Spectrum measurements are essential to RF and microwave testing. The newer signal analyzers available today deliver performance far better than those of just a few years ago. The older classic swept spectrum analyzer is a super heterodyne downconverter, perhaps with a high band front end. This downconverter feeds a filter and then an amplitude or envelope detector. The detector produces a single-ended output proportional to the magnitude of the signal, and discards phase information.

Figure 1. Modern signal analyzer with digitized IF.

In contrast, with modern spectrum analyzers, the digitizing process is moved much closer to the analyzer input, digitizing the IF signal and preserving the phase or vector information. In addition to retaining vector data, we see significant benefits from processing operations once performed in analog format, now done with more precision and flexibility using high-speed DSP (Figure 1).

Analyzer performance continues to improve over time in areas such as dynamic range, accuracy and noise specifications. In many cases, these improved performance parameters can be traded for greater measurement speed at higher performance levels than earlier analyzers.

Digital Resolution Bandwidth (RBW) Filtering

The digital IF capability of modern signal analyzers enables the application of digital filtering to the signal — providing a better shape factor and selectivity to the filter. Since the digital filters behave more ideally and consistently than analog filters, we can sweep them faster and compensate for sweeping effects.

There are two ways to look at the shape factor improvement of digital filters. Note that the maximum sweep rate of an RBW filter varies with the square of the RBW, so a filter that is 2x wider can be swept 4x faster.

  1. A filter with equivalent RBW will be several times more selective at the -60 dB level and can still sweep 3-4x faster.
  2. Where narrow RBW is needed to separate signals of different amplitudes, one can use an RBW that is several times wider. This wider filter can sweep much faster: 9x faster for an RBW filter that is 3x wider.

Reducing Analyzer Displayed Average Noise Level (DANL)

Figure 2. NFE technique improves DANL analyzer noise power calculated and subtracted real time.

Measurement processing provides new opportunities for improved dynamic range as it takes advantage of the dramatic improvements in ADC and DSP technology. However, there are additional ways to reduce the noise floor of the measurement setup. Fundamental performance improvements can be expensive so it makes no sense to give up a dB of noise figure that you could otherwise gain using some simple techniques.

At microwave frequencies losses mount quickly with cable length, quality, switching, etc. In some situations the most effective and least expensive way to improve DANL is to use better or shorter cables and move the analyzer closer to the signal to be measured.

Here are some techniques to reduce the noise floor and improve the dynamic range of spectrum measurements:

  • Reduce analyzer attenuation
  • Add preamp
  • Reduce RBW
  • Add external filtering
  • Better, shorter cables, connectors and adapters
  • Move analyzer closer to device under test (DUT)
  • Noise floor extension (NFE)

Extending the Analyzer Noise Floor

For high performance measurements, the displayed measurement reflects both the power of the signal (and its own noise floor) and the noise contributed by the analyzer’s own noise floor (except where the signal/noise ratio of the signal to be measured is more than 20 dB or so). Today’s analyzer amplitude accuracy is so good that this noise addition has become increasingly significant in measurement situations, such that both the measurement of the signal and the SNR are in error.

Figure 2 shows an example of NFE technology. It is useful for scalar measurements like spectrum or power, but it will not improve vector measurements. In the example, the scale is 3 dB/div. Note that there is virtually no error in the measurement of the leftmost tone, where the SNR in the gold trace without NFE enabled is about 15 dB. The second tone from the right, however, is measured approximately 3 dB in error. The actual signal is approximately equal to the analyzer’s non-NFE noise floor, and the resulting noise addition produces a result that is about 3 dB high. The blue trace at the bottom shows that the NFE improvement in noise floor for this 2 GHz signal (1 MHz span) is about 12 dB.

NFE Implementation

Figure 3. Measurements with and without NFE engaged and with and without a signal present. Note the variance of the measurement of analyzer noise with the NFE noise subtraction.

When subtracting the noise contributed by the analyzer we first need to accurately characterize it over the operating range of the measurements. This involves modeling the noise floor and combining the model with calibration measurements of individual analyzers to accurately estimate their noise floor. The noise level of a spectrum analyzer varies as a function of many different parameters. A list of these parameters includes: the resolution bandwidth (RBW), the input attenuation, the tuned frequency (including the amplitude correction and mixing band associated with that frequency), the display detector, and the averaging scale.

Modeling and predicting the analyzer noise in all settings and configurations (including preamp on/off etc.) involves more than a single number or a look-up table. Data points are combined with predicted frequency responses and curve fitting in some cases. The effect of analyzer temperature is also taken into account. The tolerance for these parameters includes the effects of aging or time since calibration.

It is important to keep in mind that when power levels are very small, even a small change or error in power measurement or modeling, especially a negative one, can produce a very large negative dB excursion. This brings about added variance to the measurement of the extended noise floor and is an inevitable result of noise subtraction calculations. The significantly higher variance of the analyzer noise floor with the NFE is shown in Figure 3. This variance could be reduced (or averaged or smoothed) through the use of several kinds of averaging. The best would be through the use of the average detector and a longer sweep time.

It is important to note that we are making a scalar correction in subtracting analyzer noise power. We are not subtracting instantaneous noise voltage and are, therefore, not making vector corrections. The benefit of NFE will be a dramatic improvement in the accuracy of small signals, measured near the noise floor.


No single measurement technique or advancement will solve all application challenges. Modern RF & microwave signal analysis platforms have been shown to improve test time and accuracy by providing lower noise floor levels, architectural advances and unique measurement features.

This article was written by John Hansen, Senior Application Engineer, Agilent Technologies, Santa Clara, CA. For more information, Click Here