On 23 Jun, 20:18, tum_ <atoumantsev_s...@mail.ru> wrote:
I've googled quite a lot today on the subj., basically, to find some
info on the following: how much CPU load (for the CPU mentioned in the
original post) shall I expect from a well-optimized implementation of
a software V.34 modem?
Unfortunately, all I could find so far were either assumptions and
guesses or some real (well, advertised) figures for different brands
of DSPs, ranging from 7 to 30ish MIPS. I'm not familiar with DSPs, I
can only guess that they are probably more suitable for the job in
question than general purpose ARMs but then again - DSPs are probably
very different in their performance, so its hard to make any
conclusions really...
Could someone point me to some source of relevant info for ARM9,
please. Or share his own experience.

Andres mentioned "gaoresearch or netbricks", I've been to their sites
today and they sound like real pros in this niche, but I still
couldn't find any figures for ARMs there.

trying to crosspost this to comp.arch.embedded, let's see if I succeed
this time.

On Jun 25, 8:27 pm, tum_ <atoumantsev_s...@mail.ru> wrote:
crosspost successful, thanks God.

ARM9E based cores such as ARM926EJ-S and ARM946E-S have DSP instructions
that should speed up your algorithm. If you have a high-end DSP
application then yes, you should use a screamingly fast DSP. I suspect
V.34 is small to medium but I have no V.34 experience only ADSL.

Re MIPS - ARM9 was designed to run at around 200MHz (newer silicon will
be faster so check your chip's datasheet) and ARM9 is approx 1.1
Dhrystone MIPS per MHz.

You can run the software on an simulator to find out roughly how fast it
is. If you have the older ADS tools it's called ARMulator and under RVDS
it's called RVISS (RealView Instruction Set Simulator).

IMO "how fast does it run" seems a reasonable question if you're going
to spend a few thousand on software...

Cheers,
Rich

On Jun 27, 11:57 am, Rich C <no_s...@ntlworld.com> wrote:
Thanks for the reply, Rich.

As I said in the original post, we have an uC based on arm922t running
on 180Mhz. It doesn't have DSP extensions and changing the CPU is not
an option.
The V.34 (from all those bits & pieces that I've managed to google so
far) appears to be quite a heavy thing for such a CPU. Although
definitely doable, it should be expected to eat up a very large chunk
of the CPU time, requires very serious approach to optimization and
still might render the whole system unusable (I mean, we need to be
able to do other things while the modem is working).
It also has quite a lot of configurable options which might greatly
impact the performance and hence there's a chance to get a usable
system by sacrificing certain non-mandatory options (trellis coding,
pre-coding, etc.).

The only explicit figures I've found are (from http://www.mesi.net/MESiWeb/mips.htm):
71.1 MIPS for V.34 (16-state TCM (trellis-coded modulation)) given for
ARM9e. They claim that the figures are given for highly optimized C/
Assembly mix. Considering that ARM9e is (presumably) better for the
job than ARM9, the things look not that good for us.


Yep, I know this. I also know that MIPS is a very useless thing for
comparisons/estimates like this but... that's all I've got at the
moment.

I don't have access to this but the guys in a different department are
apparently doing such tests and from what I heard they're not quite
happy. That's why I started this thread in the first place.
And the softmodem is developed by a third-party (which I can't name)
and looks like we have no access to the source code ;-)
I'm just digging this out of curiosity, trying to make my own
judgement - are the third-party guys pulling our leg with this one or
what?

you bet...

thanks again for the comments.

Can you add extra Hardware to your main CPU then?
An AT91SAM9261 has 160 kB of SRAM, ARM926EJS @ 200 MHz and
the code can be uploaded using the UART or USB during boot to the SRAM.
No need for a separate flash...




--
Best Regards,
Ulf Samuelsson
This is intended to be my personal opinion which may,
or may not be shared by my employer Atmel Nordic AB

On Jun 29, 7:51 am, "Ulf Samuelsson" <u...@a-t-m-e-l.com> wrote:
No, this would require a serious redesign of the unit. Besides, the
cost is a bit too high (sub $10 in high volumes, that's all I could
find). Fortunately, there's an option to use a hardware modem if it
becomes absolutely clear that the software modem is not going to work.
A hard modem is adding about $5 to the unit cost.
On the other hand, there's one variant of the device where a hard
modem can not be added (without a redesign of the device).