Published Nov 20, 2026 · Last updated Nov 20, 2026 · 5 min · IndieRF
Mixer Noise Figure After an LNA: 5.0 dB, Not 2.1 dB
HMC8413 LNA into LTC5549 mixer at 5.8 GHz: 4.98 dB with no image filter, 2.07 dB on a spreadsheet Friis row.
Put an HMC8413 LNA straight into an LTC5549 mixer at 5.8 GHz: a spreadsheet Friis row says 2.1 dB and the receiver is 5.0 dB. The LNA amplifies its own noise at the 2.02 GHz image as well as at the signal, and the mixer folds both into the 1.89 GHz IF; a filter with 40 dB at the image brings it back to 2.2 dB.
Where the spreadsheet loses the image
The LTC5549 data sheet, Rev. A, page 3, prints the downmixer at 5.8 GHz RF and 3.91 GHz LO with the LO doubler off: conversion loss 8.0 dB and SSB noise figure 8.1 dB, typical. Its Note 3 says the SSB figure is measured with a bandpass filter and a 2 dB pad on the input. At the image, that pad is a room-temperature resistor, so the 8.1 dB already counts one kT0 of noise at 2.02 GHz. The HMC8413 side is its Rev. 0 data sheet, Table 1 (0.01 GHz to 7 GHz): gain 19.5 dB and NF 1.9 dB typical, the same part as the IP3 note.
A spreadsheet row enters the mixer as one more Friis stage:
, 2.07 dB.
That would be right if the mixer's image port still saw kT0. Behind 19.5 dB of gain it sees the LNA's output noise, times kT0, at 2.02 GHz too, because a single-ended LNA has gain at both frequencies. Carry the two bands separately, let the mixer add its own on top of the two kT0 its SSB figure assumes, and the same lineup is
, 4.98 dB.
A Friis row has one gain and one noise figure per stage, read at one frequency, so the LNA's noise at 2.02 GHz has no cell to live in. Cascade carries a second noise band from every mixer's image back through each block to the source, and reads each block's gain and loss there.
The rule that gain in front hides the second stage breaks at an unfiltered mixer. On the spreadsheet the LTC5549 costs 0.17 dB over the LNA's 1.9 dB; in the receiver it costs 3.08 dB, because the LNA's image noise arrives with the same gain as its signal noise. That is the LNA's delivered noise counted twice, and it shrinks only when the image is filtered or rejected.
Rejection at the image does the work
Johanson prints guaranteed limits on page 1 of each filter sheet. The 5425BP15A1050001E, Revision 5.0, has 2.0 dB maximum insertion loss over 4900 to 5950 MHz and 40 dB minimum attenuation over 1280 to 3300 MHz, which holds the 2.02 GHz image. The 5515BP15C0975001E, Revision 2.0, has 1.8 dB maximum and 30 dB minimum over 500 to 4000 MHz. Each goes into Cascade as a loss curve through those printed band edges, and the engine reads it at the signal and at the image. Every number below is the Cascade engine on the linked lineup, checked against the two-band sum worked by hand.
| Between LNA and mixer | Spreadsheet Friis NF (dB) | Cascade NF (dB) |
|---|---|---|
| Nothing (LNA straight into the mixer) | 2.07 | 4.98 |
| 2 dB pad (flat, no image rejection) | 2.18 | 5.04 |
| Johanson 5515BP15C0975001E (1.8 dB, 30 dB at the image) | 2.17 | 2.17 |
| Johanson 5425BP15A1050001E (2.0 dB, 40 dB at the image) | 2.18 | 2.18 |
| 5425BP15A1050001E, then 0.5 m of LMR-100A | 2.26 | 2.26 |
The pad and the 5425BP15A1050001E both read 2.18 dB on the spreadsheet, because each is 2 dB at 5.8 GHz. Cascade puts them 2.9 dB apart. Once the filter is in, the spreadsheet and Cascade agree to 0.001 dB, since the filter hands the mixer a room-temperature image port again, which is the condition the 8.1 dB was measured under. So the spreadsheet number is right exactly when the image is filtered, and the spreadsheet row looks the same either way.
30 dB at the image is enough here. The LNA's image noise, 138 times kT0 at the filter input, comes out at 0.14 kT0 behind 30 dB, about a seventh of the filter's own thermal noise. I'd still take the 40 dB part for 0.01 dB more NF, because the same rejection is what keeps a real 2.02 GHz signal out of the IF.
The cable row is the board-to-board case: filter on the LNA board, 0.5 m of LMR-100A to the mixer. Times Microwave prints 64.1 dB/100 ft at 5800 MHz, which is 1.05 dB at 5.8 GHz. Behind the filter the cable adds its own kT0 at the image, the image stays filtered, and the NF rises 0.08 dB.
Limits of these numbers
The HMC8413's Table 1 prints one typical gain and NF for the whole 0.01 GHz to 7 GHz range, so the LNA reads the same at 2.02 GHz as at 5.8 GHz. A part with more gain at the image makes the unfiltered row worse. The filters are at their guaranteed worst case; the typical curves are better at both frequencies. A DSB data sheet figure enters Cascade's mixer field as , 3 dB higher.
Analog Devices makes the same point in its system noise-figure article: picking SSB or DSB and substituting it into the standard Friis equation is insufficient. Here that substitution is the 2.07 dB row, 2.9 dB optimistic.
References
- Analog Devices (Linear Technology), LTC5549, 2GHz to 14GHz Microwave Mixer with Integrated LO Frequency Doubler, data sheet Rev. A, page 3, AC Electrical Characteristics and Note 3.
- Analog Devices, HMC8413 Low Noise Amplifier, 0.01 GHz to 9 GHz, data sheet Rev. 0, 10/2021, page 3, Table 1.
- Johanson Technology, 5425BP15A1050001E 5.4 GHz Band Pass Filter, Revision 5.0, page 1, General Specifications.
- Johanson Technology, 5515BP15C0975001E 5.5 GHz Band Pass Filter, Revision 2.0, page 1, General Specifications.
- Times Microwave Systems, LMR-100A data sheet, attenuation table.
- D. M. Pozar, Microwave Engineering, 4th ed., chapter 10: noise figure of cascaded and passive networks.
Related
Questions? Contact
Discussion
No comments yet. Start the thread with a measurement, a correction, or a worked example.