Language/OS - Multiplatform Resource Library/LANGUAGE OS.iso/dsp/dspgroup/cis_dsp.arc/CIS_DSP1.TXT

Text File | 1988-04-27 | 85.8 KB | 2,027 lines

#: 69514 S17/TAPR NNC/DSP
    03-Feb-88  23:50:36
Sb: #DSP Wefax demodulator
Fm: Bob McGwier N4HY 74615,1366
To: ALL

Let's get this section going.  Tom and I did our traveling DSP
circus at the AMRAD meeting this past Monday to a very receptive
crowd.  Their newsletter, sent to us on a complimentary basis,
pretty much says it all.  They are picking DSP as one of he
five topics to base their future on and the newsletter membership
application asks if you are interested in DSP, PACKET NETWORKING,
Spread Spectrum, and a couple of others that slip my mind (read
that not interested ;-)).
At 3 AM, Monday morning, sitting in Tom's basement, I got the
WEFAX demodulator working.  I could hear that it was tracking
subcarrier of NOAA 9 perfectly (tape).  Then a little "C" program
gave a few hints at weather pix on the screen.  Tonight I finished
the EGA version and WOW.  Clouds, ice flows, grey scales, time
marks, the whole bit dumps to the screen.  I am quite excited
about it.  James Miller reports that he has been doing some FFT
work.  Steve Sagerian reports that the DSP56000 cards are being
shipped to me (Motorola's answer to the TMS320C25) and he is
laying out an A/D D/A board for it and procuring parts.
Dave Trulli, NN2Z works for AT&T.  He and I have written a proposal
for his boss (   ;-) ;-)  ) who is the microprocessor division
director at AT&T in Holmdel where Dave works.  This will receive
favorable consideration and then we will have some of their
DSP-32C's which is another neat looking DSP chip.  In the meantime,
the TMS320C25 board layout from TAPR/AMSAT is being delayed by the
race against time to get PACSAT ready (we just have all figured
out that we can't do five jobs at once, so we have cut back to four!)
N4HY   LET ME HEAR FROM YOU ALL

*** There is a reply:   69519

*** More ***

#: 69519 S17/TAPR NNC/DSP
    04-Feb-88  01:17:16
Sb: #69514-DSP Wefax demodulator
Fm: David Toth VE3GYQ 72255,152
To: Bob McGwier N4HY 74615,1366

Nice to see yo in this section.
I was getting lonely.
So what is the status of porting code to the PS186 that will be free-standing
(ROMable or boot-loader via radadio) ???

73, Dr. Death


#: 69518 S17/TAPR NNC/DSP
    04-Feb-88  01:16:30
Sb: 32010 Boards
Fm: Tom Clark W3IWI 71260,3640
To: all

I'm about to put in an order for a few more of the Delanco Spry 32010
25 MHz boards that we have been using to date. Now is a good time
to sign up. The cost $525 each.
Tom


#: 69937 S17/TAPR NNC/DSP
    11-Feb-88  23:55:33
Sb: #69865-DSP software
Fm: Bob McGwier N4HY 74615,1366
To: Alberto E. Zagni I2KBD 71360,3467 (X)

Terrific Alberto.
I found the last bug in the DSP display software initial offering
tonight.  It was staying locked onto the frame time very well.
I WAS CHECKING THE KEYBOARD FOR A CHARACTER BETWEEN EACH PIXEL.
It is a miracle it worked well at all.  Now the pictures stay
nicely lined up, IF YOU HAVE a decent recorder or take it directly
off the satellites.  This WEFAX-APT NOT HF Wefax, just want to make
that clear.
Another thing, I don't know where the "collective wisdon of the ages"
came down in the handbooks that a PLL will not phase lock onto the
subcarrier of the Meteor (Soviet) spacecraft.  I set up my PLL
demod (in software) at 2500 Hz and when there is sufficient modulation
we get sufficient "error voltage" (HI) to lock the phase of the loop
and the during the black (no modulation) periods, it stays locked
as the signal is phase continuous.  The pictures are much more
contrasty than NOAA pix given my current pixel mapping.  I have
not yet made any attempt to do contrast enhancement or to save the
pictures for display offline.  (EGA pix eat up a lot of room).
To work on this stuff, you will also need MASM  or some other
assembler as the display code is a 260 byte .COM file ;-)!!
Bob
.


#: 70015 S5/Amateur Satellites
    14-Feb-88  09:17:16
Sb: #69977-FO 12 HELP
Fm: Bill Coleman AA4LR 76067,2327
To: Bob McGwier N4HY 74615,1366

Bob,

Speaking of DSP, what have you been doing with it lately? I've been doing some
reading, and have learned a bit more than before. After studying FFT
algorithms, I'm curious to know more about your software spectrum analyser. How
many points in the FFT, how often you sampled points, etc?

Sounds like you have been busy writing modem code! How many different types of
signals can you demodulate now?

I'm looking to slave a 320C25 to a 68000 processor, and using 100ns ram to
interface to the '25 with 1 wait state. A TLC32041 will be used for A/D D/A
conversion. About the 32041 -- need a sample? My friend Bob got DOZENS from the
local TI rep. They like us now that we put TWO 32010s in each V-Series
SM9600....

Good luck with your demonstration!

   73, Bill


#: 70000 S5/Amateur Satellites
    14-Feb-88  00:09:51
Sb: DSP and WEFAX I
Fm: Bob McGwier N4HY 74615,1366
To: ALL

A funny thing happened on the way to the latest DSP gee whiz ;-)

I have been working on a WEFAX-APT (NOAA, Meteor,GOES, and Meteosat)
demodulator for the DSP boards available to the members of the AMSAT/TAPR
DSP team.  I claim to be the first (hey Tom, gotcha!) to copy NOAA-11
FAX transmissions outside of the people constructing it!

This statement is enough of a clue to let you know that the DSP WEFAX demod
works.  The demodulator took less than a day (Tom and I started work on
one Sunday night at his house and at three the next morning, their was the
semblance of a NOAA picture on the screen).  The display software running
on the PC has taking considerably longer and I am still not happy.

The demodulator is a simple first order phase locked loop.  This works as
the way these systems work is to encode brightness as amplitude on a
subcarrier (a tone).  A simple "Costas loop" works here with one arm
aintaining small phase error and the other returning the amplitude
information.  This is all in software on the DSP card.  The amplitude
modulates the subcarrier in an FM signal.

Bob



#: 70001 S5/Amateur Satellites
    14-Feb-88  00:10:04
Sb: DSP and WEFAX II
Fm: Bob McGwier N4HY 74615,1366
To: ALL

For all but the Soviet (Meteor) satellites, this is 2400 Hz.  For the
Meteor it is 2500Hz.  To make this change is to change a single number
in the program for the TI TMS32010 DSP chip.  It has been the wisdom of
the ages that a PLL cannot lock to the Meteor signal.  I simply do not
understand this statement as the signal appears to be phase continuous
and is pulse modulated during sync so that no phase jumps occur (see the
ARRL Handbook and Satellite Experimenters Handbook and any of Taggart's
later books).  Each of these references I have read say that because the
modulation falls below "5%"   (modulation index < 5%), that a PLL can't do
the job. A first order PLL is all that is needed here as the subcarrier
frequency is "constant" and their is no doppler effect on this as the
signal is FM.  The phase error signal during periods of high amplitude are
more than enough to achieve lock and during periods of low amplitude, the
frequency of the VCO and the phase offset remain constant as the phase error
signal is small.  Thus the PLL "flywheels" right through the low amplitude
spots. The pictures on my screen (they are better than NOAA for clouds and
ground snow cover), attest even further to the incorrectness of these
statements.

The display is done on a monochrome monitor being driven by an EGA card
capable of working with a monochrome monitor (many popular varities of
these today.  Mine is ATI's EGA wonder).  The colors are encoded as
shades of grey.  Their are 16 shades.  So the pixels are quantized to
16 possible grey levels.


#: 70009 S5/Amateur Satellites
    14-Feb-88  00:21:32
Sb: DSP and WEFAX III
Fm: Bob McGwier N4HY 74615,1366
To: ALL

Back to NOAA-11:  While working on the stuff the other day a NOAA signal
appeared on my receiver at a time I was certain that neither NOAA 9 or
10 were visible.  I started up the tracking program and looked at the
satellites of NOAA type, none were visible.  I thought, I know there hasn't
been a launch as we have been asking them could we have a ride on their
next Atlas launch.  Hmmmm.  Well, one and a half hours later, with signal
still (!) present, I knew something was afoot.  I began swinging my beam
around and found the area where the signal peaked and looked out my window
and saw it was pointed right at a tower at GE (ex-RCA) Astro.  Then I
remembered who built them.
They were testing NOAA-11 (well, it will be called that after launch) in
he labs.  You can imagine the boon to my efforts of having more than five
straight hours of WEFAX-APT signal with no antennas to be fooled with etc.
I found the worst bug in the display software and now I am cursing my tape
recorder manufacturer for not being to meet their specs on wow and flutter.

Tom and I will be doing the "Tom and Bob traveling DSP circus" at the
TAPR meeting this coming weekend.  I will be unavailable after Wednesday.

Bob N4HY

#: 70069 S5/Amateur Satellites
    14-Feb-88  22:49:34
Sb: #70000-#DSP and WEFAX I
Fm: BRUCE A. RAHN WB9ANQ 74017,2454
To: Bob McGwier N4HY 74615,1366 (X)

Bob,

All the DSP work you guys are doing sounds *GREAT*.  You have come a long way
since the AMSAT Symposium in Detroit!  I cannot wait any longer!!!! Would you
please jog my memory as to which/whoes DSP board you are using. I copied the
info down in Detroit, but as Murphy has it, I lost it.

Thanks!

Bruce

*** There is a reply:   70110

*** More ***

#: 70110 S5/Amateur Satellites
    15-Feb-88  10:52:44
Sb: #70069-DSP and WEFAX I
Fm: Bob McGwier N4HY 74615,1366
To: BRUCE A. RAHN WB9ANQ 74017,2454

Bruce:
Thanks for your interest.  This is a long way from a finished product Bruce
and we are working in "spartan conditions" if any REAL DSP board can be called
spartan conditions.  AMSAT/TAPR are offering a special buy for  TMS32010 25
Mhz boards that are manufactured by Delanco-Spry in Silver Spring, Md. (a ham
radio acquaintance of Tom and I).  We are getting the boards are a terrific
price under a group buying plan.  Please send a request for information to Tom
Clark, W3IWI, 6388 Guilford Road, Clarksville, Md.21029.  He will mail you a
form.  If you decide to purchase this board, you will send your money to the
AMSAT office where Martha is keeping a special account for this purpose.  I
don't know you and I don't want to insult anyone who has a keen interest in
this stuff but let me just say that this round is indeed for the serious
experimenter who can't stay out of his basement and away from the TV after he
is done eating dinner at night if he even remembers to eat dinner ;-).  We are
beginning to lay out plans for a more user friendly product that works with
different computers and can be optimized for different applications.  This
dovetails with our plans for a PACSAT not later than 1990 that will have a
higher bit rate than JAS-1 and our first PACSAT, which will use the PACSAT
standard.  The tariff will be the same order of magnitude of this first
generation DSP board we are using now.  We really need folks who know about
the guts of their computer and aren't afraid of tackling a new assembler even
if they don't completely understand the algorithms they are being asked to
code.  If this sounds like you, then we need you.
Bob N4HY

#: 70106 S5/Amateur Satellites
    15-Feb-88  10:52:09
Sb: #70015-FO 12 HELP
Fm: Bob McGwier N4HY 74615,1366
To: Bill Coleman AA4LR 76067,2327 (X)

Bill:
That is a WONDERFUL offer.  Yes I need as many as you can spare ;-).
Seriously, I will use whatever you send me.  Lyle is beginning to consider the
first prototype so that some software development can go on.  We are talking
in terms of 0 waits ($$$ for RAM).  I would really like to get that going if
we can fit that into the scheme of things for the next year.  The TLC320XX
series is another reason that it is much easier <<AT THIS TIME>> to use TI
parts for DSP.  I personally like the architecture of the Motorola DSP5600X
better, but the loads of software and all the glue chips that are missing if
you use TI make it very attractive in terms of "time to market".   On the FFT
we have two versions.  The FFT is sample rate dependent but we can get 4 1024
point transforms into a second if we do other things like display it and
average it (noncoherent averaging to help weak signal detection.  This lowers
the variance considerably in the amplitude of each pixel in the display).  I
saw a really neat book the other day called DIGITAL FILTERS (an innocuous
sounding title ;-) but I forget the author.  Maybe I will remember to find out
who that is and post it to you.
Bob

#: 70198 S5/Amateur Satellites
    16-Feb-88  17:57:49
Sb: #70110-#DSP and WEFAX I
Fm: Bill Bard 75366,2557
To: Bob McGwier N4HY 74615,1366 (X)

Bob, depending on the amount of work I doing at the time (we're currently
testing flight hardware with big incentives to get it out of here) I would be
willing to help get the TMS320 on a user friendly product. Are you thinking of
making it stand alone? If you plan on only using it on a IBM type computer, I
refuse to help, but could help get it working on a Mac 2. We have lots of
experience at work with the chips, using them in some real fancy modems. Later.
---Bill
PS I'd be interested in getting a copy of the software for the WEFAX modem
sometime, no hurry.

*** There are replies:  70216, 70271

*** More ***

#: 70216 S5/Amateur Satellites
    16-Feb-88  23:31:40
Sb: #70198-#DSP and WEFAX I
Fm: Bob McGwier N4HY 74615,1366
To: Bill Bard 75366,2557 (X)

The ultimate goal is to produce a piece of hardware that even a commode door
can make use of.  It will take the faster computers to make use of the WEFAX
APT, the moderate bit rate modems, the FFT software (on the fly) but we want
this to have a HUGE audience.  This is how we pay for our future projects by
making the current ones desirable.  The only way it works is we make a thing
that lots of people want and we license it to a manufacturer to make it.  I
myself do not wish it limited to PC's.  It will be limited to those computers
that have some kind of attachment to the bus (even a C-64 has this, why oh why
did Apple close the early Mac?).
Bob N4HY

*** There is a reply:   70309

*** More ***

#: 70309 S5/Amateur Satellites
    18-Feb-88  10:13:01
Sb: #70216-DSP and WEFAX I
Fm: Bill Coleman AA4LR 76067,2327
To: Bob McGwier N4HY 74615,1366

Bob,

My SE has attachment to the buss. So does a Mac II (an even better buss).

One thing I'm considering is the use of SCSI or AppleTalk to communicate with
the DSP board. SCSI is fast, but takes extra hardware to implement. AppleTalk
won't require more hardware (I plan on installing a 8530 on the board) but will
require some about of software for the board. Appletalk is about 10 time
slower, but still faster than pure serial I/O.

Still, even serial I/O at 9600 should be fast enough for WEFAX or SSTV (bet you
hadn't thought of that one....) data, if sent continuously. Problem is
processing all that serial data. That takes a lot of rather careful planning,
and I don't think it can be accomplished with a C-64.

Still, SCSI or AppleTalk are good bets. Both are pretty standardized in the
industry. For the C-64 or PC, direct buss connections are the best.

    73, Bill

*** More ***

#: 70271 S5/Amateur Satellites
    17-Feb-88  22:29:52
Sb: #70198-DSP and WEFAX I
Fm: Bill Coleman AA4LR 76067,2327
To: Bill Bard 75366,2557

Bill,

Amen to that. Here's another would-be DSP experimenter with a Mac. Mines an SE,
though.

   73, Bill

*** More ***

#: 70263 S5/Amateur Satellites
    17-Feb-88  20:59:24
Sb: #70110-DSP and WEFAX I
Fm: BRUCE A. RAHN WB9ANQ 74017,2454
To: Bob McGwier N4HY 74615,1366

Bob,

Thanks for the reply to my note.  My interest in DSP is two fold: a) an intense
personal interest with emphasis, of course, on amateur radio; and b) a
professional interest.

Of course I'm interested in the AMSAT/TAPR deal and will drop Tom a line. I'm a
basement hardware hacker (is there any other kind?) and an occasional dabbler
in assembly language programming so I'm not too afraid to tackle something like
this (I may fall flat on my face, but I'm sure to have a ball!).

My initial question to you, which I didn't word too clearly, was who builds the
DSP card you had at Detroit and what is the part/model number?

Thanks for taking the time to answer my question Bob.  Good luck in the DSP
development arena.  See you at Dayton (?).

73--Bruce


#: 70199 S5/Amateur Satellites
    16-Feb-88  17:58:02
Sb: #70001-DSP and WEFAX II
Fm: Bill Bard 75366,2557
To: Bob McGwier N4HY 74615,1366 (X)

Think about building a small circuit (or I could do it) that would demod the
wefax picture and output the picture data as a serial digital bit stream that
has header info describing the data (ie, resolution, number of pixels per
line).
Got all that? I'm interested in feeding the data into a mac for display and
processing. ---Bill


#: 70200 S5/Amateur Satellites
    16-Feb-88  17:58:11
Sb: #70009-#DSP and WEFAX III
Fm: Bill Bard 75366,2557
To: Bob McGwier N4HY 74615,1366 (X)

What were the pictures of, the lab? ---Bill

*** There is a reply:   70217

*** More ***

#: 70217 S5/Amateur Satellites
    16-Feb-88  23:31:47
Sb: #70200-#DSP and WEFAX III
Fm: Bob McGwier N4HY 74615,1366
To: Bill Bard 75366,2557 (X)

Yes, they were mostly of a light bulb across the room or in the thermal vac or
where ever they have it.  There was other stuff but it spent hours on the
light bulb.
Bob


*** There is a reply:   70233

*** More ***

#: 70233 S5/Amateur Satellites
    17-Feb-88  01:02:04
Sb: #70217-DSP and WEFAX III
Fm: Phil Karn, KA9Q 73210,1526
To: Bob McGwier N4HY 74615,1366 (X)

Bob and Bill,

Remember that the NOAA birds don't carry cameras in the conventional
sense, they have "scanning radiometers".  These contain mirrors that
simply scan a line perpendicular to the path of the moving satellite.
The other axis is "scanned" by the satellite's own orbital motion,
giving a continuous picture "strip".  So if you put a fixed object in
its field of view, you won't get a recognizable image of it, just a long
streak in the picture.

Phil


#: 70326 S5/Amateur Satellites
    18-Feb-88  23:14:30
Sb: #70216-#DSP and WEFAX I
Fm: Bill Bard 75366,2557
To: Bob McGwier N4HY 74615,1366

I guess it depends on what you want to be able to transfer to and from the
unit. Would a RS232 link at 19200 baud be acceptable or do we want an 8 bit
parallel (SCSI) bus or do we want an attachment to the bus (8/16/32 bits)?
Areas that I can help (work load permiting) are RS232, SCSI (maybe; I don't
know alot about it but what could be difficult) or attaching to the NuBus in a
Mac 2. The later may be of limited value since not alot of people will be
getting one for a while, but I might do it anyway on my own for my own use or
for use at work. I agree with the principle of keeping it as simple as
possible.
At least the Mac2 is open. With the floating point processing chip, it screams.
As soon as I get a house, I'll be buying one. Later. ---Bill

*** There is a reply:   70402

*** More ***

#: 70402 S5/Amateur Satellites
    20-Feb-88  18:48:52
Sb: #70326-#DSP and WEFAX I
Fm: Bill Coleman AA4LR 76067,2327
To: Bill Bard 75366,2557 (X)

Bill,

Running AppleTalk in the remote board is also a possibility. The first few
layers (ALAP, DDP, NBP, RTMP stub) are pretty simple to implement and would
allow more than enough bandwidth.

For using a board with a Mac, I propose connectivity to a Mac Plus is the least
common denominator. That leaves serial, Appletalk and SCSI. SCSI is about 10
times the throughput of Appletalk, which is about 20 times faster than straight
serial.

Keeping it simple might involve starting with straight serial and moving to
Appletalk if more bandwidth is needed. SCSI requires special hardware. NuBus
isn't very easy to interface to. It is powerful, but has a number of special
requirements.

If I keep talking about this, I'm going to sell myself on AppleTalk....

    73, Bill

*** There is a reply:   70474

*** More ***

#: 70474 S5/Amateur Satellites
    22-Feb-88  00:06:52
Sb: #70402-DSP and WEFAX I
Fm: Bill Bard 75366,2557
To: Bill Coleman AA4LR 76067,2327

Bill,

Maybe the way to go is to have the circuit such that I/O is handled thru some
generic 8 bit parallel port. If you want to do RS232, then the required
hardware/software for that would be added. For other I/O methods you could
change the circuit as you get time/experience like you said. I have data on
Appletalk protocols, but don't know how easy/hard it would be to implement it.
I feel better about RS232 and NuBus (I have more info on NuBus than SCSI and
feel more comfortable, providing I can get a NuBus development card which I
should be able to). RS232 may be desired since most every computer has that
interface. ---Bill

*** More ***

#: 70327 S5/Amateur Satellites
    18-Feb-88  23:14:41
Sb: #70233-DSP and WEFAX III
Fm: Bill Bard 75366,2557
To: Phil Karn, KA9Q 73210,1526

Yeah, that's right. The only way I can see it would display normal stuff would
be if they simulated the camera output with a normal type camera so that they
get something to modulate the downlink to test the RF portion. ---Bill



#: 70269 S5/Amateur Satellites
    17-Feb-88  22:29:34
Sb: #70106-FO 12 HELP
Fm: Bill Coleman AA4LR 76067,2327
To: Bob McGwier N4HY 74615,1366

Bob,

I'll see what Bob (WA1EDJ) says. I can get you a couple, and maybe as many as a
half dozen. Send me your address, and I'll let you know when I send them off to
you.

If I could bias the considerations for the DSP hardware.... After much talking
with friends who are DSP engineers, I'm convinced that the TMS32020 or C25 is
the way to go. (For now, the C30 will be when it comes out) Why? Many reasons:

1) More data RAM than the 32010. The 20 and 25 have 544 words, about 4 times as
much data RAM on chip. This allows considerably longer filters and much more
room for general hacking about.

2) Chip is faster, running with a 40 Mhz clock. Having more CPU allows running
much more complicate algoritms.

3) Enhancements to the architecture (like more AR registers -- 5 on the 20, 8
on the 25) and a more orthogonal instruction set allow easier programming.

4) FIR filters can be implemented in 2 instructions (and a table of constants)
on the 20 and 25 (RPTK; MACD). This sequence is carefully pipelined to allow
single-cycle execution. (MUCH! faster than the 32010)

5) Interfacing to A/D D/A chips such as CODECs are trivial because of the
built-in serial interface. Serial port is double buffered.

6) The 25 contains a bit reversed addressing mode that makes FFT algorithms
much shorter and faster.

7) Slow memory may be used without much penalty, even with 1 wait state.
Processor is well-pipelined. 256 words of on-chip RAM can be used to hold
program instructions if faster execution is necessary.

Overall, I find the 25 a better solution than the 32010. We have some C25
algorithm development boards at work, and they are very simple, yet very
impressive.

I haven't looked at the 56K chips.

For software development, I'd recommend a 320C25, a 32041 (you can't get 40s,
but the reference voltage circuit is trivial) and a 16-bit buss processor to
slave the DSP to. I'm using a 68000. For PC boards, the host processor will
work just fine. Adapting 16-bits to 8 is a pain, though. 68K is overkill, but I
have a development system for it. If I ROMed the DSP code, a 8031 (or some
such) would be plenty, and might be unnecessary.



*** More ***

#: 70270 S5/Amateur Satellites
    17-Feb-88  22:29:46
Sb: #70106-FO 12 HELP
Fm: Bill Coleman AA4LR 76067,2327
To: Bob McGwier N4HY 74615,1366


[Continued]

I'm interested in anything you have to say about DSP, Bob. I'm no expert, but
I'm learning! Soon, hopefully, I'll be able to do something with this
technology....

Bob, If you are handing out DSP code for ANYTHING, I'd like to give it a
look-see. Many times, all the theory becomes a lot clearer when you have
examples to follow.

     73, Bill


#: 70526 S5/Amateur Satellites
    22-Feb-88  21:34:48
Sb: #70263-DSP and WEFAX I
Fm: Bob McGwier N4HY 74615,1366
To: BRUCE A. RAHN WB9ANQ 74017,2454

Sounds perfect Bruce.  We need all the basement hackers we can stand to have
around.  I just feel it is my duty to tell folks this is the S100 era and we
are slowly building to the 68000/ 80286 era ;-).
Delanco Spry Model 10 (25 Mhz) version.
Silver Spring, Md.

Bob



#: 70528 S5/Amateur Satellites
    22-Feb-88  21:35:08
Sb: #70309-DSP and WEFAX I
Fm: Bob McGwier N4HY 74615,1366
To: Bill Coleman AA4LR 76067,2327 (X)

WRONG DSP BREATH ;-).
I was talking to some Slow Scan folks the other day about doing that.  If
you will look in the last PSR where Lyle wrote it up, he mentioned the slow
scan and WEFAX demods.  We're getting there.  I am dumping 625 bytes per
second to the screen to do the WEFAX-APT demod and I would need roughly ten
times that get slow scan to the same resolution.  The resolution would be
about 4-5 times better than the old Robot 400's so maybe that can be
sacrificed.  The next hot topics for McClarnon, Clark, myself and probably
Dan Morrison seems to be an M-ary FSK modem to get the most out of HF.  I
believe we are beginning to make an impact as we all get better at doing
this.
Bob N4HY

*** More ***

#: 70570 S5/Amateur Satellites
    23-Feb-88  10:53:50
Sb: #70474-DSP and WEFAX I
Fm: Bill Coleman AA4LR 76067,2327
To: Bill Bard 75366,2557 (X)

Bill,

I think I'm going to start off with a simple serial interface, and work up to
AppleTalk if more performance is needed. I'll just have to keep the interfaces
compatable.

I just found a great book on DSP. "Designing Digital Filters" by Charles S.
Williams. Prentice-Hall 1986. ISBN 0-13-201856-X. This book is for "the rest of
us" who are mathematical wizzos. All that is required is a working knowledge of
trigonomitry and calculus. The book even contains a review of complex numbers
and an introduction to analog filters. All notational conventions are
thouroughly explained before they are used, and the book introduces each new
concept with a easy-to-understand example.

The book starts with an overview of the impetus around designing digital
filters, discusses the issues associated with digital filtering (aliasing,
quantization). The next chapter reviews the mathematical concepts of frequency
response. The next two chapters discuss the design and implementation of
nonrecursive filters, then two chapters on recursive filters. The final two
chapters talk of polynomial modeling of digital signals and the DFT and FFT.

An excellent book for those of us who are not EEs, nor Math majors. In all, a
very readable book!

    73, Bill

*** More ***

#: 70527 S5/Amateur Satellites
    22-Feb-88  21:34:57
Sb: #70269-FO 12 HELP
Fm: Bob McGwier N4HY 74615,1366
To: Bill Coleman AA4LR 76067,2327 (X)

Your are preaching to the choir!! ;-)  We are heading towards a 25 board.
We might come out with an optimal adaptive HF modem using M-ary FSK.  While
FSK is not the optimal data encoding scheme, for the environment it will
operate in, we believe it is.  This 10 board will be used to do a 9600 BPS
FSK modem, which doesn't take those long filters, the HF modem, and a WEFAX
HF and WEFAX-APT demodulator.  Thanks for you tips, they are appreciated.
Bob
N4HY


#: 70638 S5/Amateur Satellites
    24-Feb-88  14:41:37
Sb: #70528-#DSP and WEFAX I
Fm: Bill Coleman AA4LR 76067,2327
To: Bob McGwier N4HY 74615,1366 (X)

Bob,

Damn! And I thought I had an original idea! Unfortunately, I don't get the PSR,
so I can't read what Lyle wrote. Looks like I have to find some back issues.

How are you looking at doing m-ary FSK? FFTs?

Found an excellent book on DSP. Charles Williams "Designing Digital Filters."
See my comment to Bill about it.

    73, Bill

*** There is a reply:   70674

*** More ***

#: 70674 S5/Amateur Satellites
    25-Feb-88  00:05:38
Sb: #70638-DSP and WEFAX I
Fm: Bob McGwier N4HY 74615,1366
To: Bill Coleman AA4LR 76067,2327

Yes it is a good book.  Actually, I find that the entire Prentice Hall DSP
series (Oppenheim and Schafer, Hamming (of Hamming window), others) is worth
having.  They also have a TI TMS320 course set that is interesting.
Bob


*** More ***

#: 70650 S5/Amateur Satellites
    24-Feb-88  19:31:28
Sb: #70570-DSP and WEFAX I
Fm: Bill Bard 75366,2557
To: Bill Coleman AA4LR 76067,2327

Bill,

Sounds good to me. I'm an EE and took a few graduate courses in DSP including
using the TMS320 chip so I'm familiar with it, but haven't done much at work
with it yet. Others in my department have done alot so I have a good place to
go for questions. Later. ---Bill

*** More ***

#: 70639 S5/Amateur Satellites
    24-Feb-88  14:41:45
Sb: #70527-#FO 12 HELP
Fm: Bill Coleman AA4LR 76067,2327
To: Bob McGwier N4HY 74615,1366 (X)

Bob,

Any time! 9600 FSK? I wouldn't think there's enough shift room for 9600 bps
data. Hmm.

I'm first looking at simple FSK stuff. When I get more experienced, perhaps
I'll look at more complex demodulation like SS or WEFAX. My goal is to have the
best demodulation possible -- digital or analog.

How are you planning to do 9600 bps FSK in 4 KHz of bandwidth? There's not that
much room to shift! 9600/4 is 2400 Hz, which is still pretty big! I guess it
could be done. And if it can be done, you all are the guys to do it!

   73, Bill

*** There is a reply:   70675

*** More ***

#: 70675 S5/Amateur Satellites
    25-Feb-88  00:05:46
Sb: #70639-FO 12 HELP
Fm: Bob McGwier N4HY 74615,1366
To: Bill Coleman AA4LR 76067,2327

Who said anything about 4 Khz?   Not me!  At any rate, we are going to have
to build a simple xcvr for the higher bit rates.  Rock freq control, etc.
Bob



#: 70115 S17/TAPR NNC/DSP
    15-Feb-88  11:11:18
Sb: #70062-Correction
Fm: Bob McGwier N4HY 74615,1366
To: Barry McLarnon VE3JF 71470,3651 (X)

Hey Barry:
That is great.  I wonder if I could interest you in working with me on an
M-ary FSK demod if you haven't done it already.  I am sure you have some ideas
that I don't know about that would help this immensely.  To me if would be
wonderful if we could come up with a 300 bps that used 8 or 16 different tones
and thus transmitted at 100 or 50 baud.  I like the 50 baud better for HF but
I will rely on your experience to tell me if that is the right range.  The one
problem we have with the DSP cards we are using at the moment is the size of
the DAMN data memory.  If we are going to discriminate between tones that
means filters and do be at least twenty dB down and get 8 or more of them in
the passband of a sideband XCVR, we need them to have quite a few data points.
We can do the filters using the same data and only slide them down in data
memory in the last filter so that we multiplex 8 filters on the same data
memory (say).  If we use multiple tones and encode them in a way that (say)
the third is "redundant" or some such science fiction, we can get the baud
rate down even further.  Do you have something like this in mind?  Let me set
my priorities for you in the DSP stuff now.  I have to return to the weak
signal work with Tom or he is going to have Elizabeth stick poison in the next
drink I get over at there place ;-).  It is just possible that we will do a
VERY slow m-ary FSK for this.  It will <NOT> scale up to faster speeds but
should be interesting to work with none the less.  I believe that we just
might be able to be quite a few records on UHF/VHF if (1) we do this code, (2)
we beat the noise coming out of this board, (3) there is no RF susceptibility
from high powered VHF/UHF amps (and if money grew on trees we would all be
rich).  Any thought, help, criticism, etc. you might have please let me hear
them.  BTW, really nice job on the QEX articles.  You did a great job.
Bob N4HY


#: 70492 S17/TAPR NNC/DSP
    22-Feb-88  10:03:43
Sb: #70115-Correction
Fm: Barry McLarnon VE3JF 71470,3651
To: Bob McGwier N4HY 74615,1366

Hiya Bob, hope you had a good meeting in AZ.

Thanks for the run-down on your DSP priorities, and hope we can get some
similar inputs from the other DSPeople soon.  I wish I could find more time for
this stuff, but my bod (not mention my wife :-)) won't tolerate staying up till
3 a.m.!

Anyway, here's where I'm at right now... My initial tack is to see if we can
wring anything more in the way of improved performance out of the current 300
bps 103-type "standard".  I expect some small improvements to come from
improvements in filtering, and from implementing a more robust clock recovery
scheme than the state machine in the TNC.  Hopefully, more substantial
improvement (although it obviously can't do too much to beat intersymbol
interference) will result from dual space/polarization diversity.  The basic
DSP modem is running now (including modulator, which I'm using to re-modulate
the signal for inputting into the TNC on-board modem), the remaining stumbling
block being how to control an external mux so that I can sample two receiver
outputs.  I want this to be a standalone application that can easily be ported
to another 320-based board, which means using a port on the DSP board for the
external interface.  The thought of involving the PC in this dirty work (by
using, say, one of its parallel ports) REALLY turns me off.  I wish the D-S
board had the signals for one port, at least, brought out to a connector on the
back.  I also need to round up a decent-quality tape recorder for the shack,
and get another antenna up, come Spring.

So... my intentions right now are to spend a bit more time cleaning up the
software, putting in the MUX hardware, and doing some testing with real HF
signals, and then see if anyone else wants to take the code and try it out...

On to M-ary FSK... [next message]

*** More ***

#: 70494 S17/TAPR NNC/DSP
    22-Feb-88  10:04:31
Sb: #70115-Correction
Fm: Barry McLarnon VE3JF 71470,3651
To: Bob McGwier N4HY 74615,1366

How does the following 16-ary FSK scheme grab you?  We set the sample rate to
6400 Hz, and the baud rate to 75 Hz.  Do a 64-point FFT on the central 10 ms of
the 13.33 ms baud interval, and transmit the information on 16 orthogonal tones
which fit through a typical SSB filter, say 700, 800,... 2200 Hz.  The same
receiver structure lends itself to higher-speed parallel FDM-type modem
implementation, i.e., transmission of BPSK or QPSK on 16 simultaneous tones to
get to 1200 or 2400 bps if you are willing to pay the penalty in lower average
power.  75 baud is probably the max symbol rate that should be considered.
Another possible format that would give better protection against multipath is
32-ary FSK at 37.5 baud, with 50 Hz tone spacing/20 ms windows... but that
would require a 128-point transform.

The main problem, aside from the chronic shortage of data memory, is coming up
with a synchronization scheme to keep the processing window centered in the
baud interval.  I can think of a couple of possibilities.  One is to leave a
"hole" somewhere in the middle of the transmitted spectrum by not using one of
the possible tone frequencies; e.g., transmit on any of 600, 700,... 2200
except, say, 1500 Hz.  Then the energy in the bin at 1500 Hz, after some
suitable averaging, should reach a minimum when the window is centered on the
baud interval.  Another possibility is to derive the timing from transitions on
a particular pair of tone frequencies.  The modem could be set up to frame HDLC
flags so that they would map into an alternation between the two tones. The
timing then should be locked in after the alternating pattern goes away.

Whaddaya think?

Barry


#: 70529 S17/TAPR NNC/DSP
    22-Feb-88  21:35:18
Sb: #70494-Correction
Fm: Bob McGwier N4HY 74615,1366
To: Barry McLarnon VE3JF 71470,3651 (X)

Sounds like you Clark and I are thinking like soul mates.  We all knew this
board was the S100 of our DSP project.  We are beginning the layout of the
next boards in Tuscon and we will be discussing that here I am sure as it is
the most convenient forum as to where that will lead.  I will code up the 64
point FFT and even the 128 FFT in the best optimized code I can get going.
At 6700 Hz, it will easily fit into the time frame you wanted.  I can tell
you that Rinaldo is going to love us.  I sent in article #1 for QEX on DSP.
Please consider your experiments for a future issue and if you get done in
time, put it in as one of the series.  The next one is going to be on the
WEFAX-APT demodulator.
Bob


*** More ***

#: 70611 S17/TAPR NNC/DSP
    24-Feb-88  00:40:58
Sb: #70494-Correction
Fm: Tom Clark W3IWI 71260,3640
To: Barry McLarnon VE3JF 71470,3651 (X)

Hi Barry -- just got back from Tucson last nite. Here is the m-ary
scheme I proposed: consider 32 data bits + 10 FEC bits (which should
correct 2 baddies) = 42 data bits. Put these as tones on a 20 Hz
spacing with 40 baud MSK encoding, and they occupy 820 Hz. On the
bottom end add a pilot tone (channel #0) and on the top end add an
alternating 1/0 guaranteed data clock as channel #43. This just fits
in our legal 900 Hz bandwidth, and 40 baud should make it thru any
reasonable selective fades. Data rate is then 32*40 = 1280 bits/sec
so data thruput will be lots better. The total of 44 channels can
best be detected with a 64 point complex FFT preceeded by a Hilbert
transform heterodyne (i.e. a complex mixer) whose LO freq can be
twiddled to provide AFC.
73, Tom

#: 70534 S17/TAPR NNC/DSP
    22-Feb-88  22:18:11
Sb: #Cheap Unit
Fm: Lyle Johnson, WA7GXD 76246,565
To: DSPers

Cheap DSP Notes #1

The basic design is proceeding around the ideas kicked around at my house on
Sunday.  To refresh (and expand) we are looking at the 25 MHz 320C15 (256 words
data RAM - might be good for you Delanco-Spryers who need more data memory);
the Analog Devices AD7569 8 bit analog port with track and hold, D/A, A/D and
reference, along with quick bus interface.  This analog part is dynamically
characterized for DSP applications. 8 bits of TTL digital I and O, some of
which routes to the TNC modem disconnect ala PSK modem header. 4k words of
static RAM.  A loader to dump from 27256/512/1024/etc to the RAM.  An option to
use 32k or 128k by 8 static RAM (battery backed?) with parallel port I/O to
allow dumping software from, say, a PC into the memory storage of this thing so
you don't have to burn an EPROM to test it.  VFET outputs for PTT, UP and DOWN
drivers to the radio. 82C54 programmable timer for sample clocks, etc.

Questions.  Do we need some sort of tuning indicator and "lock" LEDs?  If so,
what form should they take?  Should any "extra" digital inputs be tied to front
panel switches for software-defineable functions?  Is it OK if I make a
glichless clock-stretcher to add 80 nSec when we access the I/O ports, or do I
really have to add all the octal latches and other hoopla instead?  If it only
costs $20 more (on a $100 device, remember), should I look at a 12 bit A/D and
D/A instead?  Is an 8 bit D/A OK along with a 12 bit A/D (so we get more
dynamic range on rcve, who cares on xmit)?

Designing away...  Lyle



*** There are replies:  70613, 70621

*** More ***

#: 70613 S17/TAPR NNC/DSP
    24-Feb-88  00:41:29
Sb: #70534-#Cheap Unit
Fm: Tom Clark W3IWI 71260,3640
To: Lyle Johnson, WA7GXD 76246,565

Re timing: the important thing is the 'smoothness' of the sample/hold
preceeding the A/D. The two advantages of the 12-bit over the 8-bit
is that dynamic range is improved (and hence we are less susceptable
to QSB) and that FIR/IIR filters can be made much more accurate.
.
Re features: I'd like to suggest that the board include a 2206/2211
I/O regenerator to match 202 tones for applications slower than 1200
bits/sec much like the AEA PM-1 does. This makes it so that unmodified
TNCs can be used. I note that several of the TNCs out there in radidio
land (KAM, u21 etc) do not have easy modem disconnect capabilities.
.
The choice of 32010 vs. 32015 still needs to be considered. The extra
program RAM on the 15 (256 vs. 144 words) is very desirable, but
our development tools (i.e. Delanco-Spry board) are 32010 based.
Also the 15 costs more than the 10!
73, Tom

*** There is a reply:   70622

*** More ***

#: 70622 S17/TAPR NNC/DSP
    24-Feb-88  03:17:31
Sb: #70613-Cheap Unit
Fm: Bob McGwier N4HY 74615,1366
To: Tom Clark W3IWI 71260,3640

Tom:
I believe that if TI is consistent, the instruction set it identical. You
access the data memory after 127 by switching banks and then you are able to
address 0-127 again (on the ten only the first 16 mean anything, I guess the
15 gives you all 128).  In the case it is a plug in replacement, I would
guess it will use the identical tools we are now using.  A handful more
bucks for nearly twice the data ram enables many modems probably including
the 9600 BPS FSK/FFSK/GMSK modem Martin needs.  If it didn't I would be
disappointed and would not want us to go with the TMS3201X family anyway.

Bob


*** More ***

#: 70621 S17/TAPR NNC/DSP
    24-Feb-88  03:17:23
Sb: #70534-Cheap Unit
Fm: Bob McGwier N4HY 74615,1366
To: Lyle Johnson, WA7GXD 76246,565

If the 12 bit truly costs only 20 more, I say it never hurts.  Is the 15
pin compatible with the 10?  If so where do I get one?  I suppose that this
is really a TMS32010 with a larger bank 1 than in the TMS32010 and my guess
is that the instruction set is therefore absolutely identical since they
have worked so hard to maintain this upward compatibility thru the TMS320XXX
XXX < C30 families.  The extra data memory makes ALL the difference in the
world as to whether or not we can use this first generation for slighter
faster modems requiring adaptive equalization.  I know already that I can
fit in (say) the arm filters in the first bank for ALL popular two
dimensional signals but I couldn't get the complex equalizer coefficients in
then.  With this ability, we will literally kick most hardware
implementations that I know of out the door.  I didn't know about that chip.
Please give us details.  If it costs $30 rather $15 (;-)  it will be worth
it.  I don't know enough about the other hardware questions you are asking
to answer them.  "Put it into ainglishhhh boy". What we have found on the
Delanco Spry board is too much jitter in the sample conversion time and that
is an AD part number.  Let us proceed with care here.

Bob


#: 70656 S17/TAPR NNC/DSP
    24-Feb-88  22:04:10
Sb: comments 1 of 3
Fm: Lyle Johnson, WA7GXD 76246,565
To: dsp project

NOTE: The following comments are offered in Ainglish...

Regarding the difference in the 32010 versus the 320C15:

1) The data ram is addressed as follows:

DP = 0       dma = 0-127        data address = 0-127   (both units) DP = 1
dma = 0-15         data address = 128-143 (tms32010 only) DP = 1       dma =
0-127        data address = 128-255 (tms32015)

2) The 32015 is object code compatible and pin-for-pin interchangeable with the
32010.  The only difference is 256 words of data memory and 4k words of program
ROM or EPROM rather than 1.5k words of program ROM in the case of the 32010.

3) And, as Tom points out, it costs more money.  The EPROM version is $65 in
100s.  I am still getting a good price on the non-EPROM version.  It is also
only available as a CMOS chip.

Regarding the AD7569:

1) It is a 2 uSec A/D and a 1 uSec D/A.  Reference, etc., is on chip.  DAC is
dynamically characterized for DSP applications with a 44 dB minimum S/N ratio;
48 dB max total harmonic distortion; 55 dB typical IMD; 2's complement
interface.  A/D has 44 dB minimum S/N; 48 dB max THD; 60 dB typical IMD; flat
within 0.1 dB from DC to 200 kHz input signal at full amplitude (2.5 v pk-pk).
Track and hold acquisition time is 200 nSec.  All this in a $6 chip!


#: 70657 S17/TAPR NNC/DSP
    24-Feb-88  22:05:08
Sb: comments 2 of 3
Fm: Lyle Johnson, WA7GXD 76246,565
To: dsp project

2) To go to 12 bits A/D the cheapest approach is ML2233, which has a 30 uSec
conversion time and a 2.3 uSec track and hold acquisition time.  It needs an
external voltage reference.  The ML2233 is about $14 a copy, a reference maybe
$3 and a D/A for system output probably $5 to $13 depending on whether or not
it is 8 bit or 10 bit or 12 bit.  I am not sure if the 30 uSec conversion time
will kill us for the Martinsat application?

3) For cost, simplicity, etc., I would like to go with the AD7569.  The dynamic
range coming out of a radio, especially at audio, shouldn't come close to 40
dB, and it is no big deal to request a user to set a level to an LED on the
front panel.  I think I would rather see a level control on the front than do
the hoop$ to make a gain-programmable front end amplifier.  I would rather the
wider dynamic range A/D, I think, than the amplifier.

4) Motorola makes a linear CODEC with 13 bits resoltuoin and 9 bits accuracy,
but it is limited to a 16 kHz sample rate max. Too slow, I think, plus the
hassle of interface logic.  Costs about $7 with saples in April.

Regarding case:

1) At work we are using a nice Ten Tec enclosure that costs about $30. It is
bi.  For only about $3 more, Ten Tec punches and silkscreens the things for us,
and punches the inside chassis shelf, etc.  I want to design this board to fit
in a cheaper version of this case and pay Ten Tec to punch and silkscreen them
for us.  The switches, connectors, etc., should all mount on the PC board and
this project should wind up the nicest looking one yet.  The set up charges
will only be on the order of $100 for the whole shebang!  And they are cheap
even in lots of 1 to 5, custom.


#: 70658 S17/TAPR NNC/DSP
    24-Feb-88  22:05:45
Sb: comments 3 of 3
Fm: Lyle Johnson, WA7GXD 76246,565
To: dsp project

Regarding board:

1) For noise reasons, I am planning on a 4-layer board.

Regarding TNC interface:

1) We can do an XR2206/2211 provided we gate the 2206 output so the TNC's DCD
can function properly. Likewise, we should use the PTT from the TNC to pull a
resistor pull up to ground and put in our own watchdog timer if we are going to
interface to Kantronics stuff, and AEA multimodes (which lack a transmit
watchdog timer).  The beauty of the TNC modem disconnect is that "real" TNCs
have them and it saves parts (money) on the interface to the TNC.

2) If we are going to provide a manchester uplink for fuji and pacsat, and we
don't have access to the internal clock from the TNC via the modem disconnect,
how will we synchronize the clock to the bit stream?  The 2206/2211
combination, even on a hardwire link, adds quite a bit of jitter to the data
edges...

Lyle


#: 70672 S17/TAPR NNC/DSP
    25-Feb-88  00:05:22
Sb: #DSP, UOSAT, Receiver
Fm: Bob McGwier N4HY 74615,1366
To:  76246,565

Upon careful reflection, we don't need to do that much in the DSP box for
the 9600 BPS modem.  Since we <MUST> also provide the receiver (since when
is there a decent radio to plug one into? , i.e. K9NG) and since they appear
to have adopted the K9NG standard, we should use the PACSAT receiver for
that anyway.  Some of the stuff will have to be done in software such as bit
sync (or clock recovery).  This frees us from having to cram 10 lbs of dung
into a 5 lb can.  They want to eventually go to FFSK and GMSK modulation
schemes as experiments.  Probably by then, we will have the higher speed
DSP tools going and will be able to accomodate it.
The PACSAT receiver will go right down to the discriminator and thus all we
need to is a minimal amount of work in clock and to provide the unscrambling
function unless we decide to go directly with the K9NG modem.  In that case,
I would hope TAPR would kit a complete collection of parts for them rather
have folks scrounge for all but the 1% and caps.
Please remind me to tell Scott that since Harold is working on PS186 with
the work he and Skip are doing on the PS186, they should be enabled for 17
if they aren't already.
Bob


*** There is a reply:   70688

*** More ***

#: 70688 S17/TAPR NNC/DSP
    25-Feb-88  10:17:13
Sb: #70672-DSP, UOSAT, Receiver
Fm: HamNet SysOp Scott W3VS 76703,407
To: Bob McGwier N4HY 74615,1366

Bob,

I'll double-check and make sure Harold and Skip have access to section 17.

Scott

#: 70696 S17/TAPR NNC/DSP
    25-Feb-88  13:09:38
Sb: M-ary FSK Modems
Fm: Barry McLarnon VE3JF 71470,3651
To: Tom Clark W3IWI 71260,3640

Tom, I'm glad you brought up the question of "legal" bandwidth.  My feeling is
that we should not constrain ourselves to limits that were imposed years ago
when Baudot and mechanical teleprinters were the only game in town.  I've been
working under the tacit assumption that we will need a STA (or the equivalent
outside the U.S.) to operate the new generation of higher-speed and/or more
robust modems, and that we will be lobbying for changes in the regs to permit
packet emissions to occupy the equivalent of a voice signal.  If the SSTV types
can use up 3 kHz, and any yahoo with an AM rig is permitted to squander 6 kHz,
then it stands to reason that a superior (error-free!) mode like packet should
be allowed similar bandwidth, provided that it is making reasonably efficient
use of that bandwidth.  As far as I can tell, HF packet exists today only by
virtue of the reg that allows F1 emission with deviation up to +/- 450 Hz.  I'm
not sure that your scheme would be considered "legal" even if it did occupy
less bandwidth than this, since it is not really F1; i.e., it is multichannel
FSK rather than m-ary FSK.  We gotta go for more bandwidth, and the freedom to
try a variety of modulation and coding schemes within the confines of that
bandwidth, if we want to really push the state of the art.

[cont'd in next message]


#: 70697 S17/TAPR NNC/DSP
    25-Feb-88  13:10:58
Sb: M-ary FSK Modems
Fm: Barry McLarnon VE3JF 71470,3651
To: Tom Clark W3IWI 71260,3640

On the technical side, I'm having trouble with some of the numbers you're
quoting.  First off, 20 Hz spacing seems way too low for MSK channels running
at 40 baud.  I have some simulation studies that were done on multichannel MSK
for the MSAT program, and if you scale the results down to 40 baud, they show
that the BER would deteriorate very rapidly with spacings less than about 65
Hz.  It appears to me that if you use 20 Hz spacing, you would have to drop
down to about 12 baud in order to avoid serious interference between adjacent
MSK channels, in which case your throughput is only 32*12 = 384 bps. I also
have some misgivings about the very low average power you'd have in a waveform
with 44 simultaneous tones.

On the other hand, I think your idea of using a Hilbert transformer to provide
AFC and allow analyzing more than N/2 frequency points with an N-point
transform is a dandy one.

Barry


#: 70699 S17/TAPR NNC/DSP
    25-Feb-88  13:12:05
Sb: #70621-Cheap Unit
Fm: Barry McLarnon VE3JF 71470,3651
To: Bob McGwier N4HY 74615,1366

I'll add my vote for 12-bit conversion, on the A/D at least... the extra 4 bits
are definitely worth the bucks.  Having only 8 bits to play with would lead to
MAJOR headaches downstream.

Yes, the C15 is pin-compatible with the 10 and has the identical instruction
set.  The only differences are the increased RAM, and the fact that the C15 can
have 4K of mask-programmed ROM on-chip.  I have been using the E15 EPROM
version of the chip with the TI cross-assembler (PC V3.0) with no problems
(except that the listing output is littered with warning messages every time
data memory in the 144-255 range is referenced, and there doesn't seem to be a
switch or directive to tell the assembler to wise up!).  Haven't tried the
Allen Ashley yet though.  My feeling is, if the price of the C15 is halfways
reasonable, go for it.

I would like to see the octal latches and any other glue parts needed for the
I/O ports included on the board.  Also, please keep in mind the need for a MUX
ahead of the A/D to allow at least a couple of analog inputs, and possibly some
room on the board to add another A/D/A if required.

I gather there's been a change in direction in this development.  Wasn't the
original idea to come up with a 320C25 design, possibly using the TLC32040
analog I/O parts?

Barry


#: 70717 S17/TAPR NNC/DSP
    25-Feb-88  20:01:15
Sb: Motorola DSP Operation
Fm: Bob McGwier N4HY 74615,1366
To: ALL

Steve Sagerian, KA0YRE of Motorola (at least partially responsible for
the 68000 board network controller that CAPRA has done, I think he built it)
has really come through for the DSP project.  He arranged for the DSP
operations branch of Motorola to come up with two 56001 EXP kits.  This kit
comes with bare boards, boot ROMS ( a debugger,monitor), PAL's, several
manuals.  Just to get things rolling in a hurry they decided to be very
generous and throw in two DSP56001 chips.  This board has a 20.48 Mhz clock
and processes 10.25 million instruction per second.  Using the architecture
to its fullest one could do a 1024 pt fourier transform in 3.48 ms.  Steve
and I will be building these two units up.  We may expect further support
from Motorola as they get applications back from us.  We wish to thank
Motorola, Inc. for their generous support.
Bob
N4HY

#: 70720 S5/Amateur Satellites
    25-Feb-88  20:08:26
Sb: #70698-DSP and WEFAX I
Fm: Bob McGwier N4HY 74615,1366
To: Barry McLarnon VE3JF 71470,3651 (X)

Why, no one ever spells mine correctly ;-)  Sorry!
73
Bob


*** More ***

#: 70721 S17/TAPR NNC/DSP
    25-Feb-88  20:08:35
Sb: #70699-Cheap Unit
Fm: Bob McGwier N4HY 74615,1366
To: Barry McLarnon VE3JF 71470,3651 (X)

That is still the goal.  We want (a poor analogy but the best I could think
of) a TNC-1 followed by a TNC-2 if you catch my meaning.  All of us here
want the TMS320C25 board.  We believe that we can put a DSP widget into the
hands of the bleeding manufacturers who are looking for a new toy to sell to
the hams who are looking for a new toy thereby enriching our toy design
coffers so that we can start the cycle all over again.  Did that make sense
;-)?
Bob

#: 70788 S17/TAPR NNC/DSP
    26-Feb-88  23:51:51
Sb: #70697-#M-ary FSK Modems
Fm: Bob McGwier N4HY 74615,1366
To: Barry McLarnon VE3JF 71470,3651 (X)

The Hilbert transformer is very do-able.  In fact, look at FIR.DOC in
FIR.arc and you will see that the fir.exe can generate FIR Hilbert
transformers that are as flat as West Texas over the design region and turn
off at with a damn good shape factor outside it.  I have used them to make
modems (the 6-th Annual Hoopla).  I don't know what Tom means by MSK.  What
I thought he had in mind was on-off in those tones. In which case at a few
tens of baud, we shouldn't have trouble with 20 Khz spacing.  The average
power being 4 watts per tone when all are rolling in a 100 watt transmitter
is somewhat bothersome but don't forget the nearly 10 dB gain from coherent
averaging in each cell with the "matched filter" of the complex sinusoid at
the center of the bin.  10 log(64) or 10 log(128) is the gain right? and
throw away a few dB because it doesn't arrive at the same time and you have
to await the phasors to all sum and settle.   Did any of this make sense?
Bob


*** There is a reply:   70914

*** More ***

#: 70914 S17/TAPR NNC/DSP
    28-Feb-88  16:38:16
Sb: #70788-M-ary FSK Modems
Fm: Tom Clark W3IWI 71260,3640
To: Bob McGwier N4HY 74615,1366

My use of the words MSK was by analogy -- MSK uses 2-ary shift of
50% of the baud rate (i.e. 600 Hz for 1200 baud, which is possible
since the highest fundamental freq in 1200 baud data is 600 Hz). By
extension, 32-ary 'on/off' binary channels should be possible with
an inter-channel spacing of 1/2 the baud rate. This should correspond
to the independent channel spacing of an FFT comb detector. But if
more interchannel spacing is needed, then so be it!
A further analogy: I believe that the Telebit Trailblazer modem crams
512 channels at 4 baud into a voice-grade telephone line. They certainly
don't have any more than 2 kHz to work with (which would correspond
to a 4 Hz channel spacing). I gather that the way they get 19,200
bits/sec into this is by encoding amplitude levels on each of the
512 data channels.

#: 70913 S17/TAPR NNC/DSP
    28-Feb-88  16:37:54
Sb: #70656-comments 1 of 3
Fm: Tom Clark W3IWI 71260,3640
To: Lyle Johnson, WA7GXD 76246,565

I am trying to get a 320C15 for testing in the Delanco Spry board
just to make sure there are no GOTCHAs.
.
Re A/D jitter that Bob/Phil reported: The A/D converter is a dual-slope
device. Its conversion time does depend on the voltage. But that
is why the D/S board (and the new widget we are discussing) have
a sample and hold gate. It is poor form to simply tell the A/D converter
to make a conversion when you want the data. The proper way is to
have the timer run the S/H and issue a CPU interrupt. Then CPU reads
the data after the conversion is done.

#: 70970 S17/TAPR NNC/DSP
    29-Feb-88  15:16:09
Sb: #70721-Cheap Unit
Fm: Barry McLarnon VE3JF 71470,3651
To: Bob McGwier N4HY 74615,1366

That's what I figured... I have no problem with that approach, it just came as
a bit of a surprise since I thought the C25 design was quite far along already.


#: 70971 S17/TAPR NNC/DSP
    29-Feb-88  15:16:52
Sb: #70788-M-ary FSK Modems
Fm: Barry McLarnon VE3JF 71470,3651
To: Bob McGwier N4HY 74615,1366

Bob and Tom:

Yep, DSP Hilbert xfmrs sure beat the analog variety.

Re Tom's clarification of his scheme: I can't accept his "leap of faith" which
says you can space multiple on/off channels by 1/2 the baud rate.  In fact, the
minimum spacing at which the signals are orthogonal is EQUAL to the baud rate
(just as in the example given of the Telebit modem), so you would be doing
32*20 baud = 640 bps with 20 Hz spacing.  If you did without the FEC, you could
use 30 tones for data to give you 600 bps, and use the other two tones for
pilot tone/clock recovery.  Lotsa possibilities... but this approach is really
FDM OOK rather than m-ary FSK if you want to label it.

First off, we need to establish our goal: do we want a better 300 baud
mousetrap for HF packet (to use your analogy, a TNC-1), or are we shooting for
higher speed right off the bat?  My feeling is that we should make sure we've
got walking down pat before we try to run!  :-)

Barry

P.S. Did you catch the message from Rinaldo (#70643)?  You might have missed it
since it has no UID in the address.  Would you trust someone with such a crass
attitude towards love?  ;-)


#: 70986 S5/Amateur Satellites
    01-Mar-88  00:02:11
Sb: #70887-Motorola DSP chip
Fm: Bob McGwier N4HY 74615,1366
To: Alex Schneider 71340,27

This board is a development system and not a finished product.  It is an aid
to help code developers get their products into the field (and for Motorola
to sell more chips).  It has NO A/D and D/A so we will have to add that.
It is a 10 MIPS processor with two 52 bit accumulators and the ability to
easily handle 48 bit floating point numbers.  Fixed point is the bane of
good DSP code.  Even though it is faster, it takes a great deal more care to
get completely right.
Bob



#: 70991 S17/TAPR NNC/DSP
    01-Mar-88  00:26:05
Sb: #70971-M-ary FSK Modems
Fm: Bob McGwier N4HY 74615,1366
To: Barry McLarnon VE3JF 71470,3651 (X)

Agree that it is FDM OOK since the tones are on off and not shifting amongst
1 of n possibilities.  I missed the Rinaldo note.  I will probably not go
back and get it so please summarize.  I am curious given your statement.
On the baud rate and channel spacing, I think your numbers are more nearly
correct.  I already have coded a demod-remod that runs the pants off the
the stock tnc-2 implementation with no MF-10 and sans Eric Gustafson's mods
to the PLL demod.  It works as follows:
(1) Run two FIR over the same 33 contiguous data samples
    FIR 1 is centered at 1375 Hz and is 650 Hz between cutoff points
    FIR 2 is centered at 2025 Hz and is likewise 650 Hz between cutoff
    points.  Take the absolute value of the output of each filter and
    and compute HIGH-LOW  abs(HIGH filter output) - abs(LOW F.O.) and store.
(2) Take 21 consecutive HIGH-LOW's and FIR filter with L.P.F. with cutoff at
    data rate. This gives bandlimited bits.  Hard limiting and remoding here
    even beats the 2211, but lets eek every dB.
(3) Take the absolute value of (2) and run an exponential decay recursive
    filter to get the DC.  Subtract the value of the DC from the output of
    (2).  This gives a "tone" at the data rate (clock).  PLL  (1-st order
    frequency fixed at 1200 Hz.  Phase error = output of Sub DC in sentence
    above times   - sin(clock phase).  When the phase crosses two pi, the
    eye is maximally open (thus the reason for - sin rather than sin which
    tracks the crossings).  Take this sign and make 1200 or 2200 until the
    next crossing of 2 PI.  Now this REALLY runs a ring around the 2211 in
    all my tnc's and with all my radios.  300 bps stuff on HF is easier than
    this but we need to flatten the response of the filters more to remove
    bias.  Is this what you did or did you do a PLL demod?  In my attempts
    do a PLL demod, it worked but the bit error rate was an order of mag
    larger, comments?   Bob

#: 70995 S17/TAPR NNC/DSP
    01-Mar-88  00:55:39
Sb: #70913-comments 1 of 3
Fm: Bob McGwier N4HY 74615,1366
To: Tom Clark W3IWI 71260,3640

According to David's manual and the way we have the jumpers set on the
board. I believe we are doing exactly what you say. To wit:
Page 8.1 in the manual
The analog to digital conversion is a two step process from the software or
high level point of view.  The ADC resides at Port 3 in the 10's I/O space.
First step 1, the ADC must be told to begin the conversion.  This is done by
bringing the R/C pin (C for convert) low.  This strobe may be generated in
one of two ways.
Method 1  (paraphrasing, do it in software and do what you say is bad form)
Method 2 (the way we have them set up), Strobe generated by the output of
Timer 2.  JA in lower position.  J5 in ext position.
Method 2 is clearly the simpler method as it requires less software and does
not tie up the int pin.  etc etc

Step 2:  <<<  The ADC announces that it has completed the conversion by
pulling the STS pin and thus the BIO pin on the 10 low.  This is the BIO
interrupt.  Upon its reception the 10 branches to a section of code where it
now input the results, eg.
    in 6,3

This Method 2 and Step two is exactly how we are doing it now.  We let the
A/D set the BIO interrupt pin low and then we go and get it.  This means
that, given that the code can run between samples (few enough instructions)
that all timing is determined by the Timer 2 which I hope is stable.
Possibly we should look at it.  In my code, both the I and the O are done in
stretches of code immediately after the A/D is done since this code has a
fixed number of instructions and thus the timing is still tied to the timer.
I don't know what to make of this.  Phil has other problems with his board
that I haven't seen yet.  Bob

#: 71041 S17/TAPR NNC/DSP
    01-Mar-88  23:09:05
Sb: #70721-#Cheap Unit
Fm: Harold Price, NK6K 71635,1174
To: Bob McGwier N4HY 74615,1366 (X)

I'm not sure that's the best way to put it.  The '15 design is meant to be an
inexpensive entry-level device.  We'd like everyone to be able to have one to
allow some more interesting modulation schemes at low (<=9600) baud rates to
come into general-purpose use.  This means HF and general-access satellites,
and gee-wiz things like wefax. The '25 design is a high-end replacement for the
delanco board, to do things it can't, high speed, complex encoding, etc.  It
isn't meant to be a general consumer-level device, it's for experimenters and
high-end low-numbers applications. The '15 design would hopefully be less than
$200, the '25 design could be > $400.

*** There is a reply:   71054

*** More ***

#: 71054 S17/TAPR NNC/DSP
    02-Mar-88  00:39:29
Sb: #71041-Cheap Unit
Fm: Bob McGwier N4HY 74615,1366
To: Harold Price, NK6K 71635,1174

I agree that is the best way to put it to John Q. Ham or Joe and Mary HT. I
should be more careful in stating what my own goals are seperate from what
the project's goals are in general.  I agree that not everyone wants, needs,
can use, or can afford the C25 or 56000 or DSP-32C or whatever board.  I do
want to continue to remind everyone that this board we are talking about
doing has no commercial counterpart.  The nearest thing is a PC drop in
board and it costs $4000.  At $500 it is dirt cheap and 19.2 KB will run all
day long in it, SSTV, FAX, Spectrum analyzer, Voice (ADPCM) and ten pieces
of test equipment.  All that special hardware would cost more than the $4000
the commercial guys are getting and that is the reason they are making money
since it can be reprogrammed to do different code development.
Your point is well taken.
Bob

#: 71045 S17/TAPR NNC/DSP
    01-Mar-88  23:27:37
Sb: tcp/dsp from arpanet
Fm: Harold Price, NK6K 71635,1174
To: all

As part of TAPR's information charter, we've trying an experimental service
which makes the TCP and DSP conversations on uucp and arpanet connected systems
available to people on compuserve.  If you find this service useful, let me
know, if nobody needs it we'll expend the resources on something else.  Bdale
Garbee is gatewaying the info to me, I'm moving it here.  I'll be putting it in
DL17 because 1) uploading to DL is free, 2) people who already get it elsewhere
don't have to see it again, 3) there is no length limit.  I'll put a batch in
DL17 every few days as they come in.  For lack of anything better, I'll call
the files ymmdd.arp hp

#: 71079 S17/TAPR NNC/DSP
    02-Mar-88  08:31:21
Sb: #70991-M-ary FSK Modems
Fm: Barry McLarnon VE3JF 71470,3651
To: Bob McGwier N4HY 74615,1366

My approach to the 300 bps modem is very traditional, starting with a GOOD
bandpass filter (8th-order Butterworth, BW about 360 Hz), and using three 2nd-
order filters (mark, space, and data rate) in the discriminator.  All the
filters are IIR.  Clock recovery is not a PLL either... clock is regenerated
by a bandpass filter centered at 300 Hz into which pulses generated by zero-
crossings in the data are fired.  I should look at your first-order PLL, as it
is probably more compact, codewise.  I haven't really done any error rate
tests as yet.  I am really hurting for lack of a decent tape recorder to store
off-the-air signals... I commandeered the family cassette deck (we mostly play
CD's these days anyway, I explained), but it just doesn't cut the mustard.  It
was one of the best decks of its day, but that was 14 years ago and I think it
just has too many miles on it. <sigh>  I hate spending my hard-earned on
mundane stuff like tape recorders!


#: 71095 S5/Amateur Satellites
    02-Mar-88  11:22:13
Sb: #70986-#Motorola DSP chip
Fm: Bill Coleman AA4LR 76067,2327
To: Bob McGwier N4HY 74615,1366 (X)

Bob,

Got four of the TLC32041 chips for you! Will get them in the mail as soon as I
find a box to put them in. Each chip comes with complete literature.

If the 56K has a PCM-highway style serial port, there shouldn't be any problem
interfacing to the 32041.

    73, Bill

*** There is a reply:   71128

*** More ***

#: 71128 S5/Amateur Satellites
    02-Mar-88  21:23:02
Sb: #71095-Motorola DSP chip
Fm: Bob McGwier N4HY 74615,1366
To: Bill Coleman AA4LR 76067,2327

Wonderful Bill:
Please send them to
15 Cherry Brook Lane
East Windsor, NJ 08520
73
Bob McGwier
N4HY

#: 71126 S7/TAPR - DRNET
    02-Mar-88  21:22:47
Sb: #70643-DSP modems
Fm: Bob McGwier N4HY 74615,1366
To: Bob Schetgen 70007,3373

On Rinaldo Har!
On demod:
Okay on your setup there.  It is time for me to mail out new diskettes
anyway as I have just about a 1/2 diskette of archives with the wefax.arc
and the bel202.arc and the HF.arc that we should all look at and if you
want to I will add your stuff to the mailing so that AMSAT pays the ticket
on the mailing.  Tom and I have been eating the costs associated with this
so far but that is getting out of hand (and I am approaching the limit for
donations anyway without a complete and total accounting to uncle).  I will
take this weekend and hack together the 16 tone OOK FFT demod and see how
that works.  The nice thing I have is the ability to generate these signals
and run them through shit adders at work: group delay, amplitude foul ups,
and both colored and white noise (for those of you unfamiliar with signal
literature read NOTHING into the titles of the noise types).  We have a
telephone line simulator and one of the setups is called "modem killer".
It is. ;-)
Re: recorder
Yeah, I went out paid $$$ for a recorder and it has been worth it.

I will do my darndest to get this code to you on Monday.  The folks in
Brazil, Italy, Germany, and Japan are sending me nastygrammes wondering when
I am going to send the WEFAX-APT and BEL202 stuff.  I have clearly
underestimated the distribution strength of CIS and telemail.
Bob

*** More ***

#: 71127 S7/TAPR - DRNET
    02-Mar-88  21:22:54
Sb: #70643-DSP modems
Fm: Bob McGwier N4HY 74615,1366
To: Bob Schetgen 70007,3373

Okay Paul:
We will try to make you love us before the next digital networking
conference where we get to try to make folks think these strange hackers are
smarter than they look.
Bob



#: 71133 S17/TAPR NNC/DSP
    02-Mar-88  21:52:39
Sb: cheap board
Fm: Lyle Johnson, WA7GXD 76246,565
To: dspers

Tomorrow Chuck comes over and we will review the TMS3201x DSP design.  It is
rather ugly at present due to the need to add (1) a clock wait-state generator
and (2) an "unbalanced" address decoder in ordere to (3) glue an 82C54
programable timer onto the 3201x.  There are lots of ICs in this thing (around
40!) and I don't like it.  We may go to a PAL or some such nasty (expensive or
slow, your choice) thing to reduce the chip count and clutter.

I am looking closely at the costs associated with trading up from the 8 bit
analog I/O system ($6) to a 12 bit D/A ($10) and 13-bit A/D ($15).  In addition
to the $19 increase, we lose speed (32 uSec A/D instead of 2 uSec). It may
prove "cleaner" in terms of chip count and circuit complexity to add a simple
controller (z80/sio/ctc/EPROM/RAM) to this thing, derive the timing frm the z80
side instead of gluing the 8254 to the 3201x, and provide a simple loader
program to load software from a serial port to the unit.

I am planning on a 4-layer board for the DSP part of this.  The z80, 2206/2211
hookup, power supply, etc., can be placed on a 2-sided board to cut costs.

Anyway, we should have schematics ready to go to W3IWI, N4HY, WB6YMH, KV7B and
N7CL in a couple days for comment.

Lyle

#: 71141 S17/TAPR NNC/DSP
    02-Mar-88  23:22:24
Sb: #71054-Cheap Unit
Fm: Harold Price, NK6K 71635,1174
To: Bob McGwier N4HY 74615,1366

I agree with you completely.  I just wanted it to be clear to Joe and Mary HT
(who may someday see this stuff), that we weren't planning a
planned-obsolesence thing, or a bait-and-switch, or a fire sale, or any of the
nasty things some folks accused us of when the TNC-2 was announced.  HP,
Director, Department of Creative Paranioa.

#: 71280 S17/TAPR NNC/DSP
    05-Mar-88  11:39:14
Sb: #71045-#tcp/dsp from arpanet
Fm: Harold Price, NK6K 71635,1174
To: Harold Price, NK6K 71635,1174 (X)

Any thoughts on puting the TCP and the DSP uucp stuff in separate files?  For
now, they'll be in the same file. hp

*** There is a reply:   71303

*** More ***

#: 71303 S17/TAPR NNC/DSP
    05-Mar-88  14:43:30
Sb: #71280-tcp/dsp from arpanet
Fm: Bob McGwier N4HY 74615,1366
To: Harold Price, NK6K 71635,1174 (X)

My only question is are you sending stuff in the other direction as well?
If not I was going to upload some stuff that seemed pertinent to the folks
on the dsp-group.  If the path is bi directional then I won't bother.
Bob
(P.S. thanks)

#: 71291 S17/TAPR NNC/DSP
    05-Mar-88  14:16:01
Sb: TMS320C25 vs DSP56001
Fm: Bob McGwier N4HY 74615,1366
To: ALL

I want to compare the two processors.  Given that they are both in the same
price class, both have similar development tools available to us, we should
choose one or the other on its technical merits.

DSP56001:

Speed - 10.25 MIPS.  1024 point complex fast Fourier transform is executed
in 3.39 msec (if you populate with 45 nsec RAM).
Precision - The data paths are 24 bits wide giving 144 dB dynamic range in
the data storage with TWO accumulators 56 bits wide that give 336 dB dynamic
range for intermediate range.
Parallelism - The data ALU, address arithmetic units, and program
controller operate <IN PARALLEL> so that an instruction prefetch, a 24X24
bit multiplication, a 56 bit addition, two data moves, and two address
pointer updates using one of three types of arithmetic (linear, modulo, or
reverse carry) can be executed in a single instruction cycle.  This would
allow a four coefficient infinite impulse response (IIR) filter section
(like a Butterworth or Tschebychev) to be compute in four cycles.  I checked
and this is the theoretical minimum for a one multiplier microprocessor.
Integration - In addition to the <THREE> <INDEPENDENT> execution units, the
DSP56001 has six on chip memories, three on chip MCU style peripherals
(serial CI, Synchronous Serial I, and Host interface built in), a clock
generator and SEVEN buses (three address and four data).
Pipelining - Three three-stage instruction pipeline is essentially invisible
to the programmer (other than one well documented bug) allowing easy code
development in assembler and Motorola has its own K&R C compiler which I am
a beta test site for.  It has bugs in it but that is what beta test is all
about.
Instructions- 62 instructions.  The orthogonal syntax supports control



#: 71292 S17/TAPR NNC/DSP
    05-Mar-88  14:16:16
Sb: TMS320C25 vs DSP56001
Fm: Bob McGwier N4HY 74615,1366
To: ALL

Part II
of the parallel execution units.  The zero overhead Do and REP (repeat)
instruction instruction make writing straight line code unnecessary on this
processor, making programs smaller and hurting readability only slightly.
Last but not least, the DSP56001 comes with, built into on board ROM
(1) a boot strap loader for loading the program RAM fro byte wide memory
mapped ROM <OR> via the host interface (remember its built in).  The data X
and Y memorys are seperate buses and the on board ROMS for low memory in
these buses come with preloaded with both a Mu-law and A-law to linear
expansion tables and a FULL, FOUR QUADRANT sine wave table.  I reiterate
these are built in. It comes complete with bus arbitration signal pins built
in so that shared memory with another host is trivial.  Motorola has donated
equipment necessary to build up two units that do exactly what we wanted the
TMS320C25 fancy unit to do.  That is, they supplied the CPU board with
positions for up to 480K of memory which may be split in logical ways
between X,Y, and program memory.  It comes with a bus extender to which may
be added a host board and we need build up our own A/D D/A board which we
said we should leave to customization by end users for particular task
through game cartridges etc.  The host interface could be PC, SCSI, etc.

TMS320C25
Speed- Likewise a 10 MIPS processor (100 nsec processor) (320C1X-25 is a 6.25
MIPS processor since it has 160 ns clock).
Precision - The data path (forget paths) is 16 bits wide giving 96 dB dynamic
range and the single accumulator is 32 bits wide affording 192 dB dynamic
range.  This is the same as the 320C1X and I have run into trouble several
times because of the size of this accumulator.  Remember, a full 16X16
multiply is 32 bits.  Add a couple or three and you are in trouble.



#: 71293 S17/TAPR NNC/DSP
    05-Mar-88  14:16:31
Sb: TI vs Motorola III
Fm: Bob McGwier N4HY 74615,1366
To: ALL

Parallelism of the type most useful in DSP chips was pioneered by TI.  They
have not carried it as far as Motorola.  There is not the very advantageous
parallel data buses.  Let me explain this more carefully.  Filters in software
are weighted averages of consecutive samples and possibly previous filter
outputs.  If none of the latter it is called a finite impulse response filter.
If the latter is present it is an IIR or infinite impulse response filter.  So
what do we need to do a filter?  We need a string of consecutive numbers we
call data (samples or previous filter outputs) and we need the weights to apply
to each of these data points.  I hope you are able to envision the savings in
time of being able to store the filter coefficients in one data leg and the
data points in another data leg and fetch them in parallel while moving data in
either of the legs in parallel. The same IIR or FIR, if we ran both at exactly
ten MIPS takes a fraction of the time on the Motorola chip.  In my modems and
wefax demods etc. 95% of all operations are either FIR's (arm filters in the
modems and wefax demods) and IIR's (loop filters in the PLL's).  What did Barry
and I talk about when we described our FSK work?  He talked about his
Butterworth's (IIR) and I talked about my PLL and High LOW filters (IIR and FIR
respectively). Anything you save here is bits per second in the bank.
Integration - Again the TI fairs less well.  It has two execution units and
does prefetch three instructions and a data value.  It does have a serial and
synchronous CI but does NOT have a Host interface.  The one programming
advantage this chip has over the Motorola is that it comes with seperate
auxiliary registers for doing loops AND a 15 deep stack.  The Motorola uses the
15 deep stack to nest loops limiting both the number of loops and/or amount of
hardware stack available for other tasks. Invisible pipeline- It also has an
almost invisible pipeline but highly variable instruction completion time due
to the "Harvard" architecture and the necessity of going to great lengths to
move stuff from program memory to


#: 71294 S17/TAPR NNC/DSP
    05-Mar-88  14:16:41
Sb: #TI vs Motorola IV
Fm: Bob McGwier N4HY 74615,1366
To: ALL


I haven't had a chance to go over the ATT DSP-32C hardware manual yet but I can
tell you that ATT prices there stuff very high so that external buyers can't
compete with their own internal manufacturing of end user products. Last I
heard, the DSP-32C was $325.  We can get Motorola DSP56001's in quantity in the
$120 range (we can get them in small quantities in that range from Steve
Sagerian).  I am sure that I have left some stuff out that might make the
TMS320 look better and I hope someone will upload that or when I get to Tom's
next weekend I will finish the programming model for it if I find I have left
something significant.  I guess that I have left little doubt as to which I
would rather program for in the high end engine unless someone or something
comes along to change my mind. Bob N4HY



*** There is a reply:   71353

*** More ***

#: 71353 S17/TAPR NNC/DSP
    06-Mar-88  11:47:31
Sb: #71294-#TI vs Motorola IV
Fm: Lyle Johnson, WA7GXD 76246,565
To: Bob McGwier N4HY 74615,1366 (X)

Bob,

Most recent pricing I have sseen uts the TI cip at about $120.  I thought the
Motorola chip was around $400??

*** There is a reply:   71367

*** More ***

#: 71367 S17/TAPR NNC/DSP
    06-Mar-88  14:41:40
Sb: #71353-#TI vs Motorola IV
Fm: Bob McGwier N4HY 74615,1366
To: Lyle Johnson, WA7GXD 76246,565 (X)

Not so according to Steve, we can get the chip for in the $125 range.  I hope
that he isn't wrong as it is a much better chip but $400 would put it out of
our range for even the follow on engine.  Let me leave no doubts, 10 MIPS on
the TMS320C25 doesn't compare to the 10 MIPS on the Motorola chip because of
the way you can handle FIR's and IIR's which are the heart of the DSP
operations.  More calculations given the computational complexity on the
parallelism on each engine, I think that with carefully done code on the
Motorola chip, I could do a Costas loop demod of 19.2 KB GMSK.  This is a
coherent demod.  It what seems to be an incongruity, it would not be
possible to do the noncoherent demod because of the longer filters and
greater quantity of them needed to do the job right.  To wit:  Notice in the
filter modems Barry and I described, we needed to do bandpass filters before
the bit rate filters. In the Costas loop demod, these are both combined into
the arm filters reducing the number of filters needed by two (time).
I need 640-650 ticks per bit to do the demod on a 19.2 KB modem, and I
have 714 ticks on the 56000.  On the TMS320C25, I would need just about a
1000 ticks and I don't have them available.  We could go with a telebit
approach, but I understand they have applied for a patent.  FDM OOK even
with amplitude encoding is not unique so we will have to be careful and see
just exactly what it is they have patented.  With the Z80 chip on board,
these numbers may change somewhat, but we will have to see.  The big picture
is I can call the data and the filter coefficients into the proper places
for multiplication, accumulate previous results and shift the data down in
memory in one tick on the Motorola.  It takes two ticks to do the same thing
on the TMS320C25.  This is a direct 50% savings in number of operations in
almost any DSP implementation.  Hey: I just thought I would get everyones
juices flowing with a little discussion ;-).
Bob

*** There is a reply:   71378

*** More ***

#: 71378 S17/TAPR NNC/DSP
    06-Mar-88  17:29:38
Sb: #71367-TI vs Motorola IV
Fm: Lyle Johnson, WA7GXD 76246,565
To: Bob McGwier N4HY 74615,1366

Bob,

If the Motorola chip really is within $50 or so of the TI chip, it may well be
a beter way to go. The TI chip requires 25 nSec memory chips to run at 40 MHz
(10 MIPS), and 45 nSec memory is about 1/2 the price of 25 nSec, especially in
the higher densities.

I am aware of another option in the DSP procesor wars - the Analog Devices
ADSP-2100 DSP uP, an 8 MIPS machine.  I have ordered information on it, as well
as the Motorola chip, for further investigation.  Fujitsu and NEC are also in
this game, and I have *LOTS* of info on the Fujitsu stuff.  I will check it
out.

My experience so far has been non-technical with DSP.  I have found excellent
local support from TI, good support from Fujistsu, and wishy-washy support from
Motorola.  In the world of regular microprocessors, we have had TERRIBLE
support from Motorola.  In general, our experience at MMS is that Motorola
makes excellent silicon (but don't believe anything until you physically have
PRODUCTION silicon in your hands), terrible emulators and awkward/so-so
software tools.  Maybe things have changed (we also have one in hand and one on
order 68020-based Sun workstations), but after 9 years, Motorola STILL doesn't
make a useable emulator for the 6809 and the CMOS version of the chip, still in
the current data book, has been published as a data sheet for several YEARS and
still no silicon (we buy the Hitachi CMOS units now...). Finally, witness the
lack of V 4.0 TNC 1 software becuase GOOD software tools are either VERY
EXPENSIVE or IMPOSSIBLE to locate for the 6809.

On the other hand, if the DSP56001 chips are real, and you have in your hand
tools that actually work, it is very viable.  In spite of our problems at MMS
(we wrote our own compilers!) we have settled on the 6809 as our controller
engine.  So, don't read the above as a rejection of Motorola products, just as
a caution about chips in hand versus data sheets in the bush.

Lyle




#: 71295 S17/TAPR NNC/DSP
    05-Mar-88  14:16:54
Sb: comments 1 of 3
Fm: Bob McGwier N4HY 74615,1366
To: Lyle Johnson, WA7GXD 76246,565 (X)

I have thought about all this some more.  To do a modem one needs three
states coming from the phase detector if you are doing a coherent scheme or
an FM detector.  +1, 0, and -1.  If your filters and decisions are done
carefully after that you suffer a Van Vleck loss (Tom pounce on me if I use
the wrong name) of ~2 dB.  We are not going to build the weak signal device
with this unit I don't think.  I believe that we are shooting towards the
much larger market of large signal work on HF and the moderate signal from
something like UOSAT-C and PACSAT, and the large signal work on VHF packet.
Thirty samples thousand samples per second is big enough to do the FSK demod
but just is.  I would be happier with 40000/ sec so that I get more than
four samples per bit.  This makes lining up the clock to the maximum opening
of the eye for each bit a software job and I don't have to control the
hardware sampling from the software.  This is the primary reason for wanting
to do oversampling in the PLL that recovers the clock.  In my BEL202 modem
I am sampling at 8 samples per baud and the jitter on the clock at moderate
SNR's is zilch.  For unit 1, until someone tells me why we need 12 bits (72
dB of dynamic range) for this unit, I vote for the faster 8 bit unit.  It
does mean it is encumbent upon us to do the tough job of not wasting what
there is in noise from unwanted sources.  "A birdy on 145.01" will not do.
48 dB is enough dynamic range and I don't think we could use more below 15
Mhz on HF if we had it.
Bob

#: 71302 S17/TAPR NNC/DSP
    05-Mar-88  14:43:21
Sb: #71133-cheap board
Fm: Bob McGwier N4HY 74615,1366
To: Lyle Johnson, WA7GXD 76246,565 (X)

Great.  I uploaded comments on the 8 bit before I saw this message.  I agree
that given the targeted audience for this unit, 8 bits <IN A CLEAN DESIGN>
is enough for our purposes.  I also will attest to their not being 50 dB of
dynamic range in that coming out of either the IC's or TS's on HF.  The
problems come in DC offset, not total dynamic range.  I still believe that
for the tasks at hand, 8 bits is enough.  40 chips is really pretty big and
I like the cheap host idea.  Mike Chepponis is a Z80 king and is in the DSP
project and sounds like the perfect person to ask.  I don't think we will have
problems with Mike as we have had with other Z80 hacks.  He took Phil out to
dinner with Richard Stallman and surely most of us have read the GNU
manifesto.
Bob