CAO Hai-qing(曹海青), WANG Yu(王渝), YAO Zhi-ying(姚志英)

(1.School of Information, Capital University of Economics and Business, Beijing 100070, China;2.School of Automation, Beijing Institute of Technology, Beijing 100081, China;2.School of Logistics, Beijing Wuzi University, Beijing 101149, China)

?

Optimization of satellite searching strategy of the non-stationary antenna

CAO Hai-qing(曹海青)1,2,, WANG Yu(王渝)2, YAO Zhi-ying(姚志英)3

(1.School of Information, Capital University of Economics and Business, Beijing 100070, China;2.School of Automation, Beijing Institute of Technology, Beijing 100081, China;2.School of Logistics, Beijing Wuzi University, Beijing 101149, China)

The tiny searching step length and the satellite distribution density are the major factors to influence the efficiency of the satellite finder, so a scientific and reasonable method to calculate the tiny searching step length is proposed to optimize the satellite searching strategy. The pattern clustering and BP neural network are applied to optimize the tiny searching step length. The calculated tiny searching step length is approximately equal to the theoretic value for each satellite. In application, the satellite searching results will be dynamically added to the training samples to re-train the network to improve the generalizability and the precision. Experiments validate that the optimization of the tiny searching step length can avoid the error of locating target satellite and improve the searching efficiency.

tiny searching step length; satellite finder; patter clustering; neural network

The non-stationary satellite antenna consists of the vehicle satellite antenna, airborne antenna, ship-borne antenna,antenna on board and so on. While the loader is in a mobile state, the antenna must accurately and efficiently direct and locate the chosen satellite to ensure a high communication quality. At present the satellite finder procedure of the antenna can be concluded as shown in Fig.1. Firstly, the azimuth mechanism rotates to the required position by the azimuth servo-system. Then the elevation mechanism rotates to the required position by the elevation servo-system. At last the azimuth mechanism must rotate at a small step length named as the tiny searching step length, which is important to improve the efficiency of satellite finder, around the azimuth domain in order to ensure the antenna to locate the satellite accurately, in the process of satellite finder the auto gain control (AGC) detected by the electromagnetic sensor should reach the maximum in the end. Although the above searching strategy has been widely applied in all kinds of no-stationary antenna products, there are still some shortcomings.

Fig.1 Process of satellite finder

①Antenna locating the error target satellite

Many experiments of the above searching strategy have been done in Beijing, in which Apstar-V is the target satellite. In the process of satellite searching, the AGC detected by an electromagnetic sensor is shown in Fig.2. The azimuth mechanism firstly rotated 31.5° from the north to the south to the pointAwhich was the theoretical azimuth of Apstar-V, then the elevation mechanism rotated 38.7° from the horizontal plane to vertical plane to the pointEwhich was the theoretical elevation of Apstar-V and the AGC value became larger simultaneously, then the azimuth mechanism started to search the target at a constant tiny searching step length from the south to north, and the AGC value started to decrease gradually, then the azimuth mechanism restarted to search from the north to south and get to the pointF, the AGC value increased gradually and finally arrived at the maximum. Experiments validated that the located target satellite often was AsiaStar-VI instead of Apstar-V. There are two reasons as follows: Firstly, AsiaStar-VI is the nearest satellite to Apstar-V which can be received by all kinds of antennas in Beijing; Secondly AsiaStar-VI is newer than Apstar-V; thus as soon as the antenna pointed to the public signal coverage region between Apstar-V and AsiaStar-VI, it was easy to mistake AsiaStar-VI as Apstar-V for the AGC of Apstar-VI is larger than that of Apstar-V. In fact, the satellite searching error occurs is due to that the constant tiny-searching step length is too large. Now, the contest for space resources is becoming increasingly fierce, the intervals among adjacent geostationary satellites become smaller and smaller, thus the above satellite finder strategy with the constant tiny-searching step length can easily lead to locating error target satellite.

Fig.2 AGC of the satellite finder

②Low efficiency

Now most antenna products use the above strategy with a constant step length to locate the target satellite. In the process of searching different geostationary satellites, it must repeat searching if it finds the error target, which will influence the searching efficiency seriously.

In order to optimize the satellite finder strategy, avoid the error searching the target satellite and improve the searching efficiency, the tiny-searching step length is studied in the paper by the patter clustering and artificial neural networks based on satellite ephemeris data. Meanwhile the geostationary satellites received in Beijing are regarded as examples to validate the searching strategy.

1 Factors analysis of the searching satellite

The antenna fulfills the satellite finder task on the base of the azimuth, elevation and polarization angle calculated from the satellite ephemeris data and the position information of the antenna[1-3]. In the process of the satellite finder, the azimuth mechanism will rotate gradually at a tiny searching step length around the theoretical azimuth to locate the target satellite precisely. So the azimuth, elevation and the tiny searching step length are the main factors to influence the satellite finder, which depend on the satellites distribution density.

1.1 Distribution density of the geostationary satellites

The distribution density of the geostationary satellites has become denser and denser since many countries began to contest for the space resources. The distribution of the geostationary satellites received in Beijing is shown in Tab. 1[4], which is distinctly uneven. There are more geostationary satellites in the east longitude interval [70°,120] than that in [120°, 180°], while from the east longitude 0° to 40° there are few geostationary satellites.

Tab.1 Distribution density of geostationary satellite received in Beijing

1.2 Azimuth and elevation for the satellite finder

In the satellite searching process, the formulas of the azimuth and elevation in theory are listed as follows[5]

(1)

(2)

whereβistheazimuth; αistheelevation;Δφisthelongitudedifferenceofthetargetsatelliteandtheantenna; θisthelatitudeoftheantenna; R/Misapproximatelyequalto0.151 27,whereRistheradiusoftheearthandMisthedistancebetweenthetargetsatellitetothecenteroftheearth.Therelationshipamongtheazimuth,elevationandthelongitudeofthegeostationarysatellitereceivedinBeijingcanbeshowninFig.3.

1.3 Impact of the tiny searching step length from the distribution density of the satellite

The distribution density of the satellites influences the tiny searching step length through the azimuth interval between the adjacent satellites. Suppose the longitude of the geostationary satellite 1 is φ,andtheazimuthisβ.Thelongitudeofthegeostationarysatellite2isφ+Δlg,andtheazimuthisβ+Δβ.Thelongitudedifferencebetweenthesatellite1andtheantennaisΔφ,then

(3)

(4)

(5)

(6)

WhenΔlgvariesfrom0°to10°,thevariationoftheazimuthisshowninFig.4.Obviously,thelongitudevariationofthesatellitehastheimportantsignificancetocalculatethereasonabletinysearchingsteplength.

Fig.3 Relationship between the azimuth, elevation and the longitude of the geostationary satellites

Fig.4 Relation between the tiny searching step length and the satellite longitude

2 New strategy to calculate the tiny searching step length

A new strategy to calculate the tiny searching step length is studied based on the pattern clustering and artificial neural networks in the paper. Firstly, the geostationary satellites are classified into several classes by the pattern clustering, then the tiny searching step length (step1) is calculated by the trained neural networks according to the specific class A, meanwhile the tiny searching steps (step2or step3) are also calculated respectively by the trained neural networks according to the classes adjacent to the class A, at last the tiny searching step length (step) is decided by the minimum value among step1,step2and step3.

2.1 Pattern clustering of the geostationary satellite

The pattern clustering will classify the samples into different classes on the base of the pattern property[6].The longitude of the geostationary satellite is a main factor to influence the satellite finder according to the above study, so the clustering will depend on the longitude property mainly. For example, the geostationary satellites received in Beijing can be classified into 18 classes as shown in Tab.2, any geostationary satellite received in Beijing can be classified into the corresponding class according to this method.

2.2 Method to calculate the tiny searching step length by neural networks

According to the above study, the satellite distribution density of the different classes is different, the interval to different satellites in the same class is different too, furthermore there is a nonlinear dynamic relationship between the latitude and azimuth angle. Considering the distribution density of the geostationary satellites will be denser and denser in future, it is difficult to calculate the tiny searching step length. Because the BP neural network has the strong nonlinear mapping ability and self-learning ability, it can map the nonlinear relation between the satellite distribution and the azimuth, as well as adapt to the situation in which the geostationary satellite increases gradually. So a BP neural network is designed to optimize the tiny searching step length. The structure of the BP neural networks is shown in Fig.5, in which the longitude of the satellite, azimuth and elevation are the input elements and the tiny searching step length is the output. The number of hide layers in the network is 10 according to Ref.[7].

Tab.2 Clustering results of the geostationary satellite received in Beijing

Fig.5 Structure of BP neural network

Thirty satellites of the geostationary satellites received in Beijing are regarded as training sample sets, which are listed in Tab.3. Thirty satellites are classified into different classes according to the mentioned clustering method and used to train the BP neural network of each class. The tiny searching step length for each geostationary satellite is determined by analyzing the azimuth and elevation of the adjacent geostationary satellites. For example, the longitude of the satellite AM3 is east longitude 96.5°, the azimuth β1is29.51°northbyeast,andtheelevationα1is39.42°.OneofitsadjacentsatelliteisLuch-5B,whichlongitudeiseastlongitude95°,azimuthβ2is31.49°northbyeastandelevationα2is38.77°.TheotheradjacentsatelliteisAsiaSat-5satellite,whichlongitudeiseastlongitude100.5°,azimuthβ3is24.02°northbyeast,andelevationα3is40.94°.

SupposethetinysearchingsteplengthcorrespondingtotheAM5asstep-len,then

min(abs(β1-β2), abs(β1-β3))≤

step-len≤max(abs(β1-β2), abs(β1-β3))

(7)

where min is a function to determine the minimum of two numbers, max is a function to determine the maximum of two numbers, abs is a function to get an absolute value of a number. The other 16 geostationary satellites received in Beijing are applied to test the trained BP networks and the results are listed in Tab.4, in which the tiny searching step length is determined by the method in the paper, the theoretical tiny searching step length is determined by Eq.(7). Results validate the tiny searching step length determined by the method in the paper is very close to the theoretical value.

Tab.3 Training sample sets

Tab.4 Test sample set

2.3 Algorithm validation

The algorithm with the optimized tiny searching step length is validated by taking Apstar-V, AsiaSat-6, AsiaSat-5, AsiaSat-10 as examples. In the process of searching Apstar-V, the variation of the AGC received by the electromagnetic sensor is shown in Fig.6. All testing results show the proposed method can accurately locate the target satellite. Meanwhile the determined tiny searching step length, which is a variable for different geostationary satellites, is much smaller than the traditional satellite finder strategy which step length is a constant for different geostationary satellites.

Fig.6 AGC variation in the course of the algorithm validation

2.4 Self-learning of the BP neural network

To improve the robustness and generalizability of the designed neural network in the paper, the result of the satellites finder will be added to the training sample sets. If the searching result is correct, the information about the target satellite will be added to the training sample sets of the specific BP neural network corresponding to the target satellite belonged class. If the first searching result is wrong, the repeating satellite finder with a manual set step length will not be done until the searching result is right and the information corresponding to the correct result will also be added to the training sample sets. When a new sample is added to training sets the trained BP neural network need to be retrained to realize the dynamic feedback learning of the BP and improve the robustness and generalizability of the neural network.

3 Conclusion

The existing problems of the traditional satellite searching strategies are studied in the paper.

It has been proved that the distribution density of geostationary satellites, azimuth, elevation and tiny searching step length are the main influence factors of satellite finder searching correctness. For a geostationary satellite, the azimuth and elevation are constants; the tiny searching step length is closely related to the distribution density of the satellite. So a reasonable tiny searching step length can improve the present situation of the satellite finder. In the paper the pattern clustering and neural network are applied to determine the tiny searching step length. The geostationary satellites are classified into different classes according to the longitude; the tiny searching step length of a geostationary satellite is calculated by the trained BP neural network corresponding to the belonged class. The test results validate the difference between the calculated tiny searching step length and the theoretical tiny searching step length is very small. In application, the test examples are added to the training example sets in time, and the trained BP neural network will be retrained to improve the robustness and generalizability of BP neural network when a new geostationary satellite is added to the training sets. Now the improving searching strategy has been applied to the new small non-stationary satellite antenna. Experiments validate that the optimized strategy can avoid locating the error target satellite and improve the efficiency of satellite finder.

[1] Omihiko Y, Kohei O，Masazumi U. Highly accurate and cost-effective auto-tracking antenna system for satellite broadband communication on vessels[J].NTT Techinical Review,2005,3(9):60-71.

[2] Heev S D. Airborne satellite antenna mount and tracking system[C]∥7th African Conference in Africa, Gaborone, 2004: 289-295. DOI: 10.1109/AFRICON.2004.1406676.

[3] Tao K Y,Jin Y W, Yan Y M. Control algorithm for servo tracking system of carrier-based satellite antenna[J]. Ordnance Industry Automation,2006,25(8):55-58. (in Chinese)

[4] Cnsat. Finder parameters in China[EB/OL]. [2013-06-29]. http:∥www.cnsat.net/. (in Chinese)

[5] Wu W L, Liu X L, Ren J. The studies on critical technologies of servo system of vehicle satellite communication in motion[J]. Journal of the Hebei Academy of Sciences, 2010,27(1):31-34. (in Chinese)

[6] Bian Z Q, Zhang X G. Pattern recognition[M].Beijing: Tsinghua University Press,2008. (in Chinese)

[7] Graupe D. Principle ofartificial neural network[M].New York: World Scientific Publishing Co.,1997.

(Edited by Wang Yuxia)

10.15918/j.jbit1004-0579.201524.0318

TN 927.23 Document code: A Article ID: 1004- 0579(2015)03- 0398- 07

Received 2013- 10- 14

Supported by Academic Innovation Project of Beijing(201106149)

E-mail: cxy8888@bit.edu.cn