R1-2600109
discussion
Discussion on modulation, joint channel coding and modulation for 6GR
From Spreadtrum
Summary
Spreadtrum/UNISOC presents a technical case against adopting Probabilistic Shaping (PS), Geometric Shaping (GS), high-order uniform QAM (DL 4096QAM, UL 1024QAM), and joint channel coding and modulation for 6G Release. The document contains 2 formal proposals and 11 observations, arguing that deployment SIR/SINR distributions and severe implementation penalties render these techniques infeasible or statistically insignificant for 6G.
Position
Spreadtrum/UNISOC opposes the adoption of Probabilistic Shaping (PS) for 6G, arguing that its serial and block-based Distribution Matcher/De-matcher (DM/DDM) processing fundamentally conflicts with the TS 38.214 PDSCH processing time (N1) constraint, leading to prohibitive chip area and buffer costs. They present a technical case against supporting joint channel coding and modulation in 6G Release, though they do not preclude discussing its use case in 6G AI contexts. They question the statistical significance of DL 4096QAM and UL 1024QAM, citing system-level simulations showing that less than 3% of sub-7GHz UMa UEs and only 10-23% of mmWave UEs achieve the required SIR/SINR thresholds. They require that any evaluation of shaping gains be based on a 'Net Gain' methodology that subtracts PAPR-induced PA backoff penalties and non-ideal implementation losses from theoretical AWGN gains.
Key proposals
- Proposal 1 (Sec 2.2.2): Given that the serial and block-based nature of PS fundamentally conflicts with the essential PDSCH processing time (N1), leading to high chip area and buffer costs, PS should not be supported for 6G.
- Proposal 2 (Sec 3): Don't support joint channel coding and modulation in 6GR.
- Observation 1 (Sec 2.1): System-level simulation results indicate that in sub-7GHz UMa scenarios, less than 3% of the collected simulated UEs achieve an SIR exceeding 30.76 dB. In mmWave scenarios, only approximately 10% of the simulated UEs in UMa exhibit an SINR above 30 dB, while approximately 23% of the simulated UEs in the UMi scenario are exceeding 30 dB.
- Observation 2 (Sec 2.2): PS and GS both degrade PAPR performance relative to uniform QAM, especially for the UL DFT-s-OFDM waveform. This necessitates a 'Net Gain' evaluation methodology including PA backoff.
- Observation 3 (Sec 2.2): PS and GS both introduce substantial complexity and latency at the receiver.
- Observation 4 (Sec 2.2): PS and GS both introduce new CSI processing overhead.
- Observation 5 (Sec 2.2): PS specifically imposes a heavy physical layer storage burden due to the mandatory on-chip L1 buffer required by its block-level processing structure.
- Observation 6 (Sec 2.2.1): The theoretical gain of PS and GS is weakened by both non-ideal factors and PAPR penalties (PA backoff). Evaluations ignoring these factors create an illusion of performance improvement.
- Observation 7 (Sec 2.2.1): PS require a significant performance margin over the GS to compensate the additional implementation overhead of PS.
- Observation 8 (Sec 2.2.2): The complexity of DM/DDM is dominated by the specific requirement for high-precise integer arithmetic, which is not typically supported by mainstream DSP architectures.
- Observation 9 (Sec 2.2.2): The serial and block-based nature of DM/DDM will introduce large latency.
- Observation 10 (Sec 2.1): Considering the saturation of SINR/SIR distributions observed in system-level simulations, the potential coverage for DL 4096QAM and UL 1024QAM appears to be restricted. The study must first identify and justify the specific deployment cases where a statistically significant population of UEs can benefit from these schemes.
- Observation 11 (Sec 3): Note: it is not precluded to discuss 'joint channel coding and modulation' use case in 6G AI.