beam has been focused on a point 20 A above the antenna plane in a direction perpendicular to the antenna normal. A two-tone system has been studied as a possible retrodirective antenna which solves the problem of induced phase error as mentioned in the last section [18], Two pilot signals (frequencies/t + 2A/and/+ A/;/ is the frequency of the energy beam) are used in this two-tone system. Subtraction of one of the pilot signals (/ + 2A_/) from doubled signal of the other (2/+ 2 A/) gives a signal with frequency / and phase of 0, (Fig. 8). 0, is the phase lag of the/ frequency wave along the path difference AL. The merit of this system is that the phase of the transmitting wave can be determined without ambiguity from the two pilot signals. A retrodirective antenna using such a system was described in detail elsewhere [18]. The computer monitors and corrects the direction of the microwave beam by adjusting the individual output phases of the transmitting antennas via commands from the receiving site. Many methods have been proposed to determine correct adjustment; one is a coherent multiple tone technique documented by Chie [19]. We have studied another method using a small neural net corrected by a back-propogation learning procedure. One merit of using a neural network is the high speed of computation due to parallel processing. This is especially suitable for very large array antennas with many elements, i.e., antennas of the kind liable to be found in SPS applications. A neural net consists of an input layer, several internal hidden layers and an output layer (Fig. 9) which are connected with values of weights. The network needs a learning procedure to determine the weights. In the back-propagation learning method, test cases—in the form of patterns—are fed into the system. The weights of the system are then adjusted so as to minimize the difference between actual output patterns and
RkJQdWJsaXNoZXIy MTU5NjU0Mg==