Understanding IP3: What It Is and What It Isn’t

I use IP3 because third-order products can land uncomfortably close to the signals that created them. I do not use it as shorthand for “this receiver handles everything.” The number is meaningful only with its input or output reference, power per tone, tone frequencies and spacing, gain state, measurement plane, impedance, bandwidth and verified extrapolation region.

Two Tones Create More Than Two Output Frequencies

Take two equal sinusoidal input tones at frequencies f₁ and f₂. A perfectly linear device would reproduce only those frequencies, multiplied by gain or conversion gain. A practical amplifier, mixer or receiver contains nonlinear terms. Its third-order response can produce, among other components:

When the parent tones are close, these two IM3 products sit just outside them and can fall inside the useful passband. If the tones are farther apart, the products are farther apart too. “Close to the wanted signal” is therefore a result of the chosen tone geometry and receiver filtering, not an intrinsic property of every third-order product.

Second-order products such as f₁ + f₂ and |f₁ − f₂| may land elsewhere, but they can still be decisive in a wideband or direct-conversion system. High IP3 says nothing by itself about IP2.

The Intercept Exists on Extrapolated Lines

In the small-signal region of the familiar cubic model, each fundamental output rises approximately 1 dB when its input tone rises 1 dB. A third-order product rises approximately 3 dB. On a logarithmic power plot, extrapolate those two straight lines until they meet. That hypothetical crossing is IP3.

The device normally compresses, changes gain state or reaches another limit before the crossing. The fundamental and IM3 curves then stop following their ideal slopes. IP3 is therefore not directly reached during a valid measurement and is not proof that the hardware can accept that much input or deliver that much output.

The 1:1 and 3:1 slopes are assumptions to verify, not decorations on the graph. NIST measurements of software-defined receivers found IM3 behaviour that departed from the expected 3 dB/dB slope and used regression to characterize the actual response. A one-level calculation can look precise while extrapolating the wrong region.

IIP3 and OIP3 Name Different Reference Planes

• IIP3 is the extrapolated third-order intercept referred to the input.
• OIP3 is the same model referred to the output.

For a stage with gain GdB measured at the same frequency, load and gain state in the linear region:

The addition works because all three quantities are logarithmic. Calling GdB “linear gain” is a category error; linear power gain is a ratio and would be used in a linear-power cascade equation. For a frequency-changing mixer, use the measured conversion gain or loss and state which input and output frequencies define the reference planes.

Never compare an OIP3 from one device with an IIP3 from another as though they were the same number. Convert them to a common plane with the applicable gain, loss and impedance conditions.

The Arithmetic Is Per Tone

At one valid two-tone operating point, let:

• PIN be the input power of either equal tone in dBm;
• PFUND be the output power of the corresponding fundamental tone in dBm;
• PIM3 be the output power of its associated third-order product in dBm; and
• Δ = PFUND − PIM3 in dB.

Under the verified 1:1 and 3:1 model:

Those powers are normally per tone. Two equal, uncorrelated RF tones have a total average power 3.01 dB above either tone. Confusing per-tone power with total two-tone power shifts the reported intercept. State the convention explicitly, especially when comparing a data sheet, receiver report and bench result.

A Bounded Worked Example

Suppose a declared test produces these measured values at one reference plane:

• equal input tones: −30 dBm per tone;
• output fundamentals: −20 dBm per tone;
• corresponding IM3 products: −80 dBm each; and
• small-signal gain: 10 dB.

The separation is 60 dB, so:

The total average input power of the two equal tones is approximately −27.0 dBm, not −30 dBm. The result does not say that 0 dBm per tone is safe or that the output can deliver +10 dBm per tone linearly. It says that the measured fundamental and IM3 lines would cross there if their local slopes continued.

One point is not enough for a defensible result. Repeat at several input levels. The fundamentals should remain near a 1 dB/dB slope, the selected products near 3 dB/dB, the products above the analyser or receiver floor, and the fundamentals below compression. Report both low-side and high-side products if they differ.

Tone Spacing Selects Which Parts of a Receiver Are Exercised

There is no universal 100 kHz spacing. A component data sheet may use 1 MHz. An HF receiver report may use 20 kHz, 5 kHz or another spacing. ITU-R SM.1837-1 is the in-force test procedure for IP3 of radio-monitoring receivers and specifies a reproducible receiver method rather than leaving conditions implicit.

Spacing determines which preselector, roofing filter, mixer, IF stage or digitizer sees the large tones. Two wide-spaced tones can exercise early stages while later filters reject them. Close-spaced tones may pass farther into the chain and encounter AGC, DSP or another gain state. A receiver can therefore have several valid IP3 results—not one architecture-wide constant.

Also state centre frequency, whether both tones lie inside the receiver’s accepted bandwidth, which IM3 product is measured, and whether the product falls at the tuned frequency or another declared output. Frequency conversion and images must be mapped, not guessed.

Gain State and Operating Conditions Travel with the Number

Preamp, attenuator, AGC, dither, preselection and RF/IF gain settings can change both the noise floor and the stage that dominates IM3. Supply voltage, bias, temperature, source and load impedance, frequency and output loading can change component linearity. For an SDR, sample rate, analogue bandwidth, reference level and overload state matter too.

A high input-referred intercept in a low-gain state does not automatically mean that state gives the best weak-signal result. Reducing front-end gain often improves input-referred linearity while worsening input-referred noise. The receiving system needs both adequate noise performance and enough strong-signal margin for its actual environment.

Compression Sets a Real Boundary That IP3 Does Not

The 1 dB compression point, P1dB, is an actual measured input or output level at which gain has fallen 1 dB below the small-signal extrapolation. IP1dB and OP1dB are input- and output-referred, just as IIP3 and OIP3 are. They are different measurements.

Many amplifiers show OIP3 above OP1dB, but no universal offset applies to every topology, frequency, bias, waveform or gain state. Use the measured P1dB and saturation data. Do not infer a safe maximum input by subtracting a remembered number from IIP3.

The absolute maximum input is a survival rating, not a linearity metric. It may be set by device breakdown, protection diodes, ESD structures, switching or thermal limits. The permitted continuous, pulsed and out-of-band inputs may differ. IP3 overrides none of them.

IP3 Is Not Blocking, IP2, RMDR or ADC Full Scale

Metric or test	What it asks	Why IP3 cannot replace it
P1dB	At what level does gain compress by the declared amount?	IP3 extrapolates IM3 slopes; it does not measure actual gain compression.
Blocking	How does one strong unwanted signal degrade a weak wanted response?	The mechanism may be compression, gain switching, AGC, noise rise or another limit.
IP2 or IM2	What second-order products are generated?	A wideband or direct-conversion receiver can be limited by sums, differences or DC/near-DC products even with excellent IP3.
RMDR	How much in-band noise appears through reciprocal mixing with a strong offset signal?	Local-oscillator or sampling-clock phase noise is not predicted by IP3.
ADC clipping and SFDR	How much composite waveform and spur range can the converter process?	Hard clipping, quantization, clock jitter and converter spurs need converter-specific tests.
Absolute maximum	What input can the hardware survive?	A hypothetical intercept is not an electrical or thermal safety rating.

Rohde & Schwarz’s phase-noise guidance shows how reciprocal mixing raises in-band noise beside a strong offset signal. IEEE 1241-2023 supplies current ADC terminology and test methods. Both are reminders to name the mechanism before choosing a score.

IP3 and Noise Can Build One Conditional SFDR

IP3 does not contain a noise floor. If an input-referred noise level NIN in a declared bandwidth is added, the ideal cubic model can estimate the equal-tone level at which an input-referred IM3 product reaches that noise. The corresponding third-order spurious-free range is:

The bracketed difference is in decibels. The result inherits every assumption of the IP3 extrapolation and the noise measurement. Change bandwidth, noise figure, gain state, tone spacing or endpoint and the number changes. It does not predict a single blocker, reciprocal mixing, discrete ADC spur or clipping.

Cascaded IP3 Requires Linear Powers and a Declared Chain

For two weakly nonlinear stages with linear power gain G1 and input-referred intercept powers in watts, a common first-order cascade is:

Additional stages add corresponding gain-weighted terms. Do not insert dBm or dB directly into this linear equation. Convert powers and gains, calculate, then convert the result back to dBm.

This cascade is a planning model. Interstage filtering can attenuate the parent tones before a later nonlinear stage. Mismatch changes actual gain and termination. Products from different stages can combine with phase and may partly reinforce or cancel. Compression, AGC and switched paths break the fixed cubic model. Measure the assembled chain when the decision matters.

A Two-Tone Test Must Be Cleaner Than the Device

The usual setup uses two signal generators, source filtering, isolation or pads, a linear combiner, the device under test and a spectrum analyser or receiver output measurement. Before inserting the device, verify the combined source spectrum at the intended level. Generator outputs can interact through an inadequately isolated combiner and create the very IM3 products attributed to the device.

Analog Devices’ AN97 shows this fixture problem directly: generator isolation, matching and source filtering can set the apparent intermodulation floor. The analyser can also create products when its input mixer is overdriven. Add known output attenuation and confirm that indicated fundamentals and products change as expected without losing the product below the instrument floor.

Record at least:

• input and output reference planes, impedance and calibration/de-embedding;
• f₁, f₂, spacing, centre frequency and measured IM3 frequencies;
• power per input tone and total two-tone power convention;
• gain, conversion gain or loss at the same operating point;
• bias, supply, temperature, gain/attenuator/AGC/filter state and load;
• analyser span, resolution bandwidth, detector, attenuation, preamp and averaging;
• source, combiner and analyser IM3 margins; and
• several input levels showing the measured fundamental and product slopes before compression.

For a complete receiver, the output may be audio or digital rather than RF. State the wanted/product endpoint, receiver bandwidth and any AGC or DSP state. Compare receiver IP3 only when the methods and conditions match.

Use IP3 for the Question It Actually Answers

IP3 matters when two or more strong signals can generate third-order products in the wanted region: crowded HF bands, broadcast-rich urban sites, repeater sites, multi-transmitter stations and wideband receivers are obvious examples. It matters less when the declared products remain far below external noise and every later limit—but that conclusion comes from levels and bandwidth, not from calling a site “quiet.”

My reading order is simple: inspect actual IM3 at realistic levels, confirm the slope region, keep the per-tone convention and reference plane attached, then check P1dB, blocking, RMDR, IP2, ADC overload and noise performance separately. IP3 is useful precisely because it is narrow in meaning.

IP3 tells you where two extrapolated lines would meet. The engineering result is the measured products, slopes and operating margin long before that crossing.

Primary and Authoritative Technical Sources

• Recommendation ITU-R SM.1837-1—in-force IP3 test procedure for radio-monitoring receivers.
• Analog Devices: IP3 and Intermodulation Guide—two-tone products, IIP3/OIP3 arithmetic and cascade treatment.
• Analog Devices AN97—clean two-tone generation, source isolation, filtering, analyser limits and accurate IM3 measurement.
• Analog Devices AN-2622—declared tone spacing/level, IP2/IP3 and compression boundaries for LNAs.
• NIST: Measurements of IP3 and P1dB for Software-Defined Radios—measured slope departures, regression and receiver gain-state boundaries.
• ARRL Laboratory Test Procedures Manual—receiver two-tone IMD dynamic-range and intercept methods under declared configurations.
• Rohde & Schwarz: Understanding Phase Noise Fundamentals—reciprocal mixing as a distinct receiver limit.
• IEEE 1241-2023—active standard for ADC terminology and test methods.

Questions, antenna-factor records or height trials to share? Contact RF.Guru.