Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Computer Engineering

First Advisor

Qing Yang


Presented within this dissertation is the evolution of the research leading to the selection of small, low power architectural solutions to the University of Rhode Island’s (URI) Neural Machine Interface (NMI) algorithm. The NMI is designed to provide volitional control of an artificial limb for transfemoral amputees. The NMI algorithm is based on neuromuscular–mechanical fusion, gait phase dependent, nonlinear support vector machine (SVM) classification. URI’s NMI algorithm utilizes electromyography to detect direct commands from the human brain to the residual thigh muscles in conjunction with mechanical signals derived from loadcell to determine the user’s intended locomotion mode. Of utmost importance is the classification accuracy, since any misclassification can cause the user to stumble, possibly leading to serious injury or death. Furthermore, of importance is the development of a small and low power architectural solution, such that it can be included within the confines of the artificial limb. URI has tackled both these challenges, leading to its mobile Central Processing Unit (CPU) solution. The mobile CPU solution was the first solution with sufficient processing throughput to execute the NMI at 20ms window increments. This led to a steady state classification accuracy of 99.94%, during real-time testing, with an able bodied subject. This testing included a total of 14000+ static classifications, and is currently URI's only, 20ms window increment, state of the art algorithmic and architectural solution to undergo real time human subject testing and evaluation. iii In contrast to URI’s NMI algorithm, other state of the art algorithms provide volitional control through either echo control or solely thru intrinsic mechanical feedback. In echo control, sensors are placed within the sound leg to determine the intended locomotion mode. In most cases these sensors typically communicate wirelessly with the artificial limb to provide the feedback necessary for volitional control. This approach is disadvantaged in the fact that it requires that sensors be instrumented on the sound limb, the user must always lead with the sound limb, and the wireless communications may possibly be jammed. Current algorithms based solely on intrinsic mechanical feedback, have been shown to provide high accuracy, but have had difficulty dealing with more than two simultaneous dynamic locomotion modes (e.g. - walk, stair up, stair down, ramp up, and ramp down). Clearly URI's NMI solution has advantages over other state of the art powered lower limb prosthetic control algorithms. It provides volitional control without the need to instrument the sound limb, without the need of wireless communications, can easily detect at least seven simultaneous locomotion modes, provides smooth and highly responsive locomotion transition detection and does so with high accuracy. This accuracy can be attributed to the use of neuromuscular-mechanical fusion, SVM detection and 20ms window analysis increments. URI's small, low power, architectural solutions are leading the way towards highly accurate volitional artificial leg control of powered prosthetic devices, thereby making a bionic leg a feasible reality in the near future.