A practical recipe for successfully improving signal detection using two orthogonal monopoles in a cantenna.

A practical recipe for successfully improving signal detection using two orthogonal monopoles in a cantenna.

Great — here’s a practical recipe you can use with your setup (2 orthogonal probes, 2 LNAs, 2 SDRs). I’ll outline it in steps, then give you the minimal processing pipeline.

---

1. Hardware layout

• Antenna: two monopole probes at 90° inside your cantenna (sampling two orthogonal linear polarisations).
• Front-ends: each probe → SAWBird H1 → RTL-SDR (or RSPduo channels if you go that route).
• Calibration source: noise diode or RF tone generator, split equally to both probes (or injected before the LNAs).

---

2. Recording streams

• Use GNU Radio (or rtl_sdr + SoX) to record complex64 IQ streams from both SDRs simultaneously.
• Ensure same center frequency (1420.4058 MHz) and sample rate.
• Save with metadata in filenames:
• ProbeX_2025-09-21T1200Z_1420MHz_2Msps.c64
• ProbeY_2025-09-21T1200Z_1420MHz_2Msps.c64
• If using RTLs without shared clock: expect drift → you’ll fix later.
• If using RSPduo: channels are already coherent, no drift problem.

---

3. Calibration step

• Inject noise/tone (short burst) every 5–10 minutes.
• In processing, find this burst in both streams.
• Estimate:
  • Relative delay (cross-correlation peak).
  • Relative phase (argument of cross-spectrum at tone frequency).
• Store correction factors.

---

4. Processing pipeline (Python + NumPy/Scipy)

For each calibration interval:

1. Align streams
   • Apply fractional sample delay correction.
   • Apply complex phase rotation.
2. Form spectra
   • FFT both aligned streams → X(f), Y(f).
3. Compute cross-products
   • Auto-powers:
     • XX = X · X*
     • YY = Y · Y*
   • Cross-power:
     • XY = X · Y*
4. Form Stokes parameters
   • I = XX + YY
   • Q = XX – YY
   • U = 2 · Re(XY)
   • V = 2 · Im(XY)

---

5. Incoherent fallback (if no clock share)

• Skip delay/phase alignment.
• Just compute XX and YY, then add: I = XX + YY.
• Sensitivity improvement ~√2.
• Q/U/V won’t be reliable.

---

6. Minimal GNU Radio flowgraph

• Two RTL-SDR Source blocks (or RSPduo Source).
• Set center freq = 1420.4058 MHz, sample rate ~2 MHz.
• Record to files (File Sink) → complex64.
• Optional: FFT Sink for live spectrum view.

---

7. Python skeleton (for later analysis)

import numpy as np

# Load IQ streams

x = np.fromfile(“ProbeX.c64”, dtype=np.complex64)

y = np.fromfile(“ProbeY.c64”, dtype=np.complex64)

# — Calibration example with tone/noise burst —

# Cross-correlate

corr = np.fft.ifft(np.fft.fft(x) * np.conj(np.fft.fft(y)))

delay = np.argmax(np.abs(corr))

phase = np.angle(corr[delay])

# Apply corrections

y_corr = np.roll(y, -delay) * np.exp(-1j*phase)

# — FFTs —

X = np.fft.fft(x, n=4096)

Y = np.fft.fft(y_corr, n=4096)

# Stokes

XX = X * np.conj(X)

YY = Y * np.conj(Y)

XY = X * np.conj(Y)

I = XX + YY

Q = XX – YY

U = 2 * np.real(XY)

V = 2 * np.imag(XY)

---

⚠️ If you want to do this with two RTL-SDR Blog V3s, you’ll need to handle clock drift (resample one stream continuously). With RSPduo, you can skip that — it’s already coherent, and you can get true polarimetry much more easily.

---