Methods and devices for radio resource scheduling in radio access networks
A device may include a processor configured to provide input data that is based on channel information representative of attributes associated with one or more established communication channels of a plurality of communication devices to a trained machine learning model configured to determine a score for each of the plurality of communication devices based on the input data, wherein the score represents a likelihood of the respective communication device to be scheduled for a communication resource to perform a communication, determine, for the communication resource, one or more communication devices from the plurality of communication devices based on the determined score for each of the plurality of communication devices, and provide an output to schedule the communication resource for the one or more communication devices.
1 . A device comprising:
a memory; and
a processor configured to:
provide a graph as input data, the graph being based on channel information representative of attributes associated with one or more established communication channels of a plurality of communication devices, to a trained machine learning model configured to determine a score for each of the plurality of communication devices based on the input data, wherein the graph comprises a plurality of vertices comprising multiple input multiple output (MIMO) layer vertices, wherein each MIMO layer vertex is associated with a MIMO layer of one of the plurality of communication devices, and a communication resource vertex associated with the communication resource, wherein the graph comprises a plurality of edge sets between vertices, each edge set represent a relation associated to the vertices coupled by the edges, wherein the determined score is an attention score and represents a likelihood of the respective communication device to be scheduled for the communication resource to perform a communication;
obtain a cross-interference adjacency matrix representing a relation associated with the cross-interference between the communication devices, a user fairness adjacency matrix representing a relation associated with user fairness based on user fairness based on a priority defined for at least some of the plurality of communication devices, and an initial attention matrix representing an initial likelihood of the plurality of communication devices to be scheduled for the communication resource;
determine, for the communication resource, one or more communication devices from the plurality of communication devices based on the determined score for each of the plurality of communication devices;
provide an output to schedule the communication resource for the one or more communication devices.
2 . The device of claim 1 ,
wherein the one or more communication channels are established between a respective communication device of the plurality of communication devices and a base station or another communication device.
3 . The device of claim 1 ,
wherein the one or more communication devices comprise at least two communication devices.
4 . The device of claim 1 ,
wherein the processor is further configured to determine, for the communication resource, the one or more communication devices from the plurality of communication devices based on a threshold applied to the determined scores.
5 . The device of claim 1 ,
wherein the channel information comprises information representing, for a communication channel, at least one of a channel quality indicator (CQI), a rank indicator, a correlation matrix, a precoding matrix indicator (PMI), a precoding matrix, information representing beamforming weights, signal interference to noise ratio (SINR) per subband, SINR per a multiple input multiple output (MIMO) layer, SINR associated with a transport block, and/or information associated with measurements performed by the device for the communication channel.
6 . The device of claim 5 ,
wherein the channel information comprises the information for each sub-band of a plurality of sub-bands.
7 . The device of claim 1 ,
wherein the input data is further based on device information representative of at least one of predefined communication device priorities for at least some of the plurality of communication devices, predetermined maximum transmit power of the plurality of communication devices, predetermined average data rate of the plurality of communication devices, and/or predetermined buffer status representative of data scheduled for communication in one or more buffers of the plurality of communication devices.
8 . The device of claim 1 ,
wherein the trained machine learning model is configured to determine the determined score for the each of the plurality of communication devices based on a parameter, wherein the parameter represents the effect of at least one other communication device of the plurality of communication device to the one or more established communication channels of the respective communication device;
wherein the parameter is representative of at least one of cross-interference between the plurality of communication devices, user fairness based on a priority defined for at least some of the plurality of communication devices, or sub-band association of the plurality of communication devices representing the likelihood of the respective communication device to be scheduled for a sub-band.
9 . The device of claim 1 ,
wherein the plurality of edge sets comprising a first set of edge weights based on cross-interference, a second set of edge weights based on user fairness representing predefined priorities for at least some of the plurality of communication devices, and a third set of edge weights based on user selection score representing the likelihood of the respective communication device to be scheduled.
10 . The device of claim 1 ,
wherein the trained machine learning model comprises a graph neural network (GNN) configured to receive the input data;
wherein the attention score represents the importance of vertices associated with the respective communication device to the communication resource vertex based on an attention mechanism.
11 . The device of claim 10 ,
wherein the GNN comprises a multi-head graph attention network (MH-GAT);
wherein the MH-GAT is configured to determine the attention score based on the input data comprising a plurality of graphs for a plurality of sub-band graphs, each graph is obtained for a sub-band of the communication resource based on a portion of the channel information that is associated with the respective input and comprising a sub-band vertex as the communication resource vertex.
12 . The device of claim 11 ,
wherein the determined attention score for each communication device is based on the plurality of graphs obtained for all sub-bands of the communication resource.
13 . The device of claim 10 ,
wherein the GNN comprises one or more of layers that are configured to receive input features and provide an output including the attention score, wherein the determined score comprises the attention score that is calculated by the last layer of the one or more GNN layers.
14 . The device of claim 13 ,
wherein the one or more GNN layers comprises an L-layer diffusion convolutional GNN configured to provide output
X
s
l
according to an equation
X
s
l
=
A
s
l
F
s
l
W
a
l
+
C
s
F
s
l
W
c
l
+
Q
s
F
s
l
W
q
l
wherein
F
s
l
+
1
comprises an output of an activation function with an input
X
s
l
;
wherein s is a sub-band index, 1 is a layer index, A comprises the score adjacency matrix, F comprises the input feature vector, W a comprises machine learning model parameters for user selection score, C comprises the interference adjacency matrix, W c comprises the machine learning model parameters for cross-interference, Q comprises the information representing the priority, W c comprises machine learning model parameters for user fairness, and F comprises concatenation of input feature vectors and a feature vector defined for the sub-band.
15 . The device of claim 1 ,
wherein the processor is further configured to provide output representative of a rate allocation and/or a modulation coding scheme (MCS) selection.
16 . The device of claim 1 , further comprising:
a transceiver configured to communicate with the one or more communication devices according to the scheduled communication resource;
wherein the transceiver is configured to transmit radio communication signals to the one or more communication devices by using the communication resource.
17 . The device of claim 1 ,
wherein the processor is configured to implement the trained machine learning model at a medium access control (MAC) layer according to an open systems interconnection (OSI) model.
18 . A method comprising:
providing a graph as input data, the graph being based on channel information representative of attributes associated with one or more established communication channels of a plurality of communication devices, to a trained machine learning model, wherein the graph comprises a plurality of vertices comprising multiple input multiple output (MIMO) layer vertices, wherein each MIMO layer vertex is associated with a MIMO layer of one of the plurality of communication devices, and a communication resource vertex associated with the communication resource, wherein the graph comprises a plurality of edge sets between vertices, each edge set represent a relation associated to the vertices coupled by the edges;
obtaining a cross-interference adjacency matrix representing a relation associated with cross-interference between the communication devices, a user fairness adjacency matrix representing a relation associated with user fairness based on a priority defined for at least some of the plurality of communication devices, and an initial attention matrix representing an initial likelihood of the plurality of communication devices to be scheduled for the communication resource; wherein the trained machine learning module is configured to determine a score for each of the plurality of communication devices based on the input data, wherein the determined score for each of the plurality of communication devices is an attention score and represents a likelihood of the respective communication device to be scheduled for the communication resource to perform a communication;
determining, for the communication resource, one or more communication devices from the plurality of communication devices based on the determined score for each of the plurality of communication devices;
providing an output to schedule the communication resource for the one or more communication devices.
19 . A non-transitory computer-readable medium comprising one or more instructions which, if executed by a processor, cause the processor to:
provide a graph as input data, the graph being based on channel information representative of attributes associated with one or more established communication channels of a plurality of communication devices to a trained machine learning model, wherein the graph comprises a plurality of vertices comprising multiple input multiple output (MIMO) layer vertices, wherein each MIMO layer vertex is associated with a MIMO layer of one of the plurality of communication devices, and a communication resource vertex associated with the communication resource, wherein the graph comprises a plurality of edge sets between vertices, each edge set represent a relation associated to the vertices coupled by the edges;
obtain a cross-interference adjacency matrix representing a relation associated with cross-interference between the communication devices, a user fairness adjacency matrix representing a relation associated with user fairness based on a priority defined for at least some of the plurality of communication devices, and an initial attention matrix representing an initial likelihood of the plurality of communication devices to be scheduled for the communication resource; wherein the trained machine learning module is configured to determine a score for each of the plurality of communication devices based on the input data, wherein the determined score for each of the plurality of communication devices is an attention score and represents a likelihood of the respective communication device to be scheduled for the communication resource to perform a communication;
determine, for the communication resource, one or more communication devices from the plurality of communication devices based on the determined score for each of the plurality of communication devices;
provide an output to schedule the communication resource for the one or more communication devices.
20 . The non-transitory computer-readable medium of claim 19 ,
wherein the one or more communication devices comprise at least two communication devices.