Browsing by Author "Farrell, Todd"

Now showing 1 - 3 of 3

Open Access
ANALYSIS WINDOW INDUCED CONTROLLER DELAY FOR MULTIFUNCTIONAL PROSTHESES
(2008) Farrell, Todd; Weir, Richard F.
Many upper-limb, multifunctional prosthesis controllers analyze fixed segments of EMG data collected from the residual musculature in an attempt to discern the intended movement of the user. However, many researchers have designed controllers with little or no regard for the delay the controller will introduce when operated in real-time. If the delay is too large the prosthesis will feel sluggish and performance will suffer. Several attributes of the classifier affect the delay it will create. State-based pattern recognition classifiers typically collect EMG data in ‘analysis windows’ whose length will be defined as Ta. Class decisions based upon these collected data cannot be generated instantaneously because time is required to both record and then process the EMG. The processing time ( ) is the time from the completion of data collection until a class decision is made. The length of the window being analyzed (Ta), the microprocessor used to perform the calculations, as well as the number of channels and the number and type of features being extracted will determine . Thus must be determined empirically for each classifier. Both overlapped and disjoint analysis windows have been employed in experimental prosthesis controllers. Windows can be overlapped if the analysis windows are incremented by some amount of time (Tnew) that is greater than the processing time ( ). Overlapping the windows increases the density of class decisions which will allow majority voting. Majority voting is a post-processing strategy that has been shown to increase classifier accuracy [1-2] by analyzing the current class decision along with the n-1 previous class decisions and selecting the class that occurs most frequently in those n decisions as the controller output. The authors recently completed a study which found that 100 ms was the maximum amount of time that could be used to collect and analyze EMG signals (to maximize the classification accuracy) without substantially degrading the performance of the prosthesis [3]. This finding implies that the values of Ta, Tnew, n and should be set to ensure that the amount of time from the user’s intended change in class until the change in the output of the controller (i.e., the controller delay or ‘D’) is less than 100 ms. The goal of this work is to quantitatively define how each parameter (Ta, Tnew, n and ) affects the maximum delay as well as the range of delays introduced by the controller. Four controller configurations were examined including those that use overlapped or disjoint windows as well as those that did or did not use majority voting. Note: the data are collected with a sampling period of Ts and a frequency of 1/Ts Hz.
Open Access
REAL-TIME COMPUTER MODELING OF A PROSTHESIS CONTROLLER BASED ON EXTENDED PHYSIOLOGICAL PROPRIOCEPTION (EPP)
(2002) Farrell, Todd; Weir, Richard F.; Heckathorne, Craig W.; Childress, Dudley S.
Proprioception utilizes the physiological components of the nervous and musculoskeletal systems to allow an individual to sense the position of their limbs subconsciously. By providing a rigid connection to an object this proprioceptive ability can be extended to the object and allow the user to sense the spatial location and orientation of these objects with respect to his or her body. This concept explains how a person can use a tennis racquet to hit a tennis ball without having to observe the position of the racquet during their swing or the way a blind person uses a long cane to ‘feel’ the location of objects in their surroundings. Body-powered prostheses take advantage of this proprioceptive ability by relating the motion and position of the prosthesis to the motion and position of an intact joint of the amputee via the control cable. However, most externally powered prostheses do not have any mechanism with which to provide feedback regarding the state of the prosthesis to the proprioceptive system of the amputee. In these cases the amputee must rely on vision and other incidental sources of feedback such as motor whine and socket pressure to control their prostheses and this may place a significant cognitive load on the user.
Open Access
THE EFFECTS OF ELECTRODE IMPLANTATION AND TARGETING ON PATTERN CLASSIFICATION ACCURACY FOR PROSTHESIS CONTROL
(2008) Farrell, Todd; Weir, Richard F.
Many researchers have attempted to recognize patterns of muscle activity associated with different movements of the phantom limb and link these patterns to movements of the prosthesis. Researchers have examined a variety of different classifiers and extracted complex features from the electromyographic (EMG) signals to maximize classification accuracy. However, nearly all of these efforts used surface electrodes. Surface electrodes are advantageous because they are cheap, non-invasive and have a large pickup area. Extracting features from these recordings can allow the classifier to parse out the activity from the different muscles that sum together to produce the myoelectric signal and may increase the information available to the classifier. Alternatively, intramuscular electrodes may be advantageous for multifunctional prosthesis control because they record focally from deep muscles, provide consistent recording sites as the user changes arm orientation or dons and doffs the prosthesis and reduce crosstalk. However, only two groups have investigated intramuscular EMG for pattern recognition based control [1- 4] and only Hargrove, et al. [1] compared surface and intramuscular electrodes, recording from sixteen untargeted surface and six targeted intramuscular channels. As well as almost solely utilizing surface electrodes, previous studies in pattern recognitionbased multifunctional prosthesis control have either targeted the electrodes to specific muscles or used untargeted electrode arrays. However, no previous work has attempted to determine which approach is superior by directly comparing targeted and untargeted electrodes. Untargeted electrodes are simpler to implement and are preferable for both intramuscular and surface recordings. Socket fabrication can be simplified if the surface electrodes only need to be arranged in an array instead of targeted to specific muscles. Additionally, targeting implantable sensors (such as the IMES [5]) to specific muscles is not a trivial task and would likely require approaches such as ultrasound guidance to properly orient the implants in specific muscle bellies. Given that the effect of either electrode targeting or electrode implantation has rarely been examined, the goals of this work were to compare the classification accuracies of multifunctional prosthesis classifiers that use either surface or intramuscular EMG as well as those that use either targeted and untargeted electrodes. Further details are available in Farrell and Weir [6].