LEO Satellite Access Network (LEO-SAN) Towards 6G: Challenges and Approaches

In the existing access network architecture enabled by 3GPP NTN technology, the core network is located on the ground, while the BSs of the access network are all or partially located on the satellite-based platform [3]. This access architecture has low flexibility and large round-trip latency, which makes it inefficient to support massive users’ access. Besides, for mobile broadband service, because of the high speed of satellite movement, the coherent time between the user and BS becomes shorter. Therefore, the timeliness of access user selection in channel state deteriorates, resulting in the decrease of system throughput. In addition, the transmission delay between the satellite and ground user is much longer compared with the ground BS. The existing access mechanism based on multiple handshakes is inefficient under the condition of high latency, and the reliability and continuity of user service cannot be guaranteed. Especially for burst short packet services, re-access after an access failure can cause extra delay and signaling overhead.

The network architecture of LEO-SAN can be re-designed to address the aforementioned issues. Fig. 2 shows an enhanced service-oriented network architecture with several guidelines provided as follows. Firstly, the network connection is affected by the dynamic changes of LEO network topology, which thus requires flexible and effective architectural design according to the network environment and users’ requirements. The design concept of servitization can be considered in both satellite assess network (SAN) and core network (CN), such that the conventional SAN and CN in 5G can be merged together to facilitate easier network deployment and management. Moreover, network functions or services need to be flexible, distributed, and can be dynamically deployed in multiple nodes of different satellites to achieve efficient coordination of LEO networks. Finally, unified and efficient interface protocols can be designed to connect different network functions under the servitization framework [1].

The four-step access method is typically applied for mobile broadband service. Most of the current random access algorithms only focus on the current access moment. However, high-speed mobile satellites and flexible access users result in rapid topological changes. Most of the current access algorithms cannot adapt to the highly dynamic change of the LEO network because these algorithms make decisions only from the current handover moment. To address this issue, one possible strategy is to utilize the historical state information for future state prediction via efficient learning methods, which can improve the long-term benefits of mobile broadband access. However, user/satellite state information is not only large in scale, but also changes very quickly, leading to lower computational efficiency and worse prediction timeliness. Therefore, how to efficiently process the massive amount of data with rapid change is also essential for performance enhancement of random access.

In LEO-SAN, burst short packet access is one of the most important services of the satellite IoT. Generally, users only send a few bytes of information at a time. The most typical scheme currently studied is random access without authorization, namely the two-step access method [10]. Compared with the four-step access method, the two-step access method combines the first step and third step to reduce the number of information exchanges and simplify the access process. The preamble and user data have been transmitted together, in the current two-step access method, which helps reduce the signaling complexity and transmission delay. However, the characteristics of channel sparsity and sporadic traffic of massive user devices are ignored. Consequently, a sparse recovery mathematical model can be used to model the problem of joint channel estimation and user detection. Unfortunately, the problem is usually NP-hard. By introducing certain constraints to the objective function, the problem can be solved by some greedy algorithms and message passing algorithms [11].

In LEO-SAN, handover can be classified into intra-satellite beam handover, inter-satellite beam handover and satellite-ground beam handover, as shown in Fig. 5. The high mobility of satellites and uncertainty of massive user movement bring new difficulties and challenges to handover management. It results in frequent beam recasting and handover, which greatly increases signal delay and handover cost, and severely affects user service experience. To address these problems, the handover management strategy can be enhanced from two aspects of handover decision and handover process.

LEO-SAN usually relies on multi-beam coverage schemes to serve numerous users that may be distant from each other. It requires advanced beam resource management strategies to enhance the network performance. The beam resource management mainly concerns allocation strategies of available beam resources to massive users, which should aim at serving as many users as possible with limited beam resources on LEO satellites. Classical beam resource management schemes employ uniform resource allocation strategy for ground users. However, the uniform allocation method is inefficient under payload constraints and uneven distribution of requirements. Therefore, due to the conflict between efficient multi-dimensional resource scheduling and limited space-borne computing resources, how to coordinate beam resource scheduling in space, time, frequency, power and other domains will be a great challenge for future research.

For a single-carrier modulation system, Doppler shift will increase the difficulty of receiver demodulation, and cause performance degradation. For OFDM and other multi-carrier transmission systems, Doppler shift can cause high inter-carrier interference and deteriorate signal detection performance. To solve the above problems, the subcarrier interval can be increased to reduce the interference between subcarriers, or the pilot sequence length can be increased to improve the accuracy of Doppler estimation and compensation. However, the above methods will inevitably lead to the decrease of spectrum efficiency, which is difficult to meet the requirement of massive users for high-rate communication. Moreover, existing Doppler shift estimation and compensation algorithms depend on a high signal-to-noise ratio (SNR) threshold and slow variability of physical channel. Therefore, their performances can be degraded under the high-maneuvering and long-distance scenarios of LEO satellite communications.

In order to circumvent the challenges to mitigate Doppler effects using existing modulation methods as illustrated in Section IV-A, a new two-dimensional modulation scheme, namely OTFS, has been proposed [12]. Unlike classical time-frequency modulation formats, OTFS carries out modulation in the delay-Doppler domain, which is naturally more suitable for transmission over time-varying wireless propagation channels. This is because, by taking a series of two-dimensional transforms, the time-frequency doubly-dispersive channel is converted to the delay-Doppler domain with almost no channel fading. Fig. 6 shows the bit-error-rate (BER) performance comparison between OTFS and classical OFDM under various levels of Doppler shifts. Different relative speeds between the space and ground transceivers, i.e., $0$ km/h, $350$ km/h, $500$km/h and $1,000$km/h, are considered. In simulations, quadrature phase-shift keying (QPSK) is employed for modulation. Besides, the FFT size of OFDM is set as 256, while the size of transmitted symbol matrix in the delay-Doppler domain is set as $256\times 14$ for OTFS. It is observed that, on one hand, OFDM suffers from poor communication performance due to strong Doppler shifts. On the other hand, OTFS is capable of achieving significant performance gain over classical OFDM, which validates its superiority in terms of Doppler-shift resistance. Furthermore, almost the same BER performance can be attained by OTFS at different speeds. This further demonstrates its robustness to various levels of Dopper shift effects. However, since the received signals equal the two-dimensional circular convolution between the modulated symbols and the channel coefficients in the delay-Doppler domain, it will be challenging for efficient design of the channel estimators and equalizers. These tasks can be even harder for OTFS systems equipped with large-scale antenna arrays, which is usually practical for LEO-SAN to compensate for the severe path loss.

In contrast, VOFDM [13] has been proposed as a general format of classical OFDM and single-carrier systems with single transmit antenna, with good robustness to the time-varying channels. Interestingly, the signals in either discrete or continuous format of VOFDM coincide with that in OTFS at the transmitter side. Corresponding to the above example in Fig. 6, the OTFS transmitted sequence is equivalent to the VOFDM counterpart with the vector size of $14$. Next, we provide discussions explaining why OTFS (or VOFDM) is desirable to deal with the Doppler shift effects, from the VOFDM point of view. Explicitly, it has been shown in [13, 14] that VOFDM can achieve multipath diversity and/or signal space diversity, even with the minimum mean square error (MMSE) linear receiver in a vectorized subchannel [14]. This is because in VOFDM, at the transmitter side, a vector of information symbols is DFT (or IDFT) transformed implicitly and then, at the receiver side, the information symbols in this vector are demodulated together. More specifically, since in VOFDM, a vectorized channel matrix is pseudo-circulant, it can be diagonalized by DFT/IDFT matrix with a phase shift diagonal matrix (see formula (4.1) in [13]). Thus, this DFT (or IDFT) processing of a vector of information symbols is similar to the precoding in single antenna systems to attain signal space diversity to combat wireless fading or diagonal space-time block coding in MIMO systems to attain spatial diversity. The resultant diversity gain can effectively mitigate the time-selective channel fading caused by Doppler shifts. For more details, please see [13, 14].