15241

13698

كارشناسي ارشد

مخابرات سيستم

اصفهان : دانشگاه صنعتي اصفهان

1398

دوازده، 76ص. : مصور،‌ جدول، نمودار

حامد نريماني

محمدعلي خسروي‌فرد

علي‌محمد دوست حسيني، محمد دخيل‌عليان

1398/08/01

كتابنامه

مهندسي برق

مهندسي برق و كامپيوتر

1398/08/04

2572044

Optimimality Analysis of the Source Codes Based on Partial a Priori Information and Limited Observations Donya Sadat Rezaeishad ds rezaeishad@ec iut ac ir September 2019 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Assist Prof Hamed Narimani h narimani@cc iut ac ir Advisor Assoc Prof Seyyed Mohammadali Khosravifard khosravi@cc iut ac ir Abstract For a given memoryless source when the symbols probabilities vector P n is known the optimalcode can be obtained from the well known Huffman algorithm It is obvious that if the symbols probabilities are not precisely known then the algorithm can no longer be used The problem in thisthesis is the source coding in the absence of symbols probabilities information possessing partial information about the source features and observing a finite sequence of source output symbols One of the most primitive methods for estimating the symbols probabilities based on source outputobservation is the empirical frequency estimator that assigns a probability to each symbol propor tional to the number of symbol repeats in the observed sequence This estimator does not dependon prior source information The Add estimator is another method of estimating symbols proba bilities which provides the optimal code concerning average redundancy when we have a Dirichletdistribution over the symbols probability vector The criterion used in this study for evaluating thecode performance is the probability of that the Huffman codes from the estimated and actual probabil ity distributions do not coincide which we call it error probability of estimation Calculations showthat in some cases source coding based on pure prior information has less error probability than usingthe empirical frequency estimator Specifically for monotone sources with 4 5 and 6 symbols if thelength of the observed sequence is less than 62 25 and 11 respectively the choice of code based onthe empirical frequency estimator has more error probability than source coding using prior infor mation on average In this thesis methods are introduced for source coding that minimize the errorprobability by simultaneously using previous information and the sequence of output observations ofthe source For example for sources with 8 and 9 symbols if the length of the observed sequenceis less than 100 then the error probability can be reduced by at least 0 068 and 0 074 respectively compared to the empirical frequency method using one of the proposed methods Keywords Source coding Data compression Optimimality of the source code Empirical frequencyestimator

حامد نريماني

محمدعلي خسروي‌فرد

علي‌محمد دوست حسيني، محمد دخيل‌عليان