Cellular-free extensible distributed scheduling and resource allocation architecture and allocation method

Abstract

The invention discloses a cellular-free extensible distributed scheduling and resource allocation architecture and an allocation method. The cellular-free extensible distributed scheduling and resource allocation architecture comprises a first unit, a second unit, a third unit, a fourth unit and K pieces of UE (User Equipment), the first unit is used for resource coordination and centralized scheduling tuning, the second unit is used for realizing distributed scheduling and resource allocation, the third unit is used for time-frequency decoupling and realizing distributed precoding and receiving, and the fourth unit comprises a radio frequency remote unit RRU and an antenna unit and is used for receiving a plurality of uplink data streams sent by terminal UE. And sending the plurality of uplink data streams to the third unit, receiving a pre-coded data stream sent by the third unit, and sending the pre-coded data stream to the corresponding UE. According to the invention, the calculation load of the central processing unit can be reduced, and novel honeycomb-removed multi-layer extensible distributed scheduling and resource allocation are realized.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a cellular-free scalable distributed scheduling and resource allocation architecture and allocation method. Background Art

[0002] In modern wireless communication systems, resource allocation and scheduling are key technologies to improve the overall performance of the system. Wireless resource allocation mainly involves maximizing system throughput, reducing interference and ensuring service quality by allocating limited time-frequency, power and space resources. In cellular-free massive multiple-input multiple-output (CF-mMIMO), a large number of distributed access points collaborate to serve users through a central processing unit, effectively solving the cell edge problem of traditional cellular networks. In actual communication environments, usually only some radio frequency remote pull units can effectively serve specific terminals. The fully connected CF-mMIMO scenario is inefficient and difficult to expand, and the large dimension of the channel matrix significantly increases the computational complexity. To this end, the researchers proposed a user-centric scalable cellular-free system, defining an exclusive service RRU cluster for each user, so that the user is located at the center of the cluster, providing accurate and stable services and achieving network scalability.

[0003] However, existing resource allocation research generally relies on centralized allocation methods, that is, coordinating the transmission parameters of all RRUs based on global channel information. Although centralized methods can approach the theoretical performance limit, they have problems such as high signaling overhead, computational delay, and insufficient scalability. In addition, since the optimization problem is essentially a non-convex NP-hard problem, it is difficult to guarantee global optimality even with approximate means. Therefore, designing a low-complexity, scalable, and efficient and performance-balancing distributed resource allocation solution has become a key direction to break through the bottleneck.

[0004] Therefore, considering the backhaul link problem and the deployment of distributed and centralized units, it is necessary to combine a high-performance scalable distributed baseband signal processing architecture, design a new scalable multi-layer distributed scheduling and resource allocation framework, reduce the computing load of the central processing unit, and adopt efficient optimization strategies to improve system performance and approach a fully centralized architecture. Summary of the invention

[0005] Purpose of the invention: The present invention provides a cellular-free scalable distributed scheduling and resource allocation architecture and allocation method, which can reduce the computing load of the central processing unit and realize a new type of cellular-free multi-layer scalable distributed scheduling and resource allocation.

[0006] Technical solution: An infrastructure-less scalable distributed scheduling and resource allocation architecture according to the present invention includes: a first unit, a second unit, a third unit, a fourth unit, and K UEs; the first unit is used for resource coordination and centralized scheduling optimization, the second unit is used to implement distributed scheduling and resource allocation, the third unit is used for time-frequency decoupling to implement distributed precoding and receivers, and the fourth unit includes a radio remote unit (RRU) and an antenna unit, which is used to receive multiple uplink data streams sent by a terminal UE, send the multiple uplink data streams to the third unit, and receive the precoded data stream sent by the third unit and send the precoded data stream to the corresponding UE.

[0007] Furthermore, each UE can be associated with multiple third units, but can only be associated with one second unit and is scheduled by the only second unit, while one third unit can be associated with multiple second units. The second unit can perform multi-user scheduling and radio resource allocation according to real-time requirements. By deploying a large number of second units, the computing load of the central processing unit can be significantly reduced, a distributed scheduler can be implemented, the response speed and scalability of the system can be improved, and comprehensive "infrastructure-less" from the physical layer to the upper layer can be ensured.

[0008] Correspondingly, an infrastructure-less scalable distributed scheduling and resource allocation method includes the following steps:

[0009] Step 1, pre-define the association deployment scheme between the second unit and the third unit;

[0010] Step 2, based on the association deployment structure between the second unit and the third unit, the system determines the second unit to which each UE belongs; since each UE can only be scheduled by one second unit, the association between the UE and the second unit is equivalent to allocating each UE to a unique third unit group and can only be served by V third units in the group; at the same time, considering the limited processing capacity of each second unit, the number of UEs scheduled by each second unit does not exceed the upper limit Z.

[0011] Step 3, perform distributed pilot allocation at each second unit;

[0012] Step 4, to meet the scalability requirements, further perform distributed RRU-UE association at each second unit;

[0013] Step 5, to reduce the interference introduced by distributed scheduling, after obtaining the local results of the two tasks, send the pilot allocation result and the association situation to the first unit, and the first unit first fine-tunes the pilot allocation result according to the global information;

[0014] Step 6, the first unit further fine-tunes the RRU-UE association result according to the global information;

[0015] Step 7: Perform distributed power control at each second unit.

[0016] Furthermore, in Step 1, let the numbers of the M third units be {1, …, M}, and the numbers of the J second units be {1, …, J}. Assume M ≥ J. Each second unit schedules and manages the same number V of third units, where V < M. To ensure the balance of network load and fronthaul link; when the numbers of third units and second units are equal, i.e., M = J, since each second unit schedules a specific number V of third units, it is regarded as dividing the third units into J groups, with each group containing V third units. Since each UE can only be scheduled by one second unit, the subsequent association of UEs with second units is equivalent to roughly allocating each UE to a unique third unit group and being served only by the V third units in the group; when the numbers of third units and second units are not equal, then the M third units need to be classified into J third unit clusters first. Third units with fewer RRU mounts are preferentially merged into one third unit cluster. At this time, the numbers of third unit clusters and second units are equal, and then it is converted into the aforementioned topological association structure. Ensure that the RRU numbers of each third unit cluster are as balanced as possible.

[0017] Furthermore, in Step 2, the Intra-Group Asynchronous Sum of Large-Scale Fading Coefficient (IGA-SLSFC) is used to measure the quality of service between UEk and the third unit group scheduled by the second unit j in a wide-area asynchronous scenario, expressed as where β k,l,m represents the large-scale fading coefficient between UE k and the l-th RRU mounted on the third unit m, represents the set of third units scheduled by the second unit j, represents the set of RRUs mounted on the third unit m, and Δt k,l,m represents the asynchronous timing offset. This coefficient can be further extended to an asynchronous orthogonal frequency division multiplexing scenario, and the gain attenuation coefficient can be further optimized according to the degree to which the asynchronous timing offset exceeds the cyclic prefix range. Based on a competition mechanism algorithm, attempt to maximize the sum of IGA-SLSFC for all UEs, i.e., where represents the set of UEs scheduled by the second unit j, and the specific steps are as follows:

[0018] Step 21: Candidate UE set The set of UEs scheduled by each second unit The set of non-serving second units for each UE Calculate UE IGA-SLSFC of the third unit group corresponding to all the second units;

[0019] Step 22, UE k selects the second unit with the maximum IGA-SLSFC Schedule. At this time, if Then competition is required and it jumps to Step 23; otherwise, UE k will be directly placed into the UE set scheduled by the second unit j, that is in, that is And UE k is removed from the set of candidate UEs, that is And then it jumps to Step 24;

[0020] Step 23, when UE k needs to select its affiliated second unit through competition. The second unit j finds the UE with the minimum IGA-SLSFC in the UE set it schedules and including UE k, that is If k * = k, then the competition fails and the second unit j is included in the set of non-serving second units of UE k in, and it goes back to Step 22; otherwise, the competition is successful and the k in * is replaced with k, and UE k is removed from the set of candidate UEs In addition, UE k * is re-included in the set of candidate UEs, that is The second unit j is included in the set of non-serving second units of UE k * in; in;

[0021] Step 24, go back to Step 21 until

[0022] Furthermore, in Step 3, at each second unit, each UE it schedules first finds the RRU with the largest large-scale coefficient among its servable RRUs, that is

[0023]

[0024] This RRU is the primary RRU of this UE. Among them, j k represents the number of the second unit scheduling UE k. Then, according to

[0025]

[0026] The pilot τ is allocated to this UE, that is i k = τ. Among them, i k represents the pilot index used by UE k, represents the second unit j kThe set of UEs using pilot t among the scheduled UEs. For each UE in the above formula, the pilot with the least observable pilot contamination in its corresponding primary RRU is allocated. At this time, a local rough pilot allocation result can be obtained. Since distributed scheduling is adopted, each second unit cannot observe the global pilot contamination and can only obtain the contamination caused by the pilots shared by the UEs it schedules. Limited by this, distributed pilot allocation will inevitably generate additional interference.

[0027] Furthermore, in step 4, a distributed maximum large-scale fading association algorithm is proposed. Each second unit takes the UE as the center, and each UE it schedules is successively associated with the RRU with the largest large-scale fading coefficient that it can serve until the following formula is satisfied. At this time, a local rough RRU-UE association result is obtained;

[0028]

[0029] where the set of RRUs serving UE k in the third unit m is δ is the proportional threshold of the total large-scale coefficient. Since the distributed association scheme only pairs RRUs and UEs according to local large-scale fading coefficients, additional interference will still be introduced due to the lack of global information.

[0030] Furthermore, in step 5, a pilot allocation adjustment scheme based on the Dis coefficient is proposed. First, calculate the Dis coefficient that measures the pilot contamination when two UEs share the same pilot

[0031]

[0032] where d k =[d k,1 ,…,d k,L T and D :,k ∈{0,1} L×1 respectively represent the distances and serving relationships from UE k to all RRUs. χ k,k' represents the sum of the distance differences between two UEs to each RRU serving them. The smaller its value, the stronger the pilot contamination generated when UE k and UE k' share the same pilot. However, the absolute value of this parameter is related to the distance and association situation and will change with the scenario. Therefore, the absolute value has little reference significance, and UEs pairs that need to be adjusted should be selected according to the relative value;

[0033] At this time, calculate the z-score for each χ k,k' . The z-score is used to evaluate the distance of a sample point from the overall mean and measure how many standard deviations the original data differs from the data mean, expressed as where, and ​Indicates the mean and standard deviation of the Dis coefficient. For UE pairs, if the UE pairs share pilots, pilot reassignment is required, and the threshold is adjusted It can flexibly control the trade-off between complexity and spectral efficiency (SE) gain. The larger it is, the more UE pairs need to globally readjust the pilot assignment, and the greater the SE gain, and correspondingly, higher computing resources will be consumed.

[0034] Take out ζ in sequence k,k' The smallest UE pair that needs to be adjusted. At this time, this UE pair shares the same pilot τ, that is, i k = i k' = τ; Calculate the global pilot contamination coefficient of each pilot of the RRU with the largest large-scale fading coefficient associated with UE k and UE k' respectively. Taking UE k as an example, that is

[0035]

[0036] Among them, Represents the set of UEs globally using pilot t, different from τ k Represents the global optimal pilot of UE k at this time. Measures the pilot contamination intensity generated by the main RRU of UE k at pilot τ k ;

[0037] Compare and to select the UE with a smaller pilot contamination intensity for reassignment, that is, if then reassign pilot i to UE k k = τ k , otherwise, reassign pilot i to UE k' k' = τ k' .

[0038] Furthermore, in step 6, a RRU-UE association adjustment scheme based on the interference-aware reward (IAR) coefficient is proposed. The IAR coefficient is used to centrally optimize the RRU-UE association result. The IAR coefficient is calculated only based on the large-scale fading coefficient and can comprehensively reflect the received signal strength, interference between UEs, and pilot reuse interference.

[0039] First, according to the following formula, calculate the IAR coefficient based on the large-scale information, pilot assignment result, and the current RRU-UE association situation.

[0040]

[0041] Among them, Indicates the set of UEs sharing pilot i with UE k k of UEs, represents the set of interfering RRU for UE k, where, represents the set of RRUs associated with UE k;

[0042] Subsequently, the RRU-UE association is adjusted according to the IAR coefficient. The specific adjustment scheme is as follows: If the IAR of UE k and RRU(l,m) is higher than the IAR of the currently associated RRU, that is, it means that a better-performing RRU has been missed in the distributed RRU-UE association process, then it is further associated with this UE, that is,

[0043] Furthermore, in step 7, when considering the distributed architecture, the fractional power control of UE k calculated at the second unit is as shown in the following formula

[0044]

[0045] where, represents the maximum uplink transmission power of each UE, κ∈[0,1] is the uplink power scaling factor. When κ = 0, it means that all UEs transmit at the maximum power. The larger κ is, the higher the fairness among UEs. The difference between this formula and the conventional centralized power control is that the minimum value in the numerator of the fraction is calculated only from the UEs that can be scheduled by this second unit, rather than selecting the minimum value from all UEs globally. Its complexity does not increase with the growth of the number of UEs K, ensuring scalability.

[0046] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: In this architecture, the second unit can perform multi-user scheduling and wireless resource allocation according to real-time requirements. By deploying a large number of second units, the computing load of the central processing unit can be significantly reduced, a distributed scheduler can be implemented, and the response speed and scalability of the system can be improved; each third unit can be associated with multiple second units, ensuring a comprehensive "de-cell" from the physical layer to the upper layer; adopting the concept of distributed computing and overall local optimization, the present invention can effectively reduce the computing load of the central processing unit, ensure better system performance, and achieve a new type of de-cell multi-layer scalable distributed scheduling and resource allocation. Brief Description of the Drawings

[0047] Figure 1 is a schematic diagram of the architecture of the present invention.

[0048] Figure 2 is a schematic diagram of the scheduling and resource allocation process and execution location of the present invention.

[0049] Figure 3 is a schematic diagram of the predefined association topology structure of the second unit and the third unit of the present invention.

[0050] Figure 4 This is a schematic diagram for merging the third unit clusters in the case of the associated deployment of the second unit and the third unit in the second case of the present invention.

[0051] Figure 5 This is the result graph of the uplink users and spectral efficiency of the present invention. Detailed implementation manners

[0052] As Figure 1 and Figure 2 shown, a cell-free scalable distributed scheduling and resource allocation architecture includes: a first unit for resource coordination and centralized scheduling algorithm tuning; a second unit for implementing distributed scheduling and resource allocation; a third unit mainly responsible for spatio-temporal-frequency decoupling and implementing distributed precoding and receivers; a fourth unit including a radio remote unit (RRU) and an antenna unit for receiving multiple uplink data streams sent by a terminal (UE), sending the multiple uplink data streams to the third unit, and / or receiving precoded data streams sent by the third unit and sending the precoded data streams to the corresponding UE.

[0053] In this architecture, the second unit can perform multi-user scheduling and wireless resource allocation according to real-time requirements. By deploying a large number of second units, the computing load of the central processing unit can be significantly reduced, a distributed scheduler can be implemented, and the response speed and scalability of the system can be improved. Specifically, each UE can be associated with multiple third units, but can only be associated with one second unit and is scheduled by the only one second unit. And one third unit can be associated with multiple second units to ensure comprehensive "de-cellulization" from the physical layer to the upper layer.

[0054] Predefine the associated deployment scheme of the second unit and the third unit. Let the numbers of M third units be {1,..., M}, and the numbers of J second units be {1,..., J}. Usually, it is assumed that M≥J. To ensure the balance of network load and fronthaul links, each second unit schedules and manages the same number V of third units, where V < M.

[0055] Case 1: When the numbers of the third unit and the second unit are equal, i.e., M = J. Since each second unit schedules a specific number V of third units, it can be regarded as dividing the third units into J groups, and each group contains V third units. At this time, we predefine the associated topological structure of the second unit and the third unit as Figure 3 shown:

[0056] Taking the Figure 3 (a) scenario as an example, there are 5 third units and 5 second units, and each second unit schedules 3 third units. Then, a total of 5 third unit groups are divided, and the indices of the third units within each group are: and Each second unit will schedule a group of third units separately. Similarly, if each second unit can schedule 2 third units, the topological association structure is as shown in Figure 3 (b), which is divided into a total of 5 groups of third units. The indexes of the third units in each group are: and

[0057] Since each UE can only be scheduled by one second unit, the subsequent UE-second unit association is equivalent to roughly assigning each UE to a unique group of third units and can only be served by V third units in the group.

[0058] Case 2: When the number of third units and second units is not equal, it is necessary to first classify M third units into J clusters of third units. Third units with fewer RRU mounts are preferentially merged into one cluster of third units to ensure that the number of RRUs in each cluster of third units is as balanced as possible, as shown in Figure 4 . At this time, the number of clusters of third units is equal to the number of second units, and then it can be converted into the topological association structure of Case 1.

[0059] Based on the above second unit-third unit association deployment structure, the system determines the second unit to which each UE belongs; since each UE can only be scheduled by one second unit, the UE-second unit association is equivalent to assigning each UE to a unique group of third units and can only be served by V third units in the group. At the same time, considering the limited processing capacity of each second unit, the number of UEs scheduled by each second unit does not exceed the upper limit Z. First, define an index: Intra-Group Asynchronous Sum of Large-Scale Fading Coefficient (IGA-SLSFC) to measure the quality of service between UE k and the group of third units scheduled by second unit j in a wide-area asynchronous scenario, expressed as where β k,l,m represents the large-scale fading coefficient between UE k and the l-th RRU mounted on third unit m, represents the set of third units scheduled by second unit j, represents the set of RRUs mounted on third unit m. Δt k,l,m represents the asynchronous timing offset. This coefficient can be further extended to an asynchronous orthogonal frequency division multiplexing scenario, and the design gain attenuation coefficient can be further optimized according to the degree to which the asynchronous timing offset exceeds the cyclic prefix range.

[0060] The proposed algorithm based on the competition mechanism attempts to maximize the sum of IGA-SLSFC of all UEs, that is where represents the set of UEs scheduled by the second unit j. The specific algorithm process is as follows:

[0061] Algorithm initialization: The set of candidate UEs The set of UEs scheduled by each second unit The set of non-serving second units of each UE

[0062] Step 1: Calculate The IGA-SLSFC of the third unit group corresponding to all second units.

[0063] Step 2: UE k selects the second unit with the maximum IGA-SLSFC for scheduling. At this time, if then competition is required, and it jumps to Step 3; otherwise, UE k will be directly put into the set of UEs scheduled by the second unit j, that is and UE k is removed from the set of candidate UEs, that is and then it jumps to Step 4.

[0064] Step 3: When UE k needs to select the second unit it belongs to through competition. Specifically, the second unit j finds the UE with the minimum IGA-SLSFC in the set of UEs it schedules and including UE k, that is If k * = k, the competition fails, and the second unit j is included in the set of non-serving second units of UE k and it goes back to Step 2; otherwise, the competition succeeds, and the k in * is replaced by k, and UE k is removed from the set of candidate UEs In addition, UE k * is re-included in the set of candidate UEs, that is the second unit j is included in the set of non-serving second units of UE k * in.

[0065]

[0065] Step 4: Go back to Step 1 until

[0066] Next, perform distributed pilot allocation at each second unit. At each second unit, each UE it schedules first finds the RRU with the largest large-scale coefficient among its servable RRUs, that is

[0067]

[0068] This RRU is the primary RRU for this UE, where j k represents the second unit number for scheduling UE k, and then based on

[0069]

[0070] allocate the pilot τ to this UE, that is, i k = τ, where i k represents the pilot index used by UE k, represents the second unit j k The set of UEs that use pilot t among the UEs scheduled by j. Each of the above UEs is allocated the pilot with the least pilot pollution observable in its corresponding primary RRU. At this time, the local pilot rough allocation result can be obtained. Due to the use of distributed scheduling, each second unit cannot observe the global pilot pollution and can only obtain the pollution generated by the pilots shared by the UEs it schedules. Limited by this, the distributed pilot allocation will inevitably generate additional interference.

[0071] To meet the scalability requirements, a distributed RRU-UE association is further performed at each second unit. A distributed maximum large-scale fading association algorithm is proposed. Each second unit takes the UE as the center, and each UE it schedules is successively associated with the RRU with the largest large-scale fading coefficient that it can serve until the following formula is satisfied. At this time, the local RRU-UE rough association result can be obtained.

[0072]

[0073] Among them, the set of RRUs serving UE k in the third unit m is δ is the proportional threshold of the total large-scale coefficient. Since the distributed association scheme only pairs the RRU-UE according to the local large-scale fading coefficient, additional interference will still be introduced due to the lack of global information.

[0074] To reduce the interference introduced by distributed scheduling, after obtaining the local results of the two tasks, the pilot allocation result and the association situation are sent to the first unit. The first unit first fine-tunes the pilot allocation result according to the global information. A pilot allocation adjustment scheme based on the Dis coefficient is proposed. First, calculate the Dis coefficient that measures the pilot pollution when two UEs share the same pilot

[0075]

[0076] Among them, d k = [d k,1 , …, d k,L ​T and D :,k ∈ {0, 1} L×1 respectively represent the distances and serving relationships from UE k to all RRUs, and χ k,k' represents the sum of the distance differences between two UEs to each RRU serving them. The smaller its value, the stronger the pilot contamination will be when UE k and UE k' share the same pilot. However, the absolute value of this parameter is related to distances and association situations and will change with the scenario. Therefore, the absolute value has little reference significance, and instead, according to the magnitude of its relative value, the UE pairs that need to be adjusted are selected.

[0077] At this time, for each χ k,k' its z - score is calculated. The z - score is used to evaluate the distance of a sample point from the population mean and measure how many standard deviations the original data deviates from the data mean, expressed as where and represent the mean and standard deviation of the Dis coefficient. For UE pairs, if this UE pair shares a pilot, then pilot re - allocation is required. Adjusting the threshold can flexibly control the trade - off between complexity and spectral efficiency (SE) gain. The larger

[0078] is, the more UE pairs will need to globally re - adjust pilot allocation, the greater the SE gain, and correspondingly, higher computing resources will be consumed. k,k' The specific adjustment scheme is as follows: Sequentially take out the UE pair with the smallest ζ k that needs to be adjusted. This UE pair shares the same pilot τ at this time, that is, i k' = i

[0079] Step 1: Calculate the global pilot contamination coefficient of each pilot of the RRU with the largest large - scale fading coefficient associated with UE k and UE k' respectively. Taking UE k as an example, that is

[0080]

[0081] where represents the set of UEs globally using pilot t, different from τ k which represents the global optimal pilot of UE k at this time, measures the pilot contamination intensity generated by the main RRU of UE k in pilot τ k

[0082] Step 2: Compare and ​Select a UE with a smaller pilot contamination intensity for reallocation according to its size, that is, if Then reallocate pilot i to UE k k = τ k Otherwise, reallocate pilot i to UE k' k' = τ k' .

[0083] The first unit further fine-tunes the RRU-UE association result according to the global information. A RRU-UE association adjustment scheme based on the interference-aware reward (IAR) coefficient is proposed, and the IAR coefficient is used to centrally optimize the RRU-UE association result. The IAR coefficient is calculated only based on the large-scale fading coefficient and can comprehensively reflect the received signal strength, interference between UEs, and pilot reuse interference.

[0084] First, according to the following formula, calculate the IAR coefficient based on the large-scale information, pilot allocation result, and current RRU-UE association situation.

[0085]

[0086] Among them, represents the set of UEs sharing pilot i with UE k k , represents the set of interfering RRUs of UE k, where represents the set of RRUs associated with UE k.

[0087] Subsequently, adjust the RRU-UE association situation according to the IAR coefficient. The specific adjustment scheme is as follows: If the IAR of UE k and RRU(l,m) is higher than the IAR of the currently associated RRU, that is it means that a RRU with better performance is missed in the distributed RRU-UE association process, then further associate it with this UE, that is

[0088] Execute distributed power control at each second unit. When considering the distributed architecture, the fractional power control of UE k calculated at the second unit is shown as the following formula

[0089]

[0090] Among them, It represents the maximum uplink transmission power of each UE. κ ∈ [0, 1] is the uplink power scaling factor. When κ = 0, it means that all UEs transmit at the maximum power. The larger κ is, the higher the fairness among UEs. The difference between this formula and the conventional centralized power control is that the minimum value in the numerator of the fraction is calculated only from the UEs scheduled by this second unit, rather than selecting the minimum value from the global UEs. Its complexity does not increase with the number K of UEs, ensuring scalability.

[0091] An example of the effect is as Figure 5 shown in the uplink user and spectral efficiency result graph of the present invention. Among them, when the number of UEs is 40 and 80, four different schemes of dynamic cooperative clustering, full association of UEs with the associated third unit, distributed scheduling, and centralized tuning are compared. Figure 5 It shows that under the new cell-free scalable distributed scheduling architecture, the traditional dynamic cooperative clustering scheme is not ideal. After adopting distributed scheduling, higher performance gains can be obtained, and after centralized tuning, further performance improvement can be obtained. It fully reflects the superiority of the new cell-free multi-layer scalable distributed scheduling and resource allocation technology proposed by the present invention.

Claims

1. A cell-free scalable distributed scheduling and resource allocation architecture, characterized in that, Comprising: A first unit, a second unit, a third unit, a fourth unit, and K UEs; the first unit is used for resource coordination and centralized scheduling optimization, the second unit is used to implement distributed scheduling and resource allocation, the third unit is used for time-frequency decoupling to implement distributed precoding and receivers, and the fourth unit includes a radio remote unit (RRU) and an antenna unit, which are used to receive multiple uplink data streams sent by the terminal UE, send the multiple uplink data streams to the third unit, and receive the precoded data stream sent by the third unit and send the precoded data stream to the corresponding UE.

2. The cell-free scalable distributed scheduling and resource allocation architecture according to claim 1, characterized in that Each UE is associated with multiple third units, but can only be associated with one second unit and is scheduled by the only one second unit, while one third unit can be associated with multiple second units.

3. A distribution method applied to the cell-free scalable distributed scheduling and resource allocation architecture as described in claim 1, characterized in that, Including the following steps: Step 1: Predetermine the associated deployment scheme of the second unit and the third unit; Step 2: Based on the associated deployment structure of the second unit and the third unit, the system determines the second unit to which each UE belongs; since each UE can only be scheduled by one second unit, the association between the UE and the second unit is equivalent to allocating each UE to a unique group of third units and can only be served by V third units in the group; at the same time, considering the limited processing capacity of each second unit, the number of UEs scheduled by each second unit does not exceed the upper limit Z. Step 3: Perform distributed pilot allocation at each second unit; Step 4: To meet the requirement of scalability, further perform distributed RRU-UE association at each second unit; Step 5: To reduce the interference introduced by distributed scheduling, after obtaining the local results of the two tasks, send the pilot allocation result and the association situation to the first unit, and the first unit first fine-tunes the pilot allocation result according to the global information; Step 6: The first unit further fine-tunes the RRU-UE association result according to the global information; Step 7: Perform distributed power control at each second unit.

4. The cell-free scalable distributed scheduling and resource allocation method according to claim 3, wherein In Step 1, let the numbers of M third units be {1, …, M}, and the numbers of J second units be {1, …, J}. Assume M ≥ J. Each second unit schedules and manages the same number V of third units, where V < M; when the numbers of the third unit and the second unit are equal, i.e., M = J, since each second unit schedules a specific number V of third units, it is regarded as dividing the third units into J groups, and each group contains V third units. Since each UE can only be scheduled by one second unit, the subsequent association between the UE and the second unit is equivalent to roughly allocating each UE to a unique group of third units and can only be served by V third units in the group; when the numbers of the third unit and the second unit are not equal, it is necessary to first classify the M third units into J third unit clusters, and preferentially merge the third units with fewer RRUs mounted into one third unit cluster. At this time, the numbers of the third unit clusters and the second units are equal, and then it is converted into the aforementioned topological association structure.

5. The cell-free scalable distributed scheduling and resource allocation method according to claim 3, characterized in that In step 2, the intra-group asynchronous sum large-scale fading coefficient IGA-SLSFC is used to measure the quality of service between UE k and the set of third cells scheduled by the second cell j in a wide-area asynchronous scenario, expressed as where β k,l,m represents the large-scale fading coefficient between UE k and the l-th RRU mounted on the third cell m, represents the set of third cells scheduled by the second cell j, represents the set of RRUs mounted on the third cell m, and Δt k,l,m represents the asynchronous timing offset. Based on the competitive mechanism algorithm, an attempt is made to maximize the sum IGA-SLSFC of all UEs, that is where represents the set of UEs scheduled by the second cell j, and specifically includes the following steps: Step 21, Candidate UE Set UE Set Scheduled by Each Second Unit Set of Non-Serving Second Units of Each UE Calculate UE IGA-SLSFC of the Third Unit Group Corresponding to All Second Units; Step 22: UE k selects the second unit with the maximum IGA - SLSFC For scheduling, if at this time a competition is required, then jump to Step 23; otherwise, UE k will be directly placed into the UE set scheduled by the second unit j, that is and UE k will be removed from the set of candidate UEs, that is and then jump to Step 24; ​ Step 23. When UE k needs to select its affiliated second unit through competition. The second unit j finds the UE with the smallest IGA-SLSFC among the UEs it schedules and including UE k, that is If k * = k, the competition fails, and the second unit j is incorporated into the non-serving second unit set of UE k and go back to Step 22; otherwise, the competition succeeds, and the k in * is replaced by k, and UE k is removed from the set of candidate UEs In addition, UE k * is re-incorporated into the set of candidate UEs, that is the second unit j is incorporated into the non-serving second unit set of UE k * and so on; Step 24, go back to Step 21 until 6. The cell-free scalable distributed scheduling and resource allocation method according to claim 3, wherein In Step 3, at each second unit, each UE it schedules first finds the RRU with the largest large-scale coefficient corresponding to it among the RRUs that can serve it, that is This RRU is the primary RRU for this UE, where j k represents the second unit number for scheduling UE k, and then according to Allocate the pilot τ to the UE, i.e., i k = τ, where i k represents the pilot index used by UE k, represents the second unit j k The set of UEs that use pilot t among the UEs scheduled by, and each UE in the above formula is allocated the pilot with the least observable pilot pollution in its corresponding main RRU.

7. The cell-free scalable distributed scheduling and resource allocation method according to claim 3, wherein In step 4, a distributed maximum large-scale fading association algorithm is proposed. Each second unit takes the UE as the center, and each UE scheduled by it is successively associated with the RRU with the largest large-scale fading coefficient that it can serve until the following formula is satisfied. At this time, the local RRU-UE rough association result is obtained; Among them, the set of RRUs serving the UE k in the third unit m is δ is the proportional threshold of the total large-scale coefficient.

8. The cell-free scalable distributed scheduling and resource allocation method according to claim 3, characterized in that In step 5, a pilot allocation adjustment scheme based on the Dis coefficient is proposed. First, the Dis coefficient that measures the pilot contamination when two UEs share the same pilot is calculated where d k =[d k,1 , …, d k,L T and D :,k ∈ {0, 1} L×1 respectively represent the distances from UE k to all RRUs and the serving relationship, and χ k,k' represents the sum of the distance differences between two UEs to each RRU serving them. The smaller its value, the stronger the pilot contamination will be when UE k and UE k' share the same pilot. It is necessary to select the UE pair to be adjusted according to the magnitude of its relative value;​ At this time, for each χ k,k' calculate its z-score. The z-score is used to evaluate the distance of a sample point from the population mean and measure how many standard deviations the original data differs from the data mean, expressed as where and represent the mean and standard deviation of the Dis coefficient. For the UE pair of , if this UE pair shares a pilot, pilot reallocation is required to adjust the threshold Successively extract ζ k,k' The smallest UE pair to be adjusted. At this time, this UE pair shares the same pilot τ, that is, i k = i k' = τ; Calculate the global pilot contamination coefficient of each pilot of the RRU with the largest associated large-scale fading coefficient for UE k and UE k' respectively, that is Among them, represents the set of UEs that globally use pilot t, which is different from τ k represents the global best pilot of UE k at this time, measures the pilot contamination intensity generated by the main RRU of UE k for pilot τ k ; Compare and in terms of magnitude, and select the UE with a smaller pilot contamination intensity for reallocation, that is, if then reallocate pilot i to UE k k = τ k , otherwise, reallocate pilot i to UE k' k' = τ k' .

9. The cell-free scalable distributed scheduling and resource allocation method according to claim 3, characterized in that In step 6, an RRU-UE association adjustment scheme based on the interference-aware reward IAR coefficient is proposed, and the IAR coefficient is used to perform centralized optimization on the RRU-UE association result; Based on the large-scale information, the pilot allocation result, and the current RRU-UE association situation, the IAR coefficient is calculated, Among them, represents the set of UEs sharing pilot i with UE k k ; represents the set of interfering RRUs of UE k, where represents the set of RRUs associated with UE k; Adjust the RRU-UE association according to the IAR coefficient. If the IAR of UE k and RRU(l,m) is higher than that of the currently associated RRU, that is it indicates that a better-performing RRU has been missed in the distributed RRU-UE association process, and then it is further associated with this UE, that is 10. The cell-free scalable distributed scheduling and resource allocation method according to claim 3, wherein, In step 7, when considering the distributed architecture, the fractional power control of UE k calculated at the second unit is shown as follows wherein, represents the maximum uplink transmission power of each UE, κ ∈ [0, 1] is the uplink power scaling factor. When κ = 0, it means that all UEs transmit at the maximum power. The larger κ is, the higher the fairness among UEs is.

Images