# US 5,705,956 A · Neural network based PLL > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 5,705,956 A - Document ID: US5705956A - Status: Granted - Kind: A - Application number: 08/678,969 - Filed: 1996-07-12 - Granted: 1998-01-06 - Earliest claimed: 1996-07-12 - Assignees: National Semiconductor Corporation - Inventors: William Shields Neely - Web page: https://patentgrep.com/patent/US5705956A ## CPC - H03L 7/08: Details of the phase-locked loop - H03L 7/085: concerning mainly the frequency- or phase-detection arrangement including the filtering or amplification of its output signal (H03L7/10 takes precedence; circuits for comparing the phase or frequency of two mutually-independent oscillations H03D13/00) - H03L: AUTOMATIC CONTROL, STARTING, SYNCHRONISATION OR STABILISATION OF GENERATORS OF ELECTRONIC OSCILLATIONS OR PULSES (generation of oscillations H03B) ## Abstract A neural network controlled phase lock loop (PLL) with an artificial neural network for comparing the phase of the feedback signal from the voltage controlled oscillator (VCO) with the phase of the PLL reference signal. The VCO feedback signal and PLL reference signal are inputted to and processed by a multiple layer perception which has been trained with a model of the VCO to generate the appropriate control voltage necessary to establish and maintain the desired phase and/or frequency of the VCO output signal. ## Claims ### Claim 1 An apparatus including a neural network controlled oscillator loop comprising: a neural network signal processor configured to receive and process a reference signal and an oscillator signal and in accordance therewith provide a control signal, wherein said reference signal includes a reference phase and a reference frequency and said oscillator signal includes an oscillator phase and an oscillator frequency; and an oscillator, coupled to said neural network signal processor, configured to receive said control signal and in accordance therewith provide said oscillator signal, wherein at least one of said oscillator phase and frequency is controlled by said control signal and is established in accordance with at least one of said reference phase and frequency, respectively. ### Claim 2 (depends on 1) The apparatus of claim 1, wherein said neural network signal processor comprises a multiple layer perceptron which includes first and second inputs configured to receive said reference signal and said oscillator signal and an output configured to provide said control signal. ### Claim 3 (depends on 1) The apparatus of claim 1, wherein said neural network signal processor comprises: a plurality of preprocessors configured to receive said reference signal and said oscillator signal and in accordance therewith provide first and second pluralities of temporally related signals; and a neural network, coupled to said plurality of preprocessors, configured to receive and process said first and second pluralities of temporally related signals together simultaneously and in accordance therewith provide said control signal. ### Claim 4 (depends on 3) The apparatus of claim 3, wherein said plurality of preprocessors comprises first and second tapped delay lines configured to receive and sequentially time delay said reference signal and said oscillator signal and in accordance therewith provide first and second pluralities of sequentially time delayed signals as said first and second pluralities of temporally related signals. ### Claim 5 (depends on 3) The apparatus of claim 3, wherein said plurality of preprocessors comprises first and second shift registers configured to receive and selectively shift said reference signal and said oscillator signal and in accordance therewith provide first and second pluralities of sequentially time shifted signals as said first and second pluralities of temporally related signals. ### Claim 6 (depends on 3) The apparatus of claim 3, wherein said neural network comprises a multiple layer perception which includes first and second pluralities of inputs configured to receive said first and second pluralities of temporally related signals and an output configured to provide said control signal. ### Claim 7 (depends on 1) The apparatus of claim 1, further comprising a filter, coupled between said neural network and said oscillator, configured to filter said control signal. ### Claim 8 (depends on 1) The apparatus of claim 1, wherein said oscillator comprises a voltage controlled oscillator and said control signal comprises a control voltage. ### Claim 9 (depends on 1) The apparatus of claim 1, wherein said oscillator comprises a current controlled oscillator and said control signal comprises a control current. ### Claim 10 A neural network system which includes a neural network signal processor and is configured to train said neural network signal processor for controlling an oscillator, said neural network system comprising: a neural network signal processor configured to receive and back propagate an error signal and to receive and process a desired response signal and an example input signal and in accordance therewith provide a control signal, wherein said error signal corresponds to a difference between said desired response signal and said example input signal; an inverse circuit model, coupled to said neural network signal processor, configured to receive said desired response signal and in accordance therewith provide an inverse model signal, wherein said inverse circuit model has a transfer function associated therewith and a ratio of said control signal and said example input signal correspond to an inverse of said transfer function; and an error detection circuit, coupled to said neural network signal processor and said inverse circuit model, configured to receive and process said control signal and said inverse model signal and in accordance therewith provide said error signal. ### Claim 11 (depends on 10) The neural network system of claim 10, wherein said neural network signal processor comprises a multiple layer perceptron which includes first and second inputs configured to receive said desired response signal and said example input signal and an output configured to provide said control signal. ### Claim 12 (depends on 10) The neural network system of claim 10, wherein said neural network signal processor comprises: a plurality of preprocessors configured to receive said desired response signal and said example input signal and in accordance therewith provide first and second pluralities of temporally related signals; and a neural network, coupled to said plurality of preprocessors, configured to receive and back propagate said error signal and to receive and process said first and second pluralities of temporally related signals together simultaneously and in accordance therewith provide said control signal. ### Claim 13 (depends on 12) The neural network system of claim 12, wherein said plurality of preprocessors comprises first and second tapped delay lines configured to receive and sequentially time delay said desired response signal and said example input signal and in accordance therewith provide first and second pluralities of sequentially time delayed signals as said first and second pluralities of temporally related signals. ### Claim 14 (depends on 12) The neural network system of claim 12, wherein said plurality of preprocessors comprises first and second shift registers configured to receive and selectively shift said desired response signal and said example input signal and in accordance therewith provide first and second pluralities of sequentially time shifted signals as said first and second pluralities of temporally related signals. ### Claim 15 (depends on 12) The neural network system of claim 12, wherein said neural network comprises a multiple layer perceptron which includes first and second pluralities of inputs configured to receive said first and second pluralities of temporally related signals and an output configured to provide said control signal. ### Claim 16 (depends on 10) The neural network system of claim 10, wherein said error detection circuit comprises a signal combiner configured to receive and combine said control signal and said inverse model signal and in accordance therewith provide said error signal. ### Claim 17 (depends on 10) The neural network system of claim 10, wherein said inverse circuit model represents an inverse circuit model of a signal controlled oscillator. ### Claim 18 (depends on 10) The neural network system of claim 10, wherein said inverse circuit model represents an inverse circuit model of a serially coupled signal controlled oscillator and filter. ### Claim 19 A neural network based method of controlling an oscillator comprising the steps of: receiving and processing a reference signal and an oscillator signal and in accordance therewith generating a control signal with a neural network signal processor, wherein said reference signal includes a reference phase and a reference frequency and said oscillator signal includes an oscillator phase and an oscillator frequency; and receiving said control signal and in accordance therewith generating said oscillator signal, wherein at least one of said oscillator phase and frequency is controlled by said control signal and is established in accordance with at least one of said reference phase and frequency, respectively. ### Claim 20 (depends on 19) The method of claim 19, wherein said step of receiving and processing a reference signal and an oscillator signal and in accordance therewith generating a control signal with a neural network signal processor comprises receiving said reference signal and said oscillator signal via first and second inputs of a multiple layer perceptron and processing said reference signal and said oscillator signal to generate said control signal and outputting said control signal with said multiple layer perceptron. ### Claim 21 (depends on 19) The method of claim 19, wherein said step of receiving and processing a reference signal and an oscillator signal and in accordance therewith generating a control signal with a neural network signal processor comprises: receiving said reference signal and said oscillator signal and in accordance therewith generating first and second pluralities of temporally related signals; and receiving and processing said first and second pluralities of temporally related signals together simultaneously and in accordance therewith generating said control signal with a neural network. ### Claim 22 (depends on 21) The method of claim 21, wherein said step of receiving said reference signal and said oscillator signal and in accordance therewith generating first and second pluralities of temporally related signals comprises receiving and sequentially time delaying said reference signal and said oscillator signal and in accordance therewith generating first and second pluralities of sequentially time delayed signals as said first and second pluralities of temporally related signals. ### Claim 23 (depends on 21) The method of claim 21, wherein said step of receiving said reference signal and said oscillator signal and in accordance therewith generating first and second pluralities of temporally related signals comprises receiving and selectively shifting said reference signal and said oscillator signal and in accordance therewith generating first and second pluralities of sequentially time shifted signals as said first and second pluralities of temporally related signals. ### Claim 24 (depends on 21) The method of claim 21, wherein said step of receiving and processing said first and second pluralities of temporally related signals together simultaneously and in accordance therewith generating said control signal with a neural network comprises receiving said first and second pluralities of temporally related signals via first and second pluralities of inputs of a multiple layer perception and processing said first and second pluralities of temporally related signals to generate said control signal and outputting said control signal with said multiple layer perceptron. ### Claim 25 (depends on 19) The method of claim 19, further comprising the step of filtering said control signal. ### Claim 26 A method of training a neural network signal processor for controlling an oscillator, said method comprising the steps of: receiving and back propagating an error signal with a neural network signal processor; receiving and processing a desired response signal and an example input signal and in accordance therewith generating a control signal with said neural network signal processor; receiving said desired response signal and generating an inverse model signal in accordance with a transfer function, wherein a ratio of said control signal and said example input signal correspond to an inverse of said transfer function; and processing said control signal and said inverse model signal and in accordance therewith generating said error signal, wherein said error signal corresponds to a difference between said desired response signal and said example input signal. ### Claim 27 (depends on 26) The method of claim 26, wherein: said step of receiving and back propagating an error signal with a neural network signal processor comprises receiving and back propagating an error signal with a multiple layer perceptron; and said step of receiving and processing a desired response signal and an example input signal and in accordance therewith generating a control signal with said neural network signal processor comprises inputting said desired response signal and said example input signal to first and second inputs and outputting said control signal from an output of said multiple layer perceptron. ### Claim 28 (depends on 26) The method of claim 26, wherein: said step of receiving and back propagating an error signal with a neural network signal processor comprises receiving and back propagating an error signal with a neural network; and said step of receiving and processing a desired response signal and an example input signal and in accordance therewith generating a control signal with said neural network signal processor comprises: receiving said desired response signal and said example input signal and in accordance therewith generating first and second pluralities of temporally related signals; and processing said first and second pluralities of temporally related signals together simultaneously and in accordance therewith generating said control signal with said neural network. ### Claim 29 (depends on 28) The method of claim 28, wherein said step of receiving said desired response signal and said example input signal and in accordance therewith generating first and second pluralities of temporally related signals comprises receiving and sequentially time delaying said desired response signal and said example input signal with first and second tapped delay lines and in accordance therewith generating first and second pluralities of sequentially time delayed signals as said first and second pluralities of temporally related signals. ### Claim 30 (depends on 28) The method of claim 28, wherein said step of receiving said desired response signal and said example input signal and in accordance therewith generating first and second pluralities of temporally related signals comprises receiving and selectively shifting said desired response signal and said example input signal with first and second shift registers and in accordance therewith generating first and second pluralities of sequentially time shifted signals as said first and second pluralities of temporally related signals. ### Claim 31 (depends on 28) The method of claim 28, wherein: said step of receiving and back propagating an error signal with a neural network comprises receiving and back propagating an error signal with a multiple layer perceptron; and said step of processing said first and second pluralities of temporally related signals together simultaneously and in accordance therewith generating said control signal with said neural network comprises inputting said first and second pluralities of temporally related signals to first and second pluralities of inputs and outputting said control signal from an output of said multiple layer perceptron. ### Claim 32 (depends on 26) The method of claim 26, wherein said step of processing said control signal and said inverse model signal and in accordance therewith generating said error signal comprises combining said control signal and said inverse model signal and in accordance therewith generating said error signal. ### Claim 33 (depends on 26) The method of claim 26, wherein said step of receiving said desired response signal and generating an inverse model signal in accordance with a transfer function comprises receiving said desired response signal and generating an inverse model signal in accordance with a transfer function which represents an inverse circuit model of a signal controlled oscillator. ### Claim 34 (depends on 26) The method of claim 26, wherein said step of receiving said desired response signal and generating an inverse model signal in accordance with a transfer function comprises receiving said desired response signal and generating an inverse model signal in accordance with a transfer function which represents an inverse circuit model of a serially coupled signal controlled oscillator and filter.