A Fully Decentralized Cellular-Free Precoding Method and System Based on Graph Information Dependency

Abstract

This invention discloses a fully decentralized cellular precoding method and system based on graph information dependency. The DPU performs local channel estimation and precoding calculations, and calculates the equivalent channel of the local channel under local precoding. The DPU obtains neighbor information through a sparse fronthaul link network and obtains a directed acyclic graph on the sparse fronthaul link network as the information structure between nodes according to a fully ordered acyclic orientation. The DPU calculates cooperative precoding round-by-round according to graph information dependency, independent of information from other DPUs. This invention achieves fully decentralized distributed precoding for cellular wireless access networks by establishing information dependency in a graph structure, based on team theory, and performing one-time stepwise reasoning and information fusion on the graph. It eliminates the need for iterative computation, provides controllable latency, and offers higher flexibility, better performance, and more realistic fronthaul link assumptions.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a fully decentralized, cellular-free precoding method and system based on graph information dependence. Background Technology

[0002] Cellular networks are considered a crucial architecture for next-generation wireless communication networks. By eliminating the traditional cellular concept, they can provide the same high-quality service to all users. However, current cellular networks often still rely on a central processing unit (CPU) or CPU cluster for processing and computation. This not only requires the deployment of a large, dense, high-performance fronthaul network but also introduces higher latency. Current research and practice in academia and industry on distributed cellular networks have yielded some results. One approach is based on distributed optimization algorithms, which offer high performance upon convergence but require multiple iterations among distributed processing units, leading to significant fronthaul capacity pressure and processing latency, especially amplified in sparse fronthaul networks. Another approach utilizes heuristic or learning-based algorithms, which do not require repeated iterations but suffer from limited performance improvements or effectiveness only in specific scenarios; their practicality and deployability remain to be verified. Summary of the Invention

[0003] Purpose of the invention: This invention provides a fully decentralized cellular precoding method and system based on graph information dependence, which can solve the problem that existing cellular precoding still relies on centralized computing by a central processing unit.

[0004] Technical solution: The fully decentralized, cellular-free precoding method based on graph information dependency described in this invention includes the following steps:

[0005] Step 1: The DPU performs local channel estimation and precoding calculation, and calculates the equivalent channel of the local channel under local precoding.

[0006] Step 2: The DPU obtains neighbor information through the sparse fronthaul link network and obtains a directed acyclic graph on the sparse fronthaul link network in a fully ordered acyclic orientation manner as the information structure between nodes.

[0007] Step 3: The DPU computes collaborative precoding round by round according to graph information dependencies, which does not depend on information from other DPUs.

[0008] Furthermore, in step 1, the equivalent channel of the local channel under local precoding is calculated. ; It is the first The equivalent channel after zero-forcing precoding using local channel state information on each DPU For the number of users, Represents the complex field.

[0009] Furthermore, in step 2, each DPU generates a unique identifier and broadcasts its unique identifier through the sparse fronthaul link network, so that each DPU knows its neighbors and their unique identifiers.

[0010] Furthermore, the DPU sorts nodes based on its own unique identifier and the unique identifiers of its neighbors, according to the size of the binary string represented by the unique identifier. It then directs nodes with smaller unique identifiers to nodes with larger unique identifiers. Through this totally ordered acyclic orientation, the original undirected graph represented by the sparse forward link network is transformed into a directed acyclic graph. .

[0011] Furthermore, the directed acyclic graph (DAG) indicates the information dependencies between DPUs and avoids circular dependencies. In the DAG, DPUs are vertices, and a directed edge from DPU A to another DPU B represents that DPU A depends on information on DPU B. The dependent information is the product of the equivalent channel of other DPUs and cooperative precoding. A certain DPU depends on the first... One DPU is needed , For the first Equivalent channel of each DPU For the first The collaborative precoding that each DPU needs to compute.

[0012] Furthermore, in step 3, the DPU computes collaborative precoding round-by-round according to graph information dependencies, which does not depend on information from other DPUs, including the following steps:

[0013] Step 31: The DPU determines whether it is a source vertex in the graph information dependency structure, that is, a vertex with only incoming edges and no outgoing edges. If it is a source vertex, proceed to step 32; otherwise, proceed to step 33.

[0014] Step 32: The DPU acts as the source to compute the local cooperative precoding. Assume the first... The DPU is the source at this point. Then the cooperative precoding of this DPU is Then the DPU calculation And according to the graph information dependencies, it will be passed to the corresponding nodes, where, This indicates that it does not exist. This is the serial number of the DPU. For a directed acyclic graph The set of edges, To compute cooperative precoding information;

[0015] Step 33: The DPU is not yet the source point, as in step 4. If a DPU is not the source node, then it receives information about the source node in the current directed acyclic graph. ,in The vertex index;

[0016] Step 34: From the directed acyclic graph Delete the source vertex and its associated edges, and repeat step 31. (This applies to directed acyclic graphs.) If only one vertex remains, the computation ends after the last node's cooperative precoding is completed.

[0017] Furthermore, in step 33, after receiving the information of the source node in the current directed acyclic graph, the cooperative precoding calculation formula is corrected according to the following formula.

[0018]

[0019]

[0020] in It indicates that it exists.

[0021] Furthermore, in step 34, from the directed acyclic graph... Deleting the source vertex and its associated edges is...

[0022]

[0023]

[0024] in This represents the set subtraction operation. For a directed acyclic graph The set of vertices.

[0025] Correspondingly, a fully decentralized cellular-free precoding system based on graph information dependence includes: radio access points (APs) for each frequency band, distributed processing units (DPUs), and a sparse fronthaul link network.

[0026] The AP performs beamforming, transmitting and receiving wireless radio frequency signals in various frequency bands, digital-to-analog and analog-to-digital conversion, and waveform signal processing of the physical channel.

[0027] The DPU completes baseband processing and computing tasks and connects to the managed APs via high-performance fiber optic connections. Since the DPUs are deployed in a distributed manner in the coverage area and are close to the managed APs, it is assumed that the DPUs can perform real-time management of the connected APs.

[0028] Sparse fronthaul link networks are used to connect DPUs, meaning that DPUs exchange information through communication links independent of the radio access network, and DPUs can only exchange information with some adjacent DPUs, avoiding long-distance information transmission.

[0029] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: By establishing information dependence in a graph structure, based on team theory, the present invention performs one-time stepwise reasoning and information fusion on the graph, realizing distributed precoding of a completely decentralized non-cellular wireless access network, without iterative computation, with controllable latency, and has higher flexibility, better performance and more realistic fronthaul link assumptions. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0031] Figure 2(a) is a decentralized, cellular-free example scenario of sparse fronthaul link network connection according to the present invention.

[0032] Figure 2(b) is a graph of the dependency relationship between DPUs obtained by the present invention based on the sparse fronthaul link network.

[0033] Figure 3 This is a process diagram of calculating DPU precoding round by round according to information dependency based on Figure 2(b).

[0034] Figure 4 This is a simulation result diagram of the distributed precoding based on graph information dependence in an embodiment of the present invention. Detailed Implementation

[0035] like Figure 1 As shown, a fully decentralized, cellular-free precoding method based on graph information dependency is illustrated using the scenario in Figure 2(a). It is assumed that... This indicates the total number of DPUs. The seven DPUs are connected via a sparse fronthaul link network as shown in Figure 2(a), serving... One user. This includes the following steps:

[0036] In step S100, the DPU performs local channel estimation and precoding calculation, and calculates the equivalent channel of the local channel under local precoding. .

[0037] In step S200, the DPU obtains neighbor information through the sparse fronthaul link network and obtains a directed acyclic graph on the sparse fronthaul link network as the information structure between nodes in the form of a total order acyclic orientation.

[0038] Step S201: The seven DPUs were numbered as shown in Figure 2(a) as sequence numbers 1 to 7. Each DPU generated a unique identifier and broadcast it through the sparse fronthaul link network shown in Figure 2(a). Thus, each DPU learned about its neighbors and their unique identifiers. The table below provides an example based on the scenario in Figure 2(a). Here, it is assumed that the unique identifier generated by the DPU is the independent hot encoding of the DPU's sequence number in Figure 2(a) for simplification. Note that in practice, DPUs only recognize the unique identifier and not the sequence number; the unique identifier is a random string.

[0039] Table 1. Example table based on the scenario in Figure 2(a).

[0040]

[0041] Step S202: The DPU sorts nodes based on its own unique identifier and the unique identifiers of its neighbors. Here, we assume the sorting is based on the size of the binary string represented by the unique identifier, pointing nodes with smaller unique identifiers to nodes with larger unique identifiers. This fully ordered acyclic orientation transforms the original sparse forward link network, representing an undirected graph, into a directed acyclic graph. As shown in Figure 2(b), this directed acyclic graph defines the direction of information exchange between DPUs. Note that Figure 2(b) represents the information dependency relationship between DPUs, not the information transmission relationship. An edge between the first and second DPUs leading to the third DPU indicates that both the first and second DPUs depend on the third DPU, requiring the third DPU to transmit messages before computation can be completed. The dependent information is the product of the equivalent channel of other DPUs and cooperative precoding. For example, if a DPU depends on the first DPU... One DPU is needed Based on the information dependencies in this graph, each DPU can simplify its solution problem as follows:

[0042]

[0043] in The first one to be solved Collaborative precoding of each DPU For the first Equivalent channel of each DPU For the first The cooperative precoding that each DPU needs to compute For a directed acyclic graph The set of edges. As shown in the examples in Figure 2(a) and Figure 2(b), the problem of solving the first DPU becomes:

[0044]

[0045] in The addition terms on the left side of the formula also depend on the information of the relevant nodes.

[0046] Step S300: The DPU computes collaborative precoding round by round according to graph information dependencies, which does not depend on information from other DPUs. .

[0047] In step S301, the DPU determines whether it is a "source vertex" in the graph information dependency structure, that is, a vertex with only incoming edges and no outgoing edges. If it is a source vertex, proceed to step S302; otherwise, proceed to step S303. In the example of Figure 2(b), the 3rd and 7th DPUs are source vertices, marked with red borders.

[0048] Step S302: The DPU acts as the source point to compute the local cooperative precoding. Assume the... The DPU is the source at this point. Then the cooperative precoding of this DPU is Then the DPU calculation And according to the graph information dependencies, it will be passed to the corresponding node. In the examples of Figure 2(a) and Figure 2(b), the third DPU needs to... The data is passed to the first and second DPUs via the fronthaul link, and the same applies to the seventh DPU.

[0049] Step S303: DPU is not currently the source point, as in step... If a DPU is not the source node, then it receives information about the source node in the current directed acyclic graph. And modify the cooperative precoding calculation formula according to the following formula.

[0050]

[0051]

[0052] In the examples of Figure 2(a) and Figure 2(b), the first DPU receives... and The second DPU only receives The 4th and 5th DPUs only received Then the cooperative precoding solution formula for the first DPU becomes:

[0053]

[0054] in It still depends on the remaining nodes.

[0055] Step S304, from Delete the source vertex and its associated edges, i.e.

[0056]

[0057]

[0058] in This indicates a set subtraction operation, and the process loops through step S301, as follows: If only one vertex remains, the computation ends after calculating the last node's cooperative precoding. In the examples of Figure 2(a) and Figure 2(b), the 3rd and 7th DPUs are excluded from the graph information dependency structure after the first round of computation, resulting in the following... Figure 3 The graph information dependency structure shown in the first figure, when repeated in step S300, will yield the following results sequentially. Figure 3 The remaining subgraphs represent the graph information dependency structure.

[0059] By following these three main steps, we have completed fully decentralized, cellular-free precoding based on graph information dependencies. An example of the results can be seen below. Figure 4 . Figure 4 The "graph information dependency" curve represents the method of this invention. "No instantaneous information exchange" refers to the traditional team precoding method that uses only statistics without edge information. "Cluster information exchange" represents the clustered partial collaborative precoding currently used in academia. The capacity for sharing information between DPUs is the same for both "graph information dependency" and "cluster information exchange". However, because the "graph information dependency" method of this invention achieves globally limited collaboration, it effectively compensates for the poor inter-cluster collaboration problem of "cluster information exchange".

Claims

1. A fully decentralized, cellless precoding method based on graph information dependency, characterized in that, Includes the following steps: Step 1: The DPU performs local channel estimation and precoding calculation, and calculates the equivalent channel of the local channel under local precoding. Step 2: The DPU obtains neighbor information through the sparse fronthaul link network and obtains a directed acyclic graph on the sparse fronthaul link network in a fully ordered acyclic orientation manner as the information structure between nodes. Step 3: The DPU computes collaborative precoding round by round according to graph information dependencies, which does not depend on information from other DPUs.

2. The fully decentralized, cellless precoding method based on graph information dependency as described in claim 1, characterized in that, In step 1, the equivalent channel of the local channel under local precoding is calculated. ; It is the first The equivalent channel after zero-forcing precoding using local channel state information on each DPU For the number of users, Represents the complex field.

3. The fully decentralized, cellless precoding method based on graph information dependency as described in claim 1, characterized in that, In step 2, each DPU generates a unique identifier and broadcasts its unique identifier through the sparse fronthaul link network, so that each DPU knows its neighbors and their unique identifiers.

4. The fully decentralized, cellless precoding method based on graph information dependency as described in claim 3, characterized in that, The DPU sorts nodes based on their own unique identifier and the unique identifiers of their neighbors, according to the size of the binary string represented by the unique identifier. It then directs nodes with smaller unique identifiers to nodes with larger unique identifiers. Through this totally ordered acyclic orientation, the original undirected graph represented by the sparse forward link network is transformed into a directed acyclic graph. , For a directed acyclic graph The set of vertices, For a directed acyclic graph The set of edges.

5. The fully decentralized, cellless precoding method based on graph information dependency as described in claim 4, characterized in that, Directed acyclic graphs (DPUs) indicate information dependencies between DPUs and avoid circular dependencies. In a DPU, DPUs are vertices, and a directed edge from DPU A to another DPU B represents that DPU A depends on information on DPU B. The dependent information is the product of the equivalent channel of other DPUs and cooperative precoding. A DPU depends on the first... One DPU is needed , For the first Equivalent channel of each DPU For the first The collaborative precoding that each DPU needs to compute.

6. The fully decentralized, cellular-free precoding method based on graph information dependency as described in claim 1, characterized in that, In step 3, the DPU computes collaborative precoding round-by-round according to graph information dependencies, which does not depend on information from other DPUs. This includes the following steps: Step 31: The DPU determines whether it is a source vertex in the graph information dependency structure, that is, a vertex with only incoming edges and no outgoing edges. If it is a source vertex, proceed to step 32; otherwise, proceed to step 33. Step 32: The DPU acts as the source to compute the local cooperative precoding. Assume the first... The DPU is the source at this point. Then the cooperative precoding of this DPU is Then the DPU calculation And according to the graph information dependencies, it will be passed to the corresponding nodes, where, This indicates that it does not exist. This is the serial number of the DPU. For a directed acyclic graph The set of edges, To compute cooperative precoding information; Step 33: The DPU is not yet the source point, as in step 4. If a DPU is not the source node, then it receives information about the source node in the current directed acyclic graph. ,in The vertex index is... For the first Equivalent channel of each DPU For the first Each DPU needs to compute collaborative precoding; Step 34: From the directed acyclic graph Delete the source vertex and its associated edges, and repeat step 31. (This applies to directed acyclic graphs.) If only one vertex remains, the computation ends after the last node's cooperative precoding is completed.

7. The fully decentralized, cellular-free precoding method based on graph information dependency as described in claim 6, characterized in that, In step 33, after receiving the information of the source node in the current directed acyclic graph, the cooperative precoding calculation formula is corrected according to the following formula. in It indicates that it exists.

8. The fully decentralized, cellular-free precoding method based on graph information dependency as described in claim 6, characterized in that, In step 34, from the directed acyclic graph Deleting the source vertex and its associated edges is... in This represents the set subtraction operation. For a directed acyclic graph The set of vertices.

9. A system for implementing the fully decentralized, cellular-free precoding method based on graph information dependency as described in claim 1, characterized in that, include: Radio access points (APs), distributed processing units (DPUs), and sparse fronthaul link networks for each frequency band; The AP performs beamforming, transmitting and receiving wireless radio frequency signals in various frequency bands, digital-to-analog and analog-to-digital conversion, and waveform signal processing of the physical channel. The DPU completes baseband processing and computing tasks, and manages the APs through high-performance fiber optic connections. The DPU can manage the connected APs in real time. Sparse fronthaul link networks are used to connect DPUs, meaning that DPUs exchange information through communication links independent of the radio access network, and DPUs can only exchange information with some adjacent DPUs, avoiding long-distance information transmission.

Images