(np.float64)

n_view = min(len(x), 2000)
t = np.arange(n_view) / fs

plt.figure()
plt.plot(t, np.real(x[:n_view]))
plt.xlabel("Time, s")
plt.ylabel("Amplitude")
plt.title("Real part of IQ signal")
plt.grid(True)

nfft = 4096
X = np.fft.fftshift(np.fft.fft(x[:nfft], nfft))
f = np.fft.fftshift(np.fft.fftfreq(nfft, d=1/fs))

plt.figure()
plt.plot(f, 20 * np.log10(np.abs(X) + 1e-12))
plt.xlabel("Frequency, Hz")
plt.ylabel("Magnitude, dB")
plt.title("Spectrum of recorded tone")
plt.grid(True)

peak_index = np.argmax(np.abs(X))
peak_freq = f[peak_index]

print(f"Peak frequency: {peak_freq:.2f} Hz")

plt.show()

The script reads int16 samples, the course .ci16 format. A raw rtl_sdr file is unsigned 8-bit with an offset of 127.5, and an HDSDR recording is a WAV file; see the format table on page 07. The course readers of Lab 9.3 (ci16, cu8, cf32) and Lab 9.4 (WAV IQ) handle them for you.

5a. Run the course version and check it against a known tone¶

The same analysis, with the peak printed and the plot saved, is blocks/block_01_intro_sdr/python/iq_analysis_example.py. Before trusting it on a real recording, run it on a file whose answer you know. A 200 kHz tone sampled at 2.4 MS/s, written as .ci16, gives:

samples: 65536, FFT length: 4096, bin width: 585.9 Hz
peak offset from the tuning frequency: 199804.7 Hz
peak above the median spectrum level: 63.6 dB

The peak is 195 Hz short of 200 kHz. That is not an error: the FFT only has bins every fs/nfft = 585.9 Hz, and 199 804.7 Hz is the bin nearest to the tone.

python blocks/block_01_intro_sdr/python/iq_analysis_example.py capture.ci16 --fs 2.4e6 --out spectrum.png
python blocks/block_01_intro_sdr/python/iq_analysis_example.py capture.cu8 --dtype uint8

6. What the student should see¶

In the time domain¶

a stable waveform;
no random jumps;
periodic structure.

In the spectrum¶

the main spectral peak;
noise floor;
possibly weak parasitic components.

7. Advantages of Python in this block¶

Python is convenient because it allows the student to quickly:

change analysis parameters;
process several files;
automatically compute peak frequency;
save plots;
build simple measurement tools.

This is especially useful for later work with large amounts of IQ data.

8. What should be checked¶

During the analysis, make sure that:

the correct data type is chosen;
I and Q are not swapped;
the sampling frequency is set correctly;
a sufficiently long fragment is analyzed;
the frequency scale is interpreted correctly.

9. What can be improved in the next script version¶

Later the student may extend the script by:

adding a window function;
estimating the noise-floor level;
automatically finding several peaks;
saving results to a file;
building a spectrogram;
comparing several recordings.

9a. Exercises¶

Write the same 200 kHz tone as an rtl_sdr file (uint8, offset 127.5) and analyse it with --dtype uint8. Then analyse the same file with the default int16. The wrong type gives a confident, wrong answer: a peak near −400 kHz, about 48 dB above the median. Why does a format error not look like noise?
Change --nfft from 4096 to 65536. How close does the peak get to 200 000 Hz, and what is the bin width now?
Swap I and Q in the loader. Where does the tone move, and why?

10. What to include in the report¶

It is recommended to include:

the text or fragment of the Python script;
a time-domain plot;
a spectrum plot;
the found peak frequency;
a short conclusion.

11. Conclusions¶

After this stage the student should:

be able to read an IQ file in Python;
build a basic spectral analysis;
automatically find the frequency of the test tone;
use Python as a tool for engineering automation.