Taking advantage of knowledge your signal analyzer doesn’t have
To respect the value of your time and the limits of your patience, I try to keep these posts relatively short and tightly focused. Inevitably, some topics demand more space, and this follow up to December’s post Exchanging Information You’ve Got for Time You Need is one of those. Back then, I promised additional suggestions for adding information to the measurement process to optimize the balance of performance and speed for your needs.
Previously, I used the example of a distortion measurement, setting attenuation to minimize analyzer distortion. This post touches on the other common scenario of improving analyzer noise floor, including the choice of input attenuation. A lower noise floor is important for finding and accurately measuring small signals, and for measuring the noise of a DUT.
One of the first items to add is the tolerable amount of analyzer noise contribution and the amplitude error it can cause. If you’re measuring the noise of your DUT, the table below from my post on low SNR measurements summarizes the effects.
Examples of amplitude measurement error values—always positive—resulting from measurements made near the noise floor. Analyzer noise in the selected resolution bandwidth adds to the input signal.
Only you can decide how much error from analyzer noise is acceptable in the measurement; however, a 10 dB noise ratio with 0.41 dB error is not a bad place to start. It’s worth noting that a noise ratio of about 20 dB is required if the error is to be generally negligible.
Sadly, the input attenuation setting for best analyzer noise floor is not the same as that for best distortion performance. The amount of analyzer distortion you can tolerate is another useful factor. Reducing attenuation will improve analyzer noise floor and SNR, but at some point the cost in analyzer distortion performance may outweigh the benefit. And remember that video averaging provides a “free” noise floor benefit of 2.5 dB for measurements of CW signals.
Reducing input attenuation by 12 dB improves noise floor by a similar amount, as shown in the yellow and blue traces. Using a narrow video bandwidth (purple trace) for averaging reduces the measured noise floor but does not affect the measurement of the CW signal.
You can consult your analyzer’s specifications to find its warranted noise floor and adjust for resolution bandwidth, attenuation, etc. That approach may be essential if you’re using the measurements to guarantee performance of your own, but your specific needs are another crucial data point. If you simply want the best performance for a given configuration, you can experiment with attenuation settings versus distortion performance to find the best balance.
Many analyzer specs also include “typical” values for some parameters, and these can be extremely helpful additions. Of course, only you can decide whether the typicals apply, and whether it’s proper for you to rely on them.
If you use Keysight calibration services, they may be another source of information. Measurement results are available online for individual instruments and can include the measurement tolerances involved.
Signal analyzers themselves can be a source of information for improving measurements, and the Noise Floor Extension feature in some Keysight signal analyzers is a useful example. Each analyzer contains a model of its own noise power for all instrument states, and can automatically subtract this power to substantially improve its effective spectrum noise floor.
For microwave measurements, many signal analyzers use preselector filters to remove undesirable mixing products created in the analyzer’s downconversion process. However, these filters have some insertion loss, which increases the analyzer’s effective noise floor. A valuable nugget that you alone have is whether the mixing products or other signals will be a problem in your setup. If not, you can bypass the preselector and further improve the noise floor.
Finally, one often-overlooked tidbit is whether the signal in question is a consistently repeating burst or pulse. For these signals, time averaging can be a powerful tool. This averaging is typically used with vector signal analysis, averaging multiple bursts in the time domain before performing signal analysis. It can improve noise floor dramatically and quickly, and the result can be used for all kinds of signal analysis and demodulation.
Sorry for going on so long. There are other potential sources of information you can add, but these are some of the most useful I’ve found. If you know of others, please add a comment to enlighten the rest of us.