In this paper, we investigate how secure the TMA OFDM system is, by looking at the transmitted signal from an the viewpoint of eavesdropper. First, we propose a novel, low-complexity scheme via which the eavesdropper could defy the scrambling in the received signal and recover the transmitted symbols. We show that the symbols which the eavesdropper sees along the OFDM subcarriers are linear mixtures of the source symbols, where the mixing coefficients are unknown to the eavesdropper. Independent component analysis (ICA) could be used to obtain the mixing matrix but there would be permutation and scaling ambiguities. We show that these ambiguities can be resolved by leveraging the structure of the mixing matrix and the characteristics of the TMA OFDM system. In particular, we construct a k-nearest neighbors (KNN)-based algorithm that exploits jointly the Toeplitz structure of the mixing matrix, knowledge of data constellation, and the rules for designing the TMA ON-OFF pattern to resolve the ambiguities. In general, resolving the ambiguities and recovering the symbols requires long data. Specifically for the case of the constant modulus symbols, we propose a modified ICA approach, namely the constant-modulus ICA (CMICA), that provides a good estimate of the mixing matrix using a small number of received samples. We also propose measures which the TMA could undertake in order to defend the scrambling. Simulation results are presented to demonstrate the effectiveness, efficiency and robustness of our scrambling defying and defending schemes. Complete abstract please see in the paper.