I work on wearable systems that use muscle electrical activity to control software and measure what muscles are doing during movement. This work brings together surface electromyography (EMG), analog electronics, embedded processing, and machine learning.
It began with MuscleMate, a muscle-controlled interface developed under my supervision, and grew into a textile-integrated sensing platform. My contributions span acquisition and calibration, gesture recognition, contraction analysis, and the integration of those methods into working hardware.
Muscle activity as a control signal
MuscleMate turned muscle contractions into computer inputs. Three EMG channels fed a portable acquisition system, which calculated frequency features and transmitted them over Bluetooth. The software produced on/off commands, a continuous activation output, and gesture labels.
My natural-log transformation exposed distinct activation regions in the frequency features, giving us a useful geometry for muscle-state separation and continuous control. We also moved from a single contraction to a sequence. A recognizer could compare the pattern of activity over time, allowing the same gesture to be matched across different execution speeds. We connected the outputs to PC and Android applications, including video-game and presentation control.
MuscleMate capstone: electronics, logarithmic features, and gesture recognition →
Measuring muscles during movement
The wearable platform brought EMG into a fitted thigh band with textile electrodes and removable electronics. We combined muscle measurements with acceleration so that contraction timing and intensity could be interpreted in the context of walking, running, and cycling.
This required the physical interface and algorithms to work together. Perspiration changed electrode impedance during use. Filtering introduced delay. A contraction detector had to select the correct samples for spectral analysis. These practical details shaped the electronics, calibration experiments, buffering, and real-time processing.
Technical work
- EMG acquisition and calibration
- Instrumentation amplifiers, 24-bit multichannel conversion, embedded interfaces, and impedance measurements during exercise.
- EMG signal processing
- Contraction traces, amplitude envelopes, spectral changes during cycling, blood-lactate comparisons, and heartbeat detection.
- Movement recognition
- EMG–motion feature spaces, activity transitions, double cumulative-sum features, and five-state hidden Markov models.
- Patents
- Four issued US patents covering physiological acquisition, heartbeat detection, exercise metrics, and activity recognition.