This is what appears to be a very precise shift detection.
I was searching for ways to increase the accuracy of a shift detection by cross correlation implemented with FFT and with phase detection, and after many iterative steps with a llm and after many manual correction and modificationn to the code. I come up with this very precise implementation ( low error ) that I share in this repository.
MIT Open Source license
(base) joaocarvalho@soundofsilence:~/zed_editor/shift_1d_deteection> time ./shift_1d_detection.exe
Begin subsample vector 1D shift detection and correction ...
N : 8192
Integer shift from cross-correlation peak: -15 samples
Calculated subsample shift : -15.1234567889414393 samples
True shift : -15.1234567890123461 samples
Error : 7.0906835958339798e-11 samples
Calculated subsample shift ( after shift ): -0.0000000001314220 samples
Integer shift from cross-correlation peak: 19 samples
Calculated subsample shift : 19.1234567888950551 samples
True shift : 19.1234567890123444 samples
Error : 1.1728928939191974e-10 samples
Calculated subsample shift ( after shift ): 0.0000000000209184 samples
Integer shift from cross-correlation peak: 100 samples
Calculated subsample shift : 100.1234567889132450 samples
True shift : 100.1234567890123515 samples
Error : 9.9106500783818774e-11 samples
Calculated subsample shift ( after shift ): 0.0000000001664375 samples
Integer shift from cross-correlation peak: -100 samples
Calculated subsample shift : -100.1234567890396647 samples
True shift : -100.1234567890123515 samples
Error : 2.7313262762618251e-11 samples
Calculated subsample shift ( after shift ): 0.0000000000618456 samples
Integer shift from cross-correlation peak: 4000 samples
Calculated subsample shift : 4000.1234567889960090 samples
True shift : 4000.1234567890123799 samples
Error : 1.6370904631912708e-11 samples
Calculated subsample shift ( after shift ): 0.0000000000800355 samples
Integer shift from cross-correlation peak: -4000 samples
Calculated subsample shift : -4000.1234567889787286 samples
True shift : -4000.1234567890123799 samples
Error : 3.3651303965598345e-11 samples
Calculated subsample shift ( after shift ): -0.0000000000927685 samples
... end subsample vector 1D shift detection and correction.
real 0m0,122s
user 0m0,112s
sys 0m0,010sBest regards,
Joao Carvalho
