Optimization of the Sampling Window for Enhanced Myoelectric Prosthetic Hand Control

Lia Westermann

John and Marcia Price College of Engineering

21 Optimization of the Sampling Window for Enhanced Myoelectric Prosthetic Hand Control

Lia Westermann

Faculty Mentor: Greg Clark (Biomedical Engineering, University of Utah)

Abstract

Current myoelectric prosthetic hands exhibit limited, unnatural, and often inaccurate kinematics. To enhance their overall capability, this study focuses on optimizing the Control Command component of prosthetic operation. We hypothesize that the sampling window at which surface electromyographic (sEMG) data is processed significantly influences the signal-to-noise ratio (SNR), thereby modulating the accuracy and responsiveness of myoelectric prosthetic hand (MPH) movement. Using MATLAB, we analyzed identical sEMG signals across seven different sampling windows. The results demonstrate that the selected sampling window has a pronounced effect on the SNR of the data, directly impacting the reliability and smoothness of prosthetic actuation. Optimization of this parameter facilitates more precise biomimetic control of MPHs, enhancing dexterity and functional performance.

Introduction

Hand absence due to amputation is typically caused by trauma, disease, or congenital anomalies, and can significantly impact daily activities while diminishing quality of life [1,2]. A common solution for restoring hand function is the MPH, which is controlled by sEMG signals from residual muscle groups [3]. However, current MPHs perform at less than 50% of a human hand’s capabilities [4], and available options lack necessary functionality [5]. Despite being the most advanced and promising prostheses, MPHs remain difficult to control and often fail to meet user needs, leading to widespread dissatisfaction and device rejection [6,7].

The goal of MPHs is to mimic the human hand with a combination of dexterity, reliability, and force control [8,9]. To achieve this, the sEMG signals are processed in a prosthetic control system component known as Control Command, which involves sampling the sEMG signals within a defined sampling window [10]. Although other control parameters have been extensively researched, the optimal sampling window for sEMG signals remains inadequately studied. Control Command is a sensitive part of the system, where improper sampling can degrade performance. This study hypothesizes that the sampling window significantly affects the SNR, thereby determining the function and control capabilities of MPHs. By optimizing the sampling window, we aim to enhance the performance and kinematics of MPHs, reduce user discomfort, improve quality of life, and alleviate phantom limb pain through device acceptance.

Background

Approximately 30 million people worldwide live with limb loss, including hand absence due to amputation [1]. In the United States, around 5,000 to 6,000 major limb amputations occur annually, with one-fifth of these involving the upper limb, most commonly at the trans-radial level [2]. Amputations are caused by traumatic events, congenital anomalies, cardiovascular disease, infection, and nerve injury [3-4]. The human hand is one of the most complex organs in the body, second only to the brain, due to the more than thirty muscles in the hand and forearm that control its movement.

Robotics prostheses, including hand prostheses, significantly improve both the function and quality of life for individuals with amputations, offering psychological benefits as well [5]. MPHs are control systems based on the Peripheral Nervous System that use electrical signals from the residual muscles to generate movement [6]. While MPHs are the most functional prosthetic option to date, the complexity of the hand makes control challenging, limiting their effectiveness [7]. The primary objective of research in this field is to advance biomimetic control systems that replicate natural hand movement.

Methods

Hardware and Software

The MPH attaches to residual muscles with 32 electrodes that measure sEMG signals from the user. The electrodes amplify the signals with minimal filtering. The electrodes feature a low-pass cutoff frequency of 1500 Hz to reduce noise when the user is not actively moving their hand. The electrodes are connected to wires, which link to multiple computers running LabView (a graphical programming environment) and MATLAB (a high-level programming language) for signal processing. The LabView interface incorporates 528 channels, representing degrees of freedom, by processing data from all 32 electrodes and their inter-electrode comparisons.

Training with the MPH

To effectively use the MPH, individuals must first undergo a training session to optimize the device’s performance based on their muscle movements as detected by the sEMG signals. This training typically lasts 4-5 hours and involves performing specific hand and finger movements at varying training speeds. Upon completion, the user’s sEMG data from the entire training session is saved as a KDF file for subsequent filtering, sampling, and noise reduction in MATLAB.

EMG Data Processing

All sEMG data processing was performed using MATLAB. To analyze the effect of different sampling windows, four distinct MATLAB scripts and functions were developed. The data was processed using for-loops for each training speed and each sampling window. The tested sampling windows included 40, 100, 125, 150, 200, 250, and 300 milliseconds (ms).

Results

We examined window sizes ranging from 40-300 ms, while previous research has shown that sampling windows outside the 40-300 millisecond range can degrade MPH performance. Using MATLAB, we plotted the sEMG data for each of these seven sampling windows and analyzed the amount of noise to determine which window maximized the SNR. The sEMG data sampled at a window of 100-125 ms produced the greatest SNR.

Figure 1: EMG signal in millivolts over time in seconds when the EMG data is sampled at 40 millisecond intervals shows high noise levels. — **Figure 1.** sEMG signal is sampled at a window of 40 ms.

Figure 2: EMG signal in millivolts over time in seconds when the EMG data is sampled at 100 millisecond intervals shows low noise levels. — **Figure 2.** EMG signal in millivolts over time in seconds when the EMG data is sampled at 100 millisecond intervals shows low noise levels.

Figure 3: sEMG signal is sampled at a window of 125 ms. — **Figure 3.** sEMG signal is sampled at a window of 125 ms.

Figure 4: sEMG signal is sampled at a window of 150 ms. — **Figure 4.** sEMG signal is sampled at a window of 150 ms.

Figure 5. sEMG signal is sampled at a window of 200 ms. — **Figure 5.** sEMG signal is sampled at a window of 200 ms.

Figure 6: sEMG signal is sampled at a window of 250 ms. — **Figure 6.** sEMG signal is sampled at a window of 250 ms.

Figure 7: sEMG signal is sampled at a window of 300 ms. — **Figure 7**. sEMG signal is sampled at a window of 300 ms.

Discussion

This study demonstrated that the sampling window at which sEMG data is processed significantly influences the SNR, and consequently, the control performance of MPHs. Optimizing this parameter can improve the user’s ability to control the MPH due to background noise being reduced, enhancing functionality, reducing phantom limb pain with greater acceptance of the MPH, and improving quality of life.

We analyzed sEMG data using sampling windows ranging from 40 ms to 300 ms, based on values reported in prior studies on the Control Command portion of prosthetic control systems. The key method was to create MATLAB scripts and functions to run the sEMG signal through every sampling window. As shown in Figures 1-7, variations in the window size directly affected the noise level within the sEMG signals, confirming that sampling window plays a crucial role in determining control accuracy and system performance.

Previous research on MPHs has largely focused on pattern recognition, accurately identifying the user’s intended motion, and feature extraction to improve motion classification accuracy. In contrast, our work focused specifically on the effect of sampling windows on signal quality and device control. While prior studies narrowed the optimal sampling window to 40-300 ms, we further refined it to between 100 and 125 ms.

This research faced several limitations. The primary limitation was the use of sEMG data from a single participant, due to the extensive time required for prosthetic training. Expanding the sample size in future studies would help validate the generalizability of the results. Data processing speed was another limitation. Each MATLAB analysis run required significant computational time due to the volume of data processed. Additionally, although data was collected for four training speeds, only one was analyzed in detail. Future research could investigate how different training speeds interact with sampling windows across multiple users, aided by improved hardware and software for faster data processing.

Overall, this study highlights the importance of selecting an appropriate sEMG sampling window to enhance MPH performance. Proper sampling can strengthen the signal while reducing unwanted noise, improving overall control precision. The sampling windows identified here could inform other prosthetic systems or be adapted to individual users for personalized performance optimization.

Future work will involve real-time testing with MPH users to evaluate the practical effectiveness of these findings. By improving sEMG signal quality and control accuracy, this research contributes to the development of more reliable and functional MPHs. These advances address users’ key concerns of comfort and function, ultimately improving quality of life and reducing phantom limb pain.

Bibliography

Asogbon M, Chen S, Fang P, Geng Y, Li G, Mzurikwao D, Samuel O. Effect of Window Conditioning Parameters on the Classification Performance and Stability of EMG-Based Feature Extraction Methods. 2018 IEEE International Conference on Cyborg and Bionic Systems (CBS) 2018: 576-580.

Atzori M, Bassetto F, Caputo B, Castellini C, Elsig S, Gijsberts A, Giatsidis G, Hager A, Muller H. Electromyography data for non-invasive naturally-controlled robotic hand prostheses. Scientific Data 1 2014;1(140053).

Barone R, Ciancio A, Cordella F, Romeo R, Bellingegni A, Sacchetti R, Davalli A, Di Pino G, Ranieri F, Di Lazzaro V, Guglielmelli E, Zollo L. Control of Prosthetic Hands via the Peripheral Nervous System. Frontiers in Neuroscience 2016;10(116):1-17.

Ciancio A, Cordella F, Barone R, et al. Control of Prosthetic Hands via the Peripheral Nervous System. Frontiers in Neuroscience 2016;10(116):1-17.

Dewald H, Williams M, Anderson J, Lukyanenko P, Lambrecht J, Tyler D, Kirsch R. Stable, three degree-of-freedom myoelectric prosthetic control via chronic bipolar intramuscular electrodes: A Case Study. Journal of NeuroEngineering and Rehabilitation 2019;16(147):1-13.

Eisenberg G, Fyvie K, Mohamed A. Real-Time Segmentation and Feature Extraction of Electromyography: Towards Control of a Prosthetic Hand. International Federation of Automatic Control 2017;50(2):151-156.

Farrell T, Weir R. The Optimal Controller Delay for Myoelectric Prostheses. IEEE Transactions on Neural Systems and Rehabilitation Engineering 2007;15(1):111-118.

Fougner A, Stavdahl O, Kyberd P, Losier Y, Parker P. Control of Upper Limb Prostheses: Terminology and Proportional Myoelectric Control – A Review. IEEE Transactions on Neural Systems and Rehabilitation Engineering 2012;20(5):663-667.

Fu Q, Santello M. Improving Fine Control of Grasping Force during Hand-Object Interactions for a Soft Synergy-Inspired Myoelectric Prosthetic Hand. Frontiers in Neurorobotics 2018;11(71):1-15.

Huinink L, Bouwsema H, Plettenburg D, Bongers R, Sluis C. Learning to use a body-powered prosthesis: changes in functionality and kinematics. Journal of neuroengineering and rehabilitation 2016;13(90).

Khushaba R, Kodagoda S, Takruri M, and Dissanayake G. Toward improved control of prosthetic finders using surface electromyogram (EMG) signals. Expert Systems with Applications 2012;39(12):10731-10738.

Li W, Shi P, Yu H. Gesture Recognition Using Surface Electromyography and Deep Learning for Prostheses Hand: State-of-the-Art, Challenges, and Future. Frontiers in Neuroscience 2021;15(621885).

Mastinu E, Clemente F, Sassu P, Aszmann O, Branemark R, Hakansson B, Controzzi M, Cipriani C, Ortiz-Catalan M. Grip control and motor coordination with implanted and surface electrodes while grasping with an osseointegrated prosthetic hand. Journal of NeuroEngineering and Rehabilitation 2019;16(49)16:49.

Menon R, Di Caterina G, Lakany H, Petropoulakis L, Conway BA, Soraghan JJ. Study on Interaction Between Temporal and Spatial Information in Classification of EMG Signals for Myoelectric Prostheses. IEEE Transactions on Neural Systems and Rehabilitation Engineering 2017;25(10):1832-1842.

Noce E, Bellingegni A, Ciancio A, Sacchetti R, Davalli A, Guglielmelli E, Zollo L. EMG and ENG-envelope pattern recognition for prosthetic hand control. Journal of Neuroscience Methods 2019; 311:38-46.

Pasquina P, Perry B, Miller M, Ling G, Tsao J. Recent Advances in Bioelectric Prostheses. Neurology Clinical Practice 2015;5(2):164-170.

Phinyomark A, Phukpattaranont P, Limsakul C. Feature reduction and selection for EMG signal classification. Expert Systems with Applications 2012;39(8):7420-7431. Shi. J, Chang Q, Zheng Y. Feasibility of controlling prosthetic hand using sonomyography signal in real time: Preliminary study. Journal of Rehabilitation Research & Development 2010;47(2):87-98.

Wang Y, Tian Y, She H, Jiang Y, Yokoi H, Liu Y. Design of an Effective Prosthetic Hand System for Adaptive Grasping with the Control of Myoelectric Pattern Recognition Approach. Micromachines 2022;13(2):219.