Cortical local fields and oscillatory events are usually analyzed by estimating their spectral characteristics, but these approaches have limited ability to extract all the information carried by the signal. Applying dimensionality reduction methods such as principal component analysis can improve classification performance, help discover hidden patterns, and create new features. We aimed to assemble a multi-step analysis workflow that can transform oscillatory LFP activity into a statistical representation. After filtering and downsampling the LFP signal, waveform segments for the events of interest were collected. The data was projected from the original space to the low-dimensional principal component space. After that, a Self-Organizing Map was trained and used to cluster the segments. Finally, a 2D probability distribution (SOM profile) was calculated for related LFP segments using cluster labels. We applied the workflow to cortical slow wave, theta, and spindle oscillations recorded from the juxtasomal position of pyramidal cells and interneurons in freely moving rodents. The application of the workflow on the juxtacellular LFP events data set recorded near pyramidal cells (n=65), regular spiking (n=42), and fast-spiking (n=33) interneurons (n > 34000 down state events) revealed that the down states express a significant difference in their SOM profiles depending on the type of recorded cell. We conclude that juxtacellular LFP convey cell-type specific information. This suggests that field potentials in the network can be highly compartmentalized and can retain identities of cellular units in space and time, even if neuronal populations are in a silent state.