R1-2600191
discussion
Downlink transmission scheme for downlink shared channels
From OPPO
Summary
OPPO's contribution to 3GPP RAN1#124 on 6G downlink transmission schemes presents 34 proposals and 20 observations covering baseline PDSCH transmission modes, multi-TRP simplification, codeword-to-layer mapping, AI/ML-based cross-layer modulation and precoding, DMRS design unification and overhead reduction, and multi-RAT spectrum sharing between NR and 6G.
Position
OPPO requires NR PDSCH transmission mode (UE-transparent precoding, jointly precoded DMRS, SU-MIMO up to 8 layers, MU-MIMO) as baseline for 6G and proposes studying transmit diversity schemes (precoder cycling with partial CSI, cyclic delay diversity, SFBC) targeting high-mobility scenarios like high-speed trains. For multi-TRP, OPPO argues for simplification by prioritizing ideal-backhaul CJT, SFN-like, and DPS schemes while deprioritizing non-ideal backhaul cases, and proposes studying L1-based flexible switching with low latency and signaling overhead. On codeword-to-layer mapping, OPPO questions the value of two codewords for ranks 2-4, presenting simulation results showing only up to 2.17% UPT gain, and proposes re-studying mapping rules particularly for lower layer counts in multi-TRP contexts. OPPO strongly advocates AI/ML-based cross-layer modulation and precoding, proposing a Transformer-based autoencoder architecture that jointly optimizes CSI feedback and precoding matrix construction across a unified high-dimensional signal space, demonstrating BLER gains of 1.4-8.0 dB depending on CSI feedback type and payload size. For DMRS, OPPO proposes a single DMRS type (Type 2), a nested port structure supporting at least 24 orthogonal ports, FDM-based extension patterns (FDM-1) for additional ports, simplified port indication via start index and count, and a unified pattern enabling configurable frequency density reduction. OPPO proposes studying SIP with AI/ML receivers (MLP-Mixer for channel estimation, Transformer for detection, cascaded training with dual loss functions) that completely eliminate DMRS time-frequency overhead, demonstrating BLER and throughput gains over baseline in low-speed (3km/h) and high-speed train (300km/h) scenarios, with complexity of ~10-90 MFLOPs and models under 3.3 MB. OPPO opposes restricting 6G RS designs for NR compatibility in MRSS, instead proposing rate matching approaches.
Key proposals
- Proposal 1 (Sec 2.1): The PDSCH transmission mode from 5G/NR (transparent precoding, jointly precoded DMRS, SU-MIMO up to 8 layers and MU-MIMO) shall be the baseline for 6G.
- Proposal 6 (Sec 2.2): Multi-TRP PDSCH transmission schemes in 6G shall be simplified, prioritizing ideal backhaul cases, coherent joint transmission (CJT), SFN-like and DPS schemes, with flexible low-latency L1-based switching.
- Proposal 11 (Sec 2.3): Study the codeword-to-layer mapping rule for PDSCH, focusing on lower layer counts (2/3/4 layers) and considering both single-TRP and multi-TRP scenarios.
- Proposal 12 (Sec 2.4): Study AI/ML-based cross-layer modulation and precoding to jointly design modulation and MIMO by establishing a unified high-dimensional signal space across multiple layers.
- Proposal 13 (Sec 3.1): 6G regular DMRS design principles shall strive for a single DMRS type, aligned UL/DL design, nested port structure, potential support for more orthogonal ports, and reuse of Rel-15/18 as starting point.
- Proposal 24 (Sec 3.1.2): 6G shall only support a single DMRS type, specifically Type 2 DMRS.
- Proposal 25 (Sec 3.1.2): Support at least 24 orthogonal DMRS ports via multiple CDM groups and study the necessity of more ports for 6G DL DMRS.
- Proposal 28 (Sec 3.1.3): Consider simplified port indication signaling (e.g., starting port index plus number of ports) to support more DMRS ports without complex table design.
- Proposal 29 (Sec 3.1.4): A unified DMRS pattern supporting both higher DMRS capacity and lower frequency density (e.g., FDM-1 pattern with DMRS spanning multiple PRBs) is preferred for overhead reduction.
- Proposal 32 (Sec 3.2): Study the downlink SuperImposed Pilot (SIP) scheme, where DMRS and data share the same resource elements with a fixed power ratio (e.g., α=5% for DMRS), to completely eliminate DMRS time-frequency overhead.
- Proposal 33 (Sec 3.2): Study AI/ML-based receiver for SIP to split pilot and data, performing channel estimation and data detection jointly or channel estimation alone with legacy MMSE detection.
- Proposal 34 (Sec 4): For Multi-RAT Spectrum Sharing (MRSS) between NR and 6G, RAN1 should not restrict 6G DMRS and CSI-RS designs to be NR-compatible; instead study rate matching approaches to handle NR's dynamic transmissions.