R1-2600141
discussion
Modulation for 6GR air interface
From Huawei
Summary
This Huawei document presents 18 proposals and 72 observations for the 6G radio air interface study, systematically arguing against higher-order modulation (4096QAM) and arithmetic-coding-based probabilistic shaping (CCDM, ESS, MPDM) due to prohibitive complexity, while promoting ECC-DM as a low-cost shaping alternative and enhanced AMC for channel-dependent MCS optimization.
Position
Huawei proposes strict evaluation rules requiring full complexity, storage, and latency reporting before any performance claims are considered for 6G modulation candidates. They present a detailed technical case against arithmetic-coding-based probabilistic shaping (CCDM, ESS, MPDM), arguing these schemes require aligned quantization bit widths between transmitter and receiver—a restriction unprecedented in 3GPP specifications—and exhibit prohibitive serial processing throughput bottlenecks (~100x LDPC encoding delay) and storage overhead (e.g., 24.3M LUT for ESS). They promote ECC-DM based on polar codes as the preferred DM scheme, citing reuse of existing SCL decoder hardware, minimal standardization requirements, and vendor implementation flexibility. They argue 2D-NUC is ruled out by prohibitive demodulation complexity (14.4x LDPC decoding for 1024QAM), while enhanced AMC with channel-dependent MCS parameter selection can provide up to 1.5dB gain over pure constellation shaping in fading channels. They require RAN4 inputs on phase noise, EVM requirements, MPR/A-MPR under realistic PA models before shaping gains can be finalized.
Key proposals
- Proposal 1 (Sec 3.1): Apply rules requiring complexity, storage, and latency to be fully reported before any performance observations are made or discussed in the 6G modulation study.
- Proposal 2 (Sec 3.1): The observation on performance should not be made/discussed until the complexity, storage impact and latency have been fully collected and discussed.
- Proposal 5 (Sec 3.3.3.1): Adopt observations that probabilistic shaping requires significant changes to the legacy NR coding and modulation chain, while geometric shaping has acceptable impact.
- Proposal 8 (Sec 3.3.3.3): Adopt observations that 2D-NUC requires prohibitive demodulation complexity (14.4x for 1024QAM, 4.2x for 256QAM over LDPC decoding) and has ~20x storage overhead over 1D-NUC.
- Proposal 13 (Sec 3.3.3.4): Adopt the observation that with 2D-NUC the computational complexity of reduced ML detection (QRM-MLD) is 10~100 times that of LDPC decoding, which is unacceptable at the receiver side.
- Proposal 14 (Sec 3.3.3.4): The prohibitive demodulation complexity rules out 2D-NUC, while other shaping proposals require more discussion on benefit versus cost including complexity and storage.
- Proposal 15 (Sec 5.4): Do not consider CCDM and ESS due to extremely low area efficiency and low hardware throughput, while other shaping solutions require more justification.
- Proposal 18 (Sec 5.6): For shaping schemes, factors including phase noise, EVM requirement, MPR or A-MPR increase under realistic PA model require RAN4 confirmation or inputs.
- Observation 4 (Sec 3.2.1): For a given spectral efficiency, the optimal MCS parameters (modulation and code rate) are different between AWGN and fading channels.
- Observation 25 (Sec 3.3.2.4): ECC-DM does not need to align implementation details (quantization method, bit width) between transmitter and receiver, offering high implementation flexibility consistent with 3GPP core principles.
- Observation 28 (Sec 3.3.3.2): CCDM, ESS, and MPDM incur ~10x processing delay over LDPC decoding and ~100x processing delay over LDPC encoding due to inherent serial arithmetic nature.
- Observation 38 (Sec 3.3.3.4): In the worst-case scenario, the computational complexity of sphere decoding is prohibitive (e.g., 3×10^8x over LDPC decoding).
- Observation 68 (Sec 5.6): For DFT-s-OFDM waveform, both geometric shaping and probabilistic shaping increase PAPR, counteracting the goal of using DFT-s-OFDM, with the increase being particularly pronounced in probabilistic shaping.
- Observation 70 (Sec 6): 4096QAM modulation provides zero spectral efficiency gain assuming an E2E EVM of 1.25%.