How to know format (# of bits for fraction and integer part) of the input and output for an unscaled, pipelined/streaming io in FFT v7.1

[nic.dc]

Hi, I'm starting to use the FFT v.7 but then I would want to know the format of the bits for the input and output for unscaled method with 8 bit input and 12 bit output. It says nothing about the format in the data sheet. This format being the no. of bits for the integer part and fraction part of the input and output.

Waiting for help

Thanks

[nic.dc]

By the way, I also used Truncation rounding and used Natural order

[muzaffer]

ds260 page 2 says this: "For fixed-point inputs, the input data is a vector of N complex values represented as dual bx-bit two’s-complement numbers, that is, bx bits for each of the real and imaginary components of the data sample, where bx is in the range
8 to 34 bits inclusive."

which is all you need. Any fractional number has an "imaginary" binary point which is used to decide which bit has the weight 2^0. Other than that, the bit order is decided by two's complement behavior

[nic.dc]

Hi muzaffer! Thanks for the insight! but, how would I know where the binary point is placed for the input and output?

Thanks!

[bwiec]

Hello,

The doc also says:

Note that the core does not have a specific location
for the binary point. The output maintains the same binary point location as the input.

[nic.dc]

Thanks for that bwiec! How would I know then where the binary point from the input is?

Waiting for a reply

Thanks!

[muzaffer]

Input format is something you decide/control and depends on the filtering/processing you do before hand. You can always assume all numbers are in 1.x format with a single bit for sign and (x=n-1) bits for fraction.

[nic.dc]

Oh, so I can also assume that the format is 1x.y where: There's a single bit for sign, x no. of bits for the integer, and y no. of bits for the fraction?

Waiting for a reply
Thanks!

[muzaffer]

Yes. That's reasonable.

[nic.dc]

So, initially, I can always assume my desired format (such as the previous format 1x.y) even without filtering/processing?

And, is 1.x assumed to be the default format?

Waiting for a reply
Thanks!

[muzaffer · Accepted solution]

Effectively at the input of system you can assume the format anything, even 1.x (ie 1 sign and the rest fractional). Once you start filtering, you will need to decide how the whole part grows and/or how many bits you need to keep from the fractional part (adding multiple numbers grows the dynamic range and multiplying them increases the fractional part too) so you need to adjust the format as you go.