Daniel G
Newbie level 1
Hi,
I am confused due to the election of the correct DSP because I need to process high quality audio 24 bits long (>=16k) FFT in less than 500us.
My first approximation was the C6747 cores in floating point (as the precision meets with professional audio), but the tools and libraries are old and difficult to get. In addition, at maximum speed the FFTs required are too slow. Recently I discover the DM3730 and the c64x+ core and at 800MHz it seems to be fast enough. Now my doubts are about the precision of the fixed point DSP in the function fft_32x32 bits. As far as I know for professional audio double precision is required, but many equipment goes ahead with 48 bits accumulation/intermediate registers (double precision for 24bits).
Does the DM3730 meet the requirements? I read the datasheet but I did not found the width of the accumulators, the 6 ALU have 32-40bits width, Could be the maximum width?
Nowadays the floating point DSPs seems to be displaced (at least at texas instruments) by the fixed point ones. Another device suggestions are welcome.
Kind regards.
I am confused due to the election of the correct DSP because I need to process high quality audio 24 bits long (>=16k) FFT in less than 500us.
My first approximation was the C6747 cores in floating point (as the precision meets with professional audio), but the tools and libraries are old and difficult to get. In addition, at maximum speed the FFTs required are too slow. Recently I discover the DM3730 and the c64x+ core and at 800MHz it seems to be fast enough. Now my doubts are about the precision of the fixed point DSP in the function fft_32x32 bits. As far as I know for professional audio double precision is required, but many equipment goes ahead with 48 bits accumulation/intermediate registers (double precision for 24bits).
Does the DM3730 meet the requirements? I read the datasheet but I did not found the width of the accumulators, the 6 ALU have 32-40bits width, Could be the maximum width?
Nowadays the floating point DSPs seems to be displaced (at least at texas instruments) by the fixed point ones. Another device suggestions are welcome.
Kind regards.
