Back to results
US 11,599,799 B1Granted
Digital signal processing with neural networks
AssigneeRockwell Collins, Inc.
InventorsRyan M. Murphy, Benjamin C. Barnett
Life of this patent
Estimating the term…
Abstract
A computerized neural network training and testing environment trains and validates a neural network to produce outputs corresponding to a digital signal processing (DSP) algorithm. After validation and testing, the neural network is replicable in any processing environment, independent of device architecture. Such devices may also include a neural network having a reversed process flow to produce a digital signal corresponding to an inverse of the DSP algorithm.
Claims15
A computer apparatus comprising:
- at least one processor;
- memory connected to the processor for storing processor executable code to configure the at least one processor to:
- instantiate a neural network comprising:
- an input layer comprising at least one node, the at least one node configured to receive a digital signal input;
- at least one intermediate layer, each comprising at least one node, each node configured to receive outputs from nodes in previous layers;
- and
- an output layer comprising at least one node, the at least one node configured to receive outputs from nodes in a last intermediate layer and produce one or more output signals;
- receive a training data set comprising a plurality of digital signals and outputs, each output corresponding to a digital signal processing (DSP) algorithm output when applied to the associated digital signal;
- iteratively adjust at least one of a node composition of one or more layers or a synaptic weight of one or more nodes in at least one layer based on a training algorithm;
- convert the neural network to a data structure representing the layer structure, node composition, and synaptic weights of the neural network;
- output the data structure;
- and
- construct a reversed neural network data structure having a reversed process flow corresponding to the data structure wherein a layer order and synaptic weights are reversed.
10 more elements
A method comprising:
- instantiating a neural network comprising:
- an input layer comprising at least one node, the at least one node configured to receive a digital signal input;
- at least one intermediate layer, each comprising at least one node, each node configured to receive outputs from nodes in previous layers;
- and
- an output layer comprising at least one node, the at least one node configured to receive outputs from nodes in a last intermediate layer and produce one or more output signals;
- instantiating a reversed neural network having a reversed process flow corresponding to the neural network wherein a layer order and synaptic weights are reversed;
- receiving a digital signal;
- processing the digital signal through the neural network to produce an output corresponding to a digital signal processing (DSP) algorithm;
- receiving a signal;
- and
- processing the signal via the reversed neural network to produce a digital signal corresponding to a reversed DSP algorithm.
8 more elements
A modem comprising:
- at least one processor;
- memory connected to the processor for storing processor executable code to configure the at least one processor to:
- instantiate a neural network by receiving and converting a data structure representing a layer structure, node composition, and set of synaptic weights, the neural network comprising:
- an input layer comprising at least one node, the at least one node configured to receive a digital signal input;
- at least one intermediate layer, each comprising at least one node, each node configured to receive outputs from nodes in previous layers;
- and
- an output layer comprising at least one node, the at least one node configured to receive outputs from nodes in a last intermediate layer and produce one or more output signals;
- instantiate a reversed neural network based on a data structure having a reversed process flow as compared to the data structure wherein a layer order and synaptic weights are reversed;
- receive a digital signal;
- and
- process the digital signal through the neural network to produce an output corresponding to a digital signal processing (DSP) algorithm.
8 more elements
Dependencies are inferred from the claim text.