Electroencephalography(EEG) and magnetoencephalography (MEG) measure the electrical activity of the functioning brain using a set of sensors placed on the scalp (electrodes and magnetometers). Magneto- or electroencephalography (M/EEG) have the same biological origin, the activity of the pyramidal neurones within the cortex. The signals obtained from M/EEG are very noisy and inherently multi-dimensional, i.e. provide a vector of measurements at each single time instant. To cope with the noise, researchers, traditionally acquire measurements over multiple repetitions (trials) and average them to classify various patterns of activity. This is not optimal because of trial to trial variability. The jitter-adaptive dictionary learning method (JADL) [1] has been developed to better handle for this variability. JADL is a data-based method that learns a dictionary from a set of signals, but is currently limited to a single channel, which restricts its capacity with very noisy data such as M/EEG. In this paper, we propose an extension to the jitter-adaptive dictionary learning method, in order to handle multidimensional measurements such as M/EEG. A modified model is developed and tested using synthetically generated data set as well as real M/EEG signals. The results obtained using our model look promising, and show superior performance compared to the original single-channel JADL framework.