Predictive QoS via Channel Aggregation

The patent introduces a machine learning model as an intermediary component that bridges the gap between raw channel measurements and QoS predictions. This mediator handles the complex interdependencies across protocol layers by learning patterns from historical data, thereby achieving accurate predictions without requiring explicit modeling of all channel variations and protocol interactions.

The patent replaces traditional deterministic network analysis methods with a data-driven machine learning approach. Instead of using complex analytical models to calculate QoS parameters, the system uses trained neural networks that process channel measurements and protocol state information to predict QoS outcomes, substituting mechanical calculations with learned patterns from historical data.

The patent segments the wireless channel into multiple carriers (e.g., primary carrier and secondary carriers) and measures RSRP for each carrier separately. This segmentation allows the system to capture frequency-selective fading effects and handle channel variations across different frequency bands, improving prediction accuracy while managing complexity through structured measurement of individual carrier components.

The patent transforms single-carrier RSRP measurements into multi-dimensional feature vectors by aggregating measurements across multiple carriers and protocol layers. This dimensional expansion enriches the input space with additional characteristics (time, frequency, spatial diversity) that enable more accurate data rate predictions while the machine learning model processes these high-dimensional features efficiently.