Multiple-input multiple-output (MIMO) detector selection using neural network
A method and system for multiple-input multiple-output (MIMO) detector selection using a neural network is herein disclosed. According to one embodiment, a method includes generating a labelled dataset of features and detector labels, training a multi-layer perceptron (MLP) network using the generated labelled dataset, and selecting a detector class from a plurality of detector classes based on outputs of the trained MLP network.
1. A method of using a multi-layer perceptron (MLP) network to select a low-complexity, reliable detector, the method comprising:
receiving a signal vector and a multiple-input multiple-output (MIMO) channel matrix of resource elements (REs);
extracting channel features from the signal vector and the MIMO channel matrix;
generating a labelled dataset of the channel features and detector labels for each of the RE;
training the MLP network using the generated labelled dataset;
computing a margin associated with a maximum output value from the MLP network, wherein the computed margin is determined based on a conditional probability of detector error being less than or equal to a probability threshold value;
selecting, for an RE, a detector class from a plurality of detector classes based on a difference between the maximum output value from the MLP network and a second output value from the MLP network being less than the computed margin; and
detecting symbols in the RE using a MIMO detector corresponding to the selected detector class.
2. The method of claim 1 , wherein the labelled dataset is generated based on a log-likelihood ratio (LLR) sign.
3. The method of claim 2 , wherein the labelled dataset is further generated based on an LLR magnitude.
4. The method of claim 1 , further comprising merging classes of the plurality of detector classes based on the generated labelled dataset.
5. The method of claim 4 , wherein merging classes further comprises merging samples in a first class of the plurality of classes into a second class of the plurality of classes, wherein the second class includes fewer samples than the first class.
6. The method of claim 1 , wherein selecting the detector class further comprises minimizing a probability of detector error.
7. The method of claim 1 , wherein the channel features include at least one of eigenvalues of a channel, diagonal values of a channel matrix, and an inner product of received signals and the channel.
8. A system using a multi-layer perceptron (MLP) network to select a low-complexity, reliable detector, the system comprising:
a plurality of detectors, each of the detectors associated with a detector class from a plurality of detector classes;
a memory; and
a processor configured to:
receive a signal vector and a multiple-input multiple-output (MIMO) channel matrix of resource elements (REs);
extract channel features from the signal vector and the MIMO channel matrix;
generate a labelled dataset of the channel features and detector labels for each of the RE;
train the MLP network using the generated labelled dataset;
compute a margin associated with a maximum output value from the MLP network, wherein the computed margin is determined based on a conditional probability of detector error being less than or equal to a probability threshold value;
select, for an RE, a detector class from the plurality of detector classes based on a difference between the maximum output value from the MLP network and a second output value from the MLP network being less than the computed margin; and
detect symbols in the RE using a MIMO detector corresponding to the selected detector class.
9. The system of claim 8 , wherein the labelled dataset is generated based on a log-likelihood ratio (LLR) sign.
10. The system of claim 9 , wherein the labelled dataset is further generated based on an LLR magnitude.
11. The system of claim 8 , wherein the processor is further configured to merge classes of the plurality of detector classes based on the generated labelled dataset.
12. The system of claim 11 , wherein the processor is further configured to merge samples in a first class of the plurality of classes into a second class of the plurality of classes, wherein the second class includes fewer samples than the first class.
13. The system of claim 8 , wherein the processor is further configured to select the detector class by minimizing a probability of detector error.
14. A method of using a multi-layer perceptron (MLP) network to select a low-complexity, reliable detector, the method comprising:
receiving a signal vector and a multiple-input multiple-output (MIMO) channel matrix of resource elements (REs);
extracting channel features from the signal vector and the MIMO channel matrix;
generating a labelled dataset of the channel features and detector labels for each of the RE;
merging classes of a plurality of detector classes based on the generated labelled dataset;
training the MLP network using the generated labelled dataset;
computing a margin associated with a maximum output value from the MLP network, wherein the computed margin is determined based on a conditional probability of detector error being less than or equal to a probability threshold value;
selecting, for an RE, a detector class from the merged plurality of detector classes based on a difference between the maximum output value from the MLP network and a second output value from the MLP network being less than the computed margin; and
detecting symbols in the RE using a MIMO detector corresponding to the selected detector class.