Slow FFT in ksh93 and awk. Post: 303022348

Sponsored Content

Operating Systems OS X (Apple) Slow FFT in ksh93 and awk. Post 303022348 by wisecracker on Thursday 30th of August 2018 02:31:22 PM

08-30-2018

Registered User

Slow FFT in ksh93 and awk, ARexx and pure awk.

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:

Code:

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:

Code:

#!/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...

Quote:

Originally Posted by wisecracker
I am an AMIGA fanatic still and was searching for a basic FFT routine for Python 1.4.0 for the AMIGA.

My suggestion is to first do it in Assembler. Doing it in Python (or any other interpreted language, except probably FORTH) is like trying to build a race car - out of Lego. And, second, to do it on a DSP. I am not up to date with the latest DSPs any more, but i suggest to take a look at the C66x series from Texas Instruments or the ADSP-SC57x series from Analog Devices. Both are RT-capable floating-point DSPs and from my personal experience (TI32C80 from TI, i can't remember the name of the AD machine) both manufacturers made great procs. You might also try the still available 56k-series by Motorola. I used these (especially the 56001) a lot back in the days.

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...

Last edited by wisecracker; 09-14-2018 at 07:49 AM.. Reason: (Attached itself to this post.) Now title changed.

This User Gave Thanks to wisecracker For This Post:

wisecracker

View Public Profile for wisecracker

Find all posts by wisecracker

10 More Discussions You Might Find Interesting

1. UNIX for Dummies Questions & Answers

echo is too slow. HELP with Awk

Hello All, Below is a simple script i worte to find the 208th char in a file. If the char = "C" then I re-direct the line to a file called change.txt. If it is not "C" then I re-direct it to a file called delete.txt. My problem is I have a file 0f 500K lines. this script is very slow. I am...

2. Shell Programming and Scripting

ksh93 deprecation...

Any means of running ksh93 in a ksh88-mode? Might sound odd, but I want/need to restrict U/Win-developed scripts to correspond to the ksh88 version on my Solaris environment(s). Thanks.

3. Shell Programming and Scripting

ksh88 or ksh93

Hi all! Does anybody know how can I check if any UNIX installation has implemented ksh88 or ksh93? Thanks in advance. N�stor.

4. Shell Programming and Scripting

The builtin split function in AWK is too slow

I have a text file that contains 4 million lines, each line contains 2 fields(colon as field separator). as shown: 123:444,555,666,777,888,345 233:5444,555,666,777,888,345 623:454,585,664,773,888,345 ...... Here I have to split the second field(can be up to 40,000 fields) by comma into an...

5. Shell Programming and Scripting

cut, sed, awk too slow to retrieve line - other options?

Hi, I have a script that, basically, has two input files of this type: file1 key1=value1_1_1 key2=value1_2_1 key4=value1_4_1 ... file2 key2=value2_2_1 key2=value2_2_2 key3=value2_3_1 key4=value2_4_1 ... My files are 10k lines big each (approx). The keys are strings that don't...

6. Shell Programming and Scripting

Making a faster alternative to a slow awk command

Hi, I have a large number of input files with two columns of numbers. For example: 83 1453 99 3255 99 8482 99 7372 83 175 I only wish to retain lines where the numbers fullfil two requirements. E.g: =83 1000<=<=2000 To do this I use the following...

7. UNIX for Advanced & Expert Users

Ksh93 on Linux compatible with ksh93 on AIX

Hi Experts, I have several shell scripts that have been developed on a Linux box for korn ksh93. If we want to run this software on an AIX 6.1 box that runs ksh88 by default can we just change the she-bang line to reference /bin/ksh93 which ships with AIX as its "enhanced shell" to ensure...

8. OS X (Apple)

FFT for the AMIGA through ksh88 shell.

I don't know if anyone is interested but I have been meddling with FFT for the AMIGA. (Sadly we AMIGAns don't have these luxuries through any scripting language. Below is a Python snippet that uses the builtin 'cmath' module to work with the lowly Python 2.0.1 for the AMIGA. It is part of a...

9. AIX

Ksh93/AIX compatibility

Hi everyone ! Im trying to know from wich version of AIX KSH93 is available ? Internet tell me 6.x and 7.x AIX are available, bue what about 5.x ? Is KSH93 available on AIX 5.x ? Is it the same way to manipulate variables as KSH93 on 7.x ? Thanks for your support and have a nice day !

10. OS X (Apple)

FFT for Python 2.0.x to 3.7.0.

Hi guys... This is code that was originally designed to work on an upgraded AMIGA A1200 using Python 2.0.x. Unfortunately it broke inside much later versions, NOT because of the print statement/function but other minor subtleties. So this is the final result tested on various machines including...