User selection for MU-MIMO communications

1. A method for user selection, comprising:

for a whole set of candidate terminal devices, performing scheduling by:

obtaining a set of received signals;

determining a set of measurements for relevance among the candidate terminal devices based at least on the set of received signals and channel estimation for the set of candidate terminal devices, wherein determining the set of measurements for relevance is iteratively performed at least once until an iteration number is equal to a predetermined number and comprises:

determining residue between each received signal and its corresponding output of an artificial neuron network (ANN), having an input layer and an output layer, each of row vectors of a stream channel matrix being an input of the ANN, each of the receiving signals being an output of the ANN, an activation function for a neuron of the output layer being a linear activation function, the stream channel matrix being a product of a channel estimation matrix and a receiving beamforming matrix for the set of candidate terminal devices,

determining the set of measurements for relevance based on the residue, and

updating a set of weights of the ANN based on a gradient descent algorithm;

determining a maximum relevance measurement in the set of measurements; and

updating the set of candidate terminal devices by removing a candidate terminal device associated with the maximum relevance measurement from the set of candidate terminal devices.

2. The method of claim 1 , wherein performing scheduling includes iteratively doing until an end condition is met, the end condition including one of the following:

the number of the candidate terminal devices being less than or equal to a threshold number and channel capacity for the set of candidate terminal devices being less than channel capacity for the updated set of candidate terminal devices; and

the number of the candidate terminal devices being less than or equal to the threshold number and a convergence rate being below than a predetermined threshold, the convergence rate being determined based on a change in a sum of the measurements.

3. The method of claim 1 , wherein updating the set of weights of the ANN comprises:

determining a set of adjust rates for a learning rate based on the row vectors;

determining a set of adjusted learning rates based on the set of adjust rates; and

updating the set of weights based on the set of adjusted learning rates, the residue and the stream channel matrix.

4. The method of claim 1 , wherein determining the set of measurements for relevance comprises:

determining a relevance measurement between a first candidate terminal device and a plurality of second candidate terminal devices in the set of candidate terminal devices based on

a size of a projection of a first channel vector for the first candidate terminal device on a space constructed by channel vectors for the second candidate terminal devices, the first candidate terminal device being different from any of the second candidate terminal devices.

5. The method of claim 1 , wherein obtaining the set of received signals comprises:

determining lengths of row vectors in the stream channel matrix;

arranging the row vectors in an order of the lengths;

assigning transmitted symbols to the arranged row vectors so that different transmitted symbols are assigned to adjacent row vectors; and

determining products of the transmitted symbols and lengths of the arranged row vectors as the set of received signals.

6. A device for user selection, comprising:

at least one processor; and

at least one memory including computer program code;

the at least one memory and the computer program code configured to, with the at least one processor, cause the device to at least:

for a whole set of candidate terminal devices, perform scheduling by:

obtaining a set of received signals;

determining a set of measurements for relevance among the candidate terminal devices based at least on the set of received signals and channel estimation for the set of candidate terminal devices, wherein determining the set of measurements for relevance is iteratively performed at least once until an iteration number is equal to a predetermined number and comprises:

determining residue between each received signal and its corresponding output of an artificial neuron network (ANN), having an input layer and an output layer, each of row vectors of a stream channel matrix being an input of the ANN, each of the receiving signals being an output of the ANN, an activation function for a neuron of the output layer being a linear activation function, the stream channel matrix being a product of a channel estimation matrix and a receiving beamforming matrix for the set of candidate terminal devices,

determining the set of measurements for relevance based on the residue, and

updating a set of weights of the ANN based on a gradient descent algorithm;

determining a maximum relevance measurement in the set of measurements; and

updating the set of candidate terminal devices by removing a candidate terminal device associated with the maximum relevance measurement from the set of candidate terminal devices.

7. The device of claim 6 , wherein performing scheduling includes iteratively doing until an end condition is met, the end condition including one of the following:

the number of the candidate terminal devices being less than or equal to a threshold number and channel capacity for the set of candidate terminal devices being less than channel capacity for the updated set of candidate terminal devices; and

the number of the candidate terminal devices being less than or equal to the threshold number and a convergence rate being below than a predetermined threshold, the convergence rate being determined based on a change in a sum of the measurements.

8. The device of claim 6 , wherein updating the set of weights of the ANN comprises:

determining a set of adjust rates for a learning rate based on the row vectors;

determining a set of adjusted learning rates based on the set of adjust rates; and

updating the set of weights based on the set of adjusted learning rates, the residue and the stream channel matrix.

9. The device of claim 6 , wherein determining the set of measurements for relevance comprises:

determining a relevance measurement between a first candidate terminal device and a plurality of second candidate terminal devices in the set of candidate terminal devices based on

a size of a projection of a first channel vector for the first candidate terminal device on a space constructed by channel vectors for the second candidate terminal devices, the first candidate terminal device being different from any of the second candidate terminal devices.

10. The device of claim 6 , wherein obtaining the set of received signals comprises:

determining lengths of row vectors in the stream channel matrix;

arranging the row vectors in an order of the lengths;

assigning transmitted symbols to the arranged row vectors so that different transmitted symbols are assigned to adjacent row vectors; and

determining products of the transmitted symbols and lengths of the arranged row vectors as the set of received signals.

11. A computer program embodied on a non-transitory computer-readable medium, the computer program comprising encoded instructions which, when executed by a processor, cause the processor to carry out a method for user selection, the method comprising:

for a whole set of candidate terminal devices, performing scheduling by:

obtaining a set of received signals;

determining a set of measurements for relevance among the candidate terminal devices based at least on the set of received signals and channel estimation for the set of candidate terminal devices, wherein determining the set of measurements for relevance is iteratively performed at least once until an iteration number is equal to a predetermined number and comprises:

determining residue between each received signal and its corresponding output of an artificial neuron network (ANN), having an input layer and an output layer, each of row vectors of a stream channel matrix being an input of the ANN, each of the receiving signals being an output of the ANN, an activation function for a neuron of the output layer being a linear activation function, the stream channel matrix being a product of a channel estimation matrix and a receiving beamforming matrix for the set of candidate terminal devices,

determining the set of measurements for relevance based on the residue, and

updating a set of weights of the ANN based on a gradient descent algorithm;

determining a maximum relevance measurement in the set of measurements; and

updating the set of candidate terminal devices by removing a candidate terminal device associated with the maximum relevance measurement from the set of candidate terminal devices.