[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [linrad] AM and FM

Hi Bob and all,

> I am not sure since I have yet to look at the code, but
> it appears that Leif is saying he uses a narrow filter
> at the carrier to extract the carrier for basebanding
> before the final detection is done.  A thing one could
> do instead of a PLL is to take a high-Q resonator (IIR
> filter) which has three terms, at the carrier.
Or any other narrow filter;)

> This works well on good signals quite well if the follow on
> processing takes advantage of the independence of the
> sidebands.  However, all of us have heard a selective
> fade take the carrier and one of the sidebands out for
> seconds.
I bet the carrier does not drop so low it will not be
high above the noise in 1 Hz bandwidth;)

> A 2nd order PLL based carrier recovery algorithm
> with  a fairly narrow loop filter will "flywheel" for
> many seconds during a severe fade and it will have to be
> a deep fade indeed for the 2nd order PLL to lose lock on
> the "pure tone" that is the carrier.
Sure, and a PLL is not even needed. One could tune to within
0.1 Hz and use one's own carrier just as one would do with
two SSB receivers one for USB and the other for LSB. 

> Another thing
> PLL's can easily do is gear shift the gain or loop
> bandwidth depending on the lock condition.  This means
> that in a lock condition, one can REALLY narrow the
> bandwidth (down 1 Hz if you like) and this will take
> flywheel through a really nasty fade.
Sure, but a narrow filter with an AFC will be equivalent.
Linrad has a second AFC in the baseband besides the one
that has it's own window so I do not think a PLL solution
would produce very different results. The Linrad routines are
the way they are because they are adapted for a much more
difficult problem. Decoding morse code on the QSB peaks of
a Morse coded EME signal where path modulation and frequency
stability is not too much narrower than the modulation bandwidth.
I find it hard to believe a PLL could be helpful then.

> In the best coherent detector, one then makes an SNR
> estimation of each sideband and then votes in a weighted
> way, which sideband to believe.
Not if you want to hear what the weaker voice is
saying when two AM stations are on the same frequency;)
You assume that the objective is to produce a single channel
with the best possible estimate of what the original modulation 
was. Maybe that is what most people want, but I am sure there
are also extremists who want to hear the difficult signals........

> If both are high SNR,
> pick your favorite, if both are a bit ratty, then take
> an average.  If one fades, take the other, etc.
Well, finding out what SNR is is not trivial. What seems to
be better signal could be a computer generated interference
that is similar to the true modulation on one side of the carrier.
Very hard for the software to know. If interference
is white noise and QRN only, then this should be a good strategy,
but even better would be to split the spectrum in narrow segments 
on both sides and use this strategy for each frequency band
symmetrically around the carrier.

Since I am not BC DX listener I do not know what problems 
the user really wants to solve. Maybe, some day, when wideband 
receivers are more common, I will get recordings of interesting 
signals where more sophisticated algorithms could be useful.
It would not be difficult to set up a model of the multipath
propagation so one can use the fact that the selective fading 
travels across the frequency spectrum in a fairly regular fashion.
One could compute how the phase and amplitude distortion should 
vary with time and keep the model updated. It is obvious to me that
in cases when the selective fading dips are very deep and separated 
by a kHz or so, the phase in the sidebands reverse every kHz and
one could then remove that distortion.

Many things to experiment with for the future:)