# US 11,907,829 B2 · FMCW radar with interference signal suppression using artificial neural network > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,907,829 B2 - Document ID: US11907829B2 - Status: Granted - Kind: B2 - Application number: 17/121,295 - Filed: 2020-12-14 - Published: 2021-07-08 - Granted: 2024-02-20 - Earliest claimed: 2019-03-14 - Assignees: Infineon Technologies AG - Applicants: Infineon Technologies AG - Inventors: Paul Meissner; Franz Pernkopf; Johanna Rock; Wolfgang Roth; Mate Andras Toth; Alexander Fuchs - Web page: https://patentgrep.com/patent/US11907829B2 ## CPC - G06N 3/063: using electronic means - G01S 7/417: {involving the use of neural networks} - G06N 3/04: Architecture, e.g. interconnection topology - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/0495: Quantised networks; Sparse networks; Compressed networks - G06N 3/09: Supervised learning - G01S 7/023: {Interference mitigation, e.g. reducing or avoiding non-intentional interference with other HF-transmitters, base station transmitters for mobile communication or other radar systems, e.g. using electro-magnetic interference [EMI] reduction techniques (auxiliary means for detecting or identifying radar signals or the like G01S7/021; means for anti-jamming G01S7/36)} - G01S 7/352: {Receivers} - G01S 7/354: {Extracting wanted echo-signals (Doppler systems G01S13/50)} - G01S 7/356: {involving particularities of FFT processing} - G01S 13/343: {using sawtooth modulation} - G01S 13/4454: {phase comparisons monopulse, i.e. comparing the echo signals received by an interferometric antenna arrangement} - G01S 13/584: {adapted for simultaneous range and velocity measurements} - G01S 13/931: of land vehicles - G06N 3/045: Combinations of networks - G06N 3/048: Activation functions - G06N 3/084: Backpropagation, e.g. using gradient descent - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G01S: RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES ## Abstract A radar device may include a radar transmitter to output a radio frequency (RF) transmission signal including a plurality of frequency-modulated chirps. The radar device may include a radar receiver to receive an RF radar signal, and generate, based on the RF radar signal, a dataset including a set of digital values, the dataset being associated with a chirp or a sequence of successive chirps. The radar device may include a neural network to filter the dataset to reduce an interfering signal included in the dataset, the neural network being a convolutional neural network. At least one layer of the neural network may be a complex-valued neural network layer includes complex-valued weighting factor, where the complex-valued neural network layer is configured to perform one or more operations according to a complex-valued computation. ## Claims ### Claim 1 A radar device associated with a first vehicle, the radar device comprising: a radar receiver configured to: receive a radio frequency (RF) radar signal and, generate a digital signal based on the RF radar signal, the digital signal comprising a plurality of signal segments; and a neural network comprising a plurality of layers, each layer of the plurality of layers having one or more neurons, wherein at least one layer of the plurality of layers is a complex-valued neural network layer comprising complex-valued weighting factors selected by the neural network from a predetermined set of discrete values, wherein the complex-valued neural network layer is configured to perform one or more operations according to a complex-valued computation and the complex-valued weighting factors, and wherein an output layer of the plurality of layers has at least one neuron that delivers an output value that indicates whether a respective signal segment or a sample of the respective signal segment of the plurality of signal segments, associated with the at least one neuron, is overlaid with an interfering signal generated by a radar sensor associated with a second vehicle different from the first vehicle. ### Claim 2 (depends on 1) The radar device of claim 1, wherein each layer of the plurality of layers is a complex-valued neural network layer configured to perform one or more operations according to complex-valued computations. ### Claim 3 (depends on 1) The radar device of claim 1, wherein the neural network comprises only complex-valued neural network layers. ### Claim 4 (depends on 1) The radar device of claim 1, wherein the one or more operations to be performed by the complex-valued neural network layer include a complex-valued convolution. ### Claim 5 (depends on 1) The radar device of claim 1, wherein the one or more operations to be performed by the complex-valued neural network layer include a complex-valued batch normalization. ### Claim 6 (depends on 1) The radar device of claim 1, wherein the one or more operations to be performed by the complex-valued neural network layer include an execution of a complex-valued activation function. ### Claim 7 (depends on 6) The radar device of claim 6, wherein the complex-valued activation function is a complex-valued rectifier linear unit (ReLU) non-linearity function. ### Claim 8 (depends on 1) The radar device of claim 1, wherein the complex-valued neural network layer is a first complex-valued neural network layer and the plurality of layers include a second complex-valued neural network layer and a third complex-valued neural network layer, wherein the first complex-valued neural network layer is configured to perform a complex-valued convolution followed by an execution of a complex-valued rectifier linear unit (ReLU) non-linearity function, wherein the second complex-valued neural network layer is configured to perform a complex-valued convolution, followed by a complex-valued batch normalization, and further followed by an execution of a complex-valued ReLU non-linearity function, wherein the third complex-valued neural network layer is configured to perform a complex-valued convolution, and wherein the second complex-valued neural network layer in arranged between the first complex-valued neural network layer and the third complex-valued neural network layer. ### Claim 9 (depends on 1) The radar device of claim 1, wherein the interfering signal is a radar signal received by the radar device from an external radar device. ### Claim 10 (depends on 1) The radar device of claim 1, wherein the radar device comprises a digital signal processor configured to receive a digitized baseband signal and apply a Fourier transform to the digitized baseband signal to generate the plurality of signal segments. ### Claim 11 (depends on 1) The radar device off claim 1, wherein the plurality of signal segments corresponds to at least one of range map segments or range-Doppler map segments. ### Claim 12 (depends on 1) The radar device of claim 1, wherein the plurality of signal segments are discrete frequency values of a range map, range-Doppler map, or an angle-Doppler map, wherein each discrete frequency value corresponds to a different frequency bin. ### Claim 13 (depends on 1) The radar device of claim 1, wherein possible values for the complex-valued weighting factors selected by the neural network are limited to the discrete values in the predetermined set of discrete values. ### Claim 14 A radar device associated with a first vehicle, the radar device comprising: a radar transmitter configured to output a radio frequency (RF) transmission signal including a plurality of frequency-modulated chirps, a radar receiver configured to: receive an RF radar signal, wherein the RF radar signal is overlaid with an interfering signal that originates from a radar sensor associated with a second vehicle different from the first vehicle, generate, based on the RF radar signal, a dataset including a set of digital values, the dataset being associated with a chirp or a sequence of successive chirps; and a neural network configured to filter the dataset to reduce the interfering signal included in the dataset, the neural network is a convolutional neural network, wherein at least one layer of the neural network is a complex-valued neural network layer comprising complex-valued weighting factors selected by the neural network from a predetermined set of discrete values, wherein the complex-valued neural network layer is configured to perform one or more operations according to a complex-valued computation and the complex-valued weighting factors, wherein the neural network is configured to update the complex-valued weighting factors by selecting updated complex-valued weighting factors from the predetermined set of discrete values. ### Claim 15 (depends on 14) The radar device of claim 14, wherein each layer of the neural network is a complex-valued neural network layer configured to perform one or more operations according to complex-valued computations. ### Claim 16 (depends on 14) The radar device of claim 14, wherein the one or more operations to be performed by the complex-valued neural network layer include at least one of a complex-valued convolution, a complex-valued batch normalization, or an execution of a complex-valued activation function. ### Claim 17 (depends on 14) The radar device of claim 14, wherein the complex-valued neural network layer is a first complex-valued neural network layer and the neural network further includes a second complex-valued neural network layer and a third complex-valued neural network layer, wherein the first complex-valued neural network layer is configured to perform a complex-valued convolution followed by an execution of a complex-valued rectifier linear unit (ReLU) non-linearity function, wherein the second complex-valued neural network layer is configured to perform a complex-valued convolution, followed by a complex-valued batch normalization, and further followed by an execution of a complex-valued ReLU non-linearity function, wherein the third complex-valued neural network layer is configured to perform a complex-valued convolution, and wherein the second complex-valued neural network layer in arranged between the first complex-valued neural network layer and the third complex-valued neural network layer. ### Claim 18 A method, comprising: transmitting, by a radar device associated with a first vehicle, a radio frequency (RF) transmission signal including a plurality of frequency-modulated chirps; receiving, by the radar device, an RF radar signal, wherein the RF radar signal is overlaid with an interfering signal that originates from a radar sensor associated with a second vehicle different from the first vehicle; generating, by the radar device, based on the RF radar signal, a dataset including a set of digital values, the dataset being associated with a chirp or a sequence of successive chirps of the RF transmission signal and associated with the interfering signal; selecting, by a convolutional neural network of the radar device, complex-valued weighting factors from a predetermined set of discrete values; and filtering, by the radar device, the dataset using the convolutional neural network, the convolutional neural network being configured to reduce the interfering signal included in the dataset, wherein at least one layer of the convolutional neural network is a complex-valued neural network layer comprising the complex-valued weighting factors, wherein the complex-valued neural network layer is configured to perform one or more operations according to a complex-valued computation and the complex-valued weighting factors. ### Claim 19 (depends on 18) The method of claim 18, wherein all layers of the convolutional neural network are a complex-valued neural network layers configured to perform one or more operations according to complex-valued computations. ### Claim 20 (depends on 18) The method of claim 18, wherein the one or more operations to be performed by the complex-valued neural network layer include at least one of a complex-valued convolution, a complex-valued batch normalization, or an execution of a complex-valued activation function.