Closed jaeho3690 closed 5 months ago
Thank you for your insightful questions.
Regarding your first question, which is discussed in detail in the open review, we use phase[:, :, 1:2] because $\arg(\hat{z{1}})$ corresponds to the unit phase shift $2\pi \tau / N$ according to Equation (10), whereas the phase of $\hat{z}{0}$ is zero. This choice ensures proper phase alignment.
As for your second question, multiplying by self.linear aligns the phase as indicated by Equation (10).
If you have any further questions or need additional clarification, please feel free to ask.
Thanks again!
Thank you for sharing your impressive work!
I have two questions about your phase alignment implementation in this line of code: https://github.com/HaokunGUI/VQ_MTM/blob/6912745920397c92d3f33f9585bb690da470f365/layers/Quantize.py#L25.
Thanks in advance!