Satellite Telemetry Feature Reconstruction using Supervised Learning Algorithm  Throughput Performance

Robert P

首页> 外文期刊>Journal of spacecraft technology >Satellite Telemetry Feature Reconstruction using Supervised Learning Algorithm Throughput Performance

【24h】

Satellite Telemetry Feature Reconstruction using Supervised Learning Algorithm Throughput Performance

机译：Satellite Telemetry Feature Reconstruction using Supervised Learning Algorithm Throughput Performance

获取原文

获取原文并翻译 | 示例

掌桥外文数据库（机构版） >>

开具论文收录证明 >>

文献代查 >>

页面导航

摘要
著录项
相关主题

摘要

LEO satellites (by ISRO) launched either in Sun Synchronous Polar Orbit (SSPO) or Inclined Orbit have limited visibility (6 to 12 minutes/pass) over Ground stations. Typically, the LEO SSPO satellites pass three to four times a day over a given ground station within 14 to 15 orbits per day. This requires onboard storage of telemetry data and transmitting the same during visibility. A 1∶1 storage of telemetry onboard requires larger memory sized to the maximum number of non-visible orbits, longer transmission duration, increased transmitter bandwidth and multiple ground station availability. This becomes more complicated with the ever increasing on-orbit satellites and limited ground resources for reception. To circumvent this problem, two strategies are primarily followed. Firstly, the parameter sampling on-board is varied anywhere between once/multiple times per frame to slower sampling across master frame. Secondly, the concept of TM storage ratio (SR) ranging from 1:1 rate, to a worst case scenario of 1:16 during non-visibility as per operational convenience. This optimized trade-off scenario is acceptable for health monitoring during nominal satellite performance. However, this raises non-linear time sampling issue for supervised learning algorithms solving regression problems. This paper describes a methodology using Long Short-Term Memory (LSTM) based supervised learning to reconstruct satellite telemetry that is missed due to SR. Typically, it is observed that, only 20% to 22% of sampled telemetry data over a day is stored on-board and transmitted to ground. This methodology segregates telemetry data in time segments, does continuous localised learning from the data and localised prediction for the unavailable segment time duration, while preserving the original telemetry according to its time stamp.

著录项

来源
《Journal of spacecraft technology》 |2022年第1期|9-19|共11页
作者
Robert P;
展开▼
作者单位

Mission Planning and Operations Group, U.R. Rao Satellite Centre, Bangalore, India;

展开▼
收录信息
原文格式 PDF
正文语种英语
中图分类航天（宇宙航行）;
关键词
Machine Learning; Supervised training; Telemetry; Feature Reconstruction; Long Short-Term Memory (LSTM);

Satellite Telemetry Feature Reconstruction using Supervised Learning Algorithm Throughput Performance

摘要

著录项

相关主题

期刊订阅