Information Preserving Quantization and Decoding for Satellite-Aided 5G Communications Tobias Monsees, Dirk W¨ ubben, and Armin Dekorsy Department of Communications Engineering University of Bremen, 28359 Bremen, Germany Email: {tmonsees, wuebben, dekorsy}@ant.uni-bremen.de Abstract—We consider the uplink of a non-terrestrial network where a relay node is forwarding digitized signals via a rate limited error-prone forward link to the serving satellite. The focus of our investigations are the design of suitable low-bit resolution quantization schemes to limit the rate on the forward link and the optimization of the decoder processing at the satellite. To this end, we investigate Information Bottleneck (IB) based quantizer design with mutual information as fidelity criterion. The IB approach can be extended to consider also the error statistic of the forward link within the quantizer design. By numerical investigations we demonstrate the performance gains of this forward-aware IB quantizer in case of erroneous forward links. Furthermore, we investigate a lookup-table based decoder which is optimized for the end-to-end statistic including the access link, the quantizer and the forward link. This decoder implementation processes only discrete values using lookup-tables of small size. The numerical results show that the performance of 3-bit forward-aware IB quantizer in combination with a 3-bit discrete decoder implementation is close to the double-precision floating-point belief propagation decoder even for strongly error- prone forward links. I. I NTRODUCTION The 3rd Generation Partnership Project (3GPP) has identi- fied Non-Terrestrial Networks (NTNs) as a promising solution for the 5G service enablers Massive Machine-type Communi- cation (mMTC) and Enhanced Mobile Broadband (eMBB) and several architecture options are currently under discussions [1], [2]. In particular, the User Equipment (UE) may use a direct access link to communicate with the satellite or, alternatively, an on-ground relay node (RN) provides the access link for the UE and forwards the information between the satellite and the UEs [3]. In this paper, we focus on the Uplink (UL) of this two-hop scenario where the RN provides the access link as depicted in Fig. 1. To this end, it is assumed that the signals received on the access link are processed by the RN and digitized messages are transmitted via the rate limited Forward Link (FL) to the satellite. Here, the application of low-bit quantizers is of significant interest in order to limit the required data rate on the FL. For the satellite, a regenerative payload configuration is assumed where the user messages are estimated by decoding the Forward Error Correction (FEC) [2] code. Thus, this scenario is similar to the UL of a Cloud Radio Access Network (Cloud-RAN) system where Small Cells (SCs) forward UE messages to the Central Processing Unit (CPU) for final Relay Node UE Satellite Access Link Forward Link Fig. 1. Considered 5G architecture with relay node (RN) providing the access link for the User Equipments (UEs). processing [4], [5]. Considering modern FEC schemes like Turbo-Codes or Low Density Parity Check (LDPC) codes, the main computational complexity of the UL processing chain is required by the iterative decoder in the satellite [4]. In order to reduce the decoder complexity, we investigate discrete variants of the Message Passing (MP) decoder for LDPC codes with a low-bit representation for the internal messages [6]–[10]. As all inter- nal operations of this Lookup-Table based Message Passing (LUT-MP) decoder are replaced by Lookup-Tables (LUTs) of small size being efficiently implementable in suitable hardware architectures, this approach promises high throughput [11]. Contributions: The considered communication system im- plements a classical two-hop transmission with a compress- and-forward relay consisting of an access- and a forward link. In order to transmit compressed signals by the RN, we apply the IB method [12]–[14] to design scalar quantizers at the RN based on the access channel statistic such that the mutual information between the source signal and the RN signal is maximized. In addition, we utilize the Forward- Aware Vector Information Bottleneck (FAVIB) algorithm [15] to consider also the statistic of the forward link in the quantizer design such that the end-to-end (e2e) mutual information between the source signal and the receive signal at the satellite is maximized. Compared to common quantization schemes, both approaches will provide improved e2e performance and increase the robustness against transmission errors on the FL. Furthermore, we extend the LUT-MP approach by optimizing 978-1-7281-3627-1/19/$31.00 c  2019 IEEE