WO1998018128A1

WO1998018128A1 - Method and apparatus for multiplicative noise precompensation for magnetic recordings

Info

Publication number: WO1998018128A1
Application number: PCT/US1997/019201
Authority: WO
Inventors: Joseph Andrew O'sullivan; Ronald Scott Indeck; Marcel Wettstein Muller
Original assignee: Washington University
Priority date: 1996-10-23
Filing date: 1997-10-23
Publication date: 1998-04-30
Also published as: AU4916397A

Abstract

Electronics (40) implements the multiplicative noise model of the present invention. The diagnostic signal generator (50) processes the data signal s(t) to produce a diagnostic signal. The medium noise estimator (60) predicts repeatable noise components based on analysis of the diagnostic signal. Using the estimate of the medium noise, the precompensated signal generator (70) produces a corrected write signal.

Description

METHOD AND APPARATUS FOR MULTIPLICATIVE NOISE PRECOMPENSATION FOR MAGNETIC RECORDINGS

Government Rights To The Invention

This invention was made with government support under grants 9401697 and 9401698 awarded by the National Science Foundation. The government has certain rights in the invention.

Background And Summary of The Invention

The sources of noise in a readbac signal from a magnetic recording medium have been investigated and identified. One of those sources includes the irregu- larities and defects in the microstructure of the magnetic medium itself. For many years, the noise generated from this source has been thought, as with the noise generated from other identified sources, to be random and subject only to statistical analysis for its determina- tion. The inventors hereof have demonstrated that this noise component is instead deterministic, i.e. is permanent and repeatable, depending entirely on the transducer-medium position and on the magnetic history of the medium. As confirmed by experiments conducted by the inventors hereof, when the medium has had no signal written on it and has been recorded only with DC fields, the observed readback signals are almost identical . The magnetic contribution to the readback signal under these conditions results from spatial variations in the medium's magnetization: magnetic domains, ripple, local fluctuations of the anisotropy field and saturation magnetization. These local properties, in turn, are affected by the morphology and magnetic properties of the individual grains which make up the domain and which do not change after deposition. Hence, the noise from a nominally uniformly magnetized region measured at a fixed position on the magnetic medium is reproducible. As shown by the inventors hereof, a magnetic medium may be DC saturated and its output then measured to determine its remanent state or remanent noise. The inventors have confirmed that this remanent noise is a function of the magnetic microstructure by comparing the remanent noise after a positive DC saturation with the remanent noise after a negative DC saturation. It has been found that these waveforms are virtual "mirror images" of each other, thereby demonstrating a close correlation. Similarly, other methodologies were used to confirm that the remanent noise was determinative, repeatable, and related to the physical microstructure of the magnetic medium itself. Remanent noise arising from the permanent micro- structure exhibits identifiable features characteristic of that permanent microstructure after practically any magnetic history. See Spatial Noise Phenomena of Longi tudinal Magnetic Recording Media by Hoinville, Indeck and Muller, IEEE Transactions on Magnetics, Volume 28, No. 6, November 1992, the disclosure of which is incorporated herein by reference.

The inventive device and technique disclosed and claimed herein relies upon the discovery that the microscopic structure of the magnetic medium itself is a permanent random arrangement of microfeatures and therefore deterministic. In other words, once fabricated, the recording medium's physical microstructure remains fixed for all conventional recording processes. In particulate media, the position and orientation of each particle does not change within the binder for any application of magnetic field; in thin film media, the microcrystalline orientations and grain boundaries of the film remain stationary during the record and reproduce processes. It is the magnetization within each of these fixed microfeatures that can be rotated or modified to form the basis of the magnetic recording process. If a region of a magnetic medium is saturated in one direction by a large applied field, the remanent magnetization depends strongly on the microstructure of the medium. This remanent state is deterministic for any point on the recording surface .

Each particle or grain in the medium is hundreds to thousands of angstroms in dimension. Due to their size, a small region of the magnetic surface will contain a very large number of these physical entities. While the fabrication process may include efforts to align these entities, there is always some dispersion of individual orientations. The actual deviations will be unique to a region of the medium's surface making this orientation deterministic and making its corrupting effects susceptible to elimination. As can be appreciated by those skilled in the art, noise reduction enables increases in storage capacity and data rates, and eases the burden on transducers, magnetic media, and system design and fabrication.

Although this discovery has been made by the inventors hereof, techniques for reducing the noise itself based on this discovery have not been implemented. As this noise component of remanent noise is deterministic, it may be reliably repeated and measured at any particular point on a magnetic medium. Accordingly, the inventors have instead developed techniques which take advantage of this fact to record an information bearing signal that has been precompensated for the corrupting effects of the medium microstructure so that the recording may be read back later, free from corrupting noise, by conventional reading apparatus.

The device and method of the present invention can create uncorrupted prerecorded signals on a magnetic medium. To this end, a diagnostic signal is first written on the magnetic medium, the diagnostic signal is then read from the magnetic medium, and the readback signal is then compared with the original signal. The differences therebetween are determined to be noise, the greatest component of which is deterministic medium noise. A data signal desired to be recorded, which may or may not be the same as the diagnostic signal, is then compensated for the repeatable noise of the medium prior to being recorded back at the same location on the magnetic medium as was the diagnostic signal. Thus, after the compensated signal has been recorded onto the magnetic medium, any other readback or playback machine will produce a signal which has been compensated for remanent noise. The principles and methodologies of medium noise compensation as described above are applicable to both analog and digital magnetic recording as used in entertainment, instrumentation, and computer applications. However, special considerations apply to digital magnetic storage, in which saturation recording, including conventional longitudinal recording as well as more experimental perpendicular recording, is used virtually exclusively. The inventors among many others have demonstrated that some of the medium noise in digital saturation recording is signal dependent. This means that the effect of the magnetic medium's microstructural irregularities cannot be represented solely as additive noise, but must also comprise multiplicative noise effects. Accordingly, medium noise compensation techniques for digital recordings must be capable of diagnosing and addressing the effects of both additive and multiplicative noise.

The inventors have previously developed methodologies for precompensating data signals desired to be recorded for additive medium noise. These techniques are disclosed in U.S. Application No. 08/208,997 filed March 10, 1994, the disclosure of which is incorporated herein by reference . The inventors have now developed a model and technique, as disclosed and claimed herein, to precompensate data signals for multiplicative medium noise. The multiplicative noise model more accurately reflects the experimentally observed signal dependence of medium noise.

While the principal advantages and features of the invention have been described above, and a number of examples given, a greater understanding of the invention may be attained by referring to the drawings and the description of the preferred embodiment which follow. Brief Description of the Drawings

Figure 1 is a magnified representative depiction of the microscopic structure of a region of magnetic medium;

Figure 2 is a magnified depiction of several tracks of a magnetic medium having microscopic structure representatively shown thereon; Figure 3 is a block diagram of the write-read- write embodiment of the invention where the heads are moving to the right ;

Figure 4 is a schematic diagram of a magnetic recording system with signal -dependent medium noise; Figure 5 is a block diagram implementing the multiplicative noise model of the present invention;

Figure 6 is a block diagram of the electronics shown in Figure 3 ;

Figure 7 is a block diagram of a specific implementation of a medium noise estimator shown in Figure 6 ;

Figure 8 is a block diagram of a specific implementation of a precompensated signal generator shown in Figure 6 ; and Figure 9 is a block diagram of a specific implementation of the precompensated signal generator shown in Figure 8.

Detailed Description of the Preferred Embodiment

As shown in Figure 1, a region of magnetic medium 20 is built up with a plurality of microcrystalline structures 22 in a random pattern. This microcrystalline structure 22 is comprised of particles or grains varying from hundreds to thousands of angstroms in diameter. The view of Figure 1 is greatly enlarged and magnified in order to depict this physical phenomenon. As shown in Figure 2, this microcrystalline structure extends throughout the magnetic medium even though the magnetic medium 24 shown in Figure 2 may itself be comprised of tracks 26, 28, 30 as well known in the art.

The inventors hereof have designed a magnetic recording system and methodology that relies on a model for the magnetic recording channel which reflects the repeatability of the medium noise. In particular, two similar signals written at the same location, but at different times, yield highly correlated realizations of medium noise in the resulting readback waveforms. This results because the medium noise is caused in part by the magnetic microstructure which does not vary with time. As apparent to those skilled in the art, compensating for this medium noise will yield dramatic increases in system performance.

The inventors' methodology involves the steps of writing a diagnostic signal at a fixed location on the magnetic medium, reading the recorded signal, estimating the medium noise (or parameters in the medium noise) by comparing the read signal with the original signal to determine the differences therebetween, compensating a data signal desired to be recorded for the effects of the medium noise, and writing the precompensated data signal at the same location as was the diagnostic signal . Using this methodology, the magnetic medium will thus receive a recorded data signal which has been compensated for the remanent noise inherent therein. These compensated recordings may then be played back by any conventional playback device and produce a signal which more closely represents a recording of the original data signal on an ideal or noiseless magnetic medium. A preferred architecture for implementing this write-read-write methodology of the present invention is illustrated in Figure 3, where a plurality of conventional recording transducers 32, 34, 36 are shown moving over a magnetic medium 38. A diagnostic signal Silt) is first written onto the medium 38 with transducer 32, and the corrupted signal is then read from the medium by transducer 34. The readback signal, y_ι(t), is then used by electronics 40 to estimate the medium noise, and to generate a compensated write signal s₂(t) for recording by transducer 36 on the same portion of the medium upon which the diagnostic signal s_x(t) was previously written. Although only three transducers 32, 34, 36 are shown in Figure 3, it should be understood that a plurality of recording transducers of any number may just as easily be provided and, as taught herein, may be required in order to effect various aspects of the present invention. Because conventional transducers can be utilized, the teachings of the present invention are readily implemented using existing and available technology.

The functions performed by the electronics 40 shown generally in Figure 3 are based upon a mathematical model that accounts for the fact that medium noise has both signal and spatial dependence. It is signal dependent in the sense that its statistical properties at a particular place on the medium change with changes in the write signal . It is spatially dependent in the sense that the statistical properties depend on medium position. In digital saturation recording, where a binary digit is denoted by the presence or absence of a magnetization reversal -- called a transition -- in a bit cell, medium noise characteristics strongly depend on the number and proximity of consecutive transitions in the magnetization pattern. For example, medium noise for an isolated transition, a pair of closely spaced transitions (called a dibit) , and three closely spaced transitions (called a tribit) have different statistical properties.

Multiplicative transition noise can be decomposed into a number of modes. Experiments to date indicate that qualitatively there are three primary modes into which multiplicative transition noise can be decomposed: amplitude modulation, transition jitter (the spatial displacement of a transition from its intended location) , and pulse broadening. As an example of the spatial dependence in the presence of a signal, the inventors have experimentally demonstrated that, depending upon the location on the medium, the transition jitter in an isolated transition can have a non- zero mean value and can thus give rise to spatial dependence of medium noise. In general, medium noise can be modeled as a linear combination of the outputs of N linear systems _j ( • ) (j = 1 , 2 . . . N) , driven by the input to a magnetic recording channel. Weights of this linear combination are modeled to be random variables with their values depending on the particular place on the medium where the input signal is being written. Designating residual medium noise as n_r(t) , the total medium noise in the output signal, for input signal s(t), is given by:

N

(n_m * g)(t) = ∑θ_jLi(s)(t) + (n_r * g)(t) i=ι

Defining θ = [θ θ₂...^β _N]^τ and L(-) = [£,.(.), a(-) • • f. ^w« can write

(n_m * g)(t) = [θ^τL)(s)(t) + (n_r * g)(t).

The derivation of this model is more fully described in Joseph A. O'Sullivan et al . , Wri te-read-wri te signal precompensation techniques for magnetic recording, in Coding and Signal Processing for Information Storage, Mysore R. Raghuveer, Soheil A. Dianat, Steven W. McLaughlin, Martin Hassner, Editors, Proc . SPIE 2605, 39- 47 (1995) , the disclosure of which is incorporated herein by reference. The equation set forth above subsequently serves as the basis for design optimization of the write- read-write noise compensation device and methodology of the present invention.

A conventional magnetic recording system with signal dependent medium noise is illustrated in Figure 4. A write head 44 and a read head 46 (including any equalizer circuits that may be present) are modeled as linear time- invariant systems with impulse responses h(t) and g(t) , respectively. The random process n_d(t) represents independent signal and/or spatially dependent medium noise, and is represented by the medium noise model set forth above. A medium noise generator 48 produces n_d(t) as a function of the signal and the spatial location on the medium. A second independent noise component, n_x(t), accounts for head noise and any non- repeatable medium noise not represented by the model . An independent and random process w(t) represents additive electronic noise present in the read circuit.

According to the model, the writing of a similar waveform at two time instants and at the same physical location on the medium yields the same realization of the physical randomness in the medium noise. Thus, the first pair of write and read operations for the input signal s(t) in the write-read-write architecture produces an estimate of the medium noise for a particular signal/spatial combination. The final write signal s₂(t) can then be designed to compensate for the medium noise. Conceptually, this scheme can be thought of as using the recording channel twice, first to estimate the channel, and second to use it more efficiently. The mechanics and electronics of the system are synchronized, in a manner well-known in the art, to ensure that the third transducer 36 shown in Figure 3 writes the precompensated data signal s₂(t) at the same location at which the first transducer 32 wrote the diagnostic signal s_x(t).

The design criterion for the final write signal s₂(t) is established as follows: let n_dl(t) and n_d2(t) be the medium noise in the first and second writes . The difference between the initial and final writes is the compensating signal σ(t) = s₂ (t) - s^t). If the diagnostic signal s_α(t) is the actual data signal intended to be recorded, the design criterion is to select σ(t) to minimize the expected noise power in y₂(t) (the readback of the precompensated data signal s₂(t)) due to the medium noise and due to the compensation signal itself, given y-^t) . As derived in the earlier referenced article, the second write signal s₂(t) = s₁(t) + σ(t) can be generated as :

S₂(ω) = Sι(ω) + Σ(ω)

where the result of the first read, yχ(t), enters through the estimate θ, and where C (w) = L (w) (the Fourier transform of the vector 1 (t) representing the multiplicative noise components) divided by B (w) (the ideal channel response) .

A block diagram that implements this equation is shown in Figure 5, and uses the conditional mean θ of θ

Λ. given y_x(t), and the conditional covariance matrix Kg. This conditional mean θ depends on y_x(t) and under the Gaussian assumption equals the maximum a posteriori estimate of θ given y^t) . Standard techniques can be used to derive this estimate.

Under the Gaussian assumption and assuming the noise processes are stationary, the following equations

Λ A for θ and Kg may be easily derived:

° = J r—oomM) -Q>*_*ι)<t))dt, where the Fourier transform of ξ(t) is given by

E(ω) = K,L(ω)S(ω)

\S(ω)\*Lt(ω)K, L(ω) + \G(ω)\*S_nn(ω) + N₀ where S^ i w) is the power spectrum for the unrepeatable additive noise. The conditional covariance matrix is given by:

There may be various sources of multiplicative noise that can be included in the multiplicative noise model. Amplitude modulation, timing jitter, and pulse broadening are three effects of medium noise that fall into this category. When examined over short time durations, each of these effects may be captured through the use of one parameter. Over longer time durations, these parameters vary. The amplitude variation is modeled as a variable times the write signal, the jitter is modeled as a variable times the derivative of the write signal, and the pulse broadening is modeled as a variable times the time index times the derivative of the write signal. The write-read-write precompensation scheme for multiplicative noise is based on optimally compensating for these effects by estimating the parameters associated with each, then using these estimated parameters, along with measures of the accuracy of these estimates, in building a compensated write signal. Other sources of multiplicative noise may be compensated as well with appropriate circuits.

Figure 6 illustrates generally the electronics 40 shown in Figure 3, which implements the multiplicative noise model of the present invention. The electronics 40 includes a diagnostic signal generator 50 which processes, or passes through, the data signal s(t) desired to be recorded to produce or output the diagnostic signal s_x(t), a medium noise estimator 60 for estimating repeatable noise components of the magnetic medium based on the diagnostic signal s_x(t) and the readback y_x(t) of the diagnostic signal, and a precompensated signal generator 70 for compensating the data signal s(t) , based on the estimate of the medium noise, to produce the compensated write signal s₂(t) .

The details of the medium noise estimator 60 are shown in Figure 7. The diagnostic signal s_x(t) is provided to an ideal channel 62 having a convolution function b(t) that is equivalent to a write and read operation. The output of the ideal channel 62 is subtracted from the readback signal y_x(t) by an adder 64, and this difference is provided to a jitter estimator 66 and an amplitude estimator 68. The jitter estimator outputs a jitter compensation signal e₁ (t ) that estimates a factor by which the amplitude of the diagnostic signal was multiplied by the determinative medium noise. The amplitude estimator outputs an amplitude compensation signal θ₂(t) that estimates a factor by which the derivative of the diagnostic signal was multiplied by the determined medium noise. Although the medium noise estimator 60 shown in Figure 7 only estimates the transition jitter and amplitude modulation noise components of the medium noise, other or additional components can also be estimated as apparent to those skilled in the art, including the pulse broadening component, although perhaps with an increased system complexity and at a greater cost . As shown in Figure 8, the precompensated signal generator 70 comprises a feedback circuit employing several filters. The input data signal s (t) is provided to an adder 72 that subtracts several additional signals therefrom as described below. The output of the adder 72, which is the precompensated write signal s₂(t), is fed back to a jitter compensation filter 74a, an amplitude compensation filter 76a, and a signal shaping filter 78.

The jitter compensation filter 74a convolves the signal output of the adder 72 with a function c^t), which represents the transition jitter noise component obtained from the inverse Fourier transform of C (w) . The output of the jitter compensation filter 74a is provided to a multiplier 80 for multiplication with the jitter

A compensation signal θ_x(t) , and the output of the multiplier 80, which represents the jitter transition expected to be added to the input signal s(t) by the medium, is provided to the adder 72 for subtraction from the input data signal. Similarly, the amplitude compensation filter 76a convolves the signal output of the adder 72 with a function c₂(t), which represents the amplitude modulation noise component obtained from the inverse Fourier transform of C (w) . The output of the amplitude compensation filter 76a is provided to a multiplier 82 for multiplication with the amplitude compensation signal θ₂(t), and the output of the multiplier 82, which represents the amplitude modulation expected to be added to the input signal s(t) by the medium, is provided to the adder 72 for subtraction from the input data signal. In this manner, the input data signal s(t) can be appropriately modified so that the data signal s₂(t), once recorded on the medium and corrupted by the medium noise, more closely represents the input data signal s(t) than would the input data signal itself after being corrupted by the medium noise.

A A. If the compensation signals e₁ {t ) and θ₂ (t) precisely represented the transition jitter and amplitude modulation that will be added to the input data signal by the medium noise, the precompensated signal generator 70 as described thus far would be complete, and the signal shaping branch of the precompensated signal generator could be eliminated. Because the compensation signals θ₁(t) and θ₂(t) are merely estimates, however, and cannot perfectly estimate the multiplicative medium noise parameters due to the presence of additive medium noise, they will introduce at least some degree of error into the precompensation process. Thus, a signal shaping branch is provided in the precompensated signal filter 70 to quantitatively weigh the amount of compensation provided to the input signal based upon the level of certainty of the compensation scheme for particular frequency ranges of the signal . The signal shaping branch includes the signal shaping filter 78, which convolves the signal output of the adder 72 with a function c₃(t) obtained from the inverse Fourier transform of C (w) and the equation for the conditional covariance matrix given above. The output of the signal shaping filter 78 is provided to an adder 84, and the output of the adder 84 is provided to the adder 72 for subtraction from the input data signal s(t) . The output of the adder 84 is also provided to a jitter compensation filter 74b and to an amplitude compensation filter 76b. The jitter compensation filter 74b and the amplitude compensation filter 76b have the same time domain response c_x(t) and c₂(t) as the jitter compensation filter 74a and the amplitude compensation filter 76a, respectively. The output of the jitter compensation filter 74b is provided to a multiplier 90 for multiplication with the jitter compensation signal θ₁(t) , and the output of the multiplier 90 is provided to adder 84 for substraction from the output of the signal shaping filter 78. Similarly, the output of the amplitude compensation filter 76b is provided to a multiplier 92 for multiplication with the amplitude compensation signal θ₂(t), and the output of the multiplier 92 is provided to adder 84 for substraction from the output of the signal shaping filter 78. A specific implementation of the precompensated signal generator 70 is shown in Figure 9, where the time domain response c_x(t) of the jitter compensation filters 74a, 74b is a first derivative function, and the time domain response c₂(t) of the amplitude compensation filters 76a, 76b is represented by a straight wire with the filters 76a, 76b omitted. The time domain response c₃ (t) for the signal shaping filter 78 includes a proportional term Kl , a first derivative term times K2 , and a second derivative term times K3 , where Kl , K2 and K3 are determined from the equation for the conditional covariance matrix set forth above. As is well known in the art, it should be understood that although several of the filters shown in Figure 9 include derivative functions, direct implementation of these functions may not be desired, as some rolloff typically is desired at higher frequencies.

In the preferred embodiment of the invention, the diagnostic signal is the data signal desired to be recorded. However, this is not strictly necessary, as any diagnostic signal can be utilized from which the noise for the data signal desired to be recorded can be sufficiently estimated.

With the teachings of the present invention, both individually generated and mass produced recordings of every kind of digitally encoded information can be made for reading or playback by conventional devices already in the public's hands, including conventional computer diskette and digital tape drives. The resulting reduction in readback error rates can be exploited, as readily apparent to those of ordinary skill, to yield a corresponding increase in storage density and thus storage capacity of magnetic media.

There are various changes and modifications which may be made to the invention as would be apparent to those skilled in the art. However, these changes or modifications are included in the teaching of the disclosure, and it is intended that the invention be limited only by the scope of the claims appended hereto, and their equivalents.

Claims

What is claimed is:

1. A device for pre-compensating a data signal prior to its being recorded on a magnetic medium comprising a compensation circuit including a medium noise estimator for estimating a medium noise function in response to a read of a recorded diagnostic signal from said medium, and a pre-compensated signal generator connected to said medium noise estimator and to a source of said data signal for combining the output of the medium noise estimator and the data signal to produce a pre-compensated data signal.

2. The device of claim 1 wherein the medium noise estimator has a circuit for producing a jitter estimator signal and an amplitude estimator signal.

3. The device of claim 2 wherein the precompensated signal generator has a feedback circuit for combining the pre-compensated data signal with a jitter compensation function, an amplitude compensation function and a signal shaping function.

4. The device of claim 3 wherein the pre- compensation signal generator includes a second feedback circuit for combining a jitter compensation function and an amplitude compensation function with the said signal shaping function output.

5. The device of claim 1 further comprising a first transducer for writing said diagnostic signal on said medium, a read head for reading said medium to retrieve said diagnostic signal, and another write head for writing said pre-compensated data signal.

6. A method for producing a pre-compensated data signal comprising the steps of: reading a magnetic medium having recorded thereon a diagnostic signal; generating from said read diagnostic signal a medium noise estimation function; and combining said medium noise estimation function with a data signal to thereby produce said precompensated data signal .

7. The method of claim 6 further comprising the step of writing said diagnostic signal on said magnetic medium.

8. The method of claim 7 wherein the generating step includes the steps of : generating a jitter estimation signal, and generating an amplitude estimation signal.

9. The device of claim 8 wherein the combining step includes the step of : combining through a feedback circuit the precompensated data signal with a jitter compensation function, an amplitude compensation function, and a signal shaping function.

10. The device of claim 9 wherein the combining step includes the step of: combining through a second feedback circuit a jitter compensation function and an amplitude compensation function with said signal shaping function.