vhdl code for multiplier in filters

[sheeja]

Hi,
m doing a project on eccm techniques. since the following formulae
has to be implemented in vhdl n i m not familiar with such vhdl code i
want help on this.....


i-8 i-8
w = ( ∑ vm × va* ) ÷ ( ∑ va × va* )
k=i k=i

and Vr = vm – w × va

where vm and va are arrays consisting of complex nos: which has been
separated in to real and
imaginary for the purpose of implementing on fpga.

can u plz suggest an optimum vhdl code for the above
formulae...........


regards
sheeja

 

[Pieter Hulshoff]

sheeja wrote:

Hi,
m doing a project on eccm techniques. since the following formulae
has to be implemented in vhdl n i m not familiar with such vhdl code i
want help on this.....


i-8 i-8
w = ( ∑ vm × va* ) ÷ ( ∑ va × va* )
k=i k=i

and Vr = vm – w × va

where vm and va are arrays consisting of complex nos: which has been
separated in to real and
imaginary for the purpose of implementing on fpga.

can u plz suggest an optimum vhdl code for the above
formulae...........

I think you should first decide how you want to build this in hardware
before thinking about the needed VHDL code. What will the frequency be?
How will you set up the multiplier for real and imaginary parts? Are the
parts in integers? What FPGA will you place this in, and how much logic
can be placed between your FFs (pipelining)?

Also, I'm having trouble understanding your 1st formula. What is i? Why
does k run from i to i-8? What is indexed by k?

Kind regards,

Pieter Hulshoff

 

[Ray Andraka]

sheeja wrote:

Hi,
m doing a project on eccm techniques. since the following formulae
has to be implemented in vhdl n i m not familiar with such vhdl code i
want help on this.....


i-8 i-8
w = ( ∑ vm × va* ) ÷ ( ∑ va × va* )
k=i k=i

and Vr = vm – w × va

where vm and va are arrays consisting of complex nos: which has been
separated in to real and
imaginary for the purpose of implementing on fpga.

can u plz suggest an optimum vhdl code for the above
formulae...........


regards
sheeja

The real goal here isn't to put this in VHDL, but rather to wind up with
it in hardware. You could code it in behavioral VHDL and have something
that is not synthesizable but that is a valid VHDL model.

With that in mind, your path should start with determining how this will
map into hardware, only after that is done, then you can start realizing
that hardware with VHDL. Skipping the hardware visualization step is a
recipe for failure.

In order to map the algorithm to hardware, you will first need to
determine how fast it has to go, how many clocks per sample you have to
work with, how much precision and dynamic range is required for each
signal, among other things. You'll also need to ask yourself if this is
the most efficient way to do this in hardware, and if not what
alternative algorithms are there. Also, if the x signifies a matrix
multiply, then you need to break this down into individual complex
multiplies, and then break that down into the real and imaginary parts
unless you already have IP blocks that do that for you.

Finally, if you are not a digital hardware designer, I would strongly
suggest you employ the services of one who has a background in signal
processing as well as FPGA design. Without it, you are in for a
protracted development cycle that doesn't have a high probability of
success.