60 GHz Noise Figure Measurements: Y-Factor and Cold Source Best Practices
by Joel Nelson, RF and Microwave Applications Engineer, Keysight Technologies
E-band spectrum has seen much development in recent years, particularly the 60 GHz unlicensed band, driven by 5G backhaul and 802.11ad. Noise figure measurements at this frequency present challenges to even the most experienced test engineers. As a measurement applications engineer, I’m often asked, “what method should I use to make 60 GHz noise figure measurements?” The answer: “it depends.” It sparks the familiar tradeoff between ease of use and accuracy. Minimizing system errors and uncertainties often requires more up-front preparation and work. This tradeoff is especially applicable when comparing the two noise figure measurement methods: Y-factor and cold source. Both methods have advantages, tradeoffs, and limitations. This article covers best practices for making 60 GHz noise figure measurements using the Y-factor method with a spectrum analyzer and the cold-source method with a vector network analyzer.
Y-Factor Method Using a Spectrum Analyzer: Introduction
The Y-factor method is the go-to method for most test engineers because of the simple calibration and measurement procedure. Much of the Y-factor procedure has been automated in recent years, including automatic download of ENR data. The calibration is performed with the noise source connected to the test system. After calibration, the measurement is simply a matter of inserting the device. Spectrum analyzer noise figure measurement applications offer “one-button noise figure and gain measurements.” This simplicity can be attractive. However, it can give the user a false sense of comfort, often leading to disregard of measurement uncertainty. At current 4G cellular frequencies below 6 GHz, the measurement uncertainties can be relatively small and allow a device to meet specification. At 60 GHz, the uncertainties and errors are magnified and can no longer be ignored or disregarded.
The largest contributors to Y-factor measurement uncertainty are mismatch and noise parameter errors. Imperfections in the noise source, DUT, and test system result in signal reflections causing ripple over frequency. Generally, the higher the frequency, the higher the mismatch. Since the spectrum analyzer is unable to measure and correct for mismatch, the cascaded error can be quite large.
Noise parameters are a function of an amplifier’s source impedance. Noise generated at the device input is reflected off the test system and reenters the DUT, changing the noise figure. The device noise figure is minimized when it sees an ideal impedance, called gamma-opt. As the source impedance moves away from gamma-opt, the device noise figure increases.
Noise figure measurements are assumed to be performed in a perfect 50 ohm environment; i.e. the test system is assumed to provide perfect 50 ohm terminations to the device input and output. In practice this is rarely the case. Neither the noise source nor the measurement receiver are perfect 50 ohm terminations and the noise source impedance often changes between hot and cold states. The larger the impedance change, the more error is introduced into the measurement. As with mismatch, the spectrum analyzer is unable to measure and correct for the noise parameter effects.

Cold Source Method Using a Vector Network Analyzer: Introduction
The cold source method using a vector network analyzer addresses mismatch and noise parameter effects at the cost of increased test setup complexity. The cold source method requires two measurements: the device’s gain or S21, and its output noise power with the input terminated. The input termination or cold source can be a fixed impedance referred to as a scalar measurement, or a series of impedances referred to as a vector measurement. Modern VNAs can use impedance tuners to present multiple source impedances to the device to characterize noise parameters and accurately calculate 50 ohm noise figure. This article will focus on scalar noise figure measurements without an impedance tuner. The measurement will rely on the native source impedance of the VNA to provide the cold source termination.
Since a VNA can directly measure all 4 S-parameters, mismatch corrections can be applied to reduce measurement uncertainty. The S21 trace measured with a VNA can serve as a reference to the gain results from the Y-factor method. This comparison can also provide a visual representation of mismatch effects within the Y-factor measurement setup. The ripple pattern across frequency is typical of impedance mismatch between the device and test system. When designing and troubleshooting the test system, the goal is to minimize uncertainty while maintaining sensitivity and dynamic range.
Test Setup Comparison
Y-Factor Test Setup Considerations
The Y-factor test setup can be as straightforward as using a millimeterWave spectrum analyzer that covers the 60 GHz frequency if the analyzer has enough sensitivity. Spectrum analyzers are broadband instruments which may not have enough sensitivity at 60 GHz, so an external LNA with a preselection filter is often required. Commercial noise sources at 60 GHz typically cover a waveguide frequency range like 50-75 GHz for WR15. The preselection filter prevents undesired responses in the spectrum analyzer down-conversion process with a broadband input signal.
The test system is also subject to noise parameter and mismatch effects. The noise source and device can present different impedances to the test system, resulting in changes to the system noise figure. The system noise figure measured during the calibration step may not be the same when the device is connected. Isolators and attenuators before and after the device can be used to improve the system impedance match and noise parameter effects. However, adding attenuators will affect system sensitivity, so a balance must be found.
Cold Source Test Setup Considerations
The cold source test setup requires careful consideration to maximize dynamic range, since there is a large difference in the DUT’s output power between the S21 and output noise power measurements. For the DUT output noise power measurement, the VNA receiver noise figure needs to be 10 dB better than the DUT excess noise power, defined as DUT gain plus noise figure, minus cable loss. To improve sensitivity, modern VNAs have front panel jumpers that can be reconfigured to “reverse” the input test port coupler. This reverses the main arm and coupled arm of the coupler, increasing the signal to the receiver by the coupling factor while lowering the output power by the same amount. In this configuration, the VNA noise figure can be between 40-50 dB at 60 GHz. Therefore, the device excess noise power needs to be 50-60 dB. For most devices an external preamp and preselection filter are necessary. Using the front panel jumpers, the preamp and filter are placed between the coupler thru path output and the VNA receiver input as shown in Figure 1. This preserves the ability to measure the DUT output match.
The S21 measurement requires VNA source power levels high enough to accurately measure the small reflected signals at the device input and output, while not overloading the external preamp or VNA receivers at the DUT output. For the measurement, the VNA source output should be set to its receiver’s linear input level, minus the sum of the DUT gain and preamp gain. VNA input compression specifications are typically defined at the test port input. With the test port 2 coupler reversed, the input compression level is reduced by the coupling factor. For the calibration routine, the VNA source power can be increased by as much as the DUT gain if it does not require a change in the VNA source attenuator. An attenuator change between calibration and measurement will invalidate the calibration data. Lowering the VNA output power lowers the signal-to-noise ratio of the S11 measurement, resulting in higher measurement noise. The noise on S11 will be imparted on the corrected S21 trace. Reversing the coupler at VNA port 1 helps lower the output power while increasing the signal-to-noise ratio of S11, reducing overall measurement noise.
With both port couplers reversed, the VNA can properly measure and correct for mismatch errors. As previously mentioned, vector noise figure measurements utilize an impedance tuner to correct for noise parameter effects at the device input. Although scalar noise figure measurements do not correct for these effects at the device input, some VNAs utilize the calibration standards to characterize the noise parameters of the VNA receiver. With electronic calibration modules, multiple known impedances are presented to the VNA receiver during the calibration routine and its change in output noise power can be characterized. During the DUT measurement, the DUT output match is measured and the output noise power will be corrected for the change in noise of the VNA receiver as it is pulled by the DUT output impedance.
Test Results
For the Y-factor measurement, a Noisecom NC5115 noise source was used with a Keysight N9041B 110 GHz spectrum analyzer as the receiver. To improve the match and minimize errors, a 10 dB attenuator and a 5 dB attenuator were placed before and after the DUT, respectively. An external preamp with 35 dB gain was used to improve the spectrum analyzer sensitivity. The cold source measurement was performed using a Keysight N5247B PNA-X using the same preamp as the Y-factor test setup.

The gain measurement results are presented in Figure 2. As previously mentioned, the characteristic ripple over frequency in the gain measurement can serve as an illustration of mismatch effects. The Y-factor measurement was repeated with additional attenuation which reduced the gain ripple pattern but affected the system sensitivity, resulting in an unacceptably large error in the average noise figure. The final attenuation used provided a good compromise between system sensitivity and mismatch effects.

The noise figure results are shown in Figure 3. The cold source method resulted in a noisier trace since the VNA receiver does not have the same sensitivity as that of the spectrum analyzer. Adding a higher gain preamp could improve the sensitivity. However, this would require lowering the VNA output power, increasing noise during S11 measurements, and imparting noise on the gain measurement. Again, a balance must be found.

Both test methods require compromise to achieve an acceptable balance. The Y-factor method requires compromising system sensitivity for improvement in the mismatch effects. The cold source method requires compromising accuracy in mismatch effect corrections for improvement in system sensitivity. Since there is no perfect compromise, a good practice would be to measure more sweep points and apply smoothing to the data. The noise figure data with 8% smoothing is shown in Figure 4.
Conclusion
Returning to the original question: “What method should be used to make 60 GHz noise figure measurements?” The answer remains “it depends.” The cold source method will give the most accurate measurements, but the noise figure trace may contain noise. The Y-factor method may result in a noise figure trace with less noise, but the gain measurement will have mismatch uncertainty. The increased accuracy of the cold source method requires significantly more consideration and preparation. The end-user is then left to decide the preferred measurement technique based on the tradeoff between ease of use and accuracy.
About the Author

Joel Nelson is a RF and Microwave Applications Engineer at Keysight Technologies. He has a BSEE from the University of Utah and an MSEE from Stanford University. His Keysight career has been spent supporting customer engagements in commercial wireless and aerospace/defense accounts covering signal analyzers, signal generators, vector network analyzers, oscilloscopes, and much more.
(1502)