Overcoming Noise Figure Measurement Challenges in Fixtured, On-Wafer and ATE Environments
By David Ballo, Application Engineer, Component Test Division, Agilent Technologies

Conceptually, measuring noise figures of RF and microwave low-noise amplifiers (LNAs) is easy. The most common technique involves connecting a noise source with a known amount of excess noise to the input of an amplifier under test (AUT). This technique is commonly referred to as the Y-factor method, and it requires the measurement of two noise powers at the output of the AUT. The first measurement is done with the noise source turned off, which provides the amplifier’s input with a room-temperature 50-ohm termination. The second measurement is done with the noise source turned on, a condition where a known amount of excess noise power is injected into the amplifier, causing the output noise power to increase compared to the first measurement. From the two noise-power readings, noise figure can be calculated, giving a quantitative indication of how much noise the AUT will add to a system. In practice, making accurate noise figure measurements of LNAs at high frequencies can be challenging for two reasons: connecting to the AUT can be difficult, and when the noise source is not connected directly to the input of the amplifier, the resulting accuracy degradation is both significant and often not well understood.

There are several examples where it is impractical or impossible to connect the noise source directly to the amplifier’s input. First of all, many devices used in aerospace/defense applications and commercial microwave communications are not connectorized. For example, many transmit/receive modules used in phased-array radar systems have microstrip input and output lines, requiring test fixtures to interface to commercial coaxial-based test equipment. Another example is that of microwave monolithic integrated circuits (MMICs), which are often tested while still on the wafer on which they were fabricated, before being sealed into hermetic packages. In this case, a coaxial-to-coplanar test probe must be used to connect test equipment to the AUT. In both of these examples, the noise source cannot be connected directly to the amplifier’s input.

Even when the devices being tested have coaxial connectors, many times they are measured with automated test equipment (ATE), allowing the connection of multiple test instruments for full characterization of the AUT. For example, a network analyzer might be used to measure S-parameters and gain compression, while a spectrum analyzer is used in conjunction with signal generators and a noise source to measure intermodulation distortion and noise figure. In this scenario, a switch matrix is used between the test equipment and the AUT. Again, when measuring noise figure, the noise source cannot be connected directly to the input of the amplifier.

In these cases, when the noise source cannot be connected directly to the AUT’s input, the addition of cables, switches, test fixtures, and/or probes adds loss and causes the effective source match of the test system to degrade. While the impact of loss can be mitigated by applying a scalar correction to the excess-noise-ratio (ENR) values of the noise source, the effects of source-match degradation are not easily removed, causing a corresponding decrease in measurement accuracy. Noise figure measurements are assumed to be done with perfect 50-ohm test systems, but in real-world scenarios, this is never the case. While a noise source by itself provides a reasonable 50-ohm termination, it is not perfect. Adding extra components to connect to the AUT also adds extra reflections, which degrades the effective source match seen by the amplifier.

The impairment in source match causes two types of noise figure measurement errors. One of them is the classic error due to mismatch, arising from non-ideal system source match and amplifier input match. The input match of most high frequency LNAs is nominally 50 ohms, but the actual input match versus frequency varies around this value. This means some of the noise power from the noise source is reflected off the amplifier’s input. If the noise source supplied a perfect 50-ohm match, this reflected power would be fully absorbed, and the true 50-ohm noise figure of the LNA would be measured. However, if the noise source does not provide a perfect source match, then some of the noise power is re-reflected towards the AUT, causing more or less noise power (depending on the relative phases of the matches) to reach the input of the amplifier. The effect of this mismatch shows up as the classic ripple pattern in the test results seen if the frequency span is wide enough to show one or more cycles. Often the ripple is not seen because the frequency span of the measurement is too narrow, but the error is still in the measurement.

The other type of error introduced by imperfect system source match is not well known by many test engineers. It is due to the fact that some of the noise generated by the AUT comes out of the amplifier’s input port, where it reflects off the system source match, and reenters the amplifier. The reflected noise causes the noise figure of the AUT to change, depending on the phase of the reflected noise power and the correlation among the various noise generators within the amplifier. Thus, the measured noise figure varies as a function of the system source impedance. This effect is well understood by LNA designers, who measure the noise parameters of the individual devices used to construct the amplifier. The noise parameters tell the designer what the minimum noise figure will be for a given device, and at what impedance (gamma optimum) this minimum will occur. The noise parameters also tell the designer how the noise figure of the amplifier will change as the source impedance moves away from the optimum value. For a given impedance change, the magnitude of the resulting change in noise figure varies between amplifiers. Some amplifiers are very sensitive to changes in source impedance, and others less so. Armed with the knowledge of the device’s noise- and S-parameters, the LNA designer can go about designing matching circuitry to optimize gain and noise figure for a particular application.

The concept of noise parameters has a direct implication on our ability to accurately measure the 50-ohm noise figure. As the source impedance of the test system varies around 50 ohms, the measured noise figure of the AUT will also vary. The impact of this effect again shows up as a ripple in the measured results, indistinguishable from the ripple caused by mismatch. The more the source match changes, the larger the error introduced into the noise figure measurement. Adding components between the noise source and the AUT exacerbates this effect, even if loss compensation is used.

One way to overcome these challenges is to take advantage of the advanced architecture and calibration techniques of the modern vector network analyzer (VNA). VNAs have been used for years to accurately measure S-parameters of both connectorized and non-connectorized devices. Methods to extend vector-error correction to fixtured, on-wafer, and ATE environments have been developed so that it is not difficult to obtain accurate S-parameters in these situations, as long as proper calibration standards and techniques are used. Modern VNAs like Agilent’s PNA-X series can expand the suite of possible measurements to include intermodulation distortion by taking advantage of a built-in second RF signal source and an internal signal combiner. Recently, the ability to measure noise figure was added, with accuracy surpassing the commonly used Y-factor method.

For noise figure measurements, the PNA-X eliminates the need for a noise source by using the cold-source method. This method requires only one measurement of room-temperature noise power, along with an independent measurement of the gain of the AUT. Using vector-error correction, the VNA compensates for the mismatch effects caused by imperfect system source and load match, as well as for system frequency-response and crosstalk errors. The result is highly accurate gain and match measurements of the AUT.

To measure noise figure, an additional set of receivers was added to the PNA-X. These noise receivers are optimized for high gain and low noise figure. To overcome noise figure measurement errors caused by the non-ideal 50-ohm source match interacting with the noise generated by the AUT (the noise-parameter effect), a modified version of the cold-source method is employed that uses a standard Agilent ECal module as a variable impedance tuner, instead of as a set of electronic impedance standards. This ECal module is used to vary the source match of the test system between four and seven complex values, for each of the desired measurement frequency points. At each source impedance, a corresponding noise power measurement is made with the AUT in place. None of the source impedances presented to the AUT are exactly 50 ohms, but from the set of impedance and noise power measurements, the 50-ohm noise figure can be accurately calculated. The impedance tuner allows us to mathematically create a noise measurement system with excellent effective source match, just as mechanical or electronic calibration standards are used to provide excellent effective source match for S-parameter measurements. With the source-match-corrected cold-source technique, the effects of switches, cables, fixtures or probes are fully removed from the test results, giving the highest noise figure measurement accuracy of any commercial test system available today. Using this technique with the PNA-X, S-parameters, gain and phase compression, harmonics, intermodulation distortion, and noise figure can be accurately measured, all with a single connection to the AUT. Figure 1 compares noise figure measurement uncertainty between the Y-factor method and the cold-source method as implemented on the PNA-X. For the Y-factor method, the uncertainty is calculated in two different ways: one with the noise source connected directly to the DUT, and one with an electrical network simulating the switches and cables from an automated-test-equipment (ATE) setup placed between the noise source and the DUT (with loss correction). The PNA-X example includes the ATE network. Figure 2 shows an actual comparison between a PNA-X and an NFA measuring an amplifier in a simulated ATE environment. In this case, a 12-inch cable was used in place of a switch matrix. The ripple caused by imperfect system source match is clearly seen using the Y-factor method, but is not present in the PNA-X’s version of the cold-source method.

In conclusion, modern network analyzers like the PNA-X, with sophisticated hardware architectures and advanced calibration techniques, can overcome the difficulties and inaccuracies of using the Y-factor method for measuring noise figure in fixtured, on-wafer, or ATE environments. Using the cold-source method with a combination of vector-error correction when measuring S-parameters and source-match-correction when measuring noise power, accurate noise figure measurements can be made in any type of test environment.

About the Author
David Ballo works for Agilent Technologies’ Component Test Division in Santa Rosa, California, where he has acquired 28 years of RF and microwave measurement experience. After getting a BSEE from the University of Washington in Seattle in 1980, he spent ten years in R&D doing analog and RF circuit design on spectrum and vector-signal analyzers. Since then, he has worked in marketing, developing and presenting seminars and papers, and writing application notes and technical articles on a wide variety of network- and spectrum-analyzer measurement topics.

Agilent Technologies
www.agilent.com
TXTLINX.COM122
Email this article to a friend!