Neural network signal processor for magnetic storage channels
AssigneeNational Semiconductor Corporation
InventorWilliam Shields Neely
Life of this patent
Estimating the term…
Abstract
A neural network based signal processor for a magnetic storage channel, such as a magnetic disk drive for a computer, uses a multiple layer perceptron neural network to perform the symbol sequencing detection, equalization and decoding of information signals retrieved from the magnetic storage medium.
Claims14
A neural network system which includes a neural network based signal processor and is configured to train said neural network based signal processor for symbol sequencing detection, equalization and decoding of information signals retrieved from a magnetic storage medium, said neural network system comprising:
- a data source configured to provide a first plurality of data which includes encoding and symbol sequencing;
- a magnetic storage medium model, coupled to said data source and representing a reference magnetic storage medium, configured to receive and process said first plurality of data and in accordance therewith provide a first data signal which represents said first plurality of data, wherein said first data signal includes said encoding, said symbol sequencing and a plurality of nonlinearities, and wherein said plurality of nonlinearities corresponds to an originating of said first data signal from said reference magnetic storage medium;
- an additive noise source, coupled to said magnetic storage medium model, configured to receive said first data signal and add noise thereto and in accordance therewith provide a second data signal, wherein said second data signal includes said encoding, said symbol sequencing, said plurality of nonlinearities and said added noise;
- a neural network signal processor, coupled to said additive noise source, configured to receive and back propagate an error signal and in accordance therewith receive and perform symbol sequencing detection, equalization and decoding upon said second data signal and in accordance therewith provide a third data signal, wherein said third data signal represents a second plurality of data which corresponds to said first plurality of data with said symbol sequencing detected, said encoding decoded, said plurality of nonlinearities substantially equalized and said added noise substantially removed;
- and
- an error detection circuit, coupled to said magnetic storage medium model and said neural network signal processor, configured to receive and process said first and third data signals and in accordance therewith provide said error signal.
A method of training a neural network based signal processor for symbol sequencing detection, equalization and decoding of information signals retrieved from a magnetic storage medium, said method comprising the steps of:
- generating a first plurality of data which includes encoding and symbol sequencing;
- processing, in accordance with a magnetic storage medium model which represents a reference magnetic storage medium, said first plurality of data and in accordance therewith generating a first data signal which represents said first plurality of data, wherein said first data signal includes said encoding, said symbol sequencing and a plurality of nonlinearities, and wherein said plurality of nonlinearities corresponds to an originating of said first data signal from said reference magnetic storage medium;
- adding noise to said first data signal and in accordance therewith generating a second data signal, wherein said second data signal includes said encoding, said symbol sequencing, said plurality of nonlinearities and said added noise;
- receiving and back propagating an error signal and in accordance therewith performing symbol sequencing detection, equalization and decoding upon said second data signal and in accordance therewith generating a third data signal with a neural network signal processor, wherein said third data signal represents a second plurality of data which corresponds to said first plurality of data with said symbol sequencing detected, said encoding decoded, said plurality of nonlinearities substantially equalized and said added noise substantially removed;
- and
- processing said first and third data signals and in accordance therewith generating said error signal.
Dependencies are inferred from the claim text.