I E T E J o u r n a l o f

E d u c a t i o n
V o J 4 0 , N o s 3 &
4 , J u l y - D e c e m b e r
1 9 9 9 , p p 9 3 - 1 0 0

Line Code for Optical Fiber Communication Systems*

ROOPALI GARG
Electronics and Electrical Communications, Punjab Engineering College, Chandigarh 160 012,
India.

In this paper, different types of line coding techniques used for digital optical fiber communication have
been discussed. The substitutionary AMI codes viz BNZS and HDI£3 codes, combat the problem of loss of
information. The explanation for the preference of two level AMI codes, which includes Coded Mark
Inversion (CMI), Differential Mark Inversion (DMI) and two level class III codes, over the original three
level AMI codes is given. Code tables, properties and power spectra of 1B2B, 3B4B, 5B6B and 7B8B block
codes are given. Bit Insertion Codes—mB1C and DmB1M codes have also been described.
Substitutionary AMI codes have an edge over other codes as the problem or long sequence of zeros are
eliminated. On the same lines, block codes are preferred because of increase in rate of transmission.

The concept of using light for telecommunication is century old. Continuous efforts of scientists
world over led to a new form of communication, popularly known as Optical Fiber
Communication. For the transmission of information from the source to the receiver, various
types of codes are designed. Fundamental types of coding techniques used for digital
transmission are source, channel and line coding. Source coding digitizes the analog waveform.
It is used to reduce the redundancy in the information source output. Channel coding improves
reliability of transmission over noisy channels. Line coding not only provides reliability in
transmission but also suitable shaping of the transmitted signal power spectral density. Binary
line codes are preferred because of large bandwidth available in optical fiber communication. In
addition, these codes are less susceptible to temperature dependence of optical source and
detectors. Under these conditions, two level codes (multilevel codes) are better because of
increase in number of levels. Binary codes are designed which inserts extra symbols into the
information data stream so that the number of consecutive identical received symbols are
decreased; it also facilitates efficient timing extraction at the receiver by producing a high
density of decision level crossings.

NEED FOR LINE CODES

Line code refers to the way in which symbols are encoded by means of pulses. The simplest
method of encoding is to represent each symbol by NRZ (Non Return Zero) rectangular pulses,
also referred to as straight binary format. The second format of the output digital source is RZ
(Return to Zero). In NRZ the pulse width is equal to the symbol period whereas in RZ pulse width
is less than (Usually half Of ) symbol period.
Here, symbol period (Tsym) is the time period occupied by each symbol, once they are
modulated into waveforms. In this connection we use the term BAUD, one BAUD being defined
as one symbol per second, i.e. Rsym = 1/Tsym. In case of binary transmission we refer tb bit
period Tb and bit rate RB, i.e., RB = 1 LTB. To convert M bits into M symbols we have Tsym = M TB
or tfsym = Rt/m.
The need for encoding arises because of limitations of the straight binary format. Sometimes
there are long sequences of like symbols which create a problem of recovering the timing
information from the waveform (due to lack of transitions). So straight binary format suffers
from disappearance of information with time as there are no transitions which contain
information. Loss of timing information increases the probability of error at the output of the
digital system.

When a long string of negative input pulses is passed through an ac coupled channel, initial
rising transients pass but thereafter the output signal level decays. At the end, the output
overshoots to zero, becomes positive and then decays, as shown in Fig 1. The overshoot occurs
to bring the average value to zero. The effect is termed as “dc Wander” or “Base Line Wander”
because the dc level of the signal shifts of maintain zero mean at the output. As the nominal
level containing like digits is reduced, noise margin is reduced.

++++++++++++++++++++++++++++++++

Fig 1 A long string of negative input pulses and the corresponding output waveform.

The gist of the above discussion is that line coding overcomes the limitations of straight binary
format.

 In line coding the long sequence of similar digits is modified. This removes the limitation
of loss of timing information as in straight binary format.
 Unlike straight binary format it is not vulnerable to inter symbol interference (ISI) and
noise.
 It provides error monitoring capability which is absent in other forms.
 Straight binary format has large dc content thus resulting in base line wander. Line
coding overcomes it as the waveform containing long sequence of similar digits is
modified.

Features of a line code

ADEQUATE TIMING INFORMATION: Timing content is assessed in terms of symbol transition for
encoded data. A simple guideline is the maximum number of consecutive line elements which
can occur per unit interval.
INDEPENDENCE OF BIT SEQUENCE: The code should not impose any restriction on the message to be
transmitted.
UNIQUE DECODABILITY: Each received code word should map to a unique output word independent
of coder state.
LOW ERROR MULTIPLICATION: A simple transmission error gives rise to multiple errors in decoding
data. This is described by the error multiplication factor and error spread.
WORD ALIGNMENT INFORMATION: The decoder should be able to identify correct block structure
from the incoming serial data.
ERROR DETECTION CAPABILITY: It is used to monitor link performance and repeater performance.

I
 ROOPALIGARG: LINE CODE FOR OPTICAL FIBER COMMUNICATION SYSTEMS
9
5

SMALL LOW FREQUENCY CONTENT: It is done to facilitate ac coupling within repeaters. This helps in
minimizing base line wander.

MULTI LEVEL CODES USED IN OPTICAL FIBERS

The redundancy which has to be added to the information transmitted can be achieved by one
of the following:
 Increasing the number of amplitude levels. Ternary and other multilevel codes come
under this heading.
 Increasing the pulse repetition rate. Binary code in optical fibers come under this group.

Unipolar NRZ-L Code

Unipolar means pulses of one polarity are used, either positive or negative but not both. For
binary waveform the pulses could change between two fixed levels-binary 1 (represented by
amplitude 2A) and binary 0 (zero level) as shown in Fig 2.
For a long random binary sequence there will be as many l’s as 0’s on average, assuming
equiprobable binits, it follows that average or dc level for waveform is A. For a 1 ohm load,
power dissipated by binary 1 is 4A2 and binary 0 is zero so that average power 2A2.

+++++++++++++++++++++++++++++++++
Fig 2 Unipolar NRZ-L code for (a) binary (c) the ac waveform
The unipolar NRZ-L waveform can be considered as an ac waveform superimposed on a dc level
as shown in Fig 2. The ac waveform can be considered as a rectangular pulse varying between
+A to -A so that amplitude is still 2A. +A represents binary 1 and -A represents binary 0.

Polar NRZ-L Line Code

The polar waveform has both positive and negative pulses. For an infinitely long random binary
sequence, there will be as many l’s and 0’s on average, assuming these are equiprobable binits,
it follows that average or dc level for the waveform is zero as compared to A for unipolar
waveform. Here total signal power is A2 for 1 ohm load as compared to 2A2 for unipolar
waveform for the same swing between levels (Fig 3)

Manchester Line Code (Split Phase Code)

The Manchester code is a NRZ-L code in which bits are coded as transitions between levels + A
to -A representing binary 1 and binary 0 respectively. The transition takes place at the midpoint
on a pulse so that the dc level is entirely eliminated (as shown in Fig 4a- representing binary 1
and Fig 46-representing binary 0). Since transitions always occur, timing information derived
from these is always available but it suffers from disadvantage of requiring twice the bandwidth
of polar NRZ-L waveform code (Fig 4). Reason of the fact follows as Fig 4c represents sequence
….1111…. Manchester code, and Fig 4 D represents sequence ….1010….. in NRZ-L polar
code. With polar waveform, the highest frequency occurs when pulses of alternate polarity are
being transmitted and periodic time of the resulting square wave is 2Tb. Hence the highest
frequency expected in Manchester code is twice of that expected in polar code.

ALTERNATE MARK INVERSION (AMI) LINE CODE

In AMI code binary l’s (marks) are alternatively coded as positive (+ A) and negative levels and
binary 0 (space) are coded as zero voltage:. Thus coined as Alternate Mark Inversion (AMI). The
code is also known as Bipolar code. Even if long sequence of Fs occur, AMI code provides
transition between positive and negative levels, hence dc wander problem is eliminated. As it
uses three levels to encode two binary symbols, it is also termed as Pseudoternary Code (Fig 5).

M-Ary Encoding

-++++++++++++++++++++++
Fig 3 Polar NRZ-L code for (a) binary 1 0; (b) part of a random binary stream
++++++++++++++++
Fig 4 Manchester line code
For Af-ary encoding, we have RSYM = RB/M which clearly indicates that the symbol rate is less
than bit rate. Here, m is number of bits contained in a symbol and M = 2M. If a binary stream is
partitioned into three groups then eight levels (23 = 8) would be required corresponding to ± 4,
± 3A, ± 5A, ± 7A as shown in Fig 6.
I+++++++++++++++
Fig 5 (B) AMI coded pulse train

ADVANTAGES
It allows the information to be transmitted at a fixed bit rate over a smaller bandwidth as
compared to binary transmission. For NRZ-L waveform, the highest frequency is encountered
when the signal alternates between maximum and minimum levels. The square wave is
generated for a period of 2TB for binary system and is 2TSYM for M-ary. Frequency of square
wave is FB = RBL2 for binary system and
FM = RM/2 for M-ary.
So FM = RB /2m (because RM = RB 1M)
Or fM -fb lm (because fb - Rh /2)
Thus, M-ary system require 1/m the bandwidth required for binary transmission.

Disadvantages
It requires more power as compared to binary M) in order to maintain the same peak to peak
with excursions between adjacent levels. Secondly, the circuitry is complex.

SUBSTITUTIONARY AMI CODES

In AMI code when long sequence of l’s occur due to transition between positive and negative
values, problem of loss of timing information is eliminated, whereas when long sequence of 0’s
occur, information disappears. To combat the problem, codes are devised which replace these
long sequence of zeros by, special sequence containing several pulses, using symbols -B, 0, V
where B is pulse satisfying AMI rule, V.$S pulse violating AMI rule, 0 is no transmission. Code PF
this type are:
 BNZS (Binary N Zero Substitution)
 HDBn (High Density Bipolar n)

BNZS (Binary N Zero Substitution)

Commonly used BNZS are:
 B3ZS: Here sequence of three zeros are replaced by sequence of BOV or 00V depending
upon odd number of pulses between consecutive B and V pulses. For00V the sequence
is 00+ or 00- depending on whether previous mark is positive or negative. For BOV, it is -
0- or +0+ depending on whether previous mark is positive or negative. It is used for
transmission at 44.736 Mb/s.
 B6ZS: Here six zeros occurring together are replaced by special sequence of 0VB0VB. If
previous mark is positive, then the sequence is 0+-0-+. If previous mark is negative, the
sequence is 0- +0+ It is devised for transmission at 6.312 Mb/s.

 B8ZS; For transmission at 1.544 Mb/s, the bipolar system zero substitution format has
been devised.
 HDBn (High Density Bipolar n) Line Codes

++++++++++++++
Fig-6 M-ary coding with M = 8 levels

High density bipolar n line codes are so called because they contain a higher density of marks
(I’s) than AMI code. Most common code, under this category is HDB3. In this code sequence of
four zeros are encoded by a sequence 000V or B00V. The sequence 000V will be either 000+ if
previous marks is positive or 000- if previous mark is negative. B00V will be +00 + or -00-
depending on whether previous mark is negative or positive respectively. The sequence 000V is
used for first occurrence of 0000 and also where odd number of I’s occur between successive
0000 sequence. The B00V sequence is used where an even number of l’s occur between
successive 0000. In binary form we have
110I000011000000I010 11 100 000
 ROOPALIGARG: Lwe Code pc* Othcal Fmsi Communication Systems 97

_n
n
1
AMI 0
Two level 1 AMI
Class-1 0 (CMI)
Two level 1 AMI
Class-ll n (DMI) Two, level 1 AMI
In AMI it is
-0+OOOO-+OOOOOO-O+O-+-OOOO
000 + -
B6ZS
- 0 + 0 0 0 0 - + (L±_zJL=_± - 0 + 0 - + - Q_=-±-Q
+—0+—
HDB3
+ ~0 + 000+--*--00-00-0-i-0--f-000- 000 + -
TWO LEVEL AMI CODES
Two level AMI codes are used in low capacity optical fiber links, they are modified versions of
the original three level AMI codes. Before discussing the various types of two level AMI codes, it
is necessary to discuss “Disparity'’ and “Digital Sum”.
 Disparity and the digital sum of a code word gives a measure as to how much the code
word is balanced about the mean of transmission levels.
 Disparity is used for code words or blocks whereas digital sum is used for blocks as well
as running sequence which comprises of an arbitrary number of digits usually much
longer than a block.
 Mathematically, if 0, 1, 2 ... (r - 1) be the possible values of a digit to radix r, the disparity
of a digit is DI=AT - AT wherepf is the ith digit and a, = (r -1) -af is complement ofaf. The
disparity of N digit word is given by
+++++++++++++++++

This is equivalent to number of ones minus number of zeros.

The Digital Sum of a n digit code word is given
X a,
++++++++++++++++++++

In codes of radix r, if the digit values are taken as (- R +1, - r + 3,...... r- 3, R -1) then the digital
sum is equal to disparity.

For binary, following values are used

DJ = - 0.5 for a zero 0.5 for a zero
The Running Digital Sum (RDS) is defined as
* 1U
RDS(k)= X
 I E T E J O U R N A L O F
E D U C A T I O N , V o l 4 0 ,
98
N o s 3 & 4 , 1 9 9 9

The Digital Sum variations (DSV) is defined as difference between maximum and minimum value
of running digital sum
DSV = RDS max - RDS min.

Classification of Two Level AMI Codes (Fig 7)

(i) CLASS 1 or CMI (Coded Mark Inversion): Input bit 0 is transmitted as 01.
(it) CLASS II or DMI (Differential Mark Inversion): Space or 0 is transmitted as 01 if last 1 was
transmitted as 11 and 10 if the last 1 was transmitted as 00.
(hi) CLASS III: Space or 0 is transmitted as 01 and 10 alternately in case of successive runs of
zeros.
A zero is transmitted as. 01 if the last 1 is transmitted as 00 and as 10 if last 1 was transmitted as
11.
In two level AMI code mark (one) is transmitted as 00 and 11 alternately.
++++++++++++++++++++++++++++++++++++++

Fig 7 Examples of two level AMI coding

Some of the 5B6B coded words are:

• mBnB BLOCK CODES

In this type of codes the binary sequence to be transmitted is broken into blocks of m bits which
are then translated into binary words of N bits = m + 1. Efficiency of block code is defined as
ratio of actual information rate to the theoretical maximum information rate (MIN). Also,
increase in data rate due to conversion is given by N/M. As N increase efficiency of code
improves but increases in data rate reduces. This implies that the block length should be made
large but this may increase the hardware complexity so N is limited to value of 8.
The MBNB codes shows an error extension effect, for each bit of receiver decision error, the
whole block is in error. The maximum error spread is M bits per bit of receiver error. With the
increase in value of M the error extension and DSV increases.As M increases, the maximum
number of consecutive like elements in- creasesi hence the code carries less information. The
hardware complexity limits the value of m. The different types of MBNB codes are:
(a) 1B2B code
Here one input bit is converted into two output bits as M = 1, N = 2

Source Word Code Word

0 01
1 10

 Code words are of zero disparity.

 Efficiency = min - 1/2 = 50%
 ROQPALIGARG: LINE CODE FOR OPTICAL FIBER COMMUNICATION SYSTEMS ' 9 9

 Increase in data rate = n/m = 2

 Maximum run of successive identical digits = 2 RDS max = 0.5, RDS min = - 0.5
 So DSV = RDS max - RDS min = 1
 Two level AMI codes come under this category.
 1B2B codes employ simple code conversion rule, easy error detection, simple and
reliable timing recovery due to high level of information present. It suffers from the
disadvantage of doubling of the output data rate.

(b) 3B4B Code

In this case 3 bits of input data are converted into 4 bits of code word.

Source Word Code Word

000 0101 0101
001 1001 1001
010 0001 1110
Oil 0010 1101
100 1000 0111
101 0100 1011
110 0110 0110
111 1010 1010

Various code tables are possible but the one giving maximum value of DSV should be chosen.

 Number of code words = 2m = 23 = 8

 Efficiency = 3/4 = 75%
 Increase in data rate = 4/3 = 1.33
 Efficiency has increased from 50% to 75% and increase in data rate has decreased from 2
to 1.33 as n has increased, as was stated earlier.
 Maximum run of successive identical digits = 4 RDS max = 1, RDS min = - 1
 so DSV =2

(c) 5B6B Code

Here five bits of input data are converted into 6 bit code word.
Number of code words = 25 = 32
 I E T E J O U R N A L O F
E D U C A T I O N , V o l 4 0 ,
100
N o s 3 & 4 , 1 9 9 9

Source Word Code Word

00000 101000 010111
00001 011000 loom
00010 100100 011011
00011 ooom 000111 I

lino 001001 110110

inn 000101 111010 I

 Efficiency = 5/6 = 83.33%

 Increase in data rate = 6/5 = 1.2
 Maximum run of successive identical digits = 6 RDS max = 2, RDS min = - 2, So DSV = 4
 Here, with increase inn, efficiency has increased from 75% to 83.33% and increase in
data rate has decreased from 1.33 to 1.2
(d) 7BSB Code
Here 7 bits of input data are converted into 8 bit code word.
Source Word Code Word
0000000 11101000 11101000
0000001 00010001 11101110
0000010 00010010 11101101
0000011 00010100 11101011

1111110 11100100 11100100

1111111 01010101 01010101
 Number of code words = 27 = 128
 Some of the 7B8B code words are
 Efficiency = 7/8 = 87.5%
 Increase in data rate = 8/7 = 1.14
 Again with increase in N efficiency has increased from 83.33% to 87.5% whereas increase
in data rate has decreased form 1.2 to 1.14
 Maximum run of successive identical digits = 6 RDS max = 3, RDS min = - 3 So DSV – 6

Power Spectrum of MBNB Codes

Low frequency components are small in magnitude so ac coupling will not be a problem.
 ROQPALIGARG: LINE CODE FOR OPTICAL FIBER COMMUNICATION SYSTEMS ' 1 0 1

Error Monitoring
The receiver must be able to break the incoming digit streams into correct code words so that
the code words are converted back to the original input data stream at the receiver. To monitor
the errors at the line terminal, the methods employed are:
 Code Alphabet Violation: The prohibited words are identified in the data sequence. The
error monitoring is achieved.
 EXCESS DISPARITY MONITORING: Digital errors change the disparity of words thus bounds on
RDS are exceeded. By detecting when these bounds are exceeded error monitoring is
achieved.
BIT INSERTION CODES
At very high data rates mBnB code conversion does not prove useful, specially for large values of
m. Therefore bit insertion codes are implemented for the same. It has simpler implementation
Circuit. Here an additional bit is inserted after every m bits. Two types of bit insertion codes are
discussed below

mB1C Code
The additional bit inserted, i.e. (m + 1) th bit is taken as complement of with bit and is referred
to as c bit. The maximum number of consecutive identical elements occur when c bit and
successive wi bit elements are identical. Maximum number of consecutive identical bits are
Nmax = wi + 1. Unlike block codes mB1C does not have a defined structure. To identify the c bit
at the decoder, framing bits are needed for a long sequence of coded bits, hence their effect on
code efficiency is very small. When two successive bits of mB1C codes are passed through an
Exclusive OR (EX -OR) gate, a one is obtained after every (m + 1) bits. During error this does not
happen. Such violations are useful for error monitoring.

Differential m Binary with 1 Mark Insertion (DmB1M) code (Fig 9)

In this code a mark (or one) is inserted in (m + 1) th time slot. Mark inserted signal (d) is
converted to DmB1M Code (e) by the rule ek = dk + ek-1 where ek, dk denotes the fah bit of e and
respective Modulo-2 adder is used for the purpose.

The data is decoded by the following rule dk = ek + ek-1 The arithmetic rules for differential
encoding are:

+++++++++++++++++++++++++++++++

dk Ek-1 ek
0 0 0
0 1 1
1 0 1
i 1 0
 IETE JOURNAL OF EDUCATION, V o l 40, Nos 3 &
4,1999

When the two bits are same, the output is 0 and when they are complemented, the output is 1.
Since the mark inserted signal contains a mark in its (m+1)th time slot so dm+1 = 1.

Em+1= dm+1 + em
Em+1= 1+em
Em+1= em

Fig 9
Schematic diagram of DmB1M code converter

So the (m + 1)th bit is complement of mth bit. Thus, the DmB1M code resembles mB1C code in
its structure, and many of their properties are identical.

DmB1M has a balance of marks and spaces. Here, maximum number of consecutive identical
digits are Nmax = m + 1. As the code is unbalanced so DSV is infinite. Error monitoring can be
done by checking every (m + l)th mark_bit. Major problem of the code is error propagation due
to particular decoding rules used here. One decision error will lead to many, so additional
circuitry will be required to prevent it.

Sample operation is as follows:

+++++++++++++++++++++++++

The initial output state of modulo-2 adder is 0, reference bit is 1, so the output is 1 so now EK =1
i.e. first message bit. Also, first input bit is 1, so after differential encoding second message bit
and also feedback bit becomes 0. The operation continues in similar fashion till all the bits are
encoded.
CONCLUSION
In this paper, the various line codes used in fiber optic communication have been reviewed. The
need for line codes and the features of line codes are discussed. Multilevel codes which include
unipolar and polar NRZ-L line codes. Manchester line codes, AMI codes, M-ary coding are
described. Substitutionary AMI codes which eliminate the problem of loss of timing informa-
tion. In the former long sequence of 0’s are replaced as in BNZS and HDB3 code. Example of
BNZS, i.e. B3ZS, B6ZS and B8ZS are discussed. Two level AMI codes, used in low capacity optical
fiber links' are discussed. These include CMI, DMI and class HI codes. They are improved
versions of the original three level AMI codes. Several examples of mBnB block codes have been
considered. The code table, properties and power spectra of 1B2B, 3B4B, 5B6B and 7B8B block
codes are given. Error monitoring techniques are followed by Bit Insertion Codes : DmB1M and
mB1C.

REFERENCES
 Sujit Kumar, Study of line codes for fibre optic communication, M Tech Thesis, IIT Kanpur,
1992.
 K W Cattermole, Principles of digital line coding, Int J electronics, vol 55, no 1, pp 3-33,
1983.
 Y Takasaki, K Yamashika & Y Takashashi, Two-level AMI line coding family for optical fibre
systems, Int J Electronics, vol 55, no 1, pp 120-131,1983.
 N Yashikai, K Katagiri & T ltd, mB1C code and its performance in an optical
communication system, IEEE Trans Comm, vol 32, no 2, pp 163-168,1984.
 K Kawanishi, N Yoshikai, J Yamada & K Nakagawa, DmB1M code and its performance in a
very high speed optical transmission system, IEEE Trans Comm, vol 36, no 8, pp 951-
956,1988.
 P K Chatterjee & Joseph John, Line codes for optical fibre communication system, IIT
Kanpur.