Data transmission device

BACKGROUND OF THE INVENTION

FIELD OF THE INVENTION:

0001The present invention relates to a receiver for soft-decision and decoding for use in digital data transmission/reception applications such as automobile telephones, and particularly, to apparatus which operates using what is known as "a soft-decision device".

DESCRIPTION OF THE RELATED ART

0002In general, a transmitter/receiver for use in digital data transmission has an error correcting function for correcting data errors caused in the channel. One type of code used for correction of such errors is a convolutional code wherein information and code words correspond one by one. The method which is used most frequently for decoding this convolutional code is Viterbi decoding.

0003Viterbi decoding is a decoding method which efficiently executes the maximum-likelihood decoding utilizing a repeat structure of the convolutional code. Although described later in detail, it is possible to correct an error on the way so as to achieve the correct decoding by selecting a path where a Hamming distance is smaller, in a state transition diagram (called a trellis diagram) determined by the state of an encoder and an input code. When forming the Hamming distance, there is a method which determines input data of a decoder in binary digit, that is, 0 or 1, and a method which determines input data of a decoder not in binary digit, but in a many valued fashion. The former is called a hard decision and the latter is called a soft decision. The soft decision reflects more on the reliability of information in transmission than the hard decision.

0004Prior to explanation of a receiver having a conventional soft-decision device, the technical background will be first explained for technical understanding, and thereafter, the conventional technique will be explained.

Technical Background on FIR Filter

0005The term FIR filter is an abbreviation for a finite impulse response filter wherein the impulse response is completed in a finite duration. Fig. 22 shows the structure of an FIR filter. The FIR filter gives delays to an inputted signal in sequence by delay elements (DELAY 1, DELAY 2, ..., DELAY L in the same figure), and adds a plurality of multiplication results by an adder (SUM in the same figure) after executing multiplication by tap coefficients c₀, ..., c_L through multipliers (MULT 0, ..., MULT L). In general, a delay of the delay element is a constant value.

Technical Background on Channel Model Having Intersymbol Interference (ISI

)

0006Fig. 23 shows a channel model having ISI. This model represents the transmission line characteristic using the FIR filter. In this model, a received signal is a Channel Impulse Response (CIR) composite signal of a direct signal in the form of a transmitted signal directly received and delayed signals received with delays due to reflection or the like. In the figure, time differences between the delayed signals are given by a delay circuit DELAY (shift register). The direct signal is obtained by multiplying transmitted signal I_n and tap coefficient c₀,n by the multiplier MULT0. Here, subscript n represents a time. The delayed signals are derived by multiplying delayed transmitted signals I_n-1 ,..., I_n-L and tap coefficients c_1,n ... C_L,n through the multipliers MULT1... L, respectively. Then, outputs of the multipliers MULT0... L are synthesized by the adder SUM. Further, a composite wave outputted from the adder SUM is added with noise w_n so as to be outputted as received signal r_n.

0007In case of absence of the ISI, received signal r_n is expressed as the following equation:<maths id="math0001" num="(1)"><math display="block"><mrow><msub><mrow><mtext>r</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> = c</mtext></mrow><mrow><mtext>0,n</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> + w</mtext></mrow><mrow><mtext>n</mtext></mrow></msub></mrow></math><img file="EP0771079A2_D0001.tif" /></maths>

0008In this case, if c_0,n is known and the noise is small, I_n can be easily estimated from r_n.

0009However, according to the model of Fig. 23, if transmission sequence {I_n} is transmitted over the channel, the transmission sequence is subjected not only to the additive white Gaussian random noise (AWGN) w_n but also to the ISI. Accordingly, received signal r_n includes not only time n but also I_n prior to it. Received signal r_n at this time is expressed by the following equation:<maths id="math0002" num="(2)"><math display="block"><mrow><msub><mrow><mtext>r</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> = Σ c</mtext></mrow><mrow><mtext>i,n</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-i</mtext></mrow></msub><msub><mrow><mtext> + w</mtext></mrow><mrow><mtext>n</mtext></mrow></msub></mrow></math><img file="EP0771079A2_D0002.tif" /></maths>    wherein the sum Σ is derived for i=0, ..., L, where L represents a time length (memory length of a CIR).

0010In the transmission line model of Fig. 23, the transmission sequences are included from time n to time (n-L).

0011On the other hand, when estimating I_n from r_n, the maximum-likelihood sequence estimation is frequently used.

Technical Background on Maximum-Likelihood Sequence Estimation using Viterbi Algorithm

0012The maximum-likelihood sequence estimation using the Viterbi algorithm shown in "Maximum-likelihood sequence estimation of digital sequence in presence of intersymbol interference" (IEEE Trans. Information Theory, vol. IT-18, pp. 363-378, May 1972) written by G. D. Forney, Jr. will be explained in sequence.

0013The Viterbi algorithm is used most frequently as a decoding method of the convolutional code and equalizing the digital signals in the presence of ISI.

0014As a simple example of the convolutional code, it is assumed that a convolutional code of 2 bits is outputted relative to an input of 1 information bit. Since the 2-bit output is derived from the 1-bit input, the coding rate of this convolutional code is 1/2. In this example, if a 2-prior input is determined, an output of the next input by 0 or 1 always has a regularity. Specifically, when one bit is inputted to an encoder, a state is determined by 2-bit data already held so that a code state by the next input bit is determined per state.

0015On the other hand, according to the channel model having ISI in Fig. 23, the received signal is determined by not only the current transmitted signal but also the past transmitted signals. Accordingly, for estimating the current transmitted signal, it is also necessary to consider the past transmitted signals.

0016Candidates of transmission sequences formed by a combination of the past transmitted signals correspond to the states. If a shift register holds one symbol (L=1), there are two kinds of states, that is, the state [0] and the state [1]. On the other hand, if a shift register holds two symbols (L=2), there are four kinds of states, that is, state [0, 0], state [1, 0], state [0, 1] and state [1, 1]. In this manner, the state is expressed by a combination of the transmission sequences.

0017On the other hand, if a memory length of the Viterbi algorithm being the equalizing algorithm is given by V, state S_n at time n and state S_n-1 at time n-1 can be expressed by equations (3) and (4), respectively:<maths id="math0003" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(3)</mtext><mtd><mrow><msub><mrow><mtext>S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> = [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext>, I</mtext></mrow><mrow><mtext>n-V+2</mtext></mrow></msub><msub><mrow><mtext>, ..., I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>]</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(4)</mtext><mtd><mrow><msub><mrow><mtext>S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext> = [I</mtext></mrow><mrow><mtext>n-V</mtext></mrow></msub><msub><mrow><mtext>, I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext>, ..., I</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext>] </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0003.tif" /></maths>

0018Hereinafter, for simplification of expression, state S_n may be simply expressed as I_n-V+1 I_n-V+2 ... In. For example, state [0, 1] is expressed as 01.

0019Here, by utilizing the nature that the transmission sequences as many as V-1 from I_n-V+1 to I_n-1 in two states S_n and S_n-1 take the same values, a trellis diagram of Fig. 24 can be prepared. The trellis diagram is a state transition diagram determined by the state of the channel and the input signal.

0020In Fig. 24, when the number U of candidates of the transmitted signals is 2 and the signal takes a binary digit, 0 or 1, the number M of states becomes 4 (=U^V = 2²). Specifically, in this example, since In is 0, 1 and V=2, the trellis is formed for S_n with four states 00, 10, 01 and 11. These states are assumed to be A, B, C and D.

0021In Fig. 24, the vertical direction represents states A, B, C and D in the order named from above, while the transverse direction represents time n-1, n, n+1, n+2, n+3 and n+4 in the order named from left. Segments are drawn from each state (blank circle) to two states at the next time, respectively. For example, from state A, two segments are drawn to states A and B. A segment from state A to state A represents input data being 0 and thus represents the states before and after the data input being both 00. A line from state A to state B represents the input data being 1 and thus represents a change from state 00 before the data input to state 10.

0022Segment S_n-1/S_n in the trellis is called a branch. This branch is uniquely determined by transmission sequences I_n-V...I_n. This is expressed as equation (5):<maths id="math0004" num="(5)"><math display="block"><mrow><msub><mrow><mtext>S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> = [I</mtext></mrow><mrow><mtext>n-V</mtext></mrow></msub><msub><mrow><mtext>, I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext>, ..., I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>]</mtext></mrow></math><img file="EP0771079A2_D0004.tif" /></maths>

0023For simplification of expression like the expression of the state, branch S_n-1/S_n may be simply expressed as I_n-V I_n-V+1 ... I_n. For example, branch [1, 0, 1] is expressed as 101.

0024On the other hand, if V=L, that is, a time length (memory length of the CIR) for which the ISI influences and a memory length of the Viterbi algorithm are equal to each other, estimate value (replica) h_n of r_n represented by equation (2) can be estimated uniquely as shown by equation (6):<maths id="math0005" num="(6)"><math display="block"><mrow><msub><mrow><mtext>h</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> = Σ g</mtext></mrow><mrow><mtext>i,n</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-i</mtext></mrow></msub></mrow></math><img file="EP0771079A2_D0005.tif" /></maths>    wherein an estimate value of c_i,n is represented by g_i,n and the sum Σ is derived for i=0, ..., L.

0025In the trellis, mutually connected line S_V S_V+1 ... S_n is called a path. This path uniquely determines branches S_V/S_V+1, S_V+1/S_V+2, ..., S_n-1/S_n and transmission sequences from I₀ to I_n.

0026In Fig. 25, path S_n-1 S_n ... S_n+4 corresponding to transmission sequences [I_n-2, I_n-1, ..., I_n+4] = [1, 0, 1, 1, 0, 0, 1] is shown using the trellis of Fig. 24. Thick lines represent the path, thick blank circles represent states the path transits, and each thick line connecting between the states is a branch determined by the path. For convenience of explanation, states B and C in Fig. 25 are reversed relative to Fig. 24. Fig. 26 shows transitions of the states and the branches along the path of Fig. 25.

0027As seen from Fig. 26, data of the transmission sequences shift in order per two. For example, when the first two data 1, 0 of the transmission sequences are inputted, a state at time n-1 becomes 01. This is the state S_n-1.

0028Then, when data 1 is inputted, a state at time n becomes 10. This is the state S_n.

0029Similarly, when data 1001 is inputted, states S_n+1, ..., S_n+4 at time n+1, ..., n+4 are 11, 01, 00, 10.

0030As in the foregoing manner, the path is determined as corresponding to transmission sequences [I_n-2, I_n-1 ..., I_n+4] one by one.

0031To the contrary, if a path is determined, transmission sequences can be specified. For estimating a path from the received signals, estimation is performed at each branch. A branch metric is used for this. The branch metric represents the square error between received signal r_n and replica h_n reproduced by candidate I_n and estimate value g_i,n of the transmission sequence determined by each branch. The square error represents certainty of a transition from a state to the next state, that is, certainty of generation of a branch.

0032Replica h_n[S_n-1/S_n] of r_n can be derived from candidate values I_n-i of transmission sequences from I_n-V to I_n determined by branch S_n-1/S_n as follows:<maths id="math0006" num="(7)"><math display="block"><mrow><msub><mrow><mtext>h</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [ S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = Σ g</mtext></mrow><mrow><mtext>i,n</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-i</mtext></mrow></msub></mrow></math><img file="EP0771079A2_D0006.tif" /></maths>    wherein the sum Σ is derived for i=0 ..., L.

0033The square error between the actual received signal r_n and replica h_n[S_n-1/Sn]determined by branch [S_n-1/S_n], that is, branch metric E_n[S_n-1/S_n], can be expressed as follows:<maths id="math0007" num="(8)"><math display="block"><mrow><msub><mrow><mtext>E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = {ABS (r</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>-h</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [ S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext>])}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow></math><img file="EP0771079A2_D0007.tif" /></maths>    wherein ABS() represents a vector length in the Euclidean space. In equation (8), the square of a distance between point r_n and point h_n[S_n-1/S_n] is represented.

0034Specifically, branch metric E_n[S_n-1/S_n] is uniquely determined by branch [S_n-1/S_n].

0035Further, the sum of the square errors, that is, the sum of all the branch metrics about the branches uniquely determined along the path, is called a path metric. At each state, there are the number U (U=2 in Fig. 24) of paths. Of these paths, a path having the minimum path metric is called a survivor. The survivor exists per state.

0036The process for deriving the survivor is the ACS process. ACS is an abbreviation for Add-Compare-Select.

0037The adding process is an operation for adding survivor metric F_n-1[S_n-1] corresponding to one-time prior state S_n-1 and branch metric E_n[S_n-1/S_n] as shown in equation (9):<maths id="math0008" num="(9)"><math display="block"><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext>]</mtext></mrow></math><img file="EP0771079A2_D0008.tif" /></maths>    wherein F_n[S_n] represents a survivor metric corresponding to state S_n, and F_n[S_n-1/S_n] represents a path metric corresponding to branch S_n-1/S_n.

0038The comparing process is an operation for comparing the number U of path metrics prepared relative to each state. For the overall maximum-likelihood sequence estimation, since the number U of branches exist at each of the number M of states, the number (M·U) of path metrics are prepared.

0039The selecting process is an operation for selecting the minimum path metric at each state from the result of comparing process and selecting sequences corresponding to the selected path metric as a current survivor.

0040The foregoing is an explanation of the maximum-likelihood sequence estimation using the Viterbi algorithm. As described above, in the maximum likelihood sequence estimation shown by G. D. Forney, the received signal of a symbol rate is inputted and the ACS process is performed relative to each state recursively. After all the input signals are inputted, the survivor having the minimum survivor metric among the finally remaining survivors is determined as the maximum-likelihood path, and the only sequence (maximum-likelihood sequence) defined by this maximum-likelihood path is judged as the transmitted sequence.

Technical Background on Soft-Decision

0041The soft decision will be now explained in further detail. The convolutional coding is available as a kind of coding. The optimum decoding method of the convolutional code uses the Viterbi algorithm. A better error rate is obtained by inputting not binary data (called a hard-decision value) such as 0 or 1, but data also including reliability (called a soft-decision value), for example, 0.2, 0.9 or the like, as an input of the Viterbi algorithm. Thus, by using the soft-decision value for decoding the convolutional code, the error rate can be improved. However, since the foregoing maximum-likelihood sequence estimation can not calculate the soft-decision value, the maximum a posteriori probability estimation is frequently used for calculation of the soft-decision value.

0042The maximum a posteriori probability estimation shown in "Optimal decoding of linear codes for minimizing symbol error rate" (IEEE Trans. Information Theory, vol. IT-20, pp. 284-287, March 1974) written by L. R. Bahl and collaborator will now be explained.

0043It is assumed that the transmitted signal takes a binary digit of 0, 1. For calculating a probability of estimate value I_n-V+1 of the transmitted signal being 1 from the received signal suffering the ISI, the probability is calculated in which a path determined by the transmission sequences transits one of states, where I_n-V+1=1, among states S_n in the foregoing trellis. Specifically, probability P [I_n-V+1=1] of 1_n-V+1=1 can be expressed as follows:<maths id="math0009" num="(10)"><math display="block"><mrow><msub><mrow><mtext>P [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext> = 1] = Σ P [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>]</mtext></mrow></math><img file="EP0771079A2_D0009.tif" /></maths>    wherein P[S_n] represents the probability in which a path determined by the transmission sequences passes state S_n, and can be expressed as the product of forward probability P_f [S_n] depending on a probability of the state transition prior to time n in the trellis diagram and backward probability P_b [S_n] depending on the probability of the state transition after time n. Further, the sum Σ is derived for all the states S_n where I_n-V+1=1.

0044First, the forward recursion will be explained. The forward recursion is a process for calculating the foregoing forward probability P_f [S_n]. If probabilities P_E [S⁰_n-1/S_n] and P_E [S¹_n-1/S_n] of the branches connected to S_n and forward probabilities P_f [S⁰_n-1] and P_f [S¹_n-1] of one-time prior states S⁰_n-1 and S¹_n-1 are known, forward probability P_f [S_n] of state S_n can be derived based on equation (11):<maths id="math0010" num="(11)"><math display="block"><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = P</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>] P</mtext></mrow><mrow><mtext>E</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] + P</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>] P</mtext></mrow><mrow><mtext>E</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>]</mtext></mrow></math><img file="EP0771079A2_D0010.tif" /></maths>

0045Accordingly, based on equation (11), the forward probabilities of all the states at the respective time from the initial points, where the probabilities are known, can be calculated recursively. The probability of the branch will be explained later.

0046Similarly, if probabilities P_E[S_n/S⁰_n+1] and P_E[S_n/S¹_n+1] of the branches connected to S_n and backward probabilities P_b[S⁰_n+1] and P_b[S¹_n+1] of one-time after states S⁰_n+1 and S¹_n+1 are known, backward probability P_b[S_n] of state S_n can be derived based on equation (12):<maths id="math0011" num="(12)"><math display="block"><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = P</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext>] P</mtext></mrow><mrow><mtext>E</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext>/S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext>] + P</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext>] P</mtext></mrow><mrow><mtext>E</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext>/S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><mtext>]</mtext></mrow></math><img file="EP0771079A2_D0011.tif" /></maths>

0047Accordingly, based on equation (12), the backward probabilities of all the states at the respective time from the terminal ends, where the probabilities are known, can be calculated recursively.

0048Thus, probability P [I_n-V+1=1] at time n can be calculated using equation (13):<maths id="math0012" num="(13)"><math display="block"><mrow><msub><mrow><mtext>P [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext>=1] = Σ P</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] P</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>]</mtext></mrow></math><img file="EP0771079A2_D0012.tif" /></maths>    wherein the sum Σ is derived for all the states S_n where I_n-V+1=1.

0049If probability P [I_n-V+1=1] is no less than 0.5, I_n-V+1=1 can be estimated, while if less than 0.5, I_n-V+1=0 can be estimated.

0050The foregoing is the explanation of the maximum a posteriori probability estimation. Based on this a posteriori probability, soft-decision value y_n-V+1 can be calculated using equation (14):<maths id="math0013" num="(14)"><math display="block"><mrow><msub><mrow><mtext>y</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext> = log (P [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext> = 1] / P [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><mtext> = 0])</mtext></mrow></math><img file="EP0771079A2_D0013.tif" /></maths>    wherein P [I_n-V+1 = 0] represents a probability of I_n-V+1 = 0 and can be derived through calculation similar to P [I_n-V+1 = 1].

0051If soft-decision value y_n-V+1 is no less than 0, then I_n-V+1=1 is determined, while if less than 0, then I_n-V+1=0 is determined. This is the hard decision. The absolute value of y_n-V+1 at this time represents its reliability.

0052Next, the probability of the branch will be explained. In the foregoing channel model, the generation probability of the branch can be expressed as equation (15):<maths id="math0014" num="(15)"><math display="block"><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>E</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext>] = {1/(2π)</mtext></mrow><mrow><mtext>1/2</mtext></mrow></msup><msub><mrow><mtext>σ}· exp (-E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext>] / 2σ</mtext></mrow><mrow><mtext>2</mtext></mrow></msup><mtext>)</mtext></mrow></math><img file="EP0771079A2_D0014.tif" /></maths>    wherein E_n [S_n-1/S_n] represents the branch metric in equation (8) and σ² represents a noise power.

0053Further, if σ² is constant, by giving<maths id="math0015" num="(16)"><math display="block"><mrow><msup><mrow><mtext>A = {1/(2π)</mtext></mrow><mrow><mtext>1/2</mtext></mrow></msup><msup><mrow><mtext> σ}·exp (1/2σ</mtext></mrow><mrow><mtext>2</mtext></mrow></msup><mtext>)</mtext></mrow></math><img file="EP0771079A2_D0015.tif" /></maths> equation (15) can be rewritten as equation (17):<maths id="math0016" num="(17)"><math display="block"><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>E</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = A ·exp (-E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])</mtext></mrow></math><img file="EP0771079A2_D0016.tif" /></maths>

0054On the other hand, as shown in equation (14), since the soft-decision value depends on the ratio between P[I_n-V+1=1] and P[I_n-V+1=0], when σ² is constant, even if A=1 is given to define as equation (18), the soft-decision value is not affected.<maths id="math0017" num="(18)"><math display="block"><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>E</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = exp (-E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])</mtext></mrow></math><img file="EP0771079A2_D0017.tif" /></maths>

0055Next, the calculation of the soft-decision value based on the branch metric will be explained.

0056Probabilities P[I_n-V+1=1], P_f[S_n] and P_b[S_n] are expressed by metrics as equations (19)...(21):<maths id="math0018" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(19)</mtext><mtd><mrow><msub><mrow><mtext>P [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext>=1] = exp (-p [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><mtext>=1])</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(20)</mtext><mtd><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = exp (-p</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext>[S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(21)</mtext><mtd><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = exp (-p</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>]) </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0018.tif" /></maths>

0057Then, equations (11)...(14) can be rewritten as equations (22)...(25):<maths id="math0019" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(22)</mtext><mtd><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] =-log {exp (-p</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>]-E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>]) + exp (-p</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msup><mrow><mtext>[S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>]-E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [ S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])}</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(23)</mtext><mtd><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] =-log {exp (-p</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext>]-E</mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext>/S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext>]) + exp (-p</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext>]-E</mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext> [ S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext>/S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><mtext>])}</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(24)</mtext><mtd><mrow><msub><mrow><mtext>P [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext> = 1] =Σexp (-p</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext>[S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>]-p</mtext></mrow><mrow><mtext>b</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(25)</mtext><mtd><mrow><msub><mrow><mtext>y</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext> = p [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext> = 0]-p [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><mtext> = 1] </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0019.tif" /></maths>

0058In this manner, if the soft-decision value is calculated by the branch metric, the logarithmic and exponential operations are required to be executed frequently and a large-scale memory therefore becomes necessary so that it is difficult to perform the process in real time. Accordingly, equations (22)...(25) are therefore usually simplified for practical reasons in order to be applied to the soft-decision device.

Explanation of Receiver Having a Conventional Soft-Decision Device

0059Next, an example of a conventional soft-decision device will be explained.

0060Fig. 27 is a block diagram of the conventional soft-decision device used in the typical signal detecting device. This example is the soft-decision device which is described in an Unexamined Patent Publication No. 3-9617 and is used as an example for purposes of explanation.

0061In Fig. 27, numeral 11 denotes a received signal input terminal, 362 a CIR estimating circuit which estimates and outputs CIR g_i,n, 363-1...363-(2M) branch metric producing circuits each deriving branch metric E_n[S_n-1/S_n] based on the received signal and the CIR, 364-1...364-M ACS circuits (if i=1, 2, ..., M, the ACS circuit 364-i connects to the branch metric producing circuits 363-(2i-1) and (2i)) each executing the ACS process based on outputs of two branch metric producing circuits, 365 a shared memory connected to the ACS circuits 364-1...M, 366-1 and 366-2 minimum value selecting circuits each selecting the minimum value among outputs of the ACS circuits 364-1...M, 367 a subtraction circuit subtracting an output of the minimum value selecting circuit 366-2 from an output of the minimum value selecting circuit 366-1, and 17 a soft-decision value output terminal being an output end of this soft-decision device.

0062The branch metric producing circuits 363-1...2M have the same structure. The branch metric producing circuit 363-i is constituted by an FIR filter 3631-i, a memory 3632-i, a subtraction circuit 3633-i and a square circuit 3634-i. The branch metric producing circuit 363 performs the process corresponding to the foregoing equation (6).

0063The ACS circuits 364-1...M have the same structure. Fig. 28 is a block diagram showing the inside of the ACS circuit 364. The ACS circuit 364-i is constituted by branch metric input terminals 3641a-i and 3641b-i, one-time prior survivor metric input terminals 3642a-i and 3642b-i, adders 3643a-i and 3643b-i, a comparator/selector 3644-i, path metric output terminals 3645a-i and 3645b-i, and a current-time surviving path metric output terminal 3646-i.

0064A typical operation of the conventional soft-decision device will be explained using Fig. 27. The transmitted signal takes a binary digit of 0, 1, M(=2^V) represents the number of states and V=L.

0065First, the operation of the branch metric producing circuit 363-1 will be explained. The FIR filter 3631 receives candidates of transmission sequences corresponding to branches outputted from the memory 3632 and CIR (tap coefficients) estimated by the CIR estimating circuit 362 and outputs a replica (estimated value) of the received signal based on equation (6). The subtraction circuit 3633 receives the received signal and the replica outputted from the FIR filter 3631 and outputs a difference between the received signal and the replica. The square circuit 3634 squares the output of the subtraction circuit 3633 and outputs this value as a branch metric. The foregoing operation corresponds to equation (8). The branch metric producing circuits 363-2...363-(2M) operate similarly and output branch metrics corresponding to branches.

0066The ACS circuits 364-1...364-M as many as M share the one-time prior and current-time survivor metrics through the shared memory 365, and these path metrics are accessible mutually. Each of the ACS circuits 364-1...364-M receives branch metrics outputted from two, among the branch metric producing circuits 363-1...363-(2M), corresponding to two branches connected to the state which corresponds to the corresponding one of the ACS circuits 364-1...364-M and one-time prior survivor metrics outputted from the shared memory 365 corresponding to one-time prior states connected via those two branches, and performs the ACS process based on equation (9), and further performs addition of the ACS process so as to output path metrics corresponding to the respective branches and a current-time surviving path metric.

0067Now, the operation of the ACS circuit 364-1 will be explained in further detail using Fig. 28. The adders 3643a and 3643b add the branch metrics inputted to the branch metric input terminals 3641a and 3641b and the one-time prior survivor metrics inputted to the one-time prior survivor metric input terminals 3642a and 3642b, respectively, and output the results via the path metric output terminals 3645a and 3645b, respectively.

0068The comparator/selector 3644 inputs the path metrics outputted from the adders 3643a and 3643b and outputs one of them having a smaller value from the current-time survivor metric output terminal 3646 as a current-time survivor metric.

0069The shared memory 365 receives the outputs of the ACS circuits 364-1...364-M and sets the current-time survivor metrics as one-time prior survivor metrics. In this manner, the shared memory 365 is updated.

0070The minimum value selecting circuit 366-1 receives the path metrics as many as M corresponding to branches where the most prior transmitted signals of transmission sequences determined by the branches are 0, and outputs the minimum value thereof. The minimum value selecting circuit 366-2 receives path metrics as many as M corresponding to branches where the most prior transmitted signals of transmission sequences determined by the branches are 1, and outputs the minimum value thereof.

0071The subtraction circuit 367 outputs the result of subtraction of the minimum value outputted from the minimum value selecting circuit 366-2 from the minimum value outputted from the minimum value selecting circuit 366-1 as a soft-decision value.

0072The foregoing operations will be explained concretely using a trellis shown in Fig. 29. Here, <maths id="math0020" num=""><math display="inline"><mrow><mtext>L=V=2</mtext></mrow></math><img file="EP0771079A2_D0020.tif" /></maths> and <maths id="math0021" num=""><math display="inline"><mrow><msup><mrow><mtext>M=2</mtext></mrow><mrow><mtext>V</mtext></mrow></msup><mtext>=4</mtext></mrow></math><img file="EP0771079A2_D0021.tif" /></maths> are given to calculate a soft-decision value of transmitted signal I_n.

0073The FIR filter 3631-1 inputs candidates [I_n, I_n+1, I_n+2]=[0, 0, 0] of the transmission sequences corresponding to the branches outputted from the memory 3632-1 and CIR g₀, g₁, g₂ estimated at the CIR estimating circuit 362, and calculates replica h_n+2 [000] of the received signal.<maths id="math0022" num="(26)"><math display="block"><mrow><msub><mrow><mtext>h</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msub><mrow><mtext> [000] = g</mtext></mrow><mrow><mtext>0</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> + g</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><msub><mrow><mtext> + g</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub></mrow></math><img file="EP0771079A2_D0022.tif" /></maths>

0074The subtraction circuit 3633-1 receives received signal r_n+2 and replica h_n+2 [000] outputted from the FIR filter 351 and calculates a difference r_n+2 - h_n+2 [000] between the received signal and the replica. The square circuit 3634-1 squares r_{n+2 -} h_n+2 [000] and outputs this value as branch metric E_n+2 [000].<maths id="math0023" num="(27)"><math display="block"><mrow><msub><mrow><mtext>E</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msub><mrow><mtext> [000] = {ABS (r</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msub><mrow><mtext> -h</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msup><mrow><mtext> [000])}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow></math><img file="EP0771079A2_D0023.tif" /></maths>

0075The foregoing is the operation of the branch metric producing circuit 363-1. The other branch metric producing circuits 363-2...363-8 operate similarly and output branch metrics E_n+2 [100], E_n+2 [010], ..., E_n+2 [111] corresponding to branches 100, 010, ..., 111.

0076The ACS circuit 364-1 inputs branch metrics E_n+2 [000] and E_n+2 [100] outputted from the branch metric producing circuits 363-1 and 363-2 corresponding to branches 000 and 100 connected to state 00 which corresponds to the ACS circuit 364-1, and one-time prior surviving path metrics F_n+1 [000] and F_n+1 [100], and derives path metrics corresponding to the next branches 000 and 100 through the adding process:<maths id="math0024" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(28)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msub><mrow><mtext> [000] = E</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msub><mrow><mtext> [000] + F</mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><mtext> [00]</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(29)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msub><mrow><mtext> [100] = E</mtext></mrow><mrow><mtext>n+2</mtext></mrow></msub><msub><mrow><mtext> [100] + F</mtext></mrow><mrow><mtext>n+1</mtext></mrow></msub><mtext> [10] </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0024.tif" /></maths>

0077Path metrics F_n+2 [000] and F_n+2 [100] are outputted to the minimum value selecting circuits 366-1 and 366-2. Further, through the comparator/selector 3644 in the ACS circuit, one of these two path metrics being smaller is outputted to the shared memory 365 as current-time survivor metric F_n+2 [00] corresponding to state 00. The ACS circuits 364-2...364-4 operate similarly to output current-time survivor metrics F_n+2 [10], F_n+2 [01] and F_n+2 [11] corresponding to states 10, 01 and 11 to the shared memory 365.

0078The minimum value selecting circuit 366-1 receives path metrics F_n+2 [000], F_n+2 [010], F_n+2 [001] and F_n+2 [011], where I_n=0, among the path metrics outputted from the ACS circuits 364-1...364-4, and outputs the minimum path metric among them. The minimum value selecting circuit 366-2 receives path metrics F_n+2 [100], F_n+2 [110], F_n+2 [101] and F_n+2 [111], where I_n=1, among the path metrics outputted from the ACS circuits 364-1...364-4, and outputs the minimum path metric among them. The subtraction circuit 367 subtracts the minimum value outputted from the minimum value selecting circuit 366-2 from the minimum value outputted from the minimum value selecting circuit 366-1 and outputs the result thereof at 17 as the soft-decision value.

0079This conventional example simplifies the calculation of the soft-decision value (equations (22)...(25)) based on the foregoing maximum a posteriori probability estimation as follows:<maths id="math0025" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(30)</mtext><mtd><mrow><msub><mrow><mtext>P</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = min {(p</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>] + E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>]), (p</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>] + E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])}</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(31)</mtext><mtd><mrow><msub><mrow><mtext>P [I</mtext></mrow><mrow><mtext>n-V</mtext></mrow></msub><msub><mrow><mtext> = 1] = min {p</mtext></mrow><mrow><mtext>f</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>]} </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0025.tif" /></maths> wherein S_n represents a value when I_n-V = 1.<maths id="math0026" num="(32)"><math display="block"><mrow><msub><mrow><mtext>y</mtext></mrow><mrow><mtext>n-V</mtext></mrow></msub><msub><mrow><mtext> = p [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><msub><mrow><mtext> = 0] -p [I</mtext></mrow><mrow><mtext>n-V+1</mtext></mrow></msub><mtext> = 1]</mtext></mrow></math><img file="EP0771079A2_D0026.tif" /></maths>

0080Here, p_f [S_n] and (p_f [S_n-1] + E_n [S_n-1/S_n]) correspond to F_n [S_n] and F_n [S_n-1/S_n], respectively. Although this conventional example describes to derive soft-decision value y_n, equations (30)...(32) reference branch metric E_n [S_n-1/S_n] and derive soft-decision value y_n-V as calculated from E_n [S_n-1/S_n].

0081Further, as shown in the trellis diagram of Fig. 29, the process of this conventional example selects the minimum paths, where I_n=0 and I_n=1, among all the paths relating to all the states S_n+2 at time (n+2), and determines a difference between path metrics F_n+2 [S_n+2] of these two paths to be the soft-decision value. In this manner, by simplification, this conventional example can calculate the soft-decision value with the process of a degree similar to the maximum-likelihood sequence estimation.

0082However, in the foregoing conventional example, the logarithmic /exponential operation is replaced by the comparing/selecting process, and further, the memory length V of the trellis and the memory length L of the estimated CIR are given to be equal so that the probability to be calculated through all the sequences from the initial point to the terminal end is calculated from the partial sequences from the initial point to time (n+L). Thus, an error is caused in the soft-decision value. Due to this error, when the Viterbi decoding is performed using the soft-decision value outputted from the soft-decision device of this conventional example, the error rate of the decoded data is impaired. As used in the present application, a soft-decision value which can improve the error rate of the decoded data will be referred to as a soft-decision value with high accuracy.

0083Further, the tap coefficient in the calculation of the branch metric includes an estimation error caused by noise or the like so that this error grows with the number of taps. Accordingly, although the memory length of the estimated CIR is fixed in this conventional example, if a memory length of the CIR is reduced in the channel where the memory length of the CIR changes, the soft-decision value with high accuracy can not be achieved due to unnecessary taps.

0084Further, if the CIR varies with a lapse of time, since the estimated CIR can not follow the variation in this conventional example, a soft-decision value with high accuracy can not be achieved in the time-variable channel.

0085Further, since this conventional example calculates the soft-decision value based on the path metrics being intermediate results of the ACS process, there is a problem that the circuit scale is increased in the DSP which performs at high-speed the calculation of the branch metrics and the ACS process en bloc.

0086As described above, there have existed problems in the conventional soft-decision device in that a soft-decision value with high accuracy can not be achieved since the soft-decision value is calculated from the partial sequences of the received sequences, and further, the accuracy of the soft-decision value is further impaired in the channel where the memory length of the estimated CIR and the tap coefficient change with an lapse of time, and the circuit scale is increased in the DSP which performs the high speed calculation of the branch metrics and the ACS process en bloc using the special purpose hardware.

SUMMARY OF THE INVENTION

0087The present invention is directed to solving the foregoing problems and thus has an object to provide a soft-decision device with improved functions, wherein a soft-decision with high accuracy can be achieved. Further, a soft-decision with high accuracy can be achieved even in the channel where the memory length of the estimated CIR and the tap coefficient change with an lapse of time, and the circuit scale can be reduced in the DSP which performs at high speed the calculation of the branch metrics and the ACS process en bloc.

0088The present invention, in one embodiment thereof, utilizing a receiver, in the receiver for receiving signals including encoded data via a channel, causing conditions of the received signals to correspond to states, causing transitions from certain states to next states, which are attendant upon reception of said received signal, to correspond to branch metrics as indices of said transitions, deriving path metrics based on said branch metrics as indices of paths which are routes of said transitions when receiving said signal including encoded data in sequence, performing a soft decision of a received signal by selecting a portion of said paths as survivors based on said path metrics, and performing decoding using this result, comprises: <ul id="ul0001" list-style="none" compact="compact"><li>a soft-decision device for deriving survivor metrics per state based on said branch metrics, classifying said survivor metrics based on most prior transmitted signals in states corresponding to said survivors, and deriving a soft-decision value based on said classified survivor metrics; and</li><li>a decoding circuit for performing decoding based on said soft-decision value.</li></ul>

BRIEF DESCRIPTION OF THE DRAWINGS

0089<ul id="ul0002" list-style="none" compact="compact"><li>Fig. 1 is a block diagram of a receiver according to a first embodiment of the present invention;</li><li>Fig. 2 is a block diagram of a soft-decision device according to the first embodiment of the present invention;</li><li>Fig. 3 is a block diagram of a branch metric producing circuit of a soft-decision device according to the first embodiment of the present invention;</li><li>Fig. 4 is a block diagram of an ACS circuit of a soft-decision device according to the first embodiment of the present invention;</li><li>Fig. 5 is a block diagram of a soft-decision value producing circuit of a soft-decision device according to the first embodiment of the present invention;</li><li>Fig. 6 is a trellis diagram for explaining a soft-decision method according to the first embodiment of the present invention, wherein-V=4;</li><li>Fig. 7 is a trellis diagram for explaining a soft-decision method according to the first embodiment of the present invention, wherein-V=2;</li><li>Fig. 8 is a trellis diagram of a conventional soft-decision method for comparing a soft-decision method according to the first embodiment of the present invention, wherein-<maths id="math0027" num=""><math display="inline"><mrow><mtext>V=L=2</mtext></mrow></math><img file="EP0771079A2_D0027.tif" /></maths>;</li><li>Fig. 9 is a trellis diagram for explaining a soft-decision method according to the first embodiment of the present invention, wherein-V=2;</li><li>Fig. 10 is a graph showing BER vs. C/N for a receiver according to the first embodiment of the present invention. This graph is obtained through a computer simulation. This graph also includes plots for a conventional receiver for comparison;</li><li>Fig. 11 is a block diagram of a soft-decision device according to a second embodiment of the present invention;</li><li>Fig. 12 is a block diagram of a soft-decision device according to a third embodiment of the present invention;</li><li>Fig. 13 is a block diagram of an ACS circuit of a soft-decision device according to the third embodiment of the present invention;</li><li>Fig. 14 is a block diagram of a soft-decision value producing circuit of a soft-decision device according to the third embodiment of the present invention;</li><li>Fig. 15 is a block diagram of a soft-decision device according to a fourth embodiment of the present invention;</li><li>Fig. 16 is a block diagram of a memory length estimating circuit of a soft-decision device according to the fourth embodiment of the present invention;</li><li>Fig. 17 is a block diagram of a branch metric producing circuit of a soft-decision device according to the fourth embodiment of the present invention;</li><li>Fig. 18 is an explanatory diagram of a delay operation of a soft-decision value in a memory length estimating circuit of a soft-decision device according to the fourth embodiment of the present invention;</li><li>Fig. 19 is a block diagram of a soft-decision device according to a fifth embodiment of the present invention;</li><li>Fig. 20 is a block diagram of a memory length estimating circuit of a soft-decision device according to the fifth embodiment of the present invention;</li><li>Fig. 21 is a block diagram of a soft-decision value producing circuit of a soft-decision device according to the fifth embodiment of the present invention;</li><li>Fig. 22 is a block diagram of an FIR filter;</li><li>Fig. 23 is a diagram of a channel model;</li><li>Fig. 24 is a trellis diagram when-V=2;</li><li>Fig. 25 is a trellis diagram for explaining a conventional maximum-likelihood sequence estimation, wherein-V=2;</li><li>Fig. 26 is an explanatory diagram of a state transition corresponding to the trellis diagram of Fig. 25;</li><li>Fig. 27 is a block diagram of a conventional soft-decision device;</li><li>Fig. 28 is a block diagram of an ACS circuit of the conventional soft-decision device; and</li><li>Fig. 29 is a trellis diagram and a state transition diagram for explaining the principle of the conventional soft-decision device, wherein-V=2.</li></ul>

DESCRIPTION OF THE PREFERRED EMBODIMENTS:

First Embodiment

0090A embodiment 1 of the present invention will be explained. In the embodiment of the present invention, a transmitted signal takes a binary digit of 0 or 1. Further, in the figures, components assigned the same signs represent the same or corresponding ones.

0091The first embodiment of the present invention will now be explained.

0092Fig. 1 is a block diagram showing an example of a receiver according to the first embodiment. In the figure, 201 denotes a received signal input terminal, 202 an amplifier for amplifying a received signal, 203 an A/D converter for converting the amplified received signal from analog value to digital value, 204 a soft-decision device according to the present invention, 205 a division circuit for performing a compensation by dividing the derived soft-decision value by an output of the amplifying circuit 202, 206 a deinterleaver for rearranging the compensated soft-decision values by rule, 207 a Viterbi decoder for decoding the rearranged soft-decision values, and 208 a decoded data output terminal.

0093Next, the operation of the receiver of Fig. 1 including the soft-decision device will be explained.

0094The amplifier 202 amplifies the received signal with a gain β_n so as to control an input level of the A/D converter 203 within an appropriate range.

0095The A/D converter 203 converts the amplified received signal from an analog signal to a digital signal.

0096The soft-decision device 204 calculates the soft-decision value from the received signal converted to the digital signal. The soft-decision device 204 forms an element with features which characterize the present invention. Details of this will be described later.

0097The division circuit 205 divides the soft-decision value outputted from the soft-decision device 204 by the gain β_n outputted from the amplifier 202 so as to compensate the soft-decision value.

0098The deinterleaver 206 rearranges the corrected soft-decision values outputted from the division circuit 205 according to the rules.

0099The Viterbi decoder 207 inputs the rearranged soft-decision values and outputs the decoded data.

0100In the channel where a signal power of the received signals changes while a noise power is constant, a receiver amplifies the received signals with a gain β_n for holding the received signals at a constant level. Then, since the noise power changes, it is necessary to compensate the soft-decision value in consideration of this change.

0101Here, the noise power in an input level of the amplifying circuit 202 is given by σ². The noise power after amplification becomes (β_n·σ²). Accordingly, by normalizing the soft-decision value with (β_n·σ²), that is, by dividing the soft-decision value by the noise power (β_n·σ²) after amplification, the variation of the noise power in the amplifier 202 can be absorbed. Alternatively, since the soft-decision value has a meaning only as a relative value, it is sufficient to give σ²=1 and divide the soft-decision value by the gain β_n.

0102According to the receiver of Fig. 1, by dividing the output of the soft-decision device 202 by the gain β_n of the amplifier 202, a soft-decision with high accuracy can be achieved even in a channel where the signal power of the received signals changes while the noise power is constant.

0103Fig. 2 is a block diagram showing the internal structure of the soft-decision device of Fig. 1.

0104In Fig. 2, 11 denotes a received signal input terminal, 12 a CIR estimating circuit which estimates and outputs CIR g_i,n of the channel, 13-1...13-(2N) branch metric producing circuits provided corresponding to branches as many as 2N (here, <maths id="math0028" num=""><math display="inline"><mrow><msup><mrow><mtext>N=2</mtext></mrow><mrow><mtext>V</mtext></mrow></msup></mrow></math><img file="EP0771079A2_D0028.tif" /></maths>, V≥L), respectively, in a trellis diagram where a memory length is given by V, and deriving branch metrics E_n[S_n-1/S_n], respectively, based on the received signal and the CIR, 14-1...14-N ACS circuits (if i=1, 2, ..., N, an ACS circuit 14-i connects to branch metric producing circuits 13-(2i-1) and (2i)) corresponding to states as many as N (=2^V), respectively, in the trellis where the memory length is given by V, and each performing the ACS process using outputs of two branch metric producing circuits, 15 a shared memory connected to the ACS circuits 14-1...14-N, 16 a soft-decision value producing circuit for producing a soft-decision value based on survivor metrics F_n [S_n] outputted from the ACS circuits 14-1...14-N, and 17 a soft-decision value output terminal.

0105Here, N represents the number of states, L represents a period (CIR memory length) for which the ISI influences, and V represents a memory length of the Viterbi algorithm.

0106The branch metric producing circuits 13-1...13-(2N) have the same structure. Fig. 3 is a detailed block diagram of the branch metric producing circuit 13-1. In the figure, 21 denotes an estimated CIR input terminal, 22-1...22-(L+1) multipliers for multiplying the estimated CIR g_i,n and the transmission sequences I_n, respectively, 23 a memory for storing the transmission sequences I_n, 24 an adder for deriving the sum of outputs of the multipliers 22-1...22-(L+1), 25 a subtraction circuit for subtracting an output of the adder 24 from received signal r_n, 26 a square circuit for deriving the square of an output of the subtraction circuit 25, and 27 a branch metric output terminal.

0107The ACS circuits 14-1...14-N have the same structure. Fig. 4 is a detailed block diagram of the ACS circuit 14-1. In the figure, 31-1 and 31-2 denote branch metric input terminals, 32-1 and 32-2 input terminals of one-time prior survivor metrics, 33-1 and 33-2 adders adding branch metrics and one-time prior survivor metrics to derive path metrics, respectively, 34 a comparator/selector circuit which selects smaller one of the path metrics and outputs it as a survivor metric, and 35 a survivor output terminal.

0108Fig. 5 is a detailed block diagram of the soft-decision value producing circuit 16. In the figure, 41-1....41-(N/2) and 42-1...42-(N/2) denote survivor metric input terminals, 43-1 and 43-2 minimum value selecting circuits each selecting the minimum one from the inputted survivor metrics, and 44 a subtraction circuit.

0109Here, the survivor metrics inputted to the survivor metric input terminals 41-1...41-(N/2) correspond to those states, where the most prior symbols are "1". On the other hand, the survivor metrics inputted to the survivor metric input terminals 42-1...42-(N/2) correspond to those states where the most prior symbols are "0". Based on these relationships, the ACS circuits 14-1...14-N and the soft-decision value producing circuit 16 are connected.

0110Next, operation of the soft-decision device will be explained.

0111First, the outline of the operation of the entire device will be described and then the details of the operation will be further described.

0112The received signal is formed by a training signal for the purpose of estimating the CIR and an information signal for the purpose of transmitting the information. The training signal is assumed to be known at the side of the soft-decision device. Hereinafter, the information signal of the received signal will be referred to as the received signal, and the training signal of the received signal will be referred to as the training signal.

0113The CIR estimating circuit 12 estimates the CIR based on the training signal and the known information of the training signal and outputs g_i,n.

0114The branch metric producing circuits 13-1...13-(2N) as many as 2N receive received signal r_n and estimated CIR g_i,n outputted from the CIR estimating circuit and output branch metrics E_n [S_n-1/S_n] based on transmission sequences I_n corresponding to the branches stored in the memory 23, respectively.

0115The ACS circuit 14-1 receives the branch metrics outputted from two branch metric producing circuits corresponding to two branches connected to the state which corresponds to the ACS circuit 14-1, and one-time prior survivor metrics F_n-1 [S_n-1] outputted from the shared memory 15 corresponding to one-time prior states connected by those two branches so as to perform the ACS process and outputs a current-time survivor metric F_n [S_n]. The other ACS circuits 14-2...14-N operate similarly.

0116The shared memory 15 receives the current-time survivor metrics outputted from the ACS circuits 14-1...14-N to update the one-time prior survivor metrics.

0117The soft-decision value producing circuit 16 receives current-time survivor metrics F_n [S_n] outputted from the ACS circuits 14-1...14-N to calculate the soft-decision value and outputs the soft-decision value from the soft-decision value output terminal 17.

0118Operation of the branch metric producing circuit 13-1 will be explained with reference to Fig. 3.

0119The multipliers 22-1...22-(L+1) calculate the products of the estimated CIR g_i,n (tap coefficients of the FIR filter having the taps as many as (L+1)) inputted from the estimated CIR input terminal 21 and a portion of the candidates of the transmission sequences determined by the branches corresponding to the branch metric producing circuit 13-1 stored in the memory 23, that is, the newest transmitted signals of (L+1) symbols in the transmitted signals of (V+1) symbols determined by the branches. The adder 24 inputs the products as many as (L+1) outputted from the multipliers 22-1...22-(L+1) and calculates the sum thereof. The subtraction circuit 25 calculates a difference between the sum outputted from the adder 24 and the received signal. The square circuit 26 calculates the square of the difference outputted from the subtraction circuit 25 and outputs the result via the branch metric output terminal 27 as the branch metric.

0120Operation of the ACS circuit 14-1 will be explained using Fig. 4.

0121The branch metric input terminals 31-1 and 31-2 input the branch metrics outputted from the branch metric producing circuits 13-1 and 13-2 corresponding to two branches connected to the state which corresponds to the ACS circuit 14-1. The survivor metric input terminals 32-1 and 32-2 receive from the shared memory 15 the survivor metrics corresponding to one-time prior states connected by the two branches connected to the state which corresponds to the ACS circuit 14-1. The adders 33-1 and 33-2 calculate the sums of the branch metrics inputted via the branch metric input terminals 31-1 and 31-2 and the survivor metrics inputted via the survivor metric input terminals 32-1 and 32-2, respectively. The comparator/selector circuit 34 inputs the sums outputted from the adders 33-1 and 33-2 and compares the two sums to output the smaller one via the surviving path metric output terminal 35 as a current-time survivor metric.

0122Operation of the soft-decision value producing circuit 16 will now be explained using Fig. 5.

0123The survivor metric input terminals 41-1...41-(N/2) receive the survivor metrics outputted from the ACS circuits 14 corresponding to the states where the most prior transmitted signals forming the states are 0. On the other hand, the survivor metric input terminals 42-1...42-(N/2) receive the survivors outputted from the ACS circuits 14 corresponding to the states where the most prior transmitted signals forming the states are 1. The minimum value selecting circuit 43-1 receives the survivor metrics from the surviving path metric input terminals 41-1...41-(N/2) and selects the minimum survivor metric among them to output it. The minimum value selecting circuit 43-2 receives the survivor metrics from the survivor metric input terminals 42-1...42-(N/2) and selects the minimum survivor metric among them to output it. The subtraction circuit 44 calculates a difference between the output of the minimum value selecting circuit 43-1 and the output of the minimum value selecting circuit 43-2 and outputs the result thereof from the soft-decision value output terminal 15 as a soft-decision value.

0124Next, operation of the first embodiment will be explained using Fig. 6.

0125It is given that L=2, V=4 and N=2^V=16.

0126The branch metric producing circuits 13-1...13-32 are provided corresponding to branches S_n-1/S_n, respectively. The branch metric producing circuit 13 calculates the branch metric from equation (50) based on received signal r_n, estimated CIR g₀, g₁ and g₂, and candidates I_n-2, I_n-1 and I_n of the transmitted signals held by each branch metric producing circuit 13.<maths id="math0029" num="(50)"><math display="block"><mrow><msub><mrow><mtext>E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [ S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = [ABS {r</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> - (g</mtext></mrow><mrow><mtext>0</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> + g</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext> + g</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-2</mtext></mrow></msub><msup><mrow><mtext>)}]</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow></math><img file="EP0771079A2_D0029.tif" /></maths>    wherein I_n-2, I_n-1 and I_n represent three symbols among candidates I_n-4, I_n-3, I_n-2, I_n-1 and I_n of the transmitted signals determined by branch S_n-1/S_n. These do not depend on time n, but are constantly fixed values in each branch metric producing circuit. Further, ABS represents an absolute value sign.

0127If estimated CIR g₀, g₁ and g₂ do not change, a replica derived from estimated CIRs g₀, g₁ and g₂ and candidates I₂ [S_n-1/S_n], I₃ [S_n-1/S_n] and I₄ [S_n-1/S_n] of the transmitted signal, that is, () of equation (50), is always constant. In this case, the branch metric producing circuit 13 may be arranged that a value of replica is stored in the memory and updated only when the estimated CIRs are updated.

0128Next, the ACS producing circuits 14-1...14-16 are provided corresponding to states S_n, respectively. Concrete relationships are based on, for example, Fig. 6. The ACS producing circuit 14 receives the outputs of the branch metric producing circuits 13 corresponding to branches S⁰_n-1/S_n and S¹_n-1/S_n connected to state S_n, that is, branch metrics E_n [S⁰_n-1/S_n] and E_n [S¹_n-1/S_n]. Further, the ACS producing circuit 14 receives survivor metrics F_n-1 [S⁰_n-1] and F_n-1 [S¹_n-1] corresponding to one-time prior states S⁰_n-1 and S¹_n-1 outputted from the shared memory 15. The ACS producing circuit 14 performs the adding process as expressed by equations (51) and (52) using these inputs so as to calculate the current-time path metrics:<maths id="math0030" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(51)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext>]</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(52)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext>] </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0030.tif" /></maths>

0129Then, the ACS circuit 14 compares path metrics F_n [S⁰_n-1/S_n] and F_n [S¹_n-1/S_n] derived from the foregoing equations (51) and (52) and outputs smaller one as a current-time survivor metric F_n [S_n] corresponding to state S_n.

0130For example, the ACS circuit corresponding to state 1011 receives branch metrics E_n[01011] and E_n[11011] corresponding to branches 01011 and 11011, and survivor metrics F_n-1 [0101] and F_n-1 [1101] corresponding to one-time prior states 0101 and 1101, and calculates path metrics corresponding to branches 01011 and 11011 from equations (53) and (54):<maths id="math0031" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(53)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [01011] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [01011] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext> [0101]</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(54)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [11011] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [11011] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext> [1101] </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0031.tif" /></maths>

0131Further, the ACS circuit compares path metrics F_n [01011] and F_n [11011] and outputs smaller one as survivor metric F_n [1011] corresponding to state 1011.

0132The soft decision value producing circuit 16 receives the survivor metrics outputted from the ACS circuits 14-1...14-16 and calculates soft-decision value y_n-3 from equation (55):<maths id="math0032" num="(55)"><math display="block"><mrow><msub><mrow><mtext>y</mtext></mrow><mrow><mtext>n-3</mtext></mrow></msub><msub><mrow><mtext> = min (F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>]) - min (F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])</mtext></mrow></math><img file="EP0771079A2_D0032.tif" /></maths>    wherein the first term calculates the minimum survivor metric among the survivor metrics corresponding to the states where the most prior transmitted signals I_n-3 among the transmission sequences I_n-3, I_n-2, I_n-1 and I_n determined by states S_n become 0. The number of the transmission sequences I_n-3...I_n corresponds to N/2.

0133Further, the second term calculates the minimum survivor metric among the survivor metrics corresponding to the states where I_n-3=1.

0134Fig. 6 is a trellis diagram showing an example, wherein soft-decision value y_n of transmitted signal I_n is calculated according to this first embodiment (L=2, V=4, N=2^V=16). Path A represents a path having the minimum survivor metric among the survivor metrics corresponding to states S_n+3 where I_n=0. Path B represents a path having the minimum survivor metric among the survivor metrics corresponding to states S_n+3 where I_n=1. The soft-decision value is calculated as a path metric difference between path A and path B.

0135For facilitating understanding, Fig. 7 shows paths forming transmission sequences equal to the transmission sequences determined by path A and path B of Fig. 6, on a trellis diagram where V=2. Fig. 8, for comparison with Fig. 7, shows an example wherein soft-decision value y_n corresponding to transmitted signal I_n is calculated based on the conventional example (L=V=2, M=2^V=4), on a trellis diagram.

0136In the conventional example, soft-decision value y_n is calculated as a difference between path C (this turns out to be a survivor corresponding to state S_n+2=00) corresponding to branch S_n+1/S_n+2=000 and path D (this turns out to be a survivor corresponding to state S_n+2=10) corresponding to branch S_n+1/S_n+2=110. On the other hand, in this first embodiment, the soft-decision value is calculated further using branches S_n+2/S_n+3 so that a soft-decision value with higher accuracy can be calculated as compared with the conventional example.

0137In the conventional example of Fig. 8, the accuracy of the soft-decision value can not be enhanced by simply extending path C and path D, which will be explained using Fig. 9. In the figure, numerals over the states represent survivor metrics corresponding to the respective states. On the other hand, numerals over and under states S_n+3 represent path metrics corresponding to branches connected to states S_n+3.

0138In the conventional example (V=L=2), the soft-decision value is calculated as a metric difference between path C and path D, and soft-decision value y_n=1. However, when considering paths connected to states S_n+3, path E has the minimum path metric among the paths where I_n=0 and path F has the minimum path metric among the paths wherein I_n=1. It is seen that either of path E and path F does not become a path extending from path C or path D. Here, if the soft-decision value is calculated by extending path C and path D, y_n= 1. On the other hand, in this first embodiment (L=2, V=4), the soft-decision value is calculated as a metric difference between path E and path F, and soft-decision value y_n = - 2.

0139In this manner, by setting the memory length V of the trellis to be greater than the memory length L of the estimated CIR, the soft-decision value with accuracy higher than the conventional example can be calculated. Conventionally, it has been considered that no effects can be achieved even by setting V greater than L. However, according to this first embodiment, not only the information of branches S_n+2/S_n+3 is simply added, but also the soft-decision value is calculated by considering paths after states S_n-3 when path A and path B join as in Fig. 6, so that the improvement in accuracy can be attained.

0140In the example of Fig. 7, they join at S_n+3. On the other hand, it is possible that they do not join until a subsequent state. However, by setting a difference between the memory length V of the trellis and the memory length L of the estimated CIR to a certain great extent, most of two paths used in the calculation of the soft-decision value join. When they join, it is deemed that the soft-decision value is calculated by considering the transmission sequences to its terminal end.

0141The inventors have derived the effect of this invention quantitatively. Fig. 10 shows bit error rates (BERs) at the output of the Viterbi decoder of the receiver having the present soft-decision device, which computer simulation measures.

0142The BERs are the same in the conventional example (L=V=2) and the first embodiment where L=2 and V=3. This is because, in this case, there is no substantial difference between using the path metrics corresponding to the branches and using the survivor metrics corresponding to the states. However, as V is increased in the first embodiment, such as V=4, V=5, the BER becomes smaller than the conventional example. When V=5 or thereabouts, the BER becomes similar to the case where the transmission sequences are considered to its terminal end.

0143In Fig. 10, when C/N=3dB, BER for V=4 is about 5x10^-3 and BER for V=5 is about 4x10^-3, while BER in the conventional example is about 1x10^-2. Similarly, when C/N=5dB, BER for V=4 is about 1x10^-4 and BER for V=5 is about 2x10^-5, while BER in the conventional example is about 2x10^-4. BER in this invention is no greater than half of the conventional example.

0144In the first embodiment, the soft-decision value is calculated based on the survivor metrics (corresponding to the states) as results of the ACS process. On the other hand, in the conventional example, the soft-decision value is calculated based on the path metrics (corresponding to the branches) as intermediate results of the ACS process. In the first embodiment, it is not necessary to store the path metrics being the intermediate results or use them in calculation. Thus, the first embodiment can be easily structured, than conventional, in the DSP (digital signal processor) which can perform at high-speed the calculation of the branch metrics and the ACS process en bloc.

0145Further, although the first embodiment sets the number U, which can be taken by the transmitted signal, to 2, it can be easily extended in case of the number greater than 2. It is more practical to use a combination of -1 and 1 rather than 0 and 1 as values which can be taken. Further, it may also be arranged that a value derived by multiplying the square error between the received signal and the replica by -1 is used as a branch metric, so as to select the maximum value instead of the minimum value in the ACS process and the calculation of the soft-decision value.

0146Further, in the foregoing description, explanation has been made to an example of applying to the Viterbi decoding, it is not limited to the Viterbi decoding, but can be applied to the soft-decision decoding algorithm in general. For example, it can be applied to the algebraic decoding, the maximum a posteriori probability decoding, the maximum-likelihood decoding and the sequential decoding.

0147Further, in the first embodiment, it may be considered that no deinterleave circuit is provided. Further, when the variation of the noise power can be ignored, it is not necessary to compensate the output of the soft-decision device using the gain of the amplifier.

Second Embodiment

0148The second embodiment of the present invention will now be explained. This embodiment of the invention is a structural example wherein the transmitted signal takes a binary digit of 0 or 1. In the figures, those components which are identified by the same reference numerals and/or symbols as those in the foregoing embodiment are the same or corresponding components.

0149Fig. 11 is a block diagram showing the second embodiment of a soft-decision device of the present invention. In the figure, 11 denotes a received signal input terminal, 12 a CIR estimating circuit, 13-1...13-(2N) branch metric producing circuits provided corresponding to branches as many as 2N, respectively, in the trellis diagram where a memory length is given by V (N=2^V, V≥L), 14-1...14-N ACS circuits provided corresponding to states as many as N (=2^V), respectively, in the trellis diagram where the memory length is given by V, 15 a shared memory, 16 a soft-decision value producing circuit, and 17 a soft-decision value output terminal.

0150Further, 101 denotes a selector switch for selecting whether to supply the CIR required by the branch metric producing circuits 13 from the CIR estimating circuit 12 or an estimated CIR updating circuit 102, so as to supply them, and 102 denotes the estimated CIR updating circuit for updating the CIR based on the output of the soft-decision value producing circuit 16.

0151Next, the operation of the soft-decision device of the second embodiment will be explained using Fig. 11. The received signal is formed by a training signal for the purpose of estimating the CIR and an information signal for the purpose of transmitting the information. The training signal is assumed to be known at the side of the soft-decision device. Hereinafter, the information signal portion of the received signal will be referred to as the received signal, and the training signal portion of the received signal will be referred to as the training signal.

0152The CIR estimating circuit 12 estimates the CIR based on the training signal and the known information of the training signal. While the estimated CIR circuit 102 updates the estimated CIR depending on the soft-decision value relative to the received signal, the selector switch 101 performs the switching operation so as to input the updated estimated CIR to the branch metric producing circuits 13-1...13-(2N).

0153The estimated CIR updating circuit 102 receives the estimated CIR from the CIR estimating circuit 12, the received signal from the received signal input terminal 11 and the soft-decision value from the soft-decision value producing circuit 16, and updates the estimated CIR based on the LMS (Least Mean Square) algorithm, and then outputs the updated estimated CIR to the branch metric producing circuits 13-1...13-(2N).

0154The update of the estimated CIR is performed in the following manner. Here, the estimated CIR outputted from the estimated CIR updating circuit 102 are given by g_0,n, g_1,n, ..., g_L,n. Subscript n represents a time.

0155Soft-decision value y_n is converted to hard-decision value x_n. The estimated CIR are updated based on the algorithm (LMS algorithm) of equation (56), using hard-decision value x_n, the received signal r_n and one-time prior estimated CIR g_0,n-1, g_1,n-1, ..., g_L,n-1.<maths id="math0033" num="(56)"><math display="block"><mrow><msub><mrow><mtext>g</mtext></mrow><mrow><mtext>i,n</mtext></mrow></msub><msub><mrow><mtext> = g</mtext></mrow><mrow><mtext>i,n-1</mtext></mrow></msub><msub><mrow><mtext> + α(r</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> -Σg</mtext></mrow><mrow><mtext>j, n-1</mtext></mrow></msub><msub><mrow><mtext> · x</mtext></mrow><mrow><mtext>n-j</mtext></mrow></msub><mtext>)</mtext></mrow></math><img file="EP0771079A2_D0033.tif" /></maths>    wherein i=0, ..., L and the sum Σ is derived for j=0...L.

0156Here, α represents a step size of the update algorithm. The estimated CIR outputted from the CIR estimating circuit 12 is, as it is, used as an initial value g_i,0 of the estimated CIR in the update algorithm.

0157The branch metric producing circuits 13-1...13-(2N), the ACS circuits 14-1...14-N, the shared memory 15 and the soft-decision value producing circuit 16 operate similarly as in the first embodiment.

0158In the second embodiment, by setting the memory length V of the trellis greater than the memory length L of the estimated CIR as in the first embodiment, the soft-decision value with accuracy higher than the conventional example can be achieved.

0159Further, in the second embodiment, when the CIR varies quickly, the soft-decision value with accuracy higher than the conventional example can be calculated by recursively updating the estimated CIR.

0160Further, in the second embodiment, the estimated CIR is updated per symbol. On the other hand, by updating the CIR per several symbols, the process volume can be reduced.

0161Further, by updating the estimated CIR in the estimated CIR updating circuit per symbol while updating the estimated CIR outputted to the branch metric producing circuits per several symbols only, the process volume which would be increased for updating the estimated CIR can be reduced.

0162Further, in the second embodiment, although the LMS algorithm is used for updating the estimated CIR, the update can be performed using other algorithms.

0163Further, in the second embodiment, the soft-decision value is calculated based on the survivor metrics (corresponding to the states) as results of the ACS process. On the other hand, in the conventional example, the soft-decision value is calculated based on the path metrics (corresponding to the branches) as intermediate results of the ACS process. Thus, the second embodiment can be easily structured, than conventional, in the DSP which can high-speed calculate the calculation of the branch metrics and the ACS process en bloc.

0164Further, although the second embodiment sets the number U, which can be taken by the transmitted signal, to 2, it can be easily extended in case of the number greater than 2. It is more practical to use a combination of -1 and 1 rather than 0 and 1 as values which can be taken. Further, it may also be arranged that a value derived by multiplying the square error between the received signal and the replica by -1 is used as a branch metric, so as to select the maximum value instead of the minimum value in the ACS process and the calculation of the soft-decision value.

Third Embodiment

0165The third embodiment of the present invention will now be explained. This embodiment of the invention is a structural example wherein the transmitted signal takes a binary digit of 0 or 1. In the figures, those components which are designated by the same reference numerals and/or symbols as those in the foregoing embodiments are the same or corresponding components.

0166Fig. 12 is a block diagram showing the third embodiment of a soft-decision device of the present invention. In the figure, 11 denotes a received signal input terminal, 12 a CIR estimating circuit, 113-1...113-(2M) branch metric producing circuits as many as (2M) (M=2_L), 114-1...114-N ACS circuits provided corresponding to states as many as N (=2^V), respectively, in the trellis where the memory length is given by V, 15 a shared memory, 116 a soft-decision value producing circuit, and 17 a soft-decision value output terminal.

0167Fig. 13 is a detailed block diagram of the ACS circuit 114-1 in Fig. 11. In the figure, 121 denotes a branch metric input terminal, 122-1 and 122-2 input terminals of one-time prior survivor metrics, 123-1 and 123-2 adders, 124 a comparator/selector, 125 a survivor metric output terminal, and 126-1 and 126-2 path metric output terminals.

0168Fig. 14 is a detailed block diagram of the soft-decision value producing circuit 116 in Fig. 11. In the figure, 131-1...131-N and 132-1...132-N denote path metric input terminals, 133-1 and 133-2 minimum value selecting circuits, and 134 a subtraction circuit.

0169Next, an operation of the soft-decision device of the third embodiment will be explained using Fig. 12. The received signal is formed by a training signal for the purpose of estimating the CIR and an information signal for the purpose of transmitting the information. The training signal is assumed to be known at the side of the soft-decision device. Hereinafter, the information signal portion of the received signal will be referred to as the received signal, and the training signal portion of the received signal will be referred to as the training signal.

0170First, an outline of the operation will be explained, and thereafter, the specific process contents will be explained using equations.

0171The CIR estimating circuit 12 estimates the CIR based on the training signal and the known information of the training signal.

0172The branch metric producing circuits 113-1...113-(2M) as many as 2M receive the received signal from the received signal input terminal 11 and the estimated CIR outputted from the CIR estimating circuit 12 and output branch metrics based on the partial sequences of the transmission sequences corresponding to the branches, respectively.

0173The ACS circuit 114-1 receives a branch metric outputted from the branch metric producing circuit 113-1 which outputs the branch metric corresponding to branches connected to the state which corresponds to the ACS circuit 114-1, and one-time prior survivor metrics outputted from the shared memory 15 corresponding to one-time prior states connected by the two branches connected to the state corresponding to the ACS circuit 114-1, and outputs path metrics calculated by the adding process and a current-time survivor metric calculated by the ACS process. The other ACS circuits 14-2...14-N operate similarly.

0174The shared memory 15 receives the current-time survivor metrics outputted from the ACS circuits 114-1...114-N to update the one-time prior survivor metrics.

0175The soft-decision producing circuit 116 receives path metrics outputted from the ACS circuits 114-1...114-N to calculate the soft-decision value and outputs the soft-decision value from the soft-decision value output terminal 17.

0176Operation of the ACS circuit 114-1 will now be explained using Fig. 13.

0177The branch metric input terminal 121 receives the branch metric from the branch metric producing circuit 113-1 which outputs the branch metric corresponding to two branches connected to the state which corresponds to the ACS circuit 114-1.

0178The survivor metric input terminals 122-1 and 122-2 input from the shared memory 15 the survivor metrics corresponding to one-time prior states connected by the two branches connected to the state which corresponds to the ACS circuit 114-1.

0179The adders 123-1 and 123-2 calculate the sums of the branch metric inputted via the branch metric input terminal 121 and the survivor metrics inputted via the survivor metric input terminals 122-1 and 122-2, respectively.

0180The path metric output terminals 126-1 and 126-2 output the sums outputted from the adders 123-1 and 123-2, respectively. The comparator/selector 124 receives the sums outputted from the adders 123-1 and 123-2 and compares the two sums so as to output smaller one from the survivor metric output terminal 125 as a current-time survivor metric.

0181Operation of the soft-decision value producing circuit 116 will now be explained using Fig. 14.

0182The path metric input terminals 131-1...131-N receive the path metrics corresponding to the branches where the most prior transmitted signals of the transmission sequences determined by the branches are 0. On the other hand, the path metric input terminals 132-1...132-N receive the path metrics corresponding to the branches where the most prior transmitted signals of the transmission sequences determined by the branches are 1.

0183The minimum value selecting circuit 133-1 receives the path metrics from the path metric input terminals 131-1...131-N and selects the minimum path metric among them to output it. The minimum value selecting circuit 133-2 receives the path metrics from the path metric input terminals 132-1...132-N and selects the minimum path metric among them to output it.

0184The subtraction circuit 134 subtracts the output of the minimum value selecting circuit 133-2 from the output of the minimum value selecting circuit 133-1 and outputs this difference from the soft-decision value output terminal 15 as a soft-decision value.

0185The concrete operation of the third embodiment will be explained using Fig. 12.

0186It is given that L=2, V=4, M=2L=4 and N=2^V=16.

0187The branch metric producing circuits 113-1...113-8 are provided corresponding to combinations of 3 symbols I_n-2, I_n-1 and I_n from the right (future side) among candidates I_n-4, I_n-3, I_n-2, I_n-1 and I_n of the transmitted signals determined by branches S_n-1/S_n, respectively. The branch metric producing circuit 113 calculates the branch metric based on equation (57) using received signal r_n, estimated transmission line characteristics g₀, g₁ and g₂, and candidates I_n-2, I_n-1 and I_n of the transmission sequences corresponding to each branch metric producing circuit.<maths id="math0034" num="(57)"><math display="block"><mrow><msub><mrow><mtext>E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [ XXI</mtext></mrow><mrow><mtext>n-2</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = [ABS {r</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> - (g</mtext></mrow><mrow><mtext>0</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> + g</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext> + g</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-2</mtext></mrow></msub><msup><mrow><mtext>)}]</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow></math><img file="EP0771079A2_D0034.tif" /></maths>

0188Although the expression of branch S_n-1/S_n = I_n-4 I_n-3 I_n-2 I_n-1 I_n has been used, since the branch metric in equation (57) can be determined uniquely free of I_n-4 and I_n-3, XXI_n-2 I_n-1 I_n is used. X represents an arbitrary transmitted signal.

0189If estimated CIR g₀, g₁ and g₂ do not change, a replica derived from estimated CIR g₀, g₁ and g₂ and candidates I_n-2 I_n-1, I_n of the transmitted signals, that is, () of equation (57), is always constant. In this case, it may also be arranged that this value is stored in the memory and updated only when the estimated CIR are updated.

0190Next, the ACS circuits 114-1...114-16 are provided corresponding to states S_n, respectively. The ACS circuits 114-1...114-16 receive branch metrics E_n [ XXI_n-2 I_n-1 I_n] outputted from the branch metric producing circuits 113 each of which outputs the branch metric corresponding to both branches S⁰_n-1/S_n and S¹_n-1/S_n connected to state S_n, and survivor metrics F_n [S⁰_n-1] and F_n [S¹_n-1] corresponding to one-time prior states S⁰_n-1 and S¹_n-1 outputted from the shared memory 15, and perform the adding process as expressed by equations (58) and (59) so as to calculate the current-time path metrics:<maths id="math0035" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(58)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [ XXI</mtext></mrow><mrow><mtext>n-2</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>0</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext>]</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(59)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [ XXI</mtext></mrow><mrow><mtext>n-2</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msup><mrow><mtext> [S</mtext></mrow><mrow><mtext>1</mtext></mrow></msup><msub><mrow><mtext> </mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext>] </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0035.tif" /></maths>

0191Then, the ACS circuits 114-1...114-16 each output path metrics F_n [S⁰_n-1/S_n] and F_n [S¹_n-1/S_n] from the path metric output terminals 126-1 and 126-2 and compare these two path metrics to output smaller one as survivor metric F_n [S_n] corresponding to state S_n.

0192For example, the ACS circuit 114 corresponding to state 1011 inputs branch metric E_n [XX011], and survivor metrics F_n-1 [0101] and F_n-1 [1101] corresponding to one-time prior states 0101 and 1101, and calculates path metrics corresponding to branches 01011 and 11011 from equations (60) and (61):<maths id="math0036" num=""><math display="block"><mrow><mtable><mtr><mtd><mrow><mtable><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(60)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [01011] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [XX011] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext> [0101]</mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr><mtr><mtd><mrow><mtable><mlabeledtr><mtext>(61)</mtext><mtd><mrow><msub><mrow><mtext>F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [11011] = E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [XX011] + F</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><mtext> [1101] </mtext></mrow></mtd></mlabeledtr></mtable></mrow></mtd></mtr></mtable></mrow></mtd></mtr></mtable></mrow></math><img file="EP0771079A2_D0036.tif" /></maths>

0193The ACS circuit 114 compares path metrics F_n [01011] and F_n [11011] and outputs smaller one as survivor metric F_n [1011] corresponding to state 1011.

0194The soft decision value producing circuit 116 receives the path metrics outputted from the ACS circuits 114-1...114-16 and calculates soft-decision value y_n-4 from equation (62):<maths id="math0037" num="(62)"><math display="block"><mrow><msub><mrow><mtext>y</mtext></mrow><mrow><mtext>n-4</mtext></mrow></msub><msub><mrow><mtext> = min (F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>]) - min (F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])</mtext></mrow></math><img file="EP0771079A2_D0037.tif" /></maths>

0195Here, the first term calculates the minimum path metric among the path metrics corresponding to the branches where the most prior transmitted signals I_n-4 among transmission sequences I_n-4 I_n-3, I_n-2, I_n-1 and I_n determined by branches S_n-1/S_n become 0. The second term calculates the minimum path metric among the path metrics corresponding to the branches where I_n-4=1.

0196In the third embodiment, as in the first embodiment, by setting the memory length V of the trellis to be greater than the memory length L of the estimated CIR, the soft-decision with accuracy higher than the conventional example can be achieved. Further, by sharing the branch metric producing circuits outputting the same value, the memory length V of the trellis can be enlarged while suppressing the increment of the circuit scale.

0197Further, in the third embodiment, as in the second embodiment, by arranging to update the CIR, the soft-decision with high accuracy can be achieved even when the variation of the CIR is large.

0198Further, although the third embodiment sets the number U, which can be taken by the transmitted signal, to 2, it can be easily extended in case of the number greater than 2. It is more practical to use a combination of -1 and 1 rather than 0 and 1 as values which can be taken. Further, it may also be arranged that a value derived by multiplying the square error between the received signal and the replica by -1 is used as a branch metric, so as to select the maximum value instead of the minimum value in the ACS process and the calculation of the soft-decision value.

Fourth Embodiment

0199The fourth embodiment of the present invention will now be explained. This embodiment of the invention is a structural example wherein the transmitted signal takes a binary digit of 0 or 1. In the figures, those components which are designated by the same reference numerals and/or symbols as those in the foregoing embodiments are the same or corresponding components.

0200Fig. 15 is a block diagram showing the fourth embodiment of the present invention. In the figure, 11 denotes a received signal input terminal, 12 a CIR estimating circuit, 141 a memory length estimating circuit for estimating a memory length of a CIR having the intersymbol interference (ISI) based on estimate values of the CIR provided by the CIR estimating circuit 12, 143-1...143(2N) branch metric producing circuits provided corresponding to branches as many as 2N, respectively, in the trellis where a memory length is given by V (N=2^V, V≥L), 14-1...14-N ACS circuits provided corresponding to states as many as N (=2^V), respectively, in the trellis where the memory length is given by V, 15 a shared memory, 16 a soft-decision value producing circuit, 144 a delay circuit which is controlled by the memory length estimating circuit 141 to delay the soft-decision value depending on a shift amount of the tap coefficient of the FIR filter based on an estimation result of the memory length estimating circuit 141, and 17 a soft-decision value output terminal.

0201Fig. 16 is a detailed block diagram of the memory length estimating circuit 141 in Fig. 15. In Fig. 16, 151 represents an estimated CIR input terminal, 152-1...152-(L+1) square circuits for squaring CIR g_i, 153 an adder for deriving the sum of outputs of the square circuits 152-1...152-(L+1), 156 a multiplier for multiplying an output of the adder 153 by a constant α to output a threshold Q, 154-1...154(L+1) comparators for comparing an output of the multiplier 156 and outputs of the square circuits 152-1...152-(L+1), respectively, 155 an estimated CIR correcting circuit for determining the number of taps so as to include all the tap coefficients g_i which cause the CIR g_i (to be exact, the squares of g_i) to be greater than the threshold Q and correcting by shifting the tap coefficients, 157 an estimated CIR output terminal of the CIR, and 158 a decision value delay output terminal for outputting the shift amount of the estimated CIR correcting circuit 155 as a delay.

0202Fig. 17 is a detailed block diagram of the branch metric producing circuit 143-1 in Fig. 15. In Fig. 17, 11 represents a received signal input terminal, 161 an estimated CIR input terminal, 22-1...22-(L+1) multiplying circuits, 23 a memory, 162-1...162-L selector switches provided for the multipliers 22-2...22-(L+1), respectively, for on/off switching the outputs of the multipliers 22-2...22-(L+1) based on the output of the memory length estimating circuit 141, 24 an adder for deriving the sum of outputs of the multiplier 22-1 and the selector switches 162-1...162-L, 25 a subtraction circuit, 26 a square circuit, and 27 a branch metric output terminal.

0203Next, the operation of the soft-decision device of the fourth embodiment will be explained using Fig. 15. The received signal is formed by a training signal for the purpose of estimating the CIR and an information signal for the purpose of transmitting the information. The training signal is assumed to be known at the side of the soft-decision device. Hereinafter, the information signal portion of the received signal will be referred to as the received signal, and the training signal portion of the received signal will be referred to as the training signal.

0204First, an outline of the operation will be explained, and thereafter, the more detailed process contents will be explained using equations.

0205The CIR estimating circuit 12 estimates the CIR based on the training signal and the known information of the training signal. The memory length estimating circuit 141 estimates the estimated memory length of the CIR based on the CIR outputted from the CIR estimating circuit 12 and outputs the estimated memory length of the CIR, the estimated CIR including information about the reception, and the delay of the decision value.

0206The branch metric producing circuits 143-1...143-(2N) as many as (2N) receive the received signal and the estimated CIR including information about the estimated memory length of the CIR outputted from the memory length estimating circuit 141, and output branch metrics based on the transmission sequences corresponding to the branches, respectively.

0207The ACS circuit 14-1 receives the branch metrics outputted from two lines of the branch metric producing circuits corresponding to two branches connected to the state which corresponds to the ACS circuit 14-1, and one-time prior survivor metrics outputted from the shared memory 15 corresponding to one-time prior states connected by those two branches so as to perform the ACS process and outputs a current-time survivor metric. The other ACS circuits 14-2...14-N operate similarly.

0208The shared memory 15 inputs the current-time survivor metrics outputted from the ACS circuits 14-1...14-N to update the one-time prior survivor metrics.

0209The soft-decision producing circuit 16 inputs the current-time survivor metrics outputted from the ACS circuits 14-1...14-N to calculate the soft-decision value and outputs the soft-decision value.

0210The delay circuit 144 delays the soft-decision value outputted from the soft-decision value producing circuit 16 based on the delay of the decision value outputted from the memory length estimating circuit 141. This delayed soft-decision value is outputted from the soft-decision value output terminal 17.

0211Operation of the memory length estimating circuit 141 will be explained using Fig. 16. The square circuits 152-1...152-(L+1) calculate the squares of the respective tap coefficients (estimated CIR) outputted from the CIR estimating circuit 12.

0212The adder 153 calculates the sum of the values outputted from the square circuits 152-1...152-(L+1).

0213The multiplier 156 multiplies the sum outputted from the adding circuit 153 by α.

0214The comparators 154-1...154-(L+1) compare the values outputted from the square circuits 152-1...152-(L+1) and the multiplication result outputted from the multiplier 156.

0215The estimated CIR correcting circuit 155 corrects the estimated CIR based on the estimated CIR received from the estimated CIR input terminal 151 and the comparison results outputted from the comparators 154-1...154-(L+1), and outputs the correction result (information about the tap coefficients and the number of taps) from the estimated CIR output terminal 156.

0216Operation of the branch metric producing circuit 143-1 will now be explained using Fig. 18.

0217The multiplying circuits 22-1...22-(L+1) calculate the products of the estimated CIR (tap coefficients) received from the estimated CIR input terminal 161 and a portion of the candidates of the transmission sequences determined by the branches corresponding to the branch metric producing circuit 143-1 stored in the memory 23, that is, the (L+1) newer transmitted signals in the (V+1) transmitted signals determined by the branches. The selector switches 162-1...162-L are controlled by the estimated CIR (the number of taps) outputted from the memory length estimating circuit 142.

0218The adder 24 receives the multiplication results outputted from the multipliers 22-1...22-(L+1) via the selector switches 162-1...162-L and calculates the sum thereof.

0219The subtraction circuit 25 calculates a difference between the sum outputted from the adder 24 and the received signal inputted from the received signal input terminal 11.

0220The square circuit 26 calculates the square of the difference outputted from the subtraction circuit 25 and outputs the result via the branch metric output terminal 27 as the branch metric.

0221Further, the concrete operation of the fourth embodiment will be explained using Figs. 15...17. It is given that V=4, L=3, N=2^V=16.

0222In Fig. 16, the comparing circuits 154-1...154-4 receive the tap coefficients (estimated CIR) g₀, g₁, g₂, g₃ and compare the threshold Q and the squares of the absolute values of the tap coefficients g_i (i=0, 1, 2, 3) in the following equation (63):<maths id="math0038" num="(63)"><math display="block"><mrow><msub><mrow><mtext>Q = α [{ABS(g</mtext></mrow><mrow><mtext>0</mtext></mrow></msub><msup><mrow><mtext>)}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup><msub><mrow><mtext> + {ABS (g</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msup><mrow><mtext>)}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup><msub><mrow><mtext> + {ABS (g</mtext></mrow><mrow><mtext>2</mtext></mrow></msub><msup><mrow><mtext>)}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup><msub><mrow><mtext> + {ABS (g</mtext></mrow><mrow><mtext>3</mtext></mrow></msub><msup><mrow><mtext>)}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup><mtext>]</mtext></mrow></math><img file="EP0771079A2_D0038.tif" /></maths>

0223The estimated CIR correcting circuit 155 determines the number of taps so that the tap coefficients g which cause {ABS (g_i)}²>Q are all included, and shifts the tap coefficients to prevent {ABS (g₀)}²≤Q. This shift amount becomes the delay of he decision value. The estimated CIR correcting circuit 155 outputs the shifted tap coefficients and the effective number of taps to the branch metric producing circuits 1431...143-32 as the CIR, and outputs the delay of the decision value to the delay circuit 144.

0224An example of correcting the estimated CIR is shown in Fig. 18. Among the tap coefficients estimated by the CIR estimating circuit 12, g₁ and g₃ provide {ABS (g_i)}²>Q. Accordingly, g₁, g₂ and g₃ are the effective tap coefficients. For avoiding {ABS (g₀)}²≤Q, g₁, g₂ and g₃ are shifted to g'₀, g'₁ and g'₂, respectively. At this time, since the tap coefficients are shifted by one symbol, the delay of the decision value becomes one symbol.

0225The branch metric producing circuits 143-1...143-32 are provided corresponding to branches S_n-1/S_n, respectively. These branch metric producing circuits 143 receive received signal r_n, tap coefficients g'₀, g'₁, g'₂ and g'₃ and the effective number of taps (L_a + 1) and calculate the branch metrics from equation (64) based on candidates I_n-3, I_n-2, I_n-1 and I_n held at the branch metric producing circuits 143:<maths id="math0039" num="(64)"><math display="block"><mrow><msub><mrow><mtext>E</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n-1</mtext></mrow></msub><msub><mrow><mtext>/S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>] = {ABS (r</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> - Σg'</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext> I</mtext></mrow><mrow><mtext>n-i</mtext></mrow></msub><msup><mrow><mtext>)}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup></mrow></math><img file="EP0771079A2_D0039.tif" /></maths>    wherein the sum Σ is derived for i=0, ..., La.

0226The ACS circuits 14-1...14-N, the shared memory 15 and the soft-decision value producing circuit operate similarly as in the foregoing other embodiments.

0227The delay circuit 144 delays soft-decision value y_n by a shift amount of the tap coefficient in the memory length estimating circuit 141. Although candidates I_n-3, I_n-2, I_n-1 and I_n of the transmitted signals held at the branch metric producing circuits 143 become I_n-D-3, I_n-D-2, I_n-D-1 and I_n-D depending on the shift amount, since this sequence is constant free of time, time (n-D) is handled as time n in the foregoing explanation.

0228In the soft-decision device of the fourth embodiment, by changing the memory length L of the CIR, it is possible to follow even when the memory length of the actual CIR varies so that the soft-decision value with high accuracy can be achieved.

0229In the fourth embodiment, the memory length estimating circuit 141 estimates the memory length L of the CIR. On the other hand, by estimating also the memory length V of the trellis similarly and stopping an unnecessary ACS circuit based on this estimation value, the mean process volume of the device can be reduced.

0230Further, in the soft-decision device of the fourth embodiment, by stopping the branch metric producing circuit outputting the same value based on the estimated memory length of the CIR as in the third embodiment, the mean process volume of the device can be reduced.

0231Further, in the estimated CIR correcting circuit 155, it may be arranged to correct the estimated CIR so as to forcibly set the tap coefficient, which causes {ABS (g'_i)}²≤Q, to g_i=0. For example, since {ABS (g'₁)}²≤Q in the output of the estimated CIR correcting circuit in Fig. 22, g'₁=0 is set.

0232Further, in the estimated CIR correcting circuit 155, by setting non-effective tap coefficients (tap coefficients causing {ABS (g'_i)}²≤Q) to 0, the selector switches in the branch metric producing circuits 143-1...143(2N) can be omitted.

0233Further, in the soft-decision device of the fourth embodiment, as in the second embodiment, by arranging to update the estimated CIR recursively, even when the variation of the CIR is large, it is possible to track this so as to achieve a soft-decision with high accuracy.

0234Further, although the soft-decision device of the fourth embodiment 4 sets the number U, which can be taken by the transmitted signal, to 2, it can be easily extended in case of the number greater than 2. It is more practical to use a combination of -1 and 1 rather than 0 and 1 as values which can be taken. Further, it may also be arranged that a value derived by multiplying the square of error between the received signal and the replica by -1 is used as a branch metric, so as to select the maximum value instead of the minimum value in the ACS process and the calculation of the soft-decision value.

Fifth Embodiment

0235The fifth embodiment of the present invention will now be explained. This embodiment of the invention is a structural example wherein the transmitted signal takes a binary digit of 0 or 1. In the figures, those components which are assigned the same reference numerals and/or symbols as those in the foregoing embodiments are the same or corresponding components.

0236Fig. 19 is a block diagram showing the embodiment of a soft-decision device of the present invention. In Fig. 19, 11 denotes a received signal input terminal, 12 a CIR estimating circuit, 143-1...143-(2N) branch metric producing circuits provided corresponding to branches as many as (2N), respectively, in the trellis where a memory length is given by V (N=2^V, V≥L), 171 a memory length estimating circuit, 172 an error estimating detecting circuit for estimating an error based on the CIR from the CIR estimating circuit 12 and the received signal from the received signal input terminal 11 and outputs an estimated error to the memory length estimating circuit 171 and a soft-decision value producing circuit 176, 174-1...174-N ACS circuits provided corresponding to states as many as N (=2^V), respectively, in the trellis diagram where the memory length is given by V, 175 a shared memory, 176 the soft-decision value producing circuit, 144 a delay circuit, 17 a soft-decision value output terminal, 178 a hard-decision value producing circuit and 179 a hard-decision value output terminal.

0237Fig. 20 is a detailed block diagram of the memory length estimating circuit 171 in Fig. 19. In Fig. 20, 151 represents an estimated CIR input terminal, 152-1...152-(L+1) square circuits for squaring the estimated CIR, respectively, 154-1...154-(L+1) comparators for comparing outputs of the square circuits 152-1...152-(L+1) and an output of a multiplier 156, respectively, 155 an estimated CIR correcting circuit for correcting the CIR based on outputs of the comparators 154-1...154-(L+1) and the estimated CIR, 156 the multiplier for multiplying the estimated error power from an estimated error power input terminal 181 by a constant α, 157 an estimated CIR output terminal connected to the estimated CIR correcting circuit 155, 158 a decision value delay output terminal connected to the estimated CIR correcting circuit 155, and 181 the estimated error power input terminal for receiving the estimated error power from the error estimating circuit 172.

0238Fig. 21 is a detailed block diagram of the soft-decision value producing circuit 176 in Fig. 19. In Fig. 21, 41-1...41-(N/2) and 42-1...42-(N/2) denote survivor metric input terminals, 43-1 and 43-2 minimum value selecting circuits, 44 a subtraction circuit, 181 the estimated error power input terminal, 192 a division circuit for dividing the output of the subtraction circuit 44 by the estimated error power value from the terminal 181 so as to output soft-decision value y_n, and 197 a decision value output terminal connected to the division circuit 192.

0239Next, the operation of the soft-decision device of the fifth embodiment will be explained using Fig. 18. The received signal is formed by a training signal for the purpose of estimating the CIR and an information signal for the purpose of transmitting the information. The training signal is assumed to be known at the side of the soft-decision device. Hereinafter, the information signal portion of the received signal will be referred to as the received signal, and the training signal portion of the received signal will be referred to as the training signal.

0240The CIR estimating circuit 12 estimates the CIR based on the training signal and the known information of the training signal.

0241The error estimating circuit 172 receives the received signal and the estimated CIR outputted from the CIR estimating circuit 12 to estimate the estimated error power and outputs its estimation value.

0242The memory length estimating circuit 171 receives the CIR outputted from the CIR estimating circuit 12 and the estimated error power outputted from the error estimating circuit 172 to estimate the memory length of the CIR and outputs the estimated CIR including information about the estimated memory length of the CIR and the delay of the decision value.

0243The branch metric producing circuits 173-1...173-(2N) as many as (2N) receive the received signal and the estimated CIR including information about the memory length of the estimated CIR outputted from the memory length estimating circuit 171, and output branch metrics based on the transmission sequences corresponding to the branches, respectively.

0244The ACS circuit 174-1 receives the branch metrics outputted from two branch metric producing circuits corresponding to two branches connected to the state which corresponds to the ACS circuit 174-1, and one-time prior survivors and survivor metrics outputted from the shared memory 175 corresponding to one-time prior states connected by those two branches so as to perform the ACS process and outputs current-time survivor and survivor metric. The other ACS circuits 174-2...174-N operate similarly.

0245The shared memory 175 receives the current-time survivors and survivor metrics outputted from the ACS circuits 174-1...174-N to update the one-time prior survivors and survivor metrics.

0246The soft-decision producing circuit 176 receives the current-time survivor metrics outputted from the ACS circuits 174-1...174-N and the estimated error power outputted from the error estimating circuit 172 to perform the soft decision and outputs the soft-decision value.

0247The delay circuit 144 receives the delay of the decision value outputted from the memory length estimating circuit 171 and the soft-decision value outputted from the soft-decision value producing circuit 176 to delay the soft-decision value by the delay of the decision value, and then outputs it from the soft-decision value output terminal 17.

0248The hard-decision value producing circuit 178 receives the current-time survivors and survivor metrics outputted from the ACS circuits 1741...174-N and the delay of the decision value outputted from the memory length estimating circuit 171 and outputs the hard-decision value from the hard-decision value output terminal 179 while estimating the transmission sequences in consideration of the delay of the decision value.

0249Operation of the memory length estimating circuit 171 will now be explained using Fig. 20. The square circuits 152-1...152-(L+1) calculate the squares of the tap coefficients (estimated CIR) outputted from the CIR estimating circuit 12.

0250The multiplier 156 multiplies the estimated error power inputted from the estimated error power input terminal 181 by α.

0251The comparators 154-1...154-(L+1) compare the values outputted from the square circuits 152-1...152-(L+1) and the multiplication result outputted from the multiplier 156.

0252The estimated CIR correcting circuit 155 performs correction of the estimated CIR based on the estimated CIR inputted from the estimated CIR input terminal 151 and the comparison results outputted from the comparators 154-1...154-(L+1), and outputs the correction result (information about the tap coefficients and the number of taps) from the estimated CIR output terminal 157.

0253Operation of the soft-decision producing circuit 176 will now be explained using Fig. 21. The survivor metric input terminals 41-1...41-(N/2) receive the survivor metrics outputted from the ACS circuits corresponding to the states where the most prior transmitted signals forming the states are 0. On the other hand, the survivor metric input terminals 42-1...42-(N/2) receive the survivors outputted from the ACS circuits corresponding to the states where the most prior transmitted signals forming the states are 1.

0254The minimum value selecting circuit 43-1 receives the survivor metrics from the survivor metric input terminals 41-1...41-(N/2) and outputs the minimum survivor metric among them. The minimum value selecting circuit 43-2 receives the survivor metrics from the survivor metric input terminals 42-1...42-(N/2) and outputs the minimum survivor metric among them.

0255The subtraction circuit 44 subtracts the output of the minimum value selecting circuit 43-2 from the output of the minimum value selecting circuit 43-1.

0256The division circuit 192 divides the value outputted from the subtraction circuit 44 by the estimated error power from the estimated error power input terminal 181 and outputs this calculation result from the soft-decision value output terminal 197 as a soft-decision value.

0257Further, the specific operation of the fifth embodiment will now be explained. It is given that V=4, L=3 and N=2^V=16.

0258The CIR estimating circuit 12 estimates the tap coefficients (estimated CIR) g₀, g₁, g₂ and g₃ based on training signal r'_n and known information I'_n of the training signal.

0259The error estimating circuit 172 calculates the estimated error power from equation (65):<maths id="math0040" num="(65)"><math display="block"><mrow><msub><mrow><mtext>e =Σ{ABS (r'</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> -Σg</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext>·I'</mtext></mrow><mrow><mtext>n-i</mtext></mrow></msub><msup><mrow><mtext>)}</mtext></mrow><mrow><mtext>2</mtext></mrow></msup><msub><mrow><mtext> / (N</mtext></mrow><mrow><mtext>tra</mtext></mrow></msub><mtext> - 2)</mtext></mrow></math><img file="EP0771079A2_D0040.tif" /></maths>    wherein the sum Σ in brackets is derived for i=0, ..., 3. The sum Σ outside brackets is derived for j=3, ..., N_tra.

0260On the other hand, the estimated error power e represents the mean error power between the actual received signal and the replica of the received signal.

0261The multiplier 156 multiplies the estimated error power e by α and calculates the threshold Q (=α · e). The comparators 154-1...154-4 compare the threshold Q and the squares of the absolute values of g_i, that is, {ABS (g_i)}² (i=0, 1, 2, 3), respectively.

0262The estimated CIR correcting circuit 155 determines the number of taps so that the tap coefficients g which cause {ABS (g_i)}² >Q are all included, and shifts the tap coefficients to prevent {ABS (g₀)}²≤Q. This shift amount becomes the delay of the received signal. The estimated CIR correcting circuit 155 outputs the shifted tap coefficients, the effective number of taps to the branch metric producing circuits 143-1...143-32 as the estimated CIRs and outputs the delay of the received signal to the delay circuit 144 as the delay of the decision value.

0263The hard-decision value producing circuit 178 receives the survivors and the survivor metrics outputted from the ACS circuits 174-1...174-16 and the delay of the decision value outputted from the memory length estimating circuit 171, determines sequence along the survivor, where the survivor metric becomes minimum, to be the transmitted sequences, and outputs a signal delayed by the delay of the decision value from the hard-decision value output terminal 179 as hard-decision values.

0264The soft-decision value producing circuit 176 inputs the survivor metrics outputted from the ACS circuits 174-1...174-16 and calculates soft-decision value y_n-3 from equation (66):<maths id="math0041" num="(66)"><math display="block"><mrow><msub><mrow><mtext>y</mtext></mrow><mrow><mtext>n-3</mtext></mrow></msub><msub><mrow><mtext> = {min (F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext>]) - min (F</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><msub><mrow><mtext> [S</mtext></mrow><mrow><mtext>n</mtext></mrow></msub><mtext>])} / e</mtext></mrow></math><img file="EP0771079A2_D0041.tif" /></maths>

0265Here, the first term of the numerator calculates the minimum survivor metric among the survivor metrics corresponding to the states where the most prior transmitted signals I_n-3 among the transmission sequences I_n-3, I_n-2, I_n-1 and I_n determined by states S_n become 0. The second term of the numerator calculates the minimum survivor metric among the survivor metrics corresponding to the states where I_n-3=1.

0266In the soft-decision device of the fifth embodiment, since the soft-decision value is normalized by the estimated error power, even when the noise power changes, the proper soft-decision value can be derived so that a soft decision with high accuracy can be achieved.

0267Further, in the soft-decision device of the fifth embodiment, by changing the memory length L of the estimated CIR, a soft-decision with high accuracy can be achieved even when the memory length of the actual CIR varies.

0268Further, in the soft-decision device of the fifth embodiment, the memory length estimating circuit 141 estimates the memory length L of the estimated CIR. On the other hand, by estimating also the memory length V of the trellis similarly and stopping an unnecessary ACS circuit based on this estimation value, the mean process volume of the device can be reduced.

0269Further, in the soft-decision device of the fifth embodiment, by stopping the branch metric producing circuit outputting the same value based on the estimated memory length of the CIR as in the third embodiment, the mean process volume of the device can be reduced.

0270Further, in the soft-decision device of the fifth embodiment, as in the second embodiment, by arranging to update the estimated CIR recursively, even when the variation of the CIR is large, the soft-decision with high accuracy can be achieved.

0271Further, although the soft-decision device of the fifth embodiment sets the number U, which can be taken by the transmitted signal, to 2, it can be easily extended in case of the number greater than 2. It is more practical to use a combination of -1 and 1 rather than 0 and 1 as values which can be taken. Further, it may also be arranged that a value derived by multiplying the square error between the received signal and the replica by -1 is used as a branch metric, so as to select the maximum value instead of the minimum value in the ACS process and the calculation of the soft-decision value.

0272The hard-decision value calculated from the soft-decision value is impaired in error rate as compared with the hard-decision value determined through the maximum-likelihood sequence estimation. Accordingly, in the system where convolutionally encoded data and non-encoded data are transmitted in combination, if the hard-decision value is calculated from the result of the soft decision, the error rate of the non-encoded signal is impaired. In the soft-decision device in the fifth embodiment, since the hard decision based on the maximum-likelihood sequence estimation is performed simultaneously with the soft-decision, the error rate of the non-encoded signal can be improved in the foregoing system.