CN104113903A - Interactive cognitive learning based downlink power adjusting method and device - Google Patents

Abstract

A downlink power adjustment method and device based on interactive cognitive learning, relating to communication technology. As an intelligent body with learning ability, femtocell takes improving networking performance as the learning goal, and optimizes the network by adjusting the power allocation of resource blocks. In addition to its own learning evolution, femtocells can also conduct interactive learning based on communication similarity and professional knowledge to improve learning efficiency and save long-term energy consumption from accumulating learning experience through the air interface. Network optimization is performed by adjusting the power allocation of resource blocks, which can reduce interference in hybrid networking and improve throughput. In addition to self-learning, femtocell gateways are used as a case library, and femtocells can learn interactively based on similarity and professional knowledge to improve learning efficiency and save long-term energy consumption from accumulating learning experience through the air interface.

Description

Downlink power adjustment method and device based on interactive cognitive learning

technical field

The present invention relates to communication technology, in particular to a downlink power adjustment method and device based on interactive cognitive learning.

Background technique

The survey in the literature "Femtocells: technologies and deployment" (Zhang, J., & De la Roche, G. (2010). Femtocells: technologies and deployment (pp.1-13). New York: Wiley.) shows that in the traditional Two-thirds of the voice services and 90% of the data services in the cell area take place indoors, so the coverage and capacity of the indoor wireless access network are extremely important. However, the price of macro base stations is very expensive, and site selection, equipment installation, commissioning and maintenance will consume a lot of manpower, financial resources, material resources and time. Solving this problem by adding macro base stations will increase the operator's expenses and bring to a large number of complex network planning problems. Against such a background, emerging devices such as femtocells emerge as the times require.

Home base station, also known as femtocell base station, is a low-power wireless access point that works in the authorized frequency band and is accessed by users with existing broadband (such as DSL, cable, optical fiber). Connectivity to the mobile core network. It has the characteristics of low cost, easy installation, automatic configuration, and plug-and-play. It has the same standard and frequency band as other mobile base stations of operators, so mobile terminals such as mobile phones can be used in common use. Its transmission power is not much different from that of Wi-Fi equipment, about 10-100mW, and the coverage radius is 10-50m. It supports several active users and is used to flexibly improve indoor and outdoor wireless signal coverage and increase network capacity. However, interference inevitably exists in the mixed networking of the macro base station and the femtocell. On the one hand, femtocells, as a commodity, are freely purchased and installed by users, so that operators do not know their location distribution; scale, which will further affect the performance of the entire network. From a power point of view, without interference management, the downlink can experience the following negative situations:

(1) The signal of the home base station is too small and is submerged by the coverage of the macro cell signal, resulting in a small coverage area of the home base station and poor signal quality;

(2) If the signal of the home base station is too large relative to the signal of the macro base station, it may cause the macro cell user to lose the connection with the macro base station and enter a dead zone (dead zone), that is, the macro cell user has neither the right to access the home base station in closed mode , and cannot be connected to the macro base station; it may also cause the macro cell user to be interfered too much by the home base station, and the performance is degraded.

Traditional LTE femtocell downlink power adjustment methods mainly include the following:

Fixed power allocation mentioned in the document "Improved Decentralized Q-learning Algorithm for Interference Reduction inLTE-Femtocells" (Serrano A M G. Self-organized Femtocells: a Time Difference Learning Approach[J].2012.).

The intelligent power control (SPC) proposed in the document "Interference control for LTE Rel-9HeNB cells" (Jeju, Interference control for LTE Rel-9HeNB cells[S], 3GPP TSG RAN WG4, R4-094245, November 9th-13th, 2009) )method.

Literature "Cognition and docition in OFDMA-based femtocell networks" (Galindo-Serrano A, Giupponi L, Dohler M. Cognition and docition in OFDMA-based femtocell networks[C]//Global Telecommunications Conference (GLOBECOM 2010), 2010IEEE1.IEEE, :1-6.) mentioned iterative water filling algorithm.

These methods have the following disadvantages:

(1) First, the minimum time-frequency unit that can be allocated to users is defined as a resource block in LTE, while most traditional methods only adjust the maximum total transmit power without considering the power adjustment based on resource blocks;

(2) Secondly, in the distributed management, the method of supporting power adjustment based on resource blocks seldom takes into account the interactive learning between femtocells, and only aims at improving its own performance.

Contents of the invention

The purpose of the present invention is to provide a downlink power adjustment method and device based on interactive cognitive learning to reduce the interference generated by the home base station to the macro cell user and accelerate the realization of the process.

The downlink power adjustment method based on interactive cognitive learning includes two processes:

1. Downlink power adjustment based on interactive cognition, including:

(1) Each home base station maintains a downlink power information table, the downlink power information table uses resource blocks as the smallest unit, and is used to determine the transmission power of each resource block, and initialize all data in the downlink power information table;

(2) The home base station periodically senses the interference caused by the current transmission power configuration, and then updates the corresponding data in the downlink power information table according to the sensed information and corresponding update rules;

(3) Use the updated downlink power information table and the specified allocation method to determine the transmit power configuration for the next cycle;

(4) Steps (2) and (3) are repeated, and the ultimate goal is to make the transmission power allocated according to the sensing information the best;

2. Downlink power adjustment based on interactive learning, including:

(1) The home base station periodically reports the power information table, similarity parameters, professional knowledge and other information maintained by itself as a case to the terminal device with the aggregation function. The terminal device uses a home base station gateway, and the home base station gateway is Set a survival time for each case, after the case exceeds the survival time, it will be automatically deleted;

(2) The home base station performs active learning, and the home base station gateway performs active teaching, and the two methods are carried out simultaneously to adjust the downlink power;

The active learning refers to: for the home base station whose turn-on time is less than the threshold, it actively sends a learning application to the home base station gateway, and reports the similarity parameter and professional knowledge; Similarity, the case with a similarity above the threshold is used as an alternative case, and then the professional knowledge of the alternative cases is compared, and the highest one is selected as the final case, and the corresponding form of the case is sent to the home base station that issued the study application, and After deleting this case from the case library, the base station uses the table information to perform power adjustment based on interactive cognition. If the professional knowledge is improved, it will send a confirmation signal to the gateway and overwrite the original table to end the learning process; otherwise, The gateway selects the case with the highest professional knowledge among the current alternative cases as the final case, and sends it to the home base station that sends the learning application for learning, until the alternative case is empty, the gateway sends feedback information to the base station, and stops the learning process.

The active teaching refers to: for the latest reported case, the femtocell gateway calculates the similarity between the case and the cases in the case database, selects the case with the highest similarity, and compares the professional knowledge of the reported case and the selected case; if the difference is within Within the threshold, stop the teaching process; otherwise, send the form of the case with higher professional knowledge to the femtocell corresponding to the case with lower professional knowledge to perform power adjustment based on interactive cognition. When the knowledge level increases, a confirmation signal is sent to the gateway and the original form is overwritten to end the teaching process; otherwise, the original form is kept and the teaching process ends. This process need not be performed on all newly reported cases.

The above-mentioned downlink power adjustment method of interactive cognition and learning is cyclically executed in actual operation.

The downlink power adjustment device based on interactive cognitive learning includes three module units on the base station side: an information storage module, an information sending/receiving module, and an information processing module;

The main functions of the information storage module are: 1) store the power information table maintained by the power adjustment based on interactive cognition, the information processing module will use the power information table, and store it in the information storage module after processing; 2) temporarily store Based on case information obtained in power adjustments for interactive learning and deciding on the currently used form;

The main functions of the information sending/receiving module are: 1) Periodically receive various information fed back by neighboring cell macrocell users, and report the received value to the information processing module. For interference information, if there is no neighboring cell macro The interference of cellular users or the interference is very small (less than a predetermined threshold) and can be ignored, and the reporting value is artificially specified as 0; if there is interference from multiple adjacent macrocellular users, the one with the largest interference is selected as the reporting value. Make the final processing result satisfy that the interference to all macrocell users falls below the threshold; 2) be responsible for sending its own case and other relevant information to the femtocell gateway;

The main functions of the information processing module are: 1) in the power adjustment based on interactive cognition, process the stored information and report information at the same period as the information receiving module, specifically refer to updating table information and determining resource block transmission power; 2) In the power adjustment based on interactive learning, the self-maintenance form, similarity parameters, professional knowledge and other required information are composed into a case and handed over to the information sending/receiving module.

The invention optimizes the network by adjusting the power distribution of the resource blocks, can reduce the interference of mixed networking, and improve the throughput. In addition to self-learning, femtocell gateways are used as a case library, and femtocells can learn interactively based on similarity and professional knowledge to improve learning efficiency and save long-term energy consumption from accumulating learning experience through the air interface.

Description of drawings

FIG. 1 is a system architecture diagram of a mixed network of LTE macro base stations and home base stations in an example of the present invention.

FIG. 2 is a functional relationship diagram of a power adjustment device module for interactive cognitive learning of a home base station.

Fig. 3 is a flowchart of power adjustment based on interactive cognition of the home base station.

Fig. 4 is a flowchart of active learning based on interactive learning of the home base station.

Fig. 5 is a flowchart of active teaching based on interactive learning by the femtocell gateway.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings.

The implementation scenario of the present invention takes the mixed networking of LTE macro base stations and home base stations as an example, and the system architecture of the mixed networking of LTE macro base stations and home base stations is shown in FIG. 1 .

The present invention is a downlink power adjustment method based on interactive cognitive learning. The main process of the method is: 1. Downlink power adjustment based on interactive cognition; 2. Downlink power adjustment based on interactive learning; It is executed cyclically during operation, and is implemented based on 1. information storage module; 2. sending/receiving module; 3. information processing module. The functional relationship of the modules is shown in Figure 2.

The main functions of the information storage module are: 1) store the power information table maintained by the power adjustment based on interactive cognition, the information processing module will use the power information table, and store it in the information storage module after processing; 2) temporarily store the power information table based on interaction The case information obtained in the power adjustment of formula learning and decides the table currently used.

The main functions of the sending/receiving module are: 1) Periodically receive various information fed back by neighboring macrocell users, and report the received value to the information processing module. For interference information, if there is no neighboring macrocell user’s Interference or interference is very small (less than a predetermined threshold) and can be ignored, and the reporting value is artificially specified as 0; if there is interference from multiple adjacent macrocell users, select the one with the largest interference as the reporting value, so that the final processing The result satisfies that the interference to all macrocell users falls below the threshold; 2) is responsible for sending its own case and other relevant information to the femtocell gateway.

The main functions of the information processing module are: 1) In the power adjustment based on interactive cognition, process the stored information and report information at the same period as the information receiving module, specifically referring to updating table information and determining the transmit power of resource blocks; 2) In the power adjustment based on interactive learning, the self-maintenance table, similarity parameters, professional knowledge and other required information are composed into a case and delivered to the information sending/receiving module.

The present invention provides an embodiment. In this embodiment, the resource allocation of the home base station adopts a proportional fairness algorithm to independently allocate resource blocks to different users; in this embodiment, the downlink power adjustment based on interactive cognition is described by taking Q learning as an example.

The agents, states, actions, and rewards in Q-learning are defined as follows:

Agent: home base station, number of base stations i={1, 2,...,N}, resource block r of each base station={1,2,...,R}.

State s of the i-th home base station on the r-th resource block: s ^i,r ={I ^i,r ,Pow ⁱ }, wherein, I ^i,r is a signal-to-interference-noise ratio indicator.

${\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; if</span>}& \mathrm{<span\; class="notranslate">\; SINR</span>}<span\; class="notranslate">\; \_</span>{\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; <</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; SINR</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; if</span>}& <span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; SINR</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; SINR</span>}<span\; class="notranslate">\; \_</span>{\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; \&le;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; SINR</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\\ \mathrm{<span\; class="notranslate">\; 2</span>}& & \mathrm{<span\; class="notranslate">\; SINR</span>}<span\; class="notranslate">\; \_</span>{\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; ></span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; SIN</span>}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\end{array}\right.$

SINR_I ^i,r is the reported signal-to-interference-noise ratio of the macro-cell user in the adjacent cell to the r-th resource block of the i-th home base station. The specific method for obtaining the corresponding terminal user information from the adjacent macro-base station can refer to the patent CN 102045795A "A Implementation as described in "Methods and Devices for Obtaining Information from a Target Base Station", SINR _Th is a prescribed interference threshold, and x is a constant (in dB) for range fine-tuning;

${\mathrm{<span\; class="notranslate">\; Pow</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{ccc}\mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; if</span>}& {\mathrm{<span\; class="notranslate">\; Pow</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; <</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Pow</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span>\\ \mathrm{<span\; class="notranslate">\; 1</span>}& \mathrm{<span\; class="notranslate">\; if</span>}& <span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Pow</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; Pow</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; Pow</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}\\ \mathrm{<span\; class="notranslate">\; 2</span>}& & {\mathrm{<span\; class="notranslate">\; Pow</span>}}^{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ></span>{\mathrm{<span\; class="notranslate">\; Pow</span>}}_{\mathrm{<span\; class="notranslate">\; Th</span>}}\end{array}\right.$

Pow ⁱ is the actual transmission power of the i-th home base station, Pow _Th is the rated maximum transmission power of the home base station, and y is a constant (in dBm) for range fine-tuning. In actual situations, if the sum of the allocated power exceeds the maximum transmission power , the maximum transmit power is allocated in proportion to the power allocated by Q learning, but the state recorded in the Q value table remains unchanged, and the probability of this situation will gradually tend to 0 with the convergence process.

The quantization precision of I ^{i, r} and Pow ⁱ can be adjusted according to the actual situation.

Action a: the power level a ^{i, r} ∈ {a ₁ , a ₂ , . . . , a _M } that can be allocated to each resource block, the unit is dBm.

The return value re on the rth resource block of the i-th home base station:

${\mathrm{re}}^{i,r}=\begin{array}{}\end{array}\left\{\begin{array}{ccc}-1& \mathrm{if}& {\mathrm{Pow}}^i>{\mathrm{Pow}}_{\mathrm{Th}}\\ e^{{-(\mathrm{SINR}\_I^{i,r}-{\mathrm{SINR}}_{\mathrm{Th}})}^2-e^{(-{\mathrm{SINR}}^{i,r}/K)}}& \mathrm{if}& {\mathrm{Pow}}^i\&le;{\mathrm{Pow}}_{\mathrm{Th}},\end{array}\right.$ Wherein, K is a normal number, and SINR ^{i, r} is the signal-to-interference-noise ratio on the r-th resource block of the i-th home base station.

In addition, all the signal-to-interference and noise ratio (SINR) in this example can also be replaced by channel quality indicator (CQI).

Referring to Fig. 3, the implementation steps of downlink power adjustment based on interactive cognition in the present invention are as follows:

(1) The home base station maintains a three-dimensional Q value table, and initializes the Q value table, the state s of the resource block and the action a. Among them, the first dimension of the Q value table is state s, the second dimension is action a, and the third dimension is resource block r;

(2) The home base station periodically obtains the signal-to-interference-noise ratio (SINR) fed back by the macrocell users in the neighboring cells;

(3) Obtain the quantization value s' of the current state according to the received SINR value and the corresponding state quantization rules, and calculate the return value re,

(4) Use the following rules to update the Q value table:

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; lf</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; lf</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; re</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}{\mathrm{<span\; class="notranslate">\; max</span>}}\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

Among them, γ∈(0,1) is a constant discount factor, which reflects the importance of future returns relative to current returns, and lf∈[0,1) is a constant learning factor, which is used to control the rate of convergence;

(5) For each resource block, use the e-greedy algorithm to select and execute the action a' in this period according to the current Q value table and state s'. The specific method is: e is a small constant between 0-1, generate a random number p between 0-1, if p<e, then randomly select an action; otherwise, select the state s', corresponding to Q The action with the largest value;

(6) Update state and action s=s', a=a', and go to step (2).

In this example, the similarity parameter in the power adjustment based on interactive learning, the professional knowledge degree is defined as follows:

Similarity parameter sim: sim ^i,r ={en ^i,r ,act_en ^i,r }, where en ^i,r is the environmental similarity on resource block r corresponding to home base station i, which is represented by SINR ^i,r here, However, in other embodiments, it is not limited to using SINR ^i,r to represent; act_en ^i,r is the similarity of the impact of actions on the resource block r corresponding to home base station i on the environment, a ^{i, r} are the actions in Q learning on the corresponding resource block r corresponding to base station i, and the subscripts t and t-1 represent the current period and the previous period respectively.

Professional knowledge degree exp: exp ^{i, r} = θ ₁ ·η ⁱ + θ ₂ · con ^{i, r} where θ ₁ and θ ₂ are weights, θ ₁ + θ ₂ = 1; η ⁱ is the value of the i-th home base station energy efficiency, C is the throughput of the home base station, P is the actual transmission power, and w is the bandwidth used; con ^{i, r} is the convergence degree of the Q value table on the resource block r corresponding to the home base station i, q ^i,r is the Q value table on the resource block r corresponding to the home base station i, and the subscripts t and t-1 represent the current cycle and the previous cycle respectively.

Referring to Figures 4 and 5, the implementation steps of downlink power adjustment based on interactive learning in the present invention are as follows:

(1) The home base station periodically combines information such as the Q value table, similarity parameters, and professional knowledge into a case, reports it to the corresponding home base station gateway through the S1 interface, and stores it in the case database. Deleted from the library. Assume that the single-cycle time of power adjustment based on interactive cognition is T, the reporting cycle of the home base station is 1000T, and the survival time of each case is 1200T. When the case inventory is full, delete the cases whose survival time is below the set value (such as 500T) ;

(2) The home base station sends a learning application to the gateway, and reports the similarity parameter sim ₁ and the professional knowledge exp ₁ ; the gateway calculates the similarity δ with each case in the case base according to sim ₁ , and the calculation formula is as follows:

Among them, υ _i is the weight value, |p _i1 -p _i2 | represents the absolute value of the difference between the corresponding similarity parameters of the two cases, N is the number of similarity parameters, and 2 is taken in this example;

(3) Take the case corresponding to the similarity greater than the threshold d _Th as an alternative case. If there is no alternative case, the gateway sends corresponding feedback information to the base station and stops the learning process, otherwise, selects the case with the highest exp value among the alternative cases, sends the corresponding Q value table to the base station, and deletes the case from the alternative cases , the base station uses the Q value table to adjust the power. After 100T, it detects whether the professional knowledge is improved. If it is improved, it will overwrite the original Q value table and send a confirmation signal to the gateway. The gateway releases the alternative case and ends the learning process; if it does not increase Then send out the study application again, the gateway selects the case with the highest exp value among the current alternative cases, and sends the corresponding Q value table to the base station for learning until the alternative cases are empty, if there is still no case that meets the conditions, the gateway sends the base station Send feedback to stop the learning process.

(4) Select the latest reported case c1, and the gateway calculates the similarity between this case and each case in the case database, and the method is the same as that described in step (2). Select the case c2 corresponding to the highest similarity, and compare the professional knowledge of c1 and c2. If |exp ₁ -exp ₂ |≤exp _Th , stop the teaching process; if exp ₁ -exp ₂ >exp _Th , send the Q value table of c1 to the home base station corresponding to c2, and the base station corresponding to c2 uses the Q value table to adjust the power, if the professional knowledge is improved, then send a confirmation signal to the gateway and overwrite the original Q value table, otherwise the original Q value table is still used, and the teaching process ends; if exp ₂ -exp ₁ > exp _Th , then Send the Q value table of c2 to the home base station corresponding to c1, and the base station corresponding to c1 uses the Q value table for power adjustment. If the professional knowledge is improved, it will send a confirmation signal to the gateway and overwrite the original Q value table, otherwise it will still Use the original Q value table to end the teaching process. exp _Th is the threshold of the difference in professional knowledge.

It can be understood that a method for adjusting downlink power based on interactive cognition provided by the present invention includes the following steps:

Step 101: Periodically receive resource block-based information fed back by macrocell users in neighboring cells;

Step 102: Determine the SINR of a single resource block according to the resource block-based information;

Step 103: Update the power information table according to the SINR, and determine the transmit power of the resource block according to the power information table;

Step 104: Allocate the determined transmit power of the resource block to the corresponding HNB.

Preferably, allocating the determined resource block transmit power to the home base station corresponding to the macrocell user in the neighboring cell specifically includes the following steps, sending the determined transmit power of the resource block to the home base station corresponding to the macrocell user in the neighboring cell; wherein, the The power information table includes at least a state dimension parameter s, an action dimension parameter a, and a resource block dimension parameter r, wherein the state dimension parameter s is: a set based on the signal-to-interference-noise ratio of the home base station user resource block and the corresponding transmit power of the home base station user ; The action dimension parameter a is the power level that can be allocated to each resource block.

In addition, in the embodiment of the present invention, updating the power information table according to the SINR specifically includes the following steps:

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; lf</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; lf</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; re</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}\underset{{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}}{\mathrm{<span\; class="notranslate">\; max</span>}}\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; a</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

Among them, Q(s, a, r) is the current power information table configuration rule based on the state dimension parameter s, the action dimension parameter a, and the resource block dimension parameter r; lf is a constant learning factor used to control the convergence rate, lf∈ [0,1); γ is a constant discount factor, γ∈(0, 1); s' is the quantized value of the previous state dimension parameter, and a' is the quantized value of the previous action dimension parameter. It can be understood that the update power information table in this embodiment of the present invention may also include other update rules, such as using parameters in different ranges, or adding other factors.

A method for adjusting downlink power based on interactive learning provided by the present invention comprises the following steps:

Step 201: the femtocell combines relevant information into a case, stores it in the case database, and automatically deletes the case after the survival time exceeds;

Step 202: the home base station sends a learning application to the gateway, and after reporting the relevant information; the case base calculates its similarity with the cases in the case base;

Step 203: Select a suitable case according to the degree of similarity and professional knowledge, and send the corresponding form of the case to the home base station that issued the learning application. changes decide whether to continue the learning process;

Step 204: For the latest reported case, the femtocell gateway selects similar cases to compare the professional knowledge, if the difference is within the threshold, stop the teaching process; otherwise, send the form of the case with higher professional knowledge to the one with lower professional knowledge The home base station corresponding to that case performs power adjustment based on interactive cognition. If the professional knowledge is improved after execution, it will send a confirmation signal to the gateway and overwrite the original form; otherwise, keep the original form and end the teaching process.

Through the above solution, in the embodiment of the present invention, the smallest time-frequency unit that can be allocated to a user is a resource block, and further, the power adjustment based on the resource block enables interactive learning between the home base stations, improving the performance of the home base station itself.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented through computer program instructions and other related hardware. The program can be stored in a computer-readable storage medium. When the program is executed , may include the processes of the embodiments of the above-mentioned methods. Wherein, the storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM) or a random access memory (Random Access Memory, RAM), etc.

It can be seen that the present invention provides a downlink power adjustment method and device, and provides a variety of optional adaptation solutions. The above embodiments do not limit the technical solutions described in the present invention. The present invention has been described, and those skilled in the art should understand that all technical solutions and improvements that do not deviate from the spirit and scope of the present invention should be included in the claims of the present invention.

Claims (10)

1. the descending power method of adjustment based on interactive cognitive learning, is characterized in that comprising following process:

Process 1. is based on interactive cognitive descending power adjustment;

The descending power adjustment of process 2. based on interactive learning.

2. the descending power method of adjustment based on interactive cognitive learning as claimed in claim 1, is characterized in that in process 1, described based on interactive cognitive descending power adjustment, comprising:

(1) each Home eNodeB is safeguarded a descending power information table, and described descending power information table be take Resource Block as least unit, for determining the transmitting power of each Resource Block, all data of descending power information table is carried out to initialization;

(2) disturbed condition that the configuration of periodically perception current transmit power of Home eNodeB causes, then upgrades corresponding data in descending power information table according to the information of perception and corresponding update rule;

(3) distribution method of the descending power information table after use renewal and regulation determines the transmitting power configuration in next cycle;

(4) repeat step (2) and (3), final goal is to make each transmitting power of distributing according to perception information best.

3. the descending power method of adjustment based on interactive cognitive learning as claimed in claim 1, is characterized in that in process 2, and the described descending power adjustment based on interactive learning, comprising:

(1) Home eNodeB periodically reports the terminal equipment with aggregation feature using the power information table of self maintained, similarity parameter, professional knowledge degree as a case, described terminal equipment adopts femto gateway, femto gateway is a life span of each case setting, case surpasses after life span, automatically deletes;

(2) Home eNodeB carries out Active Learning, and femto gateway carries out active professor, and two kinds of modes are carried out simultaneously, for descending power is adjusted.

4. the descending power method of adjustment based on interactive cognitive learning as claimed in claim 3, it is characterized in that in step (2), described Active Learning refers to: the Home eNodeB that is less than threshold value for the opening time, initiatively to femto gateway, send study application, and report similarity parameter and professional knowledge degree; Gateway is according to the similarity of each case in the similarity calculation of parameter reporting and case library, case using similarity more than threshold value is as alternative case, the professional knowledge degree of more alternative case again, get soprano as selecting eventually case, the corresponding form of this case is sent to the Home eNodeB that sends study application, and this case is being deleted from case library, base station is used this form data to carry out based on after interactive cognitive power adjustment, if professional knowledge degree improves, to gateway, send confirmation signal and cover original form, finishing learning process; Otherwise gateway selects the case that in current alternative case, professional knowledge degree is the highest to select case as whole, sends to the Home eNodeB that sends study application to learn, until alternative case is empty, gateway sends feedback information to base station, stops learning process.

5. the descending power method of adjustment based on interactive cognitive learning as claimed in claim 3, it is characterized in that in step (2), described active professor refers to: for the up-to-date case reporting, femto gateway calculates the similarity of each case in this case and case library, select the highest case of similarity, relatively report case and the professional knowledge degree of selecting case; If differ in threshold value, stop professor's process; Otherwise the form of the larger case of professional knowledge degree is sent to Home eNodeB corresponding to that case that professional knowledge degree is less, to carry out based on interactive cognitive power adjustment, if professional knowledge degree improves after carrying out, to gateway, send confirmation signal and cover original form, finishing professor's process; Otherwise, retain original form, finish professor's process, this process does not need all cases that newly report to carry out.

6. the descending power method of adjustment based on interactive cognitive learning as claimed in claim 1, the descending power that it is characterized in that described interactive cognitive descending power adjustment and interactive learning is adjusted at circulation in practical operation and carries out.

7. the descending power adjusting device based on interactive cognitive learning, is characterized in that comprising three modular units in base station side: information storage module, information sending/receiving module, message processing module.

8. the descending power adjusting device based on interactive cognitive learning as claimed in claim 7, is characterized in that:

The major function of described information storage module is: the power information table that 1) storage is safeguarded based on interactive cognitive power adjustment, and message processing module will be used power information table, the information storage module of restoring after processing; 2) temporarily store the case information obtaining in the power adjustment based on interactive learning the form that determines current use;

The major function of described information sending/receiving module is: 1) periodically receive the various information by adjacent area macrocellular user feedback, and reception value is reported to message processing module, for interfere information, if do not exist adjacent area macrocellular user's interference or interference to be less than predetermined threshold value, can ignore, the regulation value of reporting is 0 artificially; If there is a plurality of adjacent area macrocellular users' interference, therefrom choose and disturb maximum conduct value of reporting, final process result is met all macrocellular users' interference is down to below threshold value; 2) be responsible for sending self case and other relevant information to femto gateway;

The major function of described message processing module is: 1) in adjusting based on interactive cognitive power,, with processing storage information and reporting information periodically, specifically refer to updating form lattice information and determine Resource Block transmitting power with information receiving module; 2), in the power based on interactive learning is adjusted, self maintained form, similarity parameter, professional knowledge degree and other information needed are formed to case and give information sending/receiving module.

9. a descending power method of adjustment, is characterized in that, comprises the following steps:

(1) periodically receive by adjacent area macrocellular user feedback based on Resource Block information;

(2) according to the described Signal to Interference plus Noise Ratio of determining single Resource Block based on Resource Block information;

(3) according to described Signal to Interference plus Noise Ratio, upgrade power information table, and according to power information table, determine the transmitting power of Resource Block;

(4) by definite Resource Block transmit power allocations, give corresponding Home eNodeB user.

10. method as claimed in claim 9, is characterized in that: by definite Resource Block transmit power allocations, to adjacent area macrocellular user, concrete steps are as follows: definite Resource Block transmitting power is sent to adjacent area macrocellular user; Wherein, described power information table at least comprises state dimension parameter s, action dimension parameter a, Resource Block dimension parameter r, wherein, state dimension parameter s is: the set of the corresponding transmitting power of the Signal to Interference plus Noise Ratio based on Home eNodeB user resources piece and this Home eNodeB user; A is for distributing to the power grade of each Resource Block for action dimension parameter;

The concrete steps of upgrading power information table according to described Signal to Interference plus Noise Ratio are as follows:

Wherein, Q (s, a, r) is the current power information table configuration rule based on state dimension parameter s, action dimension parameter a, Resource Block dimension parameter r; Lf is the constant study factor, for controlling the speed of convergence, lf ∈ [0,1); γ is constant discount factor, γ ∈ (0,1); S' is the quantized value of last next state dimension parameter, and a ' is the quantized value of front one-off dimension parameter.