Microcontrollers for DSP

Ben Bradley <ben_nospam_bradley@frontiernet.net> wrote in
news:8n37n19ghgveoo07k5ob8g9m3vr876944k@4ax.com: 

Most DSPs today function quite well as microcontrollers. For example, the 
new Sharc 21369 has I2C, 2 SPI ports, two UARTs and full data bus, 
timers, etc.

OTOH, Most microcontrollers are very limited for DSP applications.  

The DSPs tend to consume more power, cost more, are often physically 
larger.

Processors like the Blackfin are kind of hybrids. They are small, consume 
minimal power and are fast.

I think you need to clearly examine what you want the device (or 
platform) to do, and then the choices will be easier to sort out.

If you are trying to teach DSP concepts with real hardware, I would avoid 
devices that have poor DSP performance. A dsPIC for example might be fine 
for a simple motor control application, but it is less powerful that the 
DSP chips I used 10-15 years ago. 

I like (and sell SHARC based boards) in part because they are easy to 
use. It is typically much easier to code an application when you have 
single cycle 32 bit floating point and fixed point instructions 
available. 

Of course, when you have a specific application in mind, you consider all 
sorts of additional factors than just the basic DSP processing.

-- 
Al Clark
Danville Signal Processing, Inc.
--------------------------------------------------------------------
Purveyors of Fine DSP Hardware and other Cool Stuff
Available at http://www.danvillesignal.com




> On Fri, 04 Nov 2005 21:07:52 GMT, Joerg
><notthisjoergsch@removethispacbell.net> wrote:
> 
>>Hello Ross,
>>
>>> Hi. I would like to know what sorts of microcontrollers are
>>> available so that I could design some "student projects" with DSP
>>> chips. ... 
>>
>>uC or DSP? Very different beasts.
> 
>    Well, yes, sort of. There's a bit of overlap, though it appears the
> OP doesn't see the distinction between a microcontroller and a DSP.
>    There are lots of microcontrollers with traditional microprocessor
> cores. These run quite slow for DSP tasks, as they have to go through
> a lot of operations to do a typical DSP calculation.
>    There are a few microcontrollers with MAC devices stuck onto them,
> such as the DSPic and the higher-end MSP430, which makes them
> substantially faster for DSP tasks, but it may still takes several
> cycles to load/read the MAC unit.
>    Most (all?) low-end DSP nowadays ARE microcontrollers with a
> "traditional" DSP for the processor core, and usually do single-cycle
> instruction execution, including MAC's, thus they are several times
> faster than the "microcontroller-with-MAC".
> 
>>
>>You'll have to invest the time and trudge through the mfg sites. TI, 
>>Microchip, Atmel, Analog Devices etc. There is usually a budgetary
>>price listed. When you have picked one out go to Digikey and see what
>>it costs in low quantities. In the UK Farnell might be easier.
>>
>>If you can live with 12bit audio there is plenty. In the DSP arena for
>>example the TMS320LF2401. Mostly you'd have to supply your own DAC
>>though. 
>>
>>If you need very low quantities do the search the other way around.
>>Look what you local hobby shops have and then delve into the data
>>sheets to see what fits the bill. Get something on a header board to
>>save you the really fine pitch solder job. If you don't want to solder
>>much at all look for a low cost evaluation board, often available from
>>manufacturers like TI and usually including the software suite.
>>
>>Regards, Joerg
>>
>>http://www.analogconsultants.com
> 
> 

Hello Al,

> OTOH, Most microcontrollers are very limited for DSP applications.  
> 

Just had a rude awakening there. As Ben mentioned there are some with 
integrated MAC hardware. Sweet, I thought. Then I found that the MSP430 
seems to need eight cycles for a multiply and 17 cycles for 
multiply-accumulate.


> The DSPs tend to consume more power, cost more, are often physically 
> larger.
> 

That was the other awakening. Had to scratch the TMS320LF2401 from my 
list because it slurps a whopping 50mA or more. Too much for running off 
AA cells.


> Processors like the Blackfin are kind of hybrids. They are small, consume 
> minimal power and are fast.
> 

If they had somewhat decent audio ADCs on there they would be much more 
useful for an embedded application. I mean, it doesn't have to be top 
notch hifi.


> I like (and sell SHARC based boards) in part because they are easy to 
> use. It is typically much easier to code an application when you have 
> single cycle 32 bit floating point and fixed point instructions 
> available. 
> 

An issue one has to really stress in these classes is cost 
consciousness. You can't use a Lincoln Continental for everything yet 
that is what I see lots of young grads do. Unfortunately cost awareness 
isn't high on the list at universities. It should be.

Regards, Joerg

http://www.analogconsultants.com

On Thu, 10 Nov 2005 21:39:32 +0000, Joerg wrote:

> Hello Al,
> 
>> OTOH, Most microcontrollers are very limited for DSP applications.  
>> 
> 
> Just had a rude awakening there. As Ben mentioned there are some with 
> integrated MAC hardware. Sweet, I thought. Then I found that the MSP430 
> seems to need eight cycles for a multiply and 17 cycles for 
> multiply-accumulate.

Well, there are other examples, too.  Somewhat outside the bracket of the
MSP430 are "microcontrollers" based on the ARMv5 or intel's XScale cores
that have surprisingly respectable "DSP" performance.  (Recent XScale
parts can manage a four-point 16-bit dot product per cycle at hundreds of
MHz, if you line things up right.)  Really, the whole spectrum of "system
on a chip" is pretty thoroughly smeared, these days.

>> The DSPs tend to consume more power, cost more, are often physically
>> larger.
>> 
>> 
> That was the other awakening. Had to scratch the TMS320LF2401 from my
> list because it slurps a whopping 50mA or more. Too much for running off
> AA cells.

Well, some DSPs are pretty miserly.  I'm pretty sure that the TMS 55x and
54x parts are pretty stingy.  Look at the things that are used in MP3
players: 55x, ARM, coldfire.

>> Processors like the Blackfin are kind of hybrids. They are small,
>> consume minimal power and are fast.
>> 
>> 
> If they had somewhat decent audio ADCs on there they would be much more
> useful for an embedded application. I mean, it doesn't have to be top
> notch hifi.

True, there aren't many parts with codecs built in.  That seems to be the
preserve of exclusive SoC parts for large niche markets, like TVs or
phones.

>> I like (and sell SHARC based boards) in part because they are easy to
>> use. It is typically much easier to code an application when you have
>> single cycle 32 bit floating point and fixed point instructions
>> available.
>> 
>> 
> An issue one has to really stress in these classes is cost
> consciousness. You can't use a Lincoln Continental for everything yet
> that is what I see lots of young grads do. Unfortunately cost awareness
> isn't high on the list at universities. It should be.

Development cost is important in some situations, too.  There's a broad
spectrum, which is why there are so many different parts available...

-- 
Andrew

> 
> 
>> I like (and sell SHARC based boards) in part because they are easy to
>> use. It is typically much easier to code an application when you have
>> single cycle 32 bit floating point and fixed point instructions 
>> available. 
>> 
> 
> An issue one has to really stress in these classes is cost 
> consciousness. You can't use a Lincoln Continental for everything yet 
> that is what I see lots of young grads do. Unfortunately cost
> awareness isn't high on the list at universities. It should be.
> 
> Regards, Joerg
> 
> http://www.analogconsultants.com

Cost is very relative.

If you are designing for high volumes, unit (and system) cost are 
critical.

If you are designing for lower volumes, time to market, software 
development time, ease of support, etc are almost always more important.

For example, my company has a business model that works because it is 
cheaper (and less risky) for our customers to buy 100 boards from us than 
design and manufacture their own boards (even if they know how). At high 
volumes, this no longer makes sense. The economics are different.

The first mindset that I usually need to address with students and newly 
minted engineers is that their time is valuable. There is a tendency to 
trade hours for dollars (usually the wrong way). 

I know from earlier posts that Joerg is looking for a very low cost DSP 
solution. I inferred from these posts, that he has a project where unit 
cost is a critical issue and therefore I assume high volume (or at least 
high volume potential). 

I agree with Joerg that cost awareness is something that universities do 
not address well, but I think the issue is more complex than $/part.


-- 
Al Clark
Danville Signal Processing, Inc.
--------------------------------------------------------------------
Purveyors of Fine DSP Hardware and other Cool Stuff
Available at http://www.danvillesignal.com

Hello Al,

> The first mindset that I usually need to address with students and newly 
> minted engineers is that their time is valuable. There is a tendency to 
> trade hours for dollars (usually the wrong way). 
> 

I absolutely agree, and that is where even seasoned engineers are 
sometimes lacking a bit. Every effort, whether time or money, that is 
expended on a project by engineers or any other person or machine needs 
to be recorded as non-recurring up front costs. This has to be properly 
amortized per unit over a reasonable production life.


> I know from earlier posts that Joerg is looking for a very low cost DSP 
> solution. I inferred from these posts, that he has a project where unit 
> cost is a critical issue and therefore I assume high volume (or at least 
> high volume potential). 
> 

Like usual it's the latter (potential) ;-)

Meanin' there's some risk involved.


> I agree with Joerg that cost awareness is something that universities do 
> not address well, but I think the issue is more complex than $/part.
> 

Yes, but one has to remain optimistic about the potential market success 
and not wedge the whole thing into a cost corner where you can't ride 
down the cost vs quantity slope. For example, if we had an expensive 
niche DSP in there we'd have to completely re-engineer the whole 
firmware to go the next step. Else we'd remain a niche supplier ourselves.

Regards, Joerg

http://www.analogconsultants.com

Ben Bradley wrote:
> On Fri, 04 Nov 2005 21:07:52 GMT, Joerg
> <notthisjoergsch@removethispacbell.net> wrote:
>
> >Hello Ross,
> >
> >> Hi. I would like to know what sorts of microcontrollers are available
> >> so that I could design some "student projects" with DSP chips. ...
> >
> >uC or DSP? Very different beasts.
>
>    Well, yes, sort of. There's a bit of overlap, though it appears the
> OP doesn't see the distinction between a microcontroller and a DSP.
>    There are lots of microcontrollers with traditional microprocessor
> cores. These run quite slow for DSP tasks, as they have to go through
> a lot of operations to do a typical DSP calculation.

Erm, I do think that I see the distinction. The current plan is to use
a uC board for making MIDI only devices. This plan is quite well
advanced and is fairly certain to come about. The first board will have
a uC, a serial interface chip, and as little as possible else. It will
not have any A/D or D/A and will not be used for audio processing. For
processing MIDI streams, I would think that the speed of typical uC,
e.g. 32Mhz with mostly one instruction per cycle, would be adequate.
The second board will have a LCD display and a few buttons.

I'm asking questions in this board about DSP chips that might be
suitable for making portable devices that do real-time audio
processing. There have been a lot of useful replies. But I've gone
"quiet" on this thread as the uC board is now the first priority, and
I'll be then thinking about DSP boards, and revisiting this thread via
groups.google, after that is complete.

Cheers,

Ross-c

On 4 Nov 2005 10:32:32 -0800, clemenr@wmin.ac.uk wrote:

>Hi. What microcontrollers (processors with ROM and RAM and some I/O
>onboard so that you can make "devices" from them) are best for DSP?
>I've been told of a general purpose processor that I'm told has a RISC
>architecture and executes (in general) one instruction per clock cycle
>at 32Mhz. Anything better than that which isn't prohibitively expensive
>or consumers lots of current?
>
>Cheers,
>
>Ross-c

Hi,

   sadly I can't give any advice on appropriate 
microcontrollers, but something did occur to 
me.

I wonder if you've seen that terrific book titled:
"Digital Signal Processing and the Microcontroller" 
by Dale Grover and John (Jack) Deller? 

If you haven't seen that book, you should find a copy, 
somewhere, and have a look.  It's about as "real-world" 
and practical as you can get.

Good Luck,
[-Rick-]

I haven't seen that book. My university library has two copies, but
they are both out until June 2006! I'll try and obtain a copy, by
interlibrary loan if necessary.

At the moment my reading time for DSP topics is taken up by reading
"Understanding Digital Signal Processing" by an author whose name looks
remarkably similar to yours.

Cheers,

Ross-c