[hfsig] Performance of 24/192 soundcards? USB Soundcards

[Robert McGwier]

The AEA DSP2232 demodulator worked as well as it did on UO-14
and other NO DC transmitters because it had a fractionally spaced
linear filter followed by a one tap per symbols DFE.  However,
be advised that this is the higher dimensional equivalent of
"dividing by 0".

There are better approaches


"Viterbi equalizer"

or

EM algorithm.

Both of these algorithm try to jointly find the channel filter and
the data to produce a good replica of the observations.

The former attempts to jointly find the channel and find the most likely
emitted data path given the observations.

The latter attempts to jointly find the channel and the emitted data at each
symbol time that minimizes the probability of error given all of the
observations.
There is a neat algorithmic trick that uses Markovity of the data going
forward
and backward that makes this only slightly more expensive than the viterbi
algorithm but is a better system for communications systems if it is
computable.
I wrote my doctoral thesis on the nonlinear really difficult versions of
this
algorithm.  It is called nonlinear filtering and smoothing.  One can also
prove
that the EM algorithm is the Kalman filter and smoother and then channel
estimator
to improve the probability of the observations.  It is an iterative
algorithm
that can run over the data many times (unlike the viterbi or an equalizer).
In
practice,  you usually do NOT need to do too many iterations, especially
when
known data is regularly sent as in HF modems for channel training purposes.

Notice that these latter two algorithms do NOT attempt to invert the
observations
using an equalizer to get something  that looks like filtered data.  It
finds data and a channel that most looks like the noise observation.  The
difference
is in how you measure "most looks like".

It is always better when doing DSP or sound card things to just not cripple
yourself
with the restrictions of analog, only realizable circuits.

Interesting stuff.

Bob
N4HY


-----Original Message-----
From: hfsig-bounces at lists.tapr.org
[mailto:hfsig-bounces at lists.tapr.org]On Behalf Of Alexander Kurpiers
Sent: Wednesday, September 22, 2004 1:59 PM
To: TAPR HF Modes SIG Mailing List
Subject: Re: [hfsig] Performance of 24/192 soundcards? USB Soundcards


Mark Jordan wrote:
> 	Hi Anthony,
>
> 	I have sent you a 200kb ZIP file with the G3RUH circuit and partlist.
> 	It would be very interesting having a 9k6bps modem that works without
> needing DC frequency response.

This could already be done with the original modem design, you just need a
DFE in the receiver.
This is done e.g. in the ALEF NULL project (see PAM-Signaling)
ftp://ftp.tapr.org/dsp/Motorola/dsp56001/dsp_card_4/doc2.pdf

The problem is: if you have a transceiver with bad DC response a receiver
with DFE could still be used. But nobody we a "normal" receiver can copy
your signal well. So you better go and improve your TX to be "compatible"...
I would perhaps make sense to have something like that in the receiver of
the digipeater to allow the users to have a bad DC response.
Question is: is anybody still interested in 9k6 anyway :-(