Citation
Abstract
This article describes computationally intelligent neural-network and leastsquares tracking algorithms for fine pointing NASA’s 70-m Deep Space Network (DSN) antennas using the seven-channel Ka-band (32-GHz) array feed compensation system (AFCS). These algorithms process normalized inputs from the seven horns of the array in parallel and, hence, are less sensitive to variations in signal power and require much less time per pointing estimate than do conventional serial processing techniques (CONSCAN) currently employed by the DSN. An additional advantage of the parallel measurement technique described here is that mechanical scanning of the antenna is not required. It is shown that under nominal conditions a radial basis function (RBF) neural network trained with data from the seven array feed channels can point the antenna with less than 0.3-mdeg rms error, achieving significantly better pointing accuracy than the 0.8-mdeg requirement often quoted as a benchmark (corresponding to 0.1-dB signal-to-noise ratio (SNR) loss on 70-m antennas). Both neural-network and least-squares algorithms were simulated and compared using computer-generated antenna distortions. The results indicate that in acquisition mode the RBF neural network performs best at high SNRs; however, for low SNR applications, the least-squares algorithms yield better performance. All algorithms were shown to be strong candidates for providing excellent acquisition and tracking capabilities for the DSN’s 70-m antennas.
Details
- Volume
- 42-141
- Published
- May 15, 2000
- Pages
- 1–16
- File Size
- 310.9 KB