Low size, weight and power (swap) efficient hardware implementation of a wide instantaneous bandwidth neuromorphic adaptive core (NeurACore)
Described is a Neuromorphic Adaptive Core (NeurACore) cognitive signal processor (CSP) for wide instantaneous bandwidth denoising of noisy signals. The NeurACore CSP includes a NeurACore block, a globally learning layer, and a neural combiner. The NeurACore block is operable for receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network. The global learning layer is operable for adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal. Further, the neural combiner is operable for combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals.
1. A Neuromorphic Adaptive Core (NeurACore) cognitive signal processor (CSP) for wide instantaneous bandwidth denoising of noisy signals, comprising:
a NeurACore block, the NeurACore block being operable for receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network;
a global learning layer, the global learning layer operable for adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal;
a neural combiner, the neural combiner operable for combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals; and
wherein the NeurACore CSP is implemented as a Wide Instantaneous Bandwidth (WIB) NeurACore by incorporating a process that includes an adaptive core equation, output layer update equations, and weights update equations.
2. The NeurACore CSP as set forth in claim 1 , wherein in the output layer update equations, an output is created after an elementwise multiplication between weights (C) and states (X), a size of which is determined by a systems embedding factor (K).
3. The NeurACore CSP as set forth in claim 2 , wherein the output layer update equation is modified to represent hardware delays.
4. A Neuromorphic Adaptive Core (NeurACore) cognitive signal processor (CSP) for wide instantaneous bandwidth denoising of noisy signals, comprising:
a NeurACore block, the NeurACore block being operable for receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network;
a global learning layer, the global learning layer operable for adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal;
a neural combiner, the neural combiner operable for combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals; and
wherein the NeurACore block includes nodes that are operable as resonators and oscillators with a tunable frequency and tunable quality factor to denoise radio frequency and acoustic signals.
5. A Neuromorphic Adaptive Core (NeurACore) cognitive signal processor (CSP) for wide instantaneous bandwidth denoising of noisy signals, comprising:
a NeurACore block, the NeurACore block being operable for receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network;
a global learning layer, the global learning layer operable for adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal;
a neural combiner, the neural combiner operable for combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals; and
wherein the NeurACore block is configured to handle both real and complex-valued I/Q signals, with the CSP operable with matrices A and B as follows:
[
X
˙
I
X
˙
Q
]
=
A
_
_
=
[
X
I
X
Q
]
+
B
_
I
u
I
+
B
_
Q
u
Q
,
where X denotes a neural state space vector, and where I and Q subscripts denote I and Q channels of input u.
6. A computer program product for wide instantaneous bandwidth denoising of noisy signals, comprising:
a non-transitory computer readable medium having executable instructions encoded thereon, such that upon execution of the instructions, one or more processors perform operations of:
receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network;
adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal;
combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals; and
wherein mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network incorporates a process that includes an adaptive core equation, output layer update equations, and weights update equations.
7. The computer program product as set forth in claim 6 , wherein in the output layer update equations, an output is created after an elementwise multiplication between weights (C) and states (X), a size of which is determined by a systems embedding factor (K).
8. The computer program product as set forth in claim 7 , wherein the output layer update equation is modified to represent hardware delays.
9. A computer program product for wide instantaneous bandwidth denoising of noisy signals, comprising:
a non-transitory computer readable medium having executable instructions encoded thereon, such that upon execution of the instructions, one or more processors perform operations of:
receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network;
adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal;
combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals; and
wherein the neural network includes nodes that are operable as resonators and oscillators with a tunable frequency and tunable quality factor to denoise radio frequency and acoustic signals.
10. A method for wide instantaneous bandwidth denoising of noisy signals, comprising acts of:
causing one or more processors perform operations of:
receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network;
adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal;
combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals; and
wherein mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network incorporates a process that includes an adaptive core equation, output layer update equations, and weights update equations.
11. The method as set forth in claim 10 , wherein in the output layer update equations, an output is created after an elementwise multiplication between weights (C) and states (X), a size of which is determined by a systems embedding factor (K).
12. The method as set forth in claim 11 , wherein the output layer update equation is modified to represent hardware delays.
13. A method for wide instantaneous bandwidth denoising of noisy signals, comprising acts of:
causing one or more processors perform operations of:
receiving as an input a mixture of in-phase and quadrature (I/Q) signals and mapping the I/Q signals onto a neural network to determine complex-valued output weights of neural states of the neural network;
adapting the complex-valued output weights to predict a most likely next value of the input I/Q signal;
combining a set of delayed neural state vectors with the weights of the global learning layer to compute an output signal, the output signal being separate in-phase and quadrature signals; and
wherein the neural network includes nodes that are operable as resonators and oscillators with a tunable frequency and tunable quality factor to denoise radio frequency and acoustic signals.