Time delay signature (TDS) is a key issue that affects the security performance of optical chaos communication. Currently, the autocorrelation function (ACF) and delayed mutual information (DMI) are two crucial methods to extract the TDS. Several strategies have been proposed to resist statistical analysis (i.e., ACF and DMI). However, a dynamic inverse modeling method has been proposed to identify TDS by utilizing the strong nonlinear inversion capability of deep learning, which brings about new threat. To resist this analysis, we propose a stratagem that one can increase the nonlinear dimensionality of the chaos system at the transmitter and concurrently reduce the dimensionality of transmitted signals, leading to “deficiency of data dimensionality”. A three-dimensional chaotic system is designed as the transmitter and only one-dimensional signal is transmitted for communication. Furthermore, a hybrid receiver is designed. The nonlinear model of the response system is realized by cooperation of an analog electro-optical link and a neural network in digital domain. The architecture not only destroys the possibility of TDS extraction but also preserves all necessary information required for chaos synchronization with low-complexity implementation.