Fall 2011 (RA by Mohamed Hefeeda)

Sept. 20

Checked the unusual spike in PSNR results of "Toy Story" sequence around frame number 200 (above 50 db). There wasn't any bug in the code. As you can see in the trace file here [1] the quality of frames 161 - 200 with enhancement layer bit-rate of 2000 is around 52 db.

completed the writing in the format of a paper, modified the pseudocode, added a section "Practical Considerations" before experimental evaluation.
As the perceived video quality depends on the avg quality of transferred frames as well as variation of quality between successive frame I expanded the experimental Evaluation part by comparing average quality, Coefficient of variation, min_quality/ avg-quality of each scenes in frame-based, GOP-based and Scene-based optimization case. The results showed that the Scene-based optimization case yielded less variation in quality (lower coefficient of variation) and higher min_quality/ avg_quality in each scene, which was expected as we assigned equal number of enhancement layer bits to all the frames belong in the same scene. but the variation in quality between successive scene may degrade the quality of the video stream. Therefore, I computed average quality, Coefficient of variation, min_quality/ avg-quality of "'each stream of video delivered during an idle period"'. The results showed that avg_quality of frame-based optimization was higher compared to the other two cases and the coefficient of variation in this case was less than the other cases and the min_quality/ avg_quality was higher. In conclusion, frame-based optimization may result in less variation in quality throughout the current stream while resulting in more variation in a scene!

Also JC suggested that I start the write up of the thesis, and also do some research on why Markov model may be a good model for channel usage and the other existing models or prediction techniques.

I did the structure of thesis and chapters. The very first draft can be found here: []

I plan to do a study on channel/traffic model. I found the following papers useful
- [2]
- [3]
- [4]

Summer 2011 (TA course MACM 101)

Aug. 16

updated the algorithm pseudocode
implemented the algorithm with two other aggregation cases (Frame-based and GOP based). Ran the same scenario with these two cases and compared the PSNR results to scene-based optimization case.
The observation showed that although the computational cost of frame-based approach was much more than the the GOP-based approach, the PSNR results were very close to GOP-based approach.
I believe that the channel estimation still needs some work, as you can see for a number of frames, the quality drops noticeably to the quality of base layer. (where we have serious frame loss)

My report can be found here : [5]

July 20

busy with TA duties, marking midterm exam and assignment for a class of 80 students
Also as you and JC mentioned, I started writing the whole paper and applied your comments. I tried to explain the problem in more detail, in an organized way. and highlight the challenges. I also tried to explain the solution more clearly. The part remained is giving a formal form (pseudocode) of the algorithm. Also as you said, I should take into account the variation in image quality between successive images (I,B,P) frames, as the perceived quality also depends on quality variation and for the same number of bits added to I,B,P frames, the gain in quality are different.

My report till now: [6]

July 5

I finished the simulation of the problem. I formulated the problem as an optimization problem which could be modeled as a bounded knapsack problem. The problem is then solved by Reaching method of Dynamic Programming. The results show the increase in Quality (in db) by taking advantage of dynamic spectrum access, using cognitive radio networks.

My report can be found here: [7] The src code : [8]

June 8

Talked to Ahmed about 3D medical Imaging and after looking into their work and some papers for about a week, I decided that their work is so much different from what I've been doing. Ahmed said that getting to know all about it took him about 4 months, so I decided that I'd rather stay on my research topic and invest those four months into it.

I decided to go back to simulating it with Java, FGS trace files with the help of this paper : [9]

May 10

April 8

Been busy simulating my ideas but found the settings of simulation unrealistic and looking for cognitive radio simulators, I decided to change to CRCN simulator (NS-2 based). Now I'm done with the installation and set up part and I'm going through NS-2 and OTCL tutorials to see how I can simulate my scenario in this simulator environment. More details can be found in my report [11].

Jan 18

worked on simulation of the project (video streaming over CR networks), changing some parameters and trying to analyze the results.

Courses:
- CMPT 771 Internet Arch and Protocols (Grade:A)
- CMPT 820 Multimedia Systems (Grade: A-)

Courses :
- CMPT 705 Design/Analysis Algorithms (Grade: A)
- CMPT 882 Spc.Topics Art. Intelligence (Grade :A)