Well i set myself a challenge to have an FFT function using nothing but ksh93 and awk.
It took some serious jiggery pokery and concentration with all the brackets and '$' characters but here is the result. It is RADIX 2 only, but hey, show me another UNIX shell script that does it. It IS SLOW but is ideal for my needs.
It uses ksh's floating point arithmetic and awk to obtain the SINE and COSINE requirements although even awk is not needed for this as typical Taylor series for these could have been created.
This was a learning curve primarily because ksh has a few minor differences to bash; however this is the result.
(For those that know the AMIGA, I am now thinking of porting this to ARexx although it has no SINE, COSINE or SQRT. I have already done SINE and SQRT for ARexx so COSINE is just a simple SINE change.)
This is in the Public Domain, CC0, here and hopefully there will be some interest from outside our community.
The results of the code below:
Last login: Thu Aug 30 16:06:02 on ttys000
AMIGA:amiga~> cd Desktop/Code/Shell
AMIGA:amiga~/Desktop/Code/Shell> ksh
AMIGA:uw>
AMIGA:uw> ./FFT_ksh
Real floating point values:
8.5 3.637821186735 -1.70710678119 0.051425210515 1.5 -0.17274555408 -0.292893218805 0.483499156825 0.5 0.483499156845 -0.292893218815 -0.17274555407 1.5 0.05142521051 -1.7071067812 3.637821186735
Imaginary floating point values:
0 -1.93071440555 -0.5 -0.2414680083 -1 0.465638772885 0.5 -1.223607624345 0 1.223607624355 -0.499999999995 -0.46563877289 1 0.24146800831 0.500000000005 1.930714405535
Absolute values of complex numbers:
8.500000 4.118420 1.778820 0.246883 1.802780 0.496649 0.579471 1.315670 0.500000 1.315670 0.579471 0.496649 1.802780 0.246883 1.778820 4.118420
Done...
AMIGA:uw> exit
AMIGA:amiga~/Desktop/Code/Shell> _
The script:
#!/bin/ksh
# KSH Version: 'Version AJM 93u+ 2012-08-01'
# From both ${.sh.version} and ${KSH_VERSION}.
#
# Awk Version: 'awk version 20070501'
#
# OSX 10.13.6, default bash terminal running ksh.
# Darwin Barrys-MacBook-Pro.local 17.7.0 Darwin Kernel Version 17.7.0:
# Thu Jun 21 22:53:14 PDT 2018; root:xnu-4570.71.2~1/RELEASE_X86_64 x86_64
# All of this code assumes functions can use global variables, arrays, etc...
# Only constant required, 11 decimal places.
PI=3.14159265359
fft()
{
# Check for equal length real and imaginary arrays.
if [ ${#REAL_ARRAY[@]} -ne ${#IMAG_ARRAY[@]} ] || [ ${#REAL_ARRAY[@]} -le 1 ] || [ ${#IMAG_ARRAY[@]} -le 1 ]
then
echo "ERROR 1:"
echo "Number of elements, both real and imaginary, must be the same."
exit 1
fi
# Check for powers of 2.
N=${#REAL_ARRAY[@]}
if [ $(( $N&(N-1) )) -ne 0 ]
then
echo "ERROR 2:"
echo "Padding or cropping is required to the nearest power of 2 elements."
exit 2
fi
REAL=()
IMAG=()
# For each element into REAL() and IMAG() arrays...
for ((K=0; K<=$(( $N-1 )); K++))
do
SUMREAL=0.0
SUMIMAG=0.0
# For each element from REAL_ARRAY() and IMAG_ARRAY() arrays...
for ((T=0; T<=$(( $N-1 )); T++))
do
ANGLE=$(( 2.0*PI*$T.0*$K.0/$N.0 ))
# Use awk for the COSINE and SINE values...
COS_ANGLE=$( awk 'BEGIN { printf("%.11f", cos('$ANGLE')); }' )
SIN_ANGLE=$( awk 'BEGIN { printf("%.11f", sin('$ANGLE')); }' )
SUMREAL=$(( $SUMREAL+${REAL_ARRAY[$T]}*$COS_ANGLE+${IMAG_ARRAY[$T]}*$SIN_ANGLE ))
SUMIMAG=$(( $SUMIMAG-${REAL_ARRAY[$T]}*$SIN_ANGLE+${IMAG_ARRAY[$T]}*$COS_ANGLE ))
done
REAL[$K]=$SUMREAL
IMAG[$K]=$SUMIMAG
done
}
abs()
{
SQUARE=$(( (${REAL[COUNT]}**2)+(${IMAG[COUNT]}**2) ))
ABS=$( awk 'BEGIN { printf(sqrt('$SQUARE')); }' )
}
# 16 samples of single cycle square wave, floating point values must always be positive.
#
# REAL_ARRAY=( 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 )
# IMAG_ARRAY=( 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 )
#
# 8.000000 5.125830 0.000000 1.799950 0.000000 1.202690 0.000000 1.019590
# 0.000000 1.019590 0.000000 1.202690 0.000000 1.799950 0.000000 5.125830
# This has midway '0.5' value padding at the end to bring to the power of 2.
# This is the running default.
#
REAL_ARRAY=( 1.0 1.0 1.0 1.0 1.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.5 0.5 0.5 )
IMAG_ARRAY=( 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 )
#
# 8.500000 4.118420 1.778820 0.246880 1.802780 0.496650 0.579470 1.315670
# 0.500000 1.315670 0.579470 0.496650 1.802780 0.246880 1.778820 4.118420
# Standard test values.
#
# REAL_ARRAY=( 1.0 1.0 1.0 1.0 0.0 0.0 0.0 0.0 )
# IMAG_ARRAY=( 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 )
#
# 4.000000 2.613130 0.000000 1.082390 0.000000 1.082390 0.000000 2.613130
# This assumes all needed variables are global!
fft
# This lot can be discarded...
echo ""
echo "Real floating point values:"
echo "${REAL[@]}"
echo ""
echo "Imaginary floating point values:"
echo "${IMAG[@]}"
echo ""
echo "Absolute values of complex numbers:"
for COUNT in $( seq 0 1 $(( ${#REAL[@]}-1 )) )
do
abs
printf "%f " "$ABS"
done
echo ""
echo ""
echo "Done..."
Enjoy all.
Snippets like these are what keep me from going insane.
------ Post updated at 07:31 PM ------
Hi bakunin...
Thank you for the info but......
I do bizarre things like this to expand my knowledge on the limits of languages, not the limits of HW.
I even got my USB Arduino Diecimila board to talk to my AMIGA with minor HW mods, uploaded to AMINET, using BWBasic purely as a challenge, that's right BWBasic. It has had hundreds of DLs which shows people ARE interested in esoteric stuff like mine.
Real time applications are of no interest at the moment as AudioScope.sh is by definition slow as is its builtin Spec-An using Python 2.7.x with module imports of scipy and scipy.IO. As these ARE dependencies then I have written FFT code that works from Python 2.0.1 to the current 3.7.0. I now have since modified code to work from Python 1.4.0, the only version to work on a stock AMIGA A1200, also up to 3.7.0. Boy how, Python has changed. These mods can be seriously difficult to work around but now I have my building blocks to work with to adapt into AudioScope.sh for AudioScope.sh's Spec-An section without the need for dependencies.
ATM FFT on ksh was a new baby to see how easy it would be, difficult? YES; impossible? NO. I really do get enjoyment out of stretching languages to strange esoteric limits.
Can I do it in AWK? Dunno, I am not good with AWK but ARexx for the AMIGA WILL be serious fun as it does have basic floating point arithmetic but NO SINE, COSINE, SQRT, etc...