Here we propose and evaluate a method for segmenting the Linear Prediction Matrix used during the Matrix Pencil Method. This proposal considers the division of a time domain signal in <tex>$W$</tex> frequency segments that can be processed individu-ally and whose results can be put together in the final stage. The evaluation of the suggested approach is presented here in terms of the coefficient of determination <tex>$R^{2}$</tex> for the reconstructed signal, and focuses on either reducing the hardware requirements for back-scattering signal processing, its parallelisation or reduced execution time for portable GPR operations.