Computer Peripherals

School of Computer Engineering

Nanyang Technological University
Singapore

These notes are part of a 3rd year undergraduate course called "Computer Peripherals", taught at Nanyang Technological University
School of Computer Engineering in Singapore, and developed by Associate Professor Kwoh Chee Keong. The course covered
various topics relevant to modern computers (at that time), such as displays, buses, printers, keyboards, storage devices etc... The
course is no longer running, but these notes have been provided courtesy of him although the material has been compiled from
various sources and various people. I do not claim any copyright or ownership of this work; third parties downloading the material
agree to not assert any copyright on the material. If you use this for any commercial purpose, I hope you would remember where you
found it.
Further reading is suggested at the end of each chapter, however you are recommended to consider a much more modern alternative
reference text as follows:

Computer Architecture: an embedded approach

Ian McLoughlin
McGraw-Hill 2011

Chapter 7.

Magnetic Recording Fundamentals

Computer memories are divided into two types. Main or working memories including
RAM's, ROM's and other semiconductor devices which are directly addressable by the CPU,
and mass storage devices. Mass storage memories generally are not directly addressable by the
CPU, have a slower access method, are non-volatile and often removable off-line and have a
much lower cost per bit. Mass storage devices may be classified into magnetic hard disks and
floppy disks, several types of magnetic tape drives, and various kinds of optical disk drives. In
additional, there are the many varieties of card storage devices like magnetic cards, smart
cards, ROM, RAM and flash memory cards, but these will not be discussed here. Except for
the CD-ROM in which data is represented by physical indentations or pits, practically all other
write-able mass storage devices make use of the magnetic or magneto-optical (M-O)
characteristics of the recording media to store the required information. Thus we shall first
consider some basic principles used in magnetic recording and the performance of the digital
recording channel.

7.1 Magnetic Material Characteristics

When a piece of magnetic material is moved past a magnetic field, usually created by an
electromagnet, it becomes magnetised. Similarly, when the magnetised material is moved past
an unenergised coil, it induces a voltage across the coil. This is just Faraday's law of
electromagnetic induction, which relates the voltage induced to the magnetic field strength. A
number of parameters are used to characterise magnetic materials:
(a)

Coercivity Hc is the measurement the level of difficulty to magnetise the

material. For storage media, a high coercivity is desired in order that
information stored will be preserved in the presence of stray magnetic fields
that may be present. High coercivity also implies that a strong magnetic field
is needed to record information onto it. Magnets with high coercivity are
called hard magnets.

(b)

Remanence Br is the amount of magnetisation that remains after the

magnetic field is removed. Soft iron used for electromagnets are chosen to
have low remanence so that it will respond efficiently to the applied
electromagnetic field.

(c)

Magnetic domains are small regions in the magnetic media which may be
magnetised independently of adjacent regions, so that adjacent domains can
have opposite polarities. The size or granularity of these domains have an
important bearing on the density of information that can be stored.

(d)

Flux reversal occurs when a change in polarity is encountered while moving

from one domain to the next. The storage density of the media is measured
by the flux reversal per inch (frpi) or the flux change per inch (fcpi).

[Link]

Magnetic Recording Fundamentals

Figure 0-1 shows the B-H characterisation curves for soft and hard magnetic materials.
B

B
Br

Br

H
Hc

SOFT MAGNET
Hc < 10 Oe

Hc

HARD MAGNET
H c > 100 Oe

Figure 0-1. B-H magnetisation curves

7.2 Read/Write head

The "guts" of a magnetic recording system are: the write head, the magnetic medium,
and the read head. (The write head could be the same as the read head and usually has been
the case for disk drives.) The write head is driven by a current source that carries the
information to be stored. The write head radiates flux, which changes the state of magnetization of the magnetic medium immediately under the head. Actually, since the head is
moving with respect to the magnetic medium, any point on the magnetic medium retains the
state of magnetization corresponding to the last flux it experienced from the write head as the
head moves away from that point.
On a rigid disk, the disk moves in a circular motion under the head. Information is
stored on the disk in concentric tracks, the width of a track roughly being governed by the size
of the write head. The density of recording per sq inch (known as areal density) is the product
of the number of tracks per inch (tpi) and the linear density of information along a track
measured in bits per inch (bpi). Typical numbers for today's high end (i.e., expensive) rigid
disk drives are: 3,000 tpi and 30,000 bpi.
The current into the write head induces a magnetization pattern on the track immediately
below the write head. When a track is to be read, a read head is positioned over the track.
Then, the magnetization pattern "frozen" on that track radiates flux that is sensed, or "read,"
by the read head. The read head produces a voltage that is symptomatic of the magnetization
on the track being read. There are primarily two types of read head: inductive heads which
contain coils of very fine wire and which produce a voltage proportional to the time derivative
of the flux that passes through its coils, and magneto-resistive (MR) heads which produce a
voltage directly proportional to the flux sensed by the head. MR heads produce larger read
voltages than inductive heads, but have a limited dynamic range for linear operation. Only
inductive heads have been used for writing, to this date.

[Link]

Magnetic Recording Fundamentals

Consider the familiar audio cassette tape recorder in Figure 0-2. The recording/playback
head is made of easily magnetised ferrite material and has a small air-gap at the point where
it comes into contact with the recording tape, = 0. When energised, the coil winding on the
structure is used to create a strong and concentrated magnetic field on the recording media as
it moves along with a velocity v. During the playback mode, this coil detects the induced
voltages.
iw

magnetic flux

gap

media velocity
v

fringing field
n s

n s

n s

head distance
recording media

magnetic dipoles

substrate

Figure 0-2 Basic Ring read/write head.

The recording media in this case is a length of MYLAR (plastic) tape coated with a
powdered ferric oxide compound which is magnetisable and has high remanence. This layer of
magnetic material in the unmagnetised state may be conceived as made up of dipoles, tiny
magnets with N-S poles randomly positioned. Under the influence of the external magnetic
field, these dipoles will align their N-S poles in line with the applied field thus becoming
magnetised. Upon removal of the applied field some of these dipoles remain aligned.
By either increasing the rate v the media is moved across the head, or by decreasing the
granularity of the magnetic material, (i.e. making the tiny magnets smaller), we can record
faster changes in the applied magnetic field, that is, the frequency response is increased. For
digital data, the density of the stored information increase with decrease in the granularity of
the magnetic media.
With a weak field, only a small number of the dipoles retain their alignment. As the
field gets stronger, more and more of them will remain aligned, that is, the stored magnetic
field increases. For audio (analogue) recording, the variation in the audio signal levels are
recorded in this linear region of the magnetic behaviour. A saturation level is reached when
increases in the applied field does not result in a corresponding increase in the stored magnetic
field. Digital recording generally operate in the saturation region.

[Link]

Magnetic Recording Fundamentals

7.3 The Digital Read/Write Channel

Source
Encoder

Channel
Encoder

Source
Data

Output
Data

Compression
Encoder

Compression
Decoder

Error
Correction
Encoder

Error
Correction
Decoder

Modulation
Encoder

Modulation
Decoder

Signal
Generator
Write
Equalizer

Source
Decoder

Channel
Decoder

Detector
Magnetic disk
Magnetic tape
Optical Disk
CD-ROM
etc

Read
Equalizer

Storage
Device

Figure 0-3 The Digital Recording channel

Digital recording systems may be considered as communications channels. There is an input
signal and an output signal which is a transformed and noisy version of the input. Quoting
from a paper by Berlekamp "Communication links transmit information from here to there,
Computer memories transmit information from now to then." Figure 0-3 show the overall
block diagram of the digital read/write channel. The source data to be recorded or saved is
prepared by the CPU, which also provides the storage device with information concerning the
address of the storage locations. Compression and other source data preprocessing takes place
before the data passes into the channel encoder which adds error-correcting bits and converts
the data stream into a form suitable for recording. This signal is passed through the
equalisation filters, and amplified as the write current for the recording head, creating the
pattern of magnetic fluxes reversals on the storage medium.
To recover the stored data at the output, a reverse sequence takes place. Starting at the
storage medium, the flux reversals are sensed by the magnetic head. The pulses are
demodulated, equalised, decoded and finally presented at the output as the read data.
In digital recording, the magnetic medium is saturated and flux reversals are used to
represent the digital information. Figure 0-4 shows the relationship between the write signal,
flux changes and subsequent induced current in the read mode.

[Link]

Magnetic Recording Fundamentals

Write current

iw

Flux transitions

Magnetization
N

S S

N N

S S

N N

S S

N N

Read voltage

eo

Figure 0-4. Magnetic write and read waveforms.

7.3.1 Write Process
During the write process, a write current iw is passed through the coil. Since the current
needs a finite time to build up and the media is moving under the head, the result is a
magnetic transition with a finite rise-time as shown in Figure 0-4. We can represent the
change in magnetic flux mathematically as
M(x-x0)
Disks and tapes employ the longitudinal recording method illustrated above in which the
flux lines of the magnetic field are oriented in the direction of the motion of the media.
Although higher recording densities are attained using the vertical or perpendicular recording
method in which the flux lines are perpendicular to the surface of the media this method is
more difficult and correspondingly more expensive to implement. As shown above, the
direction of the write current determines the polarity of the magnetisation of the recording
medium. When the current reverses, it creates a flux reversal.

[Link]

Magnetic Recording Fundamentals

7.3.2 Read Process

During the read process, the magnetisation on the recording surface is detected by the head
and some of the magnetic flux is diverted through the coil, producing an induced voltage eo
which is proportional to the rate of change of flux as seen also in Figure 0-4. Mathematically
we can represent this read pulse as

e0 = CV
D( x ) M ( x x 0)dx
x
where D is a the efficiency of the read head. As it is, these detected current pulses are hard to
distinguish from noise pulses found in magnetic media and techniques are required to properly
encode and decode the data for magnetic recording purposes. Under ideal conditions, the peak
of the read pulse indicates the position of the flux transition. Peak detection of this signal is
implemented by first converting the signal to positive pulses with full wave rectification.
These pulses are differentiated and detected with a zero crossing detector as shown in Figure
0-5.

Read signal
e0

Differentated
output

PW50

Weaker read signal

Comparator
output
This edge
for strobe
Figure 0-5. Model of read signal used with peak detection.
Mathematical models of the read pulse are used in the design of the demodulation and
equalisation circuits, and an important parameter used in describing the pulse is the width of
the pulse at 50% amplitude, PW50. The Lorentzian model is the most common mathematical
model for an isolated transition response. The Lorentzian pulse shape can be expressed as:

[Link]

Magnetic Recording Fundamentals

h( t ) =

A
1 + (2t / PW50 ) 2

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-6

-4

-2

7.4 Peak Detection Systems

One major problem affecting the read signal is the effects of noise. It is common to
assume that the noise in the system is additive and Gaussian. Usually, it is also assumed that
the noise consists of two components: a Gaussian white noise component due to the
electronics read (i.e., receiver) side, and a Gaussian coloured component due to the medium.
The spectral characteristics of this coloured noise are essentially the same as would be
obtained from passing white noise through the linear transfer function characterizing the
system. More complicated models for the noise exist, but we will not go into the discussion.
The design of the modulation, coding and signal processing in past magnetic recording
products has been driven by the detector chosen to detect the transitions in the channel input
waveform. This detector, called a peak detector has the advantage of being both robust and
extremely simple to implement. However, by its very nature, it works best at low linear
densities. A block diagram of a typical peak detector is shown in Figure 0-6: Block diagram of
a peak detector.

H1(f)

FWR

>= T

H2(f)

d/dt

ZCD

Figure 0-6: Block diagram of a peak detector

[Link]

and

Magnetic Recording Fundamentals

The are two path through the detector. The first path is used to qualify a peak, i.e., to
ensure that the peak has sufficient amplitude. This path consists of a linear filter, Hl(f), a full
wave rectifier (FWR), and a threshold testing circuit. The bottom path is used to locate the
peak by differentiating the signal after the linear filter H2(t) which remove most of the noise
and then passing the differentiated signal through a zero crossing detector (ZCD).The system
only accepts a peak if the amplitude was large enough to the qualification test.
7.4.1 Data and Clock Recovery
Once a peak is detected by the peak detector, it is thought to be due to a transition in the
input waveform. A device called a phase-lock loop (PLL) is used to derive timing from the
position of the detected peaks. The PLL produces a clock of period Tb seconds by which to
identify channel bit intervals (sometimes called "bit cells"). Then, if an output pulse is located
in a bit interval, that bit interval is said to contain a transition.
The output of the peak detector is used as an input to the PLL, and the output clock
produced by the PLL is constantly being adjusted so that the average peak position is centered
with respect to the edges of the bit interval.
The data and clock information, stored by the schemes described later in the following
section, must be recovered and separated during the read operation. A phase-locked loop
(PLL) as shown in Figure 0-7 is used for this purpose.

Encoded
data

Mux

Phase
detector

VCO

VCO clock

Mux

System
clock

Decoder and
missing clock
detector

Read clock

Read data
Missing clock

Figure 0-7. Phase-locked loop (PLL).

The phase difference between the incoming signal frequency and the frequency of the
voltage-controlled oscillator (VCO) is averaged out by the low-pass filter, forming a DC
voltage to control the VCO. The output of the VCO is divided down to a frequency equal to
the incoming signal frequency and fed into the phase detector. The VCO and the incoming
signal are thus locked together. Any variation in frequency of the incoming signal results in a
change in the phase difference which produces a error voltage to the VCO, causing it to track
the frequency of the incoming signal.
The VCO, which is synchronised to the encoded data stream read back from the drive, is
used to generate two windows, one for the clock and the other for the data pulses. Sync bytes,
written as part of the record format are used to identify the two sets of pulses.

[Link]

Magnetic Recording Fundamentals

7.4.2 Bit Shifting[TC1]

During read back, the signals are very weak and the magnetic head may be considered to
be operating linearly. Each read pulse can be considered independently for analysis. As the
recording density increases, these read pulses are crowded together, giving rise to a source of
problem known as peak shift or bit shift. Referring to Figure 8.1,[TC2] it will be seen that at
each flux transition, the polarity of the domain changes setting two like poles adjacent to each
other. As the superposition of the two read pulse, the result is that the bits will be stored in a
position shifted from the nominal bit cell position. The actual amount of shift depends on the
bit pattern and has to be compensated for.
We can look at the problem by considering a sequence of two read pulses. When the
waveforms of the two signals are superimposed on each we notice two effects occurring.
Firstly, the amplitude of the signals is reduced. More significantly, the position of the peaks
are shifted away from each other.
Shift

d / t

Shift

Figure 8.0-8. Two adjacent bits superimposed resulting in bit shift.

If one examines the waveform produced by the linear superposition of two Lorentzian
pulses (of opposite sign) separated by PW50 seconds, one finds that this waveform will
contain two peaks separated by PW50 seconds, where > . The parameters and are
related by the formula
=

2 1+ 2 1+ 2 + 4
3

which for small becomes

1+ 2
3

For much greater than 1, is approximately equal to , but as approaches zero,

approaches a fixed, limiting distance given by the value. Thus, the peaks will be centered in
their bit interval only at low densities.

[Link]

Magnetic Recording Fundamentals

10

The usual method of eliminating errors from bit shift is by write pre-compensation. In
this technique, the amount of shift is determined from the bit pattern and a deliberate shift in
the opposite direction is introduced during recording to compensate for it. The amount of shift
is a function of the pattern as well as the proximity of the neighbouring pulses. By assuming a
mathematical model of the read pulse waveform, these values can be calculated. These
compensation values are stored in a lookup table which would list all possible bit patterns, for
say a group of eight bits, and the shift required. 'Early' and 'Late' write compensation is usually
incorporated into the controller circuitry. These errors produced by the interference between
neighbouring data bits is known as intersymbol interference (ISI).

7.5 Data Encoding

There are many ways of encoding the data as a function of flux reversals. The simplest
would be to represent each 1 with a pulse leaving the signal low for each 0. This is the returnto-zero (RZ) code as the level always drop back to the zero state. If the level for the 1 is held
high for the whole bit period, we have the special case of the non-return-to-zero (NRZ). In the
early days of magnetic recording many codes were introduced. Although many of these are
now obsolete, new codes are still being proposed mathematically in the search for improved
performance. The design of new codes take into account two main factors:
(i)

The various parameters like density, immunity to noise and timing

inaccuracies, complexity of code etc. have to be balanced against each other,
and the design is a compromise of trade-offs depending on the specific
application and environment.

(ii)

Many codes are protected by patent and sub-optimal solutions are proposed
to avoid license fees.

7.5.1 Common Code Definitions

The following section, together with Fig 0-9, defines some of the various codes used in
digital recording. Generally, the write waveforms are used in the representation. The two
levels of saturation are depicted vertically and the horizontal scale represents the distance
moved by the medium under the head, or equivalently, time. The horizontal unit used is the bit
cell width and is defined the period between two clock transitions. Within the bit cell, the
level can be "up" or "down", and there may or may not be a transition, called the data
transition. Various combinations of clock and data transitions are possible.

[Link]

Magnetic Recording Fundamentals

11

Bit cell
0

RZ

ZERO

NRZ

ZERO

ZERO

NRZI

PM

FM

MFM

ZERO

M2FM

ZERO

Fig 0-9. Recording codes.

(a)

Return to Zero (RZ)

One saturation level is designated as the zero level.

(b)

(c)

0:

there is no flux transition within the whole bit cell

1:

is represented by the occurrence of a downwards transition within the bit cell.

As the level is always at the zero level, an upwards transition is required at the
beginning of the bit cell.

RZ is functionally the same as Return to Saturation (RS). Transitions occur at

both cell boundaries and mid-cell.

Non-Return to Zero (NRZ), Non-Return to Zero-Level (NRZ-L)

0:

Down level (baseline)

1:

Up level

The Non-return to zero (NRZ) technique of recording is the simplest and most
efficient, but it is not self-clocking, and we cannot obtain the bit cell
information from it. Flux reversals occur only at mid-cells (or cell boundaries
in some implementation).

Non-Return to Zero, Invert (NRZI) or Non-Return to Zero, Mark (NRZ-M)

0:

No transition anywhere, magnetic state remains the same as in the previous bit
cell, whether up or down.

[Link]

Magnetic Recording Fundamentals

(d)

(e)

12

1:

A transition occurs in the middle of the bit cell. The direction of the transition
is not important.

A long sequence of zeroes will result in no transitions during that period and
the loss of clock synchronization. 9-track parallel tape drives can use NRZI as
the track for the parity bit will always have a transition when a long sequence
of (00)h is written.

Phase Modulation (PM), Bi-Phase, Transition (Bi-T) or Manchester code.

0:

Up-going transition in mid-cell.

1:

Down-going transition in mid-cell.

Clock transitions, at the cell boundaries, are used to enable the proper direction
for the data transition. For a series of 1's or 0's, an addition transition has to take
place at the cell boundary in order that the correct direction of the transition can
take place at the cell centre. As each bit could require up to two transitions to
encode, this is not an efficient technique. There is always a transition at midcell, and this can be used for the clocking purposes. Phase encoding (PE) is for
9-track tapes and is a simple self-clocking encoding technique.

Frequency Modulation (FM) or Bi-Phase, Mark (Bi-S).

0:

Transition at mid-cell, direction is not important.

1:

No transition at mid-cell, level is not important.

A clock transition is always present at the cell boundaries.

[Link]

Magnetic Recording Fundamentals

(f)

13

Frequency modulation (FM) is similar to PE in efficiency but does not use the
direction of the flux transition. Except for the clock transitions, FM is similar to
NRZI, with the representations for 1 and 0 interchanged. FM is the encoding
method used for single-density floppy disks.

Modified Frequency Modulation (MFM) or Miller code.

0:

No transition.

1:

Transition at mid-cell, exactly as in NRZI.

Since a sequence of zero bits will not have any transitions, introduce a flux
transition at cell boundaries to separate two adjacent 0 bits. In this way there
will be at least one flux transition in any two bit cell period, making it selfclocking. Modified FM (MFM) was introduced by IBM to increase the density
of recording by removing most of the flux transitions at bit cell boundaries in
the FM technique, thus reducing the fcpi to an average of one. MFM is the
standard encoding technique for double-density floppy disks and the earlier
Winchester hard disks. It should be observed that MFM is actually a
modification of NRZI and not FM coding.

Modified MFM (M2FM).

(g)

Modified MFM (M2FM) was introduced by Shugart Associates, one of the

earliest manufacturers of floppy disk drives. It eliminates the flux transition at
the bit cell boundary if it is preceded by a cell boundary flux transition in the
previous cell. However IBM continued to use the MFM method, which became
the de facto industry standard.

7.5.2 Run-Length-Limited Encoding (RLL)

As mentioned earlier, intersymbol interference places an upper limit on the density of the flux
transitions that a particular head-medium system can achieve. Bit shift effects can be
compensated with an early write signal performed on the code prior to the write process.
During the read process we can pass the detected pulses through a equalization filter to "slim"
them, thus reducing the effects of pulse superposition.
Another way of improving the recording density makes use of run-length limited
encoding schemes. This constrained sequences result in a two-level write waveform for which
the minimum and maximum intervals between transitions are fixed.
Run-length limited are designated as:
RLL(d, k, m, r)
and are often shortened to
RLL(d, k)
where
d=

Minimum number of consecutive zeroes allowed (including clock).

k=

Maximum number of consecutive zeroes allowed (including clock).

m = Minimum number of data bits to be encoded.

n=

Number of code bits (including clock) for each of the m data bits.

r=

Number of different word lengths in a variable length code.

[Link]

Magnetic Recording Fundamentals

14

DR = Density Ratio. Data bits per flux reversal.

bit per inch (BPI)
m
DR =
= (d +1) x
n
flux changes per inch (FCPI)
FR = Frequecy Ratio.
FR =

maximum time between transitions k + 1

=
minimum time between transitions d + 1

w = Detection window expressed as a percentage of a data bit cell

m
w = 100
n
Table 1 gives the mathematical calculations performed for a number of popular d, k
constrained codes.
Code
name

Where used

DR

FR

NRZI

Early disks, tapes

1.0

100%

FM

Floppy disks

0.5

2.0

50%

GCR

Tape

0.8

3.0

80%

MFM

Disks, IBM 3330

1.0

2.0

50%

RLL(2,7)

Disks

1.5

2.67

50%

RLL(1,7)

QIC Tape

1.33

4.0

64%

EFM

Compact Disks

10

17

1.41

3.67

47%

Table 1 Performance of some RLL codes.

We shall look at some popular RLL codes below:
7.5.3 Group-coded recording (GCR) or RLL(0,2)
GCR is another self-clocking modification of NRZI. It will be seen that we can translate
each 4-bit group into a 5-bit group with the help of Table 2.[TC3] It will be noted that the 5-bit
code are defined such that the resultant data stream will not contain more than two
consecutive zeroes. The GCR-translated data can be recorded using NRZI without introducing
any additional flux transitions and is thus very efficient. GCR is the recording method used in
high density tape drives such as the IBM 3420. The penalty to be paid is in the complexity of
the encoding/decoding logic.

[Link]

Magnetic Recording Fundamentals

Data nibble
0000
0001
0010
0011
0100
0101
0110
0111

GCR code
11001
11011
10010
10011
11101
10101
10110
10111

Data nibble
1000
1001
1010
1011
1100
1101
1110
1111

15

GCR code
11010
01001
01010
01011
11110
01101
01110
01111

Table 2 GCR encoding table.

7.5.4 Run-length-limited 1,7
The rate 2/3 (1,7) code is arguably the most popular (d,k) code in use today. Several
variations of this code exist. A simple and elegant description, due to Jacoby, begins with the
encoding table in Table 3.
Data nibble
00
01
10
11

RLL (1,7) code

101
100
001
010

Data nibble
00 00
00 01
10 00
10 01

RLL (1,7) code

101 000
100 000
001 000
010 000

Table 3 RLL(1,7) encoding table.

When the encoder is ready to encode a pair of user data bits, it "looks ahead" to the next
pair of user data bits to see if it exist in the right side of Table 3 for both pairs. If the combined
word is not found, use the left side for encoding.
As illustrated for a rate m/n code, the decoding of a code word of length n depends on
the contents of a decoder window that contains the code word in question, as well as a fixed
number of past and future code words ("look back" and "look-ahead"). In the case of the (1,7)
code, the decoder decodes the current three-bit code word by looking ahead at the next two
upcoming code words. In this way, a single incorrectly detected code symbol can propagate
into a burst of at most six user bits (in fact, the burst length does not exceed five user bits).

7.5.5 Run-length-limited 2,7

Another data encoding used together with NRZI modulation encoding in several disk
drive products is a rate 1/2 (2,7) code. One encoding and decoding tree for such a code is
given in Figure 0-10. RLL 2,7 Coding Tree.
It is easily seen that the code rate is 1/2 (every code word contains exactly twice as many
binary digits as the information sequence it represents) and that any concatenation of the
variable length code words satisfies the (2,7) constraint (each 1 in every code word is followed
by at least two 0s, and no code word begins with more than four 0s or ends in more than three
[Link]

Magnetic Recording Fundamentals

16

0s). It can also be verified that every information sequence can be decomposed uniquely into a
sequence of variable length strings in the tree. In addition, the decoding errors due to a single
code symbol in error cannot affect more than four user bits.

*
Root

(1000)

(0100)

(001000)

(100100)

(000100)

(00100100)

(00001000)

Figure 0-10. RLL 2,7 Coding Tree.

The coding tree for the RLL 2,7 code is given Figure 8.2 and Figure 8.3 shows the write
waveforms. Although each data bit is encoded into 2 channel bits, a 50% improvement in data
density is obtained when compared to MFM coding when we only consider the size of the
magnetic domain (fcpi).

Figure 0-11. Run length limited coding (2,7).

[Link]

Magnetic Recording Fundamentals

17

Prior to a few years ago, no general theory existed for the design of data encoding codes
(such as (d,k) codes) with minimum code word length, finite-state encoders, and sliding-block
decoder. Now, there is systematic technique for code construction. The method, called the
sliding-block code algorithm, allows for the design of a practical, efficient (d,k) code for any
choice of the parameters d and k.

7.6 Error Checking Techniques

Errors can enter the system at each stage of the digital recording channel. Error sources
may be classified as noise sources, which arises from random processes and interference
sources, which is deterministic. In digital recording, the most important noise sources come
from the heads, electronics and media. Interference sources of errors include electromagnetic
interference, intersymbol interference, cross-talk, and incomplete overwrite of previously
recorded data.
External interference have to be controlled by careful design and shielding. One of the
major sources of intersymbol interference is bit shifting when the recording density is very
high. This has been discussed earlier.
Head noise is generated randomly in the magnetic head as in any other electrically
resistive component. Electronics noise is generated in the components of the electronics
circuit by the resistive components and active devices such as transistors and integrated
devices. These noise sources are temperature dependent and are especially significant in the
low level analogue signal processing portions of the circuit.
Medium noise is caused by variations in the physical properties of the recording
material. In the case of magnetic recording, this could arise from impurities in the magnetic
material, uneven thickness of the coating, variations in the particle size, physical scratches on
the surface and stretching of the substrate. Similar physical defects occur in optical media.
Other sources of errors come from poor tracking, and mechanical jitter and shock.
During the write process, encoder errors produce permanent (hard) errors in the
recorded data. Permanent errors may be introduced by the media, where the magnetic
properties of the material may be poor, causing bit dropout.
When the data are being read, noise, unstable clocks, mechanical and electronic jitter all
contribute to the problem of data recovery. Other factors include crosstalk, inadequate erasure
of previously written data, tracking error, etc. Sometimes the data can be recovered by making
several attempts at reading. If eventually the data is recovered, these are known as soft errors.
In general, careful design of the low level preamplifiers for the read heads will keep
under control the head and electronics noise, so that media noise is the primary contribution to
data errors. As recording density increases, hard and soft errors created by dirt and other
contamination in magnetic disk drives also increase. Optical disks currently have a recording
density an order of magnitude greater than magnetic disks making the manufacture of defect
free media beyond the current state-of-the-art. Errors will occur, and the only solution is to
have a strategy to manage these errors so that data integrity is not compromised. Random
single bit errors can cause a number of bits of decoded data to be in error. In some cases a
string of errors may occur together; these are called burst errors.
Error detection and correction (EDAC) techniques have been extensively studied and we
will only briefly mention some of them here. When more complex codes are used, care must

[Link]

Magnetic Recording Fundamentals

18

be exercised in the design of codes such that the propagation of errors are controlled to a
known extent.
7.6.1 Block codes
Similar to data transmission applications, block codes are used to protect the stored data
from noise and random errors. Starting with k data symbols (bits), we can add (n - k) parity
symbols to construct a code word with block length n. The code is designed in such a way that
up to t corrupted symbols within the block can be detected and corrected. The (n, k) Hamming
code is an example of bit-organised codes, whereas the (n, k) Reed-Solomon (RS) codes are
used with byte-organised data.
The Hamming single-correction code is expressed as
2 (n - k) = n + 1
When a single parity bit is added to the data word, the number of possible codes is
doubled. The parity rule requires that only half of these are valid codes, the other half being
invalid as they contain single bit errors. The error words differ from the valid codes by one bit,
giving a Hamming distance of 1. By increasing the number of parity bits used, we can increase
the Hamming distance between valid codes. Assuming that when an error occurs, small errors
with fewer error bits are more likely than large multi-bit errors, then correction may be
implemented by choosing the valid code that is closest.
The (7,4) Hamming code represents the 4-bit data word by a seven-bit code word using
3 parity bits. From Table 4. we see that each code word differs from every other code word by
a distance of 3 bits. When a single-bit error occurs during the read process, then if will differ
from the correct code word by 1 bit and from every other code word by 2 bits. Thus we can
recover the correct code word using the (7,4) Hamming code when there is a maximum of 1
error bit. The code efficiency k / n, i.e., the ratio of the data bits k to the block length n, is
0.57. When (n - k) = 8, n = 63, and the code efficient for this larger block length becomes (63
- 8)/63, or about 90%.
Hamming codes are relatively simple but have limited applications, for example in
ECC-memories. More widely used in data storage as well as data communications are the
more complex Reed-Solomon codes. RS codes are discussed in the standard texts on data
communications and information theory.
Data

[Link]

Code

Magnetic Recording Fundamentals

0000
1000
0100
1100
0010
1010
0110
1110
0001
1001
0101
1101
0011
1011
0111
1111

19

0000000
1000110
0100101
1100011
0010011
1010101
0110110
1110000
0001111
1001001
0101010
1101100
0011100
1011010
0111001
1111111

Table 4 Hamming (7,4) Error correcting code..

7.6.2 Cyclic redundancy check (CRC) characters
Written together with the block of data, usually 128, 256, 512, or 1024 bytes in length is
a 16-bit CRC character. On read back, the CRC is recalculated from the read back data and
compared against the written value. Failure to verify the CRC check leads to an error routine,
which normally initiates a retry. A soft error is one which is temporary and gives a successful
read after one or more retries and is probably caused by some dust on the disk surface or by
electrical noise. Errors are considered hard or permanent when after the specified number of
retries have taken place, there is no change to the result. For disk drives, the track is marked
and added to the table of bad tracks.
CRC characters are generated using shift-registers and exclusive-OR gates that produce
an unique 16-bit number for each block of data. Usually the CRC-CCITT polynomial
x16+x12+x5+1 is used although other versions are available. Figure 8.4 shows a circuit for
implementing CRC calculation and the algorithm is shown below.

1-bit shift register

x5

x4

Data bits
to be transmitted
1010001101

Exclusive OR

x2

Figure 0-12. Shift register implementation of division by polynomial x5+x4+x2+1.

Let us assume the following:

a 10-bit message

[Link]

M = 1010001101

Magnetic Recording Fundamentals

20

a 6-bit polynomial
P = 110101
the 5-bit check character R = to be calculated.
1101010110 Q
P 110101 )101000110100000 2nM
110101
111011
110101
111010
110101
111110
110101
101100
110101
1100100
110101
01110 R
The remainder R is added to 2nM to give T = 101000110101110. At the read
channel, T is divided by P, and if the remainder is zero, it is assumed that there have been no
errors. All errors except those divisible by P will be detected.
7.6.3 Interleaving
The bit-error rate (BER) is the parameter used to specify the probability of errors in
storage devices. BER may be defined as the number of error bits divided by the total number
of bits. Magnetic storage devices typically have BER of 10-8. On the other hand, optical
storage devices like CD-ROM and WORM work with much higher recording densities, and
current manufacturing technology have difficulty achieving BER of better than 10-6
consistently. Often a physical defect occurs that is large enough to cause a sequence of many
bits to be in error. Burst errors average several bits in length but can extend to several
hundreds of bits.
The most powerful Reed-Solomon (RS) error correcting code (ECC) can correct only 8
bytes per code word which is insufficient to handle the error bursts of 100 bits or more that are
encountered in optical disks. But if, prior to the recording, several code words are combined
by interleaving, very long burst errors can be reduced to a number of shorter burst which can
be corrected by the ECC.
Rather than bits, RS codes are based on symbols, typically an 8-bit byte. If a RS is
designed to correct an error of 1 byte in a code length of eight bytes, the burst error correction
length is only 1 bit, as a two-bit error occurring at the boundary of the byte symbols would
result in a 2-byte error which is uncorrectable. However, by interleaving three code words
before storage, burst lengths of up to 17 bits could be managed as shown in Figure 8.5. Three
words are encoded and stored in memory by row, but are recorded serially by column. If a
burst error of 3 bytes should occur, the error will be distributed over the three code words and
during decoding on a row by row basis, the data can be recovered.
RS code with an interleave factor of 3 like the scheme above is used in the IBM 3370
disk storage device. They are also used in optical, high capacity tape and disk storage devices.
1 A7 A6 A5 A4 A3 A2 A1 A0
[Link]

Magnetic Recording Fundamentals

21

Words 2 B7 B6 B5 B4 B3 B2 B1 B0 (Encoded by row)

3 C7 C6 C5 C4 C3 C2 C1 C0
Encoded words
Recorded by column
C7 B7 A7 C6 B6 A6 C5 B5 A5 C4 B4 A4 C3 B3 A3 C2 B2 A2 C1 B1 A1 C0 B0 A0
3-byte burst error
1 A7 A6 A5 A4 A3 A2 A1 A0
Words 2 B7 B6 B5 B4 B3 B2 B1 B0

(Correct by row)

3 C7 C6 C5 C4 C3 C2 C1 C0
Row by row decoding correcting for single byte errors
Figure 0-13. Effect of interleaving on burst errors.

7.7 Summary
Digital magnetic recording is a unique application of magnetic recording technology, and in
these sections we can only cover briefly the range of engineering disciplines involved. The
areas of data encoding and error correction draws heavily on information theory and data
communications. For a more comprehensive treatment of the subject the reader is referred to
the following:
Digital Magnetic Recording, 2nd Ed., A.S. Hoagland and J.E. Monson, John Wiley,
N.Y. 1991.
Coding for Digital Recording, J. Watkinson, focal Press, London, 1990.
Optical Recording, A.B. Marchant, Addison Wesley, Mass, 1990.

[Link]