K-Means Cluster Study for Radiofrequency Propagation Characterization

Cody, Tyler; Roche, Rigoberto; Briones, Janette C.

The objective of this study is to design a simple method for mining radio frequency (RF) propagation data. The study explored the characteristics of a large dataset of propagation experiments conducted over the span of years and using several ground stations around the world. Furthermore, this study developed simple predictive models that can be used for link characterization and overall propagation behavior description, without the need for physical measurements on-site. It is understood that such statistical learning has several drawbacks in terms of accuracy and precision. K-means clustering was used to characterize the data set in a way never explored before in an attempt to create useful tools that reduce cost, time and risk. K-means clustering was used to characterize the data set. Cosine distance was used as a method to determine the optimal number for clustering each feature. Dependence and independence analysis was performed to explore intra and inter-sensitivity between the presented features, with respect to each other and time. Several predicative models were generated and evaluated with respect to a test set to assess a measure of prediction accuracy and precision. A simple method for data analysis was developed and tested as the basis for further studies and future refinement to produce optimal performing models.

Document ID

20180004931

Acquisition Source

Glenn Research Center

Document Type

Technical Memorandum (TM)

Authors

Date Acquired

September 6, 2018

Publication Date

July 1, 2018

Subject Category

Report/Patent Number

Funding Number(s)

Distribution Limits

Public

Public Use Permitted.

Keywords

Available Downloads

Name

Type

20180004931.pdf

STI

No Preview Available

NTRS

NTRS - NASA Technical Reports Server

Available Downloads

Related Records