relation between micro frequancy and calculate fft time?

hi all
i want to find relation between micro frequancy and calculate fft time for choosing best micro for varities of fft.for example if i have a sampling rate 1Mhz with 256 lenght of sample.what speed is optimize for choosing micro controller.?
i need to some formula.

Hello!

The only possible reply her is: it depends.
What you can be sure of is that the FFT will consume nLogs multiplication (therefore in your
case 256 log2(256) = 8x256 = 2048 multiplications.
Then, what you need to know:
- Is there a possibility of MAC operation (MAC = Multiply ACcumulate)?
- What number format are you using? Integer? Float? Double?
From this, the figures will change drastically. An example I know well: on a MSP430 (16 bit device),
an integer multiplication takes only a few clock cycles. One day, I have tried a float multiplication,
and it consumed about 300 cycles.
So it depends on your processor. If you have an FPU (e.g. STM43F429), then it will be light-fast.
If not, then it also depends on the CPU frequency, etc...

Dora.

What you can be sure of is that the FFT will consume nLogs multiplication (therefore in your
case 256 log2(256) = 8x256 = 2048 multiplications.

Click to expand...

Or slightly less, using and optimized radix-N algorithm, but there are also additions required. Details are derived in DSP text books.

You have started various threads about your "1 MS/s * 256 samples" project. The specification is still incomplete, e.g. word width was never mentioned. Also, where's the transformed data going to?

Some posts seem to suggest "real-time" operation, which I would translate in case of doubt as continuous block FFT operation without overlapping. The pure FFT arithmetic consumes several 10 MIPS, data I/O has to be added. A fast signal processor or general purpose micro (e.g. ARM9) might be able to handle it, depending on the specification details.

Hi,

DSPs are optimize for such operations.

Klaus

i konw DSPs are optimize .but i have seen some project that use pic18f with my speed.as i searched for best performance with low speed micro that has hardware multipelier , u have to write assembly code not C or sth else.for floating point u need more speed because micro has special algorithm to detect them.i find this application note for pic(AN852).its unclear for me. can i use it for 1Mhz sampling rate with lenght 256 or not.if i lose some data.how much is lost data?
https://www.google.com/url?sa=t&rct...O9dv7O7Yhr1p_kXbw&sig2=M0LE37QHzXkvIAtgKK8luQ