JPH06188785A - Automatic equalizer - Google Patents

Abstract

PURPOSE:To decrease the update arithmetic quantity of the tap coefficient of a transversal filter part and to suppress characteristic deterioration even if level variation due to a noise is caused in a tracking period. CONSTITUTION:A receiver which receives a burst signal consisting of a training period and the tracking period is provided with the transversal filter part 1 and an updating process part 3; and the transversal filter part 1 equalizes a received signal and the updating process part 3 updates the tap coefficient of the transversal filter part 1. The updating process part 3 updates the tap coefficient in the training period according to recurrent least square algorithm 50, and also updates the tap coefficient in the tracking period according to corrected recurrent least square algorithm 51 applied with an error covariance matrix at the end of the training period as an error covariance matrix regarding tap input throughout the period.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic equalizer for equalizing a received signal according to a digital modulation / demodulation method, and more particularly to a mobile communication receiver in which propagation path fluctuation is applied when the moving speed is about walking speed. It is suitable for application to the automatic equalizer in.

[0002]

2. Description of the Related Art Conventionally, as an automatic equalizer applied to a mobile communication receiver, there has been a decision feedback type shown in FIG.

In FIG. 2, a received signal converted into a digital signal has a data decision output and a transversal filter unit 1 for equalization having a front tap and a rear tap.
(Tap input vector) u (n). In the transversal filter unit 1, the tap coefficient (tap coefficient vector) h (n-
1) is also given, and the transversal filter unit 1
Is a tap coefficient h (n− for tap input u (n).
The equalized output (scalar amount) z (n) is obtained by filtering using 1) and is given to the data determination unit 2. The data determination unit 2 determines and outputs the data value (scalar amount) Y (n) of the symbol period based on the equalized output sequence z (n), and feeds it back to the rear tap of the transversal filter unit 1. To do.

The update processing section 3 comprises an error estimation section 4 and a tap coefficient update section 5. The above-described equalized output z (n) and the reference signal d (n) are given to the error estimation unit 4, and the error value (scalar amount) η included in the equalized output z (n). (N) is estimated and given to the tap coefficient updating unit 5. The tap coefficient updating unit 5 receives the given estimation error η
(N) and the information held internally, the tap coefficient h
(N) is determined and given to the transversal filter unit 1.

FIG. 3 shows the structure of a burst signal transmitted and received between a transmitter and a receiver for mobile communication. The above-described tap coefficient h (n) reflects the characteristics of the transmission path, and in order to speed up the initial pull-in of this tap coefficient h (n), a known training data sequence is introduced into the transmission symbol in mobile communication. There is something. That is, as shown in FIG. 3, one transmission burst is composed of a training period in which known training data is inserted in the receiving side and a tracking period in which original transmitting data unknown in the receiving side is inserted. The training period is provided at the head side of the transmission burst so as to be effective for initial setting of the tap coefficient. Then, as the reference signal d (n) for estimating the error included in the equalized output z (n), the symbol value of known training data is used in the training period, and the data determination output Y ( n) is used.

By the way, the transversal filter unit 1
Least Mean Squares (LMS) algorithm, Recursive Least Squares (RLS) algorithm, and the like are well known as update algorithms of tap coefficients of S. Heikin, Takebe.
Translated by Mikio, "Introduction to Adaptive Filters", published September 10, 1987,
Hyundai Engineering Co.). The RLS algorithm has a faster convergence speed of the mean square value of the estimation error than other algorithms such as the LMS algorithm, and theoretically does not have a wrong adjustment of the tap coefficient vector, so that training for initial pull-in of the tap coefficient is performed. It has excellent characteristics that it requires few symbols and can follow the fluctuation of the propagation path. Therefore, application as an updating algorithm of the tap coefficient of the transversal filter unit 1 in the receiver of the mobile communication system is being studied.

FIG. 4 shows the processing of such an RLS algorithm. As shown in FIG. 4, the RLS algorithm uses a gain vector k (n), an error covariance matrix P (n) of a tap input vector u (n), and an equalized output z.
(N), estimation error η (n), tap coefficient vector h
The calculation according to the following equations (1) to (5) for obtaining (n) is repeated (steps 600 to 604).

[0008] ^{k (n) = λ -1 P} (n-1) u (n) ÷ (1 + λ -1 u T (n) P (n-1) u (n)) ... (1) P (n) ^{= λ -1 P (n-1} ) -λ -1 k (n) u T (n) P (n-1) ... (2) z (n) = u T (n) h (n-1) ... (3) η (n) = d (n) -z (n) (4) h (n) = h (n-1) + k (n) η (n) (5) where λ is the past Is a forgetting coefficient that defines how much the information of (1) is reflected in the processing of the current time n and has a value of 1 or less. Further, the gain vector k (n) is information defining how much the information of the current input vector u (n) is reflected. "T" attached to the right shoulder of the vector notation code indicates that the vector is a transposed vector.

Among the above-mentioned operations performed by the RLS algorithm, the transversal filter unit 1 performs the operation of the equation (3), and the others are performed by the update processing unit 3. Further, in the update processing unit 3, the error estimation unit 4
The processing of equation (4) is performed, and the tap coefficient updating unit 5 performs equation (1),
The calculation is performed using the equations (2) and (5).

[0010]

The RLS algorithm has various advantages as described above, but as is clear from the above arithmetic expressions (particularly the expressions (1) and (2)), There is a problem that vector calculation and matrix calculation for the tap coefficient update calculation are complicated and the amount of calculation processing is very large. As a result, even if the update processing unit is configured by hardware or the update processing unit is configured by software, one calculation cycle (repetition cycle in FIG. 4) becomes long and real-time processing becomes difficult. I was there.

Further, as is clear from the above-mentioned arithmetic expression, the RLS algorithm updates all information one by one. Therefore, when the transmission line characteristic is noisy, the tap coefficient converged during the training period is used. There is a problem in that (estimation information of transmission path characteristics) is also adapted to level fluctuations due to noise generated during the tracking period, which is rather a factor of characteristic deterioration.

The present invention has been made in consideration of the above points, and it is possible to reduce the update calculation amount of the tap coefficient of the transversal filter unit, and suppress the characteristic deterioration even if the level fluctuation due to noise occurs in the tracking period. The present invention is intended to provide an automatic equalizer capable of performing the above.

[0013]

In order to solve such a problem, according to the present invention, a receiver for receiving a burst signal consisting of a training period and a tracking period is provided, and the received signal is equalized. In an automatic equalizer including a transversal filter unit and an update processing unit that updates the tap coefficient of the transversal filter unit, the update processing unit updates the tap coefficient according to a recursive least squares algorithm for a training period. However, regarding the tracking period, the error covariance matrix at the end of the training period is used as the error covariance matrix regarding the tap input, and the tap coefficient is updated according to the modified recursive least squares algorithm applied during that period. To do.

[0014]

In the present invention, the update processing unit for updating the tap coefficient of the transversal filter unit updates the tap coefficient according to the recursive least squares algorithm during the training period. This accelerates the convergence to the pull-in state that sufficiently reflects the transmission line characteristics.

Also, the update processing unit, as for the tracking period, as an error covariance matrix related to tap input,
The error covariance matrix at the end of the training period is updated with tap coefficients according to the modified recursive least squares algorithm applied during that period. As described above, since the error covariance matrix having a fixed value is applied in the tracking period, it is possible to reduce various calculation amounts, and it is possible to suppress the characteristic deterioration even if the level variation due to noise occurs in the tracking period.

Although the error covariance matrix having a fixed value is applied in the tracking period, the error covariance matrix sufficiently converged in accordance with the transmission line characteristics is applied in the training period. The tap coefficient can be updated appropriately, and the application of a fixed value does not cause a problem.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment in which the present invention is applied to an automatic equalizer in a receiver of a mobile communication system will be described in detail below with reference to the drawings. Here, FIG. 1 is a block diagram showing the configuration of this embodiment, and FIG. 5 is an explanatory diagram showing a coefficient updating algorithm adopted in the tracking period in this embodiment. In FIG. 1, the same parts as those in FIG. 2 are designated by the same reference numerals.

In FIG. 1, the receiver to which this automatic equalizer is applied is equipped with two antennas (not shown), and the signal received by one antenna is sampled by the sampler 6 in a cycle shorter than one symbol period (not shown). The sampled signal is sampled in a 1/2 symbol period), the received signal from the other antenna is sampled in the same cycle by the sampler 7, and each digital signal sequence is input to the common transversal filter unit 1. The data determination output output from the data determination unit 2 is also fed back and input to the transversal filter unit 1.

The sampling period selected by the samplers 6 and 7 is selected to be shorter than one symbol period in consideration of diversity input as well.
This is because an attempt was made to improve the precision of the equalized output from the.

It can be said that the transversal filter unit 1 of this embodiment is functionally provided with three transversal filters and a configuration for combining the outputs thereof.

Delay element group 101 to 1/2 symbol period
10x and tap coefficient multiplier groups 111 to 11 (x + 1)
And a common adder (summer) 40, the sampler 6
It has a transversal filter for filtering the digital signal sequence from the. In addition, a transversal for filtering the digital signal sequence from the sampler 7, which includes the delay element groups 201 to 20x in the 1/2 symbol period, the tap coefficient multiplier groups 211 to 21 (x + 1), and the common summation device 40. It has a filter. Further, the delay element groups 301 to 30y for one symbol period, the tap coefficient multiplier groups 311 to 31y, and the common summing device 40
And a transversal filter for filtering the data output sequence from the data determination unit 2. The first and second transversal filters are so-called front tap transversal filters,
In the case of this embodiment, since diversity processing is executed through filtering processing, two front-tap transversal filters are required as described above. The third transversal filter is a so-called rear tap transversal filter.

In the transversal filter unit 1,
The equalized output sequence obtained by the summer 40 is provided to the data determination unit 2 and the update processing unit 3.

The data determination unit 2 determines the data value in each symbol period based on the equalized output sequence, outputs it to the outside, and supplies it to the update processing unit 3.

Whether the update processing unit 3 of this embodiment is composed of hardware or software,
Functionally, the update processing unit 50 for the training period,
The update processing unit 51 for the tracking period. That is, the update processing unit 3 is provided with a mode signal indicating the training period or the tracking period, and when the mode signal indicates the training period, the update processing unit 50 executes the tap coefficient update process, When the signal indicates the tracking period, the update processing unit 5
1 executes the tap coefficient update process.

The update processing section 50 for the training period is
As in the conventional case, the update process is executed according to the RLS algorithm (see FIG. 4). Therefore, description of the update processing executed by the update processing unit 50 is omitted.

The update processing unit 51 for the tracking period is
The tap coefficient updating process is executed according to a coefficient updating algorithm (hereinafter, referred to as a modified RLS algorithm) in which the RLS algorithm is modified.

FIG. 5 shows the processing of such a modified RLS algorithm. As shown in FIG. 5, the modification R
Also in the LS algorithm, the gain vector k (n), the error covariance matrix P (n) of the tap input vector u (n), the equalization output z (n), the estimation error η (n), the tap coefficient vector h (n). Is repeated (step 7).
00-704).

However, the following equations (6) to (10)
As is clear from the equation, the modified RLS algorithm has a gain vector k (n) and a tap input vector u.
The arithmetic expression of the error covariance matrix P (n) of (n) is different from that of the RLS algorithm described above.

K (n) = P ₀ u (n) / ‖u (n) ‖ ² (6) P (n) = P (n-1) = P ₀ (7) z (n) = u ^T (n) h (n-1) (8) η (n) = d (n) -z (n) (9) h (n) = h (n-1) + k (n) η (n ) (10) where P ₀ is the error covariance matrix of the input vector u (n) at the end of the training period,
As shown in equation (7), it is applied as the error covariance matrix P (n) over the entire tracking period. This is the main feature of the modified RLS algorithm. Since the reference signal d (n) is in the tracking period, the data determination output Y (n) is used as it is.
Also, ‖u (n) ‖ ² in the equation (6) for obtaining the gain vector k (n) is the square of the norm of the tap input vector u (n) and is used for normalization.

Error covariance matrix P of tracking period
As (n), the reason why the error covariance matrix P ₀ at the time when the training period ends is applied is based on the following concept. Since the training period has passed, the error covariance matrix P ₀ is also in a converged state in which the transmission path characteristics are sufficiently reflected, and it is considered that there is no problem even if this is applied to the tracking period. Further, even if the transmission path characteristic fluctuates due to noise or the like, since the error covariance matrix P ₀ having a fixed value is applied, the adverse effect due to the fluctuation can be eliminated. Further, since the error covariance matrix P ₀ having a fixed value is applied, not only the information using the value of the error covariance matrix, that is, the calculation of the gain vector (see the equation (6)) can be simplified, but also the error covariance matrix Naturally, the calculation for obtaining the value of the dispersion matrix can be made unnecessary.

When a modified coefficient updating algorithm for the training period is applied to the tracking period, a coefficient updating algorithm other than the RLS algorithm may be applied as the training period coefficient updating algorithm. Since it is premised that the information is reflected in the tracking period, it is required that the transmission line characteristics be sufficiently drawn in at the end of the training period. Therefore, in the above embodiment, the RLS algorithm is applied for the training period.

Therefore, according to the above embodiment, the amount of update calculation of the tap coefficient of the transversal filter unit can be reduced, and the real-time processing can be facilitated.
In addition, it is possible to suppress the characteristic deterioration even if the level changes due to noise during the tracking period.

For example, assuming that the number of taps of the transversal filter unit 1 is M, the amount of calculation (number of multiplications) required to update the tap coefficient once is 3M ² + 3M in the RLS algorithm. The modified RLS algorithm adopted for the tracking period in this embodiment is M ² + 2M times. Therefore, when M = 12, the former requires 468 times and the latter requires 168 times, and when M = 31, the former requires 2976 times.
The latter requires 1023 multiplications, and it can be seen that the modified RLS algorithm requires about 1/3 the amount of calculation as compared with the RLS algorithm.

FIG. 6 shows equalization for a system in which the tap coefficient is fixed during the tracking period, a system for updating the coefficient according to the RLS algorithm during the tracking period, and a system for performing coefficient updating according to the modified RLS algorithm during the tracking period (see the embodiment). The characteristics are shown, and in any case, the RLS algorithm is applied during the training period.

In FIG. 6, the vertical axis represents the bit error rate (BE
R), the horizontal axis is the ratio (Ed / No) of the transmission power and the noise density per bit of the signal. Further, the operating condition of FIG. 6 is that the received signal of 16QAM to which two waves of Ray re-fading are added is input to the decision feedback type automatic equalizer having the configuration shown in FIG. 2 except for the updating method. .

As is apparent from FIG. 6, the modified RLS
The algorithm has improved characteristics when the transmission power / noise density ratio Ed / No is small and the conditions are not good. Even when the transmission power / noise density ratio Ed / No is large, the performance is comparable when the RLS algorithm is adopted. Is shown. In addition, compared with the method in which the tap coefficient is not updated, excellent characteristics are exhibited regardless of the value of the transmission power / noise density ratio Ed / No, and it can be seen that there is a great effect in improving the bit error rate.

Such equalization characteristics can be similarly applied to the automatic equalizer of the embodiment shown in FIG.

In the above embodiment, the present invention is applied to a decision feedback type automatic equalizer, but it is applied to a linear automatic equalizer having a transversal filter section without rear taps. May be.

In the above embodiment, the transversal filter unit processes two or more diversity inputs. However, the present invention is also applied to the case where the received signal input to the transversal filter unit is one sequence. can do.

[0040]

As described above, according to the present invention, the update processing unit updates the tap coefficient according to the recursive least squares algorithm for the training period, and the error covariance matrix regarding the tap input for the tracking period. The error covariance matrix at the end of the training period is updated with the tap coefficient according to the modified recursive least squares algorithm applied during that period. Therefore, the amount of calculation of the tap coefficient can be reduced as compared with the conventional method, and real-time processing can be performed. It is possible to expect a great effect in the realization of the above, and it is possible to realize an equalization characteristic superior to the conventional one even when the transmission path condition is bad, and it is possible to improve the reliability of information transmission.

[Brief description of drawings]

FIG. 1 is a block diagram showing an automatic equalizer according to an embodiment.

FIG. 2 is a block diagram showing a conventional automatic equalizer.

FIG. 3 is an explanatory diagram showing a configuration of a burst signal that arrives at a receiver.

[Fig. 4] RL adopted by a conventional automatic equalizer for the entire period
It is explanatory drawing which shows S algorithm.

FIG. 5 is an explanatory diagram showing a modified RLS algorithm adopted in a tracking period by the automatic equalizer of the embodiment.

FIG. 6 is an explanatory diagram showing the relationship between various tap coefficient update methods and equalization characteristics.

[Explanation of symbols]

1 ... Transversal filter section, 2 ... Data determination section,
3 ... Update processing unit, 50 ... Training period update processing unit (RLS algorithm adopted), 51 ... Tracking period update processing unit (modified RLS algorithm adopted).

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Office reference number FI technical display location H03H 21/00 7037-5J

Claims (1)

[Claims] 1. A transversal filter unit, which is provided in a receiver for receiving a burst signal composed of a training period and a tracking period, for equalizing a received signal and a tap coefficient of the transversal filter unit is updated. In the automatic equalizer having an update processing unit for updating the tap coefficient for the training period according to the recursive least squares algorithm, and for the tracking period as an error covariance matrix for tap input. An automatic equalizer characterized in that the error covariance matrix at the end of the training period is updated with tap coefficients according to the modified recursive least squares algorithm applied during that period.