TWI473077B - Blind source separation system - Google Patents

Description

Blind signal separation system

The present invention relates to a system and, more particularly, to a blind signal separation system.

The blind signal separation is used to analyze the mixed signal source to obtain the original signal source. For example, in the speech recognition system, the mixed signal source collected by the voice recognition system includes different original signal sources, and the conventional technology usually uses software. The algorithm separates the mixed signal sources to obtain the original signal source. However, this method requires a large number of operations in the process of training coefficients, and delays the time to obtain the original signal source.

It can be seen that the above existing methods obviously have inconveniences and defects, and need to be improved. In order to solve the above problems, the relevant fields have not tried their best to find a solution, but for a long time, no suitable solution has been developed. Therefore, how to improve the way that the conventional technology adopts the software algorithm requires a large number of operations, and delaying the acquisition of the original signal source is one of the current important research and development topics, and has become an urgent need for improvement in related fields.

It is an object of the present invention to provide a blind signal separation system for improving the manner in which the prior art uses a software algorithm, requiring a large number of operations and delaying the acquisition of the original signal source.

To achieve the above object, a technical aspect of the present invention relates to a blind signal separation system including a plurality of adaptive filters, a plurality of adders, and a plurality of message maximizing elements. In the configuration, the foregoing adaptive filters are arranged in a matrix, and each of the foregoing adders is disposed in one of the columns of the matrix, and each of the foregoing message maximizing elements is disposed in one of the columns of the matrix.

In operation, the first row of each of the foregoing adaptive filters is configured to receive the first input source, and process the first input source to output a first intermediate signal, and the foregoing adaptive filters are The second row of each column is for receiving the second input source and processing the second input source to output a second intermediate signal. Each of the adders is configured to receive the first and second intermediate signals outputted by the adaptive filters arranged in the corresponding columns, and add the corresponding first and second input signals to output A plurality of output signals. The message maximizing components are respectively configured to receive corresponding output signals output by the adders arranged in the corresponding columns, and perform a lookup table according to the corresponding output signals to output a plurality of message maximization signals.

Furthermore, the first row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the first input source and the corresponding message maximization signal to output the first An intermediate signal, wherein the second row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the second input source and the corresponding message maximization signal to output the foregoing Two intermediate signals.

According to an embodiment of the present invention, each of the foregoing adaptive filters The third row is configured to receive the third input source and the corresponding message maximization signal, and process the third input source and the corresponding message maximization signal to output a third intermediate signal.

According to another embodiment of the present invention, the foregoing adaptive filters are arranged in the first column for receiving the first input source, and the adaptive filters are arranged in the second column for receiving the second input source. And the foregoing adaptive filters are arranged in the third column for receiving the third input source and are omitted.

According to still another embodiment of the present invention, each of the foregoing adaptive filters conforms to the following formula: W ( n +1)= W (n)+ μe ( n ) X ( n ), e( n )=d (n)- X ^T ( n ) W ( n ), each of the foregoing adaptive filters includes a plurality of arithmetic units, a plurality of second adders, and a subtractor, and further, each of the foregoing operational units One includes a first multiplier, a first adder, and a second multiplier. The first adder is electrically coupled to the first multiplier, and the second multiplier is electrically coupled to the first adder. The second adders are electrically connected in series, and each of the second adders is electrically coupled to a second multiplier of each of the corresponding arithmetic units.

Operationally, in the foregoing operation unit, the first multiplier is used to multiply μe ( n ) and X ( n ), and the first adder is used to phase W (n) and μe ( n ) X ( n ) Plus, and the second multiplier is used to inner product X ( n ) and W ( n ). The second adders are used to sequentially add the aforementioned X ( n ) W ( n ) to generate X ^T ( n ) W ( n ). The subtractor is used to subtract d(n) from X ^T ( n ) W ( n ).

According to still another embodiment of the present invention, each of the foregoing adaptive filters conforms to the following formula: W ( n +1)= W (n)+ μe ( n ) X ( n ), e( n )=d (n)- X ^T ( n ) W ( n ), each of the aforementioned adaptive filters includes a plurality of arithmetic units, a Carry Save Adder (CSA), and a subtractor, and further, the foregoing Each of the arithmetic units includes a first multiplier, a first adder, and a second multiplier. The first adder is electrically coupled to the first multiplier, and the second multiplier is electrically coupled to the first adder. The carry storage adder is electrically coupled to the aforementioned second multiplier of each of the foregoing arithmetic units. The subtractor is electrically coupled to the aforementioned carry storage adder.

Operationally, in the foregoing operation unit, the first multiplier is used to multiply μe ( n ) and X ( n ), and the first adder is used to phase W (n) and μe ( n ) X ( n ) Plus, and the second multiplier is used to inner product X ( n ) and W ( n ). The carry storage adder is used to add the aforementioned X ( n ) W ( n ) to generate X ^T ( n ) W ( n ). The subtractor is used to subtract d(n) from X ^T ( n ) W ( n ).

In order to achieve the above object, another aspect of the present invention is directed to a blind signal separation system including a plurality of adaptive filters, a plurality of adders, and a plurality of message maximizing elements. In the configuration, the foregoing adaptive filters are arranged in a matrix, and each of the foregoing adders is disposed in one of the columns of the matrix, and each of the foregoing message maximizing elements is disposed in one of the columns of the matrix.

In operation, the first row of each of the foregoing adaptive filters is configured to receive the first output signal and process the first output signal. The first intermediate signal is output, and the second row of each of the foregoing adaptive filters is configured to receive the second output signal and process the second output signal to output a second intermediate signal. The foregoing adders are respectively configured to receive the first and second intermediate signals outputted by the foregoing adaptive filters arranged in the corresponding columns and the corresponding first input source or second input source, and corresponding to the foregoing And the second intermediate signal is added to the corresponding first or the second input source to output the first output signal and the second output signal, respectively. The message maximizing components are respectively configured to receive the respective first and second output signals outputted by the adders arranged in the corresponding columns, and respectively according to the foregoing first and second output signals A form is used to look up the table to output a plurality of messages to maximize the signal.

Furthermore, the first row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the first output signal and the corresponding message maximization signal to output the first An intermediate signal, wherein the second row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the second output signal and the corresponding message maximization signal to output the foregoing Two intermediate signals.

According to an embodiment of the invention, the third row of each of the foregoing adaptive filters is configured to receive the third output signal and the corresponding message maximization signal, and perform the third output signal and the corresponding message maximization signal. After processing, the third intermediate signal is output. The third column of the adders disposed in the matrix is configured to receive the first, second, and third intermediate signals and the third input source output by the foregoing adaptive filters arranged in the third column, and the first, The second and third intermediate signals are added to the third input source to output a third output signal.

According to another embodiment of the present invention, the foregoing adaptive filters are arranged in the first column for receiving the first output signal, and the adaptive filters are arranged in the second column for receiving the second output signal. The foregoing adaptive filters are arranged in the third column for receiving the third output signal and are omitted.

According to still another embodiment of the present invention, each of the foregoing adaptive filters conforms to the following formula: W ( n +1)= W (n)+ μe ( n ) X ( n ), e( n )=d (n)- X ^T ( n ) W ( n ), each of the foregoing adaptive filters includes a plurality of arithmetic units, a plurality of second adders, and a subtractor, and further, each of the foregoing operational units One includes a first multiplier, a first adder, and a second multiplier. The first adder is electrically coupled to the first multiplier, and the second multiplier is electrically coupled to the first adder. The second adders are electrically connected in series, and each of the second adders is electrically coupled to a second multiplier of each of the corresponding arithmetic units.

Operationally, in the foregoing operation unit, the first multiplier is used to multiply μe ( n ) and X ( n ), and the first adder is used to phase W (n) and μe ( n ) X ( n ) Plus, and the second multiplier is used to inner product X ( n ) and W ( n ). The second adders are used to sequentially add the aforementioned X ( n ) W ( n ) to generate X ^T ( n ) W ( n ). The subtractor is used to subtract d(n) from X ^T ( n ) W ( n ).

According to still another embodiment of the present invention, each of the foregoing adaptive filters conforms to the following formula: W ( n +1)= W (n)+ μe ( n ) X ( n ), e( n )=d (n)- X ^T ( n ) W ( n ), each of the foregoing adaptive filters includes a plurality of arithmetic units, a carry storage adder, and a subtractor, and further, each of the foregoing operational units A first multiplier, a first adder, and a second multiplier are included. The first adder is electrically coupled to the first multiplier, and the second multiplier is electrically coupled to the first adder. The carry storage adder is electrically coupled to the aforementioned second multiplier of each of the foregoing arithmetic units. The subtractor is electrically coupled to the aforementioned carry storage adder.

Operationally, in the foregoing operation unit, the first multiplier is used to multiply μe ( n ) and X ( n ), and the first adder is used to phase W (n) and μe ( n ) X ( n ) Plus, and the second multiplier is used to inner product X ( n ) and W ( n ). The carry storage adder is used to add the aforementioned X ( n ) W ( n ) to generate X ^T ( n ) W ( n ). The subtractor is used to subtract d(n) from X ^T ( n ) W ( n ).

Therefore, according to the technical content of the present invention, an embodiment of the present invention provides a blind signal separation system and implements a blind signal separation system by using an oversized integrated circuit, thereby improving a conventional technique using a software algorithm, which requires a large number of The operation, while delaying the time to get the original source.

In addition, the embodiment of the present invention can further omit the diagonal term adaptive filter, so that in addition to reducing the cost of the blind signal separation system, the operation speed of the blind signal separation system can be improved, and the blind signal separation system can be improved. effectiveness.

Furthermore, the adaptive filter of the embodiment of the present invention can further carry The adder is stored to replace the plurality of adders in the adaptive filter in which the adaptive filter is electrically connected in a row, thereby improving the operation speed of the adaptive filter and improving the efficiency of the adaptive filter.

In order to make the description of the present disclosure more complete and complete, reference is made to the accompanying drawings and the accompanying drawings. However, the embodiments provided are not intended to limit the scope of the invention, and the description of the operation of the structure is not intended to limit the order of its execution, and any device that is recombined by the components produces equal devices. The scope covered by the invention.

The drawings are for illustrative purposes only and are not drawn to the original dimensions. On the other hand, well-known elements and steps are not described in the embodiments to avoid unnecessarily limiting the invention.

One aspect of the present invention provides a blind signal separation system including a plurality of adaptive filters, a plurality of adders, and a plurality of message maximizing elements. In the configuration, the foregoing adaptive filters are arranged in a matrix, and each of the foregoing adders is disposed in one of the columns of the matrix, and each of the foregoing message maximizing elements is disposed in one of the columns of the matrix.

In operation, the first row of each of the foregoing adaptive filters is configured to receive the first input source, and process the first input source to output a first intermediate signal, and the foregoing adaptive filters are The second row of each column is for receiving the second input source, and processing the second input source A second intermediate signal is output. Each of the adders is configured to receive the first and second intermediate signals outputted by the adaptive filters arranged in the corresponding columns, and add the corresponding first and second input signals to output A plurality of output signals. The message maximizing components are respectively configured to receive corresponding output signals output by the adders arranged in the corresponding columns, and perform a lookup table according to the corresponding output signals to output a plurality of message maximization signals.

Furthermore, the first row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the first input source and the corresponding message maximization signal to output the first An intermediate signal, wherein the second row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the second input source and the corresponding message maximization signal to output the foregoing Two intermediate signals. The implementation of the blind signal separation system of the embodiment of the present invention will be explained in FIG. 1 below.

FIG. 1 is a schematic diagram of a blind signal separation system 100 according to an embodiment of the invention. As shown in FIG. 1, the blind signal separation system 100 includes a 1×1 order adaptive filter W11, a 1×2 order adaptive filter W12, a 2×1 order adaptive filter W21, and a 2×2. A Very Large Scale Integration (VLSI) architecture such as the adaptive filter W22, the first adder 110, the second adder 150, the first message maximizing component 120, and the second message maximizing component 160.

In operation, the 1×1 order adaptive filter W11 is configured to receive the first input source X1, and process the first input source X1 to output the first intermediate The first 1×2 adaptive filter W12 is configured to receive the second input source X2 and process the second input source X2 to output a second intermediate signal. Subsequently, the first adder 110 is configured to add the first intermediate signal and the second intermediate signal, and output the first output signal. The first message maximizing component 120 is configured to receive the first output signal, and output a first message maximization signal 1-2y1 according to the first output signal to perform a lookup to a form, and maximize the first message. 1-2y1 is fed back to the 1×1th order and the 1×2th order adaptive filters W11 and W12.

Furthermore, the 1×1th order and the 1×2th order adaptive filters W11 and W12 are further configured to receive the first message maximization signal 1-2y1, and the 1×1st order adaptive filter W11 is further used to An input source X1 is processed by the first message maximization signal 1-2y1 to output a first intermediate signal, and the 1×2th order adaptive filter W12 is further used to maximize the second input source X2 and the first message. After the processing of 1-2y1, the second intermediate signal is output. Then, the first adder 110 is further used to add the first intermediate signal and the second intermediate signal, and output the first output signal, so that the training coefficient is continued, and finally the first output signal can be obtained.

In addition, in operation, the 2×1 order adaptive filter W21 is configured to receive the first input source X1, and process the first input source X1 to output the first intermediate signal; the 2×2th order adaptive filter The W22 is configured to receive the second input source X2 and process the second input source X2 to output a second intermediate signal. Subsequently, the second adder 150 is configured to add the first intermediate signal and the second intermediate signal, and output the second output signal. The second message maximizing component 160 is further configured to receive the second output signal, and according to the second output signal The table is looked up to output a second message maximization signal 1-2y2, and the second message maximization signal 1-2y2 is fed back to the 2x1st order and 2x2st order adaptive filters W21 and W22.

Next, the 2×1th order and the 2×2th order adaptive filters W21 and W22 are further configured to receive the second message maximization signal 1-2y2, and the 2×1 order adaptive filter W21 is further used to The input source X1 and the second message maximization signal 1-2y2 are processed to output a first intermediate signal, and the 2×2 order adaptive filter W22 is further configured to use the second input source X2 and the second message to maximize the signal 1 -2y2 is processed to output a second intermediate signal. Then, the second adder 2 is further used to add the first intermediate signal and the second intermediate signal, and output the second output signal, so that the training coefficient is continued, and finally the second output signal can be obtained.

In this way, the embodiment of the present invention provides a blind signal separation system 100 and implements the blind signal separation system 100 by using a large integrated circuit, so as to improve the conventional technique using a software algorithm, which requires a large number of operations. Delay in getting the original source.

In an embodiment of the present invention, the third row of each of the foregoing adaptive filters of the blind signal separation system is configured to receive the third input source and the corresponding message maximization signal, and to the third input source. And the corresponding message maximization signal is processed to output a third intermediate signal. The implementation of the blind signal separation system of the embodiment of the present invention will be explained in FIG. 2 below.

2 is a schematic diagram of a blind signal separation system 200 in accordance with another embodiment of the present invention. Compared with the blind signal separation system 100 shown in FIG. 1, the blind signal separation system 200 further includes a 1×3 order adaptive filter. W13, 2×3 order adaptive filter W23, 3×1 order adaptive filter W31, 3×2 order adaptive filter W32, 3×3 order adaptive filter W33, and third adder 270 and a third message maximizing element 280.

In operation, the 1×3th adaptive filter W13 is configured to receive the third input source X3, and process the third input source X3 to output a third intermediate signal, where the first adder 210 is configured to The second and third intermediate signals are added, and the first output signal is output. The first message maximizing component 220 is configured to receive the first output signal, and output a first message maximization signal 1-2y1 according to the first output signal to perform lookup on the form, and maximize the first message 1 -2y1 is fed back to the 1st to 1st order, the 1st to 2nd order, and the 1st to 3rd order adaptive filters W11, W12, and W13.

Furthermore, the 1×1th order, the 1×2th order, and the 1×3th order adaptive filters W11, W12, and W13 are further configured to receive the first message maximization signal 1-2y1, the 1×1st order adaptability. The filter W11 is further configured to process the first input source X1 and the first message maximization signal 1-2y1 to output a first intermediate signal, and the 1×2 order adaptive filter W12 is further used to the second input source X2. The second intermediate signal is output after being processed by the first message maximization signal 1-2y1, and the 1×3th adaptive filter W13 is further configured to perform the third input source X3 and the first message maximization signal 1-2y1. After processing, the third intermediate signal is output. Then, the first adder 210 is further used to add the first, second and third intermediate signals, and output the first output signal, so that the training coefficient is continued, and finally the first output signal can be obtained.

In addition, in operation, the 2×3th adaptive filter W23 is configured to receive the third input source X3, and process the third input source X3 to output the third The third intermediate signal 240 is used to add the first, second and third intermediate signals and output the second output signal. The second message maximizing component 250 is further configured to receive the second output signal, and output a second message maximization signal 1-2y2 according to the second output signal to look up the form, and maximize the second message signal 1 -2y2 is fed back to the 2x1st order, the 2x2th order, and the 2x3th order adaptive filters W21, W22, and W23.

Secondly, the 2×1th order, the 2×2th order and the 2×3th order adaptive filters W21, W22 and W23 are further used for receiving the second message maximization signal 1-2y2, the 2×1 order adaptive filtering. The W21 is further configured to process the first input source X1 and the second message maximization signal 1-2y2 to output a first intermediate signal, and the 2×2 order adaptive filter W22 is further configured to use the second input source X2 and The second message maximization signal 1-2y2 is processed to output a second intermediate signal, and the 2×3th order adaptive filter W23 is further configured to process the third input source X3 and the second message maximization signal 1-2y2. The third intermediate signal is output afterwards. Then, the second adder 240 is further used to add the first, second and third intermediate signals, and output the second output signal, so that the training coefficient is continued, and finally the second output signal can be obtained.

In addition, in operation, the 3×1 order adaptive filter W31 is configured to receive the first input source X1, and process the first input source X1 to output the first intermediate signal, and the 3×2th order adaptive filter. The W32 is configured to receive the second input source X2, and process the second input source X2 to output a second intermediate signal, and the 3×3th order adaptive filter W33 is configured to receive the third input source X3 and The input source X3 performs processing to output a third intermediate signal. Subsequently, the third adder 270 is configured to use the first, second, and third intermediate signals Add and output a third output signal. The third message maximizing component 280 is configured to receive the third output signal, and output a third message maximization signal 1-2y3 according to the third output signal to perform lookup to the form, and maximize the third message signal 1 -2y3 is fed back to the 3x1st order, the 3x2th order, and the 3x3th order adaptive filters W31, W32, and W33.

Secondly, the 3×1st order, the 3×2th order and the 3×3th order adaptive filters W31, W32 and W33 are further used for receiving the third message maximization signal 1-2y3, the 3×1 order adaptive filtering. The W31 is further configured to process the first input source X1 and the third message maximization signal 1-2y3 to output a first intermediate signal, and the 3×2 order adaptive filter W32 is further used to the second input source X2. The third message maximizing signal 1-2y3 is processed to output a second intermediate signal, and the 3×3th order adaptive filter W33 is further configured to process the third input source X3 and the third message maximizing signal 1-2y3. The third intermediate signal is output afterwards. Then, the third adder 270 is further configured to add the first, second, and third intermediate signals, and output a third output signal, so that the training coefficient is continued, and finally the third output signal can be obtained.

In another embodiment of the present invention, the adaptive filters of the blind signal separation system are arranged in the first column for receiving the first input source, and the foregoing adaptive filters are arranged in the second column. The system for receiving the second input source and the aforementioned adaptive filters arranged in the third column for receiving the third input source is omitted. The implementation of the blind signal separation system of the embodiment of the present invention will be explained in FIG. 3 below.

FIG. 3A is a schematic diagram of a blind signal separation system 300 according to still another embodiment of the present invention. Compared to the blind signal separation system shown in Figure 2 The blind signal separation system 300 arranges the aforementioned adaptive filters in the first column for receiving the first input source (for example, the 1×1 order adaptive filter W11), and the aforementioned adaptive filtering. Arranging in the second column for receiving the second input source (for example, the 2×2th adaptive filter W22) and the foregoing adaptive filters are arranged in the third column for receiving the third input The source (for example, the 3×3th adaptive filter W33) is omitted.

In other words, the first input source transmitted to the first column of the matrix is not processed by, for example, the 1×1 order adaptive filter W11, and the second input source transmitted to the second column of the matrix does not pass. For example, the 2×2th adaptive filter W2 processes it, and the third input source transmitted to the third column of the matrix is not processed by, for example, the 3×3th adaptive filter W33. It is based on a diagonal term adaptive filter, such as the 1×1th adaptive filter W11, the 2×2th adaptive filter W22, and the 3×3th adaptive filter W33 are only used for scaling (scaling) The matrix has no effect on the solution of the blind signal separation system 300, so the diagonal term adaptive filter can be omitted, so that in addition to reducing the cost of the blind signal separation system 300, the blind signal separation system 300 can be enhanced. The speed of operation increases the efficiency of the blind signal separation system 300.

In addition, FIG. 3B is a schematic diagram of a blind signal separation system 300 according to another embodiment of the present invention, which is to reset the blind signal separation system 300 in FIG. 3A to make the line trace between electronic components more. Concisely, the production efficiency of the blind signal separation system 300 is improved.

In still another embodiment of the present invention, each of the foregoing adaptive filters conforms to the following formula: W ( n +1)= W (n)+ μe ( n ) X ( n ), e( n )=d (n)- X ^T ( n ) W ( n ), the architecture of each of the aforementioned adaptive filters 400 is shown in FIG. 4, which includes a plurality of arithmetic units 410 and a plurality of second additions The 420 and the subtractor 430, further, each of the foregoing arithmetic units 410 includes a first multiplier 412, a first adder 414, and a second multiplier 416. In the configuration, the first adder 414 is electrically coupled to the first multiplier 412 , and the second multiplier 416 is electrically coupled to the first adder 414 . The second adders 420 are electrically connected in series, and each of the second adders 420 is electrically coupled to the second multiplier 416 of each of the corresponding arithmetic units 410. .

Operationally, in the foregoing operation unit, the first multiplier 412 is used to multiply μe ( n ) and X ( n ), and the first adder 414 is used to use W (n) and μe ( n ) X ( n ) is added, and the second multiplier 416 is used to inner product X ( n ) and W ( n ). The second adder 420 is configured to sequentially add the foregoing X ( n ) W ( n ) to generate X ^T ( n ) W ( n ). Subtractor 430 is used to subtract d(n) from X ^T ( n ) W ( n ).

In still another embodiment of the present invention, each of the foregoing adaptive filters conforms to the following formula: W ( n +1)= W (n)+ μe ( n ) X ( n ), e( n )=d (n)- X ^T ( n ) W ( n ), the architecture of each of the aforementioned adaptive filters 500 is shown in FIG. 5, which includes a plurality of arithmetic units 510 and a carry storage adder 520. And a subtractor 530. Further, each of the foregoing arithmetic units includes a first multiplier 512, a first adder 514, and a second multiplier 516. The first adder 514 is electrically coupled to the first multiplier 512 , and the second multiplier 516 is electrically coupled to the first adder 514 . The carry storage adder 520 is electrically coupled to the second multiplier 516 of each of the foregoing arithmetic units. The subtractor 530 is electrically coupled to the carry storage adder 520 described above.

Operationally, in the foregoing operation unit, the first multiplier 512 is used to multiply μe ( n ) and X ( n ), and the first adder 514 is used to use W (n) and μe ( n ) X ( n ) is added, and the second multiplier 516 is used to inner product X ( n ) and W ( n ). The carry storage adder 520 is configured to add the aforementioned X ( n ) W ( n ) to generate X ^T ( n ) W ( n ). Subtractor 530 is used to subtract d(n) from X ^T ( n ) W ( n ).

It should be noted that, compared with the adaptive filter 400 shown in FIG. 4, the adaptive filter 500 shown in FIG. 5 simplifies the most logical elements in the computation path, and in detail, the adaptability The filter 500 replaces the plurality of second adders 420 electrically connected in a row in the adaptive filter 400 by the carry storage adder 520, so that the operation speed of the adaptive filter 500 can be improved and the adaptability can be improved. The efficiency of filter 500.

FIG. 6 is a schematic diagram showing a message maximizing component of a blind signal separation system according to still another embodiment of the present invention. As shown in the figure, the message maximizing component can be designed as a read-only memory (ROM) so that the blind signal separation system can perform table lookup by the message maximizing component, when the input u is checked by the table. After the unit LUT has looked up the table, it can obtain y . For example, the table lookup unit is used to query the value calculated by the input u via the sigmoid function, and then by the complement and the left shift unit Complement-two and a left shift y performs an operation to output a message maximization signal 1-2y, in other words, after quantization and binning, the value is stored in a read-only memory ROM for subsequent calculation. It is not intended to limit the invention, but merely illustrates one of the implementations of the message maximizing element.

The first adder, the second adder and the third adder in the upper open graph can be implemented by a Carry Save Adder (CSA), and the adaptive filter in the upper open graph can be minimized. The Least Mean Square (LMS) algorithm is used to implement its update behavior, but it is not intended to limit the present invention. Those skilled in the art can selectively implement appropriate electronic components or suitable algorithms according to actual needs. .

One aspect of the present invention provides a blind signal separation system including a plurality of adaptive filters, a plurality of adders, and a plurality of message maximizing elements. In the configuration, the foregoing adaptive filters are arranged in a matrix, and each of the foregoing adders is disposed in one of the columns of the matrix, and each of the foregoing message maximizing elements is disposed in one of the columns of the matrix.

In operation, the first row of each of the foregoing adaptive filters is configured to receive the first output signal, and process the first output signal to output a first intermediate signal, and the foregoing adaptive filters are The second row of each column is for receiving the second output signal and processing the second output signal to output a second intermediate signal. The foregoing adders are respectively configured to receive the first and second intermediate signals outputted by the foregoing adaptive filters arranged in the corresponding columns and the corresponding first input source or second input source, and corresponding to the foregoing And the foregoing second intermediate signal and the corresponding aforementioned first or The second input sources are added to output the first output signal and the second output signal, respectively. The message maximizing components are respectively configured to receive the respective first and second output signals outputted by the adders arranged in the corresponding columns, and respectively according to the foregoing first and second output signals A form is used to look up the table to output a plurality of messages to maximize the signal.

Furthermore, the first row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the first output signal and the corresponding message maximization signal to output the first An intermediate signal, wherein the second row of each of the foregoing adaptive filters is further configured to receive a corresponding message maximization signal, and process the second output signal and the corresponding message maximization signal to output the foregoing Two intermediate signals. The implementation of the blind signal separation system of the embodiment of the present invention will be explained in FIG. 7 below.

FIG. 7 is a schematic diagram of a blind signal separation system 700 according to another embodiment of the present invention. As shown in FIG. 7, the blind signal separation system 700 includes a 1×1 order adaptive filter W11, a 1×2 order adaptive filter W12, a 2×1 order adaptive filter W21, and a 2×2. A Very Large Scale Integration (VLSI) architecture such as the adaptive filter W22, the first adder 710, the second adder 750, the first message maximizing component 720, and the second message maximizing component 760.

In operation, the 1×1 order adaptive filter W11 is configured to receive the first output signal u1, and process the first output signal u1 to output a first intermediate signal, and the 1×2th order adaptive filter W12 Used to receive the second loss The signal u2 is output, and the second intermediate signal is processed and the second intermediate signal is output. The first adder 710 is configured to receive the first and second intermediate signals outputted by the 1×1st order and the 1×2th order adaptive filters W11 and W12 and the first input source X1, and the first and second The intermediate signal is added to the first input source X1 to output a first output signal u1. The first message maximizing component 720 is configured to receive the first output signal u1 output by the first adder 710, and perform a lookup table according to the first output signal u1 to output a first message maximization signal 1-2y1. The first message maximization signal 1-2y1 is fed back to the 1×1th order and the 1×2th order adaptive filters W11 and W12.

Furthermore, the 1×1th order and the 1×2th order adaptive filters W11 and W12 are further configured to receive the first message maximization signal 1-2y1, and the 1×1st order adaptive filter W11 is further used to An output signal u1 and the first message maximization signal 1-2y1 are processed to output a first intermediate signal, and the 1×2 order adaptive filter W12 is further used to maximize the second output signal u2 and the first message. After the processing of 1-2y1, the second intermediate signal is output. Then, the first adder 710 is further configured to add the first and second intermediate signals to the first input source X1, so that the training coefficient is continued, and finally the first output signal u1 can be obtained.

In addition, in operation, the 2×1 order adaptive filter W21 is configured to receive the first output signal u1, and process the first output signal u1 to output the first intermediate signal, and the 2×2th order adaptive filter. The device W22 is configured to receive the second output signal u2 and process the second output signal u2 to output a second intermediate signal. The second adder 750 is configured to receive the first and second intermediate signals and the second input source X2 output by the 2×1st and 2×2th adaptive filters W21 and W22, and the first and second Intermediate signal and second input The source X2 is added to output the aforementioned second output signal u2. The second message maximizing component 760 is configured to receive the second output signal u2 output by the second adder 750, and perform a lookup table according to the second output signal u2 to output the second message maximization signal 1-2y2. The second message maximization signal 1-2y2 is fed back to the 2×1th order and 2×2th order adaptive filters W21 and W22.

Furthermore, the 2×1th order and the 2×2th order adaptive filters W21 and W22 are further configured to receive the second message maximization signal 1-2y2, and the 2×1 order adaptive filter W11 is further used to An output signal u1 and a second message maximization signal 1-2y2 are processed to output a first intermediate signal, and a 2×2 order adaptive filter W22 is further used to maximize the second output signal u2 and the second message. After the processing of 1-2y2, the second intermediate signal is output. Then, the second adder 750 is further configured to add the first and second intermediate signals to the second input source X2, thus continuing the training coefficient, and finally obtaining the second output signal u2.

In this way, the embodiment of the present invention provides a blind signal separation system 700 and implements the blind signal separation system 700 by using a large integrated circuit, so as to improve the manner in which the conventional technology adopts a software algorithm, which requires a large number of operations. Delay in getting the original source.

In an embodiment of the present invention, the third row of each of the foregoing adaptive filters of the blind signal separation system is configured to receive the third output signal and the corresponding message maximization signal, and to the third output signal and The corresponding message maximization signal is processed to output a third intermediate signal. The third column of the adders disposed in the matrix is configured to receive the first, second, and third intermediate signals and the third output by the foregoing adaptive filters arranged in the third column. The input source adds the first, second and third intermediate signals to the third input source to output a third output signal. The implementation of the blind signal separation system of the embodiment of the present invention will be explained in FIG. 8 below.

FIG. 8 is a schematic diagram of a blind signal separation system 800 in accordance with still another embodiment of the present invention. Compared with the blind signal separation system 700 shown in FIG. 7, the blind signal separation system 800 further includes a 1×3 order adaptive filter W13, a 2×3 order adaptive filter W23, and a 3×1 order adaptation. The filter W31, the 3x2th adaptive filter W32, the 3x3th adaptive filter W33, the third adder 870, and the third message maximizing element 880.

In operation, the 1×3th adaptive filter W13 is configured to receive the third output signal u3 and the first message maximization signal 1-2y1, and to the third output signal u3 and the first message maximization signal 1-2y1. After processing, a third intermediate signal is output. The first adder 810 is configured to add the first, second, and third intermediate signals to the first input source X1, and output the first output signal u1. The first message maximizing component 820 is further configured to receive the first output signal u1, and output a first message maximization signal 1-2y1 according to the first output signal u1 to perform lookup to the form, and maximize the first message. The signal 1-2y1 is fed back to the 1×1th order, the 1×2th order, and the 1×3th order adaptive filters W11, W12, and W13.

Furthermore, the 1×1th order, the 1×2th order, and the 1×3th order adaptive filters W11, W12, and W13 are further configured to receive the first message maximization signal 1-2y1, the 1×1st order adaptability. The filter W11 is further configured to process the first output signal u1 and the first message maximization signal 1-2y1 to output a first intermediate signal, and the 1×2th order adaptive filter W12 is further used to the second output signal u2. And processing the first message maximization signal 1-2y1 to output the second intermediate signal, and the 1×3th order adaptive filter W13 is further configured to perform the third output signal u3 and the first message maximization signal 1-2y1. After processing, the third intermediate signal is output. Then, the first adder 810 is further used to add the first, second and third intermediate signals, so that the training coefficient is continued, and finally the first output signal u1 can be obtained.

In addition, in operation, the 2×3th adaptive filter W23 is configured to receive the third output signal u3 and the second message maximization signal 1-2y2, and maximize the signal 1 for the third output signal u3 and the second message. -2y2 is processed to output a third intermediate signal. The second adder 840 is configured to add the first, second, and third intermediate signals to the second input source X2, and output the second output signal u2. The second message maximizing component 850 is further configured to receive the second output signal u2, and output a second message maximization signal 1-2y2 according to the second output signal u2 to perform lookup to the form, and maximize the second message. The signal 1-2y2 is fed back to the 2×1th order, the 2×2th order, and the 2×3th order adaptive filters W21, W22, and W23.

Furthermore, the 2×1th order, the 2×2th order, and the 2×3th order adaptive filters W21, W22, and W23 are further configured to receive the second message maximization signal 1-2y2, and the 2×1 order adaptive The filter W21 is further configured to process the first output signal u1 and the second message maximization signal 1-2y2 to output a first intermediate signal, and the second×2 order adaptive filter W22 is further configured to use the second output signal u2. And the second message maximization signal 1-2y2 is processed to output the second intermediate signal, and the 2×3th order adaptive filter W23 is further configured to perform the third output signal u3 and the second message maximization signal 1-2y2. After processing, output the third intermediate signal. Then, the second adder 840 is further used to add the first, second and third intermediate signals, so that the training coefficient is continued, and finally the second output signal u2 can be obtained.

In addition, in operation, the 3×1 order adaptive filter W31 is configured to receive the first output signal u1 and the third message maximization signal 1-2y3, and maximize the signal 1 for the first output signal u1 and the third message. -2y3 is processed to output a first intermediate signal, and the 3x2th adaptive filter W32 is configured to receive the second output signal u2 and the third message maximization signal 1-2y3, and to the second output signal u2 and the third The message maximization signal 1-2y3 is processed to output a second intermediate signal, and the 3×3th order adaptive filter W33 is configured to receive the third output signal u3 and the third message maximization signal 1-2y3, and to the third The output signal u3 and the third message maximization signal 1-2y3 are processed to output a third intermediate signal. The third adder 870 is configured to add the first, second, and third intermediate signals to the third input source X3, and output the third output signal u3. The third message maximizing component 880 is further configured to receive the third output signal u3, and output a third message maximization signal 1-2y3 according to the third output signal u3 to perform lookup to the form, and maximize the third message. The signal 1-2y3 is fed back to the 3×1st order, the 3×2th order, and the 3×3th order adaptive filters W31, W32, and W33.

Furthermore, the 3×1st order, the 3×2th order, and the 3×3th order adaptive filters W31, W32, and W33 are further configured to receive the third message maximization signal 1-2y3, and the 3×1st order adaptability. The filter W31 is further configured to process the first output signal u1 and the third message maximization signal 1-2y3 to output a first intermediate signal, and the 3×2 order adaptive filter W32 is further configured to use the second output signal u2. And the third message maximization signal 1-2y3 is processed to output the second intermediate signal, and the 3×3th order adaptive filter W33 is further configured to perform the third output signal u3 and the third message maximization signal 1-2y3. After processing, the third intermediate signal is output. Then, the third adder 870 is further used to add the first, second and third intermediate signals, so that the training coefficient is continued, and finally the third output signal u3 can be obtained.

In another embodiment of the present invention, the foregoing adaptive filters of the blind signal separation system are arranged in the first column for receiving the first output signal, and the foregoing adaptive filters are arranged in the second column for The person receiving the second output signal and the aforementioned adaptive filters arranged in the third column for receiving the third output signal are omitted. The implementation of the blind signal separation system of the embodiment of the present invention will be explained in the following FIG. 9A.

FIG. 9A is a schematic diagram of a blind signal separation system 900 according to another embodiment of the invention. Compared with the blind signal separation system 800 shown in FIG. 8, the blind signal separation system 900 arranges the aforementioned adaptive filters in the first column for receiving the first output signal (for example, the 1st to 1st order). The adaptive filter W11), the foregoing adaptive filters are arranged in the second column for receiving the second output signal (for example, the 2×2th adaptive filter W22) and the foregoing adaptive filter arrangement The third column for receiving the aforementioned third output signal (for example, the 3×3th order adaptive filter W33) is omitted.

In other words, the first output signal transmitted to the first column of the matrix is not processed by, for example, the 1×1st order adaptive filter W11, and the second output signal transmitted to the second column of the matrix does not pass. For example, the second 2× The second-order adaptive filter W2 processes it, and the third output signal transmitted to the third column of the above matrix is not processed by, for example, the 3×3th adaptive filter W33, which is based on the diagonal The item adaptive filter, such as the 1×1th adaptive filter W11, the 2×2th adaptive filter W22, and the 3×3th adaptive filter W33, only sacling the matrix, and the blind signal The solution of the separation system 900 has no effect, so the diagonal term adaptive filter can be omitted. In addition, the cost of the blind signal separation system 900 can be improved, and the operation speed of the blind signal separation system 900 can be improved, and the blindness can be improved. The efficiency of the signal separation system 900.

In addition, FIG. 9B is a schematic diagram of a blind signal separation system 900 according to another embodiment of the present invention, which is to reset the blind signal separation system 900 in FIG. 9A to make the line trace between the electronic components more. Concisely, the production efficiency of the blind signal separation system 900 is improved.

In still another embodiment of the present invention, the equations and structures of each of the foregoing adaptive filters have been explained in the descriptions of FIGS. 4 and 5, and are not described herein. In addition, in an optional embodiment of the present invention, the implementation of the message maximization component of the blind signal separation system has been explained in the description of FIG. 6 above, and will not be described herein.

The first adder, the second adder and the third adder in the upper open graph can be implemented by a Carry Save Adder (CSA), and the adaptive filter in the upper open graph can be calculated by using LMS. The method is implemented to achieve the update behavior, which is not intended to limit the present invention. Those skilled in the art can selectively implement appropriate electronic components or suitable algorithms according to actual needs.

It will be apparent from the above-described embodiments of the present invention that the application of the present invention has the following advantages. The embodiment of the present invention provides a blind signal separation system and implements a blind signal separation system by using an oversized integrated circuit, thereby improving the conventional technique using a software algorithm, requiring a large number of operations, and delaying the acquisition of the original signal source. time.

In addition, the embodiment of the present invention can further omit the diagonal term adaptive filter, so that in addition to reducing the cost of the blind signal separation system, the operation speed of the blind signal separation system can be improved, and the blind signal separation system can be improved. effectiveness.

Furthermore, the adaptive filter of the embodiment of the present invention may further replace the plurality of adders electrically connected in a column in the adaptive filter with a carry storage adder, thereby improving the adaptive filter. The speed of operation increases the efficiency of the adaptive filter.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and the present invention can be modified and modified without departing from the spirit and scope of the present invention. The scope is subject to the definition of the scope of the patent application attached.

Complement-two and a left shift‧‧‧take complement and left shift unit

Carry Save Adder‧‧‧ Carry Storage Adder

LUT‧‧‧Checklist unit

W11‧‧‧1×1 order adaptive filter

W12‧‧‧1×2 order adaptive filter

W13‧‧‧1×3 order adaptive filter

W21‧‧‧2×1 order adaptive filter

W22‧‧‧2×2 order adaptive filter

W23‧‧‧2×3 order adaptive filter

W31‧‧‧3×1 order adaptive filter

W32‧‧‧3×2 order adaptive filter

W33‧‧‧3×3 order adaptive filter

100‧‧‧Blind signal separation system

110‧‧‧First Adder

120‧‧‧First message maximizing component

150‧‧‧second adder

160‧‧‧Second message maximizing component

200‧‧‧Blind signal separation system

210‧‧‧First Adder

220‧‧‧First Message Maximization Element

240‧‧‧second adder

250‧‧‧Second message maximizing component

270‧‧‧ third adder

280‧‧‧ Third Message Maximization Element

300‧‧‧Blind Signal Separation System

310‧‧‧First Adder

320‧‧‧First message maximizing component

340‧‧‧second adder

350‧‧‧Second Message Maximization Element

370‧‧‧ third adder

380‧‧‧ Third Message Maximization Element

400‧‧‧Adaptive filter

410‧‧‧ arithmetic unit

412‧‧‧first multiplier

414‧‧‧First Adder

416‧‧‧Second multiplier

420‧‧‧second adder

430‧‧‧Subtractor

500‧‧‧Adaptive filter

510‧‧‧ arithmetic unit

512‧‧‧ first multiplier

514‧‧‧First Adder

516‧‧‧Second multiplier

520‧‧‧ Carry Storage Adder

530‧‧‧Subtractor

700‧‧‧Blind signal separation system

710‧‧‧First Adder

720‧‧‧First Message Maximization Element

750‧‧‧second adder

760‧‧‧Second message maximizing component

800‧‧‧Blind signal separation system

810‧‧‧First Adder

820‧‧‧First message maximizing component

840‧‧‧second adder

850‧‧‧Second message maximizing component

870‧‧‧ third adder

880‧‧‧ Third Message Maximization Element

900‧‧‧Blind signal separation system

910‧‧‧First Adder

920‧‧‧First Message Maximization Element

940‧‧‧second adder

950‧‧‧Second message maximizing component

970‧‧‧ third adder

980‧‧‧ Third Message Maximization Element

The above and other objects, features, advantages and embodiments of the present invention will become more <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt; </ RTI> <RTIgt;

Figure 2 is a diagram showing a blind signal separation according to another embodiment of the present invention. Schematic diagram of the system.

FIG. 3A is a schematic diagram showing a blind signal separation system according to another embodiment of the present invention; FIG. 3B is a schematic diagram showing a blind signal separation system according to another embodiment of the present invention.

FIG. 4 is a schematic diagram showing an adaptive filter according to still another embodiment of the present invention.

FIG. 5 is a schematic diagram showing an adaptive filter according to still another embodiment of the present invention.

FIG. 6 is a schematic diagram showing a message maximizing component of a blind signal separation system according to another embodiment of the invention.

FIG. 7 is a schematic diagram showing a blind signal separation system according to still another embodiment of the present invention.

FIG. 8 is a schematic diagram showing a blind signal separation system according to still another embodiment of the present invention.

FIG. 9A is a schematic diagram showing a blind signal separation system according to another embodiment of the present invention; FIG. 9B is a schematic diagram showing a blind signal separation system according to another embodiment of the present invention.

W11‧‧‧1×1 order adaptive filter

W12‧‧‧1×2 order adaptive filter

W21‧‧‧2×1 order adaptive filter

W22‧‧‧2×2 order adaptive filter

100‧‧‧Blind signal separation system

110‧‧‧First Adder

120‧‧‧First message maximizing component

150‧‧‧second adder

160‧‧‧Second message maximizing component

Claims (10)

A blind signal separation system includes: a plurality of adaptive filters arranged in a matrix, wherein a first row of each column of the adaptive filters is configured to receive a first input source and to the first input The source is processed to output a first intermediate signal, and the second row of the columns of the adaptive filters is configured to receive a second input source, and process the second input source to output a second intermediate a plurality of adders, each of the adders being disposed in one of the arrays for receiving the first and second outputs of the adaptive filters arranged in the respective columns An intermediate signal, and adding the corresponding first and second input signals to output a plurality of output signals; and a plurality of message maximizing elements, each of the message maximizing elements being disposed in the matrix One of the columns is configured to receive corresponding output signals output by the adder arranged in the corresponding column, and perform a table lookup according to the corresponding output signal to output a plurality of message maximization signals. The first row of the columns of the adaptive filters is further configured to receive a corresponding message maximization signal, and process the first input source and the corresponding message maximization signal to output the first intermediate signal. And the second row of the columns of the adaptive filters is further configured to receive a corresponding message maximization signal, and process the second input source and the corresponding message maximization signal to output the second intermediate signal. The blind signal separation system of claim 1, wherein the third row of each of the adaptive filters is configured to receive a third input source and a corresponding message maximization signal, and to the third input source And the corresponding message maximization signal is processed to output a third intermediate signal. The blind signal separation system of claim 2, wherein the adaptive filters are arranged in the first column for receiving the first input source, and the adaptive filters are arranged in the second column for receiving The second input source and the adaptive filters are arranged in the third column for receiving the third input source and are omitted. The blind signal separation system of claim 1, wherein each of the adaptive filters conforms to the following formula: W ( n +1) = W (n) + μe ( n ) X ( n ), e ( n )=d(n)− X ^T ( n ) W ( n ), wherein each of the adaptive filters comprises: a plurality of arithmetic units, each of the arithmetic units including a first multiplication For multiplying μe ( n ) and X ( n ); a first adder electrically coupled to the first multiplier and for using W ( n ) and μe ( n ) X ( n And performing a summation; and a second multiplier electrically coupled to the first adder for inner product of X ( n ) and W ( n ); a plurality of second adders, the second The adder is electrically connected in series, and each of the second adders is electrically coupled to the second multiplier of each of the corresponding arithmetic units, and is used to sequentially X ( n ) W ( n ) is added to generate X ^T ( n ) W ( n ); and a subtractor is electrically coupled to the last one of the second adders in series and used To subtract d(n) from X ^T ( n ) W ( n ). The blind signal separation system of claim 1, wherein each of the adaptive filters conforms to the following formula: W ( n +1) = W (n) + μe ( n ) X ( n ), e ( n )=d(n)− X ^T ( n ) W ( n ), wherein each of the adaptive filters comprises: a plurality of arithmetic units, each of the arithmetic units including a first multiplication For multiplying μe ( n ) and X ( n ); a first adder electrically coupled to the first multiplier and for using W (n) and μe ( n ) X ( n And performing a summation; and a second multiplier electrically coupled to the adder for inner product of X ( n ) and W ( n ); a carry storage adder electrically coupled to the The second multiplier of each of the arithmetic units, and used to add the X ( n ) W ( n ) to generate X ^T ( n ) W ( n ); and a subtractor electrically coupled to The carry stores the adder and is used to subtract d(n) from X ^T ( n ) W ( n ). A blind signal separation system comprising: The plurality of adaptive filters are arranged in a matrix, wherein the first row of the columns of the adaptive filters is configured to receive a first output signal, and process the first output signal to output a first An intermediate signal, wherein the second row of the columns of the adaptive filters is configured to receive a second output signal, and process the second output signal to output a second intermediate signal; a plurality of adders, Each of the adders is disposed in one of the columns of the matrix, and is configured to receive the first and second intermediate signals outputted by the adaptive filters arranged in the corresponding columns and the corresponding first Input source or a second input source, and adding the corresponding first and second intermediate signals to the corresponding first or second input source to respectively output the first output signal and the second output And a plurality of message maximizing elements, each of the message maximizing elements being disposed in one of the columns of the matrix for respectively receiving the corresponding first output of the adder arranged in the corresponding column And the first Outputting a signal, and respectively performing a lookup table according to the corresponding first and second output signals to output a plurality of message maximization signals, wherein the first row of the columns of the adaptive filters is further And receiving the corresponding message maximization signal, and processing the first output signal and the corresponding message maximization signal to output the first intermediate signal, and the second row of each of the adaptive filter columns The method further receives the corresponding message maximization signal, and processes the second output signal and the corresponding message maximization signal to output the second intermediate signal. The blind signal separation system of claim 1, wherein the third row of each of the adaptive filters is configured to receive a third output signal and a corresponding message maximization signal, and to the third output signal And processing the corresponding signal maximization signal to output a third intermediate signal, wherein the third column of the adders is configured to receive the adaptive filters arranged in the third column Outputting the first, the second, and the third intermediate signals and a third input source, and adding the first, the second, and the third intermediate signals to the third input source to output the The third output signal. The blind signal separation system of claim 7, wherein the adaptive filters are arranged in the first column for receiving the first output signal, and the adaptive filters are arranged in the second column for receiving The second output signal and the adaptive filters arranged in the third column for receiving the third output signal are omitted. The blind signal separation system of claim 6, wherein each of the adaptive filters conforms to the following formula: W ( n +1) = W (n) + μe ( n ) X ( n ), e ( n )=d(n)− X ^T ( n ) W ( n ), wherein each of the adaptive filters comprises: a plurality of arithmetic units, each of the arithmetic units including a first multiplication For multiplying μe ( n ) and X ( n ); a first adder electrically coupled to the first multiplier and for using W (n) and μe ( n ) X ( n And performing a summation; and a second multiplier electrically coupled to the adder for inner product of X ( n ) and W ( n ); a plurality of second adders, the second adders Electrically connected in series, and each of the second adders is electrically coupled to the second multiplier of each of the corresponding computing units, and is used to sequentially perform the X ( n ) W ( n ) is added to generate X ^T ( n ) W ( n ); and a subtractor is electrically coupled to the last one of the second adders in series and used to d(n) minus X ^T ( n ) W ( n ). The blind signal separation system of claim 6, wherein each of the adaptive filters conforms to the following formula: W ( n +1) = W (n) + μe ( n ) X ( n ), e ( n )=d(n)− X ^T ( n ) W ( n ), wherein each of the adaptive filters comprises: a plurality of arithmetic units, each of the arithmetic units including a first multiplication For multiplying μe ( n ) and X ( n ); a first adder electrically coupled to the first multiplier and for using W (n) and μe ( n ) X ( n And performing a summation; and a second multiplier electrically coupled to the adder for inner product of X ( n ) and W ( n ); a carry storage adder electrically coupled to the The second multiplier of each of the arithmetic units, and used to add the X ( n ) W ( n ) to generate X ^T ( n ) W ( n ); and a subtractor electrically coupled to The carry stores the adder and is used to subtract d(n) from X ^T ( n ) W ( n ).