JP2908697B2 - Adaptive filter and interference elimination method using the adaptive filter. - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an adaptive filter, and more particularly to a digital mobile communication system such as a digital car telephone system, a digital portable telephone system, a digital cordless telephone system, and a digital mobile satellite communication system. The present invention relates to an adaptive filter applied to a code division multiple access system and an interference elimination system using the adaptive filter.

[0002]

2. Description of the Related Art In recent years, research, development, and commercialization of applications of spread spectrum technology to land mobile communication systems such as mobile phones and mobile phones have been performed. In such a code division multiplexing system, error rate characteristics deteriorate due to interference between users due to non-orthogonality of a spreading code assigned to each user, and a signal of a distant mobile device is masked by a signal of a nearby mobile device. Deterioration of error rate characteristics caused by the near-far problem is a major problem.

As a method for solving these problems, an interference elimination method (Mitra, U. and Poor, HV, "Adaptive R") using a linear filter (LMS filter) using a least mean square (LMS) algorithm as an adaptive algorithm.
eceiver Algorithms for Near-Far Resistant CDM
A ", Proc. Of the 3rd IEEE Int. Symp. OnPer
sonal, Indoor and Mobile Radio Commun. Bosto
n, pp.639-644, Oct. 1992.) and a layered neural network (NN) using the back propagation algorithm (BP method) (Aazhang, B., Paris,
B.-P. and Orsak, GC: "Neural Networks fo
r Multiuser Detection in Code-Division Multip
le-Access Communications ", IEEE Trans. Commun.,
vol.40, no.7, pp.1212-1222, July. 1992.).

In the interference elimination system using these adaptive filters, a case where a value obtained by sampling a received wave at a chip rate of a spread code and a case where a peak value of a matched filter output is input to the adaptive filter are considered. . Therefore, the number of input nodes of the necessary adaptive filter is equal to the number of chips of the spreading code in the former case, and equal to the number of users to be demodulated in the latter case.

An adaptive filter has a set of adjustable parameters (filter coefficients) whose parameter values are automatically set based on the estimated statistical properties of the signal being handled. Filter coefficients intends <br/> row scaling of the input signal propagating on individual connections directed from the input node to the output node.

FIG. 4 shows a configuration of a conventional interference elimination system using an adaptive filter which receives a peak value of a matched filter output as an input, and FIG. 5 shows a configuration of an adaptive filter section. Antenna (0
The spread spectrum signal S (t) received at 1), amplified by the RF amplifier (02), and converted to the baseband frequency by the frequency converter (03) is branched and matched for each user. Input to the filter (04). The peak value of the matched filter output (04
1) is an adaptive filter unit (0) prepared by the number of users.
5) are input to the input nodes (051). The input node (051) has a connection with all the output nodes (053), and each output node performs processing for only one different user. The connection between each input node (051) and each output node (053) has a fixed filter coefficient (052) that is not updated except during learning and adaptation, and the output signal of the input node (051) is scaled by the filter coefficient. After that, it is input to the output node (053). The output node (053) performs demodulation of information bits by substituting the sum of input values from all input nodes into a transfer function and comparing the result with a threshold.

The adaptive filter section includes an LMS filter using a least mean square (LMS) algorithm as an adaptive algorithm, and an RLS using a recursive least squares (RLS) algorithm.
A linear filter such as a filter or a non-linear filter such as a hierarchical NN using the BP method can be used. The LMS filter and the RLS filter use a linear function as a transfer function, and the hierarchical NN filter uses a non-linear function such as a sigmoid function as a transfer function.

The adaptive filter section (05) needs to go through a learning / adaptation process to obtain appropriate filter coefficients in order to remove interference of a propagation signal. FIG. 6 shows a control flow of the adaptive filter unit (05). Prior to the learning, each filter coefficient (052) of the adaptive filter unit (05) is set to an appropriate initial value by the adaptive filter initial value generation unit (08) (S ₁ ). In general, there are cases where zero is given as an initial value of a filter coefficient and a case where a random number whose absolute value is smaller than 1 is given. Learning and adaptation process of the adaptive filter section (05) (S ₂ ~S ₆ and S _1), the adaptive filter controller (0
No effect of noise or fading generated in 7)
The training is performed by using the pseudo matched filter output of each user as a learning pattern and using information bits assumed at the time of creating the learning pattern as a teacher signal.

[0009]

The interference cancellation system having a INVENTION Problems to be Solved] above configuration, when the cross-correlation between the signals of each user as asynchronous system changes with time, the need to relearn There is. When dealing with time-series signals, it is very important to reduce the time loss due to relearning.

An object of the present invention is to solve the above-mentioned problems of the prior art and to provide an adaptive filter having improved convergence speed and an interference elimination system using the adaptive filter.

[0011]

The speed of convergence of an adaptive filter in a learning / adaptation process depends on an initial value of a filter coefficient set before learning. Generally, in setting the initial value, the characteristics of the signal to be processed cannot be estimated, and it is difficult to predict the relationship between the individual filter coefficients after the learning has converged. However, in the method in which the peak value of the output of the matched filter is input to the adaptive filter, the value of the filter coefficient after learning is largely dependent on the cross-correlation between the signal of the user assigned to the input node and the signal of the user assigned to the output node. I can guess. The adaptive filter according to the present invention provides each filter coefficient between the input node and the output node as a cross-correlation value between a signal of the user assigned to the input node and a signal of the user assigned to the output node. The convergence speed is improved. The adaptive filter according to the invention
The ruta has the following configuration. Spread spectrum
A plurality of matched filters for inputting a heavy signal, the plurality of alignment
Each of the spread spectrum multiplexed signals obtained from the filter
From each output peak value of the channel signal,
A timing extraction section for extracting each peak timing, before
The spread spectrum obtained from the plurality of matched filters.
Temporarily store each output peak value of each channel signal of multiplexed signal
And each of the channel signals from the memory
A plurality of input nodes for receiving the output peak value and the plurality of input nodes;
All of the force nodes and their associated output nodes
Optimize the filter coefficient for each learning adaptation process
And after optimizing the filter coefficients,
An adaptive filter unit that performs a demodulation process for each channel signal, before
Extracted by the timing extraction unit for each learning adaptation process
Receiving each of the channel signals from each of the peak timings
Calculate the relative deviation of the timing, and based on the relative deviation,
Perform processing for calculating an estimated cross-correlation value between channel signals
The estimated cross-correlation estimator, an adaptive filter initial value generating unit and the front
The estimated cross-correlation value is the same as the previous value in the learning adaptation process
Of each channel signal from the previous SL memory when was
Each output peak value is output to the plurality of inputs of the adaptive filter unit.
Input to the node, and the estimated cross-correlation value is
If it is different from the previous value of
The adaptive filter is used in accordance with the estimated cross-correlation value from the estimation unit.
Set the initial value of the adaptive filter unit to the
Executing the learning adaptation process of the adaptive filter unit.
The adaptive filter
The adaptive filter set in the adaptive filter initial value generator.
The adaptive filter control according to the initial value of the adaptive filter unit.
Perform the learning adaptation process under the control of the control unit
Filter configured as follows . Also according to the invention
The interference cancellation method using an adaptive filter has the following configuration.
ing. Digital using direct spread spectrum modulation
For code division multiple access in mobile radio communications,
In the interference cancellation method using an adaptive filter in the demodulation unit,
The adaptive filter receives a spread spectrum multiplexed signal.
And a plurality of matched filters obtained from the matched filters.
Of each channel signal of the spread spectrum multiplexed signal
From each output peak value, each peak timing of each channel signal
A timing extracting unit for extracting a plurality of matching flags;
Filter of the spread spectrum multiplexed signal obtained from the filter.
A memory for temporarily storing each output peak value of the channel signal;
Each output peak value of each channel signal from the memory is
A plurality of input nodes to receive and all of the plurality of input nodes
And a plurality of output nodes coupled to each other.
While optimizing the filter coefficients for each reaction process,
After optimizing the filter coefficients, the respective channel signals
An adaptive filter unit for performing demodulation processing of the learning adaptation process
Each peak tie extracted by the timing extractor for each
Relative to the reception timing of each channel signal
Calculate the deviation, and based on the relative deviation,
Estimated cross-correlation estimation that performs processing to calculate the estimated cross-correlation value
Unit, an adaptive filter initial value generation unit, and the estimated cross-correlation value
Is the same as the previous value in the learning adaptation process
Each output peak value of each channel signal from the memory
Input to the plurality of input nodes of the adaptive filter unit,
The estimated cross-correlation value differs from the previous value in the learning adaptation process.
When it has become the said from the estimated cross-correlation estimation unit
Based on the estimated cross-correlation value, the adaptive filter initial value generation unit
The initial value of the adaptive filter section is set, and the adaptive filter section is set.
An adaptive filter for executing the learning adaptation process of the filter unit.
And an adaptive filter section , wherein the adaptive filter section includes the adaptive filter section.
The initial value of the adaptive filter unit set in the filter initial value generation unit
According to the period value, the control by the adaptive filter control unit is also performed.
And configured to perform the learning adaptation process.
An interference removal method using an adaptive filter.

[0012]

Principle of the Invention The i-th input node (i
The filter coefficient between the matched filter output value of the i-th user and the j-th output node (outputting the demodulated information bits of the j-th user) is h _ji, and the signal of the i-th user is when the cross-correlation between j-th user signal and a _ji, given by the following equation initial value h _jio filter coefficient (1).

[0013]

## _EQU1 ## h _jio = -a _ji h _ii (1) h _ii : an arbitrary constant

[0014]

According to the present invention, each filter coefficient between an input node and an output node of an adaptive filter is given as a cross-correlation value between a signal of a user assigned to an input node and a signal of a user assigned to an output node. Thus, the convergence speed of the adaptive filter can be improved, the time loss due to relearning can be reduced, and the transmission efficiency can be improved.

[0015]

1 shows a configuration of an interference canceling system using an adaptive filter according to the present invention, and FIG. 2 shows a control flow of the adaptive filter. S ₁ , S ₂ , S ₃ , and S ₆ are the same as the corresponding number steps in the operation flow of the conventional example. The output peak value (0) obtained by the matched filter (04) prepared for each user
41) is stored in the temporary memory (10) in order to make the timing at the time of input to the adaptive filter ( _S4a ). or,
Output peak value (041) is also input to the timing extraction unit (06) of each user, peak timing for each user is detected (S _7), are input to the cross correlation estimation unit (09). The cross-correlation estimating section (09) estimates the cross-correlation value (09) between the users whose spreading codes have been shifted by the relative shift of the peak timing (061).
1) is calculated (S ₂ ).

If the calculated estimated cross-correlation value (091) is the same as the cross-correlation value at the time of the previous learning, the adaptive filter control unit (07) informs the memory unit (10) of the output timing, and inputs each input of the adaptive filter. Matched filter output values (041) corresponding to the respective nodes are simultaneously input. On the other hand, if the calculated estimated cross-correlation value (091) is different from the cross-correlation value at the time of the previous learning (S _9), the newly calculated cross-correlation values in S ₀ (091) is an adaptive filter initial value generator ( 08), and gives an initial value of the filter coefficient based on the equation (1) (S ₁ ). Adaptive filter control unit (07) activates a learning and adaptation process, to train the adaptive filter using pseudo-learning pattern and the teacher signal generated by the adaptive filter control unit (07) (S _2).

FIG. 3 shows an example of the convergence characteristics in the learning / adaptation process of the method of the present invention. Balanced Gold with 31 chip length
It shows the error rate of the desired user (user 1) with respect to the number of times of learning when there are ten users to which codes are assigned. The LMS filter is used in the adaptive filter unit, and learning is performed with all user levels being 1 in an environment where no noise exists. The initial value of the connection weight in the conventional method is given by a random number between -0.1 and 0.1.
In the evaluation of the error rate, the level of the user 1 was set to 1 and the levels of other interfering users were set to 10, and the signal of the user 1 having a signal-to-noise ratio of 6 dB was demodulated.

As shown in FIG. 3 , by using the method for setting the initial value of the filter coefficient according to the present invention shown by the solid line, learning is started from a state having better error rate characteristics as compared with the case of using the conventional setting method. It can be seen that the convergence is achieved with about half the number of learnings. It is clear from the simulation results that the effect of improving the convergence characteristics according to the present invention can be obtained in an actual digital mobile communication system.

[0019]

As described above, according to the present invention,
By using a method in which each filter coefficient between an input node and an output node of the adaptive filter unit is given as a cross-correlation value between a user signal assigned to the input node and a user signal assigned to the output node, Can be improved, the time loss due to relearning can be reduced, and the transmission efficiency can be improved.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of an interference cancellation method for a spread spectrum multiplexed signal according to the present invention.

FIG. 2 is a flowchart showing an example of an operation algorithm of an adaptive filter unit used in the method of the present invention.

FIG. 3 is a diagram showing the convergence characteristics of the interference elimination method of the spread spectrum multiplexed signal according to the method of the present invention in comparison with the characteristics of the interference elimination method of the spread spectrum multiplexed signal using a conventional method in which the initial value of the filter coefficient is given by a random number. FIG.

FIG. 4 is a block diagram illustrating a configuration example of an interference cancellation method for a spread spectrum multiplexed signal using a conventional initial value setting method.

FIG. 5 is a block diagram illustrating a configuration example of an adaptive filter unit used in the conventional system of FIG. 5;

FIG. 6 is a flowchart illustrating an example of an operation algorithm of an adaptive filter unit used in a conventional method.

[Description of Signs] 1 Antenna 2 RF amplification unit 3 Frequency conversion unit 4 Matching filter 41 Matching filter output peak value 5 Adaptive filter unit 6 Timing extraction unit 61 Peak timing 7 Adaptive filter control unit 71 Output timing 8 Adaptive filter initial value generation unit 81 Initial Filter Coefficient 9 Cross-Correlation Estimator 91 Estimated Cross-Correlation Value 10 Memory 11 Information Bit

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-2-218210 (JP, A) JP-A-3-32232 (JP, A) JP-A-3-114143 (JP, A) Aazhang, B. Paris, B .; -P. and Orsak, G.S. C. : "Neural Networks for Multiuser Detection in Code-Division Multiple-Access Communications", IEEE Trans. Commun. , Vol. 40, no. 7. pp. 1212-1222, July. 1992. (58) Field surveyed (Int. Cl. ⁶ , DB name) H04J 13/04 H03H 17/00 H03H 21/00

Claims (2)

(57) [Claims] 1. A plurality of spread spectrum multiplexed signal input units.
And the spread spectrum obtained from the plurality of matched filters
From each output peak value of each channel signal of the multiplex signal,
Timing to extract each peak timing of channel signal
An extracting unit, and the spectral expansion obtained from the plurality of matched filters.
Temporarily records each output peak value of each channel signal of the scattered multiplex signal.
Memory and the output peak value of each channel signal from the memory.
A plurality of input nodes to receive and all of the plurality of input nodes
And a plurality of output nodes coupled to each other.
While optimizing the filter coefficients for each reaction process,
After optimizing the filter coefficients, the respective channel signals
An adaptive filter unit for performing a demodulation process of, and the timing extraction unit extracts the
Receiving the respective channel signals from the respective peak timings.
Calculate the relative deviation of the communication timing, and based on the relative deviation,
Perform processing to calculate the estimated cross-correlation value between each channel signal.
An estimated cross-correlation estimator, an adaptive filter initial value generator, and the estimated cross-correlation value
If the value is the same as the previous value in the learning adaptation process,
The output peak value of each channel signal from the memory
Input to the plurality of input nodes of the
The constant cross-correlation value differs from the previous value in the learning adaptation process
The estimated phase from the estimated cross-correlation estimator
The cross-correlation value allows the adaptive filter initial value generator to
Setting the initial value of the adaptive filter section,
Filter control for executing the learning adaptation process of the section
And the adaptive filter unit includes an adaptive filter initial value generation unit.
According to the set initial value of the adaptive filter unit,
Under the control of the adaptive filter control unit, the learning adaptive
An adaptive filter configured to perform the process . 2. An interference cancellation method using an adaptive filter in a demodulator for code division multiple access in digital mobile radio communication using direct spread spectrum modulation, wherein the adaptive filter receives a spread spectrum multiplexed signal. Multiple matched filters
And the spread spectrum obtained from the plurality of matched filters
From each output peak value of each channel signal of the multiplex signal,
Timing to extract each peak timing of channel signal
An extracting unit, and the spectral expansion obtained from the plurality of matched filters.
Temporarily records each output peak value of each channel signal of the scattered multiplex signal.
Memory and the output peak value of each channel signal from the memory.
A plurality of input nodes to receive and all of the plurality of input nodes
And a plurality of output nodes coupled to each other.
While optimizing the filter coefficients for each reaction process,
After optimizing the filter coefficients, the respective channel signals
An adaptive filter unit for performing a demodulation process of, and the timing extraction unit extracts the
Receiving the respective channel signals from the respective peak timings.
Calculate the relative deviation of the communication timing, and based on the relative deviation,
Perform processing to calculate the estimated cross-correlation value between each channel signal.
An estimated cross-correlation estimator, an adaptive filter initial value generator, and the estimated cross-correlation value
If the value is the same as the previous value in the learning adaptation process,
The output peak value of each channel signal from the memory
Input to the plurality of input nodes of the
The constant cross-correlation value differs from the previous value in the learning adaptation process
The estimated phase from the estimated cross-correlation estimator
The cross-correlation value allows the adaptive filter initial value generator to
Setting the initial value of the adaptive filter section,
Filter control for executing the learning adaptation process of the section
And the adaptive filter unit includes an adaptive filter initial value generation unit.
According to the set initial value of the adaptive filter unit,
Under the control of the adaptive filter control unit, the learning adaptive
An interference cancellation method using an adaptive filter, characterized in that the method is configured to execute a process .