JPH0758672A - Digital adaptive equalizer - Google Patents

Abstract

PURPOSE:To make a circuit scale small and to reduce a calculation amount by providing a successive equalization part using successive decoding algorithm. CONSTITUTION:By a received training series, the impulse response CIR of a propagation line is calculated in a propagation line impulse response calculation part 1. A replica generation part 2 generates replica signals based on the impulse response CIR of the propagation line and equalized output signals and a branch metric calculation part 3 calculates the square of the absolute value of a difference between the replica signals and reception signals through the propagation line as branch metric. A conversion table 4 converts the branch metric to Fano metric obtained from the distribution function of the branch metric for instance and the successive equalization part 5 equalizes the reception signals by applying the successive decoding algorithm such as the Fano algorithm and stack algorithm, etc.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital adaptive equalizer which applies a successive decoding algorithm to equalize a received signal. Since the signal received through the propagation path is subjected to various kinds of distortion and interference, a digital adaptive equalizer, for example, is used to equalize and identify the received signal. This digital adaptive equalizer is a decision feedback equalizer (DFE: DecisionFeed) that feeds back an equalized output signal one symbol before to cancel the influence of a delayed wave.
back equalizer) is common. A maximum likelihood sequence estimation type equalizer (MLSE; Maximum Likel) that applies a Viterbi decoding algorithm and equalizes a received signal by maximum likelihood estimation.
ihood Sequence Estimation) is also proposed. It is desired to economically realize such a digital adaptive equalizer.

[0002]

2. Description of the Related Art A decision feedback type equalizer DFE as a digital adaptive equalizer has a relatively simple structure and requires a small amount of calculation, but on the other hand, its equalization performance is not sufficient for application to digital radio communication and the like. Absent. On the other hand, the maximum likelihood sequence estimation equalizer MLSE can exhibit sufficient equalization performance even for digital wireless communication and the like. FIG. 7 is an explanatory diagram of a conventional example and shows a maximum likelihood sequence estimation type equalizer to which the Viterbi decoding algorithm is applied.

In FIG. 7, reference numeral 41 is a propagation path impulse response calculation unit, 42 is a replica generation unit, 43 is an adder, and 44.
Is a maximum likelihood sequence estimation unit, 45 is a branch metric calculation unit,
46 is an ACS circuit, 47 is a path memory, and 48 is a selector.

The selector 48, when receiving a known training sequence at the beginning of a burst signal or the like, switches the training sequence to the propagation path impulse response calculation section 41 and the replica generation section 42. The impulse response calculation unit 41 uses a known training sequence to generate an impulse response (CIR; ChannelImp) of the propagation path.
ulse Response) is calculated, and when the digital signal is received, the selector 48 is switched to the maximum likelihood sequence estimation unit 44 side, and the equalized output signal is converted into the propagation path impulse response calculation unit 41.
In addition to the replica generation unit 42, a replica signal corresponding to a delayed wave is formed, the adder 43 calculates the difference between the received signal and the replica signal, and inputs the difference to the maximum likelihood sequence equalization unit 44.

The maximum likelihood sequence estimator 44 includes a branch metric calculator 45, an ACS circuit 46 including an adder (A), a comparator (C) and a selector (S), and a path memory 47.
And the difference between the replica signal from the replica generator 42 and the received signal is output from the adder 43 and added to the maximum likelihood sequence estimator 44.

The branch metric calculator 45 calculates a branch metric for each branch, and the ACS circuit 4
The adder (A) of 6 adds the branch metric to the path metric of the last surviving path, the comparator (C) compares the path metric between the branches, and the selector (S) selects the larger path metric. Select the surviving path and select the surviving path from the path memory 4
7 and the equalized output signal is sent from the final stage of the path memory 47.

The equalized output signal is input to the replica generator 42 via the selector 48, and the replica signal corresponding to the impulse response CIR of the propagation path is generated by the propagation path impulse response calculation unit 41 and added to the adder 43. Then, the difference from the received signal is input to the maximum likelihood sequence estimation unit 44. A digital adaptive equalizer using such a Viterbi decoding algorithm is also disclosed in, for example, Japanese Patent Laid-Open No. 4-261210.

[0008]

As described above, the decision feedback type equalizer DFE as a digital adaptive equalizer of the prior art is one which feeds back the decision result and eliminates the influence of the delayed wave. Is relatively simple, but has a drawback that the equalization performance is not sufficient in data communication and the like. The maximum likelihood sequence estimation type equalizer MLSE has sufficient equalization performance, but has a large number of modulation methods. As the number of values and the delay time of the delayed wave increase,
There are drawbacks that the circuit scale including the path memory and the like becomes large, the amount of calculation exponentially increases, the equalization processing time becomes long, and the cost becomes expensive. An object of the present invention is to reduce the amount of calculation while maintaining desired equalization performance.

[0009]

A digital adaptive equalizer according to the present invention will be described with reference to FIG. 1. The digital adaptive equalizer includes a channel impulse response calculator 1 for calculating a channel impulse response CIR by a known training sequence. , A replica generation unit 2 that generates a replica signal from the equalized output signal according to the impulse response CIR of the propagation path obtained by the propagation path impulse response calculation unit 1, and the replica signal by the replica generation unit 2 and the reception via the propagation path. A branch metric calculator 3 that calculates the square of the absolute value of the difference from the signal as a branch metric, and this branch metric calculator 3
A conversion table 4 for converting the branch metric to an integer metric, and a serial equalization unit 5 for equalizing a received signal by a sequential decoding algorithm using the integer metric by the conversion table 4.

Further, the propagation path impulse response calculation unit 1 calculates the propagation path impulse response for each path extension in the successive equalization unit 5 based on the difference between the replica signal and the received signal via the propagation path. It can be configured to.

It is also possible to provide a digital adaptive equalizer corresponding to the diversity reception path and select the equalized output signal of the diversity reception path which has completed the equalization process first.

[0012]

The impulse response CIR of the propagation path is adjusted by the training sequence received prior to the reception of the digital signal.
Is calculated in the propagation path impulse response calculation unit 1. The replica generation unit 2 uses the impulse response CIR of this propagation path.
And a replica signal is generated based on the equalized output signal. The branch metric calculation unit 3 calculates the square of the absolute value of the difference between the received signal and the replica signal via the propagation path as a branch metric, and the conversion table 4 calculates this branch metric from, for example, the distribution function of the branch metric. Convert to the obtained fanometric. The successive equalization unit 5
The received signals are equalized by applying a sequential decoding algorithm such as a Fano Algorithm (Fano Algorithm) or a stack algorithm (Stack Algorithm).

Further, the propagation path impulse response calculation unit 1 calculates the propagation path impulse response based on the received signal even after calculating the propagation path impulse response CIR by the training sequence, and changes in the propagation path characteristics. By following and updating the impulse response, the equalization characteristic can be maintained.

Further, when a plurality of diversity receiving paths such as space diversity are provided, a digital adaptive equalizer is provided in each diversity receiving path, and the digital adaptive equalizer for the diversity receiving path whose equalization processing is completed first is
Since the path of the received signal has the fewest errors, the equalized output signal from the digital adaptive equalizer is selected and output.

[0015]

2 is an explanatory view of a first embodiment of the present invention, in which 11 is a channel impulse response calculation unit, 12 is a replica generation unit, 13 is a branch metric calculation unit, 14 is a conversion table, and 15 is a conversion table. A sequential equalization unit, 16 is an adder, 17 is a reception buffer, 18 is a stack control unit, 19 is a stack memory, and 20 is a sub-stack memory.

The received signal is temporarily stored in the reception buffer 17 and is sequentially read out as the equalization processing in the successive equalization unit 15 progresses. Further, the propagation path impulse response calculation unit 11
The impulse response CIR of the propagation path is calculated using a known training sequence. In addition, the branch metric calculation unit 13
Uses the square of the absolute value of the difference between the received signal from the adder 16 and the replica signal as the branch metric. This branch metric is added to the conversion table 14 and converted into an integer metric.

Further, when the stack algorithm is used, the sequential equalization section 15 uses the stack control section 18 as shown in the figure.
The stack memory 19 and the sub-stack memory 20 can be included, and the contents of the stack memory 19 can be traced back to obtain an equalized output signal. Further, the sub-stack memory 20 stores the nodes extending the path in order of increasing path metric.

FIG. 3 is an explanatory view of the stack algorithm, in which the circles indicate the nodes, and the numbers 0 to 30 therein indicate the node numbers. A case of two branches is shown in which the upper path is extended when the received symbol is assumed to be "0", and the lower path is extended when it is assumed to be "1". In addition, at the node after extending the path,
The value obtained by sequentially adding the path fanometrics is taken as the path metric. For example, at node 0, assuming that the received symbol is "0", the phanometric is -7, "1".
Assuming that it is -1, the path metric of the node 1 after extending the path (v) is -7, and the path metric of the node 2 after extending the path (i) is -1.

In this case, since the path metric of the node 2 is large, the path is extended from this node 2, and the fanometric metric at the time of assuming "0" at the node 2 and at the time of assuming "1". If both are -4, the path (i
Either i) or (iii) may be extended, but for example, the path (ii) assumed to be “0” can be determined in advance. In this case, the path metric of the node 5 is −5, which is larger than −7 of the path metric of the node 1, so that the path is extended from this node 5, and it is assumed that the node 5 is “0”. Sometimes the fanometric is -1,
If it is set to -7 when it is assumed to be "1", the path (iv) is extended because the fanometric metric when it is assumed to be "0" is large.

The path metric of the node 11 ahead of the path (iv) is -6, which is smaller than -5 of the path metric of the node 6 and returns to the node 6, and when the path is extended from this node 6, the nodes 13 and 14 are obtained. Assuming that each of the path metrics has a value of -9 as shown in the figure, the path metric of node 11 is -6, which means that the path is extended from node 11. In that case, assuming that the Fanometric is -4 both when assuming "0" and when assuming "1", the path metrics of the nodes 23 and 24 are both-.
The value becomes 10, which is smaller than −7 of the path metric of the node 1.

In this case, the path is extended after returning to the node 1, and at this node 1, it is assumed that "0" is -7, and "1" is -1. , The path (vi) is extended, and the path metric of the node 4 after that is -8. At node 4, if the fanometric is -2 when assuming "0" and 0 when assuming "1", the path (vii) assuming that the fanometric is large is extended. .

The node 1 at the end of the extended path (vii)
If 0 is assumed to be “0”, the phanometric is 2, and if “1” is assumed, it is −10.
The path (viii) assumed to be “0” is extended. Then, the path metrics of the nodes 21, 22, 23, 24 at the final stage are -6, -18, -10, -10, and the node 2
Since the path metric (-6) of 1 is the maximum, traceback is performed from this node 21 and the path (viii) indicated by the thick line,
(vii), (vi), (v) is selected and the equalized output signal is "011.
It becomes 0 ".

The above-mentioned phanometric γ _j at a certain time j is represented by γ _j = Σ [log ₂ {P (r _ji | x _ji ) / f (r _ji )}-B] (1) To be done. Note that Σ indicates that the sum of each bit from i = 1 to i = N is obtained, N is the coding rate in coding, x _ji is a transmission symbol, r _ji is a reception symbol, and P (r
_ji | x _ji ) is the probability that the received symbol r _ji is received when the transmitted symbol x _ji is transmitted, f (r _ji ) is the occurrence probability of the received symbol r _ji , B is a bias value, and this bias value B is It is known that when the received symbol is certain, the metric is selected to be a positive value, and the optimum value in decoding the error correction code is 1 / N.

The above-mentioned phanometric in the sequential equalization is slightly different from the original phanometric in the sequential decoding, but in the embodiment of the present invention, it is described as the phenometric of the same expression. To do. Assuming that independent Gaussian noises x and y with a mean value of 0 are added with equal variances σ ² to the real part component and the imaginary part component of the received signal, the joint probability distribution p (x, y of the noises x and y is added. y) becomes p (x, y) = (1 / 2πσ ² ) exp [− (x ² + y ² ) / 2σ ² ] ... (2).

In the equation (2), the noise x, y is replaced with x = R cos Θ, y = R sin Θ in polar coordinates, and the noise amplitude distribution p (R) is obtained. P (R) = (R / σ ²⁾ becomes ^{exp (-R 2 / 2σ 2)} ... (3).

The probability P (a, b) that the branch metric is equal to or greater than a and equal to or less than b is that the amplitude R is equal to or greater than a ^1/2 and equal to b ^1/2.
Since the probability is as follows, P (a, b) = ∫p (R) dR = exp (−a / 2σ ² ) −exp (−b / 2σ ² ) ... (4) In addition, ∫ indicates the integration from a ^1/2 to b ^1/2 .

Fanometric FM based on the equation (4)
(N) can be expressed as FM (n) = <α × {log (P (nδ, (n + 1) δ)) + β}> (5). Where α and β are constants, <X>
<> Indicates a symbol representing the maximum integer not exceeding X. For example, α = 8.0, β = 4.0, δ = 6.0 / 256, and n = 0 to 255 can be integers.

In the embodiment of the present invention, the fanometric FM (n) is obtained by the conversion table 14, and from the distribution function of the branch metric by the square of the absolute value calculated by the branch metric calculator 13, (5) The Fano metric obtained by the equation is stored in the conversion table 14, the branch metric is input as an address, the Fano metric is read and input to the sequential equalization unit 15.

Various methods are known for modulating the transmission signal. For example, when performing burst communication by the M-value modulation method, a training sequence added to the head of the burst is used to calculate the propagation path impulse response. The unit 11 calculates the impulse response CIR of the propagation path. Further, the replica generation unit 12 searches the sub-stack memory 20 in which the information of the nodes is stored in the descending order of the path metric, and determines M from the determination signal point at the past point in time considering the delay time of the delayed wave and the impulse response CIR. A replica signal for each of the states is generated.

The difference between the M replica signals and the received signal is obtained by the adder 16, the square of the absolute value of each of the M difference values is obtained by the branch metric calculation unit 13, and this is used as the branch metric. Output. This branch tomeric is converted into fanometric by the conversion table 14 based on the above equation (5).

The path search explanatory diagram of FIG. 3 corresponds to the case of binary value, but in the case of M value, M paths are extended, and the stack controller 18 changes from fanometric to path metric. The path having the larger path metric is used as the surviving path, the paths are sequentially extended from the node, the information of the node is stored in the stack memory 19, and the information of the node is stored in the sub-stack memory 20 in the descending order of the path metric. . Therefore, the sub-stack memory 20
By searching for, the path can be extended from the node with the largest path metric.

When the sequential equalization is completed up to the end of the burst, the stack memory 19 is traced back by referring to the sub-stack memory 20.
Will output an equalized output signal. Note that if there are many transmission errors, the process for extending the path from the backtracked node is often repeated, and the capacity of the stack memory 19 is finite, so that equalization may not be possible. In that case, the received signal is output as it is and the burst equalization processing is ended.

FIG. 4 is an explanatory view of the second embodiment of the present invention, in which the reception buffer is omitted and shown, 21 is a channel impulse response calculation section, 22 is a replica generation section, 23 is a branch metric calculation section, Reference numeral 24 is a conversion table, 25 is a serial equalization unit, 26 is an adder, 27 is a selector, 28 is a stack control unit, 29 is a stack memory, and 30 is a sub-stack memory.

In this embodiment, as in the first embodiment described above, the branch metric calculation unit 2 uses the square of the absolute value of the difference between the received signal and the replica signal as the branch metric.
3, the branch metric is converted by the conversion table 24 into the Fanometric based on the equation (5), and the successive equalization unit 25 performs the successive equalization by the stack algorithm.

When a known training sequence is received, the selector 27 adds the training sequence to the propagation path impulse response calculation unit 21 and the replica generation unit 22 to calculate the propagation path impulse response CIR, and thereafter. , The selector 27 adds the information of the past node corresponding to the delay time of the delayed wave from the successive equalization unit 25 to the propagation path impulse response calculation unit 21 and the replica generation unit 22 as well as the received signal from the adder 26. And the replica signal are added to the propagation path impulse response calculation unit 21, and the impulse response C of the propagation path is calculated every time the path is expanded by the equalization processing of one symbol.
Find the IR. As a result, the received signal can be sequentially equalized by following the characteristic change of the propagation path due to fading or the like.

In the diversity method, a method is known in which a plurality of receiving paths are formed and the receiving path having the maximum receiving level is selected, or the phases are combined and combined. The digital adaptive equalizer according to the first or second embodiment described above is provided for a plurality of reception paths of the diversity system. Then, in the reception path in which the C / N of the reception signal is large, the number of times to perform the operation of extending the path again from the node that has returned back is reduced, and therefore the equalization processing becomes the earliest. Therefore, the equalized output signal of the reception path for which the equalization processing is completed first is selected.

Further, the successive equalizers 15 and 25 are shown by the stack algorithm, but may be constructed by the Fano algorithm. In the case of this Fano algorithm, for example, the sequential equalization units 15 and 25 can be configured by the forward branch metric calculation unit, the backward branch metric calculation unit, the conversion table, the path determination control unit, and the buffer memory. .

In the sequential equalization section using this Fano algorithm, a threshold value of a plurality of stages is provided, and the path determination control section determines whether or not the path metric exceeds a certain threshold value. A search is performed, and if the path metric exceeds the threshold for the next symbol as well, a path search by forward movement is performed, and switching is performed so that the threshold also increases as the path metric increases. If the path metric is smaller than the threshold, the path search is performed in the backward direction. If the path metric does not exceed the threshold, the threshold is switched to be smaller. When the path search by the forward movement continues and the equalization ends until the end of the burst, the path determination control unit outputs an equalization output signal according to the path information stored in the buffer memory.

FIG. 5 is a C / N vs. path expansion frequency curve diagram per symbol, showing a simulation result in the case of synchronously detecting and equalizing a received signal of 8-phase PSK modulation. The horizontal axis represents C / N [dB] and the vertical axis represents the number of times. As a simulation condition, the transmission rate is 667 k symbols /
s, as fading, uncorrelated 2-wave Rayleigh, maximum Doppler frequency 10 Hz, 1 symbol delay, sample timing was assumed to be ideal timing, and frequency offset was assumed to be 0.

In the case of the conventional example shown in FIG. 7, the calculation per symbol is 6 as shown by the dotted line MLSE.
It will be four times. On the other hand, according to the embodiment of the present invention,
C / N is about 25 dB, as shown by the successive equalization of the solid line
In the above, a minimum of eight times of calculation is required for one symbol, and in the case of a propagation path with a relatively large C / N, the amount of calculation is drastically reduced compared to the conventional example. You can

FIG. 6 is a C / N vs. bit error rate characteristic curve diagram, showing a simulation result under the above conditions,
The horizontal axis represents C / N [dB] and the vertical axis represents the bit error rate BER. The thin dotted line MLSE and the thick dotted line DFE indicate the bit error rate characteristics of the maximum likelihood sequence estimation type equalizer and the decision feedback type equalizer of the conventional example, and the solid line sequential equalization is the same as the present invention. The bit error rate characteristic by an Example is shown. According to the embodiment of the present invention, the bit error rate B is higher than that of the decision feedback equalizer.
The ER is remarkably improved and is slightly inferior to that of the maximum likelihood sequence estimation type equalizer, and sufficient equalization characteristics can be obtained even in burst communication and the like.

[0042]

As described above, the present invention is a digital adaptive equalizer provided with a sequential equalization unit 5 using a sequential decoding algorithm such as a Fano algorithm or a stack algorithm, and is a maximum likelihood sequence estimation type or the like. There is an advantage that the circuit scale is smaller and the amount of calculation can be reduced as compared with the digitizer. In particular, in the case of a channel having a small BER that can be used for data communication, that is, a channel having a large C / N,
There is an advantage that the equalization processing can be performed with an extremely small calculation amount.

Further, the propagation path impulse response calculation unit 1 calculates the impulse response of the propagation path each time the path is expanded by the equalization processing of one symbol, so that the equalization characteristic is obtained with respect to the characteristic fluctuation of the propagation path due to fading or the like. Can be made to follow, and there is an advantage that the error rate can be improved.

Further, a digital adaptive equalizer having a successive equalization unit 5 is provided corresponding to the diversity reception path, and the equalization output signal of the reception path for which the equalization processing is completed first is selected to obtain the most reception state. It is possible to select an equalized output signal of a good reception path and to speed up the reception process.

[Brief description of drawings]

FIG. 1 is a diagram illustrating the principle of the present invention.

FIG. 2 is an explanatory diagram of the first embodiment of the present invention.

FIG. 3 is an explanatory diagram of a path search in the stack algorithm.

FIG. 4 is an explanatory diagram of a second embodiment of the present invention.

FIG. 5 is a C / N vs. path extension frequency curve diagram per symbol.

FIG. 6 is a C / N-bit error rate characteristic curve diagram.

FIG. 7 is an explanatory diagram of a conventional example.

[Explanation of symbols]

 1 propagation path impulse response calculation unit 2 replica generation unit 3 branch metric calculation unit 4 conversion table 5 successive equalization unit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location H04B 7/005 7741-5K 7/02 Z 4229-5K

Claims (3)

[Claims] 1. A propagation path impulse response calculation unit (1) for calculating a propagation path impulse response by a known training sequence, and equalization according to a propagation path impulse response obtained by the propagation path impulse response calculation unit (1). A replica generator (2) that generates a replica signal from an output signal, and a branch that calculates the square of the absolute value of the difference between the replica signal by the replica generator (2) and the signal received through the propagation path as a branch metric. A metric calculation unit (3), a conversion table (4) for converting the branch metric by the branch metric calculation unit (3) into an integer metric, and a sequential decoding algorithm using the integer metric by the conversion table (4). A digital signal comprising a successive equalization unit (5) for equalizing the received signal.応等 equalizer. 2. The propagation path impulse response calculation unit (1)
Is configured to calculate the impulse response of the propagation path for each path extension in the successive equalization unit (5) based on the difference between the replica signal and the received signal via the propagation path. The digital adaptive equalizer according to claim 1, wherein 3. The digital adaptive equalizer is provided corresponding to a diversity reception path, and a configuration is provided for selecting an equalization output signal of the diversity reception path for which the equalization processing is completed first. 1. The digital adaptive equalizer according to 1.