# US 11,443,167 B2 · Systems and methods for modulation classification of baseband signals using multiple data representations of signal samples > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,443,167 B2 - Document ID: US11443167B2 - Status: Granted - Kind: B2 - Application number: 16/382,912 - Filed: 2019-04-12 - Published: 2020-10-15 - Granted: 2022-09-13 - Earliest claimed: 2019-04-12 - Assignees: Motorola Solutions, Inc. - Applicants: Motorola Solutions, Inc. - Inventors: Stephen J. Govea; Nathanael P. Kuehner; David N. Taylor; Rodger W. Caruthers; Micah D. Silberstein; Gregory Agami - Web page: https://patentgrep.com/patent/US11443167B2 ## CPC - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/08: Learning methods - G06N 3/09: Supervised learning - H04L 27/12: Modulator circuits; Transmitter circuits - H04L 27/14: Demodulator circuits; Receiver circuits - H04L 27/2003: {for continuous phase modulation (frequency shift keying H04L27/10)} - H04L 27/22: Demodulator circuits; Receiver circuits - H04L 27/36: Modulator circuits; Transmitter circuits - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) ## Abstract Systems and methods for classifying baseband signals with respect to modulation type include receiving, at a consolidated neural network whose objective is modulation classification performance, a complex quadrature vector of interest including multiple samples of a baseband signal derived from a radio frequency signal of unknown modulation type, generating multiple data representations of the vector of interest, providing each data representation to one of multiple parallel neural networks in the consolidated neural network, and receiving a classification result for the baseband signal based on combined outputs of the parallel neural networks. The consolidated neural network may be trained to classify baseband signals with respect to known modulation types by receiving complex quadrature training vectors, each including samples of a baseband signal derived from a radio frequency signal of known modulation type, and comparing a classification result for the training vector to the known modulation type to determine modulation classification performance. ## Claims ### Claim 1 A method for classifying baseband signals derived from radio frequency signals with respect to modulation type, comprising: receiving, at a consolidated neural network whose objective is modulation classification performance, a complex quadrature vector of interest including a plurality of samples of a baseband signal derived from a radio frequency signal of an unknown modulation type, each sample including real and imaginary parts of the baseband signal of the unknown modulation type in the complex plane; generating a plurality of different data representations of the complex quadrature vector of interest; providing each data representation of the complex quadrature vector of interest to a respective one of a plurality of parallel neural networks of the consolidated neural network; and receiving, from the consolidated neural network, a classification result for the baseband signal derived from the radio frequency signal of the unknown modulation type. ### Claim 2 (depends on 1) The method of claim 1, further comprising, prior to receiving the complex quadrature vector of interest, training the consolidated neural network to classify baseband signals derived from radio frequency signals with respect to a plurality of known modulation types. ### Claim 3 (depends on 1) The method of claim 1, further comprising, prior to receiving the complex quadrature vector of interest, training the consolidated neural network to classify baseband signals derived from radio frequency signals with respect to a plurality of known modulation types, the training including: receiving a plurality of complex quadrature training vectors, each training vector including a plurality of samples of a baseband signal derived from a radio frequency signal of a respective one of the plurality of known modulation types; for each of the plurality of complex quadrature training vectors: generating a plurality of different data representations of the complex quadrature training vector; providing each data representation of the complex quadrature training vector to a respective one of the plurality of parallel neural networks of the consolidated neural network; receiving, as an output of the consolidated neural network, a classification result for the complex quadrature training vector with respect to the plurality of known modulation types; comparing the classification result for the complex quadrature training vector with the known modulation type of the baseband signal for which a plurality of samples is included in the complex quadrature training vector; determining modulation classification performance of the consolidated neural network dependent on the comparing; and modifying a configuration parameter of the consolidated neural network dependent on the determined modulation classification performance of the consolidated neural network. ### Claim 4 (depends on 1) The method of claim 1, wherein the classification result for the baseband signal derived from the radio frequency signal of the unknown modulation type comprises a final output vector in which each element of the final output vector indicates a probability that the unknown modulation type is a respective one of a plurality of known modulation types. ### Claim 5 (depends on 4) The method of claim 4, further comprising: outputting, by each of the plurality of parallel neural networks of the consolidated neural network, a respective constituent output vector in which each element of the constituent output vector indicates a probability that the unknown modulation type is a respective one of the plurality of known modulation types; and combining the constituent output vectors to generate the final output vector. ### Claim 6 (depends on 5) The method of claim 5, wherein combining the constituent output vectors to generate the final output vector comprises applying a respective weight to each of the constituent output vectors. ### Claim 7 (depends on 4) The method of claim 4, further comprising: outputting, by each of the plurality of parallel neural networks of the consolidated neural network, a respective intermediate output characterizing the baseband signal derived from the radio frequency signal of the unknown modulation type; and generating the final output vector dependent on the intermediate outputs. ### Claim 8 (depends on 4) The method of claim 4, further comprising: outputting, by each of the plurality of parallel neural networks of the consolidated neural network, a respective intermediate output characterizing the baseband signal derived from the radio frequency signal of the unknown modulation type; generating a consolidated output vector dependent on the intermediate outputs; providing each data representation of the complex quadrature vector of interest to a respective one of a plurality of individual neural networks of an ensemble neural network, the ensemble neural network including the plurality of individual neural networks and the consolidated neural network; outputting, by each of the plurality of individual neural networks of the ensemble neural network, a respective constituent output vector in which each element of the constituent output vector indicates a probability that the unknown modulation type is a respective one of the plurality of known modulation types; and combining the constituent output vectors and the consolidated output vector to generate the final output vector. ### Claim 9 (depends on 1) The method of claim 1, further comprising filtering the complex quadrature vector of interest prior to generating the plurality of different data representations of the complex quadrature vector of interest. ### Claim 10 (depends on 1) The method of claim 1, wherein generating the plurality of different data representations of the complex quadrature vector of interest comprises applying, to the complex quadrature vector of interest, a domain-specific transformation. ### Claim 11 (depends on 1) The method of claim 1, wherein each of the plurality of parallel neural networks of the consolidated neural network comprises one or more of a residual neural network, a convolutional layer, a pooling layer, and a fully connected layer. ### Claim 12 (depends on 1) The method of claim 1, wherein the plurality of different data representations includes an averaged Power Spectral Density (PSD) representation, a spectrogram PSD representation, an eye diagram representation, a phase representation, a phase differentiation representation, or an autocorrelation representation. ### Claim 13 A system for classifying baseband signals derived from radio frequency signals with respect to modulation type, comprising: a consolidated neural network whose objective is modulation classification performance, the consolidated neural network comprising a plurality of parallel neural networks; and a pre-processing stage configured to: receive a complex quadrature vector of interest including a plurality of samples of a baseband signal derived from a radio frequency signal of an unknown modulation type, each sample including real and imaginary parts of the baseband signal of the unknown modulation type in the complex plane; generate a plurality of different data representations of the complex quadrature vector of interest; and provide each data representation of the complex quadrature vector of interest to a respective one of the plurality of parallel neural networks of the consolidated neural network; wherein the consolidated neural network is configured to generate, dependent on the data representations of the complex quadrature vector of interest provided to the respective ones of the plurality of parallel neural networks, a classification result for the baseband signal derived from the radio frequency signal of the unknown modulation type, the classification result including a final output vector in which each element indicates a probability that the unknown modulation type is a respective one of a plurality of known modulation types. ### Claim 14 (depends on 13) The system of claim 13, wherein, prior to receiving the complex quadrature vector of interest: the pre-processing stage is further configured to: receive a plurality of complex quadrature training vectors, each training vector including a plurality of samples of a baseband signal derived from a radio frequency signal of a respective one of the plurality of known modulation types, and each sample including real and imaginary parts of the baseband signal derived from the radio frequency signal of the respective known modulation type in the complex plane; and for each of the plurality of complex quadrature training vectors: generate a plurality of different data representations of the complex quadrature training vector; and provide each data representation of the complex quadrature training vector to a respective one of the plurality of parallel neural networks of the consolidated neural network to train the consolidated neural network for modulation classification with respect to the plurality of known modulation types; receive, as an output of the consolidated neural network, a classification result for the complex quadrature training vector with respect to the plurality of known modulation types; and the consolidated neural network is further configured to, for each of the plurality of complex quadrature training vectors: compare the classification result for the complex quadrature training vector with the known modulation type of the baseband signal for which a plurality of samples is included in the complex quadrature training vector; determine modulation classification performance of the consolidated neural network dependent on the comparing; and modify a configuration parameter of the consolidated neural network dependent on the determined modulation classification performance of the consolidated neural network. ### Claim 15 (depends on 13) The system of claim 13, wherein: each of the plurality of parallel neural networks of the consolidated neural network is configured to output a respective constituent output vector in which each element of the constituent output vector indicates a probability that the unknown modulation type is a respective one of the plurality of known modulation types; and the system further comprises a concatenation stage configured to: apply a respective weight to each of the constituent output vectors; and combine the weighted constituent output vectors to generate the final output vector. ### Claim 16 (depends on 13) The system of claim 13, wherein: each of the plurality of parallel neural networks of the consolidated neural network is configured to output a respective intermediate output characterizing the baseband signal derived from the radio frequency signal of the unknown modulation type; and the consolidated neural network is configured to generate the final output vector dependent on the intermediate outputs. ### Claim 17 (depends on 13) The system of claim 13, wherein: the system further comprises an ensemble neural network including: a plurality of individual neural networks; the consolidated neural network; and a concatenation stage; each of the plurality of parallel neural networks of the consolidated neural network is configured to output a respective intermediate output characterizing the baseband signal derived from the radio frequency signal of the unknown modulation type; the consolidated neural network is configured to generate a consolidated output vector dependent on the intermediate outputs; the pre-processing stage is further configured to provide each data representation of the complex quadrature vector of interest to a respective one of the plurality of individual neural networks of the ensemble neural network; each of the plurality of individual neural networks of the ensemble neural network is configured to output a respective constituent output vector in which each element of the constituent output vector indicates a probability that the unknown modulation type is a respective one of the plurality of known modulation types; and the concatenation stage is configured to combine the constituent output vectors and the consolidated output vector to generate the final output vector. ### Claim 18 (depends on 13) The system of claim 13, further comprising a filtering stage configured to filter the complex quadrature vector of interest prior to generating the plurality of different data representations of the complex quadrature vector of interest. ### Claim 19 (depends on 13) The system of claim 13, wherein: generating the plurality of different data representations of the complex quadrature vector of interest comprises applying, to the complex quadrature vector of interest, a domain-specific transformation; and the plurality of different data representations includes an averaged Power Spectral Density (PSD) representation, a spectrogram PSD representation, an eye diagram representation, a phase representation, a phase differentiation representation, or an autocorrelation representation. ### Claim 20 (depends on 13) The system of claim 13, wherein each of the plurality of parallel neural networks of the consolidated neural network comprises one or more of a residual neural network, a convolutional layer, a pooling layer, and a fully connected layer.