`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`_____________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`_____________
`
`LSI CORPORATION and AVAGO TECHNOLOGIES U.S., INC.,
`Petitioners,
`
`v.
`
`REGENTS OF THE UNIVERSITY OF MINNESOTA,
`Patent Owner.
`_____________
`
`Case No. IPR2017-01068
`Patent No. 5,859,601
`_____________
`
`
`DECLARATION OF STEVEN W. MCLAUGHLIN
`
`
`
`
`
`306538231 v2
`
`
`
`UMN EXHIBIT 2017
`LSI Corp. et al. v. Regents of Univ. of Minn.
`IPR2017-01068
`
`
`Page 1 of 86
`
`
`
`I, Steven W. McLaughlin, hereby declare and state, that all statements made
`
`herein of my own knowledge are true and that all statements made on information
`
`and belief are believed to be true. I am over the age of 21 years and I am
`
`competent to make this declaration. These statements were made with the
`
`knowledge that willful false statements are punishable by fine or imprisonment, or
`
`both, under Section 1001 of Title 18 of the United States Code.
`
`Executed this /'2- th day of July, 2020 in Atlanta, Georgia
`
`S~?enW.cLaughlin
`
`1
`
`
`Page 2 of 86
`
`
`
`1.
`
`I have been retained by counsel for Regents of the University of
`
`Minnesota (“UMN”) as a technical expert in connection with the inter partes
`
`review (“IPR”) proceeding identified above for U.S. Patent 5,859,601 (“the ‘601
`
`Patent”). I submit this declaration in support of UMN’s response to the petition.
`
`I.
`
`BACKGROUND AND QUALIFICATIONS
`
`2.
`
`I am and have been a faculty member of the School of Electrical and
`
`Computer Engineering (“ECE”) at the Georgia Institute of Technology (“Georgia
`
`Tech”) since 1996. Currently, I am the Steve W. Chaddick chair of the ECE
`
`School. I accepted the role of dean of the College of Engineering for Georgia Tech
`
`effective September 15, 2017. From 2007 to 2012, I was vice provost for
`
`International Initiatives, a position in which I provided oversight and strategic
`
`direction for Georgia Tech’s global engagement, education, and economic
`
`development initiatives. During that time, I also served as the Steven A. Denning
`
`Chair in Global Engagement. I was a Ken Byers Professor from 2005 to 2012 and
`
`previously was deputy director of Georgia Tech-Lorraine.
`
`3.
`
`I hold a Bachelor of Science in Electrical Engineering from
`
`Northwestern University, a Master of Science in Engineering from Princeton
`
`University, and a Ph.D. from the University of Michigan.
`
`4. My research interests include communications and information
`
`theory. I have published in the areas of coding and signal processing for wireless
`
`306538231 v2
`
`2
`
`
`Page 3 of 86
`
`
`
`communications, physical layer security, quantum key distribution, and data
`
`storage. I was a co-founder of Whisper Communications, which sought to
`
`commercialize physical layer security technologies. I have published more than
`
`250 papers and I am a named inventor of approximately 36 patents. Many of my
`
`papers and patents deal with coding techniques for optical and magnetic data
`
`storage devices. I also was a Principal Scientist for Calimetrics, where I and my
`
`team developed coding, signal processing and other technologies for optical disc
`
`recording systems such as CD, DVD and BluRay. I have served as the research
`
`and thesis advisor to more than 50 students at the bachelor’s, master’s, doctoral,
`
`and postdoctoral levels. I have been awarded: the Chevalier dans l’Ordre National
`
`du Mérite (Knight of the National Order of Merit) by the President of the Republic
`
`of France in 2011; the National Science Foundation Presidential Early Career
`
`Award for Scientists and Engineers (“PECASE”); an NSF Career Award and NSF
`
`Research Initiation Award; the Technical Achievement Award from the
`
`Information Storage Industries Consortium (“INSIC”); and the Georgia Tech
`
`Graduate Student Association “Friend of the Graduate Student Award.”
`
`My curriculum vitae is provided as Appendix A hereto.
`
`II. MATERIALS REVIEWED
`
`5.
`
`I considered information from various sources in forming my opinions
`
`in this declaration. In addition to drawing from over two decades of personal
`
`306538231 v2
`
`3
`
`
`Page 4 of 86
`
`
`
`experience in the field of data storage and coding, I have also reviewed the IPR
`
`Petition and its Exhibits, including the ‘601 Patent (Ex. 1001), the Okada patent
`
`(Ex. 1007), the Soljanin Declaration (Ex. 1010) and the tables (Ex. 1011). I also
`
`reviewed the deposition transcript for Prof. Soljanin (Ex. 2011) and the exhibits
`
`used in that deposition, including Exhibits 2007 to 2010. I also reviewed UMN’s
`
`response filed herewith in detail and I agree with its analysis and conclusions about
`
`alleged anticipation by the Okada patent. I also reviewed the contemporaneously
`
`filed Declaration from Prof. Jaekyun Moon (Ex. 2016) and agree with the technical
`
`descriptions therein about magnetic recording and the operation of MTR codes.
`
`III. SUMMARY OF THE ‘601 PATENT
`
`6.
`
`The ‘601 Patent relates to digital storage systems, and in particular
`
`magnetic data storage systems. ‘601 Patent (Ex. 1001) at col. 1:9-10 (“The present
`
`invention relates in general to digital storage systems.”) and col. 2:59-61.
`
`A. HDD Basics
`
`7.
`
`Hard disk drives (“HDDs”) are a type of magnetic data storage
`
`system. The write head of a combined “read/write” head in the HDD writes data in
`
`concentric recording tracks to a circular magnetic disk in the HDD by magnetizing
`
`microscopic “bit regions” along the respective tracks on the disk. Later, when the
`
`previously-written data are read, the read head reads the data by detecting changes
`
`in the magnetic fields emanating from the big regions along a track. The read head
`
`306538231 v2
`
`4
`
`
`Page 5 of 86
`
`
`
`generates a “readback signal” that has fluctuations due to the reversals in the
`
`magnetic fields from the bit regions and a so-called “read channel” in the HDD
`
`detects the data written to the disk from the “readback signal” generated by the
`
`read head using signal processing techniques.
`
`8.
`
`Each bit region on the magnetic recording layer of the disk has a
`
`magnetic polarization that, once written by the write head, is oriented in a
`
`particular direction. The magnetic polarity of these regions can be changed from
`
`one direction to its opposite by the write head, by varying the polarity of the
`
`magnetic field emitted by the write head, in order to write the data to the disk. At
`
`the time of the ‘601 Patent, so-called “longitudinal magnetic recording” (“LMR”)
`
`was prevalent. In LMR, the bit regions are polarized in the same plane as the
`
`magnetic layer as illustrated in the diagram below.1
`
`
`1 From Mark Fischetti, “Going Vertical,” Scientific American, 2006 (Ex. 2023). In
`
`the early-to-mid 2000s, the HDD industry migrated to so-called “perpendicular
`
`magnetic recording” (“PMR”), where the bit regions are polarized perpendicular to
`
`the plane of the magnetic layer.
`
`306538231 v2
`
`5
`
`
`Page 6 of 86
`
`
`
`
`
`9.
`
`As shown in the above diagram, the polarized regions can be
`
`conceptualized as bar magnets having north (N) and south (S) poles. As such,
`
`there are four possibilities for two adjacent polarized regions―two where the
`
`magnetic orientation switches direction and two where they do not―as shown
`
`below.
`
`No Change in Magnetic Orientation
`Between Adjacent Bit Regions
`
`A Change in Magnetic Orientation
`Between Adjacent Bit Regions
`
`
`
`
`
`
`
`
`
`
`
`10.
`
`In magnetic recording, a change or reversal in the magnetic
`
`orientation between two adjacent regions is called a “transition.” The information
`
`to be stored (binary 0’s and 1’s) is recorded in the sequence of transitions and non-
`
`transitions recorded to the disk. So the ability to accurately store and read
`
`306538231 v2
`
`6
`
`
`Page 7 of 86
`
`
`
`transitions is critical to the performance of the HDD. The read head detects the
`
`transitions and the read channel determines the data written to the disk based on
`
`the detected transitions.
`
`11. The transition type shown at the top right of the chart above can be
`
`referred to as a “positive transition” and the transition type shown at the bottom
`
`right of the chart can be referred to as a “negative transition.” Note that the next
`
`transition after a positive transition has to be a negative transition and vice versa.
`
`A positive transition cannot immediately follow a positive transition and a negative
`
`transition cannot immediately follow a negative transition.
`
`12. To write the data to the disk, HDDs typically use a use a Non-Return-
`
`to-Zero Inverted (“NRZI”) system. With NRZI, at every bit region, if the data bit
`
`to be written to the disk is a “1,” the write head reverses the magnetization of the
`
`bit region on the disk being written relative to the prior, adjacent bit region,
`
`creating a magnetic transition (positive or negative). Conversely, if the data bit to
`
`be written is a “0,” the write head does not reverse the magnetization of the bit
`
`region relative to the prior region. Thus, transitions in the magnetic recording
`
`layer represent 1’s in the data sequence to be written with NRZI recording.
`
`Another known recording format (described in the ‘601 Patent) is Non-Return-to-
`
`Zero (NRZ), where a binary “1” represents a positive level in the magnetization
`
`306538231 v2
`
`7
`
`
`Page 8 of 86
`
`
`
`waveform and the binary “0” represents a negative level of the waveform. Ex.
`
`1001, col. 1:24-27.
`
`13.
`
`In most modern HDDs, the user data are modified in at least two ways
`
`prior to writing the data to the disk. First, the user data are encoded according to
`
`applicable modulation codes. A modulation code maps data bits into an encoded
`
`sequence with characteristics according to a constraint in the encoded sequence for
`
`various purposes to enable its recovery by the detector. Second, a processor
`
`converts the encoded sequences into a waveform that the write head records to the
`
`disk by magnetically polarizing the bit regions on the disk in accordance with the
`
`waveform.
`
`B.
`
`Sequence Detectors in Magnetic Recording
`
`14. To read the written data, the read head, flying above the rotating disk,
`
`detects the variations in the magnetic flux that are stored on the medium and
`
`converts the sensed magnetic fields into a continuous, analog electrical signal
`
`called the “readback signal.” To “read” written data, the read channel hardware,
`
`circuitry and algorithms process the readback signal by detecting the sequence of
`
`transitions (including the types of transitions, positive or negative) and non-
`
`transitions.
`
`15. Early read channels used “peak” or “threshold” detectors. If the
`
`readback signal, after rectification, was beyond a threshold voltage value, a
`
`306538231 v2
`
`8
`
`
`Page 9 of 86
`
`
`
`transition was detected. Ex. 2024 (S. Wang and A. Taratorin, Magnetic
`
`Information Storage Technology, Academic Press 1999) at pp. 345-347. Peak
`
`detection works adequately only at low data densities when the intersymbol
`
`interference (“ISI”) effects are small. See Ex. 2024 at p. 361. ISI is the effect
`
`when the pulse in the readback signal from one bit region interferes with the pulses
`
`from adjacent bit regions, thereby often reducing the signal strength associated
`
`with a transition, which can make the written sequence more difficult to detect. As
`
`data density is increased by packing more data into smaller regions, the transitions
`
`in the magnetic media become closer and the readback signal responses from them
`
`start to interfere with each other. As HDD manufacturers continued to increase the
`
`data density to create smaller, more powerful devices, the ISI became too great for
`
`a peak detector to accurately detect the data. Consequently, peak detectors do not
`
`work adequately at the high data density levels required by modern computing
`
`devices. Id.
`
`16. To address the problems associated with peak detectors, beginning in
`
`the early 1990s, the HDD industry migrated to more sophisticated detectors, in
`
`particular so-called “partial response maximum likelihood” or “PRML” detectors,
`
`such as Viterbi detectors. PRML detectors are digital sequence detectors that use
`
`digitized samples of the readback signal and some form of “maximum likelihood
`
`sequence detector” (“MLSD”) to both model the ISI that occurs in the readback
`
`306538231 v2
`
`9
`
`
`Page 10 of 86
`
`
`
`signal and estimate the most likely sequence of data written to the disk based on
`
`digitized samples of the readback signal. Mathematically, MLSDs maximize the
`
`conditional probability density p(y│x) across all choices of stored sequences x
`
`given the output y from the read head and select the most likely x as the decoded
`
`pattern. Ex. 2036 (B. Vasic et al., “Read Channels for Hard Drives,” Chap. 15,
`
`Coding and Signal Processing for Magnetic Recording Systems, Vasic et al. eds.,
`
`CRC Press, 2005) at p. 15-4.
`
`17. Reading the data with a MLSD, however, is not simple, in part
`
`because the number of possible stored sequences x is extremely large.
`
`Furthermore, noise and misequalization cause the readback signal samples to
`
`deviate from their expected, noiseless values. The Viterbi algorithm, for example,
`
`is one approach that considers various bit sequences and compares the
`
`corresponding expected (noise-free) partial response channel output value with the
`
`readback signal sample values. For Gaussian noise, maximizing p(y│x) is
`
`equivalent to choosing as the possible noise-free (also sometimes referred to as the
`
`expected or “target”) read channel output y sequence that is closest in distance
`
`(e.g., squared Euclidean distance) to the readback signal sample values y. Id.
`
`18. The maximization described in the prior paragraph can be
`
`accomplished with a Viterbi detector that performs a trellis-based search over a
`
`finite window (e.g., 40 readback signal samples). Below is an example of a trellis
`
`306538231 v2
`
`10
`
`
`Page 11 of 86
`
`
`
`having 8 nodes at each time instance. The 8 nodes correspond to the 8 different 3-
`
`bit sequences (e.g., 000, 001, …, 111) that represent a 3-bit history of the sequence
`
`x written to the disk (assuming a NRZI recording format for this explanation). In
`
`this example, I used a notation whereby a nontransition is indicated by a 0 and a
`
`transition in the magnetic state of the recording layer of the disk from one bit
`
`region to the next (whether a positive or negative transition) is indicated by a 1.
`
`Thus, the first node represents a sequence of nontransition-nontransition-
`
`nontransition (or 000) as the ‘history’ of input NRZI sequence. Note that there are
`
`two “branches” leaving each node, representing the fact that the next bit to be
`
`stored has to be either a ‘0’ or a ‘1’ (i.e., a non-transition or a transition). Thus,
`
`there are two branches leaving the 000 node to either (i) the node 000, indicating
`
`that the next region was a nontransition or (ii) the node 001, indicating that the
`
`next region is a transition.
`
`306538231 v2
`
`11
`
`
`Page 12 of 86
`
`
`
`
`
`19. The trellis representation shown above does not explicitly indicate
`
`whether the transitions (i.e., “1”s) are negative or positive, so the Viterbi detector
`
`can continuously update the polarity of the last transition (since a positive
`
`transition has to follow a negative transition and vice versa) for determining the
`
`appropriate target value for the branch (described further below).
`
`20. With the above trellis in mind, there is a one-to-one correspondence
`
`between possible sequences written to the disk and “paths” through the trellis. For
`
`example, a path is made up of a series of branches end-to-end and also corresponds
`
`to a potential sequence of transitions (1’s) and nontransitions (0’s) written to the
`
`disk. The red path above, for example, starts at the 001 node, meaning that written
`
`pattern is 001 up to that point, and correspondingly the first three bits in the branch
`12
`
`306538231 v2
`
`
`Page 13 of 86
`
`
`
`of the path are 0-0-1. The next node is 010, indicating that the next (4th) bit
`
`written to the disk is a 0 (i.e., the last bit represented by the 010 node). The next
`
`node is 101, indicating that the next (5th) bit written to the disk is a 1 (i.e., the last
`
`bit represented by the 101 node). And the final node is 010, indicating the final
`
`(6th) bit written to the disk is a 0 (i.e., the last bit represented by the 010 node).
`
`Thus, the red path above represents the 6-bit sequence 0-0-1-0-1-0.
`
`21. The paths also correspond to sequences of transitions and
`
`nontransitions on the magnetic disk. Using the example “bar” magnet notation
`
`above for LMR, the binary sequence 0-0-1-0-1-0 for the red path through the trellis
`
`above might correspond to the sequence of magnetized bit regions recorded on the
`
`disk shown below. The diagram below shows how the three branches shown in the
`
`trellis correspond to transitions and nontransitions of the recorded sequence.2
`
`
`2 The branches that correspond to the first three bits of the sequence in this diagram
`
`are not shown, but they have to be a sequence of nontransition-nontransition-
`
`transition, since the red path is assumed to start at the 001 node.
`
`306538231 v2
`
`13
`
`
`Page 14 of 86
`
`
`
`
`
`22. Since the trellis represents the sequence of NRZI inputs, it also
`
`represents the sequence of noise-free target channel outputs. The function of the
`
`Viterbi detector is to determine which path of noise-free target channel outputs
`
`through the trellis most closely matches the readback signal sample sequence y.
`
`To perform this function, the Viterbi detector computes so-called “branch metric
`
`values” for branches of the trellis, and sums the branch metric values for the
`
`branches along various paths through the trellis to compute a so-called “path
`
`metric value.” The branch metric value for a branch is the measure of the distance
`
`(or error) between the signal sample value(s) of the readback signal and the
`
`expected (or “target”) channel output value(s) for the branch. The target values
`
`(which could be, for example, -1, 0 and +1) depend on whether the branch
`
`represents negative transition, a nontransition, or a positive transition respectively
`
`306538231 v2
`
`14
`
`
`Page 15 of 86
`
`
`
`(which is why the Viterbi detector continuously tracks the polarity of the last
`
`transition). The target values of -1, 0 and +1 are generally for situations where
`
`there is little or no ISI. For higher densities with larger ISI, more sophisticated
`
`targets are often used.
`
`23. The squared Euclidean distance function is an example of a branch
`
`metric function that can compute the distance, i.e.:
`
`
`
`where BMV is the branch metric value (or distance), y is the readback signal
`
`sample value, and m is the target (or expected value) for the branch. The “path
`
`metric value” for a given path is the sum of the branch metric values for the
`
`branches along the path, so a path metric value can be thought of as the cumulative
`
`distance (that is, error) between the error-free target and the readback signal for the
`
`sequence of branches along a particular path. The sequence corresponding to the
`
`path with the best (lowest) path metric value is the detector’s determination of the
`
`most likely sequence of transitions/nontransitions written to the disk. Some
`
`Viterbi detectors compute branch metric values for all branches, while others
`
`compute branch metric values for some of the branches because the other branches
`
`are removed or “pruned” from the trellis to assist in limiting the number of
`
`allowable transitions.
`
`306538231 v2
`
`15
`
`
`Page 16 of 86
`
`
`
`24.
`
` The operation of sequence detectors is, therefore, very different from
`
`that of peak (or threshold) detectors.
`
`25. Also, noise in the magnetic recording system, which manifests itself
`
`in the readback signal, complicates the read process. There are several sources of
`
`noise in a HDD, but a major and growing complicating noise source in a high-
`
`density HDD is “media noise.” Media noise includes noise in the readback signal
`
`that arises from fluctuations in the medium magnetization due to such things as the
`
`differing polarities of transitions written to the disk and the specific number and
`
`density of transitions in a given region of the disk. Media noise increases with data
`
`density, which is the overwhelming trend in the HDD industry.
`
`26. Media noise arises from transitions because straight, perfect
`
`transitions between adjacent regions of oppositely polarized grains are not
`
`physically possible. Instead, the physics of energy minimization and the granular
`
`structure of the thin magnetic film used on modern HDD disks result in transitions
`
`that are jagged. These jagged transitions affect the location and shape of the
`
`positive and negative peaks in the readback signal from the read head. The degree
`
`and extent of the jaggedness (and hence the peak shift and shape of the readback
`
`signal) depends upon the sequence of symbols written to the disk. For example,
`
`sequences with more transitions will have less transition noise than sequences
`
`without as many transitions.
`
`306538231 v2
`
`16
`
`
`Page 17 of 86
`
`
`
`C. The Problem Addressed by the ‘601 Patent
`
`27. One problem experienced with sequence detectors is that there are
`
`write patterns (i.e., recorded NRZI input sequences) whose corresponding path
`
`metrics for many possible readback signals are “close” in distance. That is, there
`
`are certain recorded NRZI patterns that often result in readback signals having
`
`small numerical difference between the path metric values. This means that there
`
`are data sequences whose corresponding trellis paths are “close” numerically,
`
`which can be a dominant source of errors in decoding. Since the sequence detector
`
`chooses the path with the lowest path metric value as the most likely sequence of
`
`transitions/nontransitions recording to the magnetic recording layer, other trellis
`
`paths that have small numerical distances from the selected path are the most likely
`
`error paths. That is, small amounts of noise might cause the detector to select the
`
`path with a small numerical difference from the ideal path. These “error paths” or
`
`“error patterns” are the most likely errors to occur during the detection process.
`
`The error rate of a sequence detector can be too high (e.g., above a desired level) if
`
`the detector consistently selects error patterns associated with these “close” trellis
`
`paths and sequences. See Ex. 1001, col. 18-26 (“Errors in sequence detectors arise
`
`mostly from difficulty in distinguishing minimum distance patterns … The
`
`minimum distance patterns are those patterns corresponding to different decisions
`
`that have the minimum Euclidean distance from one another.”).
`
`306538231 v2
`
`17
`
`
`Page 18 of 86
`
`
`
`28. Figure 1 of the ‘601 Patent shows four exemplary write pattern pairs
`
`that tend to cause sequence detection errors. The four pairs are shown below.
`
`Each of the patterns has at least two consecutive transitions in the stored data
`
`pattern when written to the magnetic medium (e.g., disk). And the patterns for
`
`each pair differ in the middle three pulses; the middle three pulses are opposite for
`
`each pair.
`
`Pairs of Written Patterns from Figure 1
`of the ‘601 Patent
`
`
`
`Middle 3 pulses high-low-high
`
`Middle 3 pulses low-high-low
`
`Middle 3 pulses low-high-low
`
`Middle 3 pulses high-low-high
`
`Middle 3 pulses high-low-high
`
`Middle 3 pulses low-high-low
`
`Middle 3 pulses low-high-low
`
`Middle 3 pulses high-low-high
`
`
`
`
`
`
`
`
`
`
`
`29. As an example, the upper pattern in Pair 1 above represents a
`
`sequence of transition-transition-transition-nontransition (or T-T-T-NT), as shown
`
`below, since the signal goes from low to high, then high to low, then low to high,
`
`then stays high.
`
`306538231 v2
`
`18
`
`
`Page 19 of 86
`
`
`
`The lower pattern in Pair 1 represents a sequence of nontransition-transition-
`
`transition-transition (or NT-T-T-T), as shown below.
`
`
`
`
`
`
`30. The path metric values for each of the sequence pairs in Figure 1 of
`
`the ‘601 Patent will very often be close numerically for low and moderate noise
`
`cases. See Ex. 1001 at col. 3:53-60 (explaining that these write patterns “cause
`
`most errors in sequence detection”). In cases where the noise is sufficiently large,
`
`the sequence detector will make errors distinguishing the two patterns in each pair.
`
`31. One prior art way to partially address this problem are so-called
`
`runlength limited (“RLL”) codes. RLL codes place rigid constraints on the
`
`sequences of NRZI bits that are recorded on the disk. RLL codes have d and k
`
`constraints where d and k are, respectively, the minimum and maximum number of
`
`0s between consecutive 1s, where a 0 represents a non-transition and 1 represents a
`
`transition, respectively. RLL(d=1,k) codes were common at the time of the ‘601
`
`306538231 v2
`
`19
`
`
`Page 20 of 86
`
`
`
`Patent. Ex. 1001, col. 1:21-42. Such codes require at least one nontransition
`
`between transitions since d=1.
`
`32. With such a RLL code, many NRZI write patterns cannot be used
`
`(i.e., all write patterns with two consecutive transitions). For example, RLL(1,k)
`
`codes prohibit all the write patterns in Figure 1 of the ‘601 Patent because those
`
`write patterns all have at least two consecutive transitions. Eliminating all of the
`
`patterns with two consecutive transitions has the effect of improving the
`
`performance of the detector; its error rate improves, or equivalently, it is able to
`
`tolerate a higher level of noise. This benefit is referred to as “coding gain.”
`
`However, the coding gain from RLL(1,k) codes results in an unacceptably high
`
`“rate penalty.”
`
`33. To explain “rate penalty,” I will first explain “code rate” or
`
`“information rate” of an error correction code. The “code rate” is the proportion of
`
`the data stream that is useful (non-redundant). In many contexts, including MTR
`
`codes, the code rate can be computed as the ratio of the dataword bit size (m) to the
`
`codeword bit size (n), in other words m÷n (or m/n). A high code rate is more
`
`efficient because more real, raw data is written on the disk. Conversely, a low
`
`code rate reduces the total amount of data stored on the disk. The “efficiency” of
`
`the code is the ratio of the rate to the capacity for the channel, i.e., the upper bound
`
`on the rate at which information can be reliably transmitted over the channel. Ex.
`
`306538231 v2
`
`20
`
`
`Page 21 of 86
`
`
`
`1001, col. 5:8-10. The “rate penalty” is the difference between the capacity and
`
`the code rate. For example, if a code has a rate of 0.7500 and an efficiency of
`
`0.8640, the channel has a capacity of 0.8681 (computed as 0.7500/0.8640). And
`
`the rate penalty is 0.8681 minus 0.7500, or 0.1181 in this example.
`
`34. The rate penalty for RLL(1,k) codes that eliminate back-to-back
`
`transitions is too great to be used in a modern, high density HDD read channel.
`
`See Ex. 1001 at col. 4:15-22. In other words, the cost in data density resulting
`
`from inserting a non-data-representing “0” between each transition (since d=1)
`
`outweighs the gain from being able to more accurately read the data written to the
`
`disk. Further, when attempts were made to compensate for the large rate penalty
`
`of such a RLL code, the proposed solutions resulted in increased noise in the
`
`system which had decreased the accuracy of the reading process. Ex. 1001, col.
`
`22-24. The challenge that needed to be addressed was meeting the consecutive
`
`transition constraint with both high efficiency and good coding gain.
`
`D. The Solution of the ‘601 Patent
`
`35. To combat the problem of minimum distance errors with both high
`
`efficiency and good coding gain, the ‘601 Patent proposes a coding scheme that
`
`eliminates more than j consecutive transitions, where j > 2. For example, if j = 2,
`
`then write patterns with back-to-back transitions are permitted (unlike RLL (1,k)
`
`codes), but patterns with more than two consecutive transitions are prohibited. In
`
`306538231 v2
`
`21
`
`
`Page 22 of 86
`
`
`
`addition, the coding scheme of the ‘601 Patent simultaneously imposes an RLL k
`
`constraint on the maximum number of nontransitions (k) between transitions. Ex.
`
`1001, col. 4:46-48. The ‘601 Patent refers to codes that have the j and k constraints
`
`as “Maximum Transition Run” or “MTR” codes. As such, a MTR encoder
`
`receives a dataword of m bits in length and encodes it into an n-bit codeword,
`
`where m < n, and subject to the j and k constraints when the codewords are written
`
`to the disk. Id. at col. 4:46-60.
`
`36. The beauty of MTR codes is that the constraint on the number of
`
`consecutive transitions can be achieved with a relatively low rate penalty. Equally
`
`important is that MTR codes eliminate the “close” error patterns that I described
`
`previously (see Figure 1 of the ‘601 Patent). This means that the detector can
`
`withstand more noise and/or higher data density for relatively small rate penalty.
`
`For example, coding gains of several dB can be achieved (Ex. 1001, col. 7:38) with
`
`high rate codes. Ex. 1001, Figs. 5A-5D (showing the rates for various MTR
`
`codes).
`
`37. Figure 6 of the ‘601 Patent shows an example of a 4/5 MTR (2,8)
`
`code. In this example, 4-bit datawords are encoded into 5-bit codewords (a rate
`
`4/5=0.800 code), where the codewords, when written to the magnetic medium,
`
`satisfy the j=2 and k=8 constraints. That is, assuming the NRZI recording format,
`
`(1) there are no more than two consecutive 1s (consecutive magnetic transitions) in
`
`306538231 v2
`
`22
`
`
`Page 23 of 86
`
`
`
`the codewords when written to the magnetic medium and (2) the maximum
`
`number of consecutive 0s when the codewords are written to the disk is eight
`
`(when codeword 10000 precedes codeword 00001).
`
`
`
`
`
`
`
`306538231 v2
`
`23
`
`
`Page 24 of 86
`
`
`
`38. The 4/5 MTR (2,8) code has a rate of 0.800 and an efficiency of
`
`0.9133. See Fig. 5A (excerpt shown below).
`
`
`IV. CLAIM CONSTRUCTION AND PERSON OF ORDINARY SKILL IN
`THE ART
`39.
`I have been informed and understand that patent claims are construed
`
`in accordance with the ordinary and customary meaning of such claims as
`
`understood by one of ordinary skill in the art and the prosecution history pertaining
`
`to the patent.
`
`40. Counsel has advised me that to determine the appropriate skill level of
`
`one skilled in the art, I may consider the following factors: (a) the types of
`
`problems encountered by those working in the field and prior art solutions thereto;
`
`(b) the sophistication of the technology in question, and the rapidity with which
`
`innovations occur in the field; (c) the educational level of active workers in the
`
`field; and (d) the educational level of the inventor. I considered those factors, and
`
`306538231 v2
`
`24
`
`
`Page 25 of 86
`
`
`
`also considered the engineers that I worked with back in 1995 on data storage and
`
`coding issues and projects.
`
`41. The relevant technology field for the ‘601 Patent is data storage and,
`
`more particularly, coding techniques for digital sequence detectors. Ex. 1001, col.
`
`1:9-12. Based on this, and the factors described above, it is my opinion that a
`
`PHOSITA in the field to which the ‘601 Patent pertains would be someone
`
`working in the electrical engineering field and specializing in data coding and
`
`detection techniques used in connection with reading data from various storage
`
`media such as hard disk drives and optical media. Such a person would have had
`
`at least a bachelor’s degree in electrical engineering with three or more years of
`
`work experience in the industry. Accordingly, such as person would have studied
`
`and been familiar with traditional data coding and detection techniques and devices
`
`including RLL codes, peak detectors, and sequence detectors, such as Viterbi
`
`detectors.
`
`42. Claim 13 of the ‘601 Patent recites “generating no more than j
`
`consecutive transitions of said sequence in the recorded waveform such that j > 2”
`
`and “generating no more than k consecutive sample periods of said sequences
`
`without a transition in the recorded waveform.” These are the so-called j and k
`
`constraints of the claimed MTR codes.
`
`306538231 v2
`
`25
`
`
`Page 26 of 86
`
`
`
`43.
`
`In my opinion, one skilled in the art would understand the term
`
`“transition” in claim 13 to mean a change, and in particular a reversal, in the
`
`magnetic orientation of adjacent bit regions alo