🚀 go-pugleaf

Dear Sir/Madam,

A little while ago, I posted this message:
"Dear Friend

I have a challenging problem: designing a real-time image processing system.  
If using DSPs, such as Motorola DSP56000, the system requires about 
100 MIPS (say 100 million of multiplies and adds per second).
It needs several best DSPs on the market. The system has to be as small
as possible since it is used on an aircraft. Is there any better way to do it?
ASIC? PLD? How hard to use these techniques?
Thank you very much!!
Frank
gao@eecg.toronto.edu"

I received some interesting responses which are summarized here to entertain
those who are interested. I thank those who kindly responded to my posting.

Frank

Date:	Sat, 10 Oct 1992 00:03:41 -0400
From:	bittware@grebyn.com (Bittware)

Frank,

Since your e-mail looks like a Canada address, I'd suggest giving
bbd Electronics a call at 416-821-7800.  bbd is our rep for Canada,
as well as several other companies with products that may interest
you.

We make an ADSP-21020 based DSP board for the PC.  The ADSP-21020 
is currently the fasted DSP available at 75 MFLOPS and a 100 MFLOP
version is being qualified now and available in
small quantities.  There is also another company (Ixthos) that makes
a dual ADSP-21020 for the VME, and bbd is their rep as well.

Hope this helps!
Jim, bittware@grebyn.com


From:	hards4@evans.ee.adfa.oz.au (Brad Hards)

   From: gao@eecg.toronto.edu (Xiangyang Gao)
   Newsgroups: comp.dsp
   Date: 10 Oct 92 02:18:53 GMT

   I have a challenging problem: designing a real-time image processing system.
This is very vague - do you want to do just FFT, filtering, or something
fancy that won't map real well into DSP architecture?

   If using DSPs, such as Motorola DSP56000, the system requires about 
   100 MIPS (say 100 million of multiplies and adds per second).
   It needs several best DSPs on the market. The system has to be as small
   as possible since it is used on an aircraft. Is there any better way to do it?
A quad TMS320C40 board would be pretty small. You'll get 100 MFLOPS from
that with plenty left over. Exactly what form factor? What aircraft if you
can tell me? Where in the airframe? What operating conditions? Any
commercial board will probably need stiffening for aircraft use.

   ASIC? PLD? How hard to use these techniques? How much do the development
   tools cost? Or is there any good boards one market?
ASIC if you know exactly what the system has to do, and you have very good
tools, operators, and a shitload of cash. PLD won't make it. A Xilinx 4000
series FPGA is maybe 1/10 the logic capability at 1/4 the speed of even a
low end DSP, and twice the power requirement.

Look for boards with high end Analog Devices 21020 parts, or TMS320C3x and
TMS320C40 as host processors. The FAQ for comp.dsp has a huge range of
boards listed, including some tiny ones with single TMS320C31's which
will give about 33MFLOPS per board.

---
Brad Hards     EE4 - Australian Defence Force Academy      hards4@ee.adfa.oz.au
             "If it is stupid, and it works, it isn't stupid"
Disclaimer: If any of the above agrees with the opinions of ADFA or the RAAF,
                        it is purely coincidental. 

From:	mlelstv@specklec.mpifr-bonn.mpg.de (Michael van Elst)

In comp.dsp you write:
>I have a challenging problem: designing a real-time image processing system.  
>If using DSPs, such as Motorola DSP56000, the system requires about 
>100 MIPS (say 100 million of multiplies and adds per second).

Well, my system runs on 4 DSP56001's (up to 16) on a shared bus.

>It needs several best DSPs on the market. The system has to be as small
>as possible since it is used on an aircraft. Is there any better way to do it?

Depends on what you want to calculate and how flexible the system has to
be. A DSP sytem can be reprogrammed to arbitrary tasks as long as the
bandwidth of all subssystems is large enough. But you will of course get
higher performance with specialized chips.

>ASIC? PLD? How hard to use these techniques? How much do the development
>tools cost? Or is there any good boards one market?

ASIC development is usually quite expensive. Production cost for >1000 units
is of course cheap but ASICs are usually restricted to the single operation
they are built for. LCAs on the other side can be programmed on the fly
(the logic functions are based on RAM cells bootstrapped at powerup or reset)
but they are quite limited as each LCA has only some hundred logic functions.

Hope this helps,
-- 
Michael van Elst
UUCP:     universe!local-cluster!milky-way!sol!earth!uunet!unido!mpirbn!p554mve
Internet: p554mve@mpirbn.mpifr-bonn.mpg.de
                                "A potential Snark may lurk in every tree."