Near-optimal bounds for signal recovery from blind phaseless periodic short-time Fourier transform

We study the problem of recovering a signal $x\in\mathbb{C}^N$ from samples of its phaseless periodic short-time Fourier transform (STFT): the magnitude of the Fourier transform of the signal multiplied by a sliding window $w\in \mathbb{C}^W$. We show that if the window $w$ is known, then a generic signal can be recovered, up to a global phase, from less than 4N phaseless STFT measurements. In the blind case, when the window is unknown, we show that the signal and the window can be determined simultaneously, up to a group of unavoidable ambiguities, from less than 4N+2W measurements. In both cases, our bounds are optimal, up to a constant smaller than two.