DSP hardware

Hi,

I am newbie in this area.

I read a paper on DSP and found that there is reconfigurable DSP. And
we also can implement DSP on FPGA. My question is, is DSP a dedicated
or reconfigurable hardware or is it both? In wikipedia, DSP is said to
be dedicated. I appreaciate any feedback.

Thank you.

awa wrote:
> Hi,
> 
> I am newbie in this area.
> 
> I read a paper on DSP and found that there is reconfigurable DSP. And
> we also can implement DSP on FPGA. My question is, is DSP a dedicated
> or reconfigurable hardware or is it both? In wikipedia, DSP is said to
> be dedicated. I appreaciate any feedback.
> 
> Thank you.
> 

I can imagine someone saying "reconfigurable" when what they mean is 
"reprogrammable".

I'd say a DSP is dedicated in the sense that you cannot change its 
instruction set architecture.  On the other hand it is programmable so 
there is still a good degree of flexibility.

An FPGA is much more flexible yet.  However you trade off size, cost, 
and power for this flexibility.  For a reasonably small/simple core this 
may be a good tradeoff but I'm guessing an ARM11 implemented in an FPGA 
would be a lot of gates!

Brad

Brad Griffis skrev:
> awa wrote:
> > Hi,
> >
> > I am newbie in this area.
> >
> > I read a paper on DSP and found that there is reconfigurable DSP. And
> > we also can implement DSP on FPGA. My question is, is DSP a dedicated
> > or reconfigurable hardware or is it both? In wikipedia, DSP is said to
> > be dedicated. I appreaciate any feedback.
> >
> > Thank you.
> >
>
> I can imagine someone saying "reconfigurable" when what they mean is
> "reprogrammable".
>
> I'd say a DSP is dedicated in the sense that you cannot change its
> instruction set architecture.

That goes for just about any CPU. I suppose one might say that a DSP
is "dedicated" in the sense that it is very useful for DSP types of
applications, and very cumbersome (maybe even useless) in general
applications. 

Rune

Brad Griffis wrote:
> awa wrote:
> 
>> Hi,
>>
>> I am newbie in this area.
>>
>> I read a paper on DSP and found that there is reconfigurable DSP. And
>> we also can implement DSP on FPGA. My question is, is DSP a dedicated
>> or reconfigurable hardware or is it both? In wikipedia, DSP is said to
>> be dedicated. I appreaciate any feedback.
>>
>> Thank you.
>>
> 
> I can imagine someone saying "reconfigurable" when what they mean is 
> "reprogrammable".
> 
> I'd say a DSP is dedicated in the sense that you cannot change its 
> instruction set architecture.  On the other hand it is programmable so 
> there is still a good degree of flexibility.
> 
> An FPGA is much more flexible yet.  However you trade off size, cost, 
> and power for this flexibility.  For a reasonably small/simple core this 
> may be a good tradeoff but I'm guessing an ARM11 implemented in an FPGA 
> would be a lot of gates!
> 
> Brad

DSP is Digital Signal Processing, which is the manipulation of signals 
using numeric methods.  Any device that is the host for doing that 
numerical manipulation is a "digital signal processor".  The Term 
digital signal processor was somewhat hijacked by the microprocessor 
companies in the 1980s who developed microprocessors with a feature set 
that offered advantages over generic microprocessors for DSP 
applications, and as a result hardware solutions are left lacking a name 
that sets them apart from the software solutions.  The predominant DSP 
is the specialized microprocessor (specialized in that it has features 
like a single cycle multiply-accumulate, special looping instructions, 
or special addressing modes), but that does not preclude a hardware 
solution.  Indeed, the first digital signal processors were dedicated 
hardware systems.  Microprocessors did not play a significant part in 
signal processing until the 1980-1985 time frame.  Before that, digital 
signal processing was done on dedicated hardware designs.

The FPGA provides a low cost and compared to the early signal processing 
systems, flexible platform for hardware DSP that offers the speed 
advantages of dedicated hardware without the severe limits on making 
changes to the design.

Hi,

Thank you for all the feedbacks.

To Brad,

> I can imagine someone saying "reconfigurable" when what they mean is
> "reprogrammable".
>
> I'd say a DSP is dedicated in the sense that you cannot change its
> instruction set architecture.  On the other hand it is programmable so
> there is still a good degree of flexibility.

You highlighted a very good point. I had always thought that if
something is 'reprogrammable', it also means that it is
'reconfigurable' (because when code has changed, hence the
architecture). Do you mind to elaborate more on this?

Thank you.

"Rune Allnor" <allnor@tele.ntnu.no> wrote in news:1163684939.477105.13920
@m7g2000cwm.googlegroups.com:

> 
> Brad Griffis skrev:
>> awa wrote:
>> > Hi,
>> >
>> > I am newbie in this area.
>> >
>> > I read a paper on DSP and found that there is reconfigurable DSP. 
And
>> > we also can implement DSP on FPGA. My question is, is DSP a 
dedicated
>> > or reconfigurable hardware or is it both? In wikipedia, DSP is said 
to
>> > be dedicated. I appreaciate any feedback.
>> >
>> > Thank you.
>> >
>>
>> I can imagine someone saying "reconfigurable" when what they mean is
>> "reprogrammable".
>>
>> I'd say a DSP is dedicated in the sense that you cannot change its
>> instruction set architecture.
> 
> That goes for just about any CPU. I suppose one might say that a DSP
> is "dedicated" in the sense that it is very useful for DSP types of
> applications, and very cumbersome (maybe even useless) in general
> applications. 
> 
> Rune
> 

I used to describe DSP processors as kind of idiot savant 
microcontrollers, They were typically good for signal processing with the 
features that Rune mentions, but limited for general purpose 
applications.

I don't think this is so true any longer. Many newer DSPs have a lot of 
the features one would use in a microcontroller application. 

For example, the SHARC ADSP-21369 was two UARTs, two SPI ports, an I2C 
interfaces, Clock generators and lots of bit addressable I/O. This is all 
in addition to features that are more DSP oriented (8 SPORTs, 8 Sample 
Rate Converters, S/PDIF, etc)

It's also true that processors that are thought of as general purpose are 
becoming better at signal processing functions. The various ARM 
processors come to mind.

FPGAs are good for applications that need very fast, well defined signal 
processing. In many cases it will act as a coprocessor in a larger 
system. Most signal processing systems that we see that use FPGAs, will 
combine an FPGA with a general purpose DSP.

I would suggest that a DSP chip is easier to program for DSP 
applications.



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

Hi Al Clark,

Can I find further info on general purpose DSP on at
http://www.danvillesignal.com? I really would like to look more on
this.

Thank you.

Al Clark wrote:


> I used to describe DSP processors as kind of idiot savant 
> microcontrollers,

Straight to the point!

However, the old DSP programming didn't look bad at all compared to PIC 
or x51 programming.

> I don't think this is so true any longer. Many newer DSPs have a lot of 
> the features one would use in a microcontroller application. 

There is still a lot of overhead. The general purpose code on a DSP has 
2-3 times more fat compared to the same code for a microcontroller.

> 
> For example, the SHARC ADSP-21369 was two UARTs, two SPI ports, an I2C 
> interfaces, Clock generators and lots of bit addressable I/O. This is all 
> in addition to features that are more DSP oriented (8 SPORTs, 8 Sample 
> Rate Converters, S/PDIF, etc)

No ADC, no flash, no eeprom. The peripherals are definitely done by the 
crazy scientists.


> It's also true that processors that are thought of as general purpose are 
> becoming better at signal processing functions. The various ARM 
> processors come to mind.

Basically, the most needed thing is the fast multiplier.

> 
> FPGAs are good for applications that need very fast, well defined signal 
> processing. In many cases it will act as a coprocessor in a larger 
> system. Most signal processing systems that we see that use FPGAs, will 
> combine an FPGA with a general purpose DSP.

With FPGA, you can trade the complexity for the speed. With DSP, you can 
trade the speed for the complexity.

> 
> I would suggest that a DSP chip is easier to program for DSP 
> applications.

It's almost completely in C/C++ anyway, so it doesn't really matter in 
our days.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

awa wrote:
> Hi,
> 
> Thank you for all the feedbacks.
> 
> To Brad,
> 
>> I can imagine someone saying "reconfigurable" when what they mean is
>> "reprogrammable".
>>
>> I'd say a DSP is dedicated in the sense that you cannot change its
>> instruction set architecture.  On the other hand it is programmable so
>> there is still a good degree of flexibility.
> 
> You highlighted a very good point. I had always thought that if
> something is 'reprogrammable', it also means that it is
> 'reconfigurable' (because when code has changed, hence the
> architecture). Do you mind to elaborate more on this?

Most computers of any kind have a fixed instruction set. The number and 
sequence of instructions that make up a program depends on the program, 
but the instructions themselves depend on the processor. (Some 
processors are "microprogrammed" internally, and the microinstructions 
are in read-only memory that is part of the processor. A In a few, it is 
possible to alter the microprogram store, and those could be considered 
reconfigurable to a degree, but they are usually referred to as 
"microprogrammable".)

Imagine a family of gates, counters, registers, and all the 
paraphernalia of discrete hardware design. Imagine a breadboard that 
accepts these miniature marvels and allows them to be interconnected at 
will. In this setup, there is a close analogy between reconfiguring the 
breadboard and reprogramming a computer. Such a world exists -- it is an 
FPGA. Nevertheless, each FPGA has a fixed number of elements, and they 
are in fact arranged for use by software of a sort. If an FPGA with more 
or different facilities is needed, it's not the same part any more.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Here is my two cents.

IMHO, the key features of a DSP (in decreasing importance) are

1)  Harvard architechure RISC.
2)  Circular Buffer Support
3)  Hardware Multiplier
4)  Support for Zero overhead loops
5)  Hardware MAC
6)  Delayed Branch Jumps

Because of feature 1, digital signal processors tend to get used alot
by embedded people for non-DSP functions, like IO-routing, or for
functions that require precise timing (it is easy to count cycles with
a Harvard RISC).

Also, IMHO, alot of the fat that is generated with a DSP on general
purpose code is because the compiler/optimizer isn't as good as
compiler for the general purpose processors.  Anymore, I favor the ST
ARM9 processors for DSP (if I don't need floating point).  The ARM
optimizers are incredible, and the processor has items 1, 3, and 5 from
my list.

Finally, IMHO, I'm starting to think that FPGAs are as easy to use as
DSPs.  In an FPGA, I can code in HDL, which (in my optinion) is not as
easy to code in as C.  However, when I read the DSP compiler manual, it
says that I need to write my code in a certain way, so that the
optimizer can "realize" it into the correct instructions.  That's
starting to sound alot like an HDL.  Further, I tend go get annoyed
with the C optimizer and just program in assembly language.  Which is
easier HDL, or assembly language??