JP2017152845A - Information prediction method, information prediction system and information prediction program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce required computational complexity that is required for prediction, by realizing high prediction accuracy when predicting unknown information at a future time based on known information of a time-varying prediction object.SOLUTION: Known information of a prediction object is stored S21, first to Sth differential information are successively generated from the known information or last differential information while defining an N unit time as an interval S22. Based on the generated Sth differential information, a prediction coefficient required for prediction calculation is generated S23 and based on the generated prediction coefficient and Sth differential information, Sth differential information after the N unit time is predicted S24. Based on the predicted Sth differential information after the N unit time, first differential information after the N unit time is predicted by accumulating every differential information S25 and based on the predicted first differential information after the N unit time, unknown information after the N unit time is predicted and outputted S26.SELECTED DRAWING: Figure 6

Description

  The present invention relates to an information prediction method, an information prediction system, and an information prediction program for predicting future unknown information based on known information of a prediction target that varies with time.

  The present invention, for example, in a multi-user communication network to which transmission weight control or reception weight control is applied or in a heterogeneous communication network in which heterogeneous systems coexist, allows transmission weight and reception due to time fluctuation of a channel propagation response (channel response). The present invention relates to a propagation response for avoiding or mitigating deterioration of control accuracy due to weights or a technology for predicting information depending on the propagation response. The present invention also relates to a prediction technique applicable to prediction of audio information or video information that varies with time.

  In recent years, wireless base stations and wireless terminals having functions of wireless communication stations have been rapidly increased, supported by various factors such as convenience, miniaturization, and the spread of mobile applications. Accordingly, there is an increasing demand for highly efficient use of radio wave resources suitable for limited wireless communication.

  Conventionally, in order to avoid a decrease in throughput due to interference between radio signals transmitted by a large number of radio stations, radio resources are arranged on a frequency axis, a time axis, a space axis, or a frequency axis and a time axis. Divided use on a multi-axis combined with a space axis is the mainstream. Specifically, in the conventional interference control technology, the entire radio resources that can be used for a large number of radio stations are first divided, and each radio station performs communication within the range of radio resources obtained by the division. Interference with other radio stations is avoided.

  On the other hand, as a new interference control technology approach in recent years, multiuser MIMO (hereinafter referred to as MU-MIMO) technology (Non-Patent Documents 1 and 2), interference alignment (Interference Alignment, hereinafter referred to as IA) technology (non-patent) Documents 3 and 4) have been proposed. These interference control technologies do not divide the entire radio resources that can be used by many radio stations as in the past, but allow for interference that occurs between radio stations so that each radio station can use the entire radio resources. Yes. Then, instead of dividing, each wireless station performs interference control for interference signals other than the desired signal through transmission / reception weights and reception weights.

  In order to realize interference control by the MU-MIMO technique and the IA technique, highly accurate transmission weights and reception weights calculated based on accurate propagation responses are indispensable. On the other hand, there is a time difference between the acquisition time of the propagation response and the transmission time of the information packet, and when the time variation of the propagation path is fast, the acquired propagation response becomes obsolete with time. The accuracy of transmission weights and reception weights generated from such an obsolete inaccurate propagation response is greatly degraded, and the interference control effect by the MU-MIMO technique and the IA technique is significantly reduced.

  On the other hand, a propagation response prediction technique (Non-Patent Documents 5 and 6) has been proposed as a countermeasure against the deterioration of propagation response due to time variation of the propagation path and the deterioration of transmission / reception weight accuracy. However, any of them can only deal with an environment where the propagation path time fluctuates slowly, and is difficult to apply in an environment where the propagation path time fluctuation is relatively fast. In addition, the conventional propagation response prediction technology requires a large amount of computation, and it is very difficult to put it to practical use.

Quentin H. Spencer, A. Lee Swindlehurst, and Martin Haardt, “Zero-Forcing Methods for Downlink Spatial Multiplexing in Multiuser MIMOChannels,” IEEE Trans. On Signal Processing, vol.52, no.2, pp.461-471, Feb . 2004. David Gesbert, Marios Kountouris, Robert W. Heath Jr., Chan-Byoung Chae, and Thomas Salzer, “Shifting the MIMO Paradigm,” IEEE Signal Processing Magazine, vol.24, no.5, pp.36-46, September 2007 . V. Cadambe and S. Jafar, “Interference alignment and degrees of freedom of the K-user interference channel,” IEEE Trans. Infomation Theory, vol.54, no.8, pp.3425-3441, Aug. 2008. K. Gomadam, V. R. Cadambe and S. A. Jafar, “A Distributed Numerical Approach to Interference Alignment and Applications to Wireless Interference Networks,” IEEE Trans. Inf. Theory, vol.57, no.6, pp.3309-3322, June 2011. D. Anming, Z. Haixia and Y. Dongfeng, “Achievable rate improvement through channel prediction for interference alignment,” in Proc. Asia-Pacic Conference on Communications (APCC), pp.293-298, Aug. 2013. T. Al-Naffouri, “An EM-based forward-backward kalman filter for the estimation of time-variant channels in OFDM,” IEEE Trans. Signal Processing, vol.55, no.7, pp.3924-3930, 2007.

The following are examples of problems in conventional information prediction techniques such as propagation response of communication channels that fluctuate over time.
(1) When the time fluctuation of the communication path is relatively fast, the prediction accuracy of the propagation response is lowered, and the accuracy of the transmission weight and the reception weight generated for interference control is significantly deteriorated.
(2) The amount of computation required for prediction is large, and in order to put it to practical use, a device that can greatly reduce the amount of computation is required.

  The present invention achieves high prediction accuracy by suppressing prediction errors by limiting the range of prediction values when predicting a time-varying propagation response for interference control by, for example, MU-MIMO technology or IA technology. In addition, an object of the present invention is to provide an information prediction method, an information prediction system, and an information prediction program that can reduce the amount of calculation required for prediction.

  1st invention preserve | saves the known information of a prediction object in the information prediction method which estimates unknown information of the future time based on the known information of the prediction object which fluctuates with time, N unit time (N is an integer greater than or equal to 1) Necessary for prediction calculation based on step 1 for generating the first to S-th difference information (S is an integer of 1 or more) in order from the known information or the previous difference information, and the generated S-th difference information. Step 2 for generating a prediction coefficient, Step 3 for predicting the S-th difference information that is N unit time ahead based on the generated prediction coefficient and the S-th difference information, and the S-th difference that is N unit time ahead Based on the information, Step 4 for predicting the first difference information of N unit time ahead by accumulating the difference information of each time, and unknown information of N unit time ahead based on the first difference information of the predicted N unit time ahead Predict and And a step 5 of. In step 1, at least S + 2 pieces of known information are stored, and at least two pieces of difference information for the Sth time are generated.

  According to a second aspect of the present invention, in the information prediction system for predicting unknown information of the future time based on the known information of the prediction target that fluctuates over time, the known information of the prediction target is stored, and N unit time (N is an integer of 1 or more) Based on the first means for sequentially generating the first to S-th difference information (S is an integer of 1 or more) from the known information or the previous difference information, and the generated S-th difference information. A second means for generating a necessary prediction coefficient; a second means for predicting the S-th difference information ahead of N unit times based on the generated prediction coefficient and the S-th difference information; and a predicted N unit time Based on the difference information of the previous S time, based on the fourth means for predicting the first difference information of N unit time ahead by the accumulation of the difference information of each time, and based on the first difference information of the predicted N unit time ahead , Predict unknown information N units ahead And a fifth means for outputting. Further, the first means is configured to store at least S + 2 pieces of known information and generate at least two pieces of difference information for the Sth time.

  In the information prediction system according to the second aspect of the present invention, the transmission weights of at least one transmitting station and a plurality of receiving stations that perform wireless communication at the same frequency and at the same time depend on the propagation response of the communication channel that varies with time. In order to suppress interference signals other than the desired signal at each receiving station by setting reception weights, the propagation response or information depending on the propagation response may be predicted as information to be predicted.

  In the information prediction system according to the second aspect of the invention, a transmitting station or a receiving station or a central processing station connected to the transmitting station stores information on a known propagation response, and predicts information depending on an unknown propagation response or propagation response. It is good also as composition which performs.

  The information prediction program of the third invention causes the computer to execute processing in each means of the information prediction system of the second invention, and predicts unknown information of the future time based on the known information of the prediction target that fluctuates over time.

  In the present invention, for example, paying attention to the feature that the value width of the difference information of the propagation response is small compared to the time variation of the propagation response, for example, instead of directly predicting the unknown propagation response from the known propagation response, An unknown propagation response can be predicted by predicting the difference information and then adding the predicted difference information and the known propagation response. As described above, in the information prediction technique of the present invention, the prediction error is suppressed by limiting the range of the prediction value, and higher prediction accuracy than that of the conventional information prediction technique can be expected.

  Furthermore, in order to calculate a prediction value of a plurality of unit time ahead in the information prediction technique of the present invention, the prediction of one unit time ahead is not performed a plurality of times as in the conventional information prediction technique, but a plurality of unit time ahead is calculated. Can be calculated only once, the required amount of computation required for the prediction can be greatly reduced.

  Furthermore, since the information prediction technology of the present invention is a prediction model with a newly reduced dimension compared to the prediction model used in the conventional information prediction technology, the required amount of computation can be greatly reduced by suppressing the dimensions of the prediction calculation. .

It is a figure which shows the structural example 1 of the radio | wireless communications system to which the information prediction method of this invention is applied. It is a figure which shows the structural example 2 of the radio | wireless communications system with which the information prediction method of this invention is applied. It is a figure which shows the outline | summary of the information prediction method of this invention. It is a figure which shows the process example of the information prediction method of this invention. It is a flowchart which shows the process procedure example 1 by the information prediction method of this invention. It is a flowchart which shows the process procedure example 2 by the information prediction method of this invention. It is a figure which shows the simulation item which evaluates the information prediction method of this invention. It is a figure which shows the characteristic evaluation of a prediction error. It is a figure which shows the characteristic evaluation of a required calculation amount.

FIG. 1 shows a configuration example 1 of a wireless communication system to which the information prediction method of the present invention is applied.
In FIG. 1, there is one transmitting station 1 and K (K is an integer of 2 or more) receiving stations 2-i (i is 1 to K), and one transmitting station 1 is K receiving stations. This is a MU-MIMO configuration that simultaneously transmits information toward 2-i. In that case, the transmission station 1 transmits the desired signal addressed to each receiving station 2-i, and at the same time, the transmission weight V and the receiving station of the transmitting station 1 can be suppressed or avoided so that the interference signal to each receiving station 2-i can be suppressed or avoided. The interference signal is controlled by the reception weights U _{1 to} U _K of 2-1 to 2-K.

FIG. 2 shows a configuration example 2 of a wireless communication system to which the information prediction method of the present invention is applied.
In FIG. 2, there are K sets of transmitting stations 1-i and receiving stations 2-i, and the MU-MIMO configuration performs information transmission simultaneously. In this case, the receiving station 2-i receives K-1 interference signals from other transmitting stations in addition to the desired signal from the transmitting station 1-i that performs wireless communication. The transmission weights V _{1 to} V _K are set in the transmission stations 1-1 to ₁ -K so that the interference signal can be deleted in each reception station, and the reception weight U ₁ is set to the transmission stations 2-1 to 2-K. set ~U _K, the control of the interference signal is performed.

  Here, the radio communication system shown in FIGS. 1 and 2 may be either a single carrier system or a multicarrier system.

  The transmission weight and reception weight used for interference control are generated by predicting an unknown propagation response from a known propagation response in the transmitting station or the receiving station, or input to the central processing station 3 connected to the transmitting station. Alternatively, an unknown propagation response may be predicted and generated from a known propagation response between transmitting and receiving stations.

  As described above, in order to realize the interference control by the MU-MIMO technique or the AI technique, a highly accurate transmission weight or reception weight generated based on an accurate propagation response is indispensable. On the other hand, there is a time difference between the acquisition time of the propagation response and the transmission time of the information packet, and when the time variation of the propagation path is fast, the acquired propagation response becomes obsolete with time, and the generated transmission weight is generated. In addition, the accuracy of the reception weight is greatly deteriorated, and the effect of the interference control by the MU-MIMO technique and the AI technique is remarkably reduced.

  The problem of deterioration in accuracy of transmission weights and reception weights due to such time variation of propagation response becomes serious particularly when the time variation of the propagation path is fast. In that case, the time variation of the propagation response becomes severe, and it is very difficult to accurately predict the propagation response using the conventional prediction technique.

  The feature of the present invention is to focus on the feature that the value width of the difference information of the propagation response is small compared to the time variation of the propagation response, and instead of directly predicting the propagation response, first predict the difference information of the propagation response having a small value width, Next, the future unknown propagation response is predicted by adding the predicted difference information and the known propagation response. In the present invention, the prediction error is suppressed by limiting the range of the prediction value, and higher prediction accuracy than the conventional prediction technology can be expected.

FIG. 3 shows an overview of the information prediction method of the present invention.
In FIG. 3, conventionally, the unknown information h (t) is directly predicted as a prediction range from the known information h (t-1), h (t-2),. In the information prediction technique of the invention, the difference value d (t) is first predicted.

The difference value d (t) is
d (t) = h (t) -h (t-1)
Therefore, the information prediction technique of the present invention effectively suppresses the prediction error with respect to the h (t-1) value width portion as compared with the conventional technique.

  For example, in the prediction method in which the average prediction error is α = 20%, the future time value h (t) = 100 and the difference value d (t) = h (t) −h (t−1) = 10 to be predicted In this case, the value h (t) that can be predicted by the prior art that directly predicts h (t) is in the range of 80 to 120, whereas the value h that can be predicted by the inventive technology that predicts d (t). (t) = d (t) + h (t-1) is in the range of 98-102. Therefore, the information prediction technique of the present invention captures a value to be predicted more accurately than the conventional information prediction technique.

  Hereinafter, embodiments of the information prediction technique based on difference information according to the present invention will be described in detail. In the wireless communication system shown in FIG. 1 or FIG. 2, the prediction process is performed at any one of a transmitting station, a receiving station, and a central processing station.

Example 1
In the first embodiment, the processing station stores the known information to be predicted, generates difference information at intervals of one unit time, and predicts unknown information one unit time ahead using the front-rear Kalman filter based on the difference information. To do.

FIG. 5 shows a processing procedure example 1 by the information processing method of the present invention.
In FIG. 5, the processing station stores L pieces of known information (propagation responses h (t-1), h (t-2),..., H (tL)) to be predicted, and predicts as shown in the following equation. It is set as an input value for calculation (S11). That is, an input value d ₀ (t) for prediction calculation shown below corresponds to an initial value before calculating difference information from known information h (t).
d ₀ (t-1) ← h (t-1)
d ₀ (t-2) ← h (t-2)
:
d ₀ (tL) ← h (tL)

  Next, the processing station sequentially generates the first difference information from the known information to be predicted and the next difference information from the previous difference information until the S-th time in order by 1 unit time as an interval (S12). When S = 1, the first difference information is generated from the known information, and when S = 2, the second difference information is generated from the first difference information. The same applies to S = 3 and thereafter. Here, FIG. 4 shows an example of generation of difference information when the number of generations of difference information S = 2 and the number of known information L used for prediction calculation L = 4. As shown in FIG. 4, it can be seen that the difference information becomes gradually gentler as the difference information generation count S increases. Here, capital letters H (t) and D (t) indicate unknown information to be predicted.

Hereinafter, a generation example of difference information corresponding to the number of generations 1, 2,.
First time: D ₁ (t−1) = D ₀ (t) −d ₀ (t−1)
d ₁ (t-2) = d ₀ (t-1) −d ₀ (t−2)
d ₁ (t-3) = d ₀ (t-2) −d ₀ (t-3)
:
d ₁ (tL) = d ₀ (t−L + 1) −d ₀ (tL)

Second time: D ₂ (t−2) = D ₁ (t−1) −d ₁ (t−2)
d ₂ (t-3) = d ₁ (t-2) −d ₁ (t-3)
d ₂ (t-4) = d ₁ (t-3) −d ₁ (t-4)
:
d ₂ (tL) = d ₁ (t−L + 1) −d ₁ (tL)

S-th: D _S (tS) = D _S-1 (t-S + 1) −d _S-1 (tS)
d _S (tS-1) = d _S-1 (tS) −d _S-1 (tS-1)
d _S (tS-2) = d _S-1 (tS-1) −d _S-1 (tS-2)
:
d _S (tL) = d _S-1 (t-L + 1) −d _S-1 (tL)

  Next, the processing station generates a prediction coefficient A (t-L) necessary for the prediction calculation based on the S-th difference information generated from the known information from time t-L to time t-1 (S13). The prediction coefficient A (t-L) is determined by the prediction matrices B (t-L) and C (t-L) of the forward / backward Kalman filter depending on the known information from time t-L to time t-1.

Next, the processing station predicts the S-th difference information D _S (tS) one unit time ahead based on the generated prediction coefficient A (tL) and the known S-th difference information (S14).

Next, based on the generated S-th difference information D _S (tS) one unit time ahead, the processing station obtains the first difference information D ₁ (t−1) one unit time ahead through the accumulation of the difference information. Predict (S15).

Next, the processing station predicts difference information D ₀ (t) one unit time ahead based on the generated first difference information D ₁ (t−1) one unit time ahead, and unknown information H (t ) (S16).
D ₀ (t) = d ₀ (t-1) + D ₁ (t-1)
H (t) ← D ₀ (t)

(Example 2)
In the second embodiment, the processing station stores the known information to be predicted, generates difference information at intervals of one unit time, and predicts unknown information one unit time ahead using the forward Kalman filter based on the difference information. .

  The processing procedure of the second embodiment is the same as that of the processing procedure example 1 shown in FIG. 5, and the processing station saves the known information to be predicted and uses it as an input value for prediction calculation (S11). The first difference information from the known information to be predicted and the next difference information from the previous difference information are generated in order up to the Sth (S12).

  Next, the processing station generates a prediction coefficient A (t-L) necessary for the prediction calculation based on the S-th difference information generated from the known information from time t-L to time t-1 (S13). The prediction coefficient A (t-L) is determined by the prediction matrix B (t-L) and C (t-L) of the forward Kalman filter depending on the known information from time t-L to time t-1.

Next, as in the first embodiment, the processing station predicts the S-th difference information D _S (tS) one unit time ahead based on the generated prediction coefficient A (tL) and the known S-th difference information ( S14), based on the generated S-th difference information D _S (tS) one unit time ahead, the first difference information D ₁ (t-1) one unit time ahead is predicted through the accumulation of the difference information (S15). ) Based on the predicted first difference information D ₁ (t−1) one unit time ahead, the difference information D ₀ (t) one unit time ahead is predicted and output as unknown information H (t) ( S16).

(Example 3)
In the third embodiment, the processing station stores known information to be predicted, generates difference information at intervals of one unit time, and predicts unknown information one unit time ahead using a backward Kalman filter based on the difference information. .

  The processing procedure of the third embodiment is the same as that of the processing procedure example 1 shown in FIG. 5, and the processing station stores the known information to be predicted as an input value for the prediction calculation (S11), and then 1 unit time interval. The first difference information from the known information to be predicted and the next difference information from the previous difference information are generated in order up to the Sth (S12).

  Next, the processing station generates a prediction coefficient A (t-L) necessary for the prediction calculation based on the S-th difference information generated from the known information from time t-L to time t-1 (S13). The prediction coefficient A (t-L) is determined by the prediction matrices B (t-L) and C (t-L) of the backward Kalman filter depending on the known information from time t-L to time t-1.

Next, as in the first embodiment, the processing station predicts the S-th difference information D _S (tS) one unit time ahead based on the generated prediction coefficient A (tL) and the known S-th difference information ( S14), based on the generated S-th difference information D _S (tS) one unit time ahead, the first difference information D ₁ (t-1) one unit time ahead is predicted through the accumulation of the difference information (S15). ) Based on the predicted first difference information D ₁ (t−1) one unit time ahead, the difference information D ₀ (t) one unit time ahead is predicted and output as unknown information H (t) ( S16).

  In the above-described Examples 1 to 3, an example in which unknown information one unit time ahead is predicted is shown. However, in Examples 4 to 6 shown below, unknown information in general N unit time ahead is predicted. An example is shown. Note that N is an integer equal to or greater than 1. However, when N = 1, the first to third embodiments are obtained.

Example 4
In the fourth embodiment, the processing station stores the known information to be predicted, generates difference information at intervals of N unit times (N is an integer of 2 or more), and the future of N unit times ahead based on the difference information. Information is predicted using a front-back Kalman filter.

FIG. 6 shows a processing procedure example 2 according to the information processing method of the present invention.
In FIG. 6, the processing station stores L pieces of known information (propagation responses h (t-1), h (t-2),..., H (tL)) to be predicted, and predicts them as shown in the following equation. It is set as an input value for calculation (S21). That is, d ₀ (t) shown below corresponds to an initial value before calculating difference information from known information h (t).
d ₀ (t-1) ← h (t-1)
d ₀ (t-2) ← h (t-2)
:
d ₀ (tL) ← h (tL)

  Next, the processing station sequentially generates the first difference information from the known information to be predicted and the next difference information from the previous difference information up to the S-th time in order by N unit time as an interval (S22). Here, FIG. 4 shows an example of generation of difference information when N = 1, the number of generations of difference information S = 2, and the number of past information used for prediction L = 4. The time interval to be expanded from 1 unit time to N unit time.

Hereinafter, a generation example of difference information corresponding to the number of generations 1, 2,.
First time: D ₁ (t−1) = D ₀ (t + N−1) −d ₀ (t−1)
d ₁ (tN-1) = d ₀ (t-1) -d ₀ (tN-1)
d ₁ (tN-2) = d ₀ (t-2) −d ₀ (tN-2)
:
d ₁ (tL) = d ₀ (t−L + N) −d ₀ (tL)

Second time: D ₂ (tN-1) = D ₁ (t-1)-d ₁ (tN-1)
d ₂ (t-2N-1) = d ₁ (tN-1) −d ₁ (t-2N-1)
d ₂ (t-2N-2) = d ₁ (tN-2) −d ₁ (t-2N-2)
:
d ₂ (tL) = d ₁ (t−L + N) −d ₁ (tL)

S-th: D _S (t- (S-1) N-1) = D _S-1 (t- (S-2) N-1) -d _S-1 (t- (S-1) N-1 )
d _S (t-SN-1) = d _S-1 (t- (S-1) N-1) −d _S-1 (t-SN-1)
d _S (t-SN-2) = d _S-1 (t- (S-1) N-2) −d _S-1 (t-SN-2)
:
d _S (tL) = d _S-1 (t-L + 1) −d _S-1 (tL)

  Next, the processing station generates a prediction coefficient A (t-L) necessary for prediction calculation based on the S-th difference information generated from the known information from time t-L to time t-1 (S23). The prediction coefficient A (t-L) is determined by the prediction matrices B (t-L) and C (t-L) of the forward / backward Kalman filter depending on the known information from time t-L to time t-1.

Next, based on the generated prediction coefficient A (tL) and the known S-th difference information, the processing station calculates the S-th difference information D _S (t− (S−1) N−1) N units ahead. Prediction is made (S24).

Next, based on the generated S-th difference information D _S (tS) ahead of the N unit time, the processing station obtains the first difference information D ₁ (t−1) ahead of the N unit time through the accumulation of the difference information. Predict (S25).

Next, the processing station predicts the difference information D ₀ (t + N-1) of N unit time ahead based on the generated first difference information D ₁ (t-1) of N unit time ahead, and is unknown. Information H (t + N-1) is output (S26).
D ₀ (t + N-1) = d ₀ (t-1) + D ₁ (t-1)
H (t + N-1) ← D ₀ (t + N-1)

(Example 5)
In the fifth embodiment, the processing station stores the known information to be predicted, generates difference information at intervals of N unit time, and predicts future information ahead of N unit time using the forward Kalman filter based on the difference information. To do.

  The processing procedure of the fifth embodiment is the same as that of the processing procedure example 2 shown in FIG. 6, and the processing station stores the known information to be predicted and uses it as an input value for prediction calculation (S21), and then sets N unit time intervals. The first difference information from the known information to be predicted and the next difference information from the previous difference information are generated in order up to the Sth (S22).

  Next, the processing station generates a prediction coefficient A (t-L) necessary for prediction calculation based on the S-th difference information generated from the known information from time t-L to time t-1 (S23). The prediction coefficient A (t-L) is determined by the prediction matrix B (t-L) and C (t-L) of the forward Kalman filter depending on the known information from time t-L to time t-1.

Next, as in the fourth embodiment, the processing station, based on the generated prediction coefficient A (tL) and known S-th difference information, N-th time ahead S-th difference information D _S (t− (S−1 ) N-1) is predicted (S24), and N unit time is obtained through accumulation of the difference information each time based on the generated Nth time difference information D _S (t- (S-1) N-1) ahead of N unit time. The previous first difference information D ₁ (t-1) is predicted (S25), and based on the predicted first difference information D ₁ (t-1) N units ahead, the difference information D N units ahead ₀ (t + N-1) is predicted and output as unknown information H (t + N-1) (S26).

(Example 6)
In the sixth embodiment, the processing station stores the known information to be predicted, generates difference information at intervals of N unit time, and predicts future information ahead of N unit time using a backward Kalman filter based on the difference information. To do.

  The processing procedure of the sixth embodiment is the same as that of the processing procedure example 2 shown in FIG. 6, and the processing station stores the known information to be predicted and uses it as an input value for prediction calculation (S21). The first difference information from the known information to be predicted and the next difference information from the previous difference information are generated in order up to the Sth (S22).

  Next, the processing station generates a prediction coefficient A (t-L) necessary for prediction calculation based on the S-th difference information generated from the known information from time t-L to time t-1 (S23). The prediction coefficient A (t-L) is determined by the prediction matrices B (t-L) and C (t-L) of the backward Kalman filter depending on the known information from time t-L to time t-1.

Next, as in the fourth embodiment, the processing station, based on the generated prediction coefficient A (tL) and known S-th difference information, N-th time ahead S-th difference information D _S (t− (S−1 ) N-1) is predicted (S24), and N unit time is obtained through accumulation of the difference information each time based on the generated Nth time difference information D _S (t- (S-1) N-1) ahead of N unit time. The previous first difference information D ₁ (t-1) is predicted (S25), and based on the predicted first difference information D ₁ (t-1) N units ahead, the difference information D N units ahead ₀ (t + N-1) is predicted and output as unknown information H (t + N-1) (S26).

(Supplement)
In the first and fourth embodiments, the front and rear Kalman filters are used, in the second and fifth embodiments, the front Kalman filter is used. In the third and sixth embodiments, the rear Kalman filter is used. It is not limited by the difference in method or algorithm used.

  The “information” in the first to sixth embodiments is a propagation response that varies with time or information that depends on the propagation response. For example, transmission weights and reception weights that depend on the propagation response, and an interference matrix that represents the amount of interference as described in Equation (18) and Equation (30) of Non-Patent Document 4 correspond to information that depends on the propagation response. To do. The present invention can be applied not only to the propagation response itself, but also to any information that depends on a time-varying propagation response.

  Further, in the first to sixth embodiments, in a multi-antenna communication configuration having a plurality of transmission antennas or a plurality of reception antennas, between one transmission / reception antenna pair, between some transmission / reception antenna pairs, or between all transmission / reception antenna pairs. Application to propagation response or information depending on propagation response is possible. The MU-MIMO communication configuration and the IA communication configuration in FIGS. 1 and 2 are specific examples of the multi-antenna communication configuration.

  In addition, the propagation response or the information depending on the propagation response in the first to sixth embodiments may be either time-domain information expressed on the time axis or frequency-domain information expressed on the frequency axis. If the information is in the time domain, the present invention can be applied to each delayed wave component on the time axis. If it is information in the frequency domain, the present invention can be applied to each subcarrier component on the frequency axis.

  Here, the quantitative evaluation performed in the IA alignment communication configuration shown in FIG. 2 is shown. As specifications used for the quantitative numerical evaluation, there are three transmitting station receiving station pairs as shown in FIG. 7, and the transmitting and receiving stations here are provided with two antennas. In addition, the maximum Doppler frequency is assumed to be 100 Hz as the time variation of the radio propagation path.

  For simplicity, the conventional prediction technique uses the Forward Backward Kalman Filter technique of Non-Patent Document 6, and the prediction technique of the present invention uses the second embodiment.

FIG. 8 shows a prediction error characteristic evaluation.
The prediction error in the prediction technique of the present invention indicates the prediction accuracy of the forward Kalman filter shown in the second embodiment. The prediction error in the conventional prediction technique indicates the prediction accuracy of the front-rear Kalman filter. RMSE (Root Mean Squared Error) is used as an index representing the prediction accuracy. Furthermore, the prediction technique of the present invention shows three patterns in which the number of generations of difference information is S = 1, 2, 3. In FIG. 8, it can be seen that the prediction technique of the present invention achieves higher prediction accuracy than the conventional prediction technique. In addition, it can be confirmed that the prediction accuracy of the prediction technique of the present invention increases as the difference information generation count S increases.

FIG. 9 shows a characteristic evaluation of the required calculation amount.
The required calculation amount in the prediction technique of the present invention is that in the second embodiment. The number of four arithmetic operations is used as an index representing the required calculation amount. Furthermore, the prediction technique of the present invention shows a pattern in which the number of generations of difference information is S = 1. The order of the Kalman filter used for the prediction shows three patterns of M = 2, 3, and 4 in both the prediction technique of the present invention and the conventional prediction technique. In FIG. 9, it can be seen that the prediction technique of the present invention reduces the required amount of calculation compared to the conventional prediction technique in any Kalman filter order. It can also be confirmed that the calculation amount reduction effect becomes remarkable as the AR order increases.

  The present invention can realize, for example, a propagation response prediction process in a transmission station, a reception station, or a central processing station of a wireless communication system by a computer and a computer program that performs the above process. This computer program can be stored in a computer-readable storage medium or provided via a network.

1 Transmitting station 2 Receiving station 3 Central processing station

Claims (7)

In the information prediction method for predicting unknown information of the future time based on the known information of the prediction target that fluctuates over time,
The known information to be predicted is stored, and N-unit time (N is an integer of 1 or more) is used as an interval, and the first to S-th difference information (S is an integer of 1 or more) is sequentially ordered from the known information or the previous difference information. Step 1 to generate
Step 2 of generating a prediction coefficient necessary for prediction calculation based on the generated S-th difference information;
Based on the generated prediction coefficient and the S-th difference information, predicting S-th difference information N units time ahead;
Step 4 of predicting the first difference information of N unit time ahead by accumulating the difference information of each time based on the predicted S time difference information of N unit time ahead;
An information prediction method comprising: predicting and outputting the unknown information of N unit time ahead based on the first difference information of the predicted N unit time ahead. The information prediction method according to claim 1,
The step 1 stores at least S + 2 pieces of the known information, and generates at least two pieces of difference information for the Sth time. In an information prediction system that predicts unknown information of future time based on the known information of the prediction target that fluctuates over time,
The known information to be predicted is stored, and N-unit time (N is an integer of 1 or more) is used as an interval, and the first to S-th difference information (S is an integer of 1 or more) is sequentially ordered from the known information or the previous difference information. First means for generating
A second means for generating a prediction coefficient necessary for a prediction calculation based on the generated S-th difference information;
Based on the generated prediction coefficient and the S-th difference information, second means for predicting the S-th difference information N units ahead;
A fourth means for predicting the first difference information of N unit time ahead by accumulating the difference information of each time based on the predicted S time difference information of N unit time ahead;
An information prediction system comprising: fifth means for predicting and outputting the unknown information of N unit time ahead based on the first difference information of the predicted N unit time ahead. The information prediction system according to claim 3,
The information predicting system according to claim 1, wherein the first means is configured to store at least S + 2 pieces of the known information and generate at least two pieces of difference information for the Sth time. The information prediction system according to claim 3,
Each receiving station sets a transmission weight and a reception weight according to the propagation response of a communication path that varies in time between at least one transmitting station and a plurality of receiving stations that perform wireless communication at the same frequency and the same time. In order to suppress interference signals other than the desired signal, the information prediction system is configured to predict the propagation response or information dependent on the propagation response as the information to be predicted. The information prediction system according to claim 5,
A configuration in which a central processing station connected to the transmitting station, the receiving station or the transmitting station inputs information on the known propagation response and predicts the unknown propagation response or information dependent on the propagation response. An information prediction system characterized by being.   An information prediction program for causing a computer to execute processing in each means of the information prediction system according to claim 3 and predicting unknown information of a future time based on known information of a prediction target that varies with time.

Images