Right you are Jerry....the remaining signal after the technique I described
is applied is indeed pretty hacked up. Part of my point in talking about it
was to try and bring that very fact out to people....that the sound can have
a very odd quality (rather, lack of quality) to it after this is done.
Adaptive and predictive filters work also in some situations. But I seemed
to remember Doug telling me that they were using the technique I described
at the Hamfest way back when they introduced the Orion down at Ten Tec
central. BUT.....my memory of that conversation is NOT totally clear, and
so I do not claim to know for a fact that this is so.
Incidentally, I am familiar with the dual AGC proposal, and I have the
timewave DSP-59+ and occasionally use the AGC that it provides. Having
another AGC loop inside the radio would indeed be of great use in these
kinds of situations, and in several others.
And yes, correlated (more or less) noise can also creep in, and really mess
up the DSP process if correlation is the technique being used. Knowledge of
the signal type, as you state, can help there, even in that case. I don't
have any idea how advanced their algorithm is, and to what extent, if any,
they use a priori knowledge of the signal to improve the filter over a
simple correlation model.
Regards,
Gary, AA2IZ
----- Original Message -----
From: "Dr. Gerald N. Johnson" <geraldj@storm.weather.net>
To: <tentec@contesting.com>
Sent: Wednesday, May 03, 2006 3:45 PM
Subject: Re: [TenTec] 2.033
> On Wed, 2006-05-03 at 14:54 -0400, Gary Hoffman wrote:
> > FWIW.....not ALL DSP just narrows bandwidth.
> >
> > The way it worked when I used to work with audio DSP systems in a
previous
> > life (and the IF of the Orion is at an audio frequency) is that each
sample
> > of the A to D converter is compared to those around it, and a
correlation
> > level is established. (Excuse me that I've forgotten the correct
technical
> > language they used). Speech and tones (like CW) are highly correlated,
in
> > that each sample tends to be relatively closely related to those that
came
> > before. Random noise, of course, is not highly correlated.
> >
> > The noise is reduced by simply deleting those samples not correlated.
> >
> > Then the remainder go through D to A and come out as sound again.
> >
> > The NR reduction level control (or whatever its called) simply shifts
how
> > much correlation is required before a sample is either kept or rejected.
A
> > higher setting requires more correlation before the sample is kept. Of
> > course if you turn it up too high, then too many samples are rejected,
and
> > you get weird sounding sound after you go back from Digital to Audio.
> >
> > Has nothing whatsoever to do with a bandwidth limiting filter. That, of
> > course, is another approach, which works to a certain degree also. But
not
> > what a real DSP noise reducer does.
> >
> > Now....how much of each does the Orion do ? That I cannot tell you, not
> > being privy to the design details. Perhaps the folks at Ten Tec would
like
> > to tell us ?
> >
> > Note - years ago, when I did this, we could do it on the fly with a DSP
> > processor. I'm sure the processors they have now far outperform the
ones we
> > had then. So, its doable, and practical, essentially with a single
chip.
> >
> > 73 de Gary, AA2IZ
>
> The first book I read on noise reduction perhaps ten years ago
> concentrated on the correlation technique and I believe that is often
> used because it works but it leaves a background that is not very random
> to the ear and so many users grumble about the lack of natural sound.
> There is also the element of time delay that makes tuning the radio
> difficult with NR working.
>
> The latest book on the topic in the ISU library is "Advanced Digital
> Signal Processing and Noise Reduction" by Saeed V. Vaseghi, 3rd edition,
> copyright 2006 from John Wiley and sons. In preparation for a club
> program tomorrow, I just scanned it again. He offers at least a dozen
> different noise reduction techniques, including adaptive filters,
> predictive filters, and correlation. One technique uses a gaggle of
> Weiner filters each working on a relatively narrow spectrum looking for
> coherent voice components within the noisy spectrum. The Weiner filter's
> gain falls rapidly with poor S/N in each filter to reduce noise in that
> part of the spectrum. Other techniques include blanking and filling in
> the gaps of lost data by a myriad of interpolation or statistical
> techniques. Most all signal to noise improvers work immensely better if
> the characteristics of the signal and noise are known, but sometimes
> that's difficult in radio communications because voices and words spoken
> are so different.
>
> To add to the complications, the noise to be reduced is sometimes not
> random white noise, but is arcing from power lines, or simply mistuned
> voice signals, or hash from data or switched power circuits, or
> scratches on a 78 RPM record. On that record the hiss can be from
> granularity in the lacquer or rattling of the stylus in the groove. All
> these noise sources respond differently to noise reduction techniques
> and then think of the complications of having all these types of noise
> present at the same time. Even worse is the situation when the impulse
> character of some of the noise sources causes the roofing filters to
> ring (characteristic of the mechanical filters in the Collins S-Line)
> filling in the gaps between power line voltage peaks.
>
> At the extreme of noise reduction, if the data rate is sufficiently
> slow, one can employ a waterfall of fourier spectra and look to the
> correlation between the collection of spectra. This works great for
> extracting slow speed signals of the moon or from milliwatts ERP at VLF
> where dots may last hours. That can really reduce the bandwidth and
> still accept multiple signals. The bandwidth is effectively the
> reciprocal of the sample period of the fourier data. E. g. sample for
> 0.1 second, the bandwidth is 10 Hz, sample for 1 second, the bandwidth
> is 1 Hz... Simply narrowing the bandwidth alone in an arbitrary fashion
> doesn't always improve the S/N and if the bandwidth is narrowed before
> processes expecting fairly random noise can prevent those processes from
> working at all. Hence the measurements showing S/N decreasing with the
> narrower roofing filter. I've known of narrow CW filters (270 Hz of the
> TS-430 days from Kenwood) that were fatiguing to copy on FD because
> noise in came out as a tone and that was work picking out keyed CW
> alongside that tone. The noise was correlated by ringing of the filter
> to an annoying degree.
>
> Doing all the processing at essentially audio after converting the RF
> down to 15 KHz leaves one at the mercy of 1/f flicker noise and to the
> linearity of those multiple mixers and the spectrum shaping of the front
> end and roofing filters and the post roofing gain stages often add back
> broad band noise before the detecting process. In my FT-857D, the DSP CW
> filter is good at rejecting interference but not at improving the S/N of
> the weak CW signal, but the optional 455 KHz IF filter with the same
> bandwidth does improve the S/N of the weak CW signal to my ears.
>
> Most audio frequency range DSP need an AVC of the radio before them to
> prevent clipping from too much signal in the sampling process. But that
> can reduce the desired signal when the unwanted signal in in the
> processing bandwidth but not the output bandwidth. The classical case of
> the AVC pumping from the strong signal and hacking up the output of the
> audio filter. I've proposed publicly that there be two AGC loops, one
> for that A/D protection and another after the processing for the user
> convenience and protection. My Timewave DSP-59 offers that output AGC as
> an option. Radios should too.
>
> --
> 73, Jerry, K0CQ,
> All content copyright Dr. Gerald N. Johnson, electrical engineer
>
> _______________________________________________
> TenTec mailing list
> TenTec@contesting.com
> http://lists.contesting.com/mailman/listinfo/tentec
>
>
_______________________________________________
TenTec mailing list
TenTec@contesting.com
http://lists.contesting.com/mailman/listinfo/tentec
|