JPH0310253B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a correlation value calculation method in a reception automatic equalizer of a modulation/demodulation device.

(Prior art) Generally, MS algorithm (Mean Square
The Nth tap coefficient of the complex transversal automatic equalizer according to the algorithm is given by the following first and second equations.

C _N =C _N ^-1 -A (Y _N ^-1 ×E _Y ^-1 +X _N ^-1 ×E _X ^-1 ) ...(1) D _N =D _N ^-1 -A (X _N ^-1 ×E _Y ^-1 +Y _N ^-1 ×E _X ^-1 ) ...(2) Here, C _N and D _N are N on the real axis and imaginary axis, respectively
The N-th tap coefficients, C _N ^-1 and D _N ^-1 1 are the N-th tap coefficients on the real and imaginary axes of one sample past, respectively, and X _N ^-1 and Y _N ^-1 are the actual actual ones of one sample past, respectively. This is the Nth delay tap data on the axis and imaginary axis.
Further, E _X ^-1 and Y _Y ^-1 are the errors between the output value of the automatic equalizer one sample past and the ideal value based on the result of the likelihood judgment, respectively, projected onto the real axis and the imaginary axis. Further, A is a setting constant, which is a tap coefficient replacement gain. That is, if A is set to a large value, the tap coefficients are replaced at high speed according to the first and second equations, and when A is set to a small value, the replacement is performed slowly and minutely. Normally, A is set to a large value when the automatic equalizer initially starts up, and is set to a small value when it converges on time.

The second term on the right side of each of the first and second equations is a correlation value calculation part. In the steady convergence state of the automatic equalizer, E _X ^-1 and E _Y ^-1 become small values, and as mentioned above, A also becomes a small value, so the correlation value calculation part becomes a very small value. A specific example of these calculations is shown in FIG. 10 is a multiplier, which has M bits Y _N ^-1 , X _N ^-1 and M bits E _Y ^-1 , E _X ^-1 , A.
Multiply in order. 11 is an accumulator that stores the output from the multiplier 10 and sequentially adds or subtracts the output while outputting it. The procedure of this operation is shown in FIG.

(Problem to be Solved by the Invention) However, since the multiplication and accumulation here are performed in two's complement representation, if the correlation value calculation result becomes a very small value as described above, the two's catch - LSB
As can be seen from Figure 7, which shows the characteristics, -1/2LSB
This has the disadvantage that an offset of 100% is generated, which causes arithmetic errors and deteriorates the convergence of the automatic equalizer.

(Means for Solving the Problems) The present invention was devised in view of these problems, and by adding a constant to the accumulator in advance at the beginning of correlation value calculation, an offset of 1/2 LSB is added to the calculation result. It is an object of the present invention to provide a correlation value calculation method that can suppress the increase in convergence error when the value A corresponding to the tap coefficient replacement gain in the MS algorithm is set to a small value.

(Operation) If the value for the tap coefficient replacement gain in the MS algorithm is A, then the accumulator has a constant K=
2 ^-(M+1) /A is added and inputted, and as a result of the calculation, 2 ^-(M+1) = 1/2LSB is added to the conventional product-sum output, so A is set to a small value. Since it can be set to 1/2 LSB offset, S/N can be reduced.
Suppresses the occurrence of convergence errors due to ratio deterioration.

(Embodiment) FIG. 1 is a configuration diagram showing an embodiment of the present invention. The difference between the conventional configuration in FIG. 5 and the configuration of this embodiment is that a constant K is added and input to the accumulator 11. This K is a constant expressed as 2- ^(M+1) /A. Here, M is the bit length for the binary number at the input/output points of the multiplier 10 and the accumulator 11. Therefore, the multiplier 10 has a maximum length of 2M bits in internal calculations, and outputs an output of M bits by rounding down. The procedure of this operation is shown in FIG. The difference between this diagram and the conventional method procedure, that is, Figure 6, is that in No. 1 and No. 5, the conventional method clears the accumulator, but instead adds a constant K to the accumulator. I am typing. The final product-sum output is as follows.

A(Y _N ^-1 ×E _Y ^-1 +X _N ^-1 ×E _X ^-1 +K) = A(Y _N ^-1 ×E _Y ^-1
+X _N ^-1 ×E _X ^-1 ₎ +AK = A ⁽ _{Y N} ^-1 ×E _Y ^-1 +X ^{N -1 × E +1)} /A) A(X _N ^-1 ×E _Y ^-1 −Y _N ^-1 ×E _X ^-1 +K) = A(X _N ^-1 ×E _Y ^-1
−Y _N ^-1 ×E _X ^-1 ) ⁺ AK = A ⁽ _{X N} ^-1 ×E _Y ^-1 −Y _N ^{-1 ×} E ^(M+1) /A) As a result, the conventional product-sum output is 2 ^{- (M+1)} = 1/
2LSB will be added. Therefore, as can be seen from FIG. 3 showing the 2's complement LSB characteristic of this embodiment, the -1/2LSB offset seen in FIG. 7 has been eliminated.

If A is a small value (A≪1), as mentioned above, the correlation value calculation result will be a very small value and will be 1/
The S/N ratio for 2LSB deteriorates. However, in this embodiment, K is inversely proportional to a large value, sufficient accuracy is obtained for M bits, and 1/2LSB correction is performed accurately, so that the S/N ratio does not deteriorate.

As a result, as can be seen from Fig. 4, which shows the setting constant A - convergence error characteristics of the automatic equalizer, even if A is a small value (A≪1) compared to the conventional method, 1/
It is possible to suppress the occurrence of convergence errors due to the deterioration of the 2LSB offset S/N ratio.

(Effects of the Invention) As described above, according to the present invention, by adding a constant to the accumulator in advance at the beginning of the correlation value calculation, the 1/2 LSB offset is removed from the calculation result, and the MS algorithm It is possible to suppress an increase in convergence error when the value A for tap coefficient replacement profit is set to a small value. Along with this, since A can be set to a small value, it is possible to reduce the influence of white noise, phase jitter, impulse noise, etc. from the transmission path to which the automatic equalizer should not originally respond. Furthermore, compared to the conventional method, instead of performing a clearing operation in the initial operation of the accumulator, only the constant K is added and input, so there is no need to add or change the number of processing steps. There are also advantages.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram showing one embodiment of the present invention, and FIG.
The figure is an explanatory diagram showing the calculation procedure of Figure 1, Figure 3 is a diagram showing the two's complement-LSB characteristic of this embodiment, and Figure 4 is the setting constant A of this embodiment - convergence error of automatic equalizer Figure 5 is a configuration diagram showing a specific example of conventional correlation value calculation, Figure 6 is an explanatory diagram showing the calculation procedure of Figure 5, and Figure 7 is two's complement-LSB characteristic of the conventional example. FIG. 10...multiplier, 11...accumulator.

Claims (1)

[Claims] 1. An automatic equalizer using the MS algorithm, which consists of a multiplier whose input/output is M bits (M is an integer of 2 or more) and an accumulator that cumulatively adds the outputs of the multiplier, and which calculates correlation values in two's complement representation. In the above MS
Assuming that the value for the tap coefficient replacement gain in the algorithm is A, a constant K=
A correlation value calculation method characterized in that 2 ^-(M+1) /A is added and input.