The high Peak-to-Average Power Ratio (PAPR) of Orthogonal Frequency Division Multiplexing (OFDM) signals leads to a serious system performance degradation. To work around this issue, several algorithms have been proposed in the literature to reduce the PAPR, but, they often suffer from multiple limitations; in particular, the main issue with interleaving techniques is the spectral efficiency loss, as the transmission of a Side Information (SI) is generally required. In contrast to previous works, this paper proposes a blind interleaving technique for OFDM systems with signal space diversity. Indeed, with Rotated and Cyclically Q-Delayed (RCQD) constellations, the In-phase (I) and Quadrature (Q) components of constellations symbols are correlated, which allows the receiver to estimate the interleaver index without any SI. Moreover, to lower down the complexity burden at the receiver side, we first design a blind decoder based on the Minimum Mean Square Error (MMSE) criterion and we then propose a low complexity decoder for the Uniformly Projected RCQD (UP-RCQD) QAM, as this constellation has several interesting structural properties and achieves near optimum BER performance. Simulation results show that our proposal leads to a large PAPR reduction and to a near optimum BER performance that outperforms, over various channels, the solution currently used in DVB-T2. They also underline the good performance of the blind decoding performed with up to 98% of complexity reduction compared to the max-log Maximum Likelihood (ML) estimation.