AI-Powered Multiple Antenna Multiple Input / Multiple Output Beam Management System

Abstract

The invention is an AI-powered multiple-input multiple-output beam management system for 5G and next-generation 6G mobile communication networks, which includes at least one MIMO antenna module (1) with a beamforming capability and a 64-256 element active antenna array, and a collection module (2) that collects and preprocesses real-time channel status information, enabling dynamic beam steering, reduction of inter-beam interference, simultaneous three-dimensional beamforming and beam tracking processes in timing, height and horizontal plane axes, depending on user mobility, in Massive Multiple-Input Multiple-Output (MIMO) systems.

Description

1 TARIFF AI-Powered Multiple Antenna Multiple Input-Multiple Output Beam System Management System TECHNICAL FIELD The invention is a multi-input / multiple-output system with multiple antennas, generally powered by artificial intelligence. It is related to beam management systems. The invention is particularly useful in 5G and next-generation 6G mobile communication networks, with 64, 128 and 256 Many types of Active Antenna Systems (AAS) with antenna elements Massive Multiple-Input / Multiple-Output (Multiple-Output) In MIMO systems, dynamic beamforming depends on user movement. Direction, reduction of inter-beam interference, timing and height with horizontal 15 Simultaneous three-dimensional beamforming and beam tracking along planar axes. AI-powered multi-antenna systems that optimize processes It is related to the input-multiple output beam management system. STATE OF THE ART 20 Today, Multiple-Input Multiple-Output (MIMO) Beam management systems, especially for 5G / 5G-Advanced and 6G communications. Massive MIMO is one of the important technological fields of these systems. It involves a large number of antennas. By using the element, spatial multiplexing, beamforming and antenna gain are achieved at 25 It increases capacity, spectral efficiency, and link reliability. In current technology, artificial intelligence / machine interaction is used in 5G / 5G-Advanced Massive MIMO systems. In the learn-based beamforming approach, the base station receives information from the user. Channel status information and L1-RSRP (Layer 1 Reference Signal Received Power- 30 Using measurements such as Layer 1 Reference Signal Received Power, the appropriate beam is selected. It selects or predicts future beam changes. Thus, traditionally... Instead of scanning numerous candidate beams individually, the artificial intelligence model is used to select the beams. Selection, beam tracking, and beam switching operations can be performed more quickly. 2 However, AI models can perform real-time beam selection, beam tracking, and Beamweight optimization requires high processing power. artificial intelligence the intelligence model making correct decisions under different user, channel, mobility and environmental conditions In order to provide this, comprehensive and representative education data is needed. He hears. 5 Based on research conducted under the known state of the art, US20240291548A1 Application number [number] was found. Using artificial intelligence / machine learning The application describes how beam management is performed. The system in question, from the environment and By transferring the measurements taken from user equipment to a machine learning model, 10 It suggests suitable beams from among numerous beam options and then It establishes communication via the selected beam. Machine learning. The model offers beam options for one or more user devices, However, there is no common optimization between the beams. In current Multiple Input-Multiple Output applications, for example, 3GPP Type I / II using predefined fixed beam codebooks, such as codebooks due to limited beamset and non-optimal beam selection problem It is located; channel status information is rapidly transmitted in high-speed mobility situations. Data transfer speeds decrease due to obsolescence and beam tracking delays; very 20 Multiple Input-Multiple Output scenarios for different users Interference between the beams cannot be minimized to an optimal level; traffic Beam width is dynamically adjusted according to density and user distribution. It cannot be adjusted; the beam direction in terms of elevation angle and azimuth angle. Due to insufficient coordinated optimization, especially in multi-story buildings 25 And three-dimensional beamforming performance decreases on uneven terrain; user handover due to unpredictability of movement and reactive beam switching beam loss occurs during this time; an unnecessary number during low traffic times. Energy is wasted due to keeping the beam active, and the beam weights, Manually optimizing parameters such as tilt angle and azimuth angle 30 Therefore, site-specific adjustments take a considerable amount of time. Consequently, in multi-input-multiple-output beamforming systems with numerous antennas Improvements are being made, therefore the aforementioned disadvantages are being eliminated. 3 There is a need for new structures that will remove existing systems and provide solutions. It is heard. THE PURPOSE OF THE INVENTION The present invention meets the aforementioned requirements and overcomes all the disadvantages. Multiple input-to-multiple antennas, eliminating the need for multiple antennas and offering some additional advantages. It is related to the output beam management system. The main purpose of the invention is to enable 64, 128 and 10 generation mobile communication networks in 5G and next-generation 6G networks. Multiple Active Antenna Systems (AAS) with 256 antenna elements, for example. Massive Multiple-Input / Multiple-Output (Multiple-Output) In MIMO systems, dynamic beamforming depends on user movement. Direction, reduction of inter-beam interference, timing and height, and horizontal alignment. Simultaneous three-dimensional beamforming and beam tracking along planar axes 15 AI-powered multi-antenna systems that optimize processes The goal is to provide an input-multiple output beam management system. One purpose of the invention is to enable telecommunications operators to manage 5G Radio Access Networks. on the platforms, real-time Channel Status Information, user location and 20 using mobility data, traffic density and Quality of Service requirements Beam parameters such as azimuth angle, elevation angle, beam width, and transmission power. The goal is to dynamically optimize its parameters at the millisecond level. One aim of the invention is to create a system for urban scenarios with high user density, 25 in high-speed mobility situations and mixed indoor-outdoor environments maximizing spectrum efficiency and user experience, in return The goal is to minimize energy consumption. One of the aims of the invention is to use a deep learning-based CNN+RNN hybrid architecture to 30 To learn the optimal beam parameters from Channel Condition Information patterns and continuous beam without predefined beam codebook constraints The goal is to perform optimization in the space. 4 One aim of the invention is to analyze the user's movement trajectory using a Kalman filter and Long Short Circuit filter. Predicting the future user position using temporary memory. By predicting and implementing proactive beam steering, beam loss is reduced by 75%. is the reduction by a certain percentage. One aim of the invention is to generate a user-beam allocation graph using a Graphic Neural Network. (It aims to optimize and reduce user-to-user interference by 60%.) One purpose of the invention is to provide high capacity depending on traffic density. Smart and 10 between narrow beam and wide beam to provide wide coverage. It is about implementing a dynamic transition. One aim of the invention is to optimize the elevation angle and azimuth angle in a coordinated manner. This results in a 40% increase in data transfer speed in multi-story buildings and complex terrains. The goal is to achieve an improvement of 15%. One aim of the invention is to use reinforcement learning to teach only the necessary minimum. keeping a certain number of beams active and thus reducing energy consumption by 35% is the reduction. One aim of the invention is to automatically determine the optimal beamforming policy for each site. The advantage is that it can be learned and the need for manual adjustment can be reduced by 90%. One aim of the invention is to achieve a 45% increase in average cell data transfer rate and cell edge data transfer rate. The goal is to provide a 60% improvement in transfer speed. 25 The structural and characteristic features and all the advantages of the invention are given in the figures below. And thanks to the detailed explanation written with references to these figures, it becomes clearer. This will be understood as such. Therefore, the evaluation should also be based on these forms and details. This should be done taking the explanation into consideration. 30 BRIEF DESCRIPTION OF THE FIGURES The best way to utilize the advantages of the existing invention, together with its structure and additional elements. For understanding, it should be evaluated together with the figures explained below. is necessary. Figure 1. The invention is an artificial intelligence-powered multi-antenna multi-input-multiple-output system. This is a block diagram view of the beamforming management system. 5 REFERENCE NUMBERS 1. MIMO Antenna Module 2. Addition Module 10 3. Tracking Module 4. Prediction Engine 5. Coordination Module 6. Optimization Unit 7. Energy Management Unit 15 8. Unit of Calculation 9. Control Unit 10. Feedback loop DETAILED EXPLANATION OF THE INVENTION 20 This detailed explanation describes the invention as an AI-powered multi-antenna system. The preferred configurations for input-multiple output beam management systems are as follows: to better understand the subject and without any limiting influence It is explained in a way that will not create. 25 The invention, whose block diagram view is given in Figure 1, will enable 5G and next-generation 6G mobile devices. In communication networks, Mass Multiple-Input (MM) Multiple-output (Massive MIMO) systems, capable of beamforming, 64– At least one MIMO antenna module (1) containing a 256-element active antenna array and actual 30 an aggregation module that collects and preprocesses time-based channel status information. (2) including dynamic beam steering based on user mobility, Reducing inter-beam interference, timing, height and horizontal plane simultaneous three-dimensional beamforming and beam tracking processes on their axes 6 AI-powered optimization enabling multiple antennas, multiple inputs – multiple It is an output beam management system. The invention describes an AI-powered multi-antenna, multi-input, multi-output beam system. management system; 5  GPS (Global Positioning System) coordinates, cell handover history, and RSRP (Reference Signal Received Power) RSRQ (Reference Signal Received Quality) By combining (quality) measurements, the user's movement trajectory history can be analyzed on 10 It creates, estimates the current position, and predicts future movement. a tracking module (3) that forms its orbit,  Using the location data obtained by the tracking module (3) as input, CNN (Convolutional Neural Network) that derives spatial correlation – Convolutional Neural Network) layers and temporal correlation (temporal 15 LSTM (Long Short-Term Memory) which captures correlation) Using the (Temporary Memory) layers together for azimuth and ascent. optimal beam dimensions such as optimal beam angles, beam width, and transmission power a predictor engine that estimates its parameters (4),  To maximize overall data rate and minimize interference, 20 Multi-user system using GNN (Graph Neural Network) a coordination module that optimizes beam allocation (5),  Data such as terrain maps, building heights, or occupant floor levels using a system that optimizes both ascent and azimuth angles simultaneously. optimization unit (6), 25  By monitoring traffic load, it identifies periods of low traffic and acts accordingly. a device that enables or disables the beam energy management unit (7),  Outputs of artificial intelligence models used for beamforming optimization a calculation unit (8) that converts to antenna element weights, 30  Applying beamforming commands to the MIMO antenna module (1) a control unit (9) that provides and 7  Tracking Key Performance Indicators (KPIs) and if the model detects a decline in KPI performance, a feedback loop that triggers retraining (10) It includes. The invention describes an AI-powered multi-antenna, multi-input, multi-output beam system. In a sample application of the management system, channels (2) in a collection module situational information measurements, SRS (Sounding Reference Signal) (CSI-RS - Channel State Information Reference Signal) and CSI-RS (Channel State Information Reference Signal) Information is collected using a Reference Signal and pre-processed. Noise 10 Filtering and outlier removal (is performed. Channel for each user) The matrix is ​​extracted. A user tracking module (3), GPS (Global Positioning System – Global Positioning System coordinates, cell transition history, and RSRP (Reference Signal Received Power) / RSRQ (Reference Signal Received Quality) measurements 15 By combining them, it creates a historical trajectory of motion. Position estimation with the Kalman filter. This is done using LSTM (Long Short-Term Memory). The future trajectory of motion is predicted for 5–10 seconds ahead. A prediction engine. (4), channel status information time series and user location features as input. It receives. CNN (Convolutional Neural Network) layers spatially 20 When deriving the correlation, LSTM (Long Short-Term Memory) The layers capture the temporal correlation. The output is for azimuth and ascension. Optimal beam angles, beam width, and transmission power are determined using GNN (Graph Neural Network). Network – Graphic Neural Network) coordination module (5), user-beam allocation graph It is formed by nodes representing users and beams, and edges representing venture relationships. 25 This occurs. Multi-hop interference patterns are learned via messaging. Total data. shared user-beam to maximize speed and minimize interference allocation is optimized. Three-dimensional beamforming optimization unit (6), The angle of ascent slope is commonly in the range of 0°–90° and the azimuth direction is commonly in the range of 0°–360°. It optimizes. Inputs include terrain map, building heights, and user floor levels. It is used. Hybrid optimization based on genetic algorithms and gradient descent. is implemented. Reinforcement learning based energy management unit (7), traffic load It detects periods of low traffic by monitoring. Beam function enabled / disabled. Beam on / off policy using Q-learning for discontinuation decisions 8 It is learned. The reward function is in the form of = data rate − power cost. Beamforming weighting The computing unit (8) receives the output of the artificial intelligence model regarding the beam angles and Using antenna array geometry and direction vector equations, complex It calculates their weights. Phase shifter and amplitude values ​​are generated. Real-time. The control unit (9) adjusts the beamforming weights with a delay of 5 microseconds. It applies to the MIMO antenna module (1). Real RSRP within the scope of air verification. (Reference Signal Received Power) measurements Predicted and actual results are compared using the feedback loop. (10), data rate, SINR (Signal-to-Interference-plus-Noise Ratio). Key 10 indicators (KPIs) such as noise ratio, beam alignment error, and handover success rate. It tracks Performance Indicator (KPI) values. Retraining the model if a decline in performance is detected. It is triggered. Since the system in question involves a self-learning engine, 15 Successful beamforming configurations are stored in the experience replay buffer. Periodic model updates are performed daily for continuous improvement. This is achieved by transferring the model to similar fields through learning. In a sample application of the invention, the system produced 20 within the first 24–48 hours after commissioning. It performs basic latency profiling. UE (User Equipment) VoNR (Voice over New Radio) was launched by End-to-end latency is measured for calls, and layer-based parsing is performed. RAN (Radio Access Network) latency measurement probe, gNodeB (Next Generation NodeB – Next Generation Base Station) PDCP (Packet Data 25 Integrated into the Convergence Protocol (Packet Data Convergence Protocol) layer It collects input-output timestamp pairs for each audio packet. Transport delay its tracker displays TWAMP (Two-Way Active Measurement) at 100 ms intervals on the interface. Protocol – Two-Way Active Measurement Protocol) sends test packets back and forth. It measures the duration. UPF (User Plane Function) delay 30 telemetry collector, eBPF (extended Berkeley Packet Filter) Packet Filter) probes or local telemetry APIs (Application Programming Interfaces – Application Programming Interfaces), gRPC (Google Remote Procedure Packet processing latency via Call – Google Remote Procedure Call) 9 (per-packet processing latency) is received. IMS (IP Multimedia Subsystem) The Media Subsystem (MSS) delay analyzer is a SIP (Side-to-Side) system within IMS Core (IMS Core Network). SIP communication in the Session Initiation Protocol (Session Initiation Protocol) interface. It monitors the sessions. At the end of the first 24 hours, a baseline is established: median. The delay is 85 ms, the 95th percentile is 120 ms, and the layer contributions are RAN 30.5%. The components are: Transport 15%, UPF 10%, IMS 25%, and Propagation 20%. The system operates in 7x24 continuous monitoring mode. Call ID for each active VoNR call. A custom measurement session is created, indexed with the RAN probe and audio frame. It performs measurements every 20 ms, corresponding to its periodicity; PDCP SDU 10 (Service Data Unit) arrival time and MAC PDU (Protocol Data Unit) RAN delay is calculated by measuring the transmission time of the Unit (Protocol Data Unit). The transport tracker detects the TTL (Time To Live – TTL) in the IP header of the actual voice packets. Life Cycle) field and DSCP (Differentiated Services Code Point – By following the Differentiated Services Code Point (DSP) marking, skip 15 steps per step. It estimates the delay. The UPF collector provides telemetry per stream and high For high-volume traffic, a sampling ratio of 1:10 is applied. IMS analyzer, SIP Transaction matching is performed by correlating SIP via branch parameters. It matches the operations. The layer-based delay combiner measures all measurements in microseconds. It receives the event stream as a timestamped event stream and publishes it to a Kafka topic. 20 for each call. A delay vector is generated. The bottleneck detection engine uses the last 5 minutes of sliding time. It analyzes the delay measurements within the window. Delay per layer. The deviation from the base state is calculated. If the deviation is more than 50% and the absolute value is... A bottleneck is indicated if it exceeds ms. The engine detects the RAN bottleneck. investigates the underlying causes; cell PRB (Physical Resource Block 25 If channel (block) usage is >75%, the delay due to congestion is the average CQI (Channel Qimeter). If the Quality Indicator (Channel Quality Indicator) is <7, it is considered a delay due to poor radio quality, and if the SPS (Semi-Persistent Scheduling) inactivity ratio is >30%, it is considered scheduling inefficiency. For each bottleneck... Confidence score is calculated using Bayesian inference. Machine learning lag prediction 30 LSTM (Long Short-Term Memory) model for time series forecasting It uses a Timed Memory architecture. Input features include 30 time steps, each lasting 1 minute. Layer-by-layer delay history, cell PRB usage, active VoNR (Volume Number of Calls), average RSRP / SINR, handover rate, UPF CPU / memory. It is the use of the transport link and the use of the next 5 time steps as output (5 End-to-end delay estimation, point estimation, and confidence interval (for minutes) The model is produced using a 2-layer LSTM (0.2 interval) with 128 units each. dropout, dense output layer and MAE (Mean Absolute Error) The error function consists of the loss function. Training is done weekly, using data from the last 6 months. It is re-implemented as follows; dataset 300K samples, validation section 20% and The early stop patience is set to 10 epochs. Inference is made in real time each time. It updates every minute. For example, the estimated delay for T+5 minutes is 105 ms. With ±15 ms (95% confidence) and the current delay being 85 ms, the warning threshold is exceeded. The risk is identified. Based on the identified bottleneck and machine learning prediction, 10 Optimization actions are triggered. In the RAN bottleneck scenario, the SPS scheduler... Optimization is enabled. The current SPS configuration has a 20 ms periodicity and 50 PRB allocation size includes. Optimization unit (6) cell load and sound activity It analyzes their patterns; probability of a speech burst = 0.7, probability of silence = 0.3 It is evaluated as follows. As a decision, the SPS periodicity is reduced to 10 ms and 15 Allocations are made more frequently in order to reduce scheduling waiting times; The allocation size is reduced to 40 PRB for spectrum efficiency. gNodeB CU (Central Unit – Central Unit) to F1-AP (F1 Application Protocol) A User Equipment Context Change Request is sent via [website address], and it includes: Modified RRC (Radio Resource Control) 20 RadioBearerConfig IE with Reconfiguration (Radio Bearer Configuration Information Modified SPS via Element – ​​Radio Carrier Configuration Information Item) The configuration is located here. This configures the gNodeB DU (Distributed Unit). It implements. The expected RAN latency reduction is 45 ms → 28 ms. In a UPF bottleneck scenario, adaptive QoS (Quality of Service) The controller is enabled. An increase in UPF processing latency from 11 ms to 22 ms is detected. It is determined that the root cause is UPF CPU usage above 85% and queue depth. An increase is detected. The controller evaluates the UPF selection strategy; to gNodeB. If an alternative edge UPF with a closer and lower load is available, 30 orchestrator, SMF (Session Management Function) By sending a Session Change Request through the interface, the affected PDU (Protocol This request requires the UPF to be re-selected for Data Unit (Protocol Data Unit) sessions. SMF triggers the N2 handover procedure for UPF change and the PDU session. 11 It performs the reconstruction. Expected UPF latency reduction 22 ms → 8 ms It is ms. In a transportation bottleneck scenario, the transportation path is redirected. It is requested. The current path is gNodeB → Router1 → Router2 → UPF with 3 hops. 18 ms, the alternative path is gNodeB → Router3 → UPF with 2 hops and the expected... It includes an 11 ms delay. The orchestrator uses SDN (Software Defined Networking – Software 5). By sending a request to change the flow rule to the Defined Network (DNN) controller for VoNR traffic. It enables the use of an alternative path via a 5-tuple. The machine learning model... Predictive actions are taken based on the estimated increase in delay. For example, the cell Due to the increasing load trend, the RAN delay will be 30 within the next 5 minutes. If it is estimated that the frequency will increase from 55 ms to 55 ms, preparations for a preventive handover are carried out. 10 For UEs at the cell edge where RSRP < −105 dBm, neighboring cell measurement is triggered. and RRC Measurement Configuration is applied; handover decision threshold is temporarily set to −100 The dBm level is reduced to -102 dBm. PDCP replication for new call setups. It is enabled by default, thus reducing latency variance through redundancy. Additionally, a signal indicating encoder downgrade preference is sent to IMS, and AMR-WB (Adaptive 15 Multi-Rate Wideband (AMR-NB) 12.65 kbps → AMR-NB (Adaptive Multi-Rate Narrowband) 7.4 kbps preferred. This is done by creating smaller packets thanks to the lower bit rate, thus saving space. The interface time is reduced. As a result of these preventive measures, the estimated 55 ms. The latency increase is reduced to 38 ms. 20 The effectiveness of the implemented optimization actions is continuously evaluated. The optimization event log includes a timestamp, the identified bottleneck, and the actions taken. The action, pre-action delay, post-action delay, and activity score are recorded. The performance score is calculated as latency reduction / expected latency reduction. This log, 25 It is used as feedback for retraining the machine learning model and Optimization actions are categorized into effective / ineffective binary labels in supervised learning data. It is added to the set. Thus, the system determines which optimization is best under which network conditions. It learns that this action is more effective. For example, Cell PRB utilization >80% VE The combination of reducing SPS periodicity and predictive handover 30 is most effective under conditions of average CQI <8, providing a delay reduction of >25 ms. It can be learned. Furthermore, a suitable operating point for SPS periodicity is Bayesian. through optimization; the PDCP reordering timer is adjusted according to radio conditions. The optimal value is then determined using grid search. Thus, the system continuously updates itself. 12 It operates within a self-improving cycle. Dashboard and SLA The Service Level Agreement (SLA) monitor provides operators with It provides comprehensive visibility. Real-time metrics include ongoing call tracking. number of active VoNR calls (45), median delay (78 ms), and 95th percentile delay. (118 ms, below the 120 ms SLA target). Layer-based delay contributions are 5. The breakdown is as follows: RAN 32%, Transport 13%, UPF 9%, IMS 28%, and Propagation 18%. Latency timeline for the last 24 hours, at 5-minute resolution, 95 per hour. Percentage values ​​are monitored using bar graphs. The SLA compliance tracker... It tracks daily / weekly / monthly SLA achievement percentage; 95% of target calls. While the minimum passing time is <120 ms, the current success rate is 97.3%. Bottleneck history 10 The table includes the timestamp, detected layer, duration, applied action, and result. The information is retained. If the critical delay threshold is exceeded, a Slack / email / SMS message will be sent. Alarm notifications are generated through the integration. The system described in the invention consists of 5G gNB-CU (Central Unit) and gNB-DU 15. It is located in a decoupled architecture (Distributed Unit). Artificial intelligence. beamforming engine based on centralized orchestration within CU while performing, low-latency beam switching operations within DU. This is implemented. O-RAN (Open Radio Access Network) Within the scope of integration with its architecture, the system is a Near-RT RIC (Near-Real-Time RAN 20 Intelligent Controller – Near Real-Time Radio Access Network Intelligent Controller) It is deployed as an xApp within it. RAN KPM (Key) is accessed via the interface. Performance Metrics – Radio Access Network Key Performance Metrics) and Channel Status Information data is collected through the interface in non-real-time. Policy guidance is received. Kubernetes 25 within the scope of cloud-native deployment. Pods and horizontal scaling capability are used. GPU (Graphics Processing Unit – CUDA cores are used for CNN inference with Graphics Processing Unit (GPU) acceleration. and TensorRT optimization is applied. In the edge inference approach, such as beam tracking. Critical and delay-sensitive predictions are performed at the edge, while the three-dimensional beamforming... Complex optimization processes, such as shaping, are performed centrally. 30 The system described in this invention has alternative applications in milliwave communication. Ultra-narrow beam tracking in the 28 GHz and 39 GHz bands is used in satellite communications. Beam directional guidance towards user terminals, adaptable in radar systems. 13 Beam pattern synthesis, beamforming in 6G THz communication, wireless feedback point-to-point transport connectivity optimization, in small cells within enclosed spaces. Massive MIMO coordination, vehicle-specific communication from Vehicle to Everything. beamforming adaptation and IAB (Integrated Access and Backhaul) (Return Transport) It can be used in areas such as beam coordination. 5

Claims

14 REQUESTS 1. In 5G and next-generation 6G mobile communication networks, Multiple Inputs – Multiple In multiple output systems, the most suitable are those containing an antenna array with beamforming capability. a small MIMO antenna module (1) and 5 that collects real-time channel status information and includes an aggregation module (2) that pre-processes user mobility Dynamic beam steering depending on the configuration, reducing inter-beam interference, Simultaneous three-dimensional projection in the timing, elevation, and horizontal plane axes. Artificial intelligence that optimizes beamforming and beam tracking processes. It is a multi-input-multiple-output beamforming management system with multiple antennas, supported by 10 feature;  GPS (Global Positioning System) coordinates, cell handover history, and RSRP (Reference Signal Received Power) RSRQ (Reference Signal Received Quality) By combining (quality) measurements, the user's movement trajectory history It creates, estimates the current position, and predicts future movement. a tracking module (3) that forms its orbit,  Using the location data obtained by the tracking module (3) as input, CNN (Convolutional Neural Network – 20) that derives spatial correlation Convolutional Neural Network) layers and temporal correlation LSTM (Long Short-Term Memory) which captures correlation) Using the (Temporary Memory) layers together for azimuth and ascent. optimal beam dimensions such as optimal beam angles, beam width, and transmission power a prediction engine that estimates the parameters (4), 25  To maximize overall data rate and minimize interference. Multi-user system using GNN (Graph Neural Network) a coordination module that optimizes beam allocation (5),  Data such as terrain maps, building heights, or occupant floor levels using a 30 that optimizes both ascension and azimuth angles together. optimization unit (6),  By monitoring traffic load, it identifies periods of low traffic and acts accordingly. a device that enables or disables the beam energy management unit (7),  Outputs of artificial intelligence models used for beamforming optimization a calculation unit (8) that converts antenna element weights,  Applying beamforming commands to the MIMO antenna module (1) a control unit (9) that provides and  Tracking Key Performance Indicators (KPIs) 5 and if the model detects a decline in KPI performance, a feedback loop that triggers retraining (10) It includes.

Images