Abstract: A codebook-based beamforming architecture utilizes a random forest tree classification machine learning approach in order to circumvent the channel sensing step when designing a beamforming architecture. The random forest tree may be represented by a matrix. A content addressable memory (CAM) may be used to implement the codebook matrix based beamforming architecture using a random forest classifier allowing a near optimal codeword to be selected with a uniform delay. A photonics-based CAM may be used to exploit the high bandwidth and the low power consumption of optical components.

Abstract: An apparatus and method for computing multiplication operations involving vectors, matrices or both, using a photonic computing architecture including an optical crossbar array. A compensation device is used to compensate for non-ideal characteristics of devices in the optical crossbar array. Apparatus and methods for representing negative-valued vectors of a multiplier in the multiplication operation are also provided.

Abstract: A method of beamforming by which a signal is received by a crossbar array of phase change materials (PCM) operative to be modulated electronically to impart complex Cartesian weighting and to retain transmission parameters for future processing. A PCM crossbar array can receive a signal from carrier waves modulated by an array of antenna elements, and send a complex Cartesian weighed result to a baseband combiner. Alternatively, a PCM crossbar array can receive a signal from carrier waves modulated by a baseband precoder, and send a complex Cartesian weighed result to an array of antenna elements. Either of the baseband combiner and baseband precoder can be digital or optical. Implementations allow optically based beamforming with the retention of long-term stochastics in the PCM cells, and are therefore broadband, fast, and energy efficient, as channel state information is not required to be stored otherwise.

Abstract: A photonic device configured to perform matrix vector multiplication operations at high frequencies is provided. The vector being multiplied by the matrix is defined by vector components at specific wavelengths. The device includes a first waveguide and a second waveguide. A series of tunable microring resonators (MRRs) are coupled to the first waveguide and to a respective series of passive delay rings (PDRs), which are coupled to the second waveguide. Each MRR/PDR pair defines a tunable matrix component (tunable weight) for a respective wavelength component of the vector. A series of controllable delay elements (CDEs) such as all-pass filters are coupled to the first waveguide, upstream from the tunable MRRs. Any tuning dependent group delay caused by the MRR/PDR pairs can be compensated by controlling the CDEs such that each wavelength components has substantially a same delay as the other wavelength components.

Abstract: Matrix multiplication is performed using a photonic computing block. Matrices are preprocessed to derive representative positive, real-valued matrices. The representative matrices are multiplied using a photonic computing block in which elements of a first matrix are encoded as a signal intensity and elements of a second matrix are encoded by tuning a spectral filter to apply a corresponding weight coefficient. Pairwise multiplications of matrix elements are performed by processing different wavelength components of an overall optical signal. A balanced photodiode receives the different wavelength components following manipulation in order to complete the multiplication. A matrix multiplication can be partitioned to fit a photonic computing block. The partitions are processed in parallel by multiple photonic computing blocks then reassembled into a final matrix. The multiplication operation can be part of a matrix inversion operation.