Sorry for the delay. I don't see the typical EMG "burst" in your data but it could be due to your timescale being too short. Can you run the data acquisition on a more compressed timescale so we can see both data from no contraction and data from the contraction on the same screen? In other words, collect the data at the same sample rate but show about 5-10 seconds of data on one screen with a contraction "burst" somewhere. Then we can see the baseline noise (no contraction) and the signal (strong contraction) all on the same screen to better assess the signal-to-noise ratio.
This attached picture i took today, with strong contration and with a smaller time base. On the other hand the signal where there is no contraction looks like the attached picture (No contraction), but i removed the offest (DC voltage shown), this offest is adjustable using reohstat, i did it in order to adjust the input to the ADC which will be used further in the Processing.
Also i did some spectral analysis using NI Elvis "Dynamic Signal Analyser" and i get the results shown in the attached pictures.
In the spectrum it is shown that the signal with the highst gain is the 50 Hz(ac power lines), and also signal at 650 Hz ( i assumed it should be filtered out of the signal , and also until now i don't know the source of such frequency).
I am doing a very similar project. I will be controlling 5 degrees of freedom of a small robot arm (all servo motors) with just EMG as inputs. I have yet to buy a DAQ card. I just wanted to confirm whether the NI USB 6009 is enough to acquire 4-5 channels of EMG, from the bisceps, triceps and the gonastic and antigonastic forearm muscles respectively before I invest in any card? I reckon that 1kHz sampling should be enough for EMG. The specs for the 6009 says 48kS/s and 8 analog input channels along with 2 analog outputs. This seems fine on paper but do you think it will suffice in practice?
Thank you for your time.
Hello to everyone
Beirut I also had the same problem in my circuit with 50Hz power line noise as shown in the attached figure. The image shows an EMG burst and the baseline, both on the same screen so that we can compare the EMG signal to the noise signal. Here the time base is only 1 second. I think if you change your timebase value from 50ms to 1 sec (in Strong Contraction.png) and contract your muscle and quickly let go you will be able to clearly see the EMG burst (occurs when you contract your muscle) as well as the baseline (occurs when your muscles are at rest).I have also attached an oscilloscope image when my muscles are completely at rest. The frequency of this baseline is displayed at the bottom right corner of the oscilloscope output (50.007Hz).
The explanation I got from some reading is that ability of the Instrumentation amplifer to substract common mode signals such as 50Hz noise is limited. Ive read some papers where theyused a UAF42 notch filter set at 50Hz to supress this frequency before inputting the signal to an ADC (http://www.springerlink.com/content/g465265v5001m3t6/). In other papers they used a 50Hz notch filter within labview (http://dspace.epn.edu.ec/bitstream/123456789/113/2/P0180.pdf) .
Yes - I think you will be fine with the USB-6009. 1kHz sampling per channel is normally fine for EMG data acquisition - you should have no problem. These low cost DAQ products do not sample the analog inputs simultaneously (they are scanned) resulting in a slight phase skew from one channel to the next, but for control applications like this it is fine.
Would you say that the card’s voltage and sampling specs will be suitable rather than simply marginal to the task? Can you suggest any other cards that might be suitable? Should I consider any software overhead? And what computer (specs) would you suggest to optimally control the entire system? Are there any limitations of using a laptop with USB or PCMCIA type cards?
Thanks for your time.
You will need an EMG amplifier before the USB-6009 and most will give you an analog voltage that is perfectly suitable for acquisition by the USB-6009. The resolution, accuracy, and speed are good - not just marginal. The software overhead is not significant, and any reasonably new laptop will work fine - this is a very simple task for our DAQ hardware and LabVIEW and the USB interface.
Can anybody tell me how to design a classifier to classify different motions after feature extraction is done.Thanx
The role of the classfier is to distinguish between different motions, so you can use neural network, fuzzy logic and many other types you search google for other types that you can use.