FPGA-based muscle synergy extraction for surface EMG gesture classification

To the present day, a multitude of studies aims to understand how the Central Nervous System (CNS) translates neural pulses to muscle motor tasks, through the analysis of surface-Electromyography (sEMG) recordings. One of the most prominent methods applies the Non-Negative Matrix Factorization (NMF) to data recorded from sEMG electrodes to extract coordinated motor patterns, the so-called muscle synergies, which theorize a modular control of the muscles by the CNS. Muscle Synergies can represent a major breakthrough in prosthetics control, since they require a low number of sEMG signals to express the degrees of freedom (DoF), while current prostheses need at least two muscles for every DoF. Starting from this hypothesis, we developed an FPGA-based, real-time NMF processor to extract muscle synergies from an 8-electrode sEMG recording and feed them to a Support Vector Machine classifier which recognizes the specific movement. The FPGA covers a fundamental role, since a classical microprocessor would be too slow to execute the NMF in real time, a CPU too power hungry, and an ASIC too costly to design and produce and without the reconfiguration capabilities of FPGA. Preliminary results show a fast reaction time, with a speed-up of 24.31x versus the on-board ARM processor for the NMF and 9.91x for the SVM, and an overall increase in power efficiency of 3.1x against a Desktop CPU (TDP 65W) and 6.56x on the ARM. The classifier was trained on 5 movements (hand close, thumb close, thumb-index, middle-ring, and middle-ring-little) and obtained an accuracy in the tests around 98%.