Signal Subspace

1-D and ensemble signal subspace analysis with methods such as Esprit and RootMusic in Fortran, C, and Python

based in part upon the Spectral Analysis Lib public domain code

The core subspace code is written in Fortran 2008 and is called from other languages (Python, C). Since the programs are Fortran / Python based, they should compile and run in virtually any platform from embedded to supercomputer.

In particular, this program (Fortran, called by C or C++ optionally) works from compilers including:

• Gfortran (GCC)
• Intel oneAPI (ifort, icc, icpc)

Build

Prereqs:

• Linux: apt install liblapack-dev g++ gcc gfortran cmake
• Mac: brew install lapack gcc cmake
• Windows: use MSYS2 or Windows Subsystem for Linux

ctest -S setup.cmake -VV

Then you can test Python calling the Fortran libraries by:

pip install -e .

pytest -v

In the examples below, observe the frequency estimates printed along with their corresponding eigenvalues. A larger eigenvalue is increased confidence in that particular frequency estimate.

Fortran

ESPRIT example with noisy sinusoid

There are two versions of this program, one a full accuracy using double complex numbers, and the other using single real numbers as input. The single real (4 bytes/number) runs about 4 times faster than the double complex (16 bytes/number) program.

./f_esprit_cmpl

./f_esprit_real

C ESPRIT example with noisy sinusoid

Here is an example of calling Fortran Esprit from C, which uses real single precision float:

./c_esprit

C++ ESPRIT example with noisy sinusoid

Example of calling Fortran Esprit from C++, which uses real single precision float:

./cpp_esprit

Python

Compile Fortran ESPRIT to use from Python via f2py

pip install -e .

Selftest Fortran/C/C++/Python Esprit from Python:

pytest

Notes

/liblapack.so: undefined reference to `ATL_zgeru'

Try removing Atlas:

apt remove libatlas-base-dev

Flang / Clang / Clang++

You may need

apt install libc++abi-dev

Plots comparing Fortran to Python

python BasicEspritExample.py