5G Massive MIMO

Massive MIMO technology is one of the key tools for increasing the capacity of 5G networks. However, its efficiency directly depends on the accuracy and timeliness of Channel State Information (CSI). Under high user mobility, CSI quickly becomes outdated, and although frequent pilot transmission improves channel estimation accuracy, it also increases overhead and reduces overall network capacity.

The Laboratory team has developed a solution called PABAFT (Prediction Algorithm Based on Autoregressive Flexible TDD), which combines a channel prediction algorithm with an adaptive pilot transmission frequency control mechanism. The base station temporarily switches to more frequent pilot transmission to collect training data and then returns to the standard pilot transmission frequency, maintaining up-to-date channel information using the trained channel prediction model. Simulation results have shown that PABAFT achieves a signal-to-noise ratio close to that of perfect channel knowledge while significantly reducing resource consumption and increasing network capacity.

Another research direction focuses on evaluating channel prediction algorithms using MIMO channel simulation approaches. Machine learning–based algorithms require large volumes of high-quality data. Within this research area, the Laboratory team has developed accelerated wireless channel simulation methods that preserve model accuracy while reducing computation time by several times.

Many existing machine learning algorithms have been validated under different conditions, which makes direct comparison difficult. The Laboratory has conducted a systematic evaluation of state-of-the-art ML-based channel prediction algorithms using various widely adopted channel models. The results confirmed that the choice of channel model significantly affects conclusions about algorithm performance, highlighting the need for publicly available and realistic datasets for model training.

List of relevant papers: