[hpsdr] Mercury: Selection of frequency Segments

Wed Jan 7 07:45:37 PST 2009

Alex takes 7 independent bands each N kHz wide where N is (say) 200 kHz
wide.

With polyphase filter banks, one on each of the of the 7 independent bands,
we construct M parallel receivers in software.  If we make these 500 Hz wide
channels,  then we have  7 * 200 kHz / .5 kHz = 2800 channels!!!!!!!!  I
assure you the computation to process each of the 2800 channels is only
slightly more expensive than the 7 200 kHz channels and then only because of
the need to do fine tuning or recombining of adjacent 500 Hz channels to
make a 4 kHz channel (say).  The critical pieces needed to do this are
efficient structures for the channel filters in the polyphase filter banks.

Believe it or not,  I am about to lead us back to IIR filters.  I do this
because we need good filter shape with less computation and less latency.
Now, I had my eyes completely opened by fred harris this past summer.  I do
NOT need to sacrifice "good enough" linear phase to accomplish this AND I
get to cut latency by about 2/3's at the same time.  I take an IIR low pass
filter, follow it by an appropriately designed all pass filter and this
becomes the channel filter.  Another approach would be, if we really do need
large FIR filters,  fred harris and I proved new Nobel identities.  You can
do the polyphase filter bank where the channel filters are accomplished in
the frequency domain.  The new Nobel identity, which should have been
obvious but apparently has not been, is that you can do the IFFT on the
other side of the "Butler matrix" or polyphase synthesis/analysis DFT to
return the channels to the time domain.

This is a tremendous amount of work to put into code and people are getting
impatient for me to finish so I am pushing like mad as soon as I return from
vacation (PR is nice!!).

We are entering brave new world but let me continue to comment on other
notes:

I disagree with Ahti on the QSD.  My recent work shows that with the QSD,
and suitable modifications we can achieve un heard of IMD DR3,  BDR's that
would make even Elecraft owners happy,  <<BUT>> all of this is completely
contingent on getting suitable DDS's with low enough phase noise that our
measurements of receiver dynamics are NOT phase noise limited.   

My perfect system?  Several QSD based receivers running in parallel with
QS1R/Mercury style receiver and Penelope style DUC.  The Penelope has
sufficient dynamic range and performance to do anything I can think of.  I
can still outperform any A/D based receiver with an appropriately designed
QSD based receiver.  I have tested my wideband image rejection algorithm at
Flex and it reduces the image across the band to the noise floor and no loss
in MDS that I can detect.  The "see I can do it" implementation is in DttSP
on cgran and in the test iqtest branch for Flex.

SO with a finally good DDS with really good phase noise and low spurious
emission,  no shortcuts taken on components in the QSD (good comparators
rated well beyond the frequency range of interest),  the balanced design QSD
will still produce the better single channel receiver.

That said,  there is a a serious limitation to the QSD.  I just cannot see
it listening to 80, 40, 20, part of 15, part of 10, and ALL at the same
time.  THAT we can get with QS1R and Mercury.  I cannot wait.

Bob

ARRL SDR Working Group Chair
Member: ARRL, AMSAT, AMSAT-DL, TAPR, Packrats,
NJQRP, QRP ARCI, QCWA, FRC.
"And yes I said, yes I will Yes", Molly Bloom

-----Original Message-----
From: hpsdr-bounces at lists.hpsdr.org [mailto:hpsdr-bounces at lists.hpsdr.org]
On Behalf Of Steven Wilson
Sent: Monday, January 05, 2009 8:08 PM
To: hpsdr at lists.hpsdr.org
Subject: Re: [hpsdr] Mercury: Selection of frequency Segments

***** High Performance Software Defined Radio Discussion List *****

Bob,

Okay - Now you've got me completely confused.  (Not terribly difficult
to accomplish mind you... since I can't count above 1 being a digital
kinda guy..)

I've already posted my basic understanding of how Mercury runs. What I
"thought" Alex was providing was a brute force equivalent fitting the
same basic architecture times 7.  In my simple (minded) model that would
mean 7 separate phase accumulators driving the separate cordics doing
the DDC. 

So please oh wise and wonderful guru -  explain in simple words how we
take this to HUNDREDS of receivers?  Does it have something to do with
the width of the sampled base band? 

73 de STeve KA6S

Bob McGwier wrote:
> ***** High Performance Software Defined Radio Discussion List *****
>
> Wait until the polyphase filter bank code I am doing channelizes the N-khz
> wide channels  Alex will provide  and does so in a manner you can combine
> multiple channels easily for quantized but variable bandwidth.  With what
> Alex is doing and this code,  we can run HUNDREDS of receivers in the same
> amount of CPU time now processing the seven channels at full rate.  We
> should be shot for not doing this before now.  Again,  Alex is helping
very
> much to lead the way and being a contester myself,  I hope this upsets
every
> apple cart in town.  The stuffed shirts!  The very idea of writing a
> statistical detection of use of CW skimmer and automatic promotion from
> unassisted to assisted when this is used sets my teeth to grinding.
>
> Keep up the good work Alex!
>
> Bob
>
>
> ARRL SDR Working Group Chair
> Member: ARRL, AMSAT, AMSAT-DL, TAPR, Packrats,
> NJQRP, QRP ARCI, QCWA, FRC.
> "And yes I said, yes I will Yes", Molly Bloom
>
>
>   
>
> _______________________________________________
> HPSDR Discussion List
> To post msg: hpsdr at hpsdr.org
> Subscription help: http://lists.hpsdr.org/listinfo.cgi/hpsdr-hpsdr.org
> HPSDR web page: http://hpsdr.org
> Archives: http://lists.hpsdr.org/pipermail/hpsdr-hpsdr.org/
>
> _______________________________________________
> HPSDR Discussion List
> To post msg: hpsdr at hpsdr.org
> Subscription help: http://lists.hpsdr.org/listinfo.cgi/hpsdr-hpsdr.org
> HPSDR web page: http://hpsdr.org
> Archives: http://lists.hpsdr.org/pipermail/hpsdr-hpsdr.org/
>   

_______________________________________________
HPSDR Discussion List
To post msg: hpsdr at hpsdr.org
Subscription help: http://lists.hpsdr.org/listinfo.cgi/hpsdr-hpsdr.org
HPSDR web page: http://hpsdr.org
Archives: http://lists.hpsdr.org/pipermail/hpsdr-hpsdr.org/

 1231343137.0