Why do we use fixed point representation?

Hi

           I would like to know why dont we use double precision floating point data type as such for simulink models to be exported to FPGA ? Is it because that FPGA do not support huge memory storage for double precision? Also please add some information about why do we decide to use 16 wordlength fixed point representation in most cases.

Krishnakumar

Answer 1

[Ryan Johnson]

Floating point math is typically expensive in terms of FPGA resources. FPGA primitive DSP elements are typically designed for 16-18 bit integer formats. You can build something which uses floating point, but it will require many more primitive elements (DSP blocks, LUTs, etc) to perform your operations.

Answer 2

[Ryan Johnson]

Sure, you can do double precision floating point math in most FPGAs these days, but at the cost of resources. It's all a trade off analysis. Do you want the highest throughput? Use fixed point. Do you need high dynamic range and precision? Use floating point. I've done designs which use both in different areas.

I guess I'd say that fixed point is still "preferred", as it's natively supported by primitives, but it comes down to what's right for the job.

Ryan

Answer 3

[Krishnakumar]

Sir,

Thank you so much for your answer . But I would like to clear that if any of FPGA support double precision data. And also if still we will prefer fixed point representation due to the speed it offers in computing.

Krishnakumar