Methods and apparatus for signal correlation using a quadratic form
A method for training a machine learning (ML) model to perform signal detection to correct a carrier frequency offset associated with a signal includes training a machine learning (ML) model associated with a receiver, based on (1) data including a plurality of different carrier frequency offset values, (2) a matched filter, and (3) a linear transformation matrix, to produce a trained ML model. A signal is received at the receiver, the signal having an associated carrier frequency offset that is not known to the receiver at the time the signal is received. The method also includes identifying, at the receiver and using the trained ML model, the carrier frequency offset of the signal, and correcting the carrier frequency offset of the signal to identify a message encoded by the signal.
1. A non-transitory, processor-readable medium storing instructions that, when executed by a processor, cause the processor to:
one of:
train a machine learning (ML) model associated with a receiver, based on (1) data including a plurality of different carrier frequency offset values, (2) a matched filter, and (3) a linear transformation matrix, to produce a trained ML model, or
receive the trained ML model from a remote compute device;
receive a signal at the receiver, the signal having an associated carrier frequency offset that is not known to the receiver at the time the signal is received;
identify, at the receiver and using the trained ML model, the carrier frequency offset of the signal; and
correct the carrier frequency offset of the signal to identify a message encoded by the signal.
2. The non-transitory, processor-readable medium of claim 1 , wherein the instructions to cause the processor to train the ML model include instructions to train the ML model using a correlator defined as:
a mn =δ mn ( e iα ) m ,
where δ mn is a Kronecker delta, a mn are elements of the linear transformation matrix, and a is a free parameter to be optimized.
3. The non-transitory, processor-readable medium of claim 2 , wherein the linear transformation matrix is not an identity matrix.
4. The non-transitory, processor-readable medium of claim 1 , wherein the instructions to cause the processor to identify the carrier frequency offset of the signal include instructions to estimate the carrier frequency offset of the signal.
5. The non-transitory, processor-readable medium of claim 1 , wherein the instructions to cause the processor to train the ML model include instructions to train the ML model using a value defined as D α ·z, where D α is defined as ē iϕ z ·D θ · z , z represents an in-phase/quadrature (I/Q) point, z represents a conjugate transpose of z, D θ is an N×N diagonal unitary matrix with entries e inθ for n∈[0, N), N being an integer, and z represents a sequence of I/Q points.
6. The non-transitory, processor-readable medium of claim 1 , wherein the instructions to cause the processor to train the ML model include instructions to train the ML model in response to the signal being received at the receiver.
7. The non-transitory, processor-readable medium of claim 1 , wherein the instructions to cause the processor to train the ML model include instructions to train the ML model using a correlation having a quadratic form.
8. A method, comprising:
training a machine learning (ML) model associated with a receiver, based on (1) a plurality of different channel distortions, (2) a matched filter, and (3) one of a Toeplitz matrix or a circulant matrix, to produce a trained ML model;
receiving a signal at the receiver, the signal having an associated channel distortion that is not known to the receiver at the time the signal is received;
detecting, at the receiver and using the trained ML model, the channel distortion of the signal; and
equalizing the channel distortion of the signal to identify a message encoded by the signal.
9. The method of claim 8 , further comprising selecting the one of the Toeplitz matrix or the circulant matrix based on whether the signal includes a cyclic prefix.
10. The method of claim 8 , wherein the matched filter is a first matched filter and the trained ML model is configured to identify a second matched filter different from the first matched filter, and to retrain the trained ML model based on the second matched filter.
11. The method of claim 8 , wherein the detecting the channel distortion includes performing one of cross-correlation or auto-correlation of portions of the signal.
12. The method of claim 8 , wherein the detecting the channel distortion includes performing a correlation based on the signal, the correlation having a quadratic form.
13. The method of claim 12 , wherein the quadratic form is:
corr
(
v
_
,
w
_
)
≡
∑
i
,
j
=
1
N
a
ij
v
i
*
w
j
,
where a ij are elements of the one of the Toeplitz matrix or the circulant matrix, vi are elements of a first component of the signal, w j are elements of a second component of the signal different from the first component of the signal, and * denotes a complex conjugate.
14. The method of claim 8 , further comprising selecting the matched filter in response to receiving the signal.
15. A non-transitory, processor-readable medium storing instructions that, when executed by a processor, cause the processor to:
train a machine learning (ML) model associated with a receiver, based on (1) a plurality of different time-dependent narrow-band basis vectors, and (2) a matched filter, to produce a trained ML model;
receive a signal at the receiver, the signal having an associated interfering signal that is not known to the receiver at the time the signal is received;
identify, at the receiver and using the trained ML model, the interfering signal associated with the signal; and
subtract the interfering signal from the signal to identify a message encoded by the signal.
16. The non-transitory, processor-readable medium of claim 15 , wherein each time-dependent narrow-band basis vector from the plurality of different time-dependent narrow-band basis vectors is an element of a Shannon-Whittaker basis.
17. The non-transitory, processor-readable medium of claim 16 , wherein the trained ML model is configured to identify coefficients of a linear combination of elements of the Shannon-Whittaker basis.
18. The non-transitory, processor-readable medium of claim 15 , wherein the interfering signal is a narrowband signal.
19. The non-transitory, processor-readable medium of claim 15 , wherein the instructions to cause the processor to identify the interfering signal include instructions to perform a correlation based on the signal, the correlation having a quadratic form and not including an identity matrix.
20. The non-transitory, processor-readable medium of claim 19 , wherein the quadratic form includes a homogeneous polynomial.