The purpose of this report is to increase knowledge of how machine learning (ML) can be applied to make radio communication systems more efficient, with respect to the robustness demands on military applications, such as availability and jamming protection. The report treats three subjects: access to training data, transmitter and receiver techniques for the physical layer and time-slot allocation in radio networks. It is shown that algorithms based on neural networks are able to perform demodulation in impulse noise environments with good performance and lower computational complexity than traditional methods. Moreover, a receiver that performs channel estimation, channel equalization and demodulation through a neural network can be made more robust to jamming by adversarial training. An example of how weaknesses can be exploited to one's advantage is given, in which a transmitter uses an adversarial attack with the aim to impair performance of a hostile, publicly well-known, modulation classifier. The attack is suppressed by the communications receiver to counteract an increased bit error rate. Results show that the technique works, but improvements are required to achieve desired performance. A sufficient amount and quality of training data is crucial for the performance of ML algorithms. It is challenging to create the needed amount of data by measurements, but it has the advantage of capturing actual signal characteristics. Synthetic data is easier to generate in large amounts, but it is difficult to realistically model signal and channel effects. Use of public data sets allows for comparison with the results of others, but it is difficult to check whether they contain errors. Data augmentation is a way to expand a data set with limited amount of data. The augmented data set must be validated using relevant methods. This report also studies radio resource allocation in the form of time-slot assignment. Several publications have dealt with channel access in 5G and 6G networks, in which reinforcement learning has been emphasized as a suitable solution. How these technologies can be translated to distributed scheduling in tactical mobile networks is a task that has been initiated and is planned to continue in a future research project