Connect with us

OFDM simulation

Discussion in 'Electronic Design' started by M.Nasiriyan, Nov 8, 2003.

Scroll to continue with content
  1. M.Nasiriyan

    M.Nasiriyan Guest

    Hi.
    What do you know about parallel implementing of OFDM? Can it be in
    software or it is in hardware field only? By the other words, if we
    want to increase the efficiency by using parraleling, do we save time
    or memory or both?
     
  2. red rover

    red rover Guest

    I'm having some problems seeing posts lately.
    Is that this NG or my ISP? I cannot see the
    original post anymore.

    Parallel implementing of OFDM? Some steps have to
    be sequential so they could be pipelined, not done
    in parallel. Depending on what you call parallel,
    it is not uncommon to break up FFT/IFFTs to gain
    processing efficiencies. Do a google search on
    something like "butterfly fft" and you should find
    something on rad 4 butterflys. You can break
    a large FFT into smaller FFTs that could be done
    faster. But unless you can spread those to other
    processors or hardware you'll still end up doing
    them sequentially (but faster).

    Naturally you can do it in hardware or software, it
    just depends on the speed required and the type of
    processor available. DSP processors are obviously
    designed for these type of operations. For really heavy
    duty processing there is also dedicated FFT/IFFT processors.

    Can you give any more information on what your requirements
    are? Data rates, bandwidth, etc.

    The NG comp.dsp would have some expertise too.

    Steve
     
  3. Georg Acher

    Georg Acher Guest

    |> something on rad 4 butterflys. You can break
    |> a large FFT into smaller FFTs that could be done
    |> faster. But unless you can spread those to other
    |> processors or hardware you'll still end up doing
    |> them sequentially (but faster).

    By using smaller FFTs you can gain a major speedup, since the data addressing
    scheme is not very cache-friendly. Eg. when using a sequential "radix-32" FFT as
    the basic block for a 32K-FFT you can direct more than 60% of the data traffic to
    a very small and fast memory. When using prefetching during the calculation of
    the smaller FFT, you can hide the latency of the larger memory.

    The only drawback of that solution is the slightly confusing address generator
    ;-)
     
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day

-