Graphic Equalizer:
For the final project of the independent study when learning Digital Signal Processing (DSP). I had to build a graphic equalizer which uses parallel filters. The cascaded filters would alter certain frequencies of the audio signal. It could enhance certain ranges or diminish certain ranges. The filters were designed in Matlab and then audio song was fed to the Filter and graphic sliders were created in matlab.
* I do not own any rights to the music and it was used for only academic research purpose.
The video shows a 5 coefficients, the first coefficient C1 either diminishes (<1) the bass (range 0-70 Hz) or enhances (>1) the bass. Similarly C2 affects ranges 70 to 500 Hz. C3 affects 500 to 1000 Hz, C4 affects 1000 to 5000 Hz and C5 affects 5000 to 20000 Hz. A detailed description of how the filters were designed and how the sliders were designed can be found in the final report, whose link is provided:
Final Report
Particle Swarm Optimization:
As a part of my indpenedent study I had to visualize and understand about Particle Swarm Optimization (PSO). Particle Swarm Optimization is a population based optimization technique. It was designed based on the social behavior of bird flocking. In the flocking behavior of birds each bird behaves independently but communicate with each other to meet the objective. It is a metaheuristic meaning no assumptions are made.
The PSO tries to simulate the behavior of birds flocking and so defines multiple particles which tend to simulate the motion of the birds. Each particle is allowed to move on a predefined search space and communicate with each other to move to the appropriate solution.
The codes were initially implemented on Matlab and then tested on FPGA (Field Programmable Gate Array). The link to the final report is:
Final Report
The gif provided has a red square which defines the best position for all of the birds and blue dots are the birds moving around trying to find the other solutions. The x-axis is the solution for the equation x^2 == 0 whereas the y-axis is the cost function which is basically y = x^2. The best answer is the minimum cost which is possible only when x = 0.
Single Layer Perceptron:
As a part of my independent study I had to build a simple neural network called single layer perceptron (SLP). The single layer perceptron uses a linear iterative learning method to better create a functions which can be used to sort data.
On the gif provided there are 2 distinct data sets, one which is red and the other which is green. The SLP is trying to create a linear function which can divide the data set properly. It adjusts the line so that all the data sets are separated, until it no longer can. There are some outliers on the plots and so the SLP would never converge which is generally true for most neural network as 100% is very difficult.
The final document shows the training of SLP using PSO as well:
Final Report
For the final project of the independent study when learning Digital Signal Processing (DSP). I had to build a graphic equalizer which uses parallel filters. The cascaded filters would alter certain frequencies of the audio signal. It could enhance certain ranges or diminish certain ranges. The filters were designed in Matlab and then audio song was fed to the Filter and graphic sliders were created in matlab.
* I do not own any rights to the music and it was used for only academic research purpose.
The video shows a 5 coefficients, the first coefficient C1 either diminishes (<1) the bass (range 0-70 Hz) or enhances (>1) the bass. Similarly C2 affects ranges 70 to 500 Hz. C3 affects 500 to 1000 Hz, C4 affects 1000 to 5000 Hz and C5 affects 5000 to 20000 Hz. A detailed description of how the filters were designed and how the sliders were designed can be found in the final report, whose link is provided:
Final Report
Particle Swarm Optimization:
As a part of my indpenedent study I had to visualize and understand about Particle Swarm Optimization (PSO). Particle Swarm Optimization is a population based optimization technique. It was designed based on the social behavior of bird flocking. In the flocking behavior of birds each bird behaves independently but communicate with each other to meet the objective. It is a metaheuristic meaning no assumptions are made.
The PSO tries to simulate the behavior of birds flocking and so defines multiple particles which tend to simulate the motion of the birds. Each particle is allowed to move on a predefined search space and communicate with each other to move to the appropriate solution.
The codes were initially implemented on Matlab and then tested on FPGA (Field Programmable Gate Array). The link to the final report is:
Final Report
The gif provided has a red square which defines the best position for all of the birds and blue dots are the birds moving around trying to find the other solutions. The x-axis is the solution for the equation x^2 == 0 whereas the y-axis is the cost function which is basically y = x^2. The best answer is the minimum cost which is possible only when x = 0.
 |
| Particles (blue circles) adapting their values to get the best value (red square) |
Single Layer Perceptron:
As a part of my independent study I had to build a simple neural network called single layer perceptron (SLP). The single layer perceptron uses a linear iterative learning method to better create a functions which can be used to sort data.
 |
| Single Layer Perceptron teaching itself a function how to divide the green and red circles |
On the gif provided there are 2 distinct data sets, one which is red and the other which is green. The SLP is trying to create a linear function which can divide the data set properly. It adjusts the line so that all the data sets are separated, until it no longer can. There are some outliers on the plots and so the SLP would never converge which is generally true for most neural network as 100% is very difficult.
The final document shows the training of SLP using PSO as well:
Final Report
Comments
Post a Comment