As a hint to the practitioner, doing an FFTshift will reference FFT phase results to the center of the FFT aperture, providing the phase measurement an absolute reference if the center of the FFT window itself has an absolute time reference. An FFTshift will also greatly reduce (although not eliminate) the phase estimation error’s relationship to frequency estimation error. This is because the FFTshift will prevent the phase (of a non-exact-integer-periodic sinusoid) from flipping/alternating between adjacent FFT result bins, meaning a sinusoid “between bins” will produce a similar phase in the 2 adjacent FFT result bins, instead of being near a discontinuity.

Even with an FFTshift, phase measurement accuracy will not only depend on the signal-to-noise ratio, but the shape of any windowing. I have no mathematical proof, but seem to see better phase estimation results with Blackman-Nuttall windows, than with Hamming or Von Hann windows.

As a hint to the practitioner, doing an FFTshift will reference FFT phase results to the center of the FFT aperture, providing the phase measurement an absolute reference, if the center of the FFT window itself has an absolute time reference. An FFTshift will also greatly reduce (although not eliminate) the phase estimation error’s relationship to frequency estimation error. This is because the FFTshift will prevent the phase (of a non-exact-integer-periodic sinusoid) from flipping/alternating between adjacent FFT result bins, meaning a sinusoid “between bins” will produce a similar phase in the 2 adjacent FFT result bins, instead of being near a discontinuity.

Even with an FFTshift, phase measurement accuracy will not only depend on the signal-to-noise ratio, but the shape of any windowing. I have no mathematical proof, but seem to see better phase estimation results with Blackman-Nuttall windows, than with Hamming or Von Hann windows.

As a hint to the practitioner, doing an FFTshift will reference FFT phase results to the center of the FFT aperture, providing the phase measurement an absolute reference if the center of the FFT window itself has an absolute time reference. An FFTshift will also greatly reduce (although not eliminate) the phase estimation error’s relationship to frequency estimation error. This is because the FFTshift will prevent the phase (of a non-exact-integer-periodic sinusoid) from flipping/alternating between adjacent FFT result bins, meaning a sinusoid “between bins” will produce a similar phase in the 2 adjacent FFT result bins, instead of being near a discontinuity.

Even with an FFTshift, phase measurement will not only depend on the signal/noise ratio, but the shape of any windowing. I have no mathematical proof, but seem to see better phase estimation results with Blackman-Nuttall windows, than with Hamming or Von Hann windows.

As a hint to the practitioner, doing an FFTshift will reference FFT phase results to the center of the FFT aperture, providing the phase measurement an absolute reference if the center of the FFT window itself has an absolute time reference. An FFTshift will also greatly reduce (although not eliminate) the phase estimation error’s relationship to frequency estimation error. This is because the FFTshift will prevent the phase (of a non-exact-integer-periodic sinusoid) from flipping/alternating between adjacent FFT result bins, meaning a sinusoid “between bins” will produce a similar phase in the 2 adjacent FFT result bins, instead of being near a discontinuity.

Even with an FFTshift, phase measurement accuracy will not only depend on the signal-to-noise ratio, but the shape of any windowing. I have no mathematical proof, but seem to see better phase estimation results with Blackman-Nuttall windows, than with Hamming or Von Hann windows.