Answers are grounded in this paper and cite their sources.
In this paper, we study the degrees of freedom (DoF) of multiple-antenna channels with K specular paths, N_t transmit antennas on a half-wavelength grid and a uniform array of N_r >= 2K receive antennas, with path gains and angles unknown at both ends and redrawn in every block of T channel uses. We define the learning cost c as the number of signal dimensions per block lost to learning when T >= N_t. We prove that c >= 3/2 for every array once K >= N_t, with equality for non-uniform arrays once K >= 3N_t - 2, and already from K = N_t + 2 for every non-uniform array with N_t <= 4 and for the first explicit infinite family at that threshold, that c = 5/2 for uniform arrays with K >= N_t + 2 except at N_t = T = 2, and that c grows linearly in N_t at K = N_t + 1. For T >= N_t, these values need only K + 2 receive antennas. Here, c is half the real dimension of a space of symmetries of the transmit array, and the converses rest on genie and covering arguments adapted to these symmetries. For N_t >= 3, we also obtain the DoF of uniform arrays with K >= N_t + 2 at every T >= 2, and of every array at T = 2. When K >= N_t + 2 paths renew at staggered times known to both ends, the learning cost vanishes.
No takes yet. Share an insight, caveat, or question.
Karim Badawi (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: