DSP application

 The paper talks about biomedical signals taken from the spine, brain or muscles themselves to find out which part of the body is the signal meant to move. The processing of EMG signals is inherently tough because of their small amplitude. The transfer function used to process them to obtain signals of workable strength also changes w.r.t. the sensors i.e. surface EMGs or real wired EMGs. For both types have their own strength, they promptly decided to use two sensors for each.
 
 The program will then use a simple correlation algorithm to check inputs coming from the sensors and compare it with the preprocessed one. A good reliability can be expected when the sensors are calibrated specifically to your body. 

The applications range from sensing the movements and efficiency of muscles, to using the signals to control prosthetics, or maybe a mimicking robot used to fight(Real Steel style).


Inventor: Jaroslaw Zarychta, Winnipeg(CA)
Assignee: Respironics, INC., Pittsburgh, PA(US)
Appl. No.:09/567,795
Filed: May 9, 2000










Drive link to paper and patent
 

basic operations using DSP

The basic operations were demonstrated by JaiGanesh(EXTC) on a TMS320F28375 kit. He gave us the demo of the very basic operations like add,substract, logical shifts and multiply. With these we can implement the operations we want, for example convolution would require multiply and add; shifting can multiply/divide numbers by 2 and its powers. 

A DSP is better than a general purpose processor because of functionalities like circular shifting, dedicated high speed MACs, VLIW and other features, making it more suitable for signal processing, even real time applications.

FIR using Frequency Sampling method

Frequency Sampling method. Sounds so good on paper. If time can be sampled, why not frequency? But it was so tough. I don't think anyone could do it the designated turn, and we all sort of did it together. Can't argue was really hard to get the Phase response. But eventually, done.

The verification of the output was easy as an FIR filter always has a linear phase response, so all that had to be checked was the phase response. A linear phase response is all the confirmation needed for verification.


https://drive.google.com/drive/folders/0B0_KMgflfFtwTklLN0pLbDVnMHM

FIR using window function

The concept of window functions in itself was really awesome. It went something like"What? You can design IIR filters easily? But the response of FIR filters is better? Don't worry, Just pass it through a window and make IIR into FIR". Lovely.

Implementation-wise, waaayyy easier than FSM. FSM was tough. But window wasn't that tough.

Different windows can be implemented. Rectangular window is very easy to implement; the sinusoidal window Hamming and Hanning are the most appropriate given our As requirements; Blackman has the best As.(Somehow Blackman was not classified as a sinusoidal window)

http://sahil8rai.blogspot.com/2016/04/fir-using-frequency-sampling-method.html

Chebyshev filter design

The Chebyshev filter has ripples in the PassBand, has less hardware requirements and hence costs less. The whole idea is clearly seen in the spectrum analysis-cost trade off. A bad magnitude spectrum for good cost efficiency.

The conclusion this time was interesting. The poles of both type of filters were on the right hand side of the s-plane. Which means unstable, but then why are they used as filters??

Also the ripples in the passband(number of valley and peaks) is equal to the filter's order.

http://sahil8rai.blogspot.com/2016/04/fir-using-frequency-sampling-method.html

Butterworth filter design

The design was supposed to be done in Scilab, which I did not know. Kartik helped me with that and I think I understand the language a lot better now. I have to admit though, the theoretical calculation seemed a lot simpler than the program, which beats the purpose. But the program can be used for any order filter.

The conclusion was tricky, as Sir asked us to base it on the location of poles. Nothing unexpected happened there. The poles were inside the unit circle, one real and a conjugate pair for the LPF and two conjugate pairs for HPF.

http://sahil8rai.blogspot.com/2016/04/fir-using-frequency-sampling-method.html

Overlap add and Overlap save method

Overlap add and Overlap save method are the algorithms used to calculate the convolution result when the whole of the input is not available but the output is needed. The only need for such an algorithm would be in real time applications.

I didn't understand how breaking up the input signal into smaller parts would still give the same convoluted answer, as it should need all possible inputs given. But the program gives identical results, so I have to accept it.

http://sahil8rai.blogspot.com/2016/04/fir-using-frequency-sampling-method.html