NN-PTS: a neural network-assisted PAPR reduction technique for OTFS with high-order modulation

This paper introduces a new neural network-based partial transmit sequence (NN-PTS) approach for peak-to-average power ratio (PAPR) reduction in orthogonal time frequency space (OTFS) modulation, assessed over various modulation schemes—512-QAM, 256-QAM, and 64-QAM. The new method overcomes the PAPR and bit error rate (BER) trade-off associated with the traditional approaches. For 512-QAM, NN-PTS has an outstanding PAPR reduction to 4.8 dB at a CCDF of 10⁻⁴, surpassing traditional approaches like Selected Mapping (SLM)+PTS (8 dB), PTS (9.9 dB), and others. Likewise, for 256-QAM and 64-QAM, NN-PTS has minimum PAPR values of 3.7 dB and 1.8 dB, respectively, while having significant gains of 2.7 dB to 4.7 dB and 3.4 dB to12 dB over other techniques. In addition, BER analysis also substantiates that NN-PTS reduces PAPR while exhibiting poor BER performance. To illustrate, under 512-QAM at BER = 10⁻⁵, only 10 dB SNR is needed for NN-PTS, while it has a maximum 13 dB gain over the original OTFS signal. A similar trend also occurs for 256-QAM, where the NN-PTS achieves up to a 10.1 dB improvement in SNR compared to the traditional schemes. The outcomes prove that the NN-PTS best trades off PAPR reduction and BER efficiency under changing modulation orders, adaptable to various fading conditions. This renders the proposed NN-PTS a promising candidate for next-generation and 5 G systems, where high spectral efficiency and resilient signal quality are essential.