CN115190546A - A Handover Method of LTE-M System Based on Neural Network Prediction - Google Patents

Abstract

The invention discloses a handover switching method of an LTE-M system based on neural network prediction, which belongs to the technical field of communication and rail transit and comprises the following steps: 1) Collecting: acquiring signal intensity values of all base stations along the line trend, wherein the signal intensity values comprise sampling in an approaching state and a leaving state, and a training data set is formed; 2) Training: training the neural network with a training data set; 3) The application comprises the following steps: detecting the signal intensity of a source base station by adopting a trained neural network in the running process of the train, and calculating by using the neural network corresponding to the source base station to obtain a predicted value of the source base station; and detecting the signal intensity of the target base station, calculating by using a neural network corresponding to the target base station to obtain a predicted value of the target base station, comparing the two predicted values according to a preset algorithm, and determining whether to perform network switching on the vehicle-mounted equipment or not according to a comparison result. The invention has better adaptability to wireless propagation environment.

Description

A Handover Method of LTE-M System Based on Neural Network Prediction

technical field

The invention belongs to the technical field of communication and the technical field of rail transit.

Background technique

With the development of wireless communication technology, the wireless communication system of rail transit vehicles is developing from narrowband communication to digitalization, IPization, broadbandization and wireless communication supporting high-speed movement. In urban rail vehicle-ground communication, LTE-M technology has become the mainstream technology for the transmission of services such as vehicle-ground information, PIS, and CCTV that carry CBTC. At the same time, in order to shorten the travel time, domestic urban rail transit has gradually developed towards a high speed. Designing a handover scheme according to the characteristics of the LTE-M system and the high-speed train operating environment is of great significance to ensure the quality of the wireless communication between the train and the ground.

The current standard LTE handover decision strategy assumes that the mobile node may move forward or backward, the movement route has strong randomness, and the movement speed is not too fast. The base station usually uses the A3 event-based handover strategy when performing handover, that is, when the RSRP of the reference signal received by the adjacent base station is higher than a certain hysteresis value (Hys) of the current serving base station and maintains a certain delay time (TTT), the A3 event is triggered to initiate the handover. .

In the current LTE-M network, a mobile relay (MR) on top of the train connects the eNB and various in-vehicle devices including train control equipment. High signal strength between the MR and its serving eNB is crucial to ensure reliable communication between the train and the control center. The stronger the signal strength, the easier it is for packets to be successfully delivered over the wireless channel. A stronger signal also helps reduce the likelihood of communication drops to optimize reliability. The main purpose of the existing A3 event handover strategy design is to avoid ping-pong handover at the edge of adjacent cells when the user moves randomly at a low speed. However, in the actual high-speed urban rail operation scenario, the train always travels in one direction during the switching process, and the running trajectory is fixed. In addition, due to high-speed movement, the RSRP curves received by the MR from the serving eNB and the handover target eNB have distinct characteristics. Compared with ordinary user scenarios, it is not easy for the received RSRP values of the serving base station and the handover target base station to frequently surpass each other. . Therefore, when the A3 handover strategy is applied to the high-speed urban rail, under the preset Hys and TTT conditions, even if the mobile relay (MR) on the train is gradually approaching the adjacent cell and can establish a better quality connection with the adjacent eNB, it Still let the serving eNB maintain a poor quality connection with the MR.

The main idea of the current handover scheme combined with the artificial intelligence algorithm is to optimize the parameters of the A3 event, and the present invention adopts different technical implementation schemes.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide an LTE-M system handover scheme based on neural network prediction, so as to provide a more stable communication quality for the train.

The technical solution adopted by the present invention to solve the technical problem is the LTE-M system handover method based on neural network prediction, which is characterized in that it includes the following steps:

1) Acquisition: along the line, collect the signal strength values of each base station, including sampling in the approaching state and leaving state, to form a training data set;

2) Training: Use the training data set to train the neural network, take the signal strength values of the nearest N sampling points as input in a sliding window manner, and predict the signal strength values of the next sampling point. When the accuracy rate reaches the preset value When the training is completed, N is a preset natural number greater than 3;

3) Application: The trained neural network is used to detect the signal strength of the source base station during the running of the train, and the neural network corresponding to the source base station is used for calculation to obtain the predicted value of the source base station; to detect the signal strength of the target base station, Calculate with the neural network corresponding to the target base station to obtain the predicted value of the target base station, compare the two predicted values according to a predetermined algorithm, and decide whether to perform network handover of the vehicle-mounted device according to the comparison result.

Further, in the step 3), the predetermined algorithm is: if the signal strength value of the target base station detected at the current moment is greater than the signal strength value of the source base station, and the predicted signal strength value of the target base station is greater than the predicted signal strength value of the source base station , the network switching is started, otherwise the detection is continued.

Described step 1) is:

Acquisition: along the line, the signal strength value of each base station is collected, including the sampling in the approaching state and the leaving state; in the same acquisition process, the time interval between two adjacent samplings is a predetermined value; training dataset.

The beneficial effects of the present invention are:

1. The Recurrent Neural Network (RNN) in deep learning is used to predict the RSRP sequences of the serving eNB and the handover target eNB received during the train operation, to replace the delay time (TTT) in the traditional handover algorithm. ) to make a handover decision.

2. The training data set of the RNN comes from the historical measurement reports uploaded by the mobile stations within the service range of each eNB, so the obtained prediction network has better adaptability to the wireless propagation environment of the section where the eNB is located.

Description of drawings

1 is a schematic diagram of a handover area of the present invention;

Fig. 2 is the switching flow chart of the present invention;

FIG. 3 is a schematic diagram of the RSRP prediction effect of the mobile station receiving the source eNB according to the present invention;

FIG. 4 is a schematic diagram of the RSRP prediction effect of the mobile station receiving the target eNB according to the present invention;

5 is a schematic diagram of the mean square error of RSRP sequence prediction under different train speeds of the present invention;

6 is a cumulative distribution diagram of the switching trigger probability under the condition of the train speed of 200km/h of the present invention;

7 is a schematic diagram of the evaluation index RSRPgap in the present invention;

FIG. 8 is a comparison diagram of RSRPgap under different train speeds and different switching schemes of the present invention.

FIG. 9 is a schematic diagram of data sampling according to an embodiment.

FIG. 10 is a schematic diagram of input data of a neural network according to an embodiment.

Detailed ways

The current standard LTE handover decision strategy assumes that the mobile node may move forward or backward, the movement route has strong randomness, and the movement speed is not too fast. The base station usually uses the A3 event-based handover strategy when performing handover, that is, when the RSRP of the reference signal received by the adjacent base station is higher than a certain hysteresis value (Hys) of the current serving base station and maintains a certain delay time (TTT), the A3 event is triggered to initiate the handover. .

In the current LTE-M network, a mobile relay (MR) on top of the train connects the eNB and various in-vehicle devices including train control equipment. High signal strength between the MR and its serving eNB is crucial to ensure reliable communication between the train and the control center. The stronger the signal strength, the easier it is for packets to be successfully delivered over the wireless channel. A stronger signal also helps reduce the likelihood of communication drops to optimize reliability. The main purpose of the existing A3 event handover strategy design is to avoid ping-pong handover at the edge of adjacent cells when the user moves randomly at a low speed. However, in the actual high-speed urban rail operation scenario, the train always travels in one direction during the switching process, and the running trajectory is fixed. In addition, due to high-speed movement, the RSRP curves received by the MR from the serving eNB and the handover target eNB (base station) have distinct characteristics. Compared with ordinary user scenarios, it is not easy for the received RSRP values of the serving base station and the handover target base station. transcendent phenomenon. Therefore, when the A3 handover strategy is applied to the high-speed urban rail, under the preset Hys and TTT conditions, even if the mobile relay (MR) on the train is gradually approaching the adjacent cell and can establish a better quality connection with the adjacent eNB, it Still let the serving eNB maintain a poor quality connection with the MR.

The technical solution adopted by the present invention is to predict the signal strength value (RSRP) received by the train MR (mobile relay) through a cyclic neural network, and select an appropriate switching initiation time, so that the train can maintain the highest signal strength during the entire operation of the train. The wireless connection of the good base station improves the link quality during the handover process of the train MR.

The invention uses the RSRP measured and reported by the train MR as the training sample data of the cyclic neural network with the function of time series prediction, and the trained neural network can predict the future time point from the RSRP value at the past time point that has been measured by the train MR. to the RSRP value. If the prediction result reflects that the target base station will have stronger signal strength than the current serving base station, the handover process will be initiated immediately if the handover criterion is met, so that the train MR will be switched to the target base station with higher signal strength in time. Its effect is measured by evaluation indicators such as parameter RSRP gap and cumulative distribution of handover trigger probability.

The present invention includes the following steps:

(1) First, record the RSRP sequences received by the source base station and the handover target base station during the operation of the train MR between two adjacent base stations, and this process is completed by the network side. The sequence interval is the measurement reporting period of the mobile station.

(2) Due to the different wireless signal propagation environments within the service range of different eNBs, and the uniqueness of large-scale fading, the RSRP sequence recording and the training of the RNN should be performed independently by each eNB (each base station eNB has an identity ID) , so that it has better adaptability to RSRP prediction in the coverage area. The recurrent neural network is trained offline using the recorded RSRP sequences. Obtain the networks that predict the RSRP values of the source base station and the target base station, respectively.

(3) Periodically report measurement reports during train running. The network side outputs the RSRP prediction values of the two base stations at the next moment through the recurrent neural network according to the N times of measurement report results at the current and previous moments (N=6 is taken in the performance verification of this embodiment).

(4) When it is satisfied that the RSRP of the target base station at the current moment is higher than a certain threshold of the serving base station, and the predicted value also satisfies the corresponding threshold condition, the handover is started immediately. Otherwise, repeat the previous measurement and prediction process, and wait for the next suitable switching opportunity.

Example:

As an embodiment, the following steps are included:

1) Acquisition: along the line, collect the signal strength values of each base station, including sampling in the approaching state and leaving state, to form a training data set;

2) Training: Use the training data set to train the neural network, take the signal strength values of the nearest N sampling points as input in a sliding window manner, and predict the signal strength values of the next sampling point. When the accuracy rate reaches the preset value When the training is completed, N is a preset natural number greater than 3;

Referring to Figure 9, when the train is running, the data d1 to dn are sampled at times t1 to tn. In the sliding window method shown in Figure 10, the length of the sliding window is set to 6, and each time a group of (6 data) is input to the neural network data). The output of the neural network is the prediction. For example, input d2 to d7, and the output is the predicted value corresponding to d8.

3) Application: Using the trained neural network to detect the signal strength of the source base station (the base station behind the train) during the running of the train, obtain the detection value of the source base station, and use the neural network corresponding to the source base station to calculate and obtain The predicted value of the source base station; detect the signal strength of the target base station (the base station in front of the train), obtain the detected value of the target base station, use the neural network corresponding to the target base station to calculate, obtain the predicted value of the target base station, and compare the two according to the predetermined algorithm. According to the comparison result, it is decided whether to perform network switching on the in-vehicle device.

Further, in the step 3), the predetermined algorithm is: if the signal strength value of the target base station detected at the current moment is greater than the signal strength value of the source base station, and the predicted signal strength value of the target base station is greater than the predicted signal strength value of the source base station , the network switching is started, otherwise the detection is continued. In other words, if the detection value of the target base station is greater than the detection value of the source base station, and the predicted value of the target base station is greater than the predicted value of the source base station, then switch over, otherwise continue to detect.

Further, described step 1) is:

Acquisition: along the line, the signal strength value of each base station is collected, including the sampling in the approaching state and the leaving state; in the same acquisition process, the time interval between two adjacent samplings is a predetermined value; training dataset.

Since the base station has an identification ID, a specific base station has different identities with the train location. For example, when the train is located between base station A and base station B (base station A is behind the train, and base station B is in front of the train), base station A becomes the source base station, Base station B becomes the target base station; when the train travels between base station B and base station C, (base station B is behind the train and base station C is in front of the train), base station B becomes the source base station, base station C becomes the target base station, and so on.

Claims (4)

1. The LTE-M system handover method based on neural network prediction is characterized by comprising the following steps of:

1) Collecting: along the line trend, signal intensity value acquisition is carried out on each base station, including sampling in an approaching state and a leaving state, and a training data set is formed;

2) Training: training a neural network by using a training data set, taking the signal intensity values of the nearest N sampling points as input in a sliding window mode, predicting the signal intensity value of the next sampling point, and completing training when the accuracy reaches a preset condition, wherein N is a preset natural number which is more than 3;

3) The application comprises the following steps: detecting the signal intensity of a source base station by adopting a trained neural network in the running process of the train, and calculating by using the neural network corresponding to the source base station to obtain a predicted value of the source base station; and detecting the signal intensity of the target base station, calculating by using a neural network corresponding to the target base station to obtain a predicted value of the target base station, comparing the two predicted values according to a preset algorithm, and determining whether to perform network switching on the vehicle-mounted equipment or not according to a comparison result.

2. The LTE-M system handover method based on neural network prediction as claimed in claim 1, wherein in the step 3), the predetermined algorithm is: and if the signal intensity value of the target base station detected at the current moment is greater than the signal intensity value of the source base station, and the predicted signal intensity value of the target base station is greater than the predicted signal intensity value of the source base station, starting network switching, otherwise, continuing detection.

3. The method for LTE-M system handover based on neural network prediction as claimed in claim 1, wherein the neural network is a recurrent neural network.

4. The LTE-M system handover method based on neural network prediction as claimed in claim 1, wherein the step 1) is:

collecting: collecting signal intensity values of all base stations along the line trend, wherein the signal intensity values comprise sampling in an approaching state and a leaving state; in the same acquisition process, the time interval between two adjacent samplings is a preset value; the training data set is formed by repeated collection.

Images