Exploration of Diffusion LMS Over Static and Adaptive Combination Policy
| Autor: | Chikwendu N. Chiamaka, Alula A. Tesfaye, Amare H. Hailu, Chikwendu A. Ijeoma, A Hossin, Hailegiorgis A Bemnet |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
050101 languages & linguistics
Diffusion (acoustics) Mathematical optimization Computer science Diffusion of information 05 social sciences Relative standard deviation 02 engineering and technology Least mean squares filter Adaptive filter 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing 0501 psychology and cognitive sciences Wireless sensor network Implementation Selection (genetic algorithm) |
| Zdroj: | 2020 17th International Computer Conference on Wavelet Active Media Technology and Information Processing (ICCWAMTIP). |
| Popis: | In recent years, Diffusion LMS algorithm has been thoroughly researched. This powerful approach enables distributed problem optimization over sensor networks to be solved. The best parameter vectors for all agents is likely not to be the same in such implementations. In addition, agents typically share information through noisy links. Here, we compared the average performance of some major distributed network-based combination policies: Static Combination Policy (metropolis policy and relative variance combination policy) and adaptive Combination Policy. The diffusion of information gathered from neighbors is an important problem in creating adaptive networks in as much as the network mean-square performance is dependent on the preference for combined weights. The aspect of the best selection of the combination weights is considered. We showed the adaptive combination policy outperformed the two analyzed static combination policy over noisy links. Also note that because we considered linearly complex algorithms in a single-task LMS diffusion, we analyzed the theoretical performance within a collaborative target tracking problem. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|