Saturday, 8 December 2018

Section 8.8: Discrete Haar Wavelet Transform.

Difficult:

The most difficult part is keeping track of the indices and how they affect the equations. Memorization doesn't seem like the most practical approach. As I learn to visualize these better I think I will be able to use them out of intuition rather than memorized steps.


Interesting:

The complexity of the fast version of this algorithm is very impressive. Something that I have found fascinating in this class is learning of all the logical ways to make an algorithm run faster with less computational effort. Most of the time, like in this chapter, it requires a clever way of going about solving the problem. I am motivated to find a faster version of everything I code because of this class.


Thursday, 6 December 2018

Section 8.7: Haar Wavelets

Difficult:

Conceptually the way we approximate functions with Haar scaling function is easy to understand. Some of the notation and manipulation of these concepts are more complex. Sadly because I am still trying to understand notation that the rest of the chapter depended on, it made it more difficult to understand.

Insightful:

It is cool how well these functions can fit discontinuous functions. This is why it is superior at approximating discontinuous functions than fourier transforms. Which method would be better to approximate nonoscillating continuous functions? are there more types of functions that would not be efficient to use the two methods already studied? 

Tuesday, 4 December 2018

Section 8.6: Periodic Sampling Theorem

Difficult:

Antialiasing went over my head. But as I understand it we are setting up a new bound that is dependent on n - the number of samples. Then when we compute the equation from the samples we are guaranteed the unique solution. but I think there is more to this.

Insightful:

It is interesting how we filter sound according to human capability. It is interesting how we can have many results from aliasing. I wonder what is the restricting factor for not having enough samples.  

Saturday, 1 December 2018

Section 8.5: Convolution

Difficult:

I still don't understand the application of convolutions. I am not sure what the graph on figure 8.5.1 is supposed to be representing.
I understand that using Fourier transforms and discrete convolutions we can compute the circular convolution a lot faster. Now it would be nice to know why we are computing it. 

Interesting:

I have been wanting to learn more about convolution. I have a vague remembrance of it from math 334 and have seen it come up in research papers. I didn't know they correlated with Fourier transforms. Looking forward to understanding them better.