Third Order Intercept

Look at the datasheet for almost any RF amplifier, mixer, high-speed ADC or receiver and you will find a figure called IP3 — the third-order intercept point. This number describes how well the device handles strong signals without generating spurious products of its own. A device with a poor IP3 produces false signals that do not exist at its input; a device with a high IP3 stays clean even when several strong signals are present. Understanding IP3 lets you read datasheets intelligently, predict when intermodulation problems will occur, and understand why adding more gain at the front of a system can make its large-signal performance worse.

IP3 is not a real operating point — no device ever actually works at its intercept point. It is an extrapolated figure of merit derived from the mathematical behavior of third-order nonlinearities. The concept can seem abstract at first, but once you understand the geometry behind the two-line intersection on the gain plot, the rest of the topic falls into place naturally.

The Intercept Concept — Why We Extrapolate

Every amplifier, mixer, and active device in a signal chain has a transfer function that relates output signal level to input signal level. In the ideal world, this transfer function is perfectly linear — double the input and you double the output. Real devices are never perfectly linear; they produce distortion products. The third-order intercept point is a way of characterizing how severe that nonlinearity is.

When you apply a single tone to an amplifier and gradually increase its level, the output increases at the same rate — 1 dB of output for every 1 dB of input — as long as you stay well below the compression region. On a plot where both axes are in dBm, this relationship is a straight line with a slope of exactly 1. This is the fundamental response line.

Now apply two tones simultaneously at slightly different frequencies — say, f1 and f2. The third-order nonlinearity in the device generates intermodulation products at frequencies 2f1 − f2 and 2f2 − f1, which fall very close to the original tones and cannot be filtered out. If you measure the power of these third-order IMD products and plot them against input power, they follow a line with a slope of 3 — for every 1 dB increase in input, the third-order products increase by 3 dB. This steeper slope means the third-order product line rises much faster than the fundamental response line.

If you extend both straight lines on the plot — the fundamental (slope 1) and the third-order IMD (slope 3) — they will eventually intersect at a single point. That intersection is the IP3 point. The input power coordinate of the intersection is the Input Intercept Point (IIP3); the output power coordinate is the Output Intercept Point (OIP3).

The IP3 graph: the fundamental output (slope 1) and third-order IMD product (slope 3) are extrapolated until they intersect. The IP3 point is always above the 1 dB compression point and is never actually reached in practice.

View Larger

Here is the critical practical point: no real device ever operates at its IP3 point. Before the input reaches IIP3, the device saturates — gain collapses, the output clips, and the simple mathematical model breaks down. The device typically reaches its 1 dB compression point approximately 10 dB below OIP3. IP3 is always an extrapolated, theoretical value derived from measurements made at much lower power levels, where both the fundamental and third-order product lines are still following their ideal slopes. Because of this, IP3 is a figure of merit rather than an actual operating limit.

The value of IP3 as a figure of merit is that it lets you predict IMD product levels at any input power. Because the third-order product line has a slope of 3 and the fundamental has a slope of 1, the difference between them changes by 2 dB for every 1 dB change in input power. This is why the formula for predicting IMD levels is so clean and powerful — you only need to know the input level and the IIP3 to calculate exactly where the IMD products will land.

Input-Referred vs Output-Referred IP3

IP3 can be expressed in two ways: referred to the input of the device, or referred to the output. Both describe the same nonlinearity from different reference points, and both are useful depending on what you are trying to calculate.

IIP3 (Input Intercept Point) is the hypothetical input power level at which the extrapolated fundamental output and the extrapolated third-order IMD output would be equal. It is the x-coordinate of the intersection on the IP3 graph. For a typical low-noise amplifier (LNA), IIP3 might be +5 dBm. This means that if you could input +5 dBm to the device (without it saturating), the fundamental output and the third-order products would be equal in power — which in practice never happens, but the extrapolated math predicts it.

OIP3 (Output Intercept Point) is the same intersection expressed as output power — the y-coordinate of the intersection. OIP3 and IIP3 are related by the device's gain:

For an LNA with 15 dB of gain and an IIP3 of +5 dBm, OIP3 = 5 + 15 = +20 dBm. This makes intuitive sense: if the input intercept is at +5 dBm and the device amplifies by 15 dB, then the output intercept is at +20 dBm.

A critically useful rule of thumb connects OIP3 to the 1 dB compression point:

This approximation holds reasonably well for many amplifier types. The reason the gap is approximately 10 dB is rooted in the mathematics of amplifier transfer functions — the third-order coefficient that creates IMD products also causes gain compression, and for a power series approximation to a typical amplifier, these two effects come together at roughly 10 dB below the extrapolated intercept. For the LNA example above: P_1dB ≈ +20 − 10 = +10 dBm at the output, or equivalently −5 dBm at the input.

When comparing devices, check whether the specification is IIP3 or OIP3, and at what point in the chain it is measured. Amplifier datasheets usually quote OIP3, because the output is what a designer connects to the next stage; receiver and instrument specifications usually quote IIP3 referred to the input connector, which tells you the large-signal performance as seen by the signals you are measuring.

Cascaded IP3 and the Friis-Like Formula

A receiver front end is not a single device — it is a chain of stages: an RF bandpass filter, a low-noise amplifier, a mixer, IF filter stages, and so on. The IP3 of the entire chain is determined by a formula that resembles the Friis noise figure formula (which you covered in the noise figure lesson), but it works differently in one important way: while noise figure is dominated by the first stage, cascaded IP3 is dominated by the later stages that receive amplified signals.

The formula for cascaded IIP3 (using linear power ratios, not decibels) is:

The key insight from this formula is that each successive stage is weighted by the accumulated gain of all stages before it. If the first stage has a gain of 15 dB (factor of 31.6), then the second stage's IIP3 is divided by 31.6 when calculating its contribution to the total. A second stage with a seemingly acceptable IIP3 of +10 dBm contributes an effective IIP3 of +10 − 15 = −5 dBm to the cascade. This is why the first amplifier stage in a receiver is so critical for large-signal performance.

This calculation reveals an important design principle: high gain early in the receive chain is the enemy of good large-signal performance. This seems to contradict noise figure optimization, where you want high gain in the first stage (to set the noise floor before later, noisier stages add their contribution). The result is a fundamental tension in receiver design between low noise and good IP3.

For practical design, the ideal front end has moderate gain in the first stage — enough to establish a good noise floor but not so much that it drives later stages into IMD problems. Gain of 10–15 dB in the LNA is a common engineering compromise. High-performance receivers and spectrum analyzers often use switchable preamplifiers and attenuators so the user can reduce front-end gain when strong signals are present.

IP3 in Datasheets and Real Systems

IP3 figures range from around 0 dBm for low-cost gain blocks and RF front ends to +40 dBm or more for high-linearity amplifiers and the best receivers. These numbers have direct practical meaning for how a device behaves when strong signals are present.

The formula for predicting the power of a third-order IMD product is:

As a reference, a system with a noise floor of −130 dBm in its measurement bandwidth will start showing IMD products whenever P_IMD3 rises above roughly −130 dBm.

This is why large-signal performance matters wherever strong signals are present. A receiver near a broadcast transmitter or a cell site, or a spectrum analyzer measuring a transmitter's output, can easily see signals of −30 dBm or more. A front end with IIP3 = +10 dBm will generate clearly visible false products; one with IIP3 = +30 dBm will remain clean.

You can also work the calculation backward to find the input level at which IMD products first reach the noise floor. Set P_IMD3 = noise floor and solve for P_in:

For a front end with IIP3 = +10 dBm and noise floor = −130 dBm: P_in threshold = (−130 + 20) / 3 = −110/3 = −36.7 dBm. Any two signals stronger than about −37 dBm will produce IMD products above the noise floor. For IIP3 = +30 dBm: threshold = (−130 + 60)/3 = −70/3 = −23.3 dBm — the signals must be more than 13 dB stronger before IMD products emerge.

ADC linearity specifications

High-speed ADCs used in digital receivers and digitizing instruments specify linearity in the same terms: two-tone IMD in dBFS or dBc, SFDR, and sometimes an equivalent IIP3. An ADC's nonlinearity behaves differently from an amplifier's near full scale, where clipping takes over abruptly, so ADC datasheets give two-tone IMD measured at specific levels (for example, each tone at −7 dBFS) rather than relying on the intercept point alone. Audio converters and amplifiers specify the same property as THD+N and twin-tone IMD.

The Two-Tone Test

The standard laboratory method for measuring IP3 is the two-tone test. Two signal generators produce continuous-wave tones at slightly different frequencies — typically 1 kHz to 1 MHz apart, depending on the device and its bandwidth. Both tones are set to equal amplitude. The two generator outputs are combined through a power combiner that provides isolation between the generators (to prevent each generator's output stage from contributing its own IMD to the test), and the combined signal is fed to the device under test (DUT). The output of the DUT goes to a spectrum analyzer.

Two-tone test setup for measuring IP3. The power combiner must provide good isolation between generators to prevent generator-contributed IMD from corrupting the measurement.

View Larger

On the spectrum analyzer display, you will see the two fundamental tones at f1 and f2, and the third-order products at 2f1 − f2 and 2f2 − f1 (which fall just outside the tones on each side). You read off the per-tone input power (P_in) and the third-order IMD product power (P_IMD3), then calculate IIP3:

The quantity (P_in − P_IMD3) is the "carrier-to-IMD ratio" in dBc. For example, if P_in = −20 dBm and P_IMD3 = −80 dBm, the carrier-to-IMD ratio is 60 dBc. IIP3 = −20 + 60/2 = −20 + 30 = +10 dBm.

The test must be run at input levels well below the 1 dB compression point of the device — typically at least 10 dB below P_1dB. If the device is near compression, the third-order products will not be following the ideal slope-3 behavior, and the calculated IP3 will be pessimistic (lower than the true extrapolated value). Good practice is to confirm that the measured IMD product level changes by 3 dB for every 1 dB change in input level before declaring the test valid.

Predict Third-Order IMD Level from IP3

Working in the other direction, if you know the IIP3 of a device and the level of two signals, you can predict exactly where their third-order IMD products will land relative to the noise floor. This is useful when planning a measurement or a system — for instance, if two strong signals at 14.225 MHz and 14.230 MHz reach an amplifier, you can predict whether their third-order products at 14.220 MHz and 14.235 MHz will be visible.

IP3 vs Noise Figure Trade-Off

The noise figure and IP3 of a front end are linked by a fundamental trade-off that every designer must navigate, and that every user should understand when setting up a receiver, an instrument or a measurement.

Achieving a low noise figure requires low-noise amplification with high gain as early in the signal chain as possible. The high gain of a good LNA sets the system noise floor at a low level before the noisier stages (mixer, IF amplifiers) can add their contribution. This is exactly what the Friis noise figure formula tells you: noise figure is dominated by the first stage.

But as you just learned from the cascaded IP3 formula, high gain in the first stage is also what makes the second-stage and later-stage nonlinearities so damaging to the overall IP3. The accumulated gain before each stage weights that stage's IP3 contribution by a large factor. An LNA with 20 dB of gain reduces the second stage's effective IIP3 contribution by 20 dB — which can easily make a moderate mixer the bottleneck of the entire receive chain.

The consequence is that you cannot simultaneously optimize a receiver for the absolute best noise figure and the absolute best IP3. A receiver optimized purely for noise figure will have relatively poor large-signal performance; a receiver optimized purely for IP3 will have a higher noise floor than necessary.

The ideal front-end design aims for a moderate gain of 10–15 dB in the LNA, combined with the best possible noise figure at that gain level (typically 0.5–1.5 dB for modern LNAs), and a high IP3 in the LNA itself (IIP3 of +15 to +25 dBm is achievable with careful circuit design). The combination provides a reasonable noise floor while keeping the gain-weighted impact on the mixer's IP3 contribution manageable.

Spectrum analyzers and many receivers include switchable step attenuators — typically in 5, 6 or 10 dB steps — ahead of the first mixer. Adding 10 dB of attenuation reduces both the wanted signal and the interfering signals by 10 dB, but because IMD products change by 3 dB per 1 dB of input change, the IMD products drop by 30 dB while the wanted signal drops by only 10 dB. The signal-to-IMD ratio improves by 20 dB. When strong signals are present, 20 dB of extra spurious-free range can more than compensate for the 10 dB loss in sensitivity.