A study published in Scientific Reports presents a robust spectrogram-based automatic modulation classification (AMC) method using a deep CNN-LSTM architecture. According to the article, AMC identifies the modulation type of a given RF signal without prior knowledge of the transmitted signal's parameters, supporting applications such as cognitive radio, military communications, and signal intelligence, and enabling optimized use of spectrum resources, interference management, and accurate decoding decisions. The article notes that some AMC methods rely heavily on feature extraction, but such feature extraction and statistical signal processing techniques can present problems in complex environments, e.g., low SNRs and high levels of noise, and non-ideal conditions cannot typically provide high accuracies. The field is undergoing changes through machine learning and deep learning techniques, which the article states produce more accurate and effective AMC solutions than previous methods. Deep learning approaches using CNNs and LSTM networks are described as better suited for large and complex datasets and as having proven to outperform traditional AMC techniques, with CNNs more efficient in feature extraction and LSTMs excelling at temporal modelling. CNN-LSTM-based models are said to automatically learn both spatial and temporal features from raw data, obviating necessary manual feature generation.