RSSI Measurements of Beacon Frames from Wi-Fi Radio Waves

The data collection phase of the proposed system involves the collection of beacon frame characteristics and RSSI values from Wi-Fi APs using two Raspberry Pi devices. The purpose of this phase is to gather enough data to train the ML module of the proposed system to accurately determine the user's devices location based on these characteristics and values. To collect the data, we defined a threshold distance of 7 feet. This is the maximum distance between the user's devices that we consider acceptable for the purposes of this experiment. We then collected two datasets: one with data collected while the two Raspberry Pis were with 7 feet or less of each other, and another with data collected while the distance between the two Raspberry Pis was over 7 feet. In the first dataset collection stage, we followed the following steps:Began collecting data by placing the two Raspberry Pis 7 feet from each other.Moved the two Raspberry Pis closer and farther from each other while maintaining the distance within the predefined threshold.Repeated the data collection process at different locations to capture the variation in beacon frame characteristics and RSSI values that may exist in different environments.In the second dataset collection stage, we followed the following steps: Began collecting data by placing the two Raspberry Pis 7.5 feet from each other. This helped to determine the "gray area" between the acceptable threshold distance and the distance at which access should be denied.Moved the two Raspberry Pis closer and farther from each other while keeping the closest distance between them at 7.5 feetRepeated the data collection process at different locations to capture the variation in beacon frame characteristics and RSSI values that may exist in different environments. We collected a total of 4,825 samples of data from two Raspberry Pis (RPi 1 and RPi 2) measuring the SSID and RSSI values of 10 different WiFi APs at different locations and times. The Raspberry Pis were positioned at distances of 7.5 feet or less apart in the \textit{"authentic"} dataset and at distances of 7.5 feet or more apart in the \textit{"unauthorized"} dataset. Each dataset includes six columns: "RPi," "SSID," "Frequency (Hz)," "RSSI (dBm)," "Location," and "Label." The "RPi" column indicates which Raspberry Pi collected the data, the "SSID" column lists the name of the Wi-Fi AP, the "Frequency (Hz)" column specifies the frequency of the Wi-Fi AP in Hz, the "RSSI (dBm)" column shows the RSSI value in dBm, the "Location" column specifies the location where the data was collected, and the "Label" column is a categorical column with the value 1 or 0 for all rows, where 1 means \textit{"authentic"} and 0 means \textit{"unauthorized"}. The resulting dataset was balanced, with 2442 samples in the \textit{"authentic"} dataset and 2383 samples in the \textit{"unauthorized"} dataset. Figure \ref{dataset} shows the five top row of the \textit{"authentic"} dataset. The dataset was then prepared for the implementation phase of the experiment.

所提出系统的数据采集阶段，依托两台树莓派（Raspberry Pi）设备采集Wi-Fi接入点（Access Point, AP）的信标帧特征与接收信号强度指示（Received Signal Strength Indicator, RSSI）值。该阶段的目标是收集足量数据，以训练所提系统的机器学习（Machine Learning, ML）模块，使其可基于上述特征与数值精准判定用户设备的位置。
本次实验设定了7英尺的距离阈值，即两台设备间可接受的最大间距。随后我们采集了两组数据集：一组为两台树莓派间距不超过7英尺时采集的数据，另一组为两台树莓派间距超过7英尺时采集的数据。
在第一组数据集的采集阶段，我们遵循如下步骤：1. 将两台树莓派置于相距7英尺处，启动数据采集；2. 调整两台树莓派的间距，使其始终维持在预设阈值范围内，交替拉近与拉远二者距离；3. 在不同地点重复数据采集流程，以捕捉不同环境下信标帧特征与RSSI值可能存在的差异。
在第二组数据集的采集阶段，我们遵循如下步骤：1. 将两台树莓派置于相距7.5英尺处启动数据采集，以此确定可接受阈值与拒绝访问阈值之间的“灰色地带”；2. 调整两台树莓派的间距，保持二者最小间距为7.5英尺，同时交替拉近与拉远二者距离；3. 在不同地点重复数据采集流程，以捕捉不同环境下信标帧特征与RSSI值可能存在的差异。
本次采集总计获得4825条数据样本，由两台树莓派（记为RPi 1与RPi 2）完成，采集了10个不同Wi-Fi AP的SSID与RSSI值，采集地点与时间各不相同。其中，"可信（authentic）"数据集的树莓派间距为7.5英尺及以内，"未授权（unauthorized）"数据集的树莓派间距为7.5英尺及以上。
每组数据集均包含6个字段："RPi""SSID""Frequency (Hz)""RSSI (dBm)""Location"与"Label"。其中，"RPi"字段标识采集数据的树莓派设备；"SSID"字段列出Wi-Fi AP的名称；"Frequency (Hz)"字段指定Wi-Fi AP的工作频率，单位为赫兹；"RSSI (dBm)"字段以dBm为单位展示RSSI数值；"Location"字段标注数据采集的地点；"Label"为分类字段，取值为1或0，其中1代表"可信"，0代表"未授权"。
最终生成的数据集分布均衡，其中"可信（authentic）"数据集包含2442条样本，"未授权（unauthorized）"数据集包含2383条样本。图
ef{dataset}展示了"可信（authentic）"数据集的前5行数据。
随后，该数据集被预处理以适配实验的实现阶段。