Published Nov 20, 2026 · Last updated Nov 20, 2026 · 6 min · IndieRF

RF Cascade Spreadsheet vs Graph: 3.9 dB NF Is 6.7 dB

The Friis sheet misses 2.86 dB of NF and 28 dB of blocker margin.

A Friis sheet puts this Wi-Fi receiver at 3.87 dB NF; the graph says 6.73 dB, with the LNA's noise at the 1588 MHz image. A Band 40 blocker hurts at −9.5 dBm on the sheet and +18.7 dBm in the graph.

The lineup, built twice

The receiver is the one from the blocker note, unchanged: a Qorvo 885128 coexistence filter, an HMC8413 LNA, and an HMC213B mixer with a low-side HMC832A LO at 2000 MHz, so the IF is 412 MHz and the image sits at 1588 MHz. The 885128 guarantees 1.7 dB maximum loss on channel 1 and 32 dB minimum rejection over 1559 to 1606 MHz. The HMC8413 is 19.5 dB and 1.9 dB NF typical. The HMC213B is 10 dB conversion loss and 21 dBm IIP3, and its sheet prints no noise figure, so both builds take the passive-mixer convention, NF equal to the conversion loss. If the real NF is 2 dB higher, the sheet reads 4.04 dB and the graph 6.82 dB, so the gap between them holds within 0.1 dB.

The spreadsheet is three rows with one gain and one NF each, read at 2412 MHz, and the Friis sum:

F=F1+F2−1G1+F3−1G1G2F = F_1 + \frac{F_2 - 1}{G_1} + \frac{F_3 - 1}{G_1 G_2}

The graph is the same three blocks with the filter's printed points as a curve, the mixer's LO and its frequency, and the LO's printed phase noise. Each carrier and each noise band is read at its own frequency.

Every number, both ways

Each graph number is Cascade on the linked lineup, checked against the two-band noise sum, the Friis and intercept sums, the Rapp weak-tone form and the phase-noise integral worked by hand. Each sheet number is the Friis sheet worked by hand. "Sets it" is the stage whose ideal version (NF 0 dB or an infinite intercept, gain kept) improves the chain most.

EffectFriis spreadsheetCascade graphExplained in
Cascaded NF3.87 dB6.73 dBImage noise
Stage that sets the NFHMC8413, 1.76 dB better if ideal885128, 2.33 dB better if idealThis note
NF with 30 dB image rejection at the mixerno cell (3.87 dB)3.85 dBImage noise
Cascaded IIP3+3.03 dBm, HMC213B+3.03 dBm, HMC213B 96%IP3 of an LNA
1 dB desense, Band 40 at 2362 MHz, compression−9.5 dBm+21.8 dBmBlocker desense
1 dB desense, same blocker, reciprocal mixingno cell+18.7 dBmBlocker desense

The NF gap is 2.86 dB. The LNA has the same 19.5 dB at 1588 MHz as at 2412 MHz, so the mixer's image port sees the LNA's full output noise instead of the kT0 its SSB figure assumes, and the front end's noise is counted twice. A Friis row has no second frequency to put it in.

The blocker gap is 28 dB. The 885128 takes 1.7 dB from the wanted carrier and 33 dB from a 2362 MHz blocker, and a row holds one of those two numbers. The graph carries the blocker as its own carrier, reads the filter at its frequency, and adds the second mechanism the sheet has no cell for: the blocker mixing with the HMC832A's phase noise at 50 MHz offset and landing it on the channel. The HMC832A data sheet prints −157 dBc/Hz at 10 MHz and −162 at 100 MHz; a straight log-log line between them gives −160.5 dBc/Hz at 50 MHz offset.

The IIP3 row is where the sheet is right. Intermodulation here is set by the mixer at the wanted frequency, both builds read the filter at 2412 MHz, and they agree to the hundredth.

The graph names a different stage

The "What limits this chain" call-out above the linked budget reads "NF 6.73 dB: Qorvo 885128 coexistence filter ... sets it, 2.33 dB better if ideal, 53% of the excess noise." The sheet blames the LNA, 1.76 dB and 56%. Both are right about their own model. At 1588 MHz the filter is 32 dB of loss at 290 K, so what leaves it at the image is its own thermal noise, kT0, and the LNA amplifies that 19.5 dB into the mixer's image port. Only a noiseless filter would remove it, so the fix is the other line in the table, image rejection between the LNA and the mixer, which the mixer's Image Rejection field stands in for. 30 dB there takes the graph to 3.85 dB, 0.02 dB from the sheet, and the sheet's own number becomes right because the mixer's image port sees kT0 again, the condition an SSB noise figure assumes.

Rule of thumb: an open image behind an LNA costs close to 3 dB of NF, because the LNA delivers the same noise at the image as at the signal and the mixer folds both in. Here it is 2.89 dB (3.84 to 6.73 dB). It shrinks when the LNA has less gain at the image than at the signal, and 30 dB of image rejection takes it to within 0.01 dB.

Effects this lineup leaves to the other notes

  • Spurs: the half-IF at 2206 MHz falls between the 885128's printed 2110 to 2170 and 2300 to 2370 MHz bands, and the HMC213B spur table was measured at 3.5 GHz, so this plan has no printed level for it. See the spur chart note.
  • Gain ripple from mismatch: the channel has no named antenna, and in the 915 MHz chain the antenna port was 65% of the ripple. See the mismatch note.
  • Yield: a Monte Carlo needs printed minimum and maximum on the parts that spread, and the HMC213B prints no NF at all. See the yield note.
  • DC power and heat: this lineup has no PA, which is where the supply and the heat sink get sized. See the power-combined PA note.
  • Jitter: no ADC clock here, though the same kind of printed LO table sets one. See the jitter note.

When the sheet is the right tool

For a linear chain someone wants to edit by hand, export it. The XLSX button above the budget writes the active path with live Friis and intercept formulas, one row per stage, and it carries an image column filled with the gain the graph read off the 885128 curve at 1588 MHz. Recalculated, it reads 6.73 dB, and typing 30 into the mixer's image rejection cell gives 3.85 dB. Each row still holds one frequency, so the blocker, the LO's phase noise and any branch stay in the graph. More worked lineups are on the Cascade examples pages.

Limits of these numbers

The filter numbers are the 885128's guaranteed worst case over −30 to +85 °C; the typical curve is better at both frequencies. The HMC8413 prints one typical gain and NF for 0.01 to 7 GHz, so it reads the same at the image. Compression uses a memoryless Rapp curve on each stage. The reciprocal-mixing figure is a scalar estimate at small-signal gains that the budget does not add to its noise rows, so the budget's NF stays 6.73 dB with the blocker present.

References

  • Qorvo, 885128 2.4 GHz WLAN/BT LTE Co-Existence Filter, Preliminary Data Sheet Rev B 07-21-15, page 2, Electrical Specifications, −30 to +85 °C.
  • Analog Devices, HMC8413 Low Noise Amplifier, 0.01 GHz to 9 GHz, data sheet Rev. 0, 10/2021, page 3, Table 1.
  • Analog Devices, HMC213BMS8E Double Balanced Mixer, 1.5 GHz to 4.5 GHz, data sheet Rev. 0, 2/2018, page 3, Table 1.
  • Analog Devices, HMC832A Fractional-N PLL with Integrated VCO, 25 MHz to 3000 MHz, data sheet Rev. B, page 5, Table 1, VCO open-loop phase noise at 2 GHz.
  • D. M. Pozar, Microwave Engineering, 4th ed., chapter 10: noise figure of cascaded networks.

Questions? Contact

Discussion

No comments yet. Start the thread with a measurement, a correction, or a worked example.