Surface EMG of the hand without the palm: how many thumb channels, and where?
PavelKamenik
czech republic
I'm designing an open-source EMG suit (ADS1299, 8 channels per module, bipolar pairs).
One module covers the forearm and the hand:
5 channels in a ring round the upper forearm, at 40 % from the medial epicondyle to the
ulnar styloid (after Farrell & Weir 2008),3 on the hand: first dorsal interosseous, abductor pollicis brevis at the thenar edge,
abductor digiti minimi at the hypothenar edge, pairs 2 cm apart.
The palm stays bare (grip pressure), and the hand's electrodes would sit in a tight glove:
dry 316L steel studs in moist rings. The goal is decoding finger and thumb movement and
force for control.
Questions:
1. Off the palm, which thumb sites add information beyond APB and FDI? Is a site along the
radial side of the first metacarpal (opponens, under APB) worth its own channel, or is
it APB again?
2. Does a forearm ring see flexor pollicis longus well enough for thumb pressing, or should
one forearm channel go over FPL on purpose?
3. 5 forearm + 3 hand against 4 + 4: any data or experience?
4. Electrodes inside a glove on the back and edges of the hand: how stable is the contact
during grip?
The layout: https://github.com/Project-NIC/NIC-Crazy_Ivan/blob/main/bogatyr/CONSTRUCTION.md
Thanks for any pointers or papers.
Comments
For the 5+3 versus 4+4 question, if you can record the candidate sites together, I'd compare the two channel subsets on the same repetitions, then test again after removing and refitting the glove. Keep whole repetitions out of training, rather than splitting neighbouring windows between training and test. That would help separate the value of the extra channel from how well a particular fit was learned.
An older forearm-only reference is Fu et al. (2021): six channels, eight finger gestures, five people, with the arm resting on a table. A useful starting point, but it doesn't settle your glove layout or continuous force control.
https://doi.org/10.3389/fncom.2021.770692
Thank you for both answers, here and on the hand thread.
On the hand: the glove records all sixteen electrodes anyway, so the 5 + 3 and 4 + 4 subsets will be compared offline on the same repetitions, with a refit test after taking the glove off and putting it back; whole repetitions held out of training is now the rule, and Fu et al. 2021 is noted as the starting point.
On the artifacts: your dummy body, with matched and then deliberately unequal P and N impedances, the strip moved with the electrode ends held and then the contact disturbed with the strip still, is now the first test on the first arm. It matters more since beta 0.6: the strip is a four-layer flex with the leads between two grounds, and that ground is driven at the body's common mode by default. Your unequal pair is what will tell whether driving it earns its keep, so we will run it with the layer on ground and driven.
One question on the dummy itself: what would you use for each electrode's contact, a resistor alone or a resistor with a capacitor across it, and roughly what values for a moist electrode of carbon-filled conductive silicone? And how would you disturb a contact repeatably — a weight, a small solenoid tapping, something else?
I'd keep the first dummy resistor-only: matched inputs, then a deliberate mismatch on one side. That gives you a controlled electrical comparison while moving the strip. It does not reproduce a moving electrode contact.
For the physical contact test, I'd choose any R/C approximation from the measured impedance magnitude AND phase over the recording band, with the contact area, moisture and preload fixed. I wouldn't assign a supposedly representative pair from 'carbon-filled silicone' alone. Your material's bulk resistivity and the electrode/contact impedance are different quantities. Goyal et al. show a useful phantom-and-fitting approach, but their steel/titanium values aren't silicone values: https://pmc.ncbi.nlm.nih.gov/articles/PMC9656728/
For repeatability on the bench, a guided displacement with a stop, a fixed preload and a force trace would tell you more than an unmeasured tap. Separate normal loading from sideways movement and keep the lead routing fixed. Cömert and Hyttinen used programmed movement with monitored force; their textile-electrode results don't automatically transfer to your contacts: https://pmc.ncbi.nlm.nih.gov/articles/PMC4242479/
If you use a solenoid or motor, include a powered-but-not-moving control to check electrical pickup from the actuator. That's a proposed bench control, not a tested fix or a safety check for wearing the suit.
Ah, thanks Alex. The Goyal paper IS using steel or titanium electrodes, so I guess that can be made to work. Perhaps the paper link I posted on the other thread is less relevant now.
However, the figure 9 in that paper does show that proper skin prep is essential for the smaller 6 cm^2 electrodes (steel or titanium). The signal quality goes to hell on the 6 cm^2 electrodes, if no alcohol swap / skin prep is done prior to measurement. I wonder how much of that effect is due to the water content of the IPA, moisturizing the skin for the subsequent measurement.
One detail in section 2.3.2 changes how I'd read Figure 9: the alcohol swabs were placed directly on the electrodes during recording, rather than used to clean the skin and then removed. The discussion suggests both filling the air gaps and moistening as possible mechanisms. So this comparison doesn't isolate skin preparation, or the water content of IPA, on its own. That distinction seems worth keeping in the glove tests.