Back to results
US 12,388,685 B2Granted
Regularized interference-plus-noise covariance matrix estimation
AssigneeNokia Solutions and Networks Oy
InventorElias Aksel Raninen
Life of this patent
Estimating the term…
Abstract
Apparatuses and methods in a communication system are disclosed. A signal is received. The signal comprises demodulation reference information. The apparatus compute estimates of the channel and of interference-plus-noise vectors. The interference-plus-noise vectors are used as input for choosing a tapering matrix and for determining shrinkage target matrix and parameter. An interference plus noise covariance matrix estimator is computed and used in an equalizer, which has as input an estimate of the interference-plus-noise covariance matrix.
Claims18
An apparatus in a communication system, comprising:
- a processor;
- and a memory including instructions, the instructions, when executed by the processor, cause the apparatus to:
- control reception, with a number p of antennas, a signal comprising demodulation reference information;
- estimating n number of interference-plus-noise vectors of dimension p;
- computing, based on the estimated interference-plus-noise vectors, a sample covariance matrix S;
- determining a target matrix F based on the sample covariance matrix S and the identity matrix I;
- determining a set of tapering matrices {T1, . . . , TK}, where K is the total number of tapering matrices considered;
- choosing a tapering matrix T from the set {T1, . . . , TK} that minimizes a cross-validation loss function, wherein the cross-validation loss for each Tk, k=1, . . . , K, is computed by partitioning the estimated interference-plus-noise vectors into a given number of B folds, and, for each fold i=1, . . . , B, calculating the squared Frobenius distance between an interference-plus-noise sample covariance matrix computed using only the estimated interference-plus-noise vectors in the current fold i and a tapered interference-plus-noise sample covariance matrix computed as the elementwise (Hadamard) product between Tk and the interference-plus-noise sample covariance matrix computed using the remaining interference-plus-noise vectors in the B−1 other folds excluding fold i, and averaging over the B computed distances;
- determining the tapering matrix T using leave-one-out cross-validation as described above, where the given number of folds B equals to the number n of estimated interference-plus-noise vectors;
- determining a tapered interference-plus-noise sample covariance matrix as the elementwise (Hadamard) product of the interference-plus-noise sample covariance matrix S and the chosen tapering matrix T;
- determining an estimator {circumflex over (R)} of the interference-plus-noise covariance matrix R as a convex combination with coefficient ρ of the tapered interference-plus-noise sample covariance matrix and the target matrix F;
- determining the value for the shrinkage parameter ρ as the minimizer of the estimated mean squared error between the interference-plus-noise covariance matrix R and the estimate {circumflex over (R)} of the interference-plus-noise covariance matrix, where the shrinkage parameter ρ is restricted to be a real-valued number between 0 and 1;
- applying the estimator {circumflex over (R)} using the determined shrinkage parameter and the target matrix in the equalization of a data signal.
10 more elements
A method in an apparatus in a communication system comprising the steps of:
- controlling reception, with a number p of antennas, a signal comprising demodulation reference information;
- estimating n number of interference-plus-noise vectors of dimension p;
- computing, based on the estimated interference-plus-noise vectors, a sample covariance matrix S;
- determining a target matrix F based on the sample covariance matrix S and the identity matrix I;
- determining a set of tapering matrices {T1, . . . , TK}, where K is the total number of tapering matrices considered;
- choosing a tapering matrix T from the set {T1, . . . , TK} that minimizes a cross-validation loss function, wherein the cross-validation loss for each Tk, k=1, . . . , K, is computed by partitioning the estimated interference-plus-noise vectors into a given number of B folds, and, for each fold i=1, . . . , B, calculating the squared Frobenius distance between an interference-plus-noise sample covariance matrix computed using only the estimated interference-plus-noise vectors in the current fold i and a tapered interference-plus-noise sample covariance matrix computed as the elementwise (Hadamard) product between Tk and the interference-plus-noise sample covariance matrix computed using the remaining interference-plus-noise vectors in the B−1 other folds excluding fold i, and averaging over the B computed distances;
- determining the tapering matrix T using leave-one-out cross-validation as described above, where the given number of folds B equals to the number n of estimated interference-plus-noise vectors;
- determining a tapered interference-plus-noise sample covariance matrix as the elementwise (Hadamard) product of the interference-plus-noise sample covariance matrix S and the chosen tapering matrix T;
- determining an estimator {circumflex over (R)} of the interference-plus-noise covariance matrix R as a convex combination with coefficient ρ of the tapered interference-plus-noise sample covariance matrix and the target matrix F;
- determining the value for the shrinkage parameter ρ as the minimizer of the estimated mean squared error between the interference-plus-noise covariance matrix R and the estimate {circumflex over (R)} of the interference-plus-noise covariance matrix, where the shrinkage parameter ρ is restricted to be a real-valued number between 0 and 1;
- applying the estimator {circumflex over (R)} using the determined shrinkage parameter and the target matrix in the equalization of a data signal.
8 more elements
Dependencies are inferred from the claim text.