You are on page 1of 35

Geo-Stationary

Satellite Link
Contents:
1. Abstract

2. Introduction

3. Objectives

4. Applications and Depth and Breadth of secondary research of Satellite Communication


Systems

5. Depth of Learning using AI in Satellite Communications

6. Design of Geo-Stationary Satellite Up-link and Down-link Model for S-Band using RF
Satellite Simulink

7. MATLAB Simulation Results and its discussions

8. Results and Analysis of RF Satellite Simulink Design

9. Conclusions

10. References
Abstract:
In this paper Satellite communication link is designed from uplink and downlink keeping
geostationary orbit into consideration. We will take S-band into consideration. The uplink
power analysis is done from Ground station to uplink satellite transponder and then Downlink
power from Satellite transponder to receive station in earth receiving stations. The results of
the link budget will analyse the various Gain parameters, Energy per bit to Noise power
Spectral density, Bit error rate and Capacity of Geostationary communication Satellite.
Introduction:
Here in this RF link, we design and change uplink parameters from Ground transmitting
station to Satellite Transponders with various propagation losses like “Free space losses,
Atmospheric losses like- rain losses, Antenna de-pointing losses, Polarization losses”. Here
the Uplink signal from ground to Satellite is powerful so that it can overcome atmospheric
losses with narrow antenna beamwidth.

While on the Downlink path from Satellite to ground receiving stations it will retransmit
signal to the receiving station. While calculating link margins we have to consider various
parameters like “Antenna design, edge of the coverage zone, rain attenuation on either path,
Link outage due to solar flares due to sun spot cycles, Solar eclipses, equinoxes”.

Signal on the ground is processed from ground station to reach Satellite. The processing steps
include modulation, frequency conversion and proper amplification to reach Satellite
transponder by using narrow beam antenna. Due to free space losses signal received at the
Satellite transponder is weak and proper modulation, frequency conversion and signal
regeneration is done in transponder and then Satellite Antenna downlinks the processed
signal from satellite to various ground receiving station [1].
Objectives:
In this report we discuss the net power calculation from transmit earth station for S-Band
(Uplink =3.35 GHz and Downlink=3.15 GHz) Geostationary satellite stationed in Muscat,
Oman taking into consideration atmospheric attenuation and rain losses before reaching
uplink signal to the Satellite. Finally, we evaluate the downlink parameters from satellite
Transponders to Ground receiving stations in various location of Oman. Rest of the report
discusses the Satellite Link budget of uplink and downlink and noise and various sky noise
temperatures. Result shows all factors and calculates the status of the Geostationary Satellite
link
Satellite Link Budget Mathematical Analysis:

Here in Satellite Link design relation between Transmit power and receive power is
described. Here free space link or Radio frequency link is employed. Various Transmission
Parameters is employed like Antenna Beamwidth, Antenna Gain, Free space path loss and
basic power link equation is employed. Here System Noise is computed and parameters like
Noise Power, Noise Temperature, Noise Figure and Figure of merit are defined. The Carrier
to Noise ratio are used to define Satellite link design. The Flux density is the ratio of Power
flow and Unit Area. The Flux density and link equation can be used to calculate power
received by an earth station from a satellite transmitter with output power P t watts and driving
a lossless antenna with gain Gt, the flux density in the direction of the boresight at a distance
of R meters is

Ψ= PtGt /4πR2 (Watts/meters) … (1)

PtGt is called the Effective Isotropic Radiated Power because an Isotropic radiator with an
equivalent power equal to PtGt would produce same flux density at all directions.

For an ideal receiving antenna having an aperture of A m2 would collect a power of Pr watts
given by

Pr =Ψ*A = PtGt /4πR2(Watts) … (2)

The product PtGt is called Effective Isotropic Radiated Power since an isotropic radiator
with an equivalent power equal to PtGt would produce same flux density at all directions.

The Received Ideal Antenna Gain is given by:

Gr=4πA/λ2 => A=Gr λ2/4π … (3)

Thus

Pr=PtGtGr /(4 πR / λ)2 … (4)

Equation (4) is used to calculate Received power at any radio link. The term (4πR/ λ) 2 is
known as the path loss factor ( Lp ). It accounts for the dispersion of energy as an
electromagnetic wave travels from a transmitting source to destination in three-dimensional
space. For a real antenna however physical aperture area Ar ,Effective aperture area Aeand
aperture efficiency η A is related by equation (5)

Ae = η A Ar …(5)
For a real antenna

Pr=PtGt η A Ar /4 πR2 (Watts) … (6)

The link equation is expressed as:

Power Received= (EIRP x Received Antenna Gain)/ Path Loss (Watts) … (7)

Using Decibel notations, Equation (7) can be simplified as:

Pr=EIRP+Gr−Lp ( dbW ) … (8)

Where

EIRP = 10log ( PtGt )(dbW )

Gr =10 log(4πAe/λ2) (dB)

Lp = 20 log (4πR/λ) (dB)

Signal Attenuation:

The Path loss components of equation (8) is the algebraic sum of various loss components
such as: Losses in the atmosphere due to attenuation by air, water vapour and rain, losses at
the antenna at each side of the link and possible reduction in antenna gain due to antenna
misalignment due to poor operation of Attitude and Orbit Control subsystem of Satellite [4]
[5]. This need to be incorporated to allow sufficient system margin. Thus equation (8) can be
rewritten as equation (9):

Pr=EIRP+Gr−(lta+ lra+latm +lrain+lpol+lpt +…) … (9)

Where lta =Attenuation due to transmit antenna

lra = Attenuation due to receive antenna

latm = Attenuation due to atmosphere

lrain = Attenuation due to precipitation

lpol = Attenuation due to polarization

lpt = Antenna pointing misalignment related Attenuation


Applications and depth and Breadth of Secondary Research of
Satellite Communications Systems

Machine learning

To solve a problem in machine learning, the machine is provided the context to learn the
rules by itself. An ML is trained to rather than programmed into explicit rules. The learning
process requires data to extract pattern and hidden structures. The solution is to find the
optimal representation of the data to get closer to the expected result by searching with a
predefined space of possibilities using guidance from feedback signal where representation of
the data refers to different ways to look at or encode the data.

Three things are needed to achieve that -Input data, samples of the expected output and a way
to measure the performance of the algorithm.[2]

ML can be classified as Supervised, Unsupervised, Semi-supervised and Reinforcement


learning. During Supervised learning the training data s labelled that is tagged with correct
answer as it can check whether its predictions are right or wrong at any point in the training
process. During Image Classification, the algorithm is provided with images of different
classes and each image is tagged with the corresponding class. The supervised model learns
the patterns from training data to then able to predict labels for non-labelled data during
inferencing. These supervised learning is applied for classification and regression tasks.

Sometimes labelling is impossible due to lack of information as more unlabelled data is there.
So Unsupervised learning is employed during training. Using unlabelled data, the model can
extract hidden patterns or structures that is useful to understand a certain phenomenon or its
output could be used as an input for other models. Unsupervised Algorithms are employed
for Clustering, Anomaly detection, Association and Autoencoders.

A third technique is called Semi-supervised learning allows a mixture of nonlabelled and


labelled portions of training data. Semi-supervised learning is thus an excellent when only a
small part of the data is labelled or the labelling process is difficult or expensive. Pseudo
labelling is used to improve supervised models.

Various Supervised Algorithms are


1. SVM- Support Vector Machines which aims to find the decision boundary to classify
data inputs.[2][3] The algorithm maps the data to a high-dimensional representation
where the decision boundary is expressed as hyperplanes. The Hyperplanes is then
searched by trying to maximize the distance between the hyperplane and the nearest
data points from each class in a process called maximizing the margin.
2. Decision Trees- Decision tress is a supervised learning model that represents features
of the data as a tree by defining conditional control statements.[4][5] A robust version
is RF (Random forests) that combines various decisions trees to bring optimized
results. This involves building many different weak decision trees and then
assembling their outputs using bootstrap aggregation(bagging).[6][7]
3. Neural Networks (NN)- The Neural network has different layers of interconnected
nodes where each node is a perceptron that feeds the signal produced by a multiple
linear regression to an activation function. This function is used to add more
complexity to model by eliminating linearity. NN can be used for regression for
predicting continuous values and for classification by predicting probabilities for each
class.[8][9]

Deep Learning:
This is classified as DL that requires high computational resources.[2][10] Due to
recent computational advancement and the automation of feature engineering which
have paved the way for DL algorithm to surpass classical ML algorithm for solving
complex tasks such as Computer vision and natural Language processing.
1. CNN- Convolution Neural Networks [11][12] are deep NN that composed of
input layers, hidden convolutional layers and an output layer used in computer
vision [13] and image classification [14], object detection [15], object tracking
[16]. An activation function is chosen and followed by adding convolutional
layers. CNN architectures are defined by choosing the sizes, numbers and
positions of filters(kernels) and the activation functions.
2. RNN- Recurrent neural networks [17] are another type of neural networks in
which nodes directed graphs along a temporal sequence when previous outputs are
used as inputs. RNN uses its internal memory to process variable length sequences
of inputs. RNN are mostly used in the field of natural language processing, speech
recognition and music composition.

Reinforcement Learning:

Reinforcement learning [18] is about learning what actions to take in the hope of
maximizing rewards. The agent must search which actions bring the most rewards. These
actions can affect immediate rewards as well as subsequent awards. One of the challenges
encountered during RL is finding balance between exploitation and exploration.

To get maximum reward an RL agent must perform exploitation that is choose actions
that it has explored previously and found it to be best. Finding such actions requires exploring
solution space and try new actions.

All RL agents have explicit goals are aware of some aspects of that environment and take
actions that impact the environment. A RL agent system deals with four systems -a policy, a
reward signal, a value function and sometimes a model of the environment.
Depth of Learning using AI in Satellite Communications:

1. Beam hopping- Beam hopping is formulated as optimization problem and


modelled as Markov decision process. Deep Reinforcement learning is used to solve
beam-hopping illumination design and optimize the long term accumulated rewards of
the model Markov Decision Process (MDP). By applying Deep Neural Networks
(DNN) and Deep Reinforcement Learning (DRL) [19][20] beam hopping can achieve
greater flexibility in managing non-uniform requests throughout the lifetimes of the
satellite over the coverage area.

2. Anti -Jamming- By using Long short term memory (LSTM) which is Deep
learning Recurrent Neural Network (DL RNN) [21][22] is to learn the temporal trend
of signal. It demonstrated a reduction of overall synchronization of Frequency
Hopping Frequency Division Multiple Access (FH-FDMA).

3. Network Forecasting- Combined Fractional Auto Regressive Integrated Moving


Average (FARIMA) [23][24] with NN for internet traffic forecasting. Due to high
complexity of classical NN a least squares SVM an optimized version of SVM was
also used for forecasting.
4. Channel Modelling- Using deep CNN on satellite images and other input
parameters to predict the reference signal received power (RSRP) [25] for specific
receiver location in a specific scenario.
5. Telemetry Mining- Telemetry is the process of recording and transferring
measurements for control and monitoring. Satellite health clustering has been
performed with probabilistic clustering [26], dimensionality reduction and Hidden
Markov [27] and Regression trees [28]. While other have been developed for anomaly
detection methods like K-nearest neighbours, SVM and LSTM.[29]-[31]
6. Remote Sensing- Remote sensing is the process of extracting information about an
area, object or phenomenon by processing its reflected and emitted radiation at a
distance generally from satellite or aircraft. DL algorithm is used to classify land
coverage and crop types using Remote sensing images [32]. Zhang et al combined
DNN using a gradient boosted CNN for classification.[33]
7. Energy Management- DNN Compression usage before data training to
improve latency and save power. In the absence of solar light, Satellites are battery
energy dependent which places load on battery and shorten their lifetimes. Even DRL
allows certain energy savings compared to previous models.[34][35]

Satellite Communications Emerging Technology and Future Scope:

New Paradigm for satellite design is on the use of Technology like Computer aided design to
customise the communication Payloads.

Solar cell performance is enhanced by using Gallium arsenide or germanium multi-junction


cells. Efforts are going to improve battery like Lithium ion and fuel cell technology to
produce high powered Satellites.

Policy and regulatory issues like international trade landing rights agreements, annual
licensing fees agreements, non-tariff barriers, frequency allocation, orbit slot and
effectiveness of inter-coordination satellite system.

Security and privacy issues related with satellite communication systems to be resolved. And
there are protocols to be used for seamless interconnection of satellite, wireless and terrestrial
fibre networks. Apart from this Laser satellite beam technology is been emerging. Space
debris is one of the key concerns of international space station and different solution to tackle
it is emerging.

The advantage of Satellite communication system is remote access of areas like inaccessible
Mountainous regions of Siberia and Russia which is divided into 11 time zones and
archipelago of 3000 islands like in Indonesia.
Design of Geo-Stationary Satellite Up-link and Down-link model
for S-Band using RF Satellite Simulink:

RF Satellite Uplink

Fig 4 RF Satellite Uplink Design


RF Satellite Downlink

Figure 5. RF Satellite Downlink Design


Descriptions of design block of RF Satellite Link:

Here in this RF Satellite link design, there are many block whose details are given below

1. Bernoulli Binary Generator- This block generates random binary numbers using a
Bernoulli distribution. This block is used to generate random data bits to simulate digital
communication systems and obtain performance metrics such as bit error rate. The Bernoulli
distribution with parameter p produces zero with probability p and one with probability 1-p.
This Bernoulli distribution has mean value 1-p and variance p(1-p). The Probability of
zero parameter specifies p and can be any real number in range [0, 1].

Figure 6a Bernoulli Binary Generator

2. Rectangular Baseband QAM Modulator- The Rectangular QAM Modulator Baseband


block modulates with Mary quadrature amplitude modulation with a constellation on a
rectangular lattice. The output is a Baseband representation of the modulated signal.

Figure 6b QAM Baseband Modulator


3. Raised Cosine Receiver Filter (Square Root)- The square root raised cosine filter produces
a frequency response with unity gain at the low frequencies and complete at the higher
frequencies. It is commonly used in communication systems in pair, where the transmitter
first applies a square root raised cosine filter, and then the receiver applies a matched filter to
the block that normalizes the filter coefficients to unit energy. If you specify a Linear
amplitude filter gain, then the block scales the normalized filter coefficients using the gain
value you specify in this case.

Fig 6c Raised Cosine filter (Square)

4. HPA nonlinearity with option digital predistortion

 The parameter is used to determine how close the satellite high power amplifier is driven to
saturation. The selected back off is used to set the input and output gain of the Memory less
Nonlinearity block.

5. Free Space Path Loss


The Free Space Path Loss block simulates the loss of signal power because of the distance
between transmitter and receiver. The block reduces the amplitude of the input signal by an
amount that is determined in either of two things:
1-By the Distance (km) and Carrier frequency (MHz) parameters.
2-By the Loss (dB) parameter.

6. Phase/Frequency offset

The Phase and Frequency Offset block applies phase and frequency offsets to an incoming
signal. There are two situations for applying phase and frequency offsets to the input signal:

1-Define and set up your phase frequency offset object.


2-Apply phase and frequency offsets to the input signal according to the properties
of comment Phase Frequency Offset. The specific steps available to each object in the
toolbox.

Figure 6d RF (Phase/ Frequency Offset)

7.The IQ Imbalance Compensator

IQ imbalance compensator deals with performance issue in the design of direct conversion


receivers, also known as zero intermediate frequency (IF) or homodyne receivers. Such a
design translates the received radio frequency (RF, or pass band) signal directly from
the carrier frequency (fc) to baseband using only one mixing stage.

Figure 6e. IQ Imbalance Compensator


8.The DC Blocker

Block removes the DC component of the input signal. And it’s a components that prevent the
flow of audio and direct current (DC) frequencies while offering minimum interference to RF
signals.

9.The AGC

The automatic gain controller (AGC) block adaptively adjusts its gain level to achieve a
constant signal level at the output.

Figure 6f. The AGC

10.Phase Noise:

Phase noise is added to receiver to form complex baseband signal. This block generates
filtered phase noise according to some specified spectral mask and adds it to the input signal.

The General Parameter of Phase Noise:

1- Phase noise level (dB/Hz)


2-Frequency offset (Hz)
3-Sample rate (Hz)
Figure 6g: Phase Noise

11.Gain

The Gain block generates the output by multiplying its input by a certain gain factor. It is
multiplied by a specified value. The gain is used as a numeric value, or as a variable or
expression in the Gain parameter field. The input and gain are then multiplied, and the
result is converted to the output data type using the specified rounding and overflow
modes. Gain at downlink Receiver is used to apply gain of the receiver parabolic dish
antenna. Gain at downlink Transmitter is used to apply gain of the transmitter parabolic
dish antenna.

Figure 6h. Gain

12. Doppler correction

 It is the dispersion in measured energy caused when gamma rays from moving nuclei are
observed using a stationary detector whose opening angle is large enough to accept varying
angles of incidence. The effect gets worse as the velocity of the nucleus increases and as the
detector size (and hence opening angle seen by the projectile nucleus) is increased. Doppler
broadening can be corrected, but to do so the angle between the origin of the gamma rays and
the detection point should be known.
13. Display

Display value of variable without printing the variable name. Other way to display a
variable is to type its name.

Figure 6i. Display


Matlab Simulation Results and Discussions:
The results of simulation provide users to examine the effects of all parameters. By changing
the parameters we can change optimise the Satellite Uplink and Downlink design. Satellite
Link design budget is used for finding relationship between different parameters.
Figure 7. Throughput Calculation in Geo Stationary Satellite Link

Here in this figure 7. Throughput of Channel is increased by adopting Multiple antennas at


the Transmitter side and Multiple antennas at the receiver side. Using this Approach as the
Carrier to Noise ratio increases it increases the Channel Capacity and more users can use the
spectrum or it leads to better frequency Utilization. Here we choose Multiple Input multiple
Output Antenna arrays in both transmitter and receiver. In this scenario we have Antenna
Transmitter arrays as 16 and Antenna Receiver arrays as 16. Here Carrier frequency of
30GHz is used. It can be observed from this figure that as we increase the Signal to Noise
ratio (SNR) the channel capacity increases or in other words Throughput of the Geostationary
Satellite link increases.

Throughput of the Channel is defined by Mathematical formula: Th=1+(S/N).

Antenna arrays allows the reuse of spatial information for better coverage. Here in this Plot
we use Spatial Multiplexing. The idea of Spatial Multiplexing is to separate the channel
matrix to multiple modes so that the data stream from different elements in the transmit array
can be recovered independently from the received signal. The common way to transmit data
in a MIMO system is to uniformly split the power among transmit elements. However, the
capacity of the channel can be further improved if the channel is known at the transmitter. In
this case, the transmitter could use the water-fill algorithm to make the choice of transmitting
only in the subchannels where a satisfying SNR can be obtained. The following figure shows
the comparison of the system capacity between the two power distribution schemes. The
result confirms that the water-fill algorithm provides a better system capacity compared to the
uniform power distribution. The difference gets smaller when the system level SNR
improves.

Here it can be observed from the figure that as Signal to Noise (SNR) ratio increases towards
50dB the Channel Capacity increases towards 180 bps/Hz.
Figure 8. C/N Versus Receiving Antenna Power Pr and Transmitted Power Pt for Geo
Stationary Satellite Link

Here this figure 8 depicts graph between Carrier to Noise ratio(C/N) up and Transmitting
Power Pt from the Uplink ground station antenna and graph between Carrier to Noise
ratio(C/N)dn and Downlink Receiving Ground earth station antenna Power Pr. Here S-Band
is taken for Comparison analysis. Here the distance of Satellite is at 40,000Km distance from
Earth station. Satellite Antenna Gain is taken as 31dB. Receiver System Noise Temp is taken
at 500K.Transponder saturated output Power is taken at 80W. Earth Station Antenna
Diameter s considered as5m.Earth Station Aperture Efficiency = 68%. The Uplink Frequency
is taken at 3.35 GHz The Required C/N at the input of transponder is taken as 30dB. Here
Downlink Frequency is taken at 3.15 GHz. It has been observed from the Figure 5 that as
Transmitted Power Pt (in Watts) increases the threshold of 30kW the Carrier to Noise
ratio(C/N) up increases after crossing the threshold of 45dB. Similarly in the receiving side as
the Power of the receiving antenna Pr increases and crosses the threshold of 60MW the
Carrier to Noise ratio(C/N)dn crosses the threshold of 35dB.
Figure 9. Bit error rate vs Eb/N0 for Geo stationary Satellite link
Here figure 9 depicts the Bit Error Rate (BER) vs Energy per bit to Noise Spectral Density
(Eb/N0) for different Modulations like Binary Phase Shift Keying (BPSK), Quadrature Phase
Shift Keying (QPSK),8-PSK,16-PSK,32-PSK, Differential Binary Phase Shift Keying(D-
BPSK), Differential Quadrature Phase Shift Keying(D-QPSK),4-Quadrature Amplitude
Modulation(4-QAM),8-QAM,16-QAM and 64-QAM in Satellite Communication Link. Here
Eb/N0 range is taken as 25dB.

Here for the optimum BER different modulation techniques is adopted but it has been
observed from the figure that the performance of QPSK is best out of all modulation
techniques as it has smaller Bit error rate upto 10^-6. BPSK is better than other modulation
techniques n second scenario. The channel is considered as AWGN (Additive white Gaussian
Noise) and different uplink modulation techniques is applied in S-Band Geostationary
Satellite Link which lead to Higher Uplink margin and increases the overall Link margin and
lowest Bit error rate.
Figure 10. Estimated Outage probability vs Transmit Power (dBm) for Geo stationary
Satellite Link

Here in this Figure 10, Outage probability versus Transmit power of Ground Transmitting
station of Geostationary Satellite Link is depicted. First we set the power allocation factors as
A=0.25 and B=0.75 and make sure that the far user is given the higher fraction power. To
plot the Outage Probability versus Transmit Power initialization of power range from 0 to 40
dBm is done. Setting Bandwidth at 1MHz and generate Additive White Gaussian Noise
(AWGN) channel coefficients. To plot the outage probabilities, set target rate for each user.
For example, for user 1, we set the target rate as 1 bps/Hz and for user 2, we set the target rate
as 2 bps/Hz. Count the number of times the values calculated in step 5 fall below the target
rates and take the average. From the depicted figure above, it has been observed that Far user
Outage Probability of signal as compared to Near User Outage Probability with respect to
transmit power is more and it is clearly stated that as Uplink Ground station Transmit Power
increases the Probability of Outage decreases.
Results and Analysis of RF Satellite Simulink Design:

Figure 11. 16-QAM Constellation of RF-Satellite Link Design

Here in this figure 11.16-QAM constellation (red) with the received QAM constellation
before demodulation (yellow) s displayed. Comparing these constellation diagrams allows to
view the impact of all the RF impairments on the received signal and the effectiveness of the
compensations. We have Considered Link distance of Muscat from earth station to space as
40000km, Elevation angle=54 degrees, Longitude=24 degrees, Latitude=57degrees and
considered S-Band link with Uplink Frequency =3.35GHz and Downlink
frequency=3.15GHz and Receiver Noise temperature =500K, Path loss factor =190.5dB.
We considered Antenna efficiency as 68% and Transmitting Antenna diameter as 5m and
receiving Antenna diameter as 0.7m

Figure 12. HPA Constellations of RF-Satellite Link Design.

Here in this figure 12. Comparison of the constellation of the transmitted signal before
(yellow) and after (blue), the High Power Amplifier (HPA) is taken. The amplifier gain
makes the HPA Output signal to be larger than the HPA Input signal. Given plot enables you
to view the combined effect of both the HPA nonlinearity and digital predistortion.
Fig 13. Transmitted and Received Signal analysis of Spectrum with respect to Frequency

Here in this figure 13. Power Spectrum is obtained by double-clicking this Open Scopes
block which enables you to view the spectrum of the modulated/filtered signal (yellow) and
the received signal before demodulation (blue).
Conclusion:
The simulation results give users various parameters to optimize the Geostationary Satellite
link. We can simulate in MATLAB for Bit Error Rate, Carrier to Noise spectral density,
Outage capacity and Channel Throughput and can check Received signal Constellation and
Transmitted as well as received output using Power Spectrum Analyzer using RF Simulink
by taking into consideration Geographical location parameters. While designing a
Geostationary Satellite link we have to consider various atmospheric losses and optimize the
Satellite link margin.
References:
[1] Nadeem, Farukh, et al. “Weather effects on hybrid FSO/RF communication link.”

IEEE journal on selected areas in communications 27.9(2009).

[2] F. Chollet, “What is Deep Learning?” in Deep Learning with Python, 1st ed. New York,
NY, USA: Manning, 2017, ch. 1, pp. 3–24.

[3] C. M. Bishop, “Kernel Methods,” in Pattern Recognition and Machine Learning, 1st ed.

[4] J. R. Quinlan, “Induction of decision trees.” Machine learning 1.1, 1986, 81–106.

[5] C. M. Bishop, “Graphical Models,” in Pattern Recognition and Machine Learning, 1 st


ed. Berlin, Heidelberg, Germany: Springer-Verlag, 2006, ch. 8, pp. 359–423.

[6] L. Breiman, “Random forests,” Machine learning 45.1, 2001, pp. 5-32.

[7] L. Breiman, “Bagging predictors,” Machine learning 24.2, 1996, pp. 123-140.

[8] P. Baldi and K. Hornik, “Neural networks and principal component analysis: Learning
from examples without local minima.” Neural networks 2.1, 1989, pp. 53–58.

[9] C. M. Bishop, “Neural Networks” in Pattern Recognition and Machine Learning, 1st ed.
Berlin, Heidelberg, Germany: Springer-Verlag, 2006, ch. 5, pp. 225–290.

[10] I. Goodfellow, Y. Bengio, and A. Courville, “Introduction,” in Deep learning,


Cambridge: MIT press, 2016, ch. 1, pp. 1–26.

[11] I. Goodfellow, Y. Bengio, and A. Courville, “Convolutional Networks,” in Deep


learning, Cambridge: MIT press, 2016, ch. 9, pp. 326–366.

[12] S. Albawi, T. A. Mohammed, and S. Al-Zawi, “Understanding of a convolutional neural


network,” International Conference on Engineering and Technology (ICET), Antalya, 2017,
pp. 1–6.

[13] T. He, Z. Zhang, H. Zhang, Z. Zhang, J. Xie, M. Li “Bag of tricks for image
classification with convolutional neural networks.” Proceedings of the IEEE Conference on
Computer Vision and Pattern Recognition. 2019.
[14] Z. Zou, Z. Shi, Y. Guo, and J. Ye, “Object detection in 20 years: A survey,” arXiv
preprint arXiv:1905.05055, 2019.

[15] Q. Chu, W. Ouyang, H. Li, X. Wang, B. Liu, and N. Yu “Online multi-object tracking
using CNN-based single object tracker with spatial temporal attention mechanism,”
Proceedings of the IEEE International Conference on Computer Vision. 2017.

[16] K. R. Chowdhary, “Natural language processing,” Fundamentals of Artificial


Intelligence. Springer, New Delhi, 2020. pp. 603–649.

[17] I. Goodfellow, Y. Bengio, and A. Courville, “Sequence Modeling: Recurrent and


Recursive Nets,” in Deep learning, Cambridge: MIT press, 2016, ch. 10, pp. 367–415.

[18] RS. Sutton and GB. Andrew, “Reinforcement Learning: An Introduction,” A Bradford
Book, Cambridge, MA, USA. 2018.

[19] L. Lei, E. Lagunas, Y. Yuan, M. G. Kibria, S. Chatzinotas, and B. Ottersten, “Beam


Illumination Pattern Design in Satellite Networks: Learning and Optimization for Efficient
Beam Hopping,” in IEEE Access, vol. 8, pp. 136655-136667, 2020.

[20] X. Hu, S. Liu, X. Hu, Y. Wang, L. Xu, Y. Zhang, C. Wang, and W. Wang, “Deep
reinforcement learning-based beam Hopping algorithm in multibeam satellite systems,” IET
Communications. pp. 2485–91, Jan. 2019.

[21] S. Lee, S. Kim, M. Seo, and D. Har, “Synchronization of Frequency Hopping by LSTM
Network for Satellite Communication System,” in IEEE Communications Letters, vol. 23, no.
11, Nov. 2019, pp. 2054– 2058.

[22] C. Han, L. Huo, X. Tong, H. Wang, and X. Liu, “Spatial Anti-Jamming Scheme for
Internet of Satellites Based on the Deep Reinforcement Learning and Stackelberg Game,” in
IEEE Transactions on Vehicular Technology, vol. 69, no. 5, May 2020, pp. 5331–5342.

[23] F. Xu, Y. Lin, J. Huang, D. Wu, H. Shi, J. Song, and Y. Li, “Big Data Driven Mobile
Traffic Understanding and Forecasting: A Time Series Approach,” in IEEE Transactions on
Services Computing, vol. 9, no. 5, pp. 796-805, 1 Sept.-Oct. 2016.

[24] C. Katris, S. Daskalaki, Comparing forecasting approaches for Internet traffic, Expert
Systems with Applications, Volume 42, Issue 21, 2015, 8172–8183, ISSN 0957-4174.
[25] J. Thrane, D. Zibar, and H. L. Christiansen, “Model-Aided Deep Learning Method for
Path Loss Prediction in Mobile Communication Systems at 2.6 GHz,” in IEEE Access, vol. 8,
2020, pp. 7925–7936.

[26] T. Yairi, N. Takeishi, T. Oda, Y. Nakajima, N. Nishimura, and N. Takata, “A Data-


Driven Health Monitoring Method for Satellite Housekeeping Data Based on Probabilistic
Clustering and Dimensionality Reduction,” in IEEE Transactions on Aerospace and
Electronic Systems, vol. 53, no. 3, June 2017, pp. 1384–1401.

[27] T. Yairi, T. Tagawa, and N. Takata, “Telemetry monitoring by dimensionality reduction


and learning hidden markov model,” in Proceedings of International Symposium on Artificial
Intelligence, Robotics and Automation in Space, 2012.

[28] T. Yairi, M. Nakatsugawa, K. Hori, S. Nakasuka, K. Machida and N. Ishihama,


“Adaptive limit checking for spacecraft telemetry data using regression tree learning,” 2004
IEEE International Conference on Systems, Man and Cybernetics (IEEE Cat.
No.04CH37583), The Hague, 2004, pp. 5130–5135 vol.6.

[29] T. Shahroz, L. Sangyup, S. Youjin, L. Myeongshin, J. Okchul, C. Daewon, and S. W.


Simon, “Detecting Anomalies in Space using Multivariate Convolutional LSTM with
Mixtures of Probabilistic PCA,” 25th ACM Special Interest Group on Knowledge Discovery
and Data Mining International Conference, Alaska, USA, 2019.

[30] K. Hundman, V. Constantinou, C. Laporte, I. Colwell, and T. Soderstrom, “Detecting


Spacecraft Anomalies Using LSTMs and Nonparametric Dynamic Thresholding,” 24th ACM
Special Interest Group on Knowledge Discovery and Data Mining International Conference.
London, UK, 2018.

[31] S. Fuertes, G. Picart, JY. Tourneret, L. Chaari, A. Ferrari, and C. Richard, “Improving
Spacecraft Health Monitoring with Automatic Anomaly Detection Techniques,” 14th
International Conference on Space Operations. Daejeon, Korea, 2016.

[32] N. Kussul, M. Lavreniuk, S. Skakun, and A. Shelestov, “Deep Learning Classification of


Land Cover and Crop Types Using Remote Sensing Data,” in IEEE Geoscience and Remote
Sensing Letters, vol. 14, no. 5, pp. 778–782, May 2017.
[33] F. Zhang, B. Du, and L. Zhang, “Scene Classification via a Gradient Boosting Random
Convolutional Network Framework,” in IEEE Transactions on Geoscience and Remote
Sensing, vol. 54, no. 3, pp. 1793–1802, March 2016.

[34] H. Tsuchida, Y. Kawamoto, N. Kato, K. Kaneko, S. Tani, S. Uchida, and H. Aruga,


“Efficient Power Control for Satellite-Borne Batteries Using Q-Learning in Low-Earth-Orbit
Satellite Constellations,” in IEEE Wireless Communications Letters, vol. 9, no. 6, pp. 809-
812, June 2020.

[35] B. Zhao, J. Liu, Z. Wei, and I. You, “A Deep Reinforcement Learning Based Approach
for Energy-Efficient Channel Allocation in Satellite Internet of Things,” in IEEE Access, vol.
8, pp. 62197-62206, 2020.

[36] https://de.mathworks.com/help/comm/ref/bpskdemodulatorbaseband.html

[37]https://www.mathworks.com/help/comm/ref/comm.phasefrequencyoffset-system-
object.html

[38]https://www.mathworks.com/help/simulink/ug/simulation-data-display-and-capture-
using-scopes.html

[39] https://www.mathworks.com/help/comm/ref/comm.thermalnoise-system-object.html

[40] https://www.mathworks.com/help/matlab/ref/disp.html

[41] https://www.mathworks.com/help/comm/ug/awgn-channel.html

You might also like