Convert fixed numbers point stored in text format to fxpmath fixed point format

[rafaelnp]

Hello,

I am using fxpmath 0.4.8 on Linux and I have the following use case:

1. generate signal using numpy (e.g. sinewave with 1 or multiple fequencies)
2. generate a processed signal ("golden signal") using numpy, based on the original signal sfrom setp 1
3. convert this generated signal to fixed point using fxpmath
4. save the boh signals signal as base 2 (binary) in a file as text
5. use both files in a VHDL Testbench, performing the same processing as in step 2, and resulting signal saved as a text file, again using base 2(binary)
6. open the VHDL DUT (Design Unfer Test) generated signal file and convert to fxpmath fixed point format
7. compare visually, using e.g mathplotlib, the VHDL DUT processed signals and the fxpmath fixed point "golden signal"

I am having difficulties on the step 6, I could not find a way to convert it to fxpmath fixed point. Any help or suggestion is appreciated.

Thanks in advance.

[francof2a]

Hi!
Let me convert the issue into a discussion.
Could you give some example of the conversion you need for step 6? Input and output formats will help even more.

[rafaelnp]

To generate a sample signal, in this case just a small numpy ndarray. As the step 6, read the text file. This way the best I could get, in the end does not return the expected result...

import numpy as np
from fxpmath import Fxp

def gen_rand_fxp(N=10,  fname='./rand_fxp.txt'):

    rand_gen = np.random.default_rng()
    rand_sequence = rand_gen.uniform(-1.0, 1.0, N)

    n_fxp = Fxp(rand_sequence, dtype='fxp-s16/15')

    with open(fname, 'w') as f:
        for fp in n_fxp:
            f.write(fp.bin(frac_dot=True) + "\n")


def load_fxp(format='fxp-s16/15', fname):

    with open(fname, 'r') as f:
        lines = f.readlines()

    fpnum = Fxp(list(range(len(lines))), dtype=format)

    for i in lines:
        fpnum.__add__(i)

    print(fpnum)
    
    # return the signal to some other function...
    return fpnum


gen_rand_fxp(5)
load_fxp('fxp-s16/15', fname='./rand_fxp.txt')

[rafaelnp]

Any thoughts?

[francof2a]

You don't need to save binary data with fractional dot (let me know if it is a requirement, but it shouldn't be a problem). So, maybe is better to do:

f.write(fp.bin(frac_dot=False) + "\n")

Then, you can read each line and add 0b prefix to each and convert them in a more direct way:

def load_fxp(format='fxp-s16/15', fname='./rand_fxp.txt'):

    with open(fname, 'r') as f:
        lines = f.readlines()
    
    fpnum = Fxp(['0b' + line for line in lines], dtype=format)

    return fpnum

you could also save binary data with prefix previously or if you save with fractional dot, dtype is not required.

[francof2a]

I see there is a potential issue, about size checking when a binary format is provided as input. I'll check it soon!

By now, resizing binary string is a workaround:

fpnum = Fxp(['0b' + line[:-1] for line in lines], dtype=format)