From: MarkusVester@aol.com
Message-ID: <e6.2c69267d.2a883dc4@aol.com>
Date: Sun, 11 Aug 2002 18:23:00 EDT
Subject: LF: Noise blanking for narrowband modes?
To: rsgb_lf_group@blacksheep.org
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk
Reply-To: rsgb_lf_group@blacksheep.org
Sender: <majordom@post.thorcom.com>

Hi Alberto, Jim and LF group,

oh I just love those rainy sundays ;-). - It's been interesting to read about 
the improvements of Jason decoding with hard-limited audio. In times of heavy 
QRN, I have often observed a significant improvement of weak QRSS signals 
when connecting a pair of 1N4148 diodes parallel to the PC's line input. My 
RX has full SSB bandwidth and no AGC. The best performance occured when the 
frontend attenuator was set such that normal background noise was just below 
the threshold of the limiter. Driving the system into hard limiting did not 
seem to help any more, but tended to show blocking and IM, caused by strong 
in-band carriers and DBF39.

Narrowband reception goes along with long FFT integration times, and in a 
linear system, the effective noise level is given by the total energy of 
background and burst noise. If for example the QRN covers 10% of the 
aquisition time with a level of 30 dB above the background, it would look 
like 100-fold increased Gaussian noise, a desastrous 20 dB SNR loss. If we 
however clip the QRN at say 6 dB above background, the average noise energy 
will be increased by only 40% (1.5 dB). And the signal will virtually 
disappear during 10% of the time (-0.8dB), adding up to only 2.3 dB SNR 
degradation.

But I am wondering, if not a noise-blanking scheme would be more effective 
than simple limiting. In a software implementation with some buffering, one 
could even avoid switching clicks by smoothly reducing the gain around each 
burst (eg. a 10 ms cosine shape), keeping the spectral widening of strong 
carriers moderate. The user could preset a fraction of blanked time (e.g 10%, 
dependent on the severeness of the QRN), and the SW could then automatically 
control the threshold to approach that.

Going one step further, we could try to give an optimal estimate of a signal 
in the presence of a time-variing noise level. In analogy to a Wiener filter, 
the noise contribution of each sample (ie. 10 ms block) to the FFT input 
would then have to be inversly proportional to its SNR. If the noise level 
went up temporarily by 3 dB, we should reduce the gain by 6 dB. For high 
noise peaks, this will of course result in almost complete blanking. One nice 
feature of the scheme is that we don't have to worry about a threshold, all 
we have to do is keep track of the wideband power from the RX (assuming that 
this is noise and our weak signal is below that), and dynamically control the 
gain.

73 es gn
Markus, DF6NM