Recent advancements in DJSCC have introduced a new paradigm for semantic communications. DJSCC-based semantic communications have demonstrated notable performance, particularly in scenarios with limited bandwidth and low signal-to-noise ratios. This chapter investigates waveform optimization for semantic communication in OFDM systems. While OFDM is widely adopted for its spectral efficiency and resilience against multipath fading, the integration of neural network-based semantic encoders and decoders introduces new challenges. In particular, DJSCC enables end-to-end optimization and improved adaptability under varying channel conditions, but also leads to significantly higher PAPR due to continuous-valued semantic signals across multiple subcarriers. To address this issue, the chapter reviews both conventional and learning-based PAPR reduction techniques, including clipping, companding, SLM, PTS, PAPR-aware loss design, and clipping with retraining. By applying these methods, the excellent image reconstruction performance of DJSCC can be preserved while keeping PAPR within an acceptable range in the OFDM system. These insights underline the importance of PAPR-aware waveform design in advancing reliable and efficient semantic communication over OFDM systems, particularly for future intelligent wireless networks.