The rise of sixth-generation (6G) wireless networks brings new challenges for identifying and locating signals accurately and efficiently, especially in complex, high-data-rate OFDM environments. Traditional deep learning models like CNNs and LSTMs work well for extracting features over space and time but have difficulty with long-range dependencies. This paper presents a hybrid CNN-Transformer model that merges convolutional feature extraction with the self-attention mechanism of the Transformer to improve local and global feature learning. The CNN layers pull out essential spatial and spectral features from the OFDM signal, while the Transformer encoder captures global relationships across subcarriers and time frames. Experiments using synthetic and real OFDM datasets under AWGN and Rayleigh fading conditions show that the proposed model achieves 98.7% classification accuracy, 20% lower localization RMSE, and 25% faster inference compared to CNN-LSTM and CNN-GRU baselines. The results demonstrate the proposed model’s scalability, strength, and fit for next-generation 6G communication systems that need smart and adaptable signal processing...