Phase retrieval in high dimensions: Statistical and computational phase transitions
We consider the phase retrieval problem of reconstructing a n-dimensional real or complex signal š^ā from m (possibly noisy) observations Y_Ī¼ = | ā_i=1^n Ī¦_Ī¼ i X^ā_i/ā(n)|, for a large class of correlated real and complex random sensing matrices Ī¦, in a high-dimensional setting where m,nāā while Ī± = m/n=Ī(1). First, we derive sharp asymptotics for the lowest possible estimation error achievable statistically and we unveil the existence of sharp phase transitions for the weak- and full-recovery thresholds as a function of the singular values of the matrix Ī¦. This is achieved by providing a rigorous proof of a result first obtained by the replica method from statistical mechanics. In particular, the information-theoretic transition to perfect recovery for full-rank matrices appears at Ī±=1 (real case) and Ī±=2 (complex case). Secondly, we analyze the performance of the best-known polynomial time algorithm for this problem ā approximate message-passing ā establishing the existence of a statistical-to-algorithmic gap depending, again, on the spectral properties of Ī¦. Our work provides an extensive classification of the statistical and algorithmic thresholds in high-dimensional phase retrieval for a broad class of random matrices.
READ FULL TEXT