In the world of wireless communications, three tests serve as litmus tests for RF ATE: noise figure (NF), phase noise, and adjacent channel power ratio (ACPR). NF and ACPR demand extreme ends of the power range for both stimulus and measurement. Phase noise adds the requirements of frequency stability and synchronism for both stimulus and measurement subsystems.

This combination of requirements prevents the ATE designer from optimizing to a particular RF market segment. Rather, the intention is to deliver one system to the wireless customer that will measure all the wireless components.

Noise Figure

In short, NF is the ratio of the input-to-output signal-to-noise-ratios. NF measurements are performed on components that make up a receiver’s front end, whether they are simple low-noise amplifiers or more highly integrated RF down-converters.

Because of the mobile nature of most wireless communications appliances, the received power is very small. This factor forces the ATE to have a low thermal noise floor and contribute very little added noise power in its measurement receiver. A

-110-dBm noise floor and a 17-dB NF are representative values.

Active devices add noise to signals passing through them. Passive devices reduce signal-to-noise ratios. Both effects are represented by NF:

With the device input terminated (no RF stimulus), the output noise power is the input noise power multiplied by the gain plus any noise added by the device:

where:

The noise factor, a linear form of NF, is given by:

NF is always greater than 0 dB.

What is noise power? Noise from a resistor of any value is given by:

^{-23}J°K, Boltzmann’s Constant

Substituting the standard temperature T_{0 }of 290° Kelvin, the noise power for any value resistor is:

or expressed in dBm and scaled for measurement bandwidth, B:

Power Ratios

Device NFs are considerably lower than ATE measurement-system NFs. To overcome this measurement, the device is deliberately stimulated with two noise powers. Equation 1 is a straight line with the slope equal to the device gain and y-intercept equal to the added noise. Adding noise allows you to measure at least one point on the line. The second point is provided by either the intercept of the measurement noise floor with the line or knowledge of device gain.

Statistically Relevant Measurements

By its nature, a noise signal varies as a function of time. This poses some difficulty to maximizing test throughput. From Nyquist’s criteria, if you sample at a rate at least twice the highest frequency content, you can properly convert time-domain samples to frequency-domain spectra.

For example, for a signal bandwidth of approximately 10 kHz before digitization, sampling at 25 kHz is adequate. As you would expect, the number of samples influences measurement time and repeatability, and noise measurements are no exception.

Figure 1

illustrates the noise-power calculation over a 12.5-kHz bandwidth as a function of increasing sample size. The figure shows that a compromise in sample size is possible at the expense of accuracy. For reference, taking 16,384 samples (2

14

) at 25 kHz consumes 655 ms with a small amount of additional time to process the data. The NF measurement is an excellent indicator of sensitivity and likely to continue as a valuable production test. However, measurements of low-noise devices, repeatable to within 0.1 dB, require time and add cost to device testing.

**Copyright 1999 Nelson Publishing Inc.**

August 1999