Based on the signal reconstruction theory of linear equation groups, a novel multi-carrier technology called frequency division channel utilization (FDCU) is proposed, aimed at effectively enhancing channel capacity. Unlike other multi-carrier signals such as OFDM, the harmonics of FDCU signals are not fully orthogonal mathematically, which renders traditional Fourier signal processing methods unsuitable for their demodulation. An FDCU signal comprising three frequency components was successfully demodulated using a back-propagation (BP) neural network, demonstrating the effectiveness of BP neural networks in processing non-orthogonal signals. This approach provides a new perspective on signal application and processing methods in fields such as electronic communications, image processing, and remote sensing.