Skip to main content

Table 3 Overview of the used input signals and applied processing

From: EMG-driven control in lower limb prostheses: a topic-based systematic review

Ref. Muscles Additional sensors and feedback EMG signal processing (filter order and cut-off frequencies) Window sampling Classifier (features)
[64] Single muscle not contracting during walking NP NI Analog Thresholding (Raw Sign.)
[29] GRAC Load cell on the knee pivot\(^{{\star } {\dag }}\); footswitch\(^{{\star } {\dag }}\); RECT, BPF (NS) Analog NA (ENV)
[24,25,26,27] Knee FLEX (RFEM/VASI/ VASL), knee EXT (SEMT) Goniometer\(^{\star }\) BPF (20–500 Hz) 115–192 ms LM Network (HIST, ARC)
[56] HAMS, QUAD Joint encoder\(^{\dag }\) (visual feedback of knee joint position) HPF (1st ord, 20 Hz), RECT, LPF (1st ord, 2 Hz), NORM, PCA 2 ms QDA (ENV)
[14, 23, 62, 63] VASL, VASM, RFEM, SEMT, BICFL Joint encoder\(^{\dag }\) (visual and haptic feedback of knee joint position) BPF (2nd ord, 20–450 Hz), RECT, LPF (2nd ord, 2.5 Hz), NORM; PCA 3 ms NA (ENV)
[57] SEMT, SAR, TFL, ADDL, GRAC, RFEM, VASM, VASL, BICFL Joint encoder\(^{\dag }\) (visual feedback of knee joint position) NS 200 ms LDA (MAV, NZC, SSC, WL)
[37] GASM/SOL, TIBA Joint encoder\(^{\star }\) BPF (2nd ord, 10–500 Hz), RECT, LPF (2nd ord, 10 Hz) 50 ms NARAX (ENV)
[16] GASM, TIBA NP RECT, LPF (3rd ord, 2.5 Hz); PCA 10 ms NA (MAV)
[67, 68] GASM/GASL MoCap system\(^{\star }\); sensorized treadmill\(^{\star }\); control signal visual feedback\(^{{\star } {\dag }}\) FPF (2nd ord, 100 Hz), RECT, LPF (2nd ord, 4 Hz), NORM NS NA (ENV)
[76] VASM, BICFL, TIBA, GASL Ankle goniometer\(^{\star }\), accelerometer\(^{\star }\) RECT, LPF (NS) NS Peak detector (ENV)
[11] VASL, RFEM, SEMT, BICFL MoCap system \(^{\star }\) NORM, BPF (4th ord, 30–350 Hz), RECT, LPF (4th ord, 6 Hz), Kalman filter < 500 ms NA (ENV)
[39, 40] TIBA, GASM Joint position\(^{{\star } {\dag }}\) (visual feedback) HPF (2nd ord, 20 Hz), RECT, LPF (2nd ord, 2 Hz) 10 ms NA (ENV)
[98] GLMAX, GLMED, TFL Footswitch\(^{{\star } {\dag }}\) LPF (2nd ord, 1 KHz), HPF (3rd ord, 50 Hz), NORM NS Thresholding (ENV)
[70] VASL, VASM, RFEM, TFL, ADDL, BICFL, SEMM, SEMT Footswitch\(^{\star }\); load cell\(^{\star }\); MoCap system\(^{\star }\) NS NS Heuristic Tree (IDE, MAV, MDF, MF)
[5] TIBA, GASM, GASL Footswitch\(^{{\star } {\dag }}\); potentiometer\(^{{\star } {\dag }}\); encoder\(^{{\star } {\dag }}\); class of movement performed\(^{\star }\) (visual feedback) HPF (1st ord, 16 Hz), LPF (2nd ord, 300 Hz) 100 ms STD
[65, 66, 141, 144] SEMT, SAR, TFL, ADDL, GRAC, VASM, RFEM, VASL, BICFL Footswitch\(^{\star }\), load cell\(^{{\star } {\dag }}\) BPF (20–420 Hz) 150 ms LDA-SVM (MAV, ZCN, WL, SSC, MEC)
[53] RFEM, BICFL, SEMT, GASM, SOL 2× 6-axis force sensor\(^{\star }\), footswitch\(^{\star }\), joint encoder\(^{\star }\) RMS, BPS (20–500 Hz) NS SVM (NS)
[85] VASM, SEMT, ADDL, TFL 2× IMU\(^{\star }\) NS 200 ms Hidden Markov model (MAV, WL, ZC, SSC)
[15, 110] VASM, ADDL, TFL, SEMT MoCap system\(^{\star }\) NS NS SVM (MAV, VAR, MDF, MPF)
[32, 33, 143] SAR, RFEM, VASL, VASM, BICFL, BICFS, SEMT, TFL, ADDL, GRA 6-axis load cell\(^{{\star } {\dag }}\), 2× IMUs BPF (20–420 Hz) 150–160 ms LDA/SVM (MAV, WL, ZCN, SSC, MEC)
[86, 87] TIBA, GASL, BICF, VASL Footswitches\(^{{\star } {\dag }}\); IMU\(^{{\star } {\dag }}\) BPF (4th ord, 20–500 Hz) 100–300 ms LDA/SVM (MAV, VAR, WL, ZCN, SSC)
[120] TIBA, PERL, GASL, GASM, VASM, VASL, RFEM, BICFL Class of movement performed\(^{\star }\) (visual feedback) BPF (20–450 Hz) 250 ms LDA (MAV, ZCN, SSC, WL, ARC)
[58, 111] BICF, RFEM, VASL, VASM, SAR, GRAC, ADDL, TFL + HAMS reinnervation Load cell\(^{{\star } {\dag }}\), 2× joint encoders\(^{{\star } {\dag }}\), 2× IMU\(^{{\star } {\dag }}\) BPF (20–450 Hz) 250 ms same as [140] + Euristic FSM (MEC)
[59, 113, 114, 139, 140] SEMT, ADDL, TFL, RFEM, BICFL, SAR, GRAC, VASL, VASM Load cell\(^{{\star } {\dag }}\), 2× joint encoders\(^{{\star } {\dag }}\), 2× IMU\(^{{\star } {\dag }}\) BPF (20–450 Hz) 250–300 ms LDA/DBN (MAV, WL, ZCN, SSC, 6ord ARC, MEC)
[17] GASM, TIBA Load cell\(^{{\star } {\dag }}\), joint angle encoder\(^{{\star } {\dag }}\), 2× IMU\(^{{\star } {\dag }}\) PCA 200 ms LDA (MAV)
[106] GASL/SOL NP NI NS Thresholding (Raw Sign.)
[54, 55] FIBL, BICF NP BPF (4th ord, 20–500 Hz) 256 ms LDA/SVM/NN (38 mixed domain features)
[6] GASL, SOL, TIBA Joint angle\(^{{\star } {\dag }}\) (visual feedback) LPF (7th ord, 5 Hz), NORM \(\sim\) 1 ms NA (ENV)
[61] VASL, BICFL NP BPF (20–450 Hz), RECT, LPF (5/10 Hz), NORM NS NA (ENV)
[135, 136] HAMS, QUAD Load sensors\(^{{\star } {\dag }}\), goniometer\(^{\star }\) BPF (7th ord, 20–1000 Hz), RECT, LPF (5 Hz) NS NA (ENV)
[71, 72, 130] GASM, TIBA Joint encoder\(^{{\star } {\dag }}\), torque sensor\(^{{\star } {\dag }}\), 6× load cells\(^{{\star } {\dag }}\) HPF(4th ord, 80 Hz), RECT, AVR 150–200 ms NA (ENV)
[19] VASM, VASL, RFEM, BICF, SEMT MoCap system\(^{\star }\), sensorized treadmill\(^{\star }\), 2× IMU\(^{{\star } {\dag }}\) BPF (2nd ord, 30–300 Hz), RECT, LPF (2nd ord, 6 Hz), NORM 300 ms NA (ENV)
  1. Fields include: paper reference; muscles (input EMG muscle signals employed by the controller); additional sensors and feedback (additional sensor signals employed by the controller and possible feedback provided to the user); EMG signal processing (the sequential processing applied to the input EMG signals; in case of filters, order and cut-off frequencies are included); window sampling (the window length used for the processing and features extraction); classifier (classifier types used on the processed signals and features)
  2. NP not present, NI not implemented, NS not specified, NA not applicable, GRAC gracilis, HAMS hamstring muscles, QUAD quadriceps muscles, VASL vastus lateralis, VASM vastus medialis, RFEM rect femoris, TFL tensor fasciae latae, ADDL adductur longus, BICFL biceps femoris long arm, BICFS biceps femoris short arm, SEMM semimembranosus, SEMT semitendineus, SAR sartorius, GASM gastrocnemius medialis, GASML gastrocnemius lateralis, SOL soleus, TIBA tibialis anterioris, FIBL fibularis longus, GLMAX gluteus maximus, GLMED gluteus medius, FLEX flexors muscles, EXT extensors muscles, BPF band pass filter, HPF high pass filter, LPF low pass filter, RECT rectification, NORM max value normalization, PCA principal component analysis, RMS root mean squared signal, LDA linear discriminant analysis, NARAX non-linear autoregressive neural network with exogenous inputs, QDA quadratic discriminant analysis, ANN artificial neural network, DBN dynamic Bayes network, SVM super vector machine, LM Levenberg-Marquardt, EBA entropy-based adaptation, LIFT learning from testing data adaptation, ENV envelope, IDE integral of differential EMG, MAV mean absolute value, STD standard deviation, MDF median frequency, MF mean frequency, MPF mean power frequency, ZCN zero-crossing number, WL waveform length, SSC sign slope change, ARC autoregressive coefficients, MSAR mean square of ARC, TDAR time-domain and ARC combination, MEC features from mechanical sensors, HIST histogram bin values, VAR variance
  3. \(^{\star }\)Used during calibration phase
  4. \(^{\dag }\)Used during testing phase