JP2008182442A - Decoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce a circuit scale of a decoding device for performing decoding in a soft input and a soft output. <P>SOLUTION: An SISO decoder 67 decodes input soft value data Y<SB>t</SB>being data of a soft value which is obtained when encoding data X<SB>t</SB>obtained by encoding input data i<SB>t</SB>, routed through a channel into output soft value data λ<SB>t</SB>being data of a soft value which expresses the likelihood of the input data i<SB>t</SB>. The decoder 67 adopts a first γ memory 203 to a fourth γ memory 206, and a β memory 213 which are constituted by using registers and simultaneously perform data writing/reading as storage means for storing data used for prescribed calculation to obtain the output soft value data λ<SB>t</SB>from the input soft value data X<SB>t</SB>. The present invention is applicable to an optical disk device, etc., for example. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a decoding device, and in particular, reduces the circuit scale of a soft input / soft output decoding device that decodes input soft value data, which is soft value data, into output soft value data, which is other soft value data. The present invention relates to a decoding device that can perform the above.

  When data is recorded on a recording medium such as a disk, the data is encoded and recorded for error correction or the like.

  FIG. 1 shows an example of the configuration of a conventional optical disc apparatus.

  In the optical disc apparatus of FIG. 1, data to be recorded to be recorded on an optical disc (not shown) is supplied to the Reed-Solomon encoder 11 when data is recorded.

  Here, the data to be recorded is a bit string, and each bit takes a value of 0 or 1.

  The Reed-Solomon encoder 11 encodes data to be recorded into a Reed-Solomon code, and supplies a Reed-Solomon code as a bit string that takes one of 0 or 1 to the modulation encoder 12.

  The modulation encoder 12 modulation-codes the Reed-Solomon code from the Reed-Solomon encoder 11 into a modulation code such as a 17PP code (Parity Preserve / Prohibit Repeated Minimum Transition Runlength), for example. A 17PP code as a bit string having one of these values is supplied to an NRZI (Non Return to Zero Inversion) converter 13.

  Here, the modulation encoding by the modulation encoder 12 can be performed using, for example, a conversion table in which Reed-Solomon codes and 17PP codes are associated with each other.

  The 17PP code is a code in which any number of 0s in the range of 1 to 7 appears between 1 and 1, and 1s are not consecutive (Run Length Limited Code (1, 7)).

  The NRZI converter 13 performs NRZI conversion on the 17PP code from the modulation encoder 12, and an NRZI signal obtained as a result is supplied to the recording / reproducing system 14. The recording / reproducing system 14 writes the NRZI signal from the NRZI converter 13 in the form of pits on an optical disk (not shown).

  Here, in the NRZI conversion, when the input to the NRZI converter 13 is 0, the immediately preceding output is output as it is as an NRZI signal, and when the input is 1, the immediately preceding output is inverted, Output as NRZI signal.

  Therefore, when the input to the NRZI converter 13 is not a bit string in which at least one 0 is inserted between 1 and 1 as in a 17PP code, for example, when 1 is a continuous bit string 11111111, NRZI The output of the converter 13 is a bit string 10101010, and pits corresponding to the bit string 10101010 are formed on the optical disc. The reproduction signal reproduced from the optical disc for the pit corresponding to the bit string 10101010 is a signal having a small amplitude that switches between the H (High) level and the L (Low) level in a short cycle corresponding to 1 bit. Therefore, it becomes difficult to detect the signal level of the reproduction signal.

  On the other hand, when the input to the NRZI converter 13 is a 17PP code in which at least one 0 is inserted between 1 and 1, for example, 10101010 in which 1 and 0 are alternately arranged, NRZI conversion The output of the device 13 is 11001100, and pits corresponding to the bit string 11001100 are formed on the optical disc. And, for the pit corresponding to the bit string 11001100, the reproduction signal reproduced from the optical disc becomes a signal with a somewhat large amplitude in which the H level and the L level are switched in a somewhat long cycle corresponding to 2 bits. It becomes easy to detect the signal level of the reproduction signal.

  When the input to the NRZI converter 13 is a bit string in which 0 is continuous over a long period of time, the reproduction signal reproduced from the optical disc is a signal whose level change is gentle over a long period of time. Since there is a problem that it takes time to lock a PLL (Phase Lock Loop) that generates a clock signal necessary for the operation of each block of the optical disk apparatus or the PLL does not lock, the input to the NRZI converter 13 is 17PP. Like the code, the maximum number of consecutive zeros is also limited to a certain value.

  Next, in the optical disc apparatus of FIG. 1, when reproducing data, the recording / reproducing system 14 reproduces a reproduction signal from the optical disc. The recording / reproducing system 14 includes an A / D (Analog / Digital) converter and an equalizer (not shown), A / D-converts a reproduction signal from the optical disk, and performs processing such as waveform equalization. To output soft data. In addition, the recording / reproducing system 14 includes a PLL circuit (not shown), which generates a clock signal for operating each block of the optical disc apparatus based on the reproduced signal.

  The soft value data output from the recording / reproducing system 14 is supplied to the Viterbi decoder 16.

  Here, in the optical disk apparatus of FIG. 1, the recording / reproducing system 14 that performs recording / reproducing on an optical disk includes a PR (Partial Response) channel such as a PR121 channel, a PR1221 channel, or a PR12221 channel, an AWGN (Additive White Gaussian Noise) channel, and the like. Can be thought of as consisting of

  Therefore, the NRZI signal output from the NRZI converter 13 is added with noise through the PR channel and further through the AWGN channel in the recording / reproducing system 14, and is output as soft value data.

  The PR channel can be represented by a trellis represented by a state (state) that transitions with time and a branch that represents a transition from a state at a certain time t-1 to a state at the next time t.

  The channel sys # 11 including the NRZI converter 13 in the PR channel that can be represented by the trellis can also be represented by the trellis.

  Soft value data read out via a channel that can be represented by a trellis can be decoded into hard value data by maximum likelihood decoding.

  Therefore, the Viterbi decoder 16 decodes the soft value data from the recording / reproducing system 14 into a 17PP code which is hard value data by Viterbi decoding which is one of the maximum likelihood decodings, and supplies it to the modulation decoder 17. To do.

  The modulation decoder 17 uses, for example, a conversion table in which a Reed-Solomon code and a 17PP code are associated with each other, for example, a 17PP code that is a bit string having a value of 0 or 1 from the Viterbi decoder 16. The Reed-Solomon code is decoded and supplied to the Reed-Solomon decoder 18.

  The Reed-Solomon decoder 18 converts the Reed-Solomon code, which is a bit string having a value of 0 or 1 from the modulation decoder 17, into the original data having a value of 0 or 1. Decode to (data to be recorded) and output.

  By the way, as described above, the Viterbi decoder 16 decodes soft value data into hard value data. However, in recent years, a soft-input soft-output, which decodes soft-value data into other soft-value data. There is an increasing demand for adopting a soft-input / soft-output decoding device that performs decoding.

  Examples of the implementation algorithm of the soft input / soft output decoding device include a BCJR (Bahl-Cocke-Jeinek-Raviv) algorithm, a Max-Log-BCJR algorithm, and a SW-Max-Log-BCJR algorithm.

  For example, the BCJR algorithm is known as a decoding method that minimizes the symbol error rate after decoding a convolutional code (see Non-Patent Document 1, for example).

  The BCJR algorithm can output the likelihood of each symbol instead of outputting each symbol as a decoding result.

  Such symbol likelihood output is called soft-output.

  In recent years, research has been conducted to reduce the symbol error probability by making the decoding output of the inner code of the concatenated code and the decoding output of each iteration of the iterative decoding method soft, and decoding such soft output As a decoding method suitable for soft input / soft output decoding, the BCJR algorithm has been attracting attention.

  FIG. 2 shows a communication channel model (communication channel model) for explaining the BCJR algorithm.

  The communication channel model shown in FIG. 2 includes a convolutional encoder 21, a memoryless communication channel 22, and a decoder 23, which convolutionally encodes digital information (data) and passes through a noiseless communication channel (channel) with noise. Decode the data obtained.

That is, the convolutional encoder 21, the input data i _t of the target convolutional coding is supplied. Here, t represents the time, the input data i _t is the time t, the data of the target convolutional coding is supplied to the convolutional encoder 21.

Convolutional encoder 21 performs convolution encoding input data i _t supplied thereto, and outputs the encoded data X _t as convolutional encoding result of the input data i _t. The encoded data X _t passes through the memoryless communication path 22 to be input soft value data Y _t which is soft value data, and is supplied to the decoder 23. The decoder 23, an input soft value data Y _t, decoding the soft-input soft output decoding is performed on the output soft value data lambda _t is a data soft values representing the likelihood of the input data i _t.

  FIG. 3 shows a trellis representing the convolutional encoder 21 of FIG.

Trellis, as described above, is represented by a state S _t a transition over time, the branch from state S _t-1 of a time t-1 represents a transition to state S _t at the next time t The

  The convolutional encoder 21 includes a shift register (not shown) and performs convolutional encoding using the shift register. The number of trellis states (the number of states at time t) M is determined by the convolutional encoder 21. It is equal to the number of contents that the built-in shift register can take (number of patterns of values stored in the shift register).

In the trellis shown in FIG. 3, the state S _{t-1 at} time t-1 is the state m ′ (m ′ = 0, 1,..., M−1), and at the next time t, the input data i relative _t, is observed coded data X _t, the state S _t at time t is the state m (m = 0,1, ···, M-1) is a transition to.

Here, at time t, the state transition probability p _t (m | m ′) for transitioning from the state m ′ to the state m is expressed by Expression (1).

・・・（１）

... (1)

  Pr {A | B} is a condition under which event B occurs, and represents a conditional probability that event A will occur.

Now, convolutional coding starts from state S ₀ = 0 at time t = 0, and outputs a sequence (time series) X ₁ ^τ = X ₁ , X ₂ ,..., X _τ of encoded data X _t Then, it is assumed that the process ends in the state S _τ = 0 at time t = τ.

Further, the memoryless channel 22 receives the sequence Y ₁ ^τ = Y ₁ , Y ₂ ,..., Y _τ of the input soft value data Y _t with respect to the input of the encoded data (sequence) X ₁ ^τ. It will output.

  Here, the transition probability of the memoryless communication path 22 with noise is defined by a function R (|) satisfying the expression (2) for all times t = 1, 2,.

・・・（２）

... (2)

Here, the function R (Y _j | X _j ) represents the conversion probability that the encoded data X _{j at} time t = _j is converted into the input soft value data Y _j through the memoryless communication path 22. .

The possible values for input data i _t is, for example, when a is 0 or 1, the input soft value data ^{_{(series) Y 1 τ (= Y 1}} , Y 2, ···, Y τ) is observed when it is, the lambda _t (output soft values data is data soft values representing the) likelihood of the input data i _t at time t, when the input soft value data Y ₁ ^tau is observed, the input data i _The probability Pr {i _t = 1 | Y ₁ ^τ } that _t is 1 and the probability Pr {i _t = 0 | Y that the input data it is 0 when the input soft value data Y ₁ ^τ is observed ₁ ^τ } and is expressed by Equation (3).

・・・（３）

... (3)

In order to obtain the output soft value data λ _t of Equation (3), the probabilities α _t (m), β _t (m), and γ _t (m ′, m, i) are defined by the following equations.

・・・（４）

... (4)

・・・（５）

... (5)

・・・（６）

... (6)

Equation (4) of the probability alpha _t (m) is input soft value data Y 1 is the time at the beginning of the entire sequence Y ₁ ^tau of _t, during the period until a certain time t, the input soft value data sequence Y _{1 ^{t = Y 1, Y 2,}} ···, when Y _t is observed, in the trellis at time t, being in state m (S _t = m) probability _{Pr {S t = m; Y} 1 t} Hereinafter, this is referred to as a forward probability as appropriate. Pr {A; B} represents the probability that events A and B will occur together.

The probability β _t (m) in the equation (5) is a sequence of input soft value data Y _t from a certain time t + 1 to the last time τ of the entire sequence Y ₁ ^τ of the input soft value data Y _{t. +1} ^τ = Y _{t + 1} , Y _{t + 2} , ..., Y _{τ When τ} is observed, the probability that the trellis is in state m at time t (at time t, in state m and then Of the input soft value data Y _{t + 1} ^τ = Y _{t + 1} , Y _{t + 2} ,..., Y _τ ) from time t + 1 to time τ) Pr {Y _{t + 1} ^τ | S _t = m}, hereinafter referred to as a backward probability as appropriate.

The probability γ _t (m ′, m, i) in equation (6) is in the state m ′ at time t−1 and the input data it at the next time _t is a certain value i in the trellis. , A certain input soft value data Y _t is observed, and at time t, the probability of transition to state m (branch metric) Pr {S _t = m; Y _t ; i _t = i | S _t-1 = m '} Yes, hereinafter referred to as output transition probability as appropriate.

The forward probability α _t (m) is calculated from the input soft value data Y ₁ ^t = Y ₁ , Y ₂ ,..., Y _t from the state S ₀ = 0 at the start of encoding (t = 0). Based on the passage probability of each state 0, 1,..., M-1 at time t-1 calculated in time series.

The backward probability β _t (m) is determined from the state S _τ = 0 at the end of encoding (t = τ), and the sequence of input soft value data Y _{t + 1} ^τ = Y _{t + 1} , Y _{t + 2} ,..., Y _τ , corresponding to the passing probability of each state 0, 1,.

Further, output transition probability _{γ t (m ', m,} i) is an input soft value data Y _t at time t, the input data i _t is calculated based on the input probability value i, at time t , Corresponding to the reception probability of the output of each branch of the trellis from the state m ′ to the state m.

Here, FIG. 4 shows the output transition probability γ _t (m ′, m, i) when the number M of states is four.

Here, as described above, since the possible values for input data i _t is 2 kinds of 0 or 1, from one state S _t-1 with a time t-1, at time t the state to S _t There are two ways of transition. Then, since the number M of states is four, from a state S _t-1 at time t-1, how the transition to state S _t at time t, that is, from a state S _t-1 at time t-1 branch to state S _t at the time t is located only ways 8 (= 4 2 types ×), output transition probability _{γ t (m ', m,} i) is the probability of passing through respective branches of the eight To express.

The output soft value data λ _{t in} Expression (3) uses the forward probability α _t (m), the backward probability β _t (m), and the output transition probability γ _t (m ′, m, i). Can be calculated by:

・・・（７）

... (7)

Meanwhile, forward probability alpha _t (m) at time t in equation (4), and its one time before the time t-1 of the forward probability alpha _t-1 (m '), the input data i _t at time t is a value when a i, 'from the output transition probability gamma _t a transition to a state S _t = m at time t (m' state S _t-1 = m at time t-1, m, i) and with, It can be expressed by equation (8).

・・・（８）

... (8)

Here, in the channel model of FIG. 2, as described above, the convolutional coding is started from the state S ₀ = 0 at time t = 0, so the forward probability α ₀ (m) at time t = 0 is And represented by equation (9).

・・・（９）

... (9)

In addition, the backward probability β _t (m) at time t in Expression (5) is the backward probability β _{t + 1} (m ′) at time t + 1 one hour later and the input data i _{t at} time t + 1. _{When +1} is the value i, the output transition probability γ _{t + 1} (m, m ′, i from the state S _t = m at time t to the state S _{t + 1} = m ′ at time t + 1 ) And can be represented by the formula (10).

・・・（１０）

... (10)

Here, in the channel model of FIG. 2, as described above, the convolutional coding ends in the state S _τ = 0 at time t = τ, so the backward probability β _τ (m) at time t = τ is It is represented by Formula (11).

・・・（１１）

(11)

Furthermore, expression output transition probability gamma _t of (6) (m ', m , i) , the time t-1 of state S _t-1 = m' from the state transition to transition to state S _t = m at time t probability p _t (m | m ') and (equation (1)), coded data (state obtained by convolutionally encoding the input data i _t value i S _t-1 = m' from a state S _t = Conversion probability R (Y _t ) in which the encoded data (X) output from the convolutional encoder 21 at the time of transition to m is converted to the input soft value data Y _t at time t through the memoryless communication path 22 | X) is used to express the equation (12).

・・・（１２）

(12)

In the decoder 23 of FIG. 2, the output soft value data λ _t of the equation (7) is obtained according to the equations (8) to (12).

That is, every time the input soft value data Y _t is received, the decoder 23 obtains the output transition probability γ _t (m ′, m, i) according to the equation (12), and further outputs the output transition probability γ. _{Using t} (m ′, m, i), the forward probability α _t (m) is obtained according to Equation (8) (Equation (9)).

Further, when the decoder 23 receives all of the input soft value data series Y ₁ ^τ , the decoder 23 uses the output transition probability γ _{t + 1} (m ′, m, i) to obtain the equation (10) (equation (11)). Accordingly, the backward probability β _t (m) is obtained.

Then, the decoder 23 performs the forward probability α _t (m) determined for all times t = 1,..., Τ and all states m = 0, 1,. Using the probability β _t (m) and the output transition probability γ _t (m ′, m, i), the output soft value data for all times t = 1,. Find _λt .

According to the BCJR algorithm, the output soft value data λ _t is obtained as described above.

  By the way, in the BCJR algorithm, the calculation amount is large because it includes product calculation. Further, in the BCJR algorithm, since the encoded data needs to end at time t = τ, it is difficult to decode encoded data (continuous data) that lasts longer.

  Therefore, Max-Log-BCJR algorithm and Log-BCJR algorithm have been proposed as methods of soft-input / soft-output decoding that eliminate the product operation and reduce the amount of computation (see, for example, Non-Patent Document 2). In Non-Patent Document 2, the Max-Log-BCJR algorithm or the Log-BCJR algorithm is called a Max-Log-MAP algorithm or a log-MAP algorithm, respectively.

  Further, as a soft input / soft output decoding method for decoding continuous data, a SW-BCJR algorithm that performs sliding window processing has been proposed (see, for example, Patent Document 1 and Non-Patent Document 3).

  Hereinafter, the Max-Log-BCJR algorithm, the Log-BCJR algorithm, and the SW-BCJR algorithm will be briefly described.

  First, in the Max-Log-BCJR algorithm, the probability is expressed as a logarithm (natural logarithm) with e being the Napier's constante, and the product of the probabilities is expressed as a logarithm as shown in Equation (13). The probability sum operation is approximated by a logarithmic maximum value operation as shown in Expression (14).

・・・（１３）

... (13)

・・・（１４）

(14)

  Note that max (x, y) represents selecting the maximum value among x and y.

Now, α _t ′ (m) in Expression (15) is used as the forward probability (data representing), β _t ′ (m) in Expression (16) is used as the output probability ( Γ _t ′ (m ′, m, i) in the equation (17) is respectively adopted as the data).

・・・（１５）

... (15)

・・・（１６）

... (16)

・・・（１７）

... (17)

In this case, the forward probability α _t ′ (m) of Equation (15) can be approximated by Equation (18) by substituting Equation (8) into Equation (15).

・・・（１８）

... (18)

Further, the backward probability β _t ′ (m) of Expression (16) can be approximated by Expression (19) by substituting Expression (10) into Expression (16).

・・・（１９）

... (19)

Furthermore, the output transition probability γ _t ′ (m ′, m, i) of Expression (17) can be expressed by Expression (20) by substituting Expression (12) into Expression (17).

・・・（２０）

... (20)

  Here, in Expression (18) and Expression (19), max for i = 0,1 (max where i = 0,1 is described in the lower part) is a value that follows, and variable i is 0. Represents the selection of the maximum value among the value taken when, and the value taken when the variable i is 1.

In Expressions (18) and (19), max for m ′ represents that the subsequent value selects the maximum value among the values taken for the variable m ′ of each value. Incidentally, m 'max of about, m' status as possible values for 0, 1, ..., of the M-1, when there is an input data i _t = i, a transition to state m It is obtained for the state m ′ in which there exists.

Similarly to the equations (15) to (17), if λ _t ′ of the equation (21) is adopted as the output soft value data, the output soft value data λ _t ′ of the equation (21) By substituting equation (7) into 21), it can be approximated by equation (22).

・・・（２１）

... (21)

・・・（２２）

(22)

  Here, in Equation (22), max for m = 0, 1,..., M-1 is the value that follows, when the variable m is 0, 1,. Indicates that the maximum value among the values taken in is selected.

In Expression (22), max for m ′ indicates that the subsequent value selects the maximum value among the values taken for the variable m ′ of each value. Note that max for m ′ in the first term on the right side of the equation (22) is that the input data i _t = 1 in the states 0, 1,..., M−1 as possible values of m ′. When there is a state m ′, a transition to the state m is obtained as a target. Further, 'the max for, m' second term of m states as possible values for 0, 1, ..., of the M-1, when there is an input data i _t = 0, the state It is obtained for a state m ′ where a transition to m exists.

In the decoder 23 of FIG. 2, the output soft value data λ _t of the equation (22) can be obtained using the equations (18) to (20).

That is, every time the input soft value data Y _t is received, the decoder 23 obtains the output transition probability γ _t ′ (m ′, m, i) according to the equation (20), and further outputs the output transition probability. Using γ _t ′ (m ′, m, i), the forward probability α _t ′ (m) is obtained according to equation (18).

Further, when the decoder 23 receives all of the input soft value data series Y ₁ ^τ , the decoder 23 uses the output transition probability γ _{t + 1} ′ (m ′, m, i) to follow the backward probability according to the equation (19). Find β _t '(m).

The decoder 23 then determines the forward probabilities α _t ′ (m) obtained for all times t = 1,..., Τ and all states m = 0, 1,. Using the backward probability β _t ′ (m) and the output transition probability γ _t ′ (m ′, m, i), the output for all times t = 1,. The soft value data λ _t ′ is obtained.

According to the Max-Log-BCJR algorithm, the output soft value data λ _t ′ is obtained as described above.

  Since the Max-Log-BCJR algorithm does not include product operation, the amount of calculation can be greatly reduced compared to the BCJR algorithm.

By the way, the following equation holds for the logarithm log (e ^x + e ^y ) of the sum operation e ^x + e ^y .

・・・（２３）

... (23)

Since log (1 + e ^{− | xy |} ) on the right side of Expression (23) is a one-dimensional function having | xy | as a variable, each value has a variable | xy | and a variable | xy | function ^{log (1 + e - | xy} |) values (function values) and the association table (hereinafter referred to as function table) Once you have prepared, the sum operation e ^x + e ^y logarithm log ( e ^x + e ^y ) can be accurately obtained according to equation (23) using a function table.

  In the Log-BCJR algorithm, the sum operation of the probabilities approximated by Expression (14) in the Max-Log-BCJR algorithm is accurately obtained according to Expression (23).

Therefore, in the Log-BCJR algorithm, compared with the Max-Log-BCJR algorithm, the amount of calculation for the addition of log (1 + e ^{− | xy |} ) on the right side of Expression (23) increases. Similar to the -BCJR algorithm, the product operation is not required, and the output soft value data λ _t ′ obtained by the equation (22) is expressed by log (1 + e ^{− | xy on the} right side of the equation (23) according to the function table. If it is possible to ignore the quantization error that occurs when obtaining ^| ), it matches the logarithm of λ _t in equation (7) obtained by the BCJR algorithm.

  Next, the SW-BCJR algorithm will be described.

In the BCJR algorithm, the backward probability β _t (m) can be obtained only after the entire input soft value data series Y ₁ ^τ has been received, so the encoded data must be terminated at time t = τ. Therefore, it is difficult to decode continuous data.

Therefore, in the SW-BCJR algorithm, as with Viterbi decoding, a truncation length D is introduced, and the backward probability β _t (m) at time t after the truncation length D from time tD is set as an initial value, for example, 1 / By giving M (1 / M number of states), output soft value data λ _{tD at} time _tD is obtained.

  That is, FIG. 5 is a diagram for explaining the SW-BCJR algorithm.

The SW-BCJR algorithm, the initialization of the forward probability alpha ₀ at time t = 0 '(m) is, for example, like the BCJR algorithm, after performing in accordance with equation (9), an input soft value data Y _t Each time it is received, the following processing is performed, whereby output soft value data is obtained for continuous data.

That is, in the SW-BCJR algorithm, when the input soft value data Y _{t at} time t is received, the output transition probability γ _t (m ′, m, i) is obtained according to the equation (12).

Further, the backward probability β _t (m) is initialized to 1 / M for all states m = 0, 1,..., M−1 without using the equation (11).

Then, the backward probability β _tD (m) of the time (previous time) tD that is back in time from the time t by the truncation length D is the output transition probability γ _{t−D + 1} (m ′, m, i), γ _{t-D + 2} (m ', m, i), ..., γ _t (m', m, i) and initial value 1 / M of backward probability β _t (m) Is obtained according to the equation (10).

Furthermore, the backward probability β _tD (m) at time tD, the forward probability α _tD-1 (m) at time tD-1 obtained at time t-1 one time before, and further, the truncation length D before Using the output transition probability γ _tD (m ′, m, i) at the time tD obtained when the input soft value data Y _tD at the time _tD is received at the time tD, the time tD Output soft value data λ _tD is obtained (FIG. 5).

・・・（２４）

... (24)

Then, using the forward probability α _tD-1 (m) at time _tD-1 and the output transition probability γ _tD (m ′, m, i) at time tD, according to the equation (8), the forward probability at time tD The probability α _tD (m) is obtained. This forward probability α _tD (m) at time tD is calculated by the equation (24) when the input soft value data Y _{t + 1} at the next time _{t + 1} is received. Used to obtain value data λ _{t-D + 1} .

  Non-Patent Document 3 includes the SW-BCJR algorithm, the SW-Log-BCJR algorithm that combines the SW-BCJR algorithm and the Log-BCJR algorithm, and the SW-BCJR algorithm and the Max-Log-BCJR algorithm. SW-Max-Log-BCJR algorithm (referred to as SWAL-BCJR algorithm in Non-Patent Document 3) is also described.

  According to the SW-Log-BCJR algorithm and the SW-Max-Log-BCJR algorithm, continuous data can be decoded without product operation.

Hereafter, the same applies to the output transition probability γ _t (m ′, m, i) (γ _t ′ (m ′, m, i) which is the logarithm of γ _t (m ′, m, i)). ) Is simply written as γ. Similarly, the logarithm of the forward probability _{α t (m) (α t} (m) α t took the logarithm of the '(m) as well) and, α and, backward probability _{β t (m) (β t} (m) the same applies to beta _t (m) took, and beta, output soft value data lambda _t a (lambda _t logarithm lambda _t 'similarly taken in), and lambda, described respectively.

WO99 / 62183 pamphlet Bahl, Cocke, Jelinek and Raviv, "Optimal decoding of liner codes for minimizing symbol error rate", IEEE Trans. Inf. Theory, vol. IT-20, pp. 284-287, Mar. 1974 Robertson, Villebrun and Hoeher, "A comparison of optimal and sub-optimal MAP decoding algorithms operating in the log domain", IEEE Int. Conf. On Communications, pp. 1009-1013, June 1995 Benedetto and Montorsi, "Soft-output decoding algorithms in iterative decoding of turbo codes", TDA progress Report 42-124, Feb, 1996

  When decoding soft inputs and soft outputs, it is necessary to take the output transition probability γ and backward probability β obtained in the past to obtain the output soft value data λ, so implement the BCJR algorithm etc. The decoding device requires a memory for storing the output transition probability γ and the backward probability β.

  As such a memory, a RAM (Random Access Memory) is generally used. However, in a RAM, generally, when the number of words is small, the ratio of overhead other than memory cells storing data increases. That is, in a RAM with a small number of words, the RAM size per word (the size of the entire RAM divided equally by the number of words in the RAM) becomes large.

  Therefore, when the decoding apparatus is implemented with a soft input / soft output decoding algorithm using the truncation length D, such as the SW-Log-BCJR algorithm or the SW-Max-Log-BCJR algorithm, the truncation length D is If it is small (short), a decoding apparatus is configured using a RAM having a small number of words, and the circuit scale of the decoding apparatus becomes relatively large.

  The present invention has been made in view of such a situation, and makes it possible to reduce the circuit scale of a soft input / soft output decoding apparatus.

  A decoding device according to one aspect of the present invention provides a soft value representing the likelihood of the input data by converting the input soft value data, which is soft value data obtained by encoding the input data through the channel. A decoding device that decodes the output soft value data that is the data of the data, a calculation unit that performs a predetermined calculation for obtaining the output soft value data from the input soft value data, and a memory that stores data used for the predetermined calculation And the storage means is a memory configured by using a register and capable of simultaneously writing and reading data.

  In such a decoding device, the storage means for storing data used for the predetermined calculation for obtaining the output soft value data from the input soft value data is configured using a register, and data writing and reading are performed simultaneously. Done.

  According to one aspect of the present invention, the circuit scale of a soft input / soft output decoding device can be reduced.

  Embodiments of the present invention will be described below. Correspondences between the constituent elements of the present invention and the embodiments described in the specification or the drawings are exemplified as follows. This description is intended to confirm that the embodiments supporting the present invention are described in the specification or the drawings. Therefore, even if there is an embodiment which is described in the specification or the drawings but is not described here as an embodiment corresponding to the constituent elements of the present invention, that is not the case. It does not mean that the form does not correspond to the constituent requirements. Conversely, even if an embodiment is described here as corresponding to a configuration requirement, that means that the embodiment does not correspond to a configuration requirement other than the configuration requirement. It's not something to do.

A decoding device according to one aspect of the present invention provides:
Input data (e.g., LDPC codes as the input data i _t) encoded data was encoded (e.g., encoded data X _t) is obtained by passing through the channel (e.g., AWGN channel in Fig. 6) soft value Input soft value data (for example, input soft value data Y _t ) is decoded into output soft value data (for example, output soft value data λ) that is soft value data representing the likelihood of the input data. A decryption device,
Calculation means (for example, a γ calculation unit 101, an α calculation unit 108, a first β calculation unit 109, a second β calculation unit 110, FIG. 10) that performs a predetermined calculation for obtaining the output soft value data from the input soft value data. λ calculation unit 112),
Storage means for storing data used for the predetermined calculation (for example, the first γ memory 203, the second γ memory 204, the third γ memory 205, the fourth γ memory 206, the β memory 213 in FIG. 10);
The storage means is a memory configured by using a register (for example, the registers 303 ₁ to 303 _W in FIG. 11) capable of simultaneously writing and reading data.

One aspect of the decoding device is:
As the storage means,
In the trellis, at time t−1, in the state m ′, when the input data it at the next time _t is a certain value i, certain input soft value data Y _t is observed, and the output transitions to the state m Four γ memories (for example, the first γ memory 203, the second γ memory 204, the third γ memory 205, and the fourth γ memory 206 in FIG. 10) that store data representing the transition probability γ _t (m ′, m, i); ,
When the input soft value data series Y _{t + 1} , Y _{t + 2} ,... Is observed after a certain time t + 1, the backward probability β in the state m at the time t in the trellis. _and one β memory (for example, β memory 213 in FIG. 10) that stores data representing _t (m).

One aspect of the decoding device is:
As the calculation means,
A γ calculation unit (for example, γ calculation unit 101 in FIG. 10) that calculates data representing the output transition probability γ _t (m ′, m, i);
The output transition probability _{γ t (m ', m,} i) using the data representing the said input soft value sequence Y ₁ data, Y _2, · · ·, when Y _t is observed, in the trellis An α calculation unit (for example, α calculation unit 108 in FIG. 10) that calculates data representing the forward probability α _t (m) in the state m at time t;
Using the data representing the output transition probability γ _{t + 1} (m, m ′, i) that is in the state m at time t and transitions to the state m ′ at the next time t + 1, the backward probability β a β calculator that calculates data representing _t (m) (for example, the first β calculator 109 and the second β calculator 110 in FIG. 10);
Data representing the output transition probability γ _t (m ′, m, i), the forward probability α _t-1 (m ′) in the state m ′ at time t−1, and the backward probability β _t (m) using, lambda calculation unit for calculating the output soft values data lambda _t representing the input likelihood that the data i _t is a value i at time t (e.g., lambda calculation unit 112 of FIG. 10) and comprise a Can do.

In this case, the decoding device of one aspect is
Writing of data representing the output transition probability γ _t (m ′, m, i) calculated by the γ calculation unit, calculation of data representing the forward probability α _t (m), and the output soft value data λ A γ memory control unit that controls writing and reading of data to and from the γ memory so that reading of data representing the output transition probability γ _t (m ′, m, i) necessary for calculating _t is performed simultaneously. (For example, the γ memory control unit 202 in FIG. 10);
Writing data representing the backward probability β _t (m) calculated by the β calculating unit and reading data representing the backward probability β _t (m) necessary for calculating the output soft value data λ _t A β memory control unit (for example, a β memory control unit 211 in FIG. 10) that controls writing and reading of data to and from the β memory can be further provided.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 6 shows a configuration example of an optical disc apparatus to which the present invention is applied.

  In the optical disk apparatus of FIG. 6, when data is recorded, data to be recorded to be recorded on an optical disk (not shown) is supplied to the Reed-Solomon encoder 61.

  Here, the data to be recorded is a bit string, and each bit takes a value of 0 or 1.

  The Reed-Solomon encoder 61 encodes data to be recorded into a Reed-Solomon code, and converts the Reed-Solomon code as a bit string having a value of 0 or 1 to an LDPC (Low Density Parity Check) encoder 62. Supply.

The LDPC encoder 62 encodes the Reed-Solomon code from the Reed-Solomon encoder 61 into an LDPC code, and obtains the LDPC code as a bit string having a value of 0 or 1 obtained as a result, as input data as i _t, and supplies the modulation coder 63.

Modulation coder 63, for example, by using a conversion table that associates the LDPC code and 17PP code, an LDPC code as an input data i _t from LDPC encoder 62 modulates encoded into modulation code of 17PP code, etc. , 0 or 1 is supplied to the NRZI converter 64 as a 17PP code as a bit string.

  The NRZI converter 64 performs NRZI conversion on the 17PP code from the modulation encoder 63, and an NRZI signal obtained as a result is supplied to the recording / reproducing system 65. The recording / reproducing system 65 writes the NRZI signal from the NRZI converter 64 in the form of pits on an optical disk (not shown).

Next, in the optical disk apparatus shown in FIG. 6, when reproducing data, the recording / reproducing system 65 reproduces a reproduction signal from the optical disk. As with the recording / reproducing system 14 of FIG. 1, the recording / reproducing system 65 includes an A / D converter, an equalizer, a PLL circuit, etc. (not shown), A / D-converts the reproduction signal from the optical disc, Further, processing such as waveform equalization is performed to output soft value data (Y _t ). In the PLL circuit built in the recording / reproducing system 65, a clock signal for operating each block of the optical disc apparatus is generated based on the reproduced signal.

  Soft value data output from the recording / reproducing system 65 is supplied to a SISO (Soft In Soft Out) decoder 67.

  Here, in the optical disc apparatus of FIG. 6, the recording / reproducing system 65 that performs recording / reproducing with respect to the optical disc can be considered to be composed of a PR channel such as a PR121 channel, a PR1221 channel, or a PR12221 channel, and an AWGN channel.

  Therefore, the NRZI signal output from the NRZI converter 64 becomes a signal subjected to intersymbol interference through the PR channel in the recording / reproducing system 65, and further, noise is added through the AWGN channel to add softness. Output as value data.

  The PR channel can be represented by a trellis, and the channel sys # 21 including the NRZI converter 64 in the PR channel that can be represented by the trellis can also be represented by the trellis.

  Further, the channel from the input to the output of the modulation encoder 63 can be expressed by a valid branch and an invalid branch, that is, a time-variant trellis in which possible transitions change with time.

  A channel sys # 22 including a modulation encoder 63 that can be expressed by a time-varying trellis in a channel sys # 21 that can be expressed by a trellis can also be expressed by a trellis (time-varying trellis).

  Soft value data obtained via (via) a channel that can be represented by a trellis can be decoded into hard value data by maximum likelihood decoding. However, in the optical disc apparatus of FIG. An LDPC encoder 62 is provided immediately after the encoder 61. Therefore, it is necessary to decode the LDPC code immediately before decoding the Reed-Solomon code. Therefore, the optical disk apparatus shown in FIG. 6 is provided with an LDPC decoder 68 described later for decoding the LDPC code.

Since decoding of the LDPC code in the LDPC decoder 68 is performed with a soft value representing the likelihood of the LDPC code as an input (since it is performed for soft-input), the LDPC decoder 68 includes: soft values representing the likelihood of the LDPC code, i.e., where the need to provide as an input soft values (lambda _t) from LDPC encoder 62 represents the likelihood of the LDPC code as input data i _t to the modulation coder 63 There is.

Therefore, the SISO decoder 67 converts the soft value data (input soft value data Y _t ) from the recording / reproducing system 65 into soft value data (output soft value data λ _t ) to be supplied to the LDPC decoder 68 at the subsequent stage. The soft input / soft output to be decoded is decoded, and soft value data obtained by the decoding is supplied to the LDPC decoder 68. The SISO decoder 67 performs soft input / soft output decoding according to, for example, the SW-Log-BCJR algorithm or the SW-Max-Log-BCJR algorithm.

Here, the SISO decoder 67 corresponds to the decoder 23 in FIG. 2, and the AWGN channel of the recording / reproducing system 65 corresponds to the memoryless communication path 22 in FIG. A channel sys # 22 configured by the modulation encoder 63, the NRZI converter 64, and the PR channel of the recording / reproducing system 65 and represented by a trellis corresponds to the convolutional encoder 21 of FIG. Therefore, the channel sys # 22 is an LDPC code as an input data i _t, can be regarded as an encoder for encoding the encoded data X _t (convolutional encoder).

  The LDPC decoder 68 decodes the soft value data from the SISO decoder 67 into a Reed-Solomon code, which is a bit string having a value of 0 or 1, and supplies it to the Reed-Solomon decoder 69.

  The Reed-Solomon decoder 69 converts the Reed-Solomon code, which is a bit string having a value of 0 or 1 from the LDPC decoder 68, into the original data having a value of 0 or 1. Decrypted and output.

Next, FIG. 7, the data of the soft values supplied from the recording and reproducing system 65 of FIG. 6, as the input soft value data Y _t, soft decrypting the input soft value data Y _t to output soft values data lambda _t The example of a structure of the SISO decoder 67 which performs decoding of an input soft output is shown.

  In the SISO decoder 67 of FIG. 7, a single-port RAM is adopted as a memory for storing the output transition probability γ and the backward probability β. As the memory for storing the output transition probability γ, five (5 banks) first γRAM 103, second γRAM 104, third γRAM 105, fourth γRAM 106, and fifth γRAM 107, which are single-port RAMs, are provided, and a backward probability β As a memory for storing the first βRAM 113 and the second βRAM 114 which are two (2 banks) single-port RAMs.

That is, the SISO decoder 67 of FIG. 7 includes a control unit 100, a γ calculation unit 101, a γRAM control unit 102, a first γRAM 103, a second γRAM 104, a third γRAM 105, a fourth γRAM 106, a fifth γRAM 107, an α calculation unit 108, and a first β calculation unit 109. , The second β calculation unit 110, the βRAM control unit 111, the λ calculation unit 112, the first βRAM 113, and the second βRAM 114. For example, according to the SW-Max-Log-BCJR algorithm, the input soft value data Y _t Te, the output soft values data λ as a decoding result of the input soft value data Y _t, and outputs at four times the delay time of 4D terminating length D.

Here, a 5-bank single-port RAM is adopted as a memory for storing the output transition probability γ, and a 2-bank single-port RAM is adopted as a memory for storing the backward probability β. against _t, the output soft values data λ as a decoding result of the input soft value data Y _t, decoding apparatus for outputting at four times the delay time of 4D terminating length D, for example, T. Miyauchi, K. Yamamoto , T. Yokokawa, M. Kan, Y. Mizutani, and M. Hattori, "High-Performance Programmable SISO Decoder VLSI Implementation for Decoding Turbo Codes", in Global Telecommunications Conference, 2001 (GLOBECOM'01), San Antonio, TX, USA, vol. 1, 2001, pp. 305-309.

In the SISO decoder 67 of FIG. 7, the control unit 100 includes a γ calculation unit 101, a γRAM control unit 102, an α calculation unit 108, a first β calculation unit 109, a second β calculation unit 110, a βRAM control unit 111, and a λ calculation unit. 112 cooperate to perform a process for decoding the input soft value data Y _t into the output soft value data λ _t , the γ calculation unit 101, the γ RAM control unit 102, the α calculation unit 108, and the first β calculation unit 109. The second β calculation unit 110, the βRAM control unit 111, and the λ calculation unit 112 are controlled.

Input soft value data Y _t (D101) at time t is supplied from the recording / reproducing system 65 of FIG.

For the input soft value data Y _{t at} time t, for example, the γ calculation unit 101 performs all the branches of the trellis at time t, that is, from the state S _{t-1 at} time _{t−1, according} to the equation (20). Output transition probabilities γ (sets) for all branches representing transitions of _t to state St are calculated and supplied to the γRAM control unit 102 as data D102.

  Here, the bit width of the data D102 is the product [bits] of the number of branches of the trellis targeted by the SISO decoder 67 and the number of bits of the output transition probability γ for one branch.

  The γRAM control unit 102 writes the output transition probability γ supplied from the γ calculation unit 101 into the first γRAM 103, the second γRAM 104, the third γRAM 105, the fourth γRAM 106, or the fifth γRAM 107 in a predetermined order, and the first γRAM 103 to Read control is performed to read out the output transition probability γ from the fifth γRAM 107 in a predetermined order.

  That is, the γRAM control unit 102 performs read / write control with respect to the first γRAM 103 to read / write the output transition probability γ as the data D103. Similarly, the γRAM control unit 102 performs read / write control for reading / writing the output transition probability γ as data D104 for the second γRAM 104, and read / write control for reading / writing the output transition probability γ as data D105 for the third γRAM 105. Read / write control for reading / writing the output transition probability γ as data D106 is performed on the fourth γRAM 106, and read / write control for reading / writing the output transition probability γ as data D107 is performed on the fifth γRAM 107.

  Then, the first γRAM 103 to the fifth γRAM 107 store the output transition probability γ according to the control of the γRAM control unit 102.

  Here, each of the bit widths of the data D103 to D107 is a product [bits] of the number of branches of the trellis and the number of bits of the output transition probability γ for one branch.

  In the first γRAM 103 to the fifth γRAM 107, one word is a product [bits] of the number of branches of the trellis and the number of bits of the output transition probability γ for one branch, and the storage capacity is equal to the number of words equal to the truncation length D. Have.

  The γRAM control unit 102 uses the output transition probability γ read from the first γRAM 103 to the fifth γRAM 107 as data D108, D109, D110, or D115 as an α calculation unit 108, a first β calculation unit 109, a second β calculation unit 110, or λ It supplies to the calculation part 112.

  Here, the bit widths of the data D108, D109, D110, and D115 are all the products [bits] of the number of branches of the trellis and the number of bits of the output transition probability γ for one branch.

The data D108 supplied from the γRAM control unit 102 to the α calculation unit 108 and the data D115 supplied to the λ calculation unit 112 are output transition probabilities γ for all branches at a certain time t _α of the trellis. .

Further, the data D109 supplied from the γRAM control unit 102 to the first β calculation unit 109 is the output transition probability γ for all the branches of the trellis at a certain time t _β1 +1, and the second β calculation is performed from the γRAM control unit 102. The data D110 supplied to the unit 110 is an output transition probability γ for all branches of the trellis at a certain time t _β2 +1.

alpha calculation unit 108, using the output transition probabilities gamma (D108) for all branches of the time t _alpha of the trellis from γRAM control unit 102, for example, according to equation (18), forward to all states at time t _alpha trellis The probability α (set) is calculated and supplied to the λ calculation unit 112 as data D116.

  Here, the bit width of the data D116 is the product [bits] of the number M of trellis states (states at one time) and the number of bits of the forward probability α for one state.

On the other hand, the first β calculation unit 109 uses the output transition probabilities γ (D109) for all branches of the trellis time t _β1 +1 from the γRAM control unit 102, for example, according to the equation (19), the trellis time t _β1. The backward probability β (a set of) for all the states is calculated and supplied to the βRAM control unit 111 as data D111.

Similarly, the 2β calculator 110 also uses the output transition probabilities for all branches of the trellis at time t _.beta.2 +1 from γRAM controller 102 gamma (D110), according to equation (19), the time t _.beta.2 trellis The backward probability β (set of) for all states is calculated and supplied to the βRAM control unit 111 as data D112.

  Here, the bit width of the data D111 and D112 is the product [bits] of the number M of trellis states and the number of bits of the backward probability β for one state.

  The βRAM control unit 111 writes the backward probability β supplied from the first β calculation unit 109 and the second β calculation unit 110 to the first βRAM 113 or the second βRAM 114 in a predetermined order, and the first βRAM 113 or the second βRAM 114. From the above, read control is performed to read the backward probability β in a predetermined order.

  That is, the βRAM control unit 111 performs read / write control with respect to the first βRAM 113 to read / write the backward probability β as the data D113. Similarly, the βRAM control unit 111 performs read / write control for reading / writing the backward probability β as data D114 with respect to the second βRAM 114.

  Then, the first βRAM 113 and the second βRAM 114 store the backward probability β under the control of the βRAM control unit 111.

  Here, the bit widths of the data D113 and D114 are both the product [bits] of the number M of trellis states and the number of bits of the backward probability β for one state.

  The first βRAM 113 and the second βRAM 114 have a storage capacity corresponding to the number of words equal to the truncation length D, where one word is the product [bits] of the number M of trellis states and the number of bits of the backward probability β for one state. Have

  The βRAM control unit 111 supplies the backward probability β read from the first βRAM 113 or the second βRAM 114 to the λ calculation unit 112 as data D117.

Here, the bit width of the data D117 is the product [bits] of the number M of trellis states and the number of bits of the backward probability β for one state. As the data D117, βRAM control unit 111 to λ calculation unit retrospective probability beta fed to 112, a backward probability for all states at time trellis t _alpha beta (set of).

  The λ calculation unit 112 includes an output transition probability γ (D115) at time t supplied from the γRAM control unit 102, a forward probability α (D116) at time t-1 supplied from the α calculation unit 108, and a βRAM control unit 111. Is used to obtain the output soft value data λ at time t, for example, according to the equation (22), and supplies it to the LDPC decoder 68 (FIG. 6) as data D118.

Here, the bit width of the data D118 includes a number of bits of one time of the input data i _t, which is the product [bit] of the number of bits of the output soft values data lambda.

  Next, referring to FIG. 8, read / write control (memory management) processing for the first γRAM 103 to the fifth γRAM 107 by the γRAM control unit 102 of FIG. 7, and read / write control processing for the first βRAM 113 and the second βRAM 114 by the βRAM control unit 111. Will be described.

  Here, as described above, the first γRAM 103 to the fifth γRAM 107, the first βRAM 113, and the second βRAM 114 are all single-port RAMs, and at one time, only one of data writing or reading can be performed. Can not.

  From time t = 0 to D−1, as the processing of step S11, the γRAM control unit 102 calculates the output transition probability γ (D102) supplied from the γ calculation unit 101 from the word with the smallest address in the first γRAM 103 in time order. Write to a large word.

  From time t = D to 2D−1, as the processing of step S12, the γRAM control unit 102 calculates the output transition probability γ (D102) supplied from the γ calculation unit 101 from the word with the smallest address in the second γRAM 104 in time order. Write to a large word.

  From time t = 2D to 3D-1, as the process of step S13, the γRAM control unit 102 calculates the output transition probability γ (D102) supplied from the γ calculation unit 101 from the word with the smallest address in the third γRAM 105 in time order. Write to a large word.

  Further, as the processing of step S13, the γRAM control unit 102 reads the output transition probability γ in order from the word with the largest address to the word with the smallest address in the second γRAM 104, that is, the output transition in the order of going back in time (direction toward the past). The probability γ is read and supplied to the first β calculation unit 109 as data D109.

The first β calculation unit 109 calculates the backward probability β using the output transition probability γ from the γRAM control unit 102. The first β calculation unit 109 calculates the backward probability β (backward probability β _3D (m) at time 3D) at time t = 2D when calculating the backward probability β at time t = 2D to 3D−1. Initialize to 1 / M.

  From time t = 3D to 4D−1, as the processing of step S14, the γRAM control unit 102 calculates the output transition probability γ (D102) supplied from the γ calculation unit 101 from the word with the smallest address in the fourth γRAM 106 in time order. Write to a large word.

  Further, as the processing of step S14, the γRAM control unit 102 reads the output transition probability γ in the order of going back in time from the word with the larger address in the third γRAM 105 in the order of the smaller word, and supplies it as the data D110 to the second β calculator 110. To do.

The second β calculation unit 110 calculates the backward probability β using the output transition probability γ from the γRAM control unit 102. The second β calculator 110 calculates the backward probability β (backward probability β _4D (m) at time 4D) at time t = 3D when calculating the backward probability β at times t = 3D to 4D−1. Initialize to 1 / M.

  Further, as the processing of step S14, the γRAM control unit 102 reads the output transition probability γ in the order of going back in time from the word with the largest address of the first γRAM 103 in the order of the small word, and as the data D109, the first β calculator 109 To supply.

  The first β calculation unit 109 uses the output transition probability γ from the γRAM control unit 102 (and the backward probability β after the time obtained by the first β calculation unit 109 in step S13) to determine the backward probability β. It is calculated and supplied to the βRAM control unit 111.

  The βRAM control unit 111 supplies the backward probability β from the first β calculation unit 109 to the first βRAM 113 as the data D113, and the backward probability β is changed from the word with the larger address of the first βRAM 113 to the smaller word in order of going back in time. Write.

  From time t = 4D to 5D−1, as the processing of step S15, the γRAM control unit 102 calculates the output transition probability γ (D102) supplied from the γ calculation unit 101 from the word with the smallest address in the fifth γRAM 107 in time order. Write to a large word.

  Further, as the processing of step S15, the γRAM control unit 102 reads the output transition probability γ in the order from the word with the largest address of the fourth γRAM 106 to the word with the smallest address, and supplies it as the data D109 to the first β calculation unit 109.

The first β calculation unit 109 calculates the backward probability using the output transition probability γ from the γRAM control unit 102. The first β calculating unit 109 calculates the backward probability β (backward probability β _5D (m) at time 5D) at time t = 4D when calculating the backward probability β at time t = 4D to 5D−1. Initialize to 1 / M.

  Further, as the processing of step S15, the γRAM control unit 102 reads the output transition probability γ in order from the word with the largest address of the second γRAM 104 to the word with the smallest address, and supplies it as the data D110 to the second β calculation unit 110.

  The second β calculation unit 110 uses the output transition probability γ from the γRAM control unit 102 (and the backward probability β of the time after being obtained by the second β calculation unit 110 in step S14) to determine the backward probability β. It is calculated and supplied to the βRAM control unit 111.

  The βRAM control unit 111 supplies the backward probability β from the second β calculation unit 110 to the second βRAM 114 as data D114, and the backward probability β is changed from the word with the larger address of the second βRAM 114 to the smaller word in order of going back in time. Write.

  At the same time, the γRAM control unit 102 reads the output transition probability γ in the order from the word with the smallest address in the first γRAM 103 and supplies it to the α calculation unit 108 as data D108 and to the λ calculation unit 112 as data D115. Supply.

  The α calculation unit 108 calculates the forward probability α using the output transition probability γ supplied from the γRAM control unit 102 as data D108, and supplies the forward probability α to the λ calculation unit 112 as data D116.

  At this time, the βRAM control unit 111 reads the backward probability β in the order from the word with the smallest address of the first βRAM 113 to the large word, that is, reads the backward probability β in the order of time, and sends it to the λ calculation unit 112 as data D117. Supply.

  The λ calculation unit 112 is supplied from the γRAM control unit 102 as data D115, the output transition probability γ supplied from the α calculation unit 108 as data D116, and the βRAM control unit 111 as data D117. The output soft value data λ is calculated using the backward probability β and is output as data D118.

  From time t = 5D to 6D−1, as the processing of step S16, the γRAM control unit 102 calculates the output transition probability γ (D102) supplied from the γ calculation unit 101 from the word with the smallest address in the first γRAM 103 in time order. Write to a large word.

  Further, as the processing of step S16, the γRAM control unit 102 reads the output transition probability γ in the order of going back in time from the word with the largest address of the fifth γRAM 107 in the order of the small word, and supplies it to the second β calculator 110 as the data D110. To do.

  The second β calculation unit 110 calculates the backward probability β using the output transition probability γ from the γRAM control unit 102. In calculating the backward probability β at time t = 5D to 6D−1, the second β calculation unit 110 initializes the backward probability β to 1 / M at time t = 5D.

  Further, as the processing of step S16, the γRAM control unit 102 reads the output transition probability γ in the order of going back in time from the word with the largest address to the word with the smallest address in the third γRAM 105, and uses the first β calculation unit 109 as the data D109. To supply.

  In the first β calculation unit 109, the backward probability β is calculated using the output transition probability γ from the γRAM control unit 102 and supplied to the βRAM control unit 111.

  The βRAM control unit 111 supplies the backward probability β from the first β calculation unit 109 to the first βRAM 113 as the data D113, and the backward probability β is changed from the word with the larger address of the first βRAM 113 to the smaller word in order of going back in time. Write.

  At the same time, the γRAM control unit 102 reads the output transition probability γ in the order from the word having the smallest address in the second γRAM 104 to the word having the largest address, and supplies the output transition probability γ to the α calculation unit 108 as data D108 and to the λ calculation unit 112 as data D115. Supply.

  The α calculation unit 108 calculates the forward probability α using the output transition probability γ supplied as the data D108 from the γRAM control unit 102, and supplies the forward probability α to the λ calculation unit 112 as data D116.

  At this time, the βRAM control unit 111 reads the backward probability β in the order of time from the word with the smallest address in the second βRAM 114 in the order of the word, and supplies it to the λ calculation unit 112 as data D117.

  The λ calculation unit 112 is supplied from the γRAM control unit 102 as data D115, the output transition probability γ is supplied from the α calculation unit 108 as data D116, and the βRAM control unit 111 is supplied as data D117. The output soft value data λ is calculated using the backward probability β and is output as data D118.

  Thereafter, similarly, read / write control processing for the first γRAM 103 to the fifth γRAM 107 and read / write control processing for the first βRAM 113 and the second βRAM 114 are performed.

  That is, for a variable j having an integer value of 5 or more, at time t = j × D or (j + 1) × D−1, the output transition probability γ (D102) in time order obtained by the γ calculation unit 101 is The first γRAM 103 to the fifth γRAM 107 are written from a word having a small address to a word having a large address.

  Here, the target of writing of the output transition probability γ is the first γRAM 103 when the variable j = 5, the second γRAM 104 when the variable j = 6, and the third γRAM 105 when the variable j = 7. When j = 8, it is the fourth γRAM 106, when the variable j = 9 is the fifth γRAM 107, and when the variable j = 10, it is the first γRAM 103. Thereafter, the same pattern is repeated.

  At the same time, output transition probabilities γ are read out in order from the largest address to the smallest word in the first γRAM 103 to the fifth γRAM 107, and the first β calculation unit 109 or the first The 2β calculation unit 110 calculates a backward probability β that is not stored in the first βRAM 113 and the second βRAM 114 (hereinafter referred to as a backward probability β1 as appropriate). In calculating the backward probability β1, the backward probability β1 is initialized to 1 / M at time t = (i + 5) × D.

  Here, the output transition probability γ used for the calculation of the backward probability β1 is the target of reading the fifth γRAM 107 when the variable j = 5, the first γRAM 103 when the variable j = 6, and the variable j = 7. The second γRAM 104, the variable j = 8 is the third γRAM 105, the variable j = 9 is the fourth γRAM 106, and the variable j = 10 is the fifth γRAM 107. Thereafter, the same pattern is repeated.

  The backward probability β1 is calculated by the second β calculation unit 110 when the variable j = 5, the first β calculation unit 109 when the variable j = 6, and the second β calculation unit when the variable j = 7. 110. Thereafter, the same pattern is repeated.

  At time t = j × D to (j + 1) × D−1, the output transition probability γ in the order of going back in time is read out in the order from the word with the largest address in the first γRAM 103 to the fifth γRAM 107, Using the output transition probability γ, the first β calculation unit 109 or the second β calculation unit 110 calculates the backward probability β (hereinafter referred to as the backward probability β2 as appropriate) stored in the first βRAM 113 and the second βRAM 114. The

  Further, the backward probability β2 is written (stored) in the order of going back in time from the word having the larger address in the first βRAM 113 or the second βRAM 114 to the smaller word.

  Here, the output transition probability γ used for the calculation of the backward probability β2 is the target of reading the third γRAM 105 when the variable j = 5, the fourth γRAM 106 when the variable j = 6, and the variable j = 7. The fifth γRAM 107, the variable j = 8 is the first γRAM 103, the variable j = 9 is the second γRAM 104, and the variable j = 10 is the third γRAM 105. Thereafter, the same pattern is repeated.

  The backward probability β2 is calculated by the first β calculation unit 109 when the variable j = 5, the second β calculation unit 110 when the variable j = 6, and the first β calculation unit when the variable j = 7. 109. Thereafter, the same pattern is repeated.

  Further, the writing probability of the backward probability β2 is the first βRAM 113 when the variable j = 5, the second βRAM 114 when the variable j = 6, and the first βRAM 113 when the variable j = 7. Thereafter, the same pattern is repeated.

  At the same time, the output transition probability γ in the order of time is read from the first γRAM 103 to the fifth γRAM 107 in ascending order of the address, and is supplied to the α calculator 108 as data D108, and as data D115. The λ calculation unit 112 is supplied.

  The α calculation unit 108 calculates the forward probability α using the output transition probability γ supplied as the data D108, and supplies it as the data D116 to the λ calculation unit 112.

  At this time, the backward probability β2 in the order of time is read in the order from the word with the smallest address in the first βRAM 113 or the second βRAM 114, and supplied to the λ calculator 112 as data D117.

  The λ calculation unit 112 calculates the output soft value data λ using the output transition probability γ supplied as the data D115, the forward probability α supplied as the data D116, and the backward probability β2 supplied as the data D117. Are output as data D118.

  Here, the target of reading the output transition probability γ used for the calculation of the forward probability α and the output soft value data λ is the second γRAM 104 when the variable j = 5, and the third γRAM 105 when the variable j = 6. Yes, the variable j = 7 is the fourth γRAM 106, the variable j = 8 is the fifth γRAM 107, the variable j = 9 is the first γRAM 103, and the variable j = 10 is the second γRAM 104. Thereafter, the same pattern is repeated.

  The target of reading the backward probability β2 is the second βRAM 114 when the variable j = 5, the first βRAM 113 when the variable j = 6, and the second βRAM 114 when the variable j = 7. Thereafter, the same pattern is repeated.

  Next, with reference to FIG. 9, read / write control for the first γRAM 103 to the fifth γRAM 107 and read / write control for the first βRAM 113 and the second βRAM 114 will be further described.

  The first from the left in FIG. 9 shows the progress of calculation of the output transition probability γ, the forward probability α, the backward probability β (β1 and β2), and the output soft value data λ.

  That is, in the first from the left in FIG. 9, the horizontal axis indicates the output transition probability γ, the forward probability α, the backward probability β (β1 and β2), and the output soft value, with the left to right direction as the direction of travel of the trellis time. The section (position) of the trellis where the data λ is calculated is represented, and the vertical axis represents the time (the downward direction from the top is the traveling direction of the time).

  In the first line from the left in FIG. 9, the line L11 indicates the progress of calculation of the output transition probability γ.

  For the output transition probability γ, the output transition probability γ at the time t is calculated as the time t elapses.

  In the first from the left in FIG. 9, lines L12 and L13 indicate the progress of calculation of the backward probability β.

  The backward probability β starts after waiting for a time 2 × D that is twice the truncation length D from the timing when the calculation of the output transition probability γ is started (time t = 0). From t = 2D, the time is calculated in the order of time 2D in the order of time.

  Then, the backward probability β for the first time D in the backward probability β for the time 2D (for example, if the backward probability β for the time t = 0 to 2D-1 is the time t = The backward probabilities (D through 2D-1) are discarded without being written into the first βRAM 113 and the second βRAM 114, and thus are not used in the calculation of the output soft value data λ.

  On the other hand, in the backward probability β for the time 2D, the backward probability β for the subsequent time D in the order of going back in time (for example, if the backward probability β for the time t = 0 to 2D−1, the time t = The backward probability (0 to D-1) is written into the first βRAM 113 or the second βRAM 114 for use in the calculation of the output soft value data λ.

  The line L12 indicates the progress of calculation of the backward probability β1 of the backward probability β that is not written to the first βRAM 113 and the second βRAM 114, and the line L13 is the backward probability written to the first βRAM 113 or the second βRAM 114. The progress of the calculation of β2 is shown.

  The backward probabilities β1 and β2 with an initial value of 1 / M are calculated by shifting (sliding) the start timing by the cutoff length D (sliding window processing) as shown by lines L12 and L13.

  In the first line from the left in FIG. 9, the line L14 indicates the progress of calculation of the forward probability α and the output soft value data λ.

  The calculation of the forward probability α and the output soft value data λ is started immediately after the backward probability β (β2) necessary for calculating the output soft value data λ is obtained.

  The second from the left in FIG. 9 shows the progress of data reading / writing from the first γRAM 103 to the fifth γRAM 107.

  In the second from the left in FIG. 9, the horizontal axis represents the addresses of the first γRAM 103 to the fifth γRAM 107, and the vertical axis represents the time as in the first from the left in FIG.

  In the second line from the left in FIG. 9, a line L21 indicates the progress of writing the output transition probability γ to the first γRAM 103 to the fifth γRAM 107.

  The output transition probability γ is written from the first γRAM 103 to the fifth γRAM 107 to the larger word in the order of time from the word with the smallest address.

  In the second line from the left in FIG. 9, lines L22 and L23 indicate the progress of reading from the first γRAM 103 to the fifth γRAM 107 of the output transition probability γ used for calculating the backward probability β.

  Reading the output transition probability γ used for the calculation of the backward probability β from the first γRAM 103 to the fifth γRAM 107 is twice the cutoff length D from the timing when the calculation of the output transition probability γ is started (time t = 0). Start after waiting for time 2xD.

  Then, the reading of the output transition probability γ used for the calculation of the backward probability β is performed in the order of going back in time, that is, from the first γRAM 103 to the fifth γRAM 107 from the large address to the small word, and the output transition for time 2D. The probability γ is read out.

  In the second line from the left in FIG. 9, the line L22 indicates the output transition probability γ for the first time D in the order of going back in time among the output transition probabilities γ for the time 2D (for example, time t = 0 to 2D− If the output transition probability γ is 1, the reading progress status at the time t = D to the output transition probability 2D-1) is shown. The output transition probability γ for the first time D is shown from the left in FIG. This is used for the calculation of the backward probability β1 described in the first line L12.

  Further, the line L23 indicates that the backward transition probability β for the time 2D is the output transition probability γ for the later time D (for example, the output transition probability γ for the time t = 0 to 2D-1 in the order of going back in time). If there is, the progress of reading of the output transition probability (time t = 0 to D-1) is indicated, and the output transition probability γ for the subsequent time D will be described by the first line L13 from the left in FIG. Used for calculating the backward probability β2.

  In the second line from the left in FIG. 9, a line L24 indicates the progress of reading from the first γRAM 103 to the fifth γRAM 107 of the output transition probability γ used in the calculation of the forward probability α and the output soft value data λ.

  The reading of the output transition probability γ used for the calculation of the forward probability α and the output soft value data λ is performed when calculating the forward probability α and the output soft value data λ described in the first line L14 from the left in FIG. In addition, the processing is performed in time order, that is, from the first γRAM 103 to the fifth γRAM 107 with the smallest address to the largest word.

  The third from the left in FIG. 9 (first from the right) shows the progress of data reading / writing with respect to the first βRAM 113 and the second βRAM 114.

  In the third from the left in FIG. 9, the horizontal axis represents the addresses of the first βRAM 113 and the second βRAM 114, and the vertical axis represents the time as in the first from the left in FIG.

  In the third line from the left in FIG. 9, a line L31 indicates the progress of writing the backward probability β to the first βRAM 113 or the second βRAM 114.

  The backward probability β is written from the word with the larger address of the first βRAM 113 or the second βRAM 114 to the smaller word in order of going back in time. The backward probability β written to the first βRAM 113 or the second βRAM 114 is only the backward probability β2 described in the first line L13 from the left in FIG.

  In the third line from the left in FIG. 9, a line L32 indicates the progress of reading from the first βRAM 113 or the second βRAM 114 of the backward probability β2 used for the calculation of the output soft value data λ.

  The backward probability β2 used for the calculation of the output soft value data λ is read in the order of time when calculating the output soft value data λ described in the first line L14 from the left in FIG. 9, that is, the first βRAM 113 or The second βRAM 114 is performed from a small word to a large word.

  Here, as described above, since the first γRAM 103 to the fifth γRAM 107, the first βRAM 113, and the second βRAM 114 are single-port RAMs, data writing and reading cannot be performed at the same time. Only data writing or reading is performed.

  That is, in FIG. 9, for example, when attention is paid to time t = 4D to 5D, as shown in the second from the left in FIG. 9, in the first γRAM 103, the output used for calculating the forward probability α and the output soft value data λ. Only the reading of the transition probability γ (line L24) is used, and in the second γRAM 104, only the reading of the output transition probability γ used for calculating the backward probability β2 (line L23) is output, and the fourth γRAM 106 is used to calculate the backward probability β1. Only the transition probability γ is read (line L22), and in the fifth γRAM 107, only the output transition probability γ is written (line L21).

  Further, as shown third from the left in FIG. 9, in the first βRAM 113, only the backward probability β2 used for the calculation of the output soft value data λ (line L32) is read, and in the second βRAM 114, the backward probability β2 is written ( Only line L31) is carried out respectively.

  As described above, when the RAM (the first γRAM 103 to the fifth γRAM 107, the first βRAM 113, and the second βRAM 114) is employed as the storage means for storing the output transition probability γ and the backward probability β used for calculating the output soft value data λ. As described above, if the truncation length D is small, the number of words is small. Therefore, the SISO decoder 67 is configured using a RAM with a large overhead, and the circuit scale becomes large.

  FIG. 10 shows another configuration example of the SISO decoder 67 of FIG.

  In the figure, portions corresponding to those in FIG. 7 are denoted by the same reference numerals, and description thereof will be omitted below as appropriate.

  That is, the SISO decoder 67 of FIG. 10 is provided with a γ calculation unit 101, an α calculation unit 108, a first β calculation unit 109, a second β calculation unit 110, and a λ calculation unit 112. However, instead of the control unit 100, the control unit 200 replaces the γRAM control unit 102, and the γ memory control unit 202 replaces the five first γRAMs 103 to the fifth γRAM 107 with four first The 1γ memory 203 to the fourth γ memory 206 are provided in place of the βRAM control unit 111, the β memory control unit 211 is provided, and one β memory 213 is provided in place of the two first βRAMs 113 and the second βRAM 114. This is different from the case of FIG.

  Therefore, in FIG. 7, five RAMs (first γRAM 103 to fifth γRAM 107) are required as storage means for storing the output transition probability γ, whereas in FIG. 10, four memories (first γ memory) are required. 203 to the fourth γ memory 206) constitutes the SISO decoder 67. Further, in FIG. 7, two RAMs (first βRAM 113 and second βRAM 114) are required as storage means for storing the backward probability β, whereas in FIG. 10, one memory (β memory 213) is required. Thus, the SISO decoder 67 is configured.

In the SISO decoder 67 of FIG. 10 configured as described above, for example, as in the case of FIG. 7, the input soft value data Y _{t is} subjected to the input soft value according to the SW-Max-Log-BCJR algorithm. output soft values data λ as a decoding result value data Y _t, and outputs at four times the delay time of 4D terminating length D.

That is, in the SISO decoder 67 of FIG. 10, the control unit 200 includes a γ calculation unit 101, an α calculation unit 108, a first β calculation unit 109, a second β calculation unit 110, a λ calculation unit 112, a γ memory control unit 202, and The β memory control unit 211 cooperates so as to perform processing for decoding the input soft value data Y _t into the output soft value data λ _t , a γ calculation unit 101, an α calculation unit 108, a first β calculation unit 109, The second β calculation unit 110, λ calculation unit 112, γ memory control unit 202, and β memory control unit 211 are controlled.

Input soft value data Y _t (D101) at time t is supplied from the recording / reproducing system 65 of FIG.

The γ calculation unit 101 calculates the output transition probability γ for all branches of the trellis at time t in accordance with the equation (20) with respect to the input soft value data Y _t at time t, and as the data D102, the γ memory control unit 202.

  The γ memory control unit 202 writes the output transition probability γ supplied from the γ calculation unit 101 into the first γ memory 203, the second γ memory 204, the third γ memory 205, or the fourth γ memory 206 in a predetermined order. Then, read control is performed to read the output transition probability γ from the first γ memory 203 to the fourth γ memory 206 in a predetermined order.

  That is, the γ memory control unit 202 performs read / write control with respect to the first γ memory 203 to read / write the output transition probability γ as the data D203. Similarly, the γ memory control unit 202 performs read / write control for reading / writing the output transition probability γ as data D204 with respect to the second γ memory 204, and the output transition probability γ as data D205 with respect to the third γ memory 205. Read / write control for reading / writing is performed on the fourth γ memory 206 for reading / writing the output transition probability γ as data D206.

  Then, the first γ memory 203 to the fourth γ memory 206 store the output transition probability γ according to the control of the γ memory control unit 202.

  Here, the bit widths of the data D203 to D206 are all the product [bits] of the number of branches of the trellis and the number of bits of the output transition probability γ for one branch.

  Similarly to the first γRAM 103 to the fifth γRAM 107 in FIG. 7, the first γ memory 203 to the fourth γ memory 206 are each a product of the number of trellis branches and the number of bits of the output transition probability γ for one branch [ Bit] and has a storage capacity equal to the number of words equal to the truncation length D.

  The γ memory control unit 202 uses the output transition probability γ read from the first γ memory 203 to the fourth γ memory 206 as data D108, D109, D110, or D115, as an α calculation unit 108, a first β calculation unit 109, and a second β calculation. To the unit 110 or the λ calculation unit 112.

Here, the data D108 supplied from the γ memory control unit 202 to the α calculation unit 108 and the data D115 supplied to the λ calculation unit 112 are output transition probabilities γ for all branches at a certain time t _α of the trellis. It is.

The data D109 supplied from the γ memory control unit 202 to the first β calculation unit 109 is the output transition probability γ for all branches of the trellis at a certain time t _β1 +1. The data D110 supplied to the 2β calculation unit 110 is the output transition probability γ for all branches of the trellis at a certain time t _β2 +1.

The α calculation unit 108 uses the output transition probabilities γ (D108) for all the branches at the time t _α of the trellis from the γ memory control unit 202, and uses the output transition probabilities γ (D108) for all the states at the time t _α of the trellis according to the equation (18). α is calculated and supplied to the λ calculation unit 112 as data D116.

On the other hand, the 1β calculation unit 109, gamma all branches of the trellis at time t _.beta.1 +1 output transition probability gamma using (D 109) for the memory control unit 202, in accordance with equation (19), the trellis at time t _.beta.1 The backward probability β for all states is calculated and supplied to the β memory control unit 211 as data D111.

Similarly, the 2β calculator 110 also, gamma all branches of the trellis at time t _.beta.2 +1 output transition probability gamma using (D110) for from the memory control unit 202, in accordance with equation (19), the time of the trellis t _.beta.2 The backward probability β for all the states is calculated and supplied to the β memory control unit 211 as data D112.

  The β memory control unit 211 writes the backward probability β supplied from each of the first β calculation unit 109 and the second β calculation unit 110 into the β memory 213 in a predetermined order, and from the β memory 213, the backward probability Read control for reading β in a predetermined order is performed.

  That is, the β memory control unit 211 performs read / write control with respect to the β memory 213 to read / write the backward probability β as data D213.

  The β memory 213 stores the backward probability β under the control of the β memory control unit 211.

  Here, the bit width of the data D213 is the product [bits] of the number M of trellis states and the number of bits of the backward probability β for one state.

  Similarly to the first βRAM 113 and the second βRAM 114 in FIG. 7, the β memory 213 is censored by multiplying the number of trellis states M by the number of bits of the backward probability β for one state [bits]. It has a storage capacity equal to the number of words equal to the length D.

  The β memory control unit 211 supplies the backward probability β read from the β memory 213 to the λ calculation unit 112 as data D117.

Here, as the data D117, beta retrospective probability beta is supplied to the λ calculating section 112 from the memory control unit 211, a backward probability beta for all states at time trellis t _alpha.

  The λ calculation unit 112 includes an output transition probability γ (D115) supplied from the γ memory control unit 202, a forward probability α (D116) supplied from the α calculation unit 108, and a backward probability supplied from the β memory control unit 211. Using β, the output soft value data λ is obtained according to the equation (22) and supplied to the LDPC decoder 68 (FIG. 6) as data D118.

  In the SISO decoder 67 of FIG. 10, four first γ memories 203 to fourth γ memory 206 as storage means for storing the output transition probability γ, and one β memory as storage means for storing the backward probability β. Each of the memories 213 is a memory configured by using a register and capable of simultaneously writing and reading data.

  That is, FIG. 11 shows a configuration example of the first γ memory 203 of FIG.

The first γ memory 203 includes an address converter 301, W selectors 302 ₁ to 302 _W , W registers 303 ₁ to 303 _W , and a selector 304.

Here, the number W of the selectors 302 ₁ to 302 _{W and} the number W of the registers 303 ₁ to 303 _W are equal to the cutoff length D.

  An address is supplied to the address converter 301.

  Here, the address is supplied from the γ memory control unit 202 (FIG. 10), and the γ memory control unit 202 controls reading and writing of data with respect to the first γ memory 203 by supplying the address to the first γ memory 203.

As an address, an integer value ranging from 0 to W-1 equal to the cutoff length D-1 is supplied. Therefore, the number of bits of the address is log ₂ W bits.

The address converter 301 supplies the log ₂ W-bit address supplied from the γ memory control unit 202 to a W-bit bit string, in which only one of the W bits is 1 and all other bits are The bit string of 0 is converted into a bit string of 0, the bit string of the W bit is used as a word selection signal, the k-th bit (k-th bit) from the least significant bit of the word selection signal, and the selector 302 _k (k = 1, 2,. ..Supply to W).

That is, if the value represented by the address of log ₂ W bits is A, the address converter 301 is a bit string of W bits in which only the A + 1 bit from the least significant bit is 1 and all other bits are 0. As the word selection signal, the k-th bit of the word selection signal is supplied to the selector 302 _k .

The selector 302 _k (k = 1, 2,..., W) is supplied with the k-th bit of the word selection signal from the address converter 301 and also written into the first γ memory 203 from the γ memory control unit 202. As write data, an output transition probability γ of one word is supplied. Here, if the product of the number of branches of the trellis and the number of bits of the output transition probability γ for one branch is expressed as N, the output transition probability γ of one word is an N-bit bit string, and the trellis branch The output transition probability γ is several minutes.

When the kth bit of the word selection signal from the address converter 301 is 1, the selector 302 _k sets the output transition probability γ of one word (N bits) as write data from the γ memory control unit 202. Select and supply to register 303 _k .

The selector 302 _k selects the output transition probability γ of one word (N bits) latched by the register 303 _k when the k-th bit of the word selection signal from the address converter 301 is 0. , And supplied to the register 303 _k .

The register 303 _k is composed of, for example, flip-flops that can latch the output transition probability γ of one word (N bits), and latch the output transition probability γ of one word supplied from the selector 302 _k. And supplied to the selector 302 _k and the selector 304.

Therefore, when the k-th bit of the word selection signal supplied from the address converter 301 to the selector 302 _k is 1 for a certain address A, the selector 302 _k writes data from the γ memory control unit 202. and 1 word output transition probabilities as a data gamma is selected, is supplied to the register 303 _k, the register 303 _k, gamma output transition probabilities gamma of 1 word from the memory controller 202 is latched. That is, as a result, the output transition probability γ of 1 word is written in the address A.

On the other hand, when the k-th bit of the word selection signal supplied from the address converter 301 to the selector 302 _k is 0 for a certain address A, in the selector 302 _k , the register 303 _k is latched 1 is selected output transition probabilities γ word is supplied to the register 303 _k, the register 303 _k, the output transition probability of the 1-word γ is latched. That is, as a result, the output transition probability γ of one word stored at the address other than the address A is maintained as it is.

The selector 304 is supplied with the output transition probability γ of one word latched by the register 303 _k . Further, the selector 304 is supplied with the same address A as that supplied to the address converter 301 from the γ memory control unit 202 (FIG. 10).

The selector 304 outputs the one-word output transition probability γ latched by the register 303 _{A + 1} as the address A out of the one-word output transition probability γ latched by the _W registers 303 ₁ to 303 _W. γ is selected and supplied to the γ memory control unit 202 as read data read from the address A.

In the first γ memory 203 configured as described above, for example, when the address A and the output transition probability γ of one word as write data are supplied from the γ memory control unit 202, the address A is converted into an address converter. It is supplied to the 301 and the selector 304, the output transition probability γ of one word as the write data is supplied to the W number of selectors 302 ₁ to 302 _W.

The selector 304 to which the address A is supplied selects the one-word output transition probability γ latched by the register 303 _{A + 1} as the address A out of the registers 303 ₁ to 303 _W and reads it from the address A. The read data is supplied to the γ memory control unit 202.

On the other hand, the address converter 301 converts the address A into a W-bit word selection signal in which only the A + 1-th bit from the least significant bit is 1, and all other bits are 0, and the word selection signal The kth bit is supplied to the selector 302 _k .

Among the selectors 302 ₁ to 302 _W , the selector 302 _k to which the word selection signal whose k-th bit is 1 is supplied from the address converter 301, that is, in the selector 302 _{A + 1} , from the γ memory control unit 202, An output transition probability γ of 1 word as write data is selected and supplied to the register 303 _{A + 1} .

In the register 303 _{A + 1} , the output transition probability γ of one word from the selector 302 _{A + 1} is latched, and thereby, the output transition probability γ of one word is written to the address A.

As described above, the first γ memory 203 reads the output transition probability γ of one word written in the address A (the output transition probability γ of one word latched by the selector 302 _{A + 1} ). Output) and the writing of the output transition probability γ of one word to the same address A (the latch of the output transition probability γ of one word as write data in the selector 302 _{A + 1} ) Can be done simultaneously by specification.

The second γ memory 204 to the fourth γ memory 206 and the β memory 213 other than the first γ memory 203 are configured in the same manner as the first γ memory 203. However, the register 303 _k of the β memory 213 is composed of as many flip-flops as can latch the backward probability β of one word.

  Next, referring to FIG. 12, read / write control processing for the first γ memory 203 to the fourth γ memory 206 by the γ memory control unit 202 of FIG. 10, and read / write control processing for the β memory 213 by the β memory control unit 211. Will be described.

  Here, as described above, the first γ memory 203 to the fourth γ memory 206 and the β memory 213 are all configured as shown in FIG. 11, and simultaneously write and read data at one time. This is a memory capable of read modify write (read modify write) in which data can be read from that address and written to that address immediately after the address is specified once.

  From time t = 0 to D−1, as the processing of step S31, the γ memory control unit 202 sets the output transition probability γ (D102) supplied from the γ calculation unit 101 to the addresses of the first γ memory 203 in time order. Write from a small word to a large word.

  From time t = D to 2D−1, as the processing of step S32, the γ memory control unit 202 sets the output transition probability γ (D102) supplied from the γ calculation unit 101 to the addresses of the second γ memory 204 in time order. Write from a small word to a large word.

  From time t = 2D to 3D−1, as the processing of step S33, the γ memory control unit 202 sets the output transition probability γ (D102) supplied from the γ calculation unit 101 to the addresses of the third γ memory 205 in time order. Write from a small word to a large word.

  Further, as the processing of step S33, the γ memory control unit 202 reads the output transition probability γ in order from the word with the largest address of the second γ memory 204 to the word with the smallest address, that is, in order of going back in time (direction toward the past). The output transition probability γ is read out and supplied to the first β calculation unit 109 as data D109.

The first β calculation unit 109 calculates the backward probability β in the order of going back in time using the output transition probability γ from the γ memory control unit 202. The first β calculation unit 109 calculates the backward probability β (backward probability β _3D (m) at time 3D) at time t = 2D when calculating the backward probability β at time t = 2D to 3D−1. Initialize to 1 / M.

  From time t = 3D to 4D−1, as the processing of step S34, the γ memory control unit 202 sets the output transition probability γ (D102) supplied from the γ calculation unit 101 to the addresses of the fourth γ memory 206 in time order. Write from a small word to a large word.

  Further, as the processing of step S34, the γ memory control unit 202 reads the output transition probability γ in the order of going back in time from the word with the largest address of the third γ memory 205 to the word with the smallest address, and as the data D110, the second β calculation unit 110.

The second β calculation unit 110 uses the output transition probability γ from the γ memory control unit 202 to calculate the backward probability β in the order of going back in time. The second β calculator 110 calculates the backward probability β (backward probability β _4D (m) at time 4D) at time t = 3D when calculating the backward probability β at times t = 3D to 4D−1. Initialize to 1 / M.

  Further, as the processing of step S34, the γ memory control unit 202 reads the output transition probability γ in the order of the word having the larger address from the first γ memory 203 in the order of the smaller word, and supplies it to the first β calculation unit 109 as data D109. To do.

  The first β calculation unit 109 uses the output transition probability γ from the γ memory control unit 202 (and the backward probability β obtained by the first β calculation unit 109 in step S33), and in step S33, performs the first β calculation. The backward probability β at a time further back from the backward probability β obtained by the unit 109 is calculated and supplied to the β memory control unit 211.

  The β memory control unit 211 supplies the backward probability β from the first β calculation unit 109 to the β memory 213 as data D213, and the backward probability β from the word with the larger address in the β memory 213 in order of going back in time. Write to a small word.

  From time t = 4D to 5D−1, as the processing of step S35, the γ memory control unit 202 reads the output transition probability γ in time order from the word with the smallest address to the word with the largest address in the first γ memory 203, and data D108 Is supplied to the α calculation unit 108 and is also supplied as the data D115 to the λ calculation unit 112. At the same time, the output transition probability γ (D102) supplied from the γ calculation unit 101 is set in time order in the first γ memory 203. Write from a small word to a large word.

  That is, in the first γ memory 203, read-modify-write is performed in which the stored output transition probability γ is read and the newly obtained output transition probability γ is written simultaneously.

  Further, as the processing of step S35, the γ memory control unit 202 reads the output transition probability γ in the order of going back in time from the word with the largest address of the fourth γ memory 206 in the order of the small word, and as the data D109, the first β calculation unit 109.

The first β calculation unit 109 uses the output transition probability β from the γ memory control unit 202 to calculate the backward probability β in the order of going back in time. The first β calculating unit 109 calculates the backward probability β (backward probability β _5D (m) at time 5D) at time t = 4D when calculating the backward probability β at time t = 4D to 5D−1. Initialize to 1 / M.

  Furthermore, as the processing of step S35, the γ memory control unit 202 reads the output transition probability γ in the order of going back in time from the word with the largest address of the second γ memory 204 to the word with the smallest address, and as the data D110, This is supplied to the 2β calculator 110.

  The second β calculation unit 110 uses the output transition probability γ from the γ memory control unit 202 (and the backward probability β obtained by the second β calculation unit 110 in step S34), and in step S34, performs the second β calculation. The backward probability β at a time further back from the backward probability β obtained by the unit 110 is calculated and supplied to the β memory control unit 211.

  At this time, the β memory control unit 211 reads the backward probability β written in step S34 in the order from the word with the smallest address of the β memory 213 to the largest word, that is, reads the backward probability β in time order, The data D117 is supplied to the λ calculation unit 112, and at the same time, the backward probability β in the order of going back the time from the second β calculation unit 110 is supplied to the β memory 213 as the data D213, and the backward probability β is Are written in ascending order from the word with the smallest address in the β memory 213 to the word with the larger address.

  That is, in the β memory 213, read-modify-write is performed in which the stored backward probability β is simultaneously read and the newly obtained backward probability β is written.

  Here, in the read-modify-write in the β memory 213 in step S35, the newly determined backward probability β in the order of going back in time is written from a word with a smaller address in the β memory 213 toward a larger word. In this case, the β memory 213 stores a backward probability β at a later time at a smaller address (a backward probability β at an earlier time is stored at a larger address).

  In step S35, the α calculation unit 108 further calculates the forward probability α using the output transition probability γ read from the first γ memory 203 by the read-modify-write and supplied as data D108. And supplied to the λ calculation unit 112 as data D116.

  The λ calculation unit 112 outputs the output transition probability γ read from the first γ memory 203 by the γ memory control unit 202 by read-modify-write and supplied as data D115, and the forward probability α supplied from the α calculation unit 108 as data D116. The β memory control unit 211 calculates the output soft value data λ using the backward probability β read out from the β memory 213 by read-modify-write and supplied as data D117, and outputs it as data D118.

  At time t = 5D to 6D−1, as the processing of step S36, the γ memory control unit 202 reads the output transition probability γ in the order of going back in time from the word with the smallest address to the word with the largest address in the second γ memory 204, The data D108 is supplied to the α calculation unit 108, and the data D115 is supplied to the λ calculation unit 112. At the same time, the output transition probability γ (D102) supplied from the γ calculation unit 101 is set to the second γ in order of time. Write from small word to large word in memory 204.

  That is, in the second γ memory 204, read-modify-write is performed in which the stored output transition probability γ is read and the newly obtained output transition probability γ is written simultaneously.

  Further, as the process of step S36, the γ memory control unit 202 reads the output transition probability γ in the order of going back in time from the word with the largest address of the first γ memory 203 in the order of the small word, and as the data D110, the second β calculation unit 110.

The second β calculation unit 110 uses the output transition probability β from the γ memory control unit 202 to calculate the backward probability β in the order of going back in time. In calculating the backward probability β at the time t = 5D to 6D−1, the second β calculating unit 110 calculates the backward probability β (the backward probability β _6D (m) at the time 6D) at the time t = 5D. Initialize to 1 / M.

  Further, as the processing of step S36, the γ memory control unit 202 reads the output transition probability γ in the order of going back in time from the word with the largest address of the third γ memory 205 to the word with the smallest address, and as the data D109, the first β It supplies to the calculation part 109.

  The first β calculation unit 109 uses the output transition probability γ from the γ memory control unit 202 (and the backward probability β obtained by the first β calculation unit 109 in step S35), and in step S35, performs the first β calculation. The backward probability β at a time further back from the backward probability β obtained by the unit 109 is calculated and supplied to the β memory control unit 211.

  At this time, the β memory control unit 211 reads the backward probability β written in step S35 in order from the word with the largest address of the β memory 213 to the smallest word, that is, reads the backward probability in the time order, The data D117 is supplied to the λ calculation unit 112, and at the same time, the backward probability β in the order of going back the time from the first β calculation unit 109 is supplied to the β memory 213 as the data D213, and the backward probability β is Are written in ascending order from the word with the largest address in the β memory 213 to the word with the smallest address.

  That is, in the β memory 213, read-modify-write is performed in which the stored backward probability β is simultaneously read and the newly obtained backward probability β is written.

  Here, in the read-modify-write in the β memory 213 in step S36, the newly obtained backward probability β in the order of going back in time is written from a word with a larger address in the β memory 213 toward a smaller word. In this case, in the β memory 213, the backward probability β of the earlier time is stored at a smaller address (the backward probability β of the later time is stored at a larger address).

  In step S36, the α calculation unit 108 further calculates the forward probability α using the output transition probability γ read from the first γ memory 203 by the read-modify-write and supplied as the data D108. And supplied to the λ calculation unit 112 as data D116.

  The λ calculation unit 112 outputs an output transition probability γ read from the second γ memory 204 by the γ memory control unit 202 by read-modify-write and supplied as data D115, and a forward probability α supplied from the α calculation unit 108 as data D116. The β memory control unit 211 calculates the output soft value data λ using the backward probability β read out from the β memory 213 by read-modify-write and supplied as data D117, and outputs it as data D118.

  Thereafter, similarly, read / write control processing for the first γ memory 203 to the fourth γ memory 206 and read / write control processing for the β memory 213 are performed.

  That is, for a variable j having an integer value of 5 or more, from time t = j × D to (j + 1) × D−1, from a word with a smaller address in the first γ memory 203 to the fourth γ memory 206 toward a larger word. , The output transition probability γ in time order used for the calculation of the forward probability α and the output soft value data λ is read and supplied to the α calculation unit 108 as data D108, and the λ calculation unit as data D115 112, and at the same time, the output transition probability γ (D102) in time order obtained by the γ calculating unit 101 is written from a word with a small address to a word with a large address in the first γ memory 203 to the fourth γ memory 206. Light is done.

  Here, for example, the read-modify-write that simultaneously reads and writes the output transition probability γ is the second γ memory 204 when the variable j = 5, and the third γ memory 205 when the variable j = 6. The variable j = 7 is the fourth γ memory 206, the variable j = 8 is the first γ memory 203, and the variable j = 9 is the second γ memory 204. Thereafter, the same pattern is repeated.

  At the same time, output transition probabilities γ are read in the order of going back in time from the word with the largest address in the first γ memory 203 to the fourth γ memory 206 in order from the smallest word, and the first β calculation unit is used by using the output transition probability γ. 109 or the second β calculation unit 110 calculates a backward probability β1 of the backward probability β that is not stored in the β memory 213. In calculating the backward probability β1, the backward probability β1 is initialized to 1 / M at time t = (i + 5) × D.

  Here, for example, the output transition probability γ used for the calculation of the backward probability β1 is to be read from the first γ memory 203 when the variable j = 5 and from the second γ memory 204 when the variable j = 6. Yes, the variable j = 7 is the third γ memory 205, the variable j = 8 is the fourth γ memory 206, and the variable j = 9 is the first γ memory 203. Thereafter, the same pattern is repeated.

  The backward probability β1 is calculated by, for example, the second β calculator 110 when the variable j = 5, the first β calculator 109 when the variable j = 6, and the second β when the variable j = 7. The calculation unit 110. Thereafter, the same pattern is repeated.

  At time t = j × D to (j + 1) × D−1, the output transition probability γ in the order of going back in time from the word with the largest address in the first γ memory 203 to the fourth γ memory 206 is further increased. Using the output transition probability γ, the backward probability β2 stored in the β memory 213 among the backward probability β is calculated by the first β calculation unit 109 or the second β calculation unit 110.

  Here, for example, the output transition probability γ used for the calculation of the backward probability β2 is the third γ memory 205 when the variable j = 5, and the fourth γ memory 206 when the variable j = 6. Yes, the variable j = 7 is the first γ memory 203, the variable j = 8 is the second γ memory 204, and the variable j = 9 is the third γ memory 205. Thereafter, the same pattern is repeated.

  Further, the backward probability β2 is calculated by the first β calculation unit 109 when the variable j = 5, the second β calculation unit 110 when the variable j = 6, and the first β when the variable j = 7, for example. The calculation unit 109. Thereafter, the same pattern is repeated.

  From time t = j × D to (j + 1) × D−1, the backward probability β2 is read from the β memory 213 and supplied to the λ calculation unit 112 as data D117. Read-modify-write is performed in which the backward probability β2 calculated by the 1β calculation unit 109 or the second β calculation unit 110 is written to the β memory 213.

  Here, in the read-modify-write that simultaneously reads / writes the backward probability β2 with respect to the β memory 213, for example, when the variable j is an odd number, the backward probability β2 in the direction from the larger word to the smaller word in the β memory 213. When the variable j is an even number, the backward probability β2 is read and written in the direction from the smaller word to the larger word in the β memory 213.

  On the other hand, the α calculation unit 108 calculates the forward probability α using the output transition probability γ supplied as the data D108, and supplies it as the data D116 to the λ calculation unit 112.

  Then, the λ calculation unit 112 uses the output transition probability γ supplied as the data D115, the forward probability α supplied as the data D116, and the backward probability β2 supplied as the data D117 to generate the output soft value data λ. Calculated and output as data D118.

  Next, the read / write control for the first γ memory 203 to the fourth γ memory 206 and the read / write control for the β memory 213 will be further described with reference to FIG.

  The first from the left in FIG. 13 shows the progress of calculation of the output transition probability γ, the forward probability α, the backward probability β (β1 and β2), and the output soft value data λ.

  In the first from the left in FIG. 13, the horizontal axis represents the output transition probability γ, the forward probability α, the backward probability β (β1 and β2), and the output soft value data λ, with the direction from left to right as the travel direction of the trellis time. Represents the section of the trellis for which is calculated, and the vertical axis represents the time.

  The first from the left in FIG. 13 is the same as the first from the left in FIG.

  Therefore, in the first from the left in FIG. 13, the line L11 indicates the progress of the calculation of the output transition probability γ, and the line L12 indicates the progress of the calculation of the backward probability β1 of the backward probability β that is not written in the β memory 213. The line L13 indicates the progress of calculation of the backward probability β2 written in the β memory 213, and the line L14 indicates the progress of calculation of the forward probability α and the output soft value data λ.

  The second from the left in FIG. 13 shows the progress of data reading / writing from the first γ memory 203 to the fourth γ memory 206.

  In the second from the left in FIG. 13, the horizontal axis represents the addresses of the first γ memory 203 to the fourth γ memory 206, and the vertical axis represents the time as in the first from the left in FIG. 13.

  In the second line from the left in FIG. 13, the line L51 indicates the progress of writing the output transition probability γ to the first γ memory 203 to the fourth γ memory 206.

  The output transition probability γ is written in the order of time from the word with the smallest address in the first γ memory 203 to the fourth γ memory 206 to the larger word.

  In the second line from the left in FIG. 13, lines L52 and L53 indicate the progress of reading from the first γ memory 203 to the fourth γ memory 206 of the output transition probability γ used for calculating the backward probability β.

  The output transition probability γ used for calculating the backward probability β is read from the first γ memory 203 to the fourth γ memory 206 from the timing (time t = 0) when the calculation of the output transition probability γ is started. It will be started after waiting for 2xD.

  Then, the reading of the output transition probability γ used for the calculation of the backward probability β is performed in the order of going back in time, that is, from the first γ memory 203 to the fourth γ memory 206 from the word having the larger address to the word having the smaller time 2D. The minute output transition probability γ is read.

  In the second line from the left in FIG. 13, the line L52 indicates the output transition probability γ for the first time D in the order of going back in time among the output transition probabilities γ for time 2D (for example, time t = 0 to 2D− If the output transition probability γ is 1, the reading progress status at time t = D or 2D-1) is shown. The output transition probability γ for the first time D is shown from the left in FIG. This is used to calculate the backward probability β1 represented by the first line L12.

  Further, the line L53 indicates an output transition probability γ for the subsequent time D in the backward probability β for the time 2D (for example, the output transition probability γ for the time t = 0 to 2D-1). If there is, the progress status of reading at time t = 0 to D-1) is shown, and the output transition probability γ for time D after this is represented by the first line L13 from the left in FIG. Used to calculate the backward probability β2.

  In the second line from the left in FIG. 13, a line L54 indicates the progress of reading from the first γ memory 203 to the fourth γ memory 206 of the output transition probability γ used for the calculation of the forward probability α and the output soft value data λ. ing.

  The reading of the output transition probability γ used for the calculation of the forward probability α and the output soft value data λ is performed when calculating the forward probability α and the output soft value data λ described in the first line L14 from the left in FIG. In addition, the processing is performed in time order, that is, from the first γ memory 203 to the fourth γ memory 206 to the larger word from the smallest address.

  After time t = 4D, the output transition probability γ is read from the first γ memory 203 through the fourth γ memory 206 represented by the line L54, and the output transition probability γ with respect to the first γ memory 203 through the fourth γ memory 206 represented by the line L51. Read-modify-write, in which writing is performed simultaneously, is performed.

  The third from the left in FIG. 13 (first from the right) shows the progress of reading / writing data to / from the β memory 213.

  In the third from the left in FIG. 13, the horizontal axis represents the address of the β memory 213, and the vertical axis represents the time as in the first from the left in FIG.

  In the third line from the left in FIG. 13, the line L61 indicates the progress of writing the backward probability β (β2) to the β memory 213.

  The writing of the backward probability β (β2) to the β memory 213 starts at time 3D as shown by the line L61.

  The backward probability β (β2) is written in the β memory 213 in the order of going back in time. The backward probability β in the order of going back in time is written from a word having a larger address in the β memory 213 to a smaller word. On the other hand, writing is performed alternately from a word having a small address in the β memory 213 toward a word having a large address in the cycle of the cutoff length D.

  Note that the backward probability β written to the β memory 213 is only the backward probability β2 represented by the first line L13 from the left in FIG.

  In the third line from the left in FIG. 13, a line L62 indicates the progress of reading from the β memory 213 of the backward probability β2 used in the calculation of the output soft value data λ.

  Reading of the backward probability β2 from the β memory 213 is started from the time 4D as shown as the line L62. In this line L62, the backward probability β2 written in the β memory 213 before the cutoff length D is reached. Is read out.

  After time t = 4D, read-modify-write is performed in which reading of the backward probability β2 from the β memory 213 represented by the line L62 and writing of the backward probability β2 represented by the line L61 are performed simultaneously.

  The backward probability β2 is read from the β memory 213 represented by the line L62 in the reverse order of the writing order when the backward probability β2 is written in the β memory 213. Thus, the backward probability β2 is determined in time order. , Λ calculation unit 112.

  Here, the first β calculation unit 109 or the second β calculation unit 110 calculates the backward probability β2 in the order of going back in time, and the backward probability β2 in the order of going back in time is a word having a large address in the β memory 213. The writing from the small word to the small word and, conversely, the writing from the small word in the β memory 213 toward the large word are repeated alternately.

  Therefore, the backward probability β2 is read from the β memory 213 in the reverse order of the writing order when the backward probability β2 is written to the β memory 213, so that the backward probability β2 is Read sequentially.

As described above, encoded data X _t (for example, channel # 22, which is a convolutional encoder), which is obtained by encoding input data i _t (for example, LDPC code to be input to channel # 22, which is a convolutional encoder) SISO decoding encoded data X _t to) the channel (e.g., an input soft value data Y _t obtained by way of the AWGN channel), decodes the output soft value data lambda _t representing the likelihood of the input data i _t In the device 67, the output soft value data λ _t is obtained from the input soft value data Y _t , for example, the output transition probability γ used for the predetermined calculation such as Expression (18), Expression (19), Expression (22) , as storage means for storing the backward probability beta, to register 303 ₁ is configured with a 303 _W, to the data write and read at the same time (read-modify-write) capable first 1γ memory 203, second 4γ Mori 206, and since the adopted β memory 213, it is possible to reduce the circuit scale of the SISO decoder 67.

  That is, the storage means for storing the output transition probability γ by adopting a memory capable of read-modify-write configured using a register as the storage means for storing the output transition probability γ is shown in FIG. As described above, four memories, ie, the first γ memory 203 to the fourth γ memory 206 are sufficient, and when the single-port RAM shown in FIG. 7 is adopted, the five first γRAMs 103 to the fifth γRAM 107 are required. Thus, the number of storage means for storing the output transition probability γ can be reduced by one.

  Similarly, as a storage means for storing the backward probability β, a storage means configured by using a register and capable of read-modify-write is used, so that the storage means for storing the backward probability β is as shown in FIG. Furthermore, only one memory of the β memory 213 is required, and when the single-port RAM shown in FIG. 7 is adopted, the two first βRAMs 113 and the second βRAM 114 are required, whereas the backward probability β Can be reduced by one.

  In addition, when the number of words is small, the single-port RAM has a large proportion of overhead other than memory cells, as described above. The memory that can be used for read-modify-write can be reduced in circuit scale, and from this point of view, by adopting a memory that can be used for read-modify-write, the SISO decoder 67 The circuit scale can be reduced.

  Further, as described with reference to FIG. 12, the output probability γ used for calculating the forward probability α and the output soft value data λ from the first γ memory 203 to the fourth γ memory 206, and the γ calculation Read modification write for simultaneously writing the output transition probability γ obtained by the unit 101 to the first γ memory 203 to the fourth γ memory 206 and reading of the backward probability β2 used by the λ calculation unit 112 from the β memory 213 In addition, the first γ memory 203 through the first γ memory 203 through the first γ memory 203 so that the backward modification β 2 calculated by the first β calculation unit 109 or the second β calculation unit 110 is written to the β memory 213 at the same time. By performing read / write control on the 4γ memory 206 and the β memory 213, as shown in FIG. At the same processing speed as in the case of FIG. 9 which employs the RAM of Gurupoto (performance), it is possible to perform decoding of SISO.

  Note that the SISO decoder 67 in FIG. 10 can perform soft input / soft output decoding in accordance with, for example, the SW-Log-BCJR algorithm in addition to the SW-Max-Log-BCJR algorithm.

In the present embodiment, the input soft value data Y _t obtained from the optical disc is decoded into the output soft value data λ _t , but the present invention is applied to the input soft value data obtained from a recording medium such as an optical disc. another case of decoding the value data Y _t to output soft values data lambda _t, for example, when decoding an input soft value data Y _t that is transmitted via a wireless or wired transmission medium to the output soft value data lambda _t It can also be applied to.

  Furthermore, the embodiment of the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the gist of the present invention.

It is a block diagram which shows the structure of an example of the conventional optical disk apparatus. It is a figure which shows the model of the communication path explaining a BCJR algorithm. 3 is a diagram showing a trellis representing a convolutional encoder 21. FIG. In a case where the number M of states is four, which is a diagram showing an output transition probability _{γ t (m ', m,} i). It is a figure explaining SW-BCJR algorithm. 1 is a block diagram illustrating a configuration example of an optical disc device to which the present invention is applied. It is a block diagram which shows the structural example of the SISO decoder 67 using single port RAM. It is a figure explaining the read / write control process with respect to the 1st gamma RAM103 thru / or the 5th gamma RAM107, and the read / write control process with respect to the 1st beta RAM113 and the 2nd betaRAM114. It is a figure explaining the read / write control process with respect to the 1st gamma RAM103 thru / or the 5th gamma RAM107, and the read / write control process with respect to the 1st beta RAM113 and the 2nd betaRAM114. It is a block diagram which shows the structural example of the SISO decoder 67 using a register | resistor. 3 is a diagram illustrating a configuration example of a first γ memory 203. FIG. FIG. 5 is a diagram for explaining read / write control processing for a first γ memory 203 to a fourth γ memory 206 and read / write control processing for a β memory 213. FIG. 5 is a diagram for explaining read / write control processing for a first γ memory 203 to a fourth γ memory 206 and read / write control processing for a β memory 213.

Explanation of symbols

  21 convolutional encoder, 22 memoryless channel, 23 decoder, 61 Reed-Solomon encoder, 62 LDPC encoder, 63 modulation encoder, 64 NRZI converter, 65 recording / reproducing system, 67 SISO decoder, 68 LDPC decoder , 69 Reed-Solomon decoder, 101 γ calculation unit, 102 γRAM control unit, 103 1st γRAM, 104 2nd γRAM, 105 3rd γRAM, 106 4th γRAM, 107 5th γRAM, 108 α calculation unit, 109 1st β calculation unit, 110 1st 2β calculation unit, 111 βRAM control unit, 112 λ calculation unit, 113 first βRAM, 114 second βRAM, 202 γ memory control unit, 203 first γ memory, 204 second γ memory, 205 third γ memory, 206 fourth γ memory, 211 β Memory control unit, 213 β memory

Claims (4)

Decodes input soft value data, which is soft value data obtained by encoding data encoded through the channel, into output soft value data, which is soft value data representing the likelihood of the input data. In the decoding device
Calculation means for performing a predetermined calculation for obtaining the output soft value data from the input soft value data;
Storage means for storing data used for the predetermined calculation,
The decoding unit is a memory configured by using a register and capable of simultaneously writing and reading data. The decoding device according to claim 1, wherein the calculation means performs calculation according to a SW-Log-BCJR algorithm or a SW-Max-Log-BCJR algorithm. As the storage means,
In the trellis, at time t−1, in the state m ′, when the input data it at the next time _t is a certain value i, certain input soft value data Y _t is observed, and the output transitions to the state m Four γ memories for storing data representing the transition probability γ _t (m ′, m, i);
When the input soft value data series Y _{t + 1} , Y _{t + 2} ,... Is observed after a certain time t + 1, the backward probability β in the state m at the time t in the trellis. The decoding device according to claim 2, further comprising: one β memory that stores data representing _t (m). As the calculation means,
A γ calculator that calculates data representing the output transition probability γ _t (m ′, m, i);
The output transition probability _{γ t (m ', m,} i) using the data representing the said input soft value sequence Y ₁ data, Y _2, · · ·, when Y _t is observed, in the trellis An α calculation unit for calculating data representing a forward probability α _t (m) in the state m at time t;
Using the data representing the output transition probability γ _{t + 1} (m, m ′, i) that is in the state m at time t and transitions to the state m ′ at the next time t + 1, the backward probability β a β calculator for calculating data representing _t (m);
Data representing the output transition probability γ _t (m ′, m, i), the forward probability α _t-1 (m ′) in the state m ′ at time t−1, and the backward probability β _t (m) used, and a lambda calculating section for calculating the output soft values data lambda _t representing the likelihood that the input data i _t is a value i at time t,
Writing of data representing the output transition probability γ _t (m ′, m, i) calculated by the γ calculation unit, calculation of data representing the forward probability α _t (m), and the output soft value data λ A γ memory control unit that controls writing and reading of data to and from the γ memory so that reading of data representing the output transition probability γ _t (m ′, m, i) necessary for calculating _t is performed simultaneously. When,
Writing data representing the backward probability β _t (m) calculated by the β calculating unit and reading data representing the backward probability β _t (m) necessary for calculating the output soft value data λ _t The decoding device according to claim 3, further comprising: a β memory control unit that controls writing and reading of data to and from the β memory so as to be performed simultaneously.

Images