这篇文章的实验范式和我们的实验范式很像。
Notes
为了提高分解性能，观测向量y(n)可以进行K个延迟。
The method has been tested in many different experimental setups [9], [10], [16]–[21], [29], [31], [41], [45], [47] also in moderate dynamic contractions [8], [11], [12], [26], [40] and yielded up to 70 simultaneously active MUs.
Experiments
- All the subjects performed wrist flexions-extensions. Six subjects performed also pronations-supinations and four subjects performed ulnar-radial deviation.
- The subjects followed the trapezoidal (2 s ramp up + 5 s plateau + 2s ramp down + 5 s rest) force profiles. In each measurement, we recorded ten repetitions of the selected movement.
- The hdEMG signals were recorded by two arrays of 5×13 electrodes with diameter of 1 mm and inter-electrode distance of 8 mm (OT Bioelettronica, Italy).
Data analysis
- 取30s作为一段，对每一段进行分解，然后把这些分解向量作用总的120s，再移除重复的MU，选择PNR高的MU。
- 对每个MU进行动作的归类，比如flexion相关的MU归为G1，extension相关的MU归为G2，剩下的归为G3。
- 肌肉兴奋是非负性的，非常符合NMF的假设。这意味着每个DOF需要估计至少两个分量F，即每个运动方向一个分量。
- 没有一种测试的分解技术能够完全补偿由于动态肌肉收缩引起的MUAP变化，尽管在研究手腕运动的情况下，这些变化相对较小，如前面[8]所述。