Roy Lewallen wrote:
Joel Kolstad wrote:
Roy,

"Roy Lewallen" wrote in message
...
This is covered in detail in _Modern Spectrum Analyzer Theory and
Applications_ by Morris Engelson (Artech House, 1984).

Does Engelson's idea of "modern" include digitizing spectrum analyzers, which
have already taken over most of the mid- to high-end of test equipment and --
as with digital scopes -- will most likely soon be in all but the lowest of
the low-end of instruments?

No, the book was written before those came about. I fudged a little by
qualifying my posting with "conventional" spectrum analyzer to exclude
newer technologies, but it sounds like "conventional" is rapidly
advancing to become the newer types.

I got out of the SA development world just as digital techniques were
beginning to develop -- my last patent (# 5,629,703), in fact, dealt
with a way of reducing distortion in an A/D converter intended for use
in a SA-like instrument. At that time, we were anticipating doing a
conventional sweeping down-conversion, then digitizing it at a 25 MHz
IF. Predictably, the digitization point has been moving toward the front
of the instrument since then.

I haven't followed the technology since, but I'm sure the new ones use
sampling techniques combined with an FFT, which is another way of
imperfectly representing what the real spectrum is like. (Actually, you
can never perfectly represent the spectrum of a real waveform, because
any spectrum which is finite in frequency span has to have existed for
an infinite time. Any modulated waveform is way outside this category.)
Sampling produces its own, different, sorts of artifacts which are
different from the ones produced by sweeping, although there are some
similarities. You have to be at least as careful, and maybe more so, in
interpreting a sampled waveform as one which is from a swept filter.

I'd be surprised if someone hasn't written an equivalent book to cover
the new type instruments, although a lot of the material in Morris'
books is still valid and teaches a lot about the nature of spectra.
Morris was the driving force behind Tektronix's entry into the spectrum
analyzer market, and the group's chief engineer and architect for a long
time. He'd retired by the time I joined the group, and I only met him
when he taught a couple of one-week courses based on his books. He's
truly one of the experts in the field.

Roy Lewallen, W7EL

Indeed, we've been making spectral analysis equipment using FFT
techniques since the early 80's or even a little longer. I believe the
HP3577 from that era uses digital IF techniques, though I never got
familiar with its "guts." Things have come a very long way since then.
Just as Roy says, there's a whole different set of things to
understand to get accurate, meaningful measurements out of them. Even
knowing what I do, I still get surprised by the instrument's response
to particular signals sometimes.

With respect to signals in finite bandwidths having to have existed
forever, in theory that's true, but in practice, noise from sources
beyond the signal of interest will always overwhelm those portions of
the signal that are very far from its main power, for a huge proportion
of practical signals. We'll keep pushing for higher spurious-free
dynamic range and lower noise floors in the instruments, but it's
probably fair to say they'll never be perfect, and signals are
essentially always polluted by external noise sources as well, if only
thermal noise in external resistances.

I believe you can find some information on the Agilent website about
factors to consider when making FFT-based spectral measurements. I
probably have a pdf or two hiding somewhere in my archives covering
some of this, too.

Modern analyzers are backed up by some very impressive software that
can help you analyze all sorts of things about modulation, signal
variation with time, etc. It goes FAR beyond simple spectral displays
of FFT results.

One of the key features of FFT-based spectral analysis is that you
don't miss as much as a swept analyzer with narrow IF does. For
example, if you have a signal that keys on for a millisecond out of
every second or so at 100MHz, it might indeed be radiating a broad
spectrum but there's observable energy only within a few kilohertz of
100MHz. An analyzer sweeping 50-150MHz with a 1kHz resolution
bandwidth will take quite a while to do its sweep, several seconds, and
may very well not be "looking at" 100Mhz, or close enough to it, at the
time the pulsed signal comes up to be able to see it. But an FFT-based
analyzer that's sampling fast enough to capture the whole 50-150MHz
band at once can easily find the 100MHz signal every time it pulses on.
The FFT is, quite literally, a large bank of parallel filters/energy
detectors at evenly spaced frequencies. In combination with a
"windowing" function, the effective shape of those filters can be
adjusted.

Cheers,
Tom