# Re: Cleaning up the spectra

• Subject: Re: Cleaning up the spectra
• From: Leif Asbrink <sm5bsz.com; leif@xxxxxxxxxxxxxxxx>
• Date: Thu, 13 Mar 2008 13:56:16 +0100

```Hello Michel,

> Despite spectacular results with the autospur removal feature, there
> are still a lot of "untamed" spurs, birdies visible; see e.g. the
> screenshot taken during the last 2m EME Dubus CW activity event.
> (bottom of the homepage)
>
> It is annoying to click on a carrier to find out it is a birdy...so I
> wonder what else could be tried to improve the situation besides an
> optimisation of the automatic spur removal feature parameters?
> (this is with an inactive photovoltaic cells inverter)
There are two spurs close to the green line at 048.57. The high resolution
graph has averaged 30 times in a bandwidth of about 5Hz and the lines
at 048.65 and 048.52 do not show any broadening. These lines are
however very weak, (S+N)/N is 3 dB or less in about 3 Hz.

I do not think the spur removal routine can work on spurs that
are as near the noise floor as these ones. Not with the current
requirements for spur S/N that are built into the code.

The spur removal routine has a parameter for the spur time constant.
It then collects the average of spectra for 3 times this time.
Assuming 5 Hz per pixel in your spectrum, the spectra arrive
at a rate of 10 per second if the window is sine squared which
means that the searched spectrum would be identical to what
you have in your high resolution graph for a spur time constant
of 1 second.

In spursub.c you may find these lines:
// anything that is 15dB/sqrt(3*spur_speknum) above the noise floor has to be a signal.
// Locate the max point.
With a 1 second time constant it means that anything 2.8 dB above
the noise floor has to be a signal.

Now, locating the noise floor is not trivial, there may be all sorts
of other signals than spurs present...

What Linrad does is to split the spectrum in 32 parts and then
locate the minimum point in each one of them. The noise floor
RMS value is then assumed to be  7dB/sqrt(3*spur_speknum) above
the minimum point. The criterion for a signal is therefore
3.9 dB abov the smallest point.

In your screen dump, a spur is removed at 048.23. This spur is not
very strong and probably not very stable so the spur removal has
removed some of the noise on this frequency. As a consequence the
none of the other spurs is sufficiently high above the "noise floor"

In case you made a recording - or can make a recording - I would
happily add it to my library of test files. Maybe there are things that
could be improved. (I am sure there are....)

****************************************

One important thing:
http://www.sm5bsz.com/linuxdsp/install/dlevel.htm
Your screen dump looks like figure 11 here. This indicates an
error that causes loss of S/N on your system. The above link
should explain it so you can eliminate this problem.
(Hint. Have a look at the waterfall with the Delta44 unplugged.)

73

Leif / SM5BSZ

--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to linrad@xxxxxxxxxxxxxxxx
To unsubscribe from this group, send email to linrad-unsubscribe@xxxxxxxxxxxxxxxx
-~----------~----~----~----~------~----~------~--~---

```