MRO

apps/mobility_robustness_optimization/mobility_robustness_optimization.py

@@ -0,0 +1,293 @@
+import pandas as pd
+import numpy as np
+import matplotlib.pyplot as plt
+from typing import Any, Dict, List, Tuple
+
+from radp.digital_twin.utils import constants
+from radp.digital_twin.utils.gis_tools import GISTools
+from radp.digital_twin.rf.bayesian.bayesian_engine import (
+    BayesianDigitalTwin,
+    NormMethod,
+)
+from notebooks.radp_library import get_percell_data
+from radp.digital_twin.utils.cell_selection import perform_attachment
+
+
+class MobilityRobustnessOptimization:
+    """
+    A class to perform Mobility Robustness Optimization (MRO) using Bayesian Digital Twins. This class integrates
+    user equipment (UE) data with cell topology to predict the received power at various UE locations and
+    determines the optimal cell attachment based on these predictions. The class uses Bayesian modeling to
+    accurately forecast signal strength, accounting for factors such as distance, frequency, and antenna characteristics.
+    """
+
+    def __init__(
+        self,
+        ue_data: pd.DataFrame,
+        topology: pd.DataFrame,
+        prediction_data: pd.DataFrame,
+        tx_power_dbm: int = 23,
+    ):
+        self.ue_data = ue_data
+        self.topology = topology
+        self.tx_power_dbm = tx_power_dbm
+        self.full_data = self._preprocess_ue_topology_data()
+        self.prediction_data = prediction_data
+        self.bayesian_digital_twins = {}
+
+    def _connect_ue_to_all_cells(self, pred_data: bool = False) -> pd.DataFrame:
+        """
+        Connects each user equipment (UE) entry to all cells in the topology for each tick,
+        effectively creating a Cartesian product of UEs and cells, which includes data from both sources.
+        """
+        # Create copies to avoid modifying class attributes directly
+        if pred_data:
+            ue_data_tmp = self.prediction_data.copy()
+        ue_data_tmp = self.ue_data.copy()
+        topology_tmp = self.topology.copy()
+
+        # Remove the 'cell_' prefix and convert cell_id to integer if needed
+        if self.topology["cell_id"].dtype == object:
+            self.topology["cell_id"] = (
+                self.topology["cell_id"].str.replace("cell_", "").astype(int)
+            )
+        ue_data_tmp["key"] = 1
+        topology_tmp["key"] = 1
+        combined_df = pd.merge(ue_data_tmp, topology_tmp, on="key").drop("key", axis=1)
+
+        return combined_df
+
+    def _calculate_received_power(
+        self, distance_km: float, frequency_mhz: int
+    ) -> float:
+        """
+        Calculate received power using the Free-Space Path Loss (FSPL) model.
+        """
+        # Convert distance from kilometers to meters
+        distance_m = distance_km * 1000
+
+        # Calculate Free-Space Path Loss (FSPL) in dB
+        fspl_db = 20 * np.log10(distance_m) + 20 * np.log10(frequency_mhz) - 27.55
+
+        # Calculate and return the received power in dBm
+        received_power_dbm = self.tx_power_dbm - fspl_db
+        return received_power_dbm
+
+    def _preprocess_ue_topology_data(self) -> pd.DataFrame:
+        full_data = self._connect_ue_to_all_cells()
+        full_data["log_distance"] = full_data.apply(
+            lambda row: GISTools.get_log_distance(
+                row["latitude"], row["longitude"], row["cell_lat"], row["cell_lon"]
+            ),
+            axis=1,
+        )
+
+        full_data["cell_rxpwr_dbm"] = full_data.apply(
+            lambda row: self._calculate_received_power(
+                row["log_distance"], row["cell_carrier_freq_mhz"]
+            ),
+            axis=1,
+        )
+
+        return full_data
+
+    def _preprocess_ue_training_data(self) -> pd.DataFrame:
+        data = self.full_data.copy()
+        train_per_cell_df = [x for _, x in data.groupby("cell_id")]
+        n_cell = len(self.topology.index)
+
+        metadata_df = pd.DataFrame(
+            {
+                "cell_id": [cell_id for cell_id in self.topology.cell_id],
+                "idx": [i + 1 for i in range(n_cell)],
+            }
+        )
+        idx_cell_id_mapping = dict(zip(metadata_df.idx, metadata_df.cell_id))
+        desired_idxs = [1 + r for r in range(n_cell)]
+
+        n_samples_train = []
+        for df in train_per_cell_df:
+            n_samples_train.append(df.shape[0])
+
+        train_per_cell_df_processed = []
+        for i in range(n_cell):
+            train_per_cell_df_processed.append(
+                get_percell_data(
+                    data_in=train_per_cell_df[i],
+                    choose_strongest_samples_percell=False,
+                    n_samples=n_samples_train[i],
+                )[0][0]
+            )
+
+        training_data = {}
+
+        for i, df in enumerate(train_per_cell_df_processed):
+            train_cell_id = idx_cell_id_mapping[i + 1]
+            training_data[train_cell_id] = df
+
+        for train_cell_id, training_data_idx in training_data.items():
+            training_data_idx["cell_id"] = train_cell_id
+            training_data_idx["cell_lat"] = self.topology[
+                self.topology["cell_id"] == train_cell_id
+            ]["cell_lat"].values[0]
+            training_data_idx["cell_lon"] = self.topology[
+                self.topology["cell_id"] == train_cell_id
+            ]["cell_lon"].values[0]
+            training_data_idx["cell_az_deg"] = self.topology[
+                self.topology["cell_id"] == train_cell_id
+            ]["cell_az_deg"].values[0]
+            training_data_idx["cell_carrier_freq_mhz"] = self.topology[
+                self.topology["cell_id"] == train_cell_id
+            ]["cell_carrier_freq_mhz"].values[0]
+            training_data_idx["relative_bearing"] = [
+                GISTools.get_relative_bearing(
+                    training_data_idx["cell_az_deg"].values[0],
+                    training_data_idx["cell_lat"].values[0],
+                    training_data_idx["cell_lon"].values[0],
+                    lat,
+                    lon,
+                )
+                for lat, lon in zip(
+                    training_data_idx["latitude"], training_data_idx["longitude"]
+                )
+            ]
+
+        return training_data
+
+    def _preprocess_prediction_data(self) -> pd.DataFrame:
+        data = self._connect_ue_to_all_cells(pred_data=True)
+
+        data["log_distance"] = data.apply(
+            lambda row: GISTools.get_log_distance(
+                row["latitude"], row["longitude"], row["cell_lat"], row["cell_lon"]
+            ),
+            axis=1,
+        )
+        data["cell_rxpwr_dbm"] = data.apply(
+            lambda row: self._calculate_received_power(
+                row["log_distance"], row["cell_carrier_freq_mhz"]
+            ),
+            axis=1,
+        )
+
+        data["relative_bearing"] = data.apply(
+            lambda row: GISTools.get_relative_bearing(
+                row["cell_az_deg"],
+                row["cell_lat"],
+                row["cell_lon"],
+                row["latitude"],
+                row["longitude"],
+            ),
+            axis=1,
+        )
+        return data
+
+    def training(self, maxiter: int) -> List[float]:
+        """
+        Trains the Bayesian Digital Twins for each cell in the topology using the UE locations and features
+        like log distance, relative bearing, and cell received power (Rx power).
+        """
+        training_data = self._preprocess_ue_training_data()
+        bayesian_digital_twins = {}
+        loss_vs_iters = []
+        for train_cell_id, training_data_idx in training_data.items():
+            bayesian_digital_twins[train_cell_id] = BayesianDigitalTwin(
+                data_in=[training_data_idx],
+                x_columns=["log_distance", "relative_bearing"],
+                y_columns=["cell_rxpwr_dbm"],
+                norm_method=NormMethod.MINMAX,
+            )
+            self.bayesian_digital_twins[train_cell_id] = bayesian_digital_twins[
+                train_cell_id
+            ]
+            loss_vs_iters.append(
+                bayesian_digital_twins[train_cell_id].train_distributed_gpmodel(
+                    maxiter=maxiter,
+                )
+            )
+        return loss_vs_iters
+
+    def predictions(self) -> Tuple[pd.DataFrame, pd.DataFrame]:
+        """
+        Predicts the received power for each User Equipment (UE) at different locations and ticks using Bayesian Digital Twins.
+        It then determines the best cell for each UE to attach based on the predicted power values.
+        """
+        prediction_data = self._preprocess_prediction_data()
+        full_prediction_df = pd.DataFrame()
+
+        # Loop over each 'tick'
+        for tick, tick_df in prediction_data.groupby("tick"):
+            # Loop over each 'cell_id' within the current 'tick'
+            for cell_id, cell_df in tick_df.groupby("cell_id"):
+                # Perform the Bayesian prediction
+                pred_means_percell, _ = self.bayesian_digital_twins[
+                    cell_id
+                ].predict_distributed_gpmodel(prediction_dfs=[cell_df])
+
+                # Assuming 'pred_means_percell' returns a list of predictions corresponding to the DataFrame index
+                cell_df["pred_means"] = pred_means_percell[0]
+
+                # Include additional necessary columns for the final DataFrame
+                cell_df["tick"] = tick
+                cell_df["cell_id"] = cell_id
+
+                # Append the predictions to the full DataFrame
+                full_prediction_df = pd.concat(
+                    [full_prediction_df, cell_df], ignore_index=True
+                )
+        full_prediction_df = full_prediction_df.rename(
+            columns={"latitude": "loc_y", "longitude": "loc_x"}
+        )
+        predicted = perform_attachment(full_prediction_df, self.topology)
+
+        return predicted, full_prediction_df
+
+
+# Scatter plot of the Cell towers and UE Locations
+
+
+def mro_plot_scatter(df, topology):
+    # Create a figure and axis
+    plt.figure(figsize=(10, 8))
+
+    plt.scatter([], [], color="grey", label="RLF")
+
+    # Define color mapping based on cell_id for both cells and UEs
+    color_map = {1: "red", 2: "green", 3: "blue"}
+
+    # Plot cell towers from the topology dataframe with 'X' markers and corresponding colors
+    for _, row in topology.iterrows():
+        color = color_map.get(
+            row["cell_id"], "black"
+        )  # Default to black if unknown cell_id
+        plt.scatter(
+            row["cell_lon"],
+            row["cell_lat"],
+            marker="x",
+            color=color,
+            s=200,
+            label=f"Cell {row['cell_id']}",
+        )
+
+    # Plot UEs from df without labels but with the same color coding
+    for _, row in df.iterrows():
+        color = color_map.get(
+            row["cell_id"], "black"
+        )  # Default to black if unknown cell_id
+        if row["sinr_db"] < -2.9:  # REMOVE COMMENT WHEN sinr_db IS FIXED
+            color = "grey"  # Change to grey if sinr_db < 2
+
+        plt.scatter(row["loc_x"], row["loc_y"], color=color)
+
+    # Add labels and title
+    plt.xlabel("Longitude (loc_x)")
+    plt.ylabel("Latitude (loc_y)")
+    plt.title("Cell Towers and UE Locations")
+
+    # Create a legend for the cells only
+    handles, labels = plt.gca().get_legend_handles_labels()
+    by_label = dict(zip(labels, handles))
+    plt.legend(by_label.values(), by_label.keys())
+
+    # Show the plot
+    plt.show()