Physical Layer Modulation Benchmarking: Classical QAM Baselines for AI-Based 6G Transceiver Evaluation

The integration of artificial intelligence into sixth-generation (6G) wireless physical layers demands rigorously validated classical performance baselines against which AI-driven schemes can be objectively measured. This paper establishes comprehensive validated performance baselines for 16-QAM, 64-QAM, and 256-QAM modulation schemes using the NVIDIA Sionna differentiable simulation framework. We provide complete mathematical formulation of constellation mapping, power normalization, channel modeling, and minimum-distance detection, followed by extensive bit error rate (BER) analysis across SNR ranges from 0 to 30 dB under AWGN, Rayleigh, and Rician fading channels. Through systematic performance characterization, we quantify fundamental trade-offs between spectral efficiency and error resilience across modulation orders. To further validate the practical utility of the established base-lines, we integrate LDPC channel coding consistent with 5G NR standards and demonstrate a neural receiver case study where the classical baselines serve as objective reference points for AI-based evaluation. The obtained BER performance closely matches theoretical expectations, validating implementation correctness. This work provides essential reference benchmarks for AI-based transceiver evaluation, practical guidelines for adaptive modulation in heterogeneous 6G scenarios, and identifies inherent limitations of fixed modulation schemes that motivate learning-based approaches.