Stationary Wavelet Processing and Data Imputing in Myoelectric Pattern Recognition on a Low-Cost Embedded System

Pattern recognition-based decoding of surface electromyography allows for intuitive and flexible control of prostheses but comes at the cost of sensitivity to in-band noise and sensor faults. System robustness can be improved with wavelet-based signal processing and data imputing, but no attempt has been made to implement such algorithms on real-time, portable systems. The aim of this work was to investigate the feasibility of low-latency, wavelet-based processing and data imputing on an embedded device capable of controlling upper-arm prostheses. Nine able-bodied subjects performed Motion Tests while inducing transient disturbances. Additional investigation was performed on pre-recorded Motion Tests from 15 able-bodied subjects with simulated disturbances. Results from real-time tests were inconclusive, likely due to the low number of disturbance episodes, but simulated tests showed significant improvements in most metrics for both algorithms. However, both algorithms also showed reduced responsiveness during disturbance episodes. These results suggest wavelet-based processing and data imputing can be implemented in portable, real-time systems to potentially improve robustness to signal distortion in prosthetic devices with the caveat of reduced responsiveness for the typically short duration of signal disturbances. The trade-off between large-scale signal corruption robustness and system responsiveness warrants further studies in daily life activities.