Robust detection of below-lesion, non-invasive electromyographic signals

In the majority of people experiencing "complete" spinal cord injury (total or near total para or quadraplegic) some small percentage of axons survive across the site of injury (lesion). Although the response from this small population of neurons is insufficient to facilitate active motor functions, neural impulses are potentially transmitted, under volitional control, below the level of the spinal cord injury. The ability to use these impulses to augment residual muscle functions, control the environment, or to reinforce the learning of new actions, would have profound implications on the rehabilitation and social reintegration of people with spinal cord or other neurological injuries. This investigation develops signal processing techniques for subthreshold electromyographic (EMG) signals, collected non-invasively (without breaking the skin) and below-lesion (below the spinal cord injury). Such signals are buried deep in noise, but the signal (motor unit action potential trains) and noise (spurious neuronal firings and biological interference) are shown to be separable using nonlinear techniques for which the usual signal to noise ratio is irrelevant. Then, controllers based on these techniques will enable the use of weak neural impulses to perform functions necessary for self reliant existance, and will lead to development of prosthetic/orthotic control methodologies.