Tip:
Highlight text to annotate it
X
Modern spectrum analyzers digitize data early on in the measurement process. As a result
of digitization, the spectrum shown on the display is not truly continuous, but divided
into discreet bands called buckets. Each bucket may span several data points, but only one
data point is chosen to represent each bucket. The continuous trace on the screen is formed
by joining together these discreet representative points. The rules which determine how representative
points are selected for each bucket are set by the detector type. When you start up an
Agilent spectrum analyzer, you are shown the noise floor of a full span measurement. The
detector type is normal. But we will revisit this detector type later. First, we will examine
the simplest three detectors – peak, negative peak and sample. This detector present conveniently
generates three traces which each use one of the detector types we are interested in.
The peak detector always displays the highest amplitude that was measured inside of a bucket.
It is a good choice for strong continuous signals, but also increases the apparent noise
level. The negative peak detector always displays the lowest amplitude that was measured in
a bucket. The negative peak detector is used for a few specific measurements like EMC testing,
but is rarely used for most scenarios. The sample detector displays the data point closest
to the center of each bucket. It is a good choice when measuring noise-like signals,
but can miss steady narrow signals like sinusoids. Now we will source a 6 GHz sinusoid and see
how the different detectors respond. The peak detector sees the sinusoid even in this wide
span. However, the sample and the negative peak detectors are not registering a signal.
While these detectors are useful, they can miss narrow signals, especially in wide-span
measurements. The default detector is the normal detector, which is actually a combination
of the peak and negative peak detectors. When a signal is present in a bucket, the normal
detector acts as a peak detector. When only noise is present, the buckets alternate between
negative peak and peak detection. To see the result of this detection scheme, we will change
trace 2 to a normal detector and compare it to peak and negative peak. Overlaid like this,
we can see how the normal detector combines peak and negative peak data. It has also successfully
detected the 6 GHz sinusoid we are sourcing. The normal detector is a good general-use
detector because it does not miss signals and also shows the amplitude range of noise.
In this scenario, we are trying to measure a weak signal with a slow FM modulation. We
are tracking the signal with a continuous peak search, but the noise is strongly influencing
our measurement. Often, we will implement trace averaging to squelch the noise variance
around weak signals. Let us see what happens when we try that here.
Trace averaging uses multiple sweeps to average, so we cannot use it on a moving signal like
this one. We need to reduce our variance in real time instead of over multiple sweeps.
We can accomplish this with the average detector. The average detector averages all the amplitude
values in a bucket and displays this point as the representative value. This yields a
real-time variance reduction at the expense of increased sweep time. We have improved
our measurement quality by reducing the variance around our signal. We can further improve
the variance by adjusting how the bucket values are averaged. Power averaging performs an
RMS average, which is very useful for finding the RMS power of complex signals. Voltage
averaging performs a linear average of voltage data, which is often called for in EMI testing.
Log power averaging performs an average on the decibel values of power, which leads to
a 2.5 dB drop of the noise level. We will switch to log power averaging to take advantage
of that 2.5 dB drop in noise level. By just modifying our detector settings, we have greatly
improved the signal-to-noise ratio of this measurement.