R1-2600190
discussion
Discussion on modulation, joint channel coding and modulation for 6GR
From OPPO
Summary
This OPPO 3GPP RAN1 contribution analyzes modulation schemes for 6GR, providing 26 observations and 4 proposals. OPPO argues against higher-order modulations beyond 1024QAM (DL) and 256QAM (UL) due to stringent EVM and limited applicability, supports geometric shaping (GS) over probabilistic shaping (PS) citing PS's performance degradation in fading/multi-layer scenarios and high complexity, and proposes studying AI/ML-based demodulation and cross-layer modulation.
Position
OPPO proposes limiting the 6GR modulation scheme to up to 1024QAM for DL and 256QAM for UL, presenting technical arguments against 4096QAM based on its -38 dB EVM requirement and marginal applicability shown by system-level SINR CDFs where only the top 1% of UEs exceed 29 dB SINR in UMa at 7 GHz. OPPO supports geometrical shaping (GS) including both 1D-NUC and 2D-NUC with specified and downloadable constellations, emphasizing 2D-NUC achieves 0.6 dB and 0.4 dB gain over uniform QAM in AWGN and Rayleigh channels respectively, while 1D-NUC requires similar demapping complexity to uniform QAM. OPPO presents a comprehensive technical case against probabilistic shaping (PS), detailing performance loss of CCDM-based PS (0.7 dB in Rayleigh fading, 2.4 dB with 100 REs, 3.1 dB with 50 REs), 4.82x–9.51x additional complexity from DM, ~1 dB PAPR increase with DFT-s-OFDM, and significant Tx/Rx chain impacts including code rate constraints and RV redesign. OPPO proposes studying AI/ML-based demodulation for different constellations and cross-layer modulation jointly with MIMO precoding for geometric shaping, showing AI/ML demodulator gains of 0.1 dB for 256QAM and 0.2 dB for 256 2D-NUC over max-log.
Key proposals
- Proposal 1 (Sec 2.1.2): Up to 1024-ary constellation for DL and up to 256-ary constellation for UL are supported for 6GR modulation scheme.
- Proposal 1 (Sec 2.2.2): Support geometrical shaping for 6GR modulation scheme, including: 1D-NUC and 2D-NUC; Specified constellations and non-specified downloadable constellations.
- Observation 10 (Sec 2.3.1.1): PS-based modulation could have significant impact on Tx/Rx chain design, including code rate design, interleaving, scrambling, and redundancy version (RV) design, etc.
- Observation 13 (Sec 2.3.1.2): In i.i.d. Rayleigh fading channel, the performance gain of PS significantly declines. Ideal-DM-based PS and CCDM-based PS can have 0.4 dB and 0.7 dB performance loss compared with uniform QAM at 10% BLER.
- Observation 18 (Sec 2.3.1.3): Taking QAM demodulation as baseline, CCDM and D-CCDM introduce 4.82x and 9.51x additional complexity for PS scheme, respectively.
- Observation 20 (Sec 2.3.1.4): PS-based modulation exhibits ~1 dB PAPR increase with DFT-s-OFDM waveform compared with NR uniform QAM scheme.
- Proposal 1 (Sec 3): Study AI/ML based demodulation for different constellations.
- Proposal 1 (Sec 4.1): To jointly design modulation and MIMO for better performance, study the cross-layer modulation for geometric shaping.
- Observation 26 (Sec 5): For the enhanced AMC supporting multiple combinations of modulation order and coding rate with the same spectral efficiency, the benefits of enhanced AMC should be further studied when integrating the UE CQI feedback procedure.
- Observation 1 (Sec 2.1.1): For UL 1024QAM, the Tx EVM should be further relaxed compared with DL 1024QAM. When assuming Tx EVM 3.0%+ Rx EVM 3.0%, SNR to achieve 10% BLER increases by 2.7 dB and 12.81 dB for MCS 23 and MCS 24, respectively, compared with 256QAM. The BLER never reaches 10% for MCS 25 and MCS 26 assuming the same EVM level for UL 1024QAM.
- Observation 2 (Sec 2.1.1): 4096QAM requires approximately -38 dB EVM, reflecting a 6 dB reduction compared to the NR 1024QAM EVM requirement. This stringent EVM requirement for 4096QAM would cause deployment barriers and further increase hardware cost.
- Observation 5 (Sec 2.2.1): In AWGN and Rayleigh fading channels, 2D-NUC could achieve 0.6 dB and 0.4 dB gain at 10% BLER compared with uniform QAM, respectively.
- Observation 15 (Sec 2.3.1.2): Compared with uniform QAM, PS could exhibit performance gain in 1-layer transmissions but exhibits performance loss for 2-layer transmissions for both ideal and realistic channel estimation methods.
- Observation 21 (Sec 2.3.2): Compared with fully-shaped PS, the performance of partially-shaped PS may slightly deteriorate with lower DM computational complexity.
- Observation 24 (Sec 3): 256 2D-NUC with AI/ML based demodulator outperforms 256QAM with max-log demodulator about 0.8dB.