Hardware-oriented Memory-limited Online Fastica Algorithm and Hardware Architecture for Signal Separation

This paper presents a hardware-oriented memory-limited online FastICA algorithm and its hardware architecture and implementation for eight-channel electroencephalogram (EEG) signal separation. The online algorithm integrates the data overlapping, garbage detection, channel permutation, and momentum-controlled weight update schemes to stabilize the order of the decomposed source signals across time. This study also realizes the algorithm into a hardware architecture and implementation with a core area of 1.469x1.469 mm2 in a TSMC 90 nm process. The resulting power dissipation for eight-channel EEG signal separation is 65 mW@100 MHz at 1V.