Is a shared clock required for interferometry with two software-defined radios, one per receiver?
Short answer: it depends on how you plan to combine the two channels:
- If you only sum powers (incoherent combining) — e.g., make spectra from each chain separately and average/add them — you don’t need a shared clock. You’ll get ~√2 (~1.5 dB) sensitivity improvement with two independent systems.
- If you want to combine voltages coherently, do polarimetry (Stokes), beamform, or cross-correlate the two orthogonal probes, then yes, a common clock (and time alignment) is important. Without it, the two RTLs will drift in sample rate and phase, making the complex relationship between channels wander; you can sometimes fix that in software with a calibration tone/noise diode and continuous resampling, but it’s fiddly and fragile.
Practical options for your setup
- Use hardware with a shared reference
- Your RSPduo already provides two phase-locked tuners and simultaneous sampling — ideal for two orthogonal probes at H I.
- Alternatively, a coherent multi-channel RTL platform (KrakenSDR/KerberosSDR) shares a 28.8 MHz ref and solves most of the pain.
- If you must use two RTL-SDR Blog V3 sticks
- They each run from their own 28.8 MHz oscillator by default. For coherence you’d want them locked to the same 28.8 MHz reference (hardware mod; not turnkey on V3). Even then you still need:
- Deterministic start or calibration: inject a pulsed noise diode or a single pilot tone split to both RF chains to measure relative delay/phase vs. frequency.
- Ongoing correction: track small residual sample-rate offsets in software and continuously adjust one stream (fractional resampler) before forming cross-spectra.
- If you skip the clock mod, treat it as incoherent and just average powers.
- They each run from their own 28.8 MHz oscillator by default. For coherence you’d want them locked to the same 28.8 MHz reference (hardware mod; not turnkey on V3). Even then you still need:
- Front-end notes for your “two orthogonal monopoles in a cantenna”
- Two probes 90° apart can sample orthogonal linear polarisations in the dominant TE₁₁ mode, but mind probe depth/rotation, return loss, and isolation.
- Your Nooelec SAWBird H1 is LNA+SAW only (no LO), so it doesn’t affect coherence; just keep gain states matched and avoid compression.
- A split noise diode at the feed (or just ahead of the LNAs) is the easiest way to calibrate both chains simultaneously.
What to do based on your goal
- Just improve SNR: make spectra on each path → average/add → no shared clock required.
- Do Stokes I,Q,U,V or any vector/phase-aware combining: prefer RSPduo (you have one), or a coherent RTL array; otherwise be prepared for clock-share mods + calibration + software correction.
Why is a shared clock not required if you only sum powers (incoherent combining)
Just improve SNR: make spectra on each path → average/add → no shared clock required.
Because you’re not combining voltages, you’re combining powers.
- After you detect power (|x|²), the phase information is gone. A shared clock only matters when you need a stable relative phase/time between channels (coherent combining, cross-correlation, Stokes from XY). For power-summing, phase drift doesn’t enter.
- Two independent receivers looking at the same sky give two independent noise realizations. If you make a spectrum from each chain and then average/add the powers, the noise averages down as √N (radiometer equation), so two chains give ~√2 (~1.5 dB) SNR improvement — without requiring phase or sample-clock coherence.
- Sample-rate/LO offsets only shift the frequency axis, not the instantaneous power in each channel. You can register the spectra later (e.g., align known carriers/lines or interpolate onto a common frequency grid) and then average. No common reference is needed for that.
Caveats:
- Make sure the chains are independent (separate LNAs; don’t split after one LNA) so their receiver noise isn’t correlated.
- Match gains or weight the powers (e.g., inverse-variance) before adding.
- If you care about narrow spectral features, frequency-align the spectra before averaging so a ppm LO error doesn’t smear the line.
TL;DR: Power detection removes phase, so incoherent (power) combining benefits from independence, not coherence — hence no shared clock required.