See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324422547

ARMA-Based Adaptive Coding Transmission Over Millimeter-Wave Channel

for Integrated Satellite-Terrestrial Networks

Article  in  IEEE Access · April 2018

DOI: 10.1109/ACCESS.2018.2825256

CITATIONS READS

18 765

5 authors, including:

Shushi Gu Jian Jiao

Harbin Institute of Technology (Shenzhen) Harbin Institute of Technology
86 PUBLICATIONS   298 CITATIONS    176 PUBLICATIONS   827 CITATIONS   

SEE PROFILE SEE PROFILE

Zixuan Huang Qinyu Zhang

UNSW Sydney Harbin Institute of Technology Shenzhen Graduate School
6 PUBLICATIONS   59 CITATIONS    414 PUBLICATIONS   2,498 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Hybrid Carrier based on WFRFT View project

National Natural Sciences Foundation of China (NSFC) View project

All content following this page was uploaded by Zixuan Huang on 27 September 2018.

The user has requested enhancement of the downloaded file.

 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2018.DOI

ARMA-based Adaptive Coding

Transmission over Millimeter-Wave
Channel for Integrated
Satellite-Terrestrial Networks
SHUSHI GU, JIAN JIAO, (Member, IEEE), ZIXUAN HUANG, SHAOHUA WU, (Member, IEEE),
QINYU ZHANG, (Senior Member, IEEE)
Communication Engineering Research Center, Harbin Institute of Technology (Shenzhen), Guangdong 518055, China
Corresponding authors: Jian Jiao (jiaojian@hit.edu.cn); Qinyu Zhang (zqy@hit.edu.cn)
This work was supported in part by the National Natural Sciences Foundation of China under Grant 61771158, Grant 61701136, Grant
61525103 and Grant 61371102, and in part by the Shenzhen Basic Research Program under Grant JCYJ20170811154309920, Grant
JCY2017081114233370, Grant JCYJ20170811160142808 and Grant JCYJ20160328163327348.

ABSTRACT The integrated satellite-terrestrial network can provide broadband wireless access in a
wide coverage, efficient, and cost-effective manner, and is regarded as one of the most promising
infrastructures for future heterogeneous network towards fifth generation (5G). With the development
of next generation of high throughput satellites (HTS), the application of the millimeter-wave (mmWave)
band HTS is viewed as a vital role in the future integrated satellite-terrestrial network. Considering the
rain attenuation is the dominant fading factor of the mmWave channel, we first propose a practical time-
varying rain attenuation prediction model based on the autoregressive–moving-average (ARMA) model.
Then, we develop an adaptive coding transmission (ACT) scheme based on the analog fountain codes
(AFC) combine the ARMA mmWave channel prediction model. The key parameters are selected based on
a tradeoff between the decoding failure probability, block length, and overhead for our ARMA-based ACT
scheme. Simulation results show that our proposed ACT scheme can effectively improve the throughput.

INDEX TERMS Integrated satellite-terrestrial network, millimeter-wave channel, autoregressive-moving-

average model, rain attenuation, adaptive coding transmission

I. INTRODUCTION mmWave channel, several optimization schemes have been

proposed [7]–[9]. In particular, the rain attenuation at the
ILLIMETER-WAVE (mmWave) band high through-
M put satellites (HTS) are viewed as backbone relays to
enhance the throughput of the integrated satellite-terrestrial
Ka-band is much higher than at the X-band (in the order
of 1.5 ∼ 4.5 dB/km higher for a rain rate ranging from
10 ∼ 30 mm/hour [10]). Thus, appropriate link margins
networks towards fifth generation (5G) [1], [2]. The archi-
should be allocated to take into account these atmospheric
tecture of future integrated satellite-terrestrial networks is
impairments and to reduce the target outage probability [11].
shown in Fig. 1, where several mmWave band (e.g., Ka/Q/V-
band) high-throughput geostationary Earth orbit (GEO)
There are a number of advanced link designs that have
satellites can be connected to form the backbone network,
been proposed in the literature to mitigate the effect of
and to support global ubiquitous broadband wireless access
rain attenuation and system noise temperature on the Ka-
of the integrated space/air/terrestrial information network in
band HTS return channel [12], [13]. Moreover, compared
the future [3], [4].
to the terrestrial communications, the space nodes in the
However, the mmWave band HTS return channel is integrated satellite-terrestrial networks are limited by re-
known to be much more sensitive to the atmospheric at- source limitations with respect to performing complicated
tenuation than the conventional S/C/X-band channels [5], rain attenuation prediction algorithms [14]. In general, a
[6]. In order to further enhance the throughput of the timely decision on the short-term weather state greatly

VOLUME 4, 2016 1

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

Inter-satellite laser links

Lunar
Ka/Q/V-band beam/beamforming Orbiter
Ka/Q/V-band free-space links
Ka/Q/V-band links with rain attenuation
Mars Orbiter

GEO High-Throughput Satellites

Backbone Networks

Low Earth Orbit

lation
on
Satellites Constellation

Remote Sensing
Space
Satellite
Station

Terrestrial Wireless
Navigation Sensor Networks
Earth
Station
Cellular
Networks

FIGURE 1. The architecture of future high-throughput integrated satellite-terrestrial networks.

complicates communication system operations [15]. represent a more suitable scheme for the integrated satellite-
Furthermore, space communications suffer from the long terrestrial networks with large-capacity data and multi-hop
propagation delay, and conventional Automatic Repeat- transmissions [25], [26] than the existing polar codes [27],
reQuest (ARQ) schemes are inefficient. In fact, the Con- spinal codes, Low-Density Parity-Check (LDPC) codes, and
sultative Committee for Space Data System (CCSDS) File turbo codes.
Delivery Protocol (CFDP) and the Licklider Transport Pro- In this paper, we choose the autoregressive–moving-
tocol (LTP) are designed with a negative-acknowledgement average (ARMA) model [28] to design our forecasting
(NAK) mechanism [16], [17], and the space nodes can algorithm in Section II. This model can predict the N -state
obtain the delayed channel state information (CSI). Markov chain of rain attenuation on the Ka-band link as the
Therefore, in order to ensure a certain reliability in the yellow links in Fig. 1, for which the prediction error may
performance of the Ka-band HTS communications, one approach zero. In Section III, we propose an adaptive coding
needs to consider the tradeoffs between transmission latency, transmission (ACT) scheme over the Ka-band channel based
reliability, and onboard power allocations for the HTS on the analog fountain code (AFC), which is a rateless code
nodes [18]. Authors in [19] designed a simple estimation that combines the Luby Transform (LT) code with m-ary
scheme of CSI over the Ka-band channel with a delayed- modulation [29]. The proposed ACT scheme can approach
feedback ACK/NAK message. For that, a transmit power the Shannon limit over a wide signal-to-noise ratio (SNR)
allocation scheme can be proposed that maximizes the long- range. In Section IV, we analysis the ACT scheme based
term throughput [20]. on the ARMA prediction model for the Ka-band link in the
Hence, a rain attenuation prediction model to deal with a integrated satellite-terrestrial networks as shown in Fig. 1. In
time-varying channel state is needed, and the Markov state Section V, we show through simulations that the proposed
model could be used to describe the Ka-band mmWave link ACT scheme can increase system throughput and improve
state [21]. It is important to design a forecast algorithm transmission efficiency. Finally, conclusions are drawn in
based on the N -state Markov model to predict the weather Section VI.
state for the code-rate adaption on the sender side.
Moreover, with the estimation scheme of CSI, the space II. N -STATE MARKOV MODEL AND FORECASTING
nodes can choose the appropriate code rate channel codes A. PRELIMINARIES OF THE ARM A MODEL
to enhance the transmission performance. Considering the In this section, we design an N -state Markov chain to
CCSDS long erasure code (LEC) recommendation [22], model the rain attenuation variation at Ka-band channel. We
and existing related rateless coding schemes [23], [24], set the length of the transmission time slot is 90 seconds, and
approaching-capacity rateless codes with linear complexity we choose a 3 seconds sample interval due to the round trip
2 VOLUME 4, 2016

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

time of the Moon-to-Earth is about 2.64 seconds. Therefore, 1

we need to sample 30 times in each time slot and use

the average value to model the Markov channel. Moreover, 0.5

Autocorrelation
due to the huge distance and long propagation delay in
the integrated satellite-terrestrial networks, the transmitter 0
can only obtain delayed CSI from feedback. Therefore, in
this paper, our transmission system model adopts 1 time
−0.5
slot feedback CSI to design a practical time-varying rain
attenuation prediction model based on the ARMA model
for the integrated satellite-terrestrial networks. −1
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
In our previous work [30], we used the Van de Kam- Lag k

p (VDK) distribution-based model to design an N -state (a)

Markov chain. In this paper, by using the relationship 1
between the fade-slope standard deviation [28] and attenua-
tion level [31] from fade-slope statistics of rain attenuation

Partial Autocorrelation
0.5
measured from Xi’an, China on March 14, 2010, we use the
recommended statistical model in the International Telecom-
munication Union (ITU) standards to design an N -state 0

Markov chain to generate the rain attenuation time series

xt , which can simulate a given duration rainfall event to −0.5
help us design and verify the ARMA prediction algorithm.
Our channel estimation algorithm is designed by utilizing
−1
the ARMA model. In the ARM A(p, q) model, p is the order 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Lag k
of the autoregressive (AR) part AR(p), and q is the order
(b)
of the moving average (MA) part M A(q). The AR(p) part
is used to predict the future value, which is only related to FIGURE 2. The calculation result of (a) the ACF and (b) the PACF.
its previous predicted values, thus,
p
X
AR(p) : xt = φi xt−i + εt , (1) respectively, of the primary data xt . Thus, the first step
i=1
of establishing a rain attenuation forecasting ARM A(p, q)
where φ1 , φ2 , ..., φp are the parameters of the AR(p) model, model is to obtain the ρbk and ϕbk,k of sequence xt .
and εt is white noise. The M A(q) part is used to calculate On one hand, the ACF ρbk shows the overall correlation
the current accumulative errors of the predicted values by of xt and xt−k , and the equation is given as follows:
the AR(p) part, where
cov(xt , xt−k ) E(xt xt−k ) − E(xt )E(xt−k )
q ρbk = 2
= . (4)
σ σ2
X
M A(q) : xt = εt + θi εt−i . (2)
i=1 On the other hand, the PACF ϕ bk,k shows the immedi-
θ1 , θ2 , ..., θq are the parameters of the M A(q) model, and ate data dependency of xt and xt−k after removing all
εt is white noise. indirect correlations brought by the intermediate variables
Hence, the ARM A(p, q) model is established by com- xt−1 , xt−2 , ..., xt−k+1 . By using the Yule–Walker equation,
bining the above two parts and is given as follows: the PACF ϕ bk,k can be easily simplified as follows. First, for
p q j = 1, 2, ..., k, we have
X X
ARM A(p, q) : xt = φi xt−i + θi εt−i + εt , (3) ρbj = ρbj−1 ϕ bk,2 + · · · + ρbj−k ϕ
bk,1 + ρbj−2 ϕ bk,k ,
i=1 i=1
k
X (5)
where εt is white noise which follows a normal distribution = ρbj−r ϕ
bk,r ,
N (0, σ 2 ) with zero mean and variance σ 2 . r=1

B. KEY PARAMETERS OF THE PROPOSED ARM A

and ρbj = ρb−j . Then, the set of ρbj in Equation (5) can be
MODEL
solved by representing these equations in the matrix form
with j > 0 and ρb0 = 1. Thus Equation (5) is turned to
The key to establish our rain attenuation prediction
ARM A(p, q) model is the course of estimating and solving
    
ρb1 1 ρb1 · · · ρbk−1 ϕ
bk,1
the parameters in Equation (3). These parameters are closely  ρb2   ρb1 1 · · · ρbk−2 
 ϕ
 bk,2 
 ···  =  ···  · · · .
   
related to the autocorrelation function (ACF) ρbk and the ··· ··· ··· 
partial autocorrelation function (PACF) ϕ bk,k of sequence ρbk ρbk−1 ρbk−2 · · · 1 ϕ
bk,k
xt . ρbk and ϕbk,k indicate the trends and characteristics, (6)
VOLUME 4, 2016 3

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

Fig. 2 shows the values of ACF ρbk and PACF ϕ bk,k 15

corresponding to lag k, where lag is the number of afterward

10
time windows and the rain attenuation series are generated

Rain attenuation (dB)

from the N -state Markov chain. By the characteristics of 5

these two value series, we can initially determine the order

0
of the ARM A(p, q) model according to the principles
in [30]. The characteristics of the two value series can be -5

indicated by truncation, which is a situation where the time

-10
series will be zero after several lags. Origin rain attenuation series
First-order differential rain attenuation series
From Fig. 2(b), we can see that the value series of PACF -15
0 5 10 15 20 25 30 35 40 45 50
ϕ
bk,k is truncated temporarily at lag k = 5, which means Index of the ACF series

that the PACF ϕ bk,k value is almost equal to 0 when lag

(a)
k = 5 and the PACF ϕ bk,k value is close to 0 with Lag > 5.
However, the ACF ρbk value is oscillatory, as shown in Fig. 1

2(a), and we can see that the value series of ACF ρbk is not
truncated. In this case, as the the ACF ρbk value series are

Autocorrelation
0.5
decreased after lag k > 5, we choose the lag k = 6 as a
preliminary selection.
Furthermore, we can verify our preliminary selection for
0
the ACF by using the first order of the differential equation,
and an example of the improved ACF series to remove the
oscillatory by using the first order differential equation is
−0.5
shown in Fig. 3. The ACF of the first-order differential 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Lag k
rain attenuation series shown in Fig. 3(b) verifies that, if
the selected order of ACF is correct, the ρbk value series (b)
will be truncated after the original rain attenuation series is
FIGURE 3. The improved (a) Rain attenuation series and (b) its first-order differential.
processed by the first order differential equation.
Therefore, if the ACF ρbk and PACF ϕ bk,k value series are
truncated by lag k1 = 6 and k2 = 5, we can calculate the
order of ARM A(p, q) from p = k1 − k2 , q = k1 . The result
γk = E(xt xt−k ), for k ∈ {0, 1, ..., p}, and we solve θk as
of the order for our ARM A(p, q) model is: p = 1, q = 6.
follows:
Therefore, we can substitute the original data series xt into

the ARM A(1, 6) model as follows:  γ0 = σ 2 (θ0 θ0 + θ1 θ1 + · · · + θq θq )

γ1 = σ 2 (θ0 θ1 + θ1 θ2 + · · · + θq−1 θq )

ARM A(1, 6) : xt =φ1 xt−1 + θ1 εt−1 + · · · + θ6 εt−6 + εt ,



···

(7) . (9)
where εt is normal distribution N (0, σ 2 ) with zero mean 
 γk = σ 2 (θ0 θk + θ1 θ1+k + · · · + θq−k θq )
 ···

and variance σ 2 .


γq = σ 2 (θ0 θq + θ1 θ1+q + · · · + θ0 θq )


C. CALCULATION METHOD OF THE COEFFICIENTS

FOR THE PROPOSED ARM A(1, 6) MODEL
There are q + 1 equations in (9), where q undetermined
coefficients θk∈{1,2,...,p} , and one unknown parameter σ 2 .
Then, to calculate the rest coefficients φk , θk and σ 2 in the
In our ARM A(1, 6) model, p = 1, q = 6, φ1 = γ7 /γ6
ARM A(1, 6) model, such as those left in Equation (3), once
and θ0 = 1. To calculate these numerical results from the
again, we solve φk by utilizing the Yule–Walker equation
nonlinear equations (9), we can utilize several linear iter-
in a simple way. Thus,
  ation algorithms, such as the Newton–Raphson algorithm.
φ1 In this paper, we use the same N -state Markov channel to
 φ2  generate four rainfall events where the maximum rain fading
 ···  =
 
is 10 dB. Then, we use the ARMAX function in Matlab to
φp calculate the coefficients of our ARM A(1, 6) model, and
−1 
the results of the coefficient values are shown in Table 1.
 
γq γq−1 · · · γq−p+1 γq+1
 γq+1 γq · · · γq−p+2  Then, we can adopt the four data series shown in Table
  γq+2 .
 
1 as the initial input to generate the N -state Markov chain.

 ··· ··· ··· ···   ··· 
γq+p−1 γq+p−2 · · · γq γq+p As the length of the Markov chain sequences is set as 50,
(8) we can obtain the four sequences which correspond to their
Therefore, we have p coefficients as φk∈{1,2,...,p} , where respective estimation parameters by using the ARM A(1, 6)
γk is autocovariance function of the ARMA model, and model. To analyze each set of the previous 25 pilots of each
4 VOLUME 4, 2016

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

12
Predicted rain attenuation series
10 Origin rain attenuation series
Residual error series

Rain attenuation (dB)

8

-2

-4
0 5 10 15 20 25 30 35 40 45 50
Index of series

FIGURE 4. Comparison of the prediction values and the original value.

TABLE 1. Coefficient results of different serials in ARMA(1, 6).

1

Parameter Serial 1 Serial 2 Serial 3 Serial 4

φ1 0.848 –0.530 0.577 0.560

Autocorrelation
θ1 0.027 –1.685 –0.205 –0.268 0.5
θ2 –0.431 0.907 –0.391 –0.271
θ3 –0.078 –0.308 –0.147 –0.101
θ4 0.034 0.299 –0.121 0.012
θ5 0.220 –0.219 0.00004 –0.109 0
θ6 –0.020 0.043 –0.136 0.338

−0.5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Lag k
sequence, we have
(a)
xt =0.848xt−1 + εt + 0.027εt−1 − 0.431εt−2 −
(10)
0.078εt−3 + 0.034εt−4 + 0.22εt−5 − 0.02εt−6 . 1
Partial autocorrelation

We can use the latest 25 data values as pilots, and then 0.5
compare them to the upcoming 25 data values in order to
measure the prediction accuracy of our ARM A(1, 6) model.
The estimation error simulation result of the first sequence in 0
Table 1 is shown in Fig. 4, and the corresponding ACFs and
PACFs of the first-order differential rain attenuation series
are shown in Fig. 5. −0.5
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Lag k
In general, quantization of the veracity of forecast results
firstly includes the absolute error of predicted points, which (b)
is the difference between the predicted values and actual val-
ues. It has the same order as the prediction data. Secondly, FIGURE 5. (a) The ACF and (b) PACF of the first-order differential rain attenuation.
the relative error of predicted points is included, which is
the ratio of the deviation and the actual results. There is a
difference in the deviation between the predicted results and
actual results, the relative error has no unit and is usually III. ANALOG FOUNTAIN CODES AND OPTIMIZATION
expressed as a percentage. We use the mean absolute error Moreover, the conventional channel codes with fixed
to measure the prediction error. When the rain attenuation code-rate and the block length are unable to meet the time-
fading depth is 10 dB, the average prediction error is 0.0227 varying N -state wireless communication channel capacity
dB when using our ARM A(1, 6) model with respect to the at all times [32], especially if the duration of each channel
N -state Markov chain, taking the mean absolute errors. We state in Ka-band channel is very short (i.e., several seconds
will verify the performance of our ARM A(1, 6) prediction or minutes). When the rate is larger than the capacity,
model later in Section V. transmissions are said to unreliable [33], and when the rate
is smaller than the capacity, transmissions are said to be
VOLUME 4, 2016 5

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

need to be designed such that the stream of coded symbols

b1 b2 b3 ... bk
follows a Gaussian-like distribution.
Hence, both the degree distribution and weight set can
ɷ1,1 ɷ1,2 ɷ2,3 ɷ3,3 affect the performance of the AFC scheme. On one hand,
the optimization of the degree distribution is used to find
ɷ3,k ɷN-1,k ɷN,2 out the optimization condition via And-Or tree analysis, and
then solve the optimal results by utilizing linear program-
S1 S2 S3 ... SN-1 SN ming [36]. In the previous study [37], we found out that the
degree distribution of the AFC scheme is not the main factor
FIGURE 6. Analog fountain code (AFC) encoding scheme. with respect to the decoding performance. In contrast, in the
finite block length regime, a constant degree distribution is
preferred, e.g., all the AFC coded nodes [Sj ]j=1,2,...,N are
inefficient. with Ω(x) = x8 as in the practical application, without
Alternatively, when CSI is known a priori at the transmit- considering the unequal error protection [38], [39].
ter side, adaptive coded modulation (ACM) can be used [34].
B. WEIGHT SET OPTIMIZATION
However, ACM can only obtain a step-like throughput. In
this section, we consider a rateless code known as the On the other hand, the design of the AFC weight set
AFC for which the CSI is not needed at the transmitter. is important, an optimum selection weight set can improve
Instead, the rate is determined on the fly and, thus, achieves the decoding performance. An optimum weight set is that
a seamless performance across all channel states. which leads to the weight distribution of the received coded
symbols closer to the Gaussian distribution, e.g., the AFC
with weight set {±4, ±3, ±2, ±1} has a better decoding per-
A. AFC ENCODER
formance than with the weight set {±4, ±4, ±2, ±1} [36].
The adaptive coding modulation (ACM) is proposed to
enhance the transmission efficiency over the time-varying C. AFC DECODER
wireless channels. Recently in [29], authors proposed the
In this paper, considering the huge noise and frequent in-
AFC scheme, which is a combination of the binary fountain
terruptions of the Ka-band GEO satellites relaying commu-
code (FC) with the idea behind m-ary modulation. For this,
nications in the the integrated satellite-terrestrial backbone
the information bits are first modulated using binary phase-
network, we chose the compression perception belief propa-
shift keying (BPSK).
gation (CSBP) decoding algorithm to enhance the decoding
In a manner similar to binary FCs, AFCs can be rep- performance. The algorithm is detailed in [40]. By using the
resented in the form of a bipartite graph as shown in CSBP decoding algorithm, and choosing degree distribution
Fig. 6. In Fig. 6, the circles represent the information Ω(x) = x8 and the weight set ws = {±4, ±3, ±2, ±1},
symbols [bi ]i=1,2,...,k . They are commonly referred to as we will compare the decoding performance with different
variable nodes. The squares represent the coded symbols block lengths and code-rates. Furthermore, we compared the
[Sj ]j=1,2,...,N and are commonly referred to as check nodes. decoding performance with the LDPC codes in [41], where
Pd is chosen
First, a positive integer
d
from a predetermined the number of variable nodes is k = 100, and the code-rate
distribution Ω(x) = d Ωd x , where d is commonly of the LDPC codes are 0.95, 0.8 and 0.6, respectively.
referred to as the degree, and Ωd is the corresponding The simulation results of the decoding failure probability
degree probability. Then, d information symbols are chosen versus SNR are shown in Fig. 7(a), where the number of
uniformly at random and multiplied by d coefficients chosen variable nodes in the AFC scheme is k = 100, 200, and the
uniformly at random from a predetermined set ws . Finally, code-rate is R = k/n = {1, 0.8, 0.6}. When R = 1, the bit
the coded symbol is the sum of these weighted symbols. In error rate (BER) of the AFC with k = 100, 200 is about
particular, the j th coded symbol Sj can be expressed as 10−3 orders of magnitude when the SN R ≥ 10 dB. When
dj R = 0.6, the BER of the AFC is decreased to 10−5 when
Sj =
X
wj,i bi , (11) the SN R ≥ 5 dB, and the BER can decrease to 10−6 when
i=1
the SN R ≥ 10 dB.
As shown in Fig. 7, the AFC schemes have error floor
where dj is the chosen degree, and wj,i are the randomly problems compared to the LDPC codes. The LDPC code
selected weights from the set ws . with lower code-rate can achieve a very low error floor if the
Therefore, the binary elements “1” in conventional LT SNR is larger than its decoding threshold value. However,
codes generator matrix G are replaced by the corresponding the achievable rate of the fixed rate LDPC codes can perform
real weights wj,i in the AFC generator matrix. well only in a small range of SNRs.
Intuitively, the performance of AFC is dependent on the Fig. 8 shows the achievable rate of the AFC schemes
proper design of the degree distribution Ω(x) and the weight and LDPC codes from DVB-S2 [41] with different rate and
set ws [35]. From the Shannon limit, these two parameters modulation types in the AWGN channel, versus the SNR
6 VOLUME 4, 2016

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

100
k=100, R=1
k=100, R=0.8
A. DESIGN OF THE ADAPTIVE CODING TRANSMISSION
10-1
k=100, R=0.6
k=200, R=1 SCHEME
k=200, R=0.8

10-2
k=200, R=0.6
LDPC k=100, R=0.95
We assume a Ka-band GEO satellites relaying the com-
LDPC k=100, R=0.8
LDPC k=100, R=0.6 munication link can be divided into discrete transmission
10-3
time windows with constant duration, and the rain atten-
BER

10-4
uation is modeled by the N -state Markov chain. Without
loss of generality, we select a constant duration of each
10-5
time window (90 seconds) and a transmission data rate
10-6 (20kB/s), and then the number n of the coded symbols can
10-7
be transmitted in a time window is determined.
0 2 4 6 8 10 12
SNR (dB) According to the Markov theory, assuming that the sta-
(a) tionary distribution of N states is P~ = (P 1 , P 2 , . . . , P N ),
100 then P~ = P~ Q, where Q is the matrix of transition proba-
bility of the Markov chain, and P 1 + P 2 + · · · + P N = 1.
10-1 Moreover, we assume pi is the BER of the ith state in the N -
state Markov chain and p1 ≤ p2 ≤ ... ≤ pi ≤ ... ≤ pN . Let
decoding failure probability

10-2 L denotes the length of the coded symbol, and Pi denotes

the symbol loss probability in the ith Markov state, we have
10-3 k=100, R=1
k=100, R=0.8
k=100, R=0.6
Pi = 1 − (1 − pi )L . (12)
k=200, R=1
10-4 k=200, R=0.8
k=200, R=0.6
LDPC k=100, R=0.95 Then, let ni and ki denote the expected number of
LDPC k=100, R=0.8

10-5
LDPC k=100, R=0.6 the received coded symbols and the recovery information
0 2 4 6 8 10 12
SNR (dB) symbols in a transmission time window in the ith Markov
(b) state, respectively. The relationship between the number of
the transmitted symbols n and the received symbols ni is
FIGURE 7. Decoding performance of AFC and LDPC codes: (a) BER; (b) DFP.
ni = (1 − Pi ) · n, (13)

and the overhead Oi of the ith Markov state in a transmis-

4
sion time window can be calculated as
Shannon Limit

3.5
AFC
LDPC R=1/2, BPSK n − ki
LDPC R=1/2, QPSK Oi = . (14)
LDPC R=1/2, 8PSK ki
3 LDPC R=2/3, BPSK
LDPC R=2/3, QPSK
LDPC R=2/3, 8PSK Therefore, in our ACT scheme, the sender can predict
Rate(bits per symbol

2.5 LDPC R=5/6, BPSK

LDPC R=5/6,
LDPC R=5/6,
QPSK
8PSK
the weather state and then adaptively select the applicable
2
coding overhead according to the ARM A(1, 6) prediction
1.5 result and Equation (14). Hence, for the N -state Markov
1
chain with the stationary distribution of N states P 1 ≤
P 2 ≤ . . . ≤ P N , the average coding overhead of the ACT
0.5
scheme for the N -state Markov channel is
0 PN
0 2 4 6 8 10 12
n− P i ki
SNR(dB)
Oa = PN i=1 . (15)
i
FIGURE 8. Achievable rate of the AFC schemes and the LDPC codes versus i=1 P ki
signal-to-noise ratios (SNRs) for BER = 10−4 .
Let Fe denotes the average prediction error of our
ARM A(1, 6) model. From the above assumption and anal-
ysis, if the sender predicts the next transmission time
for BER = 10−4 . Clearly, the maximum achievable rate of window is in the 1st Markov state (i.e., the worst state),
AFC scheme outperforms the LDPC codes in the simulation then the expected recovery information symbols in the next
range of SNRs. It is important to note that the fixed rate transmission time window is k1 . If the sender predicts the
LDPC codes can be optimized for a specific SNR, however, next transmission time window is in the 2nd Markov state,
they are not optimal for other SNRs. then the expected recovery information symbols in the next
transmission time window is k2 (1 − Fe /N ). If the sender
predicts the next transmission time window is in the 3rd
IV. ADAPTIVE CODING TRANSMISSION SCHEME Markov state, then the expected recovery information sym-
BASED ON THE ARMA (1, 6) MODEL bols in the next transmission time window is k3 (1−2Fe /N ).
VOLUME 4, 2016 7

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

Therefore, the throughput G of the ACT scheme can be 100

20 pilots
calculated as follows 50 pilots
80 pilots
N 10-1
X (i − 1)Fe
G = P 1 k1 + ki (P i −

Prediction error (dB)

). (16)
N
i=2 10-2

In contrast, if the sender chooses to transmit with a

fixed overhead, such as the ith Markov state overhead, the 10-3

throughput of the fixed scheme can be modeled as

 10-4
G1 = k1 ,
Pi−1 (17)
Gi = ki · (1 − j=1 P j ), for i > 1.
0 5 10 15 20 25 30
This can help the space nodes in the integrated satellite- Fading depth of the rain attenuation (dB)

terrestrial networks tradeoffs between reliability and limited (a)

communication resources.
0.3
B. COMPLEXITY ANALYSIS OF AFC IN THE ADAPTIVE

Predicetion error (dB)

0.25
CODING TRANSMISSION SCHEME
The complexity of our ACT scheme is an important 0.2
factor for determining the corresponding coding parameters
0.15
of the AFC for the integrated satellite-terrestrial network
communications. We consider the AFC with k variable 0.1
nodes, and assume the corresponding generated matrix G,
0.05
for which the maximum number of nonzero elements in each with ARMA (1, 6) model
column and row is C and R, respectively. The complexity 0
without ARMA model
of generating G is O(C · k). Moreover, each encoded node 1 2 3 4 5 6
Predicted steps
7 8 9 10

is connected with at most R variable nodes. Then, the table

size is 2R for every encoded node, and the complexity of the (b)
AFC with block length n AFC is O(n). On the decoder side, FIGURE 9. Impact of the autoregressive–moving-average (ARMA) model
we adopt the CSBP decoding algorithm which is a kind of estimation to the ACT scheme: (a) different number of the pilots; (b) prediction
error.
updated belief propagation (BP) algorithm. In each iteration
decoding round, the maximum computation complexity
PR is
O(C · W), which is a constant and W = 1 + i=1 | wi |.
Therefore, the total computation complexity of the CSBP became more precise. Considering the tradeoffs between
algorithm is O(C · k · W). In a well-designed AFC scheme, reliability, computational complexity and buffer consump-
W and C are both constant, and the AFC scheme can be tions, we selected 25 pilots and 10 dB rain attenuation.
considered to have linear computation complexity. The prediction errors versus predicted steps are shown
in Fig. 9(b). The prediction error slowly increases with
the predicted steps, which indicates that our ARM A(1, 6)
V. SIMULATIONS AND DISCUSSION
model can perform well in the case the sender can not
A. PREDICTION ERROR ANALYSIS OF THE ARMA (1, 6)
receive channel state feedback in a short time.
MODEL
We first present the Monte Carlo simulation results to
evaluate the performance of our ARM A(1, 6) prediction B. CHANNEL ESTIMATION AFFECTION IN THE
model. Moreover, consider that the sender can obtain the ADAPTIVE CODING TRANSMISSION SCHEME
delayed CSI by NAK feedback in the integrated satellite- Based on the above discussion, considering the prediction
terrestrial network communications, and it can use the feed- error in our ARM A(1, 6) model, in this subsection we
back channel states of the last-n time windows to enhance simulate the mean absolute errors with the prediction error.
the prediction error of the ARM A(1, 6) prediction model. Assuming the number of variable nodes k is 100, the
However, the tradeoffs between reliability, computational simulation result of the decoding failure probability with
complexity and buffer consumptions should be considered. a error range from –2 dB to 2 dB on the different rain
Fig. 9(a) shows the one-step prediction error of our attenuation SNR values 10 dB, 15 dB, 20 dB, and 25 dB
ARM A(1, 6) model with a five-state Markov chain for are shown in Fig. 10. Each simulation data point runs 1000
seven different rain attenuation SNR values: 0 dB, 5 dB, times. As shown in Fig. 10, the estimation error has little
10 dB, 15 dB, 20 dB, 25 dB, and 30 dB. Each simulation effect on the the decoding performance of the ACT scheme
data point is running 1000 times. Obviously, as the number if our ARM A(1, 6) model can limit the rain attenuation
of the pilots is increased from 20 to 80, the prediction result prediction error in ±2 dB, which can be achieved in the
8 VOLUME 4, 2016

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

−6
x 10
3
18
16
10 dB 14
2.5

Throughput (kB/s)
12
decoding failure probability

10

2 8
6
15 dB
4
1.5 20 dB 2
0
ACT Fixed rate 1 Fixed rate 2 Fixed rate 3 Fixed rate 4 adaptive LT codes [23]
25 dB
1
FIGURE 11. Throughput simulation of different coding schemes.

0.5
−2 −1 0 1 2
Mean absolute error of estimation (dB) dB is the case without channel fading. Corresponding to
FIGURE 10. Impact of different signal-to-noise ratios (SNRs) on the adaptive
the selected four rain attenuation values, we can obtain four
coding transmission scheme. channel states: SN R1 = 7 dB, SN R2 = 9 dB, SN R3 = 11
dB, and SN R4 = 15 dB. By using BPSK modulation, we
TABLE 2. Simulation parameters. LT: Luby Transform. derived the relation of the BER and SNR as follows
1
Parameter Values PeDP SK = exp(−SN R). (19)
Transmission power 32 W 2
Data rate 20kB/s
Time window duration 90 s
Accordingly, we have four BERs which correspond to
Prediction error 0.0227 dB the four states: p1 = 3.3 × 10−3 , p2 = 1.7 × 10−4 ,
BER corresponding to the four-state 10−3 , 10−4 , 10−6 , 10−14 p3 = 1.7 × 10−6 , and p4 = 9.2 × 10−15 . In this simulation,
Ka-band channel the communication link is modeled by a four-state Markov
Overhead of the ACT scheme 0.67, 0.25, 0, 0
Overhead of the fixed scheme 0.67, 0.25, 0, 0 chain; we keep the receiver decoding failure probability less
Overhead of the adaptive LT coding 5.17, 2.97, 0.463, 0.45 than 10−6 and calculate the code-rate in Table 2.
scheme [23]
Then, we compare our ACT scheme with the four fixed-
rate schemes and adaptive LT coding scheme. The simula-
most situations. tion result is shown in Fig. 11. The decoding overhead of the
first fixed scheme is 0.67, that of the second fixed scheme
C. PERFORMANCE OF THE ADAPTIVE CODING is 0.25, and the third and fourth fixed schemes have fixed
TRANSMISSION SCHEME overheads 0 each of them, they correspond to the channel
In this subsection, we simulate the performance of our state BERs of 10−3 , 10−4 , 10−6 , 10−14 , respectively. As
ACT scheme. According to the record of the rain attenuation shown in Fig. 11, the throughput performance results of
in Xi’an, China on March 14, 2010, where the rain attenua- the four fixed-rate schemes are related not only to the code-
tion fading depth was 8 dB, we select four-state rain fading rate, but also are related to the channel states. We find that
with 0 dB, 4 dB, 6 dB, and 8 dB. Therefore, a four-state the throughput of the proposed ACT scheme is over 15%
Markov chain can be modeled [28], and the corresponding greater than for the fixed rate schemes and the adaptive LT
Markov transition matrix is given as follows: coding scheme.
 
0.9986 0.0014 0.0000 0.0000 VI. CONCLUSION
 0.0007 0.9986 0.0007 0.0000 
P= In this paper, we analyzed the rain attenuation of the Ka-
 0.0000 0.0007 0.9986 0.0007  . (18)

band channel and modeled an N -state Markov channel to
0.0000 0.0000 0.0014 0.9986 capture a channel capacity that randomly ranged from a
According to the Markov random process theory, the good to bad state for future integrated satellite-terrestrial
stationary probability of each state is P 1 = 0.1667, P 2 = network communications. Then, by using the ARMA model,
0.3333, P 3 = 0.3333, P 4 = 0.1667. We simulate the we proposed the practical time-varying rain attenuation pre-
performance of our ACT scheme and compare it with diction algorithm ARM A(1, 6) model for the modeled N -
corresponding fixed rate coding scheme and the adaptive LT state Markov chain channel, which can reduce the prediction
coding scheme in [23]. The system parameters are provided error probability to an error range of about 0.0027 dB
in Table 2. error in a 10 dB rain attenuation fading depth. Moreover,
The Ka-band link communication parameters are selected by analyzing the overhead of the AFC schemes under
from the typical the integrated satellite-terrestrial network certain decoding failure probabilities corresponding to the
communications, which are referred from [6]. The trans- predicted SNR from our ARM A(1, 6) prediction model, we
mission power is 32 W, we can assume that SN R0 = 15 designed an ACT scheme based on the AFC coding scheme,
VOLUME 4, 2016 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

and the key parameters of AFC were selected based on a [17] Q. Yu, R. Wang, K. Zhao, W. Li, X. Sun, J. Hu, and X. Li, “Modeling
tradeoff between decoding failure probability, block length, RTT for DTN protocol over asymmetric cislunar space channels,” IEEE
Systems Journal, vol. 10, no. 2, pp. 556–567, Jun. 2016.
and maximal code-rate for the integrated satellite-terrestrial [18] T. de Cola, E. Paolini, G. Liva, and G. Calzolari, “Reliability options for
network communications. Simulation results show that our data communications in the future deep-space missions,” Proceedings of
proposed scheme can effectively improve throughput and the IEEE, vol. 99, no. 11, pp. 2056–2074, Nov. 2011.
efficiency with respect to existing fixed-rate coding schemes. [19] G. Wang, S. Burleigh, R. Wang, L. Shi, and Y. Qian, “Scoping contact
graph-routing scalability: investigating the system’s usability in space-
vehicle communication networks,” IEEE Vehicular Technology Magazine,
ACKNOWLEDGMENT vol. 11, no. 4, pp. 46–52, Dec. 2016.
Shushi Gu, Zixuan Huang, and Shaohua Wu contributed [20] J. Jiao, X. Sui, S. Gu, S. Wu, and Q. Zhang, “Partially observable markov
decision process-based transmission policy over Ka-band channels for
equally to this work. space information networks,” Entropy, vol. 19, p. 510, 2017.
[21] L. Castanet, T. Deloues, and J. Lemorton, “Channel modelling based on
REFERENCES N-state markov chains for satcom systems simulation,” in Proceedings
of Twelfth International Conference on Antennas and Propagation, 2003
[1] M. Jia, X. Gu, Q. Guo, W. Xiang, and N. Zhang, “Broadband hybrid (ICAP 2003). (Conf. Publ. No. 491), Exeter, UK, Mar. 2003, pp. 119–122.
satellite-terrestrial communication systems based on cognitive radio to-
[22] CCSDS, Erasure correcting codes for use in nearearth and
ward 5G,” IEEE Wireless Communications, vol. 23, pp. 96–106, Dec.
deep-space communications, document CCSDS Standard
2016.
131.5-O-1, Orange Book, Nov. 2014. [Online]. Available:
[2] J. Jiao, H. Gao, S. Wu, and Q. Zhang, “Performance analysis of space
https://public.ccsds.org/Pubs/131x5o1.pdf.
information networks with backbone satellite relaying for vehicular net-
works,” Wireless Communications and Mobile Computing, vol. 2017, pp. [23] J. Jiao, Q. Guo, and Q. Zhang, “Packets interleaving CCSDS file delivery
1–13, Dec. 2017. protocol in deep space communication,” IEEE Aerospace and Electronic
[3] J. Jiao, Y. Hu, Q. Zhang, and S. Wu, “Performance modeling of LTP- Systems Magazine, vol. 26, no. 2, pp. 5–11, Feb. 2011.
HARQ schemes over OSTBC-MIMO channels for hybrid satellite terres- [24] K. Zhang, Q. Zhang, and J. Jiao, “Bounds on the reliability of RaptorQ
trial networks,” IEEE Access, vol. 6, pp. 5256–5268, 2018. codes in the finite-length regime,” IEEE Access, vol. 5, pp. 24766–24774,
[4] D. Bell, S. Allen, N. Chamberlain, M. Danos, C. Edwards, R. Gladden, 2017.
D. Herman, S. Huh, P. Ilott, T. Jedrey, T. Khanampornpan, A. Kwok, [25] J. Jiao, Y. Yang, B. Feng, S. Wu, Y. Li, and Q. Zhang, “Distributed
R. Mendoza, K. Peters, S. Sburlan, M. Shihabi, and R. Thomas, “MRO rateless codes with unequal error protection property for space information
relay telecom support of Mars science laboratory surface operations,” in networks,” Entropy, vol. 19, p. 38, 2017.
Proceedings of 2014 IEEE Aerospace Conference, Big Sky, MT, USA, [26] S. Gu, J. Jiao, Q. Zhang, Z. Yang, W. Xiang and B. Cao, “Network-
Mar. 2014, pp. 1–10. coded rateless coding scheme in erasure multiple-access relay enable
[5] IOAG-SISG, Recommendations on a strategy for space internetworking, communications,” IET Communications, vol. 8, no. 4, pp. 537–545, Mar.
document IOAG Technical Report, Nov. 2008. [Online]. Available: http- 2014.
s://cwe.ccsds.org/ioag/Final Products/SISG Reportv1.4 FINAL.pdf [27] B. Feng, Q. Zhang, and J. Jiao, “An efficient rateless scheme based on the
[6] I. Sung, J. Gao, “CFDP performance over weather-dependent Ka-band extendibility of systematic polar codes,” IEEE Access, vol. 5, pp. 23223–
channel,” in Proceedings of the American Institute of Aeronautics and 23232, 2017.
Astronautics Conference, Rome, Italy, 2006, pp. 1–21. [28] ITU, Propagation data and prediction methods required
[7] N. Adams, D. Copeland, A. Mick, and N. Pinkine, “Optimization of deep- for the design of Earth-space telecommunication systems,
space Ka-band link schedules,” in Proceedings of 2014 IEEE Aerospace document ITU-R P.618–12, Geneva, 2015. [Online]. Available:
Conference, Big Sky, MT, USA, Mar. 2014, pp. 1–7. www.srrc.org.cn/spreadmodel/PDFile/R-REC-P.618-12-201507-I!!PDF-
[8] S. Maleki, S. Chatzinotas, B. Evans, K. Liolis, J. Grotz, A. Vanelli-Coralli, E.pdf.
and N. Chuberre, “Cognitive spectrum utilization in Ka band multibeam [29] M. Shirvanimoghaddam, Y. Li, and B. Vucetic, “Near-capacity adaptive
satellite communications,” IEEE Communications Magazine, vol. 53, no. analog fountain codes for wireless channels,” IEEE Communications
3, pp. 24–29, Mar. 2015. Letters, vol. 17, no. 12, pp. 2241–2244, Dec. 2013.
[9] J. Jiao, R. Wang, S. C. Burleigh, Q. Xie, A. Zhuang, A. Sabbagh, and K [30] C. Chen, J. Jiao, H. Wu, Y. Li, and Q. Zhang, “Adaptive rateless coding
Zhao, “Reliable deep-space file transfers: How data transfer can be ensured scheme for deep-space Ka-band communications,” in Proceedings of
within a single round-trip interval,” IEEE Vehicular Technology Magazine, 2016 IEEE International Conference on Wireless for Space and Extreme
vol. 12, no. 4, pp. 86–94, Dec. 2017. Environments (WiSEE), Aachen, Germany, Sept. 2016, pp. 77–81.
[10] J. L. Gao, “On the performance of adaptive data rate over deep space Ka-
[31] V. D. Kamp, “Climatic radiowave propagation models for the design of
band link: Case study using Kepler data,” in Proceedings of 2016 IEEE
satellite communication systems,” Netherlands: Eindhoven University of
Aerospace Conference, Big Sky, MT, USA, Mar. 2016, pp. 1–7.
Technology, 1999.
[11] CCSDS, Licklider transmission protocol (LTP) for CCSDS, documen-
t CCSDS Standard 734.1-B-1, May 2015. [Online]. Available: http- [32] M. Shojafar, N. Cordeschi, and E. Baccarelli, “Energy-efficient adap-
s://public.ccsds.org/Pubs/734x1b1.pdf. tive resource management for real-time vehicular cloud services,” IEEE
Transactions on Cloud Computing, to be published, doi: 10.1109/TC-
[12] P. Ferreira, R. Paffenroth, and A. Wyglinski, “Interactive multiple model
C.2016.2551747.
filter for land-mobile satellite communications at Ka-band,” IEEE Access,
vol. 5, pp. 15414–15427, 2017. [33] F. Xin, L. Huang, Y. Huo, C. Hu, and Y. Tian, “Space power synthesis-
[13] K. Zoltn, “Channel adaptive output back-off setting of non-linear power- based cooperative jamming for unknown channel state information,” in
amplifiers for high throughput multi-spot beam satellite systems,” Interna- Proceedings of the 12th International Conference on Wireless Algorithms,
tional Journal of Satellite Communications and Networking, vol. 33, no. Systems, and Applications (WASA), Guilin, China, Jun. 2017, pp. 483–
5, pp. 391–404, 2015. 495.
[14] L. Shi, J. Jiao, A. Sabbagh, R. Wang, Q. Yu, J. Hu, H. Wang, S. C. [34] N. Cordeschi, D. Amendola, M. Shojafar, and E. Baccarelli, “Distributed
Burleigh, and K. Zhao, “Integration of Reed-Solomon codes to licklider and adaptive resource management in cloud-assisted cognitive radio vehic-
transmission protocol (LTP) for space DTN,” IEEE Aerospace and Elec- ular networks with hard reliability guarantees,” Vehicular Communications
tronic Systems Magazine, vol. 32, no. 4, pp. 48–55, Apr. 2017. vol. 2, no. 1, pp. 1–12, Jan. 2015.
[15] R. Wang, M. Qiu, K. Zhao, and Y. Qian, “Optimal RTO timer for best [35] M. Shirvanimoghaddam, Y. Li, B. Vucetic, J. Yuan, and P. Zhang, “Binary
transmission efficiency of DTN protocol in deep-space vehicle communi- compressive sensing via analog fountain coding,” IEEE Transactions on
cations,” IEEE Transactions on Vehicular Technology, vol. 66, no. 3, pp. Signal Processing, vol. 63, no. 24, pp. 6540–6552, Dec. 2015.
2536–2550, Mar. 2017. [36] M. Shirvanimoghaddam, Y. Li, M. Dohler, B. Vucetic, and S. Feng, “Prob-
[16] CCSDS, CCSDS file delivery protocol (CFDP), document CCSDS S- abilistic rateless multiple access for machine-to-machine communication,”
tandard 727.0-B-4, Blue Book, Jan. 2007. [Online]. Available: http- IEEE Transactions on Wireless Communications, vol. 14, no. 12, pp.
s://public.ccsds.org/Pubs/727x0b4.pdf. 6815–6826, Dec. 2015.

10 VOLUME 4, 2016

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2825256, IEEE Access

S. Gu et al.: ARMA-based Adaptive Coding Transmission over Millimeter-Wave Channel for Integrated Satellite-Terrestrial Networks

[37] R. Abbas, M. Shirvanimoghaddam, Y. Li, and B. Vucetic, “Random access SHAOHUA WU (S’07–M’11) received the Ph.D.
for M2M communications with QoS guarantees,” IEEE Transactions on degree from Harbin Institute of Technology in
Communications, vol. 65, no. 7, pp. 2889–2903, Jul. 2017. 2009, in communication engineering. From Apr.
[38] R. Abbas, M. Shirvanimoghaddam, Y. Li, and B. Vucetic, “Performance 2009 to Jun. 2011, he was a PostDoc at the
analysis and optimization of LT codes with unequal recovery time and Department of Electronics and Information En-
intermediate feedback,” in Proceedings of 2016 IEEE International Con- gineering, Shenzhen Graduate School of Harbin
ference on Communications (ICC), Kuala Lumpur, Malaysia, May 2016, Institute of Technology (HITSZS), where he has
pp. 1–6.
been working till now. He has been an As-
[39] R. Abbas, M. Shirvanimoghaddam, Y. Li, and B. Vucetic, “Analysis on
sociate Professor of HITSZS since Jul. 2012.
LT codes for unequal recovery time with complete and partial feedback,”
in Proceedings of 2016 IEEE International Symposium on Information His current research interests include wireless
Theory (ISIT), Barcelona, Spain, Jul. 2016, pp. 305–309. image/video transmission, deep space communication, IR-UWB rang-
[40] H. Cui, C. Luo, and K. Tan, “Seamless rate adaptation for wireless ing/localization/communication, and 5G wireless transmission technolo-
networking,” in Proceedings of the 14th ACM international conference on gies.
Modeling analysis and simulation of wireless and mobile systems, Miami,
FL, USA, Oct. 2011.
[41] ETSI, Digital video broadcast (DVB); second generation framing struc-
ture, channel coding and modulation systems for broadcasting, interactive
services, news gathering and other satellite broadband applications (DVB-
S2), document EN 302 307 (ETSI) (V1.2.1), Aug. 2009. [Online].

SHUSHI GU received the M.S and Ph.D. degree

in communication engineering from Harbin In-
stitute of Technology (HIT) in 2012 and 2016,
respectively. He is currently a Post-Doctoral Re-
search Fellow with the Communication Engineer-
QINYU ZHANG (M’08–SM’12) received the
ing Research Centre, Shenzhen Graduate School,
bachelor’s degree in communication engineering
HIT, Shenzhen, China. His current interests in-
from the Harbin Institute of Technology (HIT)
clude erasure codes, network and channel cod-
in 1994, and the Ph.D. degree in biomedical
ing, space information networks, and distributed
and electrical engineering from the University of
storage system.
Tokushima, Japan, in 2003. From 1999 to 2003,
he was an Assistant Professor with the University
of Tokushima. From 2003 to 2005, he was an
Associate Professor with the Shenzhen Graduate
School, HIT, and was the Founding Director of
JIAN JIAO (M’16) received the M.S and the Communication Engineering Research Center with the School of
Ph.D. degree in communication engineering from Electronic and Information Engineering. Since 2005, he has been a Full
Harbin Institute of Technology (HIT) in 2007 and Professor, and serves as the Dean of the EIE School. His research interests
2011, respectively. From 2011 to 2015, he was a include aerospace communications and networks, wireless communications
Post-Doctoral Reseach Fellow with the Commu- and networks, cognitive radios, signal processing, and biomedical engi-
nication Engineering Research Centre, Shenzhen neering. He is on the Editorial Board of some academic journals, such
Graduate School, HIT, Shenzhen, China. From as the Journal on Communications, KSII Transactions on Internet and
2016 to 2017, he was with the School of Elec- Information Systems, and Science China: Information Sciences. He was
trical and Information Engineering, University of the TPC Co-Chair of the IEEE/CIC ICCC’15, the Symposium Co-Chair of
Sydney, Sydney, Australia, as a China Scholarship the IEEE VTC’16 Spring, an Associate Chair for Finance of ICMMT’12,
Council Visiting Scholar. Since 2017, he is an assistant professor in the and the Symposium Co-Chair of CHINACOM’11. He has been a TPC
Department of Electrical and Information Engineering of HIT Shenzhen Member for INFOCOM, ICC, GLOBECOM, WCNC, and other flagship
Graduate School. His current interests include error control codes, space conferences in communications. He was the Founding Chair of the IEEE
information networks, random multiple access, and machine-to-machine Communications Society Shenzhen Chapter. He has received the National
communications. He received the the Shenzhen High Level Talent Program Science Fund for Distinguished Young Scholars, the Young and Middle-
Award in 2015. Aged Leading Scientist of China, and the Chinese New Century Excellent
Talents in University, and obtained three scientific and technological awards
from governments.

ZIXUAN HUANG is currently a Master studen-

t at the Communication Engineering Research
Centre, School of Electrical And Information En-
gineering, Harbin Institute of Technology Shen-
zhen, China. She received her bachelor degree
in Electronic Information Science and Technol-
ogy from Hefei University of Technology, Hefei
in 2016. Her research interests include space
information networks, error control codes and
machine-to-machine communications.

VOLUME 4, 2016 11

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
View publication stats http://www.ieee.org/publications_standards/publications/rights/index.html for more information.