Citation
Abstract
The downlink analyzer (DLA) is a hybrid learning and monitoring system combining classical signal-processing and connectionist (i.e., neural network) pattern classification. It learns to detect and diagnose anomalous operations in the downlink portion of the Deep Space Network, NASA’s communications link to all unmanned spacecraft. To meet its learning and monitoring objectives, the DLA must process data sequences gathered from throughout the downlink. Many sequences have low signal-to-noise ratios, and the sample rates for different information sources vary widely. This article describes the technologies employed by the DLA, focusing on aspects that are novel. In the process of learning to date, the DLA has discovered a previously unknown downlink failure mode, providing the first direct evidence that it can teach human engineers, scientists, and operators subtle but important factors that lead to the loss of scientific data.
Details
- Volume
- 42-126
- Published
- August 15, 1996
- Pages
- 1–19
- File Size
- 592.4 KB