Artificial neural network read channel

• US 5,699,487 A
• Filed: 07/07/1995
• Issued: 12/16/1997
• Est. Priority Date: 07/07/1995
• Status: Expired due to Fees

• Current Assignee: Hewlett-Packard Company (HP Inc.)
• Original Assignee: Hewlett-Packard Company (HP Inc.)
• Inventors: Richardson, Kenneth G.
• Primary Examiner(s): Hafiz, Tariq R.
• Application Number:
  US08/499,152
• Time in Patent Office: 893 Days
• Field of Search: 360/46, 395/20-25, 395/27, 382/155-159
• US Class Current: 706/20
• CPC Class Codes: G11B 20/10009 Improvement or modification of read or write signals
• CPC Class Codes: G11B 20/10518 using neural networks
• CPC Class Codes: G11B 5/09 Digital recording
• CPC Class Codes: H04L 2025/03464 Neural networks
• CPC Class Codes: H04L 25/03165 using neural networks
• CPC Class Codes: H04L 7/0062 detection of error based on data decision error, e.g. Mueller type detection
• CPC Class Codes: H04L 7/02 Speed or phase control by the received code signals, the signals containing no special synchronisation information H04L7/0075 takes precedence
• CPC Class Codes: H04L 7/10 Arrangements for initial synchronisation

Abstract

A magnetic read channel employs an artificial neural network for reconstruction of a recorded magnetic signal and its corresponding synchronization signal. A magnetic read head receives magnetic signals from a magnetic recording media such as a magnetic tape or disk and converts it to an electronic signal. A preamplifier receives and amplifies the electronic signal from the magnetic read head to produce an amplified electronic signal. A delay line receives the amplified electronic signal from the preamplifier, storing delayed successive representations of the received signal. An artificial neural network receives the delayed successive representations from the delay line for reconstruction of the originally recorded data signal. Prior to use in an application, the artificial neural network is trained via a training method with a known training set of corresponding simultaneously generated data and clock pairs. Training the network with data having such known clock (synchronization) signal enables extraction of the synchronization signal from its nonlinear properties hidden within its corresponding data.

Claims

1. A magnetic read channel comprising:

a magnetic read head for converting a recorded magnetic signal to an electronic signal;

a preamplifier for receiving and amplifying said electronic signal to produce an amplified electronic signal;

delaying and storing means for receiving said amplified electronic signal, delaying successive representations of the received signal, and storing the delayed signal representations;

an artificial neural network for receiving as input said delayed signal representations, classifying said input, and producing, based on said delayed signal representations, at least one reconstructed data signal and at least one reconstructed synchronization signal;

wherein said artificial neural network is trained via training means which train said artificial neural network utilizing known input data and corresponding pairs of training data and clock synchronization signals.

2. The magnetic read channel of claim 1, said artificial neural network comprising a plurality of neurons each having:

neuron input means for receiving a plurality of inputs,

weight assignment means for assigning weights to each of said plurality of inputs,

neuron computational means for performing simple computations, and neuron output means for producing at least one output signal.

3. The magnetic read channel of claim 2, said plurality of neurons each further comprising:

output error computational means for receiving an output signal produced by said neuron output means and a desired response signal, and computing an output error signal.

4. The magnetic read channel of claim 2, said plurality of neurons being arranged in a feedforward network topology having a set of input layer neurons, at least one set of hidden layer neurons, and a set of output layer neurons, wherein the neuron output means of at least one of the neurons in a given layer connects to the neuron input means of at least one of the neurons in the immediately subsequent layer.

5. The magnetic read channel of claim 2, being fully interconnected, wherein:

each stored delayed successive signal representation is coupled to said neuron input means of each input layer neuron;

each neuron output means from each of said input layer neurons is coupled to the neuron input means of each hidden layer neuron in the immediately subsequent hidden layer;

each neuron output means from each of said hidden layer neurons is coupled to the neuron input means of each hidden layer neuron in the immediately subsequent hidden layer, if said immediately subsequent hidden layer exists;

each neuron output means from each hidden layer neuron of the hidden layer immediately prior to the output layer is coupled to the neuron input means of each output layer neuron;

the neuron output means of at least one of said output layer neurons represents a final reconstructed data signal; and

the neuron output means of at least one of said output layer neurons represents a final reconstructed synchronization signal.

6. The magnetic read channel of claim 2, being sparsely connected, wherein:

each stored delayed successive signal representation is coupled to said neuron input means of each input layer neuron;

the neuron output means from at least one of said input layer neurons is coupled to the neuron input means of at least one hidden layer neuron in the immediately subsequent hidden layer;

the neuron output means from at least one of said hidden layer neurons is coupled to the neuron input means of at least one hidden layer neuron in the immediately subsequent hidden layer, if said immediately subsequent hidden layer exists;

the neuron output means from at least one hidden layer neuron of the hidden layer immediately prior to the output layer is coupled to the neuron input means of at least one output layer neuron;

the neuron output means of at least one of said output layer neurons represents a final reconstructed data signal; and

the neuron output means of at least one of said output layer neurons represents a final reconstructed synchronization signal.

7. The magnetic read channel of claim 1, said delaying and storing means comprising a delay line having a plurality of delay line taps, wherein each of said delay line taps contains a single representation of said amplified electronic signal.

8. The magnetic read channel of claim 1, said training means for training said artificial neural network comprising test data generation means and a training algorithm.

9. The magnetic read channel of claim 8, said training algorithm comprising backpropagation.

10. A magnetic read channel comprising:

a magnetic read head for converting a recorded magnetic signal to an electronic signal;

a preamplifier for receiving and amplifying said electronic signal to produce an amplified electronic signal;

delaying and storing means for receiving said amplified electronic signal to produce stored delayed successive signal representations;

an artificial neural network for receiving, detecting, filtering and reconstructing said stored delayed successive signal representations to produce at least one reconstructed data signal and at least one reconstructed synchronization signal, said artificial neural network comprising a plurality of neurons each comprising neuron input means for receiving a plurality of inputs, weight assignment means for assigning weights to each of said plurality of inputs, neuron computational means for performing simple computations, and neuron output means for producing at least one output signal, and said artificial neural network being interconnected wherein each stored delayed successive signal representation is coupled to the neuron input means of each input layer neuron, each neuron output means from at least a portion of said input layer neurons is coupled to at least a portion of the neuron input means of each hidden layer neuron in the immediately subsequent hidden layer, each neuron output means from at least a portion of said hidden layer neurons is coupled to the neuron input means of at least a portion of the hidden layer neurons in the immediately subsequent hidden layer if said immediately subsequent hidden layer exists, each neuron output means from at least a portion of the hidden layer neurons of the hidden layer immediately prior to the output layer is coupled to the neuron input means of at least a portion of the output layer neurons, and the neuron output means of at least one of said output layer neurons represents a final reconstructed data signal and at least one of said output layer neurons represents a final synchronization signal; and

wherein said artificial neural network is trained via training means comprising test data generation means to generate known input data and corresponding pairs of training data and clock synchronization signals and a training algorithm comprising backpropagation.