Topband: AGC (and now other work)

Oleg Skydan skidan at mail.ints.net
Wed Feb 11 23:20:24 EST 2004


Thank you for a very informative post.

I tried the SD-ROM algorithm (it was described in a paper I downloaded asome
time ago), but I was not
impressed with its performance. Possibly it was due to the QRN type of
noise, where each pulse usually
consists of many short pulses. I did not know an algorithm which can
effectively remove such type of noise.

Want to try to make a DSP NR which will work like a conventional NB, but
with the careful blanking pulse
shaping.

As for the adaptive noise reduction, I tryed LMS/NLMS algorithms and found
them almost useless.
Now I have abandoned the LMS algorithm and use the autocorrelation one. As
far as I know the idea was
introduced by the SP9VRC in his code for the EVM56002. It is not in the
perfect condition now (I will try to
 improove it further), but I often use it. I will make some sample records
and put them on my web site on
week-end.

Also a variable level of noise reduction is a very useful feature. Sometimes
the signal is very distorted at
the maximal NR, so reducing the NR level makes amazing things (aspecially
with the speech signals).

As for the notch/autonotch. I am using a very good manual notch filter (very
sharp), based on the two tap LMS
filter described in the Doug KF6DX QEX 1998 articles. The autonotch is a
modified NLMS algorithm. It performs
well on the clean tones, unfortunately most of them are modulated with hum,
so it does not rejects them
completely. I can change settings and make it more agressive, but then it
starts to remove parts of speech :(.

Have tryed a bank of two tap LMS filters tunable by the FFT. Works well, but
it needs a clever algorithm to recognize
steady cariers and useful signals.

So some (maybe a lot of ?) work is needed here too.

Both notch filters are before AGC of course.

> Fortunately: from the Echo cancelling world, there comes a new collection
of algorithms called Affine Projection >Algorithms which solve our problems.
They converge almost as fast as RLS, they are only a couple of times more
>expensive than LMS and really do a beautiful job.  When I am done, I
believe this will be the first application of this >algorithm to automatic
noise reduction in receivers.  From my receiver recordings (IF recording and
AF recordings) and >simulation and design in MATLAB, this algorithm performs
spectacularly well.

It is something new for me.

Regards,
Oleg
73 de UR3IQO



More information about the Topband mailing list