`460 DlOflAL ('OMMUNICAnONS
`
`bound on R, such that P,,->O as n~:c, For unquantized (soft-decision)
`bound on R, such that P,,->O as n~:c, For unquantized (soft-decision)
`decoding, R" is given as
`decoding, R" is given as
`
`2
`R = log
`2 1 _ e--f./Nl !
`{}
`
`(8-1-102)
`(8-1-102)
`
`where 'to! No = R, Y. is the SNR per dimension_ This result was derived in
`where 'to! No = R, Y. is the SNR per dimension_ This result was derived in
`Section 7-2.
`Section 7-2.
`On the other hand, if the output of the demodulator is quantized to Q levels
`On the other hand, if the output of the demodulator is quantized to Q levels
`prior to decoding, the Chernoff bound may be used to upper-bound the
`prior to decoding, the Chernoff bound may be used to upper-bound the
`ensemble average binary error probability Pis;, 8m ) defined in Section 7-2. The
`ensemble average binary error probability Pis;, 8m ) defined in Section 7-2. The
`result of this derivation is the same upper bound on Pe given in (8-1-101) but
`result of this derivation is the same upper bound on Pe given in (8-1-101) but
`with R" replaced by R Q , where
`with R" replaced by R Q , where
`
`(8-1-103)
`(8-1-103)
`
`In (8-1-103), {Pi} are the prior probabilities of the two signals at the input to
`In (8-1-103), {Pi} are the prior probabilities of the two signals at the input to
`the channel and {P(i I j)} denote the transition probabilities of the channel. For
`the channel and {P(i I j)} denote the transition probabilities of the channel. For
`example. in the case of a binary symmetric channel, we have p, = po = t
`example. in the case of a binary symmetric channel, we have p, = po = t
`P(O 10) = p(llt) = 1- P, and P(O 11) = p(ll 0) = p. Hence,
`P(O 10) = p(llt) = 1- P, and P(O 11) = p(ll 0) = p. Hence,
`
`where
`where
`
`2
`2
`RQ =log21 +v'4p(1_p) Q=2
`RQ =log21 +v'4p(1_p) Q=2
`
`p = Q(v'2Yb RJ
`p = Q(v'2Yb RJ
`
`(8-:-104)
`(8-:-104)
`
`(8-1-105)
`(8-1-105)
`
`A plot of RQ versus 10 log ('lJNol is illustrated in Fig, 8-1-15 for Q =2 and
`A plot of RQ versus 10 log ('lJNol is illustrated in Fig, 8-1-15 for Q =2 and
`Q = 00 (soft-decision decoding). Note that the difference in decoder perfor(cid:173)
`Q = 00 (soft-decision decoding). Note that the difference in decoder perfor(cid:173)
`mance between unquantized soft-decision decoding and hard-decision decod(cid:173)
`mance between unquantized soft-decision decoding and hard-decision decod(cid:173)
`ing is approximately 2 dB. In fact. it is easily demonstrated again that as
`ing is approximately 2 dB. In fact. it is easily demonstrated again that as
`'lJNll~O. the
`loss
`in performance due to hard-decision decoding
`is
`'lJNll~O. the
`loss
`in performance due to hard-decision decoding
`is
`
`1.0
`1.0
`N O.9
`O.9
`N
`H 0.8
`H 0.8
`~ 0.7
`~ 0.7
`oc~ 0.6
`oc~ 0.6
`1 0.5
`1 0.5
`,z 0.4 g 0.3
`,z 0.4
`
`"
`
`0.2
`0.1
`0.1
`
`FlGURE 11-1-15 Comparison of Ro (soft·decision decoding) with RQ (hard(cid:173)
`FlGURE 11-1-15 Comparison of Ro (soft·decision decoding) with RQ (hard(cid:173)
`decision decoding.) as a function of the SNR per dimension.
`decision decoding) as a function of the SNR per dimension.
`
`o
`o
`-5
`5
`IOlogtlJNn) (dB)
`IOlogtlJNn) (dB)
`
`10
`10
`
`TQ Delta Exhibit 2006 (Part 2 of 2)
`Cisco Systems, Inc. v. TQ Delta LLC
`IPR2016-01021
`
`470
`
`
`
`f'HAPlfR it BLOCK AND CONVOLUTIONAL CHANNEL COVES 461
`f'HAPlfR it BLOCK AND CONVOLUTIONAL CHANNEL COVES 461
`
`to log", ~Jr = 2 dB, which is the same decibel difference that was obtained in
`to log", ~Jr = 2 dB, which is the same decibel difference that was obtained in
`our comparison of the channel capacity relations. We mention that about 1 dB
`our comparison of the channel capacity relations. We mention that about 1 dB
`of this loss can be recovered by quantizing the output of the demodulator to
`of this loss can be recovered by quantizing the output of the demodulator to
`three levels instead of two (see Problem 7-11). Additional improvements are
`three levels instead of two (see Problem 7-11). Additional improvements are
`possible by quantizing the output into more than three levels. as shown in
`possible by quantizing the output into more than three levels. as shown in
`Section 7-3.
`Section 7-3.
`
`8-1-7 Bounds on Minimum Distance of Linear Block Codes
`8-1-7 Bounds on Minimum Distance of Linear Block Codes
`The expressions for the probability of error derived in this chapter for
`The expressions for the probability of error derived in this chapter for
`sofl-decision and hard-decision decoding of linear binary block codes clearly
`sofl-decision and hard-decision decoding of linear binary block codes clearly
`indlcate the importance that the minimum distance parameter plays in the
`indlcate the importance that the minimum distance parameter plays in the
`performance of the code. If we consider soft-decision decoding, for example,
`performance of the code. If we consider soft-decision decoding, for example,
`the upper bound on the error probability given by (8-1-52) indicates that, for a
`the upper bound on the error probability given by (8-1-52) indicates that, for a
`given code rate K = kin, the probability of error in an A WON channel
`given code rate K = kin, the probability of error in an A WON channel
`decrease', exponentially with dm •n • When this bound is used in conjunction with
`decrease', exponentially with dm •n • When this bound is used in conjunction with
`the lower bound on d m " given below, we obtain an upper bound on PM that
`the lower bound on d m " given below, we obtain an upper bound on PM that
`can be achieved by many known codes. Similarly, we may use the upper bound
`can be achieved by many known codes. Similarly, we may use the upper bound
`given by (8-1-82) for the probability of error for hard-decision decoding in
`given by (8-1-82) for the probability of error for hard-decision decoding in
`conjunction with the lower bound on d min to obtain an upper bound on the
`conjunction with the lower bound on d min to obtain an upper bound on the
`error probability for
`linear binary block codes on the binary symmetric
`error probability for
`linear binary block codes on the binary symmetric
`channel.·
`channel.·
`On the other hand, an upper bound on dmin can be used to determine a
`On the other hand, an upper bound on dmin can be used to determine a
`lower bound on the probability of error achieved by the best code. For
`lower bound on the probability of error achieved by the best code. For
`example, suppose that hard-decision decoding is employed. In this case, we
`example, suppose that hard-decision decoding is employed. In this case, we
`have the two lower bounds on PM given by (8-1-86) and (8-1-87). with the
`have the two lower bounds on PM given by (8-1-86) and (8-1-87). with the
`former being the tighter. When either one of these two bounds is used in
`former being the tighter. When either one of these two bounds is used in
`conjunction with an upper bound on d min the result is a lower bound on PM for
`conjunction with an upper bound on d min the result is a lower bound on PM for
`the best (n. k) code. Thus, upper and lower bounds on d min are important in
`the best (n. k) code. Thus, upper and lower bounds on d min are important in
`assessing the capabilities of codes.
`assessing the capabilities of codes.
`A simple upper bound on the minimum distance of an (n. k) binary or
`A simple upper bound on the minimum distance of an (n. k) binary or
`non-binary linear block code was given in (8-1-14) as d min ,;; n - k + 1. It is
`non-binary linear block code was given in (8-1-14) as d min ,;; n - k + 1. It is
`convenient to normalize this expression by the block size n. That is.
`convenient to normalize this expression by the block size n. That is.
`
`1
`d min
`1
`d min
`~ ~ (l - R,.) + -
`~ ~ (l - R,.) + -
`n
`n
`n
`n
`
`(8-1- j(}6)
`(8-1- j(}6)
`
`where Rc is the code rate. For large n. the factor lin can be neglected.
`where Rc is the code rate. For large n. the factor lin can be neglected.
`If a code has the largest possible distance, i.e., d min = n - k + 1, it is called a
`If a code has the largest possible distance, i.e., d min = n - k + 1, it is called a
`maximum-distance-separable code. Except for the trivial repetition-type codes.
`maximum-distance-separable code. Except for the trivial repetition-type codes.
`there are no binary maximum-separable codes. In fact, the upper bound in
`there are no binary maximum-separable codes. In fact, the upper bound in
`(8-1-106) is extremely loose for binary codes. On the other hand. nonbinary
`(8-1-106) is extremely loose for binary codes. On the other hand. nonbinary
`codes with d nin = n - k + 1 do exist. For example, the Reed-Solomon codes,
`codes with d nin = n - k + 1 do exist. For example, the Reed-Solomon codes,
`which comprise a subclass of BCH codes, are maximum-distance-separable.
`which comprise a subclass of BCH codes, are maximum-distance-separable.
`In addition to the upper bound given above, there are several relatively
`In addition to the upper bound given above, there are several relatively
`
`471
`
`
`
`462
`462
`
`l)I(lliAL (O:'tot~1lXIC .. HIOl'\iS
`l)I(lliAL (O:'tot~1lXIC .. HIOl'\iS
`
`tight hounds on the minimum distance of linear block codes. We shall briefly
`tight hounds on the minimum distance of linear block codes. We shall briefly
`descrihe four important bounds. three of which are upper bounds and the
`descrihe four important hounds. three of which are upper bounds and the
`other a lower bound. The derivations of these bounds are lengthy and are not
`other a lower hound. The derivations of these hounds are lengthy and are not
`of particular interest in our subsequent discussion. The interested reader may
`of particular interest in our subsequent discussion. The interested reader may
`Tefer 10 Chapler 4 of the hook by Peterson and Weldon (1972) for those
`Tefer 10 Chapler 4 of the hook by Peterson and Weldon (1972) for those
`derivations.
`derivations.
`One upper hound on the minimum distance can be obtained from the
`One upper hound on the minimum distance can be obtained from the
`inequality in [8-1-83). By taking the logarithm of both sides of (8-1-83) and
`inequality in [8-1-83). By taking the logarithm of hoth sides of (8-1-83) and
`dividing by If. we obtain
`dividing by If. we obtain
`' (n)
`' (n)
`1 - R.- ;. - log, 2:
`.
`1 - R.- ;. - log, 2:
`.
`I
`I
`n
`n
`
`;00-0
`;00-0
`
`I
`I
`
`(8-1-\07)
`(8-1-\07)
`
`Since the error-correcting capability of the code. measured by I, is related to
`Since the error-correcting capability of the code. measured by I, is related to
`the minimum distance. the above relation is an upper bound on the minimum
`the minimum distance. the ahove relation is an upper bound on the minimum
`distance. It is called the Hamming upper bOllnd.
`distance. It is called the Hamming upper bOllnd.
`The asymptotic form of {8-1-107) is obtained by letting n-+ ce. Now, for any
`The asymptotic form of {8-1-107) is obtained by letting n-+ ce. Now, for any
`II. let t" be the largest integer I for which (8-1-107) holds. Then, it can be shown
`II. let t" be the largest integer I for which (8-1-107) holds. Then, it can be shown
`(Peterson and Weldon. 1972) that as n.- x , the ratio tin for any (n, k) block
`(Peterson and Weldon. 1972) that as n.- x , the ratio tin for any (n, k) block
`code cannot exceed r,,/n, where tuln satisfies the equation
`code cannot exceed r,,/n, where tuln satisfies the equation
`
`1- Rc = H(lo/n)
`1- Rc = H(lo/n)
`
`(8-1-108)
`(8-1-108)
`
`and H(xl is the binary entropy function defined by (3-2-10)_
`and H(xl is the binary entropy function defined by (3-2-10)_
`The generalization of the Hamming bound to nonbinary codes is simply
`The generalization of the Hamming bound to nonbinary codes is simply
`
`1 - Rc;' !.Iog., [± (.~)(q - I)i]
`1 - Rc;' !.Iog., [± (.~)(q -I)i]
`
`n
`n
`
`(=0 1
`(=0 1
`
`(8-1-109)
`(8-1-109)
`
`Another upper hound. developed by Plotkin (1960), may be stated as
`Another upper bound. developed by Plotkin (1960), may be stated as
`follows. The number of check digits required to achieve a minimum distance
`follows. The number of check digits required to achieve a minimum distance
`d mh in an (n. k) linear block code satisfies the inequality
`d mh in an (n. k) linear block code satisfies the inequality
`
`qdm ;, -I
`qdm ;, -I
`-I-loud·
`-I-loud·
`n-k:;;'
`n-k:;;'
`oq mm
`oq mm
`q-
`q-
`I
`I
`
`(8-1-110)
`(8-1-110)
`
`where q is the alphabet size. When the code is binary, (8-1-110) may be
`where q is the alphabet size. When the code is binary, (8-1-110) may be
`expressed as
`expressed as
`
`) 1 (
`dmin (1
`) 1 (
`dmin (1
`- - 1 - -2d . log2 d min
`- - 1 - -2d . log2 d min
`,,;; -
`,,;; -
`2
`2
`n
`n
`min
`min
`
`2)
`2)
`
`1 - R,. + -
`1 - R,. + -
`n
`n
`
`In the limit asn.- lC with dminln ,,;;~, (8-1-110) reduces to
`In the limit asn.- lC with dminln ,,;;~, (8-1-110) reduces to
`
`(8-1-111)
`(8-1-111)
`
`472
`
`
`
`CHAYff-1{ It llU)('fo( A~D ('():-';VGL1'TJO:\AL CHA':-;I-L \ '()DES
`llU)('i\. A~D ("():-';VGL1'TJO:\AL CHA':-;I-L. \ "()DES
`CHA:>·ff·K K
`
`463
`463
`
`Finally. there is another tight upper bound on the minimum distance
`Finally, there is another tight upper bound on the minimum distance
`obtained by Elias (Berlekamp. I 96i'i). It may be expressed in its asymptotic
`obtained by Elias (Berlekamp. I 96i'i). It may be expressed in its asymptotic
`form as
`form as
`
`(~-1-112)
`(~-1-112)
`
`where the parameter A ,s related Ie) the code rate through the equation
`where the parameter A ,s related (e) the code rate through the equation
`
`R.= I +Alog,A -tl-A)log,(I-A). (hA~1 (~-I-I13)
`R,= I +Alog,A -tl-A)log,(I-A). (hA~1 (~-I-II3)
`
`Lower bounds on the minimum distance of (n. k) block codes also exist. In
`Lower bounds on the minimum distance of (fl, k) block codes also exist. In
`particular. binary block codes exist that have a normalized minimum distance
`particular. binary block codes exist that have a normalized minimum distance
`that asymptotically satisfies the inequality
`that asymptotically satisfies the inequality
`
`(8-1-1141
`(8-1-1141
`
`where a is related u the code rate through the equation
`where a is related u the code rate through the equation
`
`R, = I -
`R, = I - fi(a)
`f/( a. )
`
`= 1 + a log, (f + (I - (Y) log, (l - a).
`= I + a log, '" + (I - (Y) log, (l - a).
`
`(8-1-115)
`(8-1-115)
`
`ThiS lower bound is a specie! case of a lower bound developed by Gilbert
`ThiS lower bound is a special case of a lower bound developed by Gilbert
`(1952) and Varsharmov (1957). which applies to non binary and b:nary block
`(1952) and Varsharmov (1957). which applies to non binary and b:nary block
`codes.
`codes.
`The asymptotic bounds given above arc plotted in Fig. 8·1-16 for binar)
`The asymptotic bounds given above arc plotted in Fig. 8·1-16 for binar)
`codes. Also plotted in the figure for purposes of comparison are curves of the
`codes. Also plotted in the figure for purposes of comparison are curves of the
`minimum distance as a function of code rate for BCH codes of block lengths
`minimum distance as a function of code rate for BCH codes of block lengths
`n = 31 and 63. We observe that for n = 31 and 63, the normalized minimum
`n = 31 and 63. We observe thai for n = 31 and 63, the normalized minimum
`distance falls well above the Varsharmov-Gilbcrl lower bound. As the block
`distance falls well above the Varsharmov-Gilbcrt lower bound. As the block
`lcn~lh n increa,cs. \he efficiencv of the BCH codes diminishes. For example,
`lcn~th n increa,cs. \he efficiencv of the BCH codes diminishes. For example.
`when n = 1023, the curve for the normalized minimum distance falls close to
`when n = 1023. the curve for the normalized minimum distance falls close to
`
`05
`0.5
`
`n
`n
`
`Plot!.;!.'. urpcr bo"nd
`Plot!.;!.'. urpcr bo"nd
`
`Il~
`l:h..t\
`l:h..t\~_....;;;:
`""","-~
`i Uppl!: b"una
`i Uppl!: b"una
`
`() I r Gilhel1-Varsbrmn\"
`() I r Gilhel1-Varsbrmnv
`
`lo .... er hound
`lo .... er hound
`
`FIC;l.:RE 8-1-16
`FIC;L:RE 8-1-16
`
`l;ppt:r and lower hounds un normalizl'd minimum
`l;ppt:r and lower hounds un normalizl'd minimum
`Jistance as.a function of code fJte.
`Jistance a'S a function of code fJte.
`
`"
`"
`
`0.2
`0.2
`
`OfJ
`OfJ
`:1~
`:1~
`Code r,lk' N,
`Code r,lk' N,
`
`0,,;,\
`0·';'\
`
`1.0
`1.0
`
`473
`
`
`
`464
`464
`
`DIGITAL COMMF"ICATIONS
`DIGITAL COMMF"ICATIONS
`
`the Varsharmov-Gilbert bound. As n increases beyond n = 1023. the normal(cid:173)
`the Varsharmov-Gilbert bound. As n increases beyond n = 1023. the normal(cid:173)
`ized minimum distance of the BCH codes continues to decrease and falls below
`ized minimum distance of the BCH codes continues to decrease and falls below
`the Varsharmov-Gilbert bound. That is, dm;n/n approaches zero as n tends to
`the Varsharmov-Gilbert bound. That is, dm;n/n approaches zero as n tends to
`infinity. Consequently the BCH codes. which are the most important class of
`infinity. Consequently the BCH codes. which are the most important class of
`cyclic codes. are not very efficient at large block lengths.
`cyclic codes. are not very efficient at large block lengths.
`
`8-1-8 Nonbinary Block Codes and Concatenated Block
`8-1-8 Nonbinary Block Codes and Concatenated Block
`Codes
`Codes
`
`A non binary block code consists of a set of fixed-length code words in which
`A non binary block code consists of a set of fixed-length code words in which
`the elements of the code words are selected from an alphabet of q symbols.
`the elements of the code words are selected from an alphabet of q symbols.
`denoted by {a, 1,2 ..... q - I}. Usually. q = 2*, so that k information bits are
`denoted by {a, 1,2 ..... q - I}. Usually. q = 2*, so that k information bits are
`mapped into one of tbe q symbols. The length of the non binary code word is
`mapped into one of tbe q symbols. The length of the non binary code word is
`denoted by N and the number of information symbols encoded into a block of
`denoted by N and the number of information symbols encoded into a block of
`N symbols is denoted by K. The minimum distance of the nonbinary code is .
`N symbols is denoted by K. The minimum distance of the nonbinary code is .
`denoted by D m ;". A systematic (N, K) block code consists of K information
`denoted by D m ;". A systematic (N, K) block code consists of K information
`symbols and N - K parity check symbols.
`symbols and N - K parity check symbols.
`Among the various types of nonbinary linear block codes, the Reed(cid:173)
`Among the various types of nonbinary linear block codes, the Reed(cid:173)
`Solomon codes are some of the most important for practical applications. As
`Solomon codes are some of the most important for practical applications. As
`indicated previously, they comprise a subset of the BCH codes, which in turn
`indicated previously, they comprise a subset of the BCH codes, which in turn
`are a class of cydic codes. These codes are described by the parameters
`are a class of cydic codes. These codes are described by the parameters
`
`N=q-!=2A-1
`N=q-!=2"-1
`K = 1.2.3, ' " , N - I
`K = 1.2.3, ' " , N - I
`Dm;n=N- K + 1
`Dm;n=N- K + 1
`
`R,=K/N
`R,=K/N
`
`Such a code is guaranteed 10 correct up to
`Such a code is guaranteed 10 correct up to
`
`1= U(Dm ," -
`1= U(Dm ," -
`
`l)J
`l)J
`=U(N - K)J
`=U(N - K)J
`
`(8-1-116)
`(8-1-116)
`
`(8-1-117)
`(8-1-117)
`
`symbol errors. Of course, these codes may be extended or shortened in the
`symbol errors. Of course, these codes may be extended or shortened in the
`manner described previously for binary block codes.
`manner described previously for binary block codes.
`The weight distribution {A,} of the class of Reed-Solomon codes is known.
`The weight distribution {A,} of the class of Reed-Solomon codes is known.
`The coefficients in the weight enumerating polynomial are given as
`The coefficients in the weight enumerating polynomial are given as
`
`where D == D m ," and q = 2*,
`where D == D m ," and q = 2*,
`One reason for the importance of the Reed-Solomon codes is their good
`One reason for the importance of the Reed-Solomon codes is their good
`
`(8-I-WI)
`(8-I-WI)
`
`474
`
`
`
`CHAPTER K: BLOCK AND CON\iOLUTlO~AL {,HA~NL CODES
`CHAPTER K: BLOCK AND CON\iOLUTlO~AL {,HA~NL CODES
`
`46S
`46S
`
`distance properties. A second reason for their importance is the existence of
`distance properties. A second reason for their importance is the existence of
`efficient hard-decision decoding algorithms, which make it possible to imple(cid:173)
`efficient hard-decision decoding algorithms, which make it possible to imple(cid:173)
`ment relatively long codes in many practical applications where coding is
`ment relatively long codes in many practical applications where coding is
`desirable.
`desirable.
`A nonbinary code is particularly matched to an M-ary modulation technique
`A nonbinary code is particularly matched to an M-ary modulation technique
`transmitting the 2' possible symbols. Specifically, M-ary orthogonal
`for
`transmitting the 2' possible symbols. Specifically, M-ary orthogonal
`for
`signaling. e.g., M-ary FSK, is frequently used. Each of the 2* symbols in the
`signaling. e.g., M-ary FSK, is frequently used. Each of the 2* symbols in the
`q-ary alphabet .is mapped to one of the M = 2* orthogonal signals. Thus, the
`q-ary alphabet .is mapped to one of the M = 2* orthogonal signals. Thus, the
`transmission of a code word is accomplished by transmitting IV orthogonal
`transmission of a code word is accomplished by transmitting IV orthogonal
`signals, where each signal is selected from the set of M = 2' possible signals.
`signals, where each signal is selected from the set of M = 2' possible signals.
`The optimum demodulator for such a signal corrupted by A WGN consists
`The optimum demodulator for such a signal corrupted by A WGN consists
`of M matched filters (or cross-correIa tors) whose outputs are passed to the
`of M matched filters (or cross-correIa tors) whose outputs are passed to the
`decoder, either in the form of soft decisions or in the form of hard deciSions. If
`decoder, either in the form of soft decisions or in the form of hard deciSions. If
`hard decisions are made by the demodulator, the symbol error probability f"
`hard decisions are made by the demodulator, the symbol error probability f"
`and the code parameters are sufficient to characterize the performance of the
`and the code parameters are sufficient to characterize the performance of the
`decoder. In fact, the modulator, the A WGN channel, and the demodulator
`decoder. In fact, the modulator, the A WGN channel, and the demodulator
`form an equivalent discrete (M-ary) input, discrete (M-ary) output, symmetric
`form an equivalent discrete (M-ary) input, discrete (M-ary) output, symmetric
`memoryless channel characterized by the transition probabilities p. = 1 - PM
`memoryless channel characterized by the transition probabilities p. = 1 - PM
`and f,,/(M - 1). This channel model, which is illustrated in Fig. 8-1-17. is a
`and f,,/(M - 1). This channel model, which is illustrated in Fig. 8-1-17. is a
`generalization of the BSe.
`generalization of the BSe.
`The performance of the hard-decision decoder may be characterized by the
`The performance of the hard-decision decoder may be characterized by the
`following upper bound on the code word error probability:
`following upper bound on the code word error probability:
`P. "". ~ (IV)' r (1- P )N-,
`P. "". ~ (IV)' r (1- P )N-,
`
`(L .J .M
`(L .J .M
`I
`I
`1'-' -<-- 1
`1'-' -<-- 1
`
`M
`M
`
`(8-1-119)
`(8-1-119)
`
`where I is the number of errors guaranteed to be corrected by the code.
`where I is the number of errors guaranteed to be corrected by the code.
`When a code word error
`is made. the corresponding symbol error
`When a code word error
`is made. the corresponding symbol error
`probability is
`probability is
`[ " ' IV ' .
`[ " ' IV ' .
`p, .• = ~ ~ i( . 1p',,(I- f,,)"
`p, .• = ~ ~ i( . 1p',,(I- f,,)"
`
`'"Y 1= t t l
`ill 1= t t l
`
`I J
`I J
`
`(8-1-120)
`(8-1-120)
`
`FIGURE 8-1-17 M-ary input. M-ary output, s},mrnetric merroryless
`FIGURE 8-1-17 M-ary input. M-ary output, s},mrnetric merroryless
`channel.
`channel.
`
`1 - p.w
`M - , .t::-----..::~ __ 30 "1 _ .
`PM
`p= M- 1
`
`475
`
`
`
`466
`466
`
`DIGITAL COMMU!"Ic.,-\TTO~S
`DIGITAL COMMU!"Ic.,-\TTO~S
`
`Furthermore, if the symbols are converted to binary digits, the hit error
`Furthermore, if the symbols are converted to binary digits, the hit error
`probability corresponding to (8-1-120) is
`probability corresponding to (8-1-120) is
`
`(8-1-121)
`(8-1-121)
`
`Example 8-1-13
`Example 8-1-13
`
`I = 31 Reed-Solomon code
`Let us evaluate the performance of an N = 2' -
`I = 31 Reed-Solomon code
`Let us evaluate the performance of an N = 2' -
`with D m ," = 3, 5, 9, and 17, The corresponding values of K are 29, 21, n.
`with D m ," = 3, 5, 9, and 17, The corresponding values of K are 29, 21, n.
`and 15. The modulation is ll/! = 32 orthogonal FSK with noncoherem
`and 15. The modulation is ll/! = 32 orthogonal FSK with noncoherem
`detection at the receiver.
`detection at the receiver.
`The probability of a symbol error is given by (5-4-46), and may he
`The probability of a symbol error is given by (5-4-46), and may he
`expressed as.
`expressed as.
`
`(8-,-122)
`(8-,-122)
`
`where y is the SNR per code symbol. By using (8-1-122) in (8-1-120l and
`where y is the SNR per code symbol. By using (8-1-122) in (8-1-120l and
`combining the result with (8-1-121), we obtain the bit error probability. The
`combining the result with (8-1-121), we obtain the bit error probability. The
`results of these computations are plotted in Fig. 8-1-1/;. Note that the more
`results of these computations are plotted in Fig. 8-1-1/;. Note that the more
`powerful codes (large D mi,) give poorer performance at low SNR per bit
`powerful codes (large D mi,) give poorer performance at low SNR per bit
`than the weaker codes. On the other hand, at high SNR, the more powerful
`than the weaker codes. On the other hand, at high SNR, the more powerful
`codes give better perfolmance. Hence, there are crossovers among the
`codes give better perfolmance. Hence, there are crossovers among the
`various codes, as illustrated for example in Fig. 8-1-18 for the t = 1 and 1 = 8
`various codes, as illustrated for example in Fig. 8-1-18 for the t = 1 and 1 = 8
`codes. Crossovers also occur among the I = I, 2, and 4 codes at smaller
`codes. Crossovers also occur among the I = I, 2, and 4 codes at smaller
`values of SNR per bit. Similarly, the curves for t = 4 and 8 and for 1=1-: and
`values of SNR per bit. Similarly, the curves for t = 4 and 8 and for 1=1-: and
`2 cross in the region of high S"lR. This is the characteristic behavior for
`2 cross in the region of high S"lR. This is the characteristic behavior for
`noncoherent detection of the coded waveforms.
`noncoherent detection of the coded waveforms.
`
`If the demodulator docs nOl make a hard deciSIon on each sl'mbol. buL
`If the demodulator docs nOl make a hard deciSIon on each sl'mbol. buL
`
`~o- (
`~o- (
`
`c:
`c:
`'.::-
`:O.j
`'.::-
`:O.j
`E
`E
`
`"'FIGURE 8.1-18
`"'FIGURE 8.1-18
`
`Perforrna!lce of ~everal i\- =-: 1L (-crror correcting Reeu-So\omon
`Perforrna!lce of ~everal i\- =-: 1L (-crror correcting Reeu-So\omon
`code5 wi!:') 32-ary FSK moduhtion on an A. WGN channel
`code5 wi!:') 32-ary FSK moduhtion on an A. WGN channel
`(nHncohe~enr demodulation).
`(nHncohe~enr demodulation).
`
`10
`
`4.0
`
`(,';':0
`';:-.JR ~I h';,ll"JBl
`
`476
`
`
`
`CHA:1TlR.s BLOCK A~D ["O!\:VOUTIO:-..lAl CHA'''Fl \ O[)l'~ 467
`CHA:>TlR.s BLOCK A~D ["O!\:VOUTIO:-..lAl CHA'''Fl \ O[)l'~ 467
`
`Outer
`Outer
`encoder
`encoder
`! \ Kl
`! \ Kl
`
`Input
`Input
`data
`data
`
`FlGURE 8-t-J9
`FlGURE 8-t-J9
`
`Jolt;!
`Jolt;!
`Block diagram uf a communications ~yslem employing a concatenated c\)dc.
`Block diagram uf a communications ~yslem employing a concatenated c\)dc.
`
`the d.::coder,
`to
`the unquantized matched filter outputs
`instead, passe,
`the d.::coder,
`to
`the unquantized matched filter outputs
`instead, passe,
`soft-decision decoding can he performed. This decoding involves the formation
`soft-decision decoding can he performed. This decoding involves the formation
`of q" = 2"< correlation metrics. where each metric corresponds to one of the
`of q" = 2"< correlation metrics. where each metric corresponds to one of the
`q" code words and consists of a sum of /II matched filter outputs corr.:spon<.iing
`q" code words and consists of a sum of /II matched filter outputs corr.:spon<.iing
`to the N code symt>ols. The matched filter outputs may be added coherently. or
`to the N code symt>ols. The matched filter outputs may be added coherently. or
`they may be envelope-cetected and then added, or they may he 'quare law
`they may be envelope-cetected and then added, or they may he 'quare law
`detected and then added. If coherent detection is used and the channel noise is
`detected and then added. If coherent detection is used and the channel noise is
`AV .... GN, the computation of the probability of error is a straightforward
`AV .... GN, the computation of the prohability of error is a straightforward
`extension of the hinary case considered in Section 8-1-4. On the other hand.
`extension of the hinary case considered in Section 8-1-4. On the other hand.
`when envelope detection or square-law detection and noncoherent comhining
`when envelope detection or square-law detection and noncoherent combining
`are used to fonn the decision variables, the computation of the decoder
`are used to form the decision variables, the computation of the decoder
`performance is considerably more complicated.
`performance is considerably more complicated.
`
`Concatenated Block Codes A concatenated code consists 01 tWe) separate
`Concatenated Block Codes A concatenated code consists 01 tWe) separate
`codes which are combined to form a larger code. Usually one code is selected
`codes which are combined to form a larger code. Usually one code is selected
`to be nonbinarv and the other is binary. The two codes are concatt:nated as
`to be nonbinarv and the other is binary. The two codes are concatt:nated as
`illustrated in Fig. 8-1-\9. The Ilonbinary (N, K i code forms the <'uter code and
`illustrated in Fig. 8-1-\9. The nonbinary (N, K; code forms the <'uter code and
`the binary code forms the inner code. Code words are formed by subdividing a
`the binary code forms the inner code. Code words are formed by subdividing a
`block of kK information bits into K groups, called symbols, where each symhol
`block of kK information bits into K groups, called symbols, where each symhol
`consists of k bits. The K k-bit symbols are encoded into N k-bit symbols by the
`consists of k bits. The K k-bit symbols are encoded into N k-bit symbols by the
`outer encoder, as is usually done with a nonbinary code. The inner encoder
`outer encoder, as is usually done with a nonbinary code. The inner encoder
`takes each k-bit symbol and encodes it into a binary block code of length fl.
`takes each k-bit symbol and encodes it into a binary block code of length fl.
`Thus we obtain a concatenated block code having a block length nf Nfl bits and
`Thus we obtain a concatenated block code having a block length nf Nfl bits and
`containing kK information bits. That is, we have created an equivalent
`containing kK information bits. That is, we have created an equivalent
`(Nn, Kk) long binary code. The bits in each code word are transmitted over
`(Nn, Kk) long binary code. The bits in each code word are transmittC!d over
`the channel by means of PSK or, perhaps. by FSK.
`the channel by means of PSK or, perhaps. by FSK.
`We also indicate that the minimum distance of til, concatenated code is
`We also indicate that the minimum distance of til, concatenated code is
`dm;nDm;m where Dm'n is the minimum distance of the outer code and d""n is the
`dm;nDm;m where Dm'n is the minimum distance of the outer code and d""n is the
`minimum distance of the inner code. Furthermore, the rate of 'he t'oneaten(cid:173)
`minimum distance of the inner code. Furthermore, the rate of 'he t'oneaten(cid:173)
`ated code is KkiNfl, which is equal to the product of the two code rates.
`ated code is KkiNfl, which is equal to the product of the two code rates.
`A hard-decision decoder for a concatenated code is conveniently separated
`A hard-decision decoder for a concatenated code is conveniently separated
`into an inner decoder and an outer decoder. The inner decoder takes the hard
`into an inner decoder and an outer decoder. The inner decoder takes the hard
`decisions on each group of n bits, corresponding to a code word of the inner
`decisions on each group of n bits, corresponding to a code word of the inner
`code, and makes a decision on the k information bits based on maximum(cid:173)
`code,