Electroencephalogram (EEG) signal is recorded as a multidimensional dataset, which can be interpreted using autoregressive (AR) models. In this way, it is possible to extract relevant information about the signal, which can be used for Brain Computer Inteface (BCI) classification. Yule-Walker equations can be derived from the AR equation, which yields the autocovariance matrix that is a symmetric positive definite matrix (SPD) which is classified using Riemann geometry approach.. We can also think of formulating the problem from another point of view: since the autocovariance matrix is a matrix of delayed covariances, we can obtain the same result by creating an embedding of the original system in a high dimensional space. Thus, it seems natural to connect our approach with the delay embedding theorem proposed by Takens in the context of dynamical systems. Such an embedding method depends on two parameters: the delay parameter and the embedding dimension D, respectively the lag and the order in the context of AR model. We use grid search to find various parameters of the compared approaches. We test our approach on several datasets and compare it to the state of the art.