Published Nov 20, 2026 · Last updated Nov 20, 2026 · 5 min · IndieRF
LNA Spread and Receiver Yield: 81% Meet −142 dBm
Typical values call it a pass; the worst-case stack calls it a fail.
A PSA4-5043+ LNA ahead of an SX1262 meets a −142 dBm LoRa SF12 spec on 80.7% of builds, its gain and NF spread across the data sheet limits read as ±3σ. A spreadsheet says pass at typical (−142.1 dBm) and fail at the worst-case stack (−141.69 dBm).
Three answers from one data sheet row
The PSA4-5043+ sheet, Rev. D, prints its 1.0 GHz row at 5 V as gain 16.5 dB minimum, 18.4 typical, 20.2 maximum, and NF 0.75 dB typical, 1.1 maximum, with no NF minimum. That is the nearest row with limits to 915 MHz. Gain falls and NF rises with frequency on this part (22.1 dB and 0.65 dB typical at 0.5 GHz), so using the 1.0 GHz row here errs on the pessimistic side.
The sheet prints limits and no distribution. This note reads min and max as ±3σ of a normal spread about typical, gain and NF independent. On that reading 80.7% of builds meet −142 dBm and 95.4% meet a 1.1 dB system NF; spread uniformly between the limits instead (NF mirrored about typical, 0.40 to 1.10 dB), the same builds give 64% and 77.9%.
The SX1262 sheet prints sensitivity, not NF: −137 dBm at SF12 and 125 kHz, where the demodulator needs −20 dB SNR. Backing the NF out of with B the 125 kHz channel gives 6.01 dB. That is a convention, and it moves the answer almost as far as the choice of spread does: at 5.01 dB the sensitivity yield is 89.2%, at 7.01 dB it is 65.8%.
Friis puts the pair at
so the radio's 6.01 dB costs 0.15 dB behind 18.4 dB of gain. A spreadsheet row at typical values gives 0.90 dB and −142.1 dBm, a pass by 0.1 dB, and the LNA buys 5 dB over the radio's own −137 dBm. The worst-case stack, minimum gain with maximum NF, gives 1.32 dB and −141.69 dBm: a fail, and the design gets sent back. Neither number says how many boards miss. The graph does, because the requirement sits on the SX1262 stage and the Monte Carlo draws the LNA's gain and NF for every trial.
How the yield is worked
With σ = (20.2 − 16.5)/6 = 0.62 dB on gain and σ = (1.1 − 0.75)/3 = 0.12 dB on NF, drawn independently, a build with gain g passes the NF limit L when the LNA's NF sits under . The yield is that probability averaged over the gain spread:
The sensitivity spec is the same integral with L = −142 + 143.006 = 1.006 dB. Worked by quadrature, Y is 0.8075 for sensitivity and 0.9543 for NF. The linked run, 4000 seeded trials, passes 3231 and 3823, which is 80.8% and 95.6%; four binomial σ at 4000 trials is ±2.5 points at 81% and ±1.3 at 95%, and both counts sit inside a quarter of that. A different Seed moves the sensitivity count by about ±25, one binomial σ.
A spec at −142.0 dBm and one at −141.91 dBm (the 1.1 dB NF limit) are 0.09 dB apart, about 1% of free-space range, and 15 points of yield. Near −142 dBm every 0.05 dB of spec is worth about 10 points, so write the limit after the yield run, then move it or the part.
Rule of thumb: a worst-case stack of two independent ±3σ limits is a (1 − Φ(3))² event, about 2 parts per million, so a design that fails only there still ships nearly every board. It breaks when the parameters are correlated (gain and NF from one die share a process corner, and the sheet prints no correlation) and when the limits come from a screened test, which trims the tails the ±3σ reading assumes.
Other limits on the same receiver
Each row is the NF limit at the SX1262, the sensitivity it equals at SF12, and the yield under each reading of the data sheet limits, worked by quadrature. Every row passes at typical and fails the worst-case stack.
| NF limit (dB) | Sensitivity (dBm) | Yield, ±3σ normal | Yield, uniform |
|---|---|---|---|
| 1.00 | −142.01 | 79.3% | 63.1% |
| 1.05 | −141.96 | 89.5% | 70.5% |
| 1.10 | −141.91 | 95.4% | 77.9% |
| 1.15 | −141.86 | 98.3% | 85.3% |
| 1.20 | −141.81 | 99.5% | 92.7% |
| 1.25 | −141.76 | 99.9% | 97.8% |
The two readings differ by 17 points at −142 dBm because a uniform spread puts as many parts at the limit as at typical. If the vendor publishes a measured histogram, it replaces both columns.
Cascade holds the yield in memory only, so the link carries the requirement and the Monte Carlo set-up and the reader presses Run. The draws are independent and the SX1262's 6.01 dB stays fixed; to spread it too, add a variable for it in a saved file.
References
- Mini-Circuits, PSA4-5043+ Low Noise, High IP3 Monolithic Amplifier, 50 Ω, 0.05 to 4 GHz, data sheet Rev. D (ECO-026573), page 2, Electrical Specifications at +25 °C, VD = +5.0 V, 1.0 GHz row.
- Semtech, SX1261/2 Long Range, Low Power, sub-GHz RF Transceiver, DS.SX1261-2.W.APP Rev. 1.2, June 2019, Table 3-8 (RXS_LB, SF12, 125 kHz) and Table 6-1 (LoRa demodulator SNR).
Related
Questions? Contact
Discussion
No comments yet. Start the thread with a measurement, a correction, or a worked example.