Artificial intelligence in the process of radio signal intelligence

Artificial intelligence (AI) has long been on the rise in a variety of fields. In automatic image and speech recognition, for example, it has been doing very well for years and is what made modern applications in this area possible in the first place. That is why we at PROCITEC started investigating and successfully using AI methods some time ago as part of the further development of our radio signal intelligence solutions.

AI is very often based on the machine learning (ML) approach. Here, an algorithm independently learns to fulfill a task, whereby the solution path is not predetermined. The algorithm finds this itself through a special learning phase.

ML saw a major breakthrough with the application of deep neural networks (DNN). These can now solve increasingly complex tasks through so-called Deep Learning (DL). They deliver very good results, especially in the areas of classification and pattern recognition.

Central steps in reconnaissance of radio signals are their detection and classification. It therefore makes sense to use neural networks for these types of problems as well. There are now numerous suggestions in the literature. These classification approaches are very often proposed for use in the field of cognitive radio systems, where tasks such as spectrum interference monitoring and dynamic spectrum access are primarily to be solved.

The core process of radio signals intelligence includes not only theDetection and classification of signals. This is only an intermediate step on the way to extracting signal content and therefore cannot be considered in isolation. The process can be roughly divided into the following steps:

• Detection of Signals Of Interest (SOI)
• Classification of SOIs
  • Determination of the modulation type
  • Determination of the method of transmission
• demodulation and decoding(extraction of signal content)

Based on the detection of individual SOIs, they are then classified. The type of modulation and finally the transmission method of the signal are determined here. If the latter is known, the signal can be demodulated and decoded.

The classification can be done on the basis of ML approaches or classically, by defining and directly measuring characteristic parameters of the individual modulation types or transmission methods. The latter is also referred to as an expert system.

At PROCITEC, we have been using the process described here for detecting radio signals in our software products for years. So far, expert systems for the classification of SOIs have been used successfully. In the course of the further development and constant improvement of our products, we have started to integrate DNN-based approaches into the process.

In a first step, for modulation types not yet supported in our software, it was checked whether DNN-based classification leads to good results and how these can be integrated into the existing process parallel to our expert system. The basis for this was formed by approaches known from the literature.

A crucial challenge when using a DNN is to provide a sufficient number of statistically meaningful signal examples for its learning phase (training). Since the decision for a class in a DNN, in contrast to the expert system, is not transparently comprehensible, great attention must be paid to which characteristic properties the training data set contains and whether this data reflects reality.

In addition, when generating the data set, it is important to vary certain signal parameters (e.g. noise, frequency offset, etc.) in a targeted manner so that the DNN can later react robustly to such variations in the signals in real use. It has been shown that an adequate data set can only be generated with reasonable effort using artificial signals.

Based on the training data generated in this way, we have successfully learned a DNN and evaluated it for classification using real radio signals. The work was carried out in close cooperation with a local university.

The new DNN was then integrated into the existing software. Great attention was paid to the interaction with the existing components in the overall process. After successful quality tests, thego2signals productsthe PROCITEC about firstAI based classifiers.

The experiences from the development described here have shown that the use of DNNs for the classification of radio signals has great potential. We will therefore continue to push this topic at PROCITEC and assume that the right combination of proven classic approaches and AI-based technologies will deliver better classification results in the future.

 

PROCITEC Ltd
sales@procitec.de
www.procitec.de