Re: Re: [oc] fft on hardware??or DSP??

[zeyaohan <zeyaohan@263.net>]

Hi, Jdalton:
    Quite good, I wonder whether you can put your thoughts of your design----about your fastest FFT processor's skeleton on OpenCores? And I do think parallel structure is more difficult to control than pipeline structure.
	I don't know whether you will develop floating points processor or only integer points FFT. About Xilinx VirtexII, they embedded MAC into the VirtexII, so it can run as the speed of ASIC.



>> Hi folks,
>
>Hello
>
>> 1. What is the speed of fastest FFT availabe in the market till now?
>
>Not sure.
>
>> 2. I have seen few xilinx FFT documents for virtex FPGAs saying 1024
>> FFT in arround 60us. Other DSP vendors giving much speed FFTs than
>> FPGAs.
>> So which is suitable for fastest FFT, either DSP or FPGA or dedicated
>> chip.
>> what are the limitations in both.
>
>The FFT I am designing for opencores should do around 1 point 
>every 10ns in an FPGA.  This translates to 10us for 1024 points.  
>Essentially it will be 'real time', allowing a continuous stream of data
>to be  processed.
>
>> 3. Can I get faster FFTs in FPGA than DSPs?
>
>My guess is for long transforms, an FPGA will generally
>have a higher thoughput, as it can do more in parallel.
>A DSP could potentially have a shorter latency due to
>more highly optimised multipliers.  For very short transforms
>a DSP with multiple ALUs might be faster than an FPGA as it can do
>everything in parallel with optimised multipliers.
>
>> 4. What are the advantages of FFT on FPGA than in DSP?
>
>See previous answer.
>
>Best wishes
>John
>

Best Regards


--
To unsubscribe from cores mailing list please visit http://www.opencores.org/mailinglists.shtml