.like() seems to fail on complex numbers

[mousecat98]

Hi all,
I am facing an issue with fxpmath's .like() functionality. In my case, I perform some fixed-point signal processing using numpy's convolve()-method. Using ordinary datatype definitions, the result is a complex number which is correct. However, using ahead datatype definitions and the .like()-function the resulting number is real-valued.

import numpy as np
from fxpmath import Fxp

# datatype definition
TAP = Fxp(None, dtype='fxp-s32/24-complex')
SIGNAL = Fxp(None, dtype='fxp-s32/24-complex')

# signal
signal = np.array([(1 + 1j), (1 - 1j), (-1 + 1j), (-1 - 1j)], dtype=complex)
signal_fxp = Fxp(signal).like(SIGNAL)
signal_fxp1 = Fxp(signal, dtype='fxp-s32/24-complex')

# generate filter
filt = np.arange(0.1, 0.5, 0.1)
filt_fxp = Fxp(filt).like(TAP)
filt_fxp1 = Fxp(filt, dtype='fxp-s32/24')

# convolve signal and filter
out = np.convolve(signal_fxp, filt_fxp, 'same')
out1 = np.convolve(signal_fxp1, filt_fxp1, 'same')

# debug output
print("Signal FXP")
signal_fxp.info(verbose=3)
signal_fxp1.info(verbose=3)
print(100 * "-")
print("Filter FXP")
filt_fxp.info(verbose=3)
filt_fxp1.info(verbose=3)
print(100 * "-")
print("Output FXP")
out.info(verbose=3)
out1.info(verbose=3)

This code results in the following console output:

Signal FXP
	dtype		=	fxp-s32/24-complex
	Value		=	[ 1.  1. -1. -1.]
	extended_prec	=	True

	Signed		=	True
	Word bits	=	32
	Fract bits	=	24
	Int bits	=	7
	Val data type	=	<class 'float'>

	Upper		=	127.99999994039536
	Lower		=	-128.0
	Precision	=	5.960464477539063e-08
	Overflow	=	saturate
	Rounding	=	trunc
	Shifting	=	expand

	dtype		=	fxp-s32/24-complex
	Value		=	[ 1.+1.j  1.-1.j -1.+1.j -1.-1.j]
	extended_prec	=	True

	Signed		=	True
	Word bits	=	32
	Fract bits	=	24
	Int bits	=	7
	Val data type	=	<class 'complex'>

	Upper		=	127.99999994039536
	Lower		=	-128.0
	Precision	=	5.960464477539063e-08
	Overflow	=	saturate
	Rounding	=	trunc
	Shifting	=	expand

----------------------------------------------------------------------------------------------------
Filter FXP
	dtype		=	fxp-s32/24
	Value		=	[0.09999996 0.19999999 0.29999995 0.39999998]
	inaccuracy	=	True
	extended_prec	=	True

	Signed		=	True
	Word bits	=	32
	Fract bits	=	24
	Int bits	=	7
	Val data type	=	<class 'float'>

	Upper		=	127.99999994039536
	Lower		=	-128.0
	Precision	=	5.960464477539063e-08
	Overflow	=	saturate
	Rounding	=	trunc
	Shifting	=	expand

	dtype		=	fxp-s32/24
	Value		=	[0.09999996 0.19999999 0.29999995 0.39999998]
	inaccuracy	=	True
	extended_prec	=	True

	Signed		=	True
	Word bits	=	32
	Fract bits	=	24
	Int bits	=	7
	Val data type	=	<class 'float'>

	Upper		=	127.99999994039536
	Lower		=	-128.0
	Precision	=	5.960464477539063e-08
	Overflow	=	saturate
	Rounding	=	trunc
	Shifting	=	expand

----------------------------------------------------------------------------------------------------
Output FXP
	dtype		=	fxp-s25/24
	Value		=	[ 0.29999995  0.39999998  0.39999998 -0.09999996]

	Signed		=	True
	Word bits	=	25
	Fract bits	=	24
	Int bits	=	0
	Val data type	=	<class 'float'>

	Upper		=	0.9999999403953552
	Lower		=	-1.0
	Precision	=	5.960464477539063e-08
	Overflow	=	saturate
	Rounding	=	trunc
	Shifting	=	expand

	dtype		=	fxp-s25/24-complex
	Value		=	[ 0.29999995+0.10000002j  0.39999998+0.19999993j  0.39999998+0.20000005j
 -0.09999996-0.30000001j]

	Signed		=	True
	Word bits	=	25
	Fract bits	=	24
	Int bits	=	0
	Val data type	=	<class 'complex'>

	Upper		=	0.9999999403953552
	Lower		=	-1.0
	Precision	=	5.960464477539063e-08
	Overflow	=	saturate
	Rounding	=	trunc
	Shifting	=	expand

Thanks for your help!

[francof2a]

Hello @mousecat98
I can reproduce the error. Thanks for the issue, I'll try to find a solution for that.

I will have news soon.

[francof2a]

I've fix the problem. It was not about like method but using None to initialize the template. As workaround you might try replace None by complex(0) (or even any complex number).

By the way, this is gonna to be fixed in the next release (0.4.6).

Thank you!

[francof2a]

Issue solved in v0.4.6