Soundcard spectrum analyzer through regenerative frequency division

Hi,

I'm trying to build a spectrum analyzer for the shortwave spectrum that
uses a simple computer soundcard for data acquisition.

Specifically what I'd like to do is compress the mean signal levels from DC
to 30MHz to DC to 30 kHz, which my sound card should be able to sample. I
have no need to be able to rebuild the signals into an intelligible
format. What I hope to do is merely to see at what frequencies
transmissions are taking place. With 30Mhz compressed to 30kHz it should
be a simple matter of reading the frequency and mentally multiplying it
with 1000.

Extensive googling suggested regenerative dividers might do the trick, but
if it is too broad a range I'd also be very happy with steps of 3MHz
sparated over a 10-way switch. However, I don't quite know how to go about
designing something like this. I'm much in need of guidance here, which
I'm hoping you will provide me with. I don't even have a proper, specific
question I could ask... Help?

 

I think you're asking for quite a lot: you want to divide the
frequencies but maintain the amplitudes. For the most part, all the
methods I know of that divide frequency are grossly nonlinear and do
not preserve the amplitudes; they can't even preserve distinct
frequencies. I know of one method, but implementing it would be way
more difficult than just building a simple digitzer followed by an
FPGA with FFT code in it. Another method that's been used is to build
a simple version of a swept spectrum analyzer: just feed a ramp to a
VCO, VCO output to a mixer, to convert your band of interest (spread
over time) to one frequency, where you detect the amplitude. Plot
that amplitude versus time, synchronized with the ramp you feed the
VCO. There are modules available to take care of much of the hard
parts for you. If the sweep is fast enough, you can even use your
sound card to digitize the detected amplitude, though it would be nice
to not have the sound card AC coupled as it almost certainly is. You
could do a software "DC restore" on a retrace blanking level, though.

Cheers,
Tom

 

Alas, on the wretched hour of Thu, 22 Feb 2007 09:25:50 -0800 Tom Bruhns
thusly proclaimed:

Yes, I suspect sweeping the bands is how regular spectrum analyzers work,
but the problem with that is that I can't use a normal audio spectrum
analyzer to plot the result. I'd have to write rather complex software to
make it work, (unless this is the module you speak of) and I suspect,
though I'm clearly no expert, that it would be easier to do it in hardware.

Maintaining the amplitudes is just a matter of retaining them in relation
to each other so that when I see a peak on the analyzer I'll know that it
is most likely a signal rather than noise. I don't need an absolute value,
nor an exact frequency, as long as I don't have to tune the entire
shortwave spectrum to find it.

Could you please name a few of the methods you mentioned you knew so I
can study them? This is something of a learning project for me, far from
mission-critical, so I'd be quite happy with something that
merely resembles what I'm trying to achieve. Then, when I have a clue of
what it is I'm actually doing, I could try building something that would
not be a general embarrasment. )

I have a few ideas of what I'd like to be able to eventually achieve with
this thing in the long run. It essentially amounts to a Software Defined
Radio wherein you can observe the entire shortwave band on your
computer monitor, and then click on some part to 'zoom in' on a frequency,
choose a demodulator and listen to the transmission. Though I'll likely
never reach that far, I'd still like to take the first step.

Thank you very much for your reply.

 

What a fascinating question!

The following link may provide some ideas:

http://adsabs.harvard.edu/abs/1981STIN...8228519C
Frequency compression of wideband signals using a distributed sampling
technique

Reading the abstract, it appears you would sample, say every kilohertz,
from some base frequency (probably not DC, but maybe 100 kHz) to 30 MHz
and store the amplitude of each sample. (VCO stepped by 1kHz) Each
sample would allow you to record the peak or average amplitude found
within some bandwidth (say 1 kHz). These amplitudes could be easily
digitized and stored on a computer.

From there, it is simple to display each sample amplitude as a one
hertz ampitude on your sound card (1000:1 compression). By playing with
the sampling bandwidths, etc., you can improve resolution. Actually,
there is no need to use the sound card since the data can be plotted
directly on the computer screen.

This would allow you to incorporate variable zoom, or perform other
software analyses on the stored data.

You might encounter problems in detecting short-duration signals that
occur between samples.

In any case, you will probably want to employ logarithmic compression of
amplitudes so that weak, as well as strong, signals can be displayed
simultaneously.



Good luck!

Chuck

 

Alas, on the wretched hour of Thu, 22 Feb 2007 15:02:56 -0500 chuck
thusly proclaimed:

Well yes... Say I have an A/D converter running at a sampling rate of 60M
samples per second I think I should theoretically be able to completely
represent the shortwave band and do all sort of mojo with it... in
software. That paper was written in '81, but nowadays my 'pooter can
calculate FFTs fast enough to probably process the entire band in
real-time, and if I get my GFX card to do the presentation I can
over-sample and anti-alias it to my screen to show me the entire thing at
60 fps. - All of which I'd get at zero effort if I managed to do the
frequency compression in analog. With for example audio compressors I
could still do a lot of the mojo, and the device would not be limited to
the platform it's written for. I could even use analog devices with it...

But I'm beginning to suspect you're essentially right anyway. From my
googling it seems frequency compression has only been done in software, so
it might be I'm chasing a red herring here... Then again, this being a
learning project it'd just mean I'd be working with more modern tools.

Thank you very much for your reply. I'll begin looking into it immediately.

 

The method of "frequency compression" I am thinking of is not
practical at the frequencies you're interested in, I'm afraid, and it
really is a big task to implement it. I might try to think of an
analog of the technique that would work at HF radio frequencies, but I
don't hold much hope for that.

There's a grass-roots effort to make a SDR system using the LTC2208, I
believe it is. You should search for that. You can leverage the work
of a lot of others that way, and contribute to that effort. They may
well have both hardware and software already that will do what you
want and more.

Cheers,
Tom

 

What would you use at the "front end" to sync the scan-rate
of the receiver to the sweep rate of the software? Either
the radio has to tell the software "I am on this frequency"
or the software has to tell the radio "go to this frequency".
Many of todays radios provide a nice digital interface, so it
shouldn't be hard - but it has to be part of the design.

The rate that the radio can scan and produce a reliable
signal will be a limiting factor. That's not news, but
has to be considered in the design.

Anyway, I'm thinking a straight audio recording from T0 to Tn
time with the receiver frequency under PC control. You could
display the frequency range and rate and audio level on the
screen real time, and record the info on the hard drive for a
later look. It makes a poor man's spectrum analyzer of sorts.
And it lets you "play" with the front end to figure out its
capabilities for use in a "real" spectrum analyzer design.

Ed

 

Alas, on the wretched hour of Thu, 22 Feb 2007 13:07:58 -0800 Tom Bruhns
thusly proclaimed:

It must be the Mercury project of HPSDR... Looks rather interesting,
though I wonder if I truly need all those FPGAs. I must study it! =)

Thank you ver much for your replies.

 

There's no easy way to "compress" the frequency range.

Also using a very fast A/D isnt too practical-- the radio signals vary
over a huge amplitude range. You'd need like a 16-bit 60MHz A/D
converter, and you can't find those at Radio Shack.

 

the only way I can think of to do this would be to take snapshots at 100MHz
sample rate and play them back at 100kHz rate to your soundcard

you'll get noise at around the sample length so it won't be clean all the
way to DC

 

AD8307: dc-500MHz, 92 dB dynamic range; $12.63 at Jameco in single
quantities.

 

Response time: 400 nanoseconds

 

Not unreasonable to build a block of say 30 (or more) bandpass filters.
Centre frequencies staggered as felt fit.
Add 30 ways of signal scanning, analogue switch I.Cs connecting each
channel in turn to Chuck's AD8307. Take the audio output from that.
Could even add a single 1V audio tone as channel 0 to provide a zero
frequency, spectral reference line.
john

 

Hook that array of filters up to a PIC or AVR to do the grunt work and stuff
the data back to the PC over serial -- its got the A/D converter built in,
so you just need the filters, analog switches, and an integrator/peak
detector (i.e. you listen to each band for a short period of time, looking
for a peak or summing the levels during that time, then you sample). The A/D
is a lot slower than any stand alone A/D chip, but you are just sampling the
levels, not the entire spectrum. At max resolution the Atmega168 can do 15k
samples per second.. for 30 bands, thats 5000 scans per second max -- more
than adequate.

If you really wanted to go all out -- put a peak detector on each filter,
and have the controller clear the detector after it samples -- that will get
you continuous monitoring of each band during the scan -- you wont miss
anything while checking other bands.

Run the uplink to the PC at 2 bytes per sample + a few bytes overhead for
framing = 64 bytes.. on 115kbps serial (11 kbytes/sec) will get you some
where around 170 scans per second -- again more than adequate (or drop
the sample resolution to 8 bits and double your scan rate)

 

Alas, on the wretched hour of Sat, 24 Feb 2007 00:10:07 +0000 John
Barrett thusly proclaimed:

Oh my, I thought the conversation had died. Glad to be proven wrong!

I've studied the HPSDR project, but it seems their hardware will be vey
expensive. As such, the comparatively dirt-cheap AD8307 looks very
interesting. I do however get the impression that it doesn't sample the
entire bandwith simultaneously, so it couldn't work as an SDR on its
lonesome. Might have to still use the LTC2208 for that, but no reason to
be in a hurry.

Rather than downsample the output of the chip I'd rather connect it to one
of the high-speed ports on my computer and then have it sort through it
all. That way I'd be dealing with a minimum of hardware and consequently a
minimum of complexity. I'll have to start studying the issue in more
detail.

Thank you all very much for your replies! Glad to see sucha response.

Sincerely,

 

You don't want a regenerative frequency divider; you won't get enough
frequency division nor (because it is really an oscillator) any useful
amplitude information. It is a bit like the timebase of an oscilloscope
synchronising on a higher frequency signal. Not what you want. Logic
circuit dividers and phase locked loops with frequency multipliers in
the loop all lack important features - and all need a sweep frequency to
tune them because they only divide one frequency at a time.

I agree with others that compressing the 30MHz to 30kHz for your
soundcard and software to analyse with a FFT or whatever is a difficult
way of getting the desired result. You would need to sample at huge
speed and good accuracy, feed it into some dual-ported (or interleaved
access) memory external to the PC (because it has to be at greater speed
than you could reasonably feed into a PC via standard interfaces), then
squirt it out to the PC at a much slower speed (or something else at
least as difficult). In other words the hardware you will need, just to
make the signal nice for the PC, will be much more than if you had done
a whole spectrum analyzer without the PC.

One relatively small modification to your design goal could make life a
lot easier: feed the soundcard with a time-domain representation of the
spectrum, not a frequency-domain one. So you would have a tiny amount
of hardware to do part of what a conventional spectrum analyzer might do
(ramp wave to control a VCO, VCO output to product detector/balanced
modulator, with low pass filter and fast-responding AC detector, feeding
the soundcard input). Simple oscilloscope-for-soundcard software will
then be able to display the spectrum easily, or you could sample the
waveform a certain number of milliseconds into the ramp to get the
amplitude at a given frequency. The ramp to control the frequency could
even come from a soundcard's output if you like, so you can sweep it
slowly some times to get better resolution (otherwise, to cover 30MHz at
maybe 10Hz you're doing an average of 300kHz per millisecond which is
way too much). If you were sampling and converting 30MHz down to 30kHz
you'd still have the problem that you can miss signals if you scan too
quickly/not have enough samples.

What do you want to do with the results? Do you want to be able to
distinguish stations 10kHz apart? Do you want to see the whole 30MHz
spectrum at once, or zoom in to parts of it? Do you mind waiting a few
seconds before you see the spectrum? How low in frequency do you really
need to go? Of course if you only need to see the spectrum 30kHz at a
time (e.g. 31.03MHz to 31.06MHz) there are easy-ish methods that still
use FFT software on the PC, but shift the frequency axis along.


Mark A

 

The datasheet I see is a logarithmic amplifier, not a A/D.