R
Member
Datasheet link budgets quote MCL = Tx + G − Rx-sens, and 155–168 dB makes it look like magic. Field success still fails because those numbers are the *noise-channel ideal*, not the *modulation-class* reality. Here's the checklist that closes the gap.
1. Sensitivity numbers vs the duty they must fill
LoRa SX127x at SF12/BW125 quotes ~−137 dBm (some "marketing −148 dBm" at SF12/BW62.5). That's under AWGN. Real fade in a plant:
- Rayleigh flat fading requires ~10–14 dB extra margin even for FSK at 1 %, even worse for chirp spread; SF12 in mobility has Doppler sensitivity;
- Interference floor: your SF12 is the "receiver ear" that hears every higher-SF collision — shared bands (ISM) mean PLRs you can't budget for with sensitivity alone; plan for an *interference margin* (6–10 dB in dense deployments) as real as cable loss.
2. The "invisible" losses
- Antenna mismatch + enclosure detuning (see the monopole thread here) — measured TRP/TRI, not S11, tells truth;
- Coiled-wire antenna near metal shelving: −6..−10 dB routinely;
- Cable: 5 m low-loss = ~1.5 dB at 868, fine, unless UV-sore connectors oxidize — that's your "why Q3 failures".
3. Fade-margin methodology that survives an audit
1. Start with the worst *sensitivity* of a production lot (−135 not −137);
2. Subtract measured TRP/TIR budget of final product;
3. Compute predicted coverage with at least log-normal shadowing σ 8–10 dB (free-space/ITU models optimistic in buildings);
4. Add margin: implementation loss (clock errors, PA compression → ACI), interference, multi-path notches;
5. Verify with 3 metrics in the field: RSSI histogram, SNR vs position, and — always PLR at fixed duty cycles for days. Link budget is a hypothesis; PLR logs are the receipt.
The "it works until the rack is stocked" case: liquid-loaded shelving shifts 3–6 dB and moves the null. Include *loaded* configuration in the test plan, or promise nothing to ops.
Anyone doing RF planning of IoT networks who's compared measured vs predicted PLR by vendor: what bias correction (distance exponent vs σ) do your models need in practice? My experience says ITU-R P.1238 indoor models over-predict coverage by 15–20 dB in warehouse aisles.
1. Sensitivity numbers vs the duty they must fill
LoRa SX127x at SF12/BW125 quotes ~−137 dBm (some "marketing −148 dBm" at SF12/BW62.5). That's under AWGN. Real fade in a plant:
- Rayleigh flat fading requires ~10–14 dB extra margin even for FSK at 1 %, even worse for chirp spread; SF12 in mobility has Doppler sensitivity;
- Interference floor: your SF12 is the "receiver ear" that hears every higher-SF collision — shared bands (ISM) mean PLRs you can't budget for with sensitivity alone; plan for an *interference margin* (6–10 dB in dense deployments) as real as cable loss.
2. The "invisible" losses
- Antenna mismatch + enclosure detuning (see the monopole thread here) — measured TRP/TRI, not S11, tells truth;
- Coiled-wire antenna near metal shelving: −6..−10 dB routinely;
- Cable: 5 m low-loss = ~1.5 dB at 868, fine, unless UV-sore connectors oxidize — that's your "why Q3 failures".
3. Fade-margin methodology that survives an audit
1. Start with the worst *sensitivity* of a production lot (−135 not −137);
2. Subtract measured TRP/TIR budget of final product;
3. Compute predicted coverage with at least log-normal shadowing σ 8–10 dB (free-space/ITU models optimistic in buildings);
4. Add margin: implementation loss (clock errors, PA compression → ACI), interference, multi-path notches;
5. Verify with 3 metrics in the field: RSSI histogram, SNR vs position, and — always PLR at fixed duty cycles for days. Link budget is a hypothesis; PLR logs are the receipt.
The "it works until the rack is stocked" case: liquid-loaded shelving shifts 3–6 dB and moves the null. Include *loaded* configuration in the test plan, or promise nothing to ops.
Anyone doing RF planning of IoT networks who's compared measured vs predicted PLR by vendor: what bias correction (distance exponent vs σ) do your models need in practice? My experience says ITU-R P.1238 indoor models over-predict coverage by 15–20 dB in warehouse aisles.
