In this paper, we introduce a novel method for constructing synthetic, but realistic, data of four Electroencephalography (EEG) channels. The data generation technique relies on imitating the relationships between real EEG data spatially distributed over a closed-circle. The constructed synthetic dataset establishes ground truth that can be used to test different source separation techniques. The work then evaluates three projection techniques – Principal Component Analysis (PCA), Independent Component Analysis (ICA) and Canonical Component Analysis (CCA) – for source identification and noise removal on the constructed dataset. These techniques are commonly used within the EEG community. EEG data is known to be highly sensitive signals that get affected by many relevant and irrelevant sources including noise and artefacts. Since we know ground truth in a synthetic dataset, we used differential evolution as a global optimisation method to approximate the “ideal” transform that need to be discovered by a source separation technique. We then compared this transformation with the findings of PCA, ICA and CCA. Results show that all three techniques do not provide optimal separation between the noisy and relevant components, and hence can lead to loss of useful information when the noisy components are removed.