IndieRF RF Cascade Analyzer
A 2 dB filter, a 20 dB LNA with a 1.5 dB noise figure, and a 12 dB driver with a 6 dB noise figure cascade to 3.59 dB. Put the LNA first and the same three parts read 1.66 dB. The canvas above reads gain, Friis noise figure, P1dB, IP3, and SFDR at every stage of a receive or transmit chain, with splitters, combiners, and mixers. It is free, and the math runs in your browser.
Noise figure cascades by dividing each stage's excess noise by the gain in front of it. With F and G as linear ratios (F = 10NF/10):
F = F1 + (F2 − 1)/G1 + (F3 − 1)/(G1G2) + …
A passive part's noise figure equals its loss, so the filter is a 2 dB noise figure stage with −2 dB of gain. This is the example at 2.45 GHz in 20 MHz, read stage by stage from the same solver the canvas uses.
| Stage | Gain (dB) | NF (dB) | Cum. gain (dB) | Cum. NF (dB) | NF added (dB) | Cum. IIP3 (dBm) |
|---|---|---|---|---|---|---|
| Band-pass filter | −2.0 | 2.00 | −2.0 | 2.00 | 2.00 | n/a |
| LNA | 20.0 | 1.50 | 18.0 | 3.50 | 1.50 | 12.0 |
| Driver amplifier | 12.0 | 6.00 | 30.0 | 3.59 | 0.09 | −5.1 |
The filter alone reads 2.00 dB. The LNA lifts the chain to 3.50 dB because the filter's loss sits in front of all of the LNA's gain. The driver adds only 0.09 dB, since 18 dB of net gain is ahead of it. Move the filter behind the LNA and the chain reads 1.66 dB: the same filter costs 0.1 dB there and 2.0 dB in front. On a receiver that is about 2 dB of sensitivity. A preselector in front of the LNA can still be worth that price when a nearby transmitter would overload the LNA, and the budget tells you what the protection costs.
Gain in front hides the second stage. With 20 dB of gain ahead of it, a 10 dB NF second stage adds about 0.3 dB to a 1 dB NF first stage. With 10 dB ahead, the same stage adds about 2.3 dB. It stops working when the front end is lossy: a filter has negative gain, so it hides nothing and every stage behind it counts in full.
Intercepts add as reciprocals in milliwatts after each stage is referred back to the chain input. A stage's IIP3 is its OIP3 minus its gain.
1/IIP3 = 1/IIP31 + G1/IIP32 + G1G2/IIP33 + …
The example reads an IIP3 of −5.1 dBm at the input and an OIP3 of 24.9 dBm at the output. The driver's own OIP3 is +25 dBm, so the filter and the LNA move the answer by less than 0.1 dB. The last stage sets chain linearity because it sees all the gain. Cascaded P1dB uses the same reciprocal sum at the output: take each stage's output P1dB plus the gain behind it, then add the reciprocals in milliwatts. The LNA's +20 dBm becomes +32 dBm at the chain output and the driver's +15 dBm is lower, so the chain output P1dB is +14.9 dBm and the driver is the first stage to compress. The sum assumes every stage compresses softly.
OIP3 sits about 10 dB above OP1dB. Good for a class-A small-signal amplifier, and it is the gap both amplifiers in the example use. It breaks for feedback amplifiers, class-AB parts, and mixers, where the gap can be 5 to 15 dB. Read the datasheet before you lean on it.
SFDR = (2/3)(OIP3 − N), where N is the integrated noise at the output in the channel bandwidth. The example has an OIP3 of 24.9 dBm and −67.4 dBm of output noise in 20 MHz, so SFDR is 61.5 dB. N falls 10 dB per decade of bandwidth, so cutting the bandwidth tenfold buys about 6.7 dB of SFDR. Treat NF, IM3, and SFDR with caution once a stage compresses, because the small-signal equations stop holding.
A receiver that splits to two chains, or two transmitters that merge into one antenna, does not fit a table. The canvas is a graph. A splitter divides power by 10·log10(N) plus its excess loss, a combiner is an explicit node, and every signal keeps its source and frequency, so two carriers at different frequencies are never added as if they were in phase. Budgets for a branch are read along the path you select.
Each link opens the canvas with that chain loaded. Edit any block and the budget recalculates. Every template solved, with its level diagram.
Intermediate · 2.45 GHz RF · 140 MHz IF
Preselector → limiter → LNA → mixer → IF filter → VGA → ADC
A protected 2.45 GHz receiver front end with a 140 MHz IF strip, AGC, and ADC.
41.6 dB gain and 3.57 dB noise figure at the ADC.
Starter · 1.2 GHz center · 200 MHz bandwidth
Preselector → limiter → LNA → attenuator → VGA → anti-alias filter → ADC
A mixerless 1.2 GHz receive chain that trades front-end noise figure against ADC headroom.
20.2 dB gain and 14.25 dB noise figure at the ADC.
Intermediate · 28 GHz · 100 MHz channel
1 GHz IF → upconverter → driver → PA → antenna
A 28 GHz FR2 transmitter with explicit IF, LO, upconversion, driver, PA, and output filtering.
Advanced · 2.2 GHz · 20 MHz channel
Driver → 2-way split → dual PA → coherent combine → coupler
A coherent split-amplify-combine module with branch phase controls and coupled-power monitoring.
Advanced · 27.5–30.5 GHz · 1 GHz steps
1 GHz IF → tunable LO/upconverter → driver → PA → BPF → line → antenna
A tunable 27.5–31 GHz uplink transmitter with five-point component characterization and a ready-to-run analysis.
Checked against hand math: the cases, the tolerance and what the check does not cover.
Convert each stage's noise figure and gain from dB to linear, then add each stage's excess noise divided by the gain ahead of it: F = F1 + (F2 − 1)/G1 + (F3 − 1)/(G1·G2), and NF = 10·log10(F). A 2 dB band-pass filter, a 20 dB gain LNA with a 1.5 dB NF, and a 12 dB gain amplifier with a 6 dB NF cascade to 3.59 dB. The canvas does this at every stage, so you can read the running NF down the budget table.
Every later stage's noise is divided by the gain in front of it. The 6 dB NF amplifier in the example adds 0.09 dB to the chain because 18 dB of net gain sits ahead of it. Loss is the other half of the story: a 2 dB filter ahead of the LNA costs 2.0 dB of system NF, and the same filter behind the LNA costs 0.1 dB. The chain reads 3.59 dB with the filter first and 1.66 dB with the LNA first.
Refer each stage's IIP3 back to the chain input by dividing by the gain ahead of it, then add the reciprocals in milliwatts: 1/IIP3 = 1/IIP3₁ + G1/IIP3₂ + G1·G2/IIP3₃. The example chain has an input-referred IIP3 of -5.1 dBm, which is 24.9 dBm at the output. The last stage sets the result because it sees all the gain, so a 25 dBm OIP3 driver after a 30 dBm OIP3 LNA lands within 0.1 dB of the driver alone. The canvas applies this equation on a single-source path. Where two signals meet at a combiner, it follows the coherent or incoherent setting you pick.
The example chain's output P1dB is 14.9 dBm. Move each stage's output P1dB to the chain output by adding the gain behind it, then add the reciprocals in milliwatts: 1/OP1dB = 1/(OP1dB₁·G₂) + 1/OP1dB₂ for two stages. The LNA's 20 dBm becomes 32 dBm at the output of the 12 dB driver and the driver's own is 15 dBm, so the chain lands 0.1 dB under the driver alone and the driver is the first stage to compress. Two equal terms land 3 dB under either one. The sum is an approximation that assumes every stage compresses softly. The canvas names the first stage to compress and shows the headroom from the signal level to the cascaded value.
SFDR = (2/3)·(OIP3 − N), where N is the integrated noise at the output in the same bandwidth. The example chain has an OIP3 of 24.9 dBm and -67.4 dBm of output noise in 20 MHz, so SFDR is 61.5 dB. Change the bandwidth and N moves 10 dB per decade, which is why a narrower channel buys SFDR.
A matched 50 Ω scalar budget. Enter each block's return loss and the budget adds a mismatch ripple bound, worst case and RSS: 10 dB against 10 dB reads +0.92 / −0.83 dB per interface. It takes reverse isolation as infinite, so a passive part between two mismatched ports can pass a reflection back and push the gain past it. There is no return-loss cascade and no stability check, so a VNA or a circuit simulator is still the check before a board spin. A filter without a loss curve is a scalar loss with a matching noise figure. With a curve, each carrier takes the loss at its own frequency and the noise rows integrate the curve across the channel. Distortion is IM3 only. A mixer noise figure is single-sideband as entered. Image noise counts the mixer's own image rejection and every curve upstream read at the image.
Yes. The canvas is a graph, not a table with a fixed number of rows. A splitter can feed two receive chains, a combiner can merge two transmitters, and each signal keeps its source and frequency through the lineup, so two carriers at different frequencies are never added as if they were in phase. Metrics for a branch are read along the path you select.
It is free. The calculation, the examples, CSV export, and share links need no account. The math runs in your browser, and an unsigned session autosaves in that browser so a refresh keeps your work. A share link carries the design in the URL hash, which the browser does not send to the server. Signing in only adds saving a project to the cloud.