C55x5: Block Floating Point

Roman Arnet

Hello,
We are considering for a future product a C55x5 DSP that comes along with a FFT coprocessor. In order to perform part of calculations in the frequency domain, we intend to implement the Constant-Overlap-and-Add (COLA) method as referred in http://www.ti.com/lit/an/sprabb6b/sprabb6b.pdf chapter 7.
However, having checked the API of the copro I am not sure if the DSP really fits to what we need as block floating point (e.g. http://www.ti.com/lit/an/spra948/spra948.pdf) seems not to be natively supported.
Is there a technique how we could profit from the FFT coprocessor to implement an efficient COLA that supports block floating point?
Any suggestions would be helpful.

over 11 years ago

0 Mark M over 11 years ago

TI__Mastermind 30110 points

Hi,

The C55xx FFT hardware accelerator only supports 16-bit fixed point.

Floating point is not native on C55x but can be computed in software (relatively expensive).

FFTs can also be computed in software using code from DSPLIB: http://www.ti.com/tool/sprc100

Hope this helps,
Mark

Processors

Processors forum

C55x5: Block Floating Point