Environment-Aware Coverage Optimization for Space-Ground Integrated Maritime Communications

,


I. INTRODUCTION
With the rapid development of marine industries, such as marine tourism, offshore aquaculture and oceanic mineral exploration, there is a growing demand for broadband maritime communication services [1]. Satellites have large coverage areas to enable global maritime communications, but most of them offer currently only narrowband services, e.g., a maximum rate of 492 kbps provided by the widely-used international maritime satellites (Inmarsat).
The associate editor coordinating the review of this manuscript and approving it for publication was Zhenyu Zhou . Besides, although some high-throughput satellites have been being planned, high-power transmitters and high-sensitivity receivers are always required for satellite terminals (STs) to overcome the high propagation loss [2]. On the ground side, by setting up base stations (BSs) along the coast, the shorebased maritime communication network (MCN) can provide broadband services for coastwise mobile terminals (MTs). However, its coverage area is usually limited. The farthest coverage distance reported in the literature is only 100 km, which even cannot support maritime activities in the exclusive economic zone (EEZ) [3]. Therefore, it is of great interest to build a space-ground integrated network for maritime communications, so as to achieve the coordination gain, taking full advantage of both satellites and shore-based networks [4].
Due to the limited number of geographically available onshore BS sites, and the strong mobility of low-Earth orbit (LEO) and medium-Earth orbit (MEO) satellites, the spaceground integrated MCN has a highly non-cellular structure, which is different from terrestrial cellular networks. As a result, there always exist blind zones despite of network planning. In addition, when a terrestrial BS covers the MTs far from shore with high power, it will inevitably cause strong co-channel interference to the close-range MTs served by the neighboring BSs, as well as the STs nearby [5]. The ergodic sum capacity is often restricted by the blind zones and the coverage areas with severe interference, posing great challenges to the design of practical MCNs. Therefore, it is critically important to reconsider the coverage optimization issue for space-ground integrated MCNs [6].
Generally, the coverage performance of space-ground integrated networks can be optimized with the cooperative design of signals transmitted by different BSs. In [7], the authors optimized the deployment of BSs and analyzed the gain of directional beamforming. In [8], a cognitive spectrum sharing and frequency reuse scheme was proposed to improve intercell fairness. More recently, the authors of [9] proposed a beamforming scheme based on the uplink-downlink duality theory to decrease the co-channel interference. These studies all require full channel state information (CSI) at the transmitter. To reduce the overhead of acquiring CSI, the authors in [10] exploited large-scale CSI instead of full CSI for coverage optimization. All of the above schemes, however, have not fully utilized the wireless propagation environment information, such as the geographic location of scatterers, due to the complicated distribution pattern and unpredictable movement of the scatterers.
Different from terrestrial networks, geographical environment characteristics could be exploited for coverage optimization in the MCN. Previous channel modeling studies have suggested that maritime channels consist of only a few strong propagation paths due to the limited number of scatterers [11]. Some scatterers mounted in fixed locations, such as the lighthouses, reefs, and islands, can be observed using remote sensing techniques [12]. Besides, the locations of moving scatterers such as vessels and sea waves can be estimated using vessel detectors for maritime surveillance [13] and the trochoid model [14], respectively. Therefore, the transmit antenna correlation can be acquired from the marine environment, such as the azimuth angle between the scatterers and the transmit antennas.
In this paper, we optimize the coverage performance of the MCN using the path loss and the transmit antenna correlation estimated from the environmental information, i.e., the location and mobility information of scatterers. For all the MTs within the coverage, the ergodic sum capacity is maximized by optimizing the input covariance and the precoding matrix, while providing the STs with guaranteed quality of service (QoS). We derive the correlation matrix from the azimuth angle of departure (AoD) of each scattering path, and express the ergodic sum capacity without the smallscale CSI. The optimization problem is proved a complicated non-convex one. We introduce a tight upper bound to approximate the ergodic sum capacity, and propose an iterative algorithm to solve the problem by solving a convex subproblem in each iteration. We evaluate the performance of the proposed environment-aware scheme using Matlab and Visualyse, where the real-world geographic information of a coastal area around the Chongming Island of China is used to estimate the CSI. Simulation results reveal that the proposed scheme can greatly improve the ergodic sum capacity and achieve dynamic coverage for the non-cellular space-ground integrated MCN. The main contributions of this paper are summarized as follows: where the information of marine environment is collected and used for more precise channel estimation.
• Using the environmental information, a cooperative coverage optimization algorithm is proposed for the spaceground integrated MCN. The ergodic sum capacity of all the MTs within the coverage is maximized, while the STs are provided with guaranteed QoS.
• The performance of the proposed scheme is evaluated using the real-world geographic information of a coastal area of China. The impact of BS configuration and network topology on the coverage performance is also discussed.
The remainder of this paper is organized as follows: Section II introduces the system model of an environmentaware space-ground integrated MCN. In Section III, the formulation of the cooperative coverage optimization problem is presented, and an iterative algorithm is proposed to solve the problem. Section IV presents simulation results and further discussions. Finally, Section V concludes this paper.
Throughout this paper, lightface symbols represent scalars, while boldface symbols denote vectors and matrices. C M ×N represents an M × N complex matrix, and CN (0, σ 2 ) denotes the complex Gaussian distribution with zero mean and σ 2 variance. I N represents an N × N identity matrix, and O denotes a zero matrix.
[A] (p,q) represents the element in the p-th row and the q-th column of A. diag {A 1 , . . . , A N } denotes a block diagonal matrix composed by A 1 , . . . , A N .
|a| represents the magnitude of a. A H denotes the transpose conjugate of A. E (·), det (·) and tr (·) represent the expectation operator, the determinant operator and the trace operator, respectively.

II. SYSTEM MODEL
As depicted in Fig. 1, we consider the downlink transmission of a frequency division duplexing MCN consisting of K onshore BSs. Each BS k (k = 1, 2, . . . , K ) consists of N k correlated antenna arrays (CAAs) equipped with L antenna There are also J STs suffering co-channel interference.
We propose an environment-aware MCN structure to achieve dynamic coverage, as depicted in Fig. 2. The information of marine environment is gathered via a signalling network, where the locations of fixed scatterers (such as lighthouses, reefs and islands) are obtained from remote sensing satellites, and the locations of moving scatterers (such as vessels) are estimated using vessel detectors and shipping lanes. Using the environmental information, BSs can then estimate the CSI (e.g. the path loss and the transmit antenna correlation) and generate dynamic beams accordingly, user terminals with directional antennas can also adjust the antenna patterns to match the desired signals.
The received signal of the k-th MT can be expressed as where H (k,i) ∈ C M k ×N i L is the channel gain from the i-th (i = 1, 2, . . . , K ) BS to the k-th MT, and n (k) ∼ CN (0, σ 2 I M k ) denotes the receiver noise. In a practical shorebased MCN, large antenna arrays are always equipped at the BSs to extend the range of coverage, and hybrid precoding can be used instead of full digital precoding to reduce the hardware cost and complexity. Therefore, the transmitted signal for the i-th MT x (i) ∈ C N i L can be written as where s (i) is the normalized transmitted symbol, b (i) and w (i) are the baseband precoding matrix and the radio frequency (RF) precoding matrix, respectively. Each element of w (i) satisfies the magnitude constraint w (i) (p,q) = 1 √ N i L . We assume a total transmit power constraint P (k) max for each BS k, i.e., Generally, H (k,i) can be expressed as [15] is the small-scale fading matrix, whose entries follow independent and identical distribution with the average propagation loss l (k,i) n (n = 1, 2, . . . , N i ) satisfying [11] l (k,i) where λ denotes the carrier wavelength, d (k,i) n is the distance between the n-th CAA of the i-th BS and the k th MT, h (i) t VOLUME 8, 2020 In the following, we give an estimation of T (k,i) using the geographic location information of scatterers. As maritime channels are discovered to have a limited number of scatterers, we adopt the widely-used geometric Saleh-Valenzuela channel model [16]. The number of scatterers is L s , and each scatterer contributes a single propagation path between the BS and the MT. Under this model, the channel gain from the n-th CAA of the i-th BS to the k-th MT can be expressed as where α (k,i) n,l ∼ CN 0, P R and ϕ (k,i) n,l denote the complex gain and the AoD of the l-th path (as depicted in Fig. 1), respectively. Assuming uniform linear arrays are deployed at the BS and the antenna elements are closely packed at the MT, the m-th (m = 1, 2, . . . , M k ) row of a BS can be written as where d is the distance between antenna elements. The channel covariance matrix can thus be expressed in (9), as shown at the bottom of this page. Therefore, T (k,i) n ∈ C L×L is estimated to be the normalized form of R The total covariance of the interference plus noise at the j-th (j = 1, 2, . . . , J ) ST can be expressed as where L (j,i) and T (j,i) are the path loss and the transmit antenna correlation from the i-th BS to the j-th ST, respectively. To better evaluate this interference, the space environment information is used. Particularly, assuming the antennas of a ST are always pointing to the satellite, the antenna directivity θ (j,i) n can be estimated from the satellite's position (as depicted in Fig. 1), and L (j,i) can be calculated according to (6). In addition, the AoD of each scattering path from the n-th CAA of the i-th BS to the j-th ST ϕ (j,i) n,l can be obtained from the locations of scatterers, and T (j,i) can be estimated accordingly.
In Table 1, we summarize the aforementioned environmental information on the sea and in space, as well as the channel parameters estimated from these information.

III. COOPERATIVE COVERAGE OPTIMIZATION
In this section, we optimize the coverage performance of the MCN using the aforementioned channel parameters estimated from the environmental information. Our objective is 89208 VOLUME 8, 2020 to maximize the ergodic sum capacity for the MTs within the coverage by optimizing the precoding matrix cooperatively, while providing the STs with guaranteed QoS. The ergodic sum capacity can be written as where = diag (1) , . . . , (K ) , and is the total covariance of the interference plus noise at the k-th MT. The optimization problem is formulated as where σ (j) max denotes the interference threshold of the j-th ST.
There are three major challenges of solving the optimization problem. Firstly, to optimize with only the path loss L (k,i) |k, i = 1, 2, . . . , K and the transmit antenna correlation T (k,i) |k, i = 1, 2, . . . , K , we have to calculate the expectation over the instantaneous sum capacity with respect to the small-scale fading S (k,i) |k, i = 1, 2, . . . , K , and it is difficult to express C ( ) straightly without the expectation operator. Secondly, due to the mutual coupling of interference as shown in (11) and (13), C ( ) is not a concave function of , and (14) is a complicated non-convex problem. Thirdly, (k) is a function of w (k) and b (k) subject to (14e) and (14f), and it is intractable to obtain the optimal value of the coupled variables (k) , w (k) , and b (k) .
To tackle the first challenge, we adopt the random matrix theory and introduce an upper bound of C ( ), which is proved to be tight in the low signal-to-noise ratio (SNR) regime [15]. Defining (k) = L (k,k) T (k,k) L (k,k) (k = 1, . . . , K ), we can express the upper bound as where both (16) and are concave functions of .
It should be noted that C ( ) is still a non-concave function of . To transform (14) into a convex problem, we have to replace ξ 2 ( ) with a linear function of . To this end, we use the first-order Taylor expansion to approximate ξ 2 ( ), and propose an iterative algorithm to solve (14) by solving a convex subproblem in each iteration.
At a certain point¯ , the first-order Taylor expansion of ξ 2 ( ) can be expressed aŝ and = diag (1) , . . . , (K ) . C ub ( ) can thus be approximated as whereξ 2 is a linear function of , and C sub ¯ is a concave function of . Therefore, the original problem (14) can be solved iteratively. The subproblem solved in each iteration s is formulated as To solve (21) in each iteration, we have to further decouple the variables (k) , b (k) , and w (k) . It is noted that by optimizing (k) , we actually provide a digital precoding scheme for the BSs, while a hybrid precoding scheme is obtained by optimizing b (k) and w (k) . Inspired by this, we first obtain the optimal solution of (k) by solving (21a)-(21d), which is a convex problem. Then, sub-optimal solutions of b (k) and w (k) can be obtained from (k) by solving the following optimization problem [17]: The proposed algorithm is described in detail in Algorithm 1. In practical MCNs where large antenna arrays are equipped at the BSs, hybrid precoding can be used instead of full digital precoding to reduce the hardware cost and complexity by reducing the number of RF chains. Therefore, Update s = s + 1.

IV. PERFORMANCE EVALUATION AND DISCUSSIONS
The MCN simulated below consists of 3 onshore BSs, 3 MTs and 2 STs within 50 km offshore. We assume N k = 3, L = 3 and M k = 2 for all k (k = 1, 2, . . . , K ), and M j = 1 for all j (j = 1, 2, . . . , J ). The carrier frequency and the bandwidth are set to be 1.9 GHz and 20 MHz, respectively. The transmit and receive antenna heights are assumed to be 50 m and 10 m, respectively. The noise power density is −174 dBm/Hz. The receiving antenna gain is given as where g (0) is the boresight gain, and θ th is the angle between −3 dB points of the main lobe [18]. First, we evaluate the ergodic sum capacity under different transmit power constraints. We compare the performance of the proposed environment-aware scheme with that of the space-environment-aware scheme in [19], the  non-environment-aware scheme in [15], and the equal power allocation scheme as (k) = P (k) max N k L k I. As shown in Fig. 3, the proposed scheme achieves larger ergodic sum capacity than the other schemes. Using the environmental information in space, [19] outperforms the non-environment-aware scheme and the equal power allocation scheme, due to the use of directional antennas at the transmitter/receiver. Furthermore, the performance gap can be greatly enlarged for maritime communications using the proposed scheme. The performance gain mainly comes from the utilization of marine environment information, i.e., the location and mobility information of scatterers. With the knowledge of transmit antenna correlation derived from the above information, the proposed scheme both enhances the desired signal and mitigates the co-channel interference.
The shore-based MCN has to cover a vast area with a limited number of BS sites. It usually adopts high-powered BSs for remote transmission, resulting in low energy efficiency. Besides, previous studies have suggested that energy efficiency can be severely degraded by the strong co-channel interference, and there is a tradeoff between energy efficiency and spectral efficiency [20]- [23]. For this reason, we simulate the energy efficiency using different schemes in Fig. 4. It can be seen that the proposed environment-aware scheme achieves higher energy efficiency than the non-environmentaware schemes, although it aims to maximize the ergodic sum capacity. On the other hand, the energy efficiency decreases with the increase of P (k) max , as the Shannon capacity grows logarithmically with the signal-to-interference-plus-noise ratio (SINR), and a larger value of P (k) max leads to stronger interference.  Further, we evaluate the coverage performance of the proposed scheme using Visualyse Professional 7, where the real-world geographic information of a coastal area around the Chongming Island of China is used for simulation [24]. Fig. 6 shows the SINR of a MT moving within the area, where directional transmit antennas and omnidirectional receive antennas are used. It can be observed that the proposed scheme can achieve dynamic coverage by both enhancing the desired signal and mitigating the co-channel interference. Specifically, we compare the performance when a singlebeam transmit antenna (i.e. L = 1) and phased-array transmit antennas are equipped at the BS. As can be seen in Fig. 6(a), the beam generated by BS 1 does not strictly point to MT 1. Instead, it has to be misaligned to avoid interfering with ST 1, which will inevitably reduce the SINR of MT 1.
The use of phased-array antennas can address this issue to some extent. As depicted in Fig. 6(b) and Fig. 7, the sidelobes of the beam are suppressed to mitigate the co-channel interference to certain angles, while the mainlobe of the beam can still point to the desired direction to maximize the SINR of the MT. In other words, with phased-array antennas equipped at the BS, the transmit antenna correlation derived from marine environment information can be utilized to achieve better coverage performance. VOLUME 8, 2020 In traditional scenarios, shore-based MCNs mainly cover offshore areas, while maritime satellites serve the users farther away. However, MTs can be farther offshore than STs in extreme cases. As depicted in Fig. 8, the proposed dynamic coverage optimization scheme also has good performance for this network topology. Indeed, as shown in Fig. 8(b), if BS 2 serves ST 2 (which is nearer than MT 2), it will inevitably interfere with MT 3. Therefore, it is better for BS 2 to serve MT 2, leaving ST 2 to the satellites. In Fig. 9, we give an overall evaluation of the coverage performance of the spaceground integrated MCN. It can be observed that the shorebased MCN and the maritime satellite are complementary in terms of coverage areas when working in a coordinated way. In other words, the proposed scheme can break the boundaries between the coverage areas of shore-based MCNs and maritime satellites, and achieve dynamic coverage for different network topologies. It is different from traditional scenarios where shore-based MCNs and satellites cover offshore areas and areas farther away from the coast, respectively.

V. CONCLUSION
In this paper, we have exploited the marine environment information for coverage optimization in the space-ground integrated MCN. Particularly, the location and mobility information of scatterers on the sea have been utilized to estimate the transmit antenna correlation. On that basis, we have formulated a non-convex problem for the optimization of input covariances and precoding matrixes, to maximize the ergodic sum capacity for the MTs while providing the STs with guaranteed QoS. We have introduced a tight upper bound of the ergodic sum capacity, and proposed an iterative algorithm to solve the problem by solving a convex subproblem in each iteration. Simulation results have shown that the proposed scheme can significantly improve the ergodic sum capacity compared with existing approaches. Besides, it can break the boundaries between the coverage areas of shore-based MCNs and maritime satellites, achieving dynamic coverage for different network topologies. We have further pointed out that, the performance gain mainly comes from the utilization of marine environment information, which implies a brand new way for coverage optimization in practical MCNs.
JIANHUA LU (Fellow, IEEE) received the B.S.E.E. and M.S.E.E. degrees from Tsinghua University, Beijing, China, in 1986 and 1989, respectively, and the Ph.D. degree in electrical and electronic engineering from The Hong Kong University of Science and Technology, Hong Kong, in 1998.
Since 1989, he has been with the Department of Electronic Engineering, Tsinghua University, where he is currently a Professor. He was granted funding as a National Distinguished Young Scholar, and headed the Innovative Research Group of the National Natural Science Foundation of China. He was a Chief Scientist of the National Basic Research Program (973) of China. He has made influential contributions to China's national major engineering programs, such as the Lunar Program. He has published over 300 technical articles in international journals and conference proceedings. His current research interests include broadband wireless communications, multimedia signal processing, and satellite communications.
Dr. Lu is a member of the Chinese Academy of Sciences. He also serves as an Executive Member of the Council of Chinese Institute of Electronics. He has been an Active Member of professional societies. He