A power amplifier nonlinear correction algorithm for broadband high-order modulation signals

Abstract

A power amplifier nonlinear correction algorithm suitable for broadband high-order modulated signals relates to the technical field of channel nonlinear equalization, comprising: obtaining a baseband signal and generating a timing signal and a timing synchronization signal after synchronous preprocessing; generating an equalization input vector according to the timing signal and the timing synchronization signal; constructing an OS-ELM equalization unit for processing the equalization input vector and generating an equalization output result; extracting the equalization output vector after the equalization output result is processed by a buffer; constructing a symbol decision unit for generating a symbol decision soft value according to the timing signal and the equalization output vector; constructing an online training unit for generating an online label vector according to the timing signal and the symbol decision soft value, and generating a hidden layer output matrix according to the timing signal and the equalization input vector; and finally generating a weight matrix by combining the online label vector, the hidden layer output matrix and the hidden layer output matrix of the historical sample; and updating the equalization output result by using the weight matrix; and the calculation complexity is low.

Description

Technical Field

[0001] The invention relates to the technical field of channel nonlinear equalization, and in particular to a power amplifier nonlinear correction algorithm suitable for broadband high-order modulation signals. Background Art

[0002] As a key technology in data transmission systems, the quality of equalization technology has a direct impact on subsequent signal demodulation and decoding, signal-to-noise ratio estimation and rate modulation. For high-order modulated signals, the modulation format with non-constant mode characteristics faces serious power amplifier nonlinear distortion problems, so equalization is more difficult.

[0003] At the same time, in modern communication systems, the performance of equalization is usually improved by sacrificing the pilot / training sequence overhead at the expense of bandwidth; however, this algorithm is difficult to apply to high-throughput data transmission systems, and the constant modulus characteristics of the pilot / training sequence make it difficult to establish a correct nonlinear correction model, which puts forward more stringent requirements for the equalization of broadband high-order modulation systems. At present, in order to reduce the overhead of the pilot / training sequence, blind equalization algorithms such as the least mean square (LMS) algorithm, the recursive least square (RLS) algorithm, and the constant modulus (CM) algorithm and its improved algorithm are generally used; and for the problem of power amplifier nonlinear correction, the existing technologies include nonlinear equalization algorithms based on the Volterra series model, the Hammerstein model algorithm and its improved algorithm, and polynomial-based nonlinear equalization algorithms.

[0004] However, the LMS algorithm performs LS operation on the decision result of the equalizer output and its input sample to obtain the equalization filter coefficient under the MSE criterion; however, the algorithm relies on decision direction, and the existence of low signal-to-noise ratio or nonlinear distortion will inevitably lead to erroneous decision direction, thereby deteriorating the equalization performance.

[0005] Although the RLS-based equalization algorithm does not need to store past input samples, it still requires the decision results of the equalizer output to update the filter coefficients under the LS criterion, and still does not consider the impact of nonlinear distortion and low signal-to-noise ratio.

[0006] Although the CM-based equalization algorithm is not affected by the decision direction, it is not suitable for high-order amplitude-phase modulation systems (such as N-QAM and M-APSK modulation); its improved algorithms such as multi-mode (Multi modulus, MM) and decision direction constant modulus (Decision direction constant modulus, DD CM) algorithms are suitable for high-order amplitude-phase modulation systems, but are inevitably affected by the signal-to-noise ratio and do not consider the influence of the nonlinear distortion of the power amplifier.

[0007] In response to the nonlinear distortion of power amplifiers, the equalization algorithm based on the Volterra series model is gradually being eliminated due to the complex modeling method and high computational complexity of the Volterra series model, as well as its unsuitability for modern nonlinear distortion models.

[0008] Equalization algorithms based on the Hammerstein model, such as LMS / RLS, are still affected by the signal-to-noise ratio due to the existence of decision pointing, and there is no improved CM algorithm based on the Hammerstein model. There is an urgent need for a nonlinear correction algorithm for broadband high-order modulated signal power amplifiers.

[0009] In summary, the above-mentioned equalization algorithms have some insurmountable problems for broadband high-order modulated data transmission communication systems affected by power amplifier nonlinear distortion; therefore, we propose a power amplifier nonlinear correction algorithm with low computational complexity. Summary of the invention

[0010] The object of the present invention is to provide a power amplifier nonlinear correction algorithm suitable for broadband high-order modulated signals, which has low computational complexity.

[0011] The present invention is achieved through the following technical solutions:

[0012] A power amplifier nonlinear correction algorithm suitable for broadband high-order modulated signals, comprising:

[0013] After acquiring the baseband signal and synchronous preprocessing, a timing signal with 1 times the symbol clock is generated. A timing synchronization signal that is times the symbol clock;

[0014] generating an equalization input vector according to the timing signal and the timing synchronization signal;

[0015] Construct an OS-ELM equalization unit to perform linear / nonlinear channel equalization on the equalization input vector and generate an equalization output result;

[0016] After the equalization output result is processed by the buffer, the equalization output vector is extracted;

[0017] Constructing a symbol decision unit for generating a symbol decision soft value according to the timing signal and the equalization output vector;

[0018] An online training unit is constructed to generate an online label vector according to the timing signal and the symbol judgment soft value, and to generate a hidden layer output matrix according to the timing signal and the equalization input vector; and finally a weight matrix is ​​generated by combining the online label vector, the hidden layer output matrix and the hidden layer output matrix of the historical samples;

[0019] Use the weight matrix to update the equalization output results.

[0020] Furthermore, the calculation formula of the timing signal and the timing synchronization signal is:

[0021] ;

[0022] ;

[0023] ;

[0024] in, is the baseband signal after DDC, It is a timing signal of 1 times symbol clock; for times the symbol clock, and It is a multiple of 4, obtained from engineering experience. It is the timing synchronization signal.

[0025] Furthermore, the calculation formula of the balanced input vector is:

[0026] ;

[0027] in, Input vector to the equalization module.

[0028] Furthermore, the OS-ELM equalization unit includes an input layer, a hidden layer and an output layer.

[0029] Furthermore, the calculation formula of the balanced output result is:

[0030] ;

[0031] in is the input weight matrix of the OS-ELM equalization unit, is the hidden layer bias vector of the OS-ELM equalization unit, is the output weight vector of the OS-ELM equalization unit, is the hidden layer activation function of the OS-ELM equalization unit, It is the equalization output result of OS-ELM equalization unit.

[0032] Furthermore, the calculation formula of the balanced output vector is:

[0033] ;

[0034] is the balanced output vector.

[0035] Furthermore, the symbol decision unit includes a convolutional layer and a fully connected layer.

[0036] Furthermore, the calculation formula of the symbol decision soft value is:

[0037] ;

[0038] in, is the symbol decision soft value, and They are respectively represented as the real and imaginary parts of the symbol decision soft value; is the output weight vector of the full connection for real part decision; is the fully connected input weight matrix of the real part decision; is the fully connected hidden layer bias vector for real part decision; is the output weight vector of the full connection for imaginary part judgment; is the fully connected input weight matrix of the imaginary part decision; is the fully connected hidden layer bias vector for imaginary part decision;

[0039] and are the real and imaginary part convolution layer output vectors respectively, and the calculation formula is;

[0040] ;

[0041] According to the balanced output vector , the decision unit CNN layer output and The calculation formula is expressed as:

[0042] ;

[0043] Among them, the and They are respectively the real part operation and the imaginary part operation; It is the first The convolution kernel weights of the convolution kernels, and and Constitute the demodulation symbols.

[0044] Furthermore, the online label vector The calculation formula is:

[0045] ;

[0046] The hidden layer output matrix The calculation formula is:

[0047] .

[0048] Furthermore, the calculation formula of the weight matrix is:

[0049] .

[0050] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0051] The present invention discloses a power amplifier nonlinearity correction algorithm suitable for broadband high-order modulated signals, which utilizes an OS-ELM equalization unit and a symbol decision unit to effectively improve the bit error rate performance of broadband high-order modulated signals in power amplifier nonlinear distortion scenarios while reducing computational complexity.

[0052] Specifically, the OS-ELM equalization unit equalizes the input signal affected by linear / nonlinear channel distortion according to the timing signal and synchronization signal, thereby improving the anti-fading performance of high-speed reception; and obtains the equalized input vector through the buffer to further improve the correctness of symbol decision; finally, the online training unit is used to process past samples and current samples to realize adaptive real-time update of the weight matrix and improve the accuracy of the output of the OS-ELM equalization unit.

[0053] Moreover, this algorithm targets the communication scenarios of broadband high-order modulated signals. According to the OS-ELM equalization unit, symbol decision unit and online training unit, a broadband high-order modulated signal adaptive equalizer is formed. It has the advantages of resistance to noise, ISI linear distortion and power amplifier nonlinear distortion, thereby improving the transmission distance of transmission and reception while reducing the influence of power amplifier nonlinear distortion. It brings many implementation plans for large-bandwidth and high-speed data transmission receivers, which is of great significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the process of the present invention.

[0055] Figure 2 It is a schematic diagram of the structure of the present invention.

[0056] Figure 3 Schematic diagram of the OS-ELM equalization unit of the present invention.

[0057] Figure 4 It is a schematic diagram of the decision module of the present invention.

[0058] Figure 5 It is a flow chart of the online training unit of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0060] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0062] In the description of the present invention, it should be noted that if the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside" and the like appear, the orientation or position relationship indicated is based on the orientation or position relationship shown in the accompanying drawings, or is the orientation or position relationship in which the product of the application is usually placed when used. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0063] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0064] Example 1

[0065] As attached Figure 1-2 A power amplifier nonlinear correction algorithm suitable for broadband high-order modulation signals is shown, comprising:

[0066] After acquiring the baseband signal and synchronous preprocessing, a timing signal with 1 times the symbol clock is generated. A timing synchronization signal that is times the symbol clock; wherein It is the oversampling multiple after timing synchronization, which is set by engineering experience and is generally 2 to 4;

[0067] generating an equalization input vector according to the timing signal and the timing synchronization signal;

[0068] Construct an OS-ELM equalization unit to perform linear / nonlinear channel equalization on the equalization input vector and generate an equalization output result;

[0069] After the equalization output result is processed by the buffer, the equalization output vector is extracted;

[0070] A decision unit is constructed to generate a symbol decision soft value according to the timing signal and the equalization output vector;

[0071] An online training unit is constructed to generate an online label vector according to the timing signal and the symbol judgment soft value, and to generate a hidden layer output matrix according to the timing signal and the equalization input vector; and finally a weight matrix is ​​generated by combining the online label vector, the hidden layer output matrix and the hidden layer output matrix of the historical samples;

[0072] Use the weight matrix to update the equalization output results.

[0073] According to some preferred embodiments of the present invention, the OS-ELM equalization unit comprises an input layer, a hidden layer and an output layer, which can be obtained according to the following process:

[0074] According to the synchronization preprocessing module, the symbol clock timing signal and A synchronization signal of times the symbol clock; It is the oversampling multiple after timing synchronization, which is set based on engineering experience and is generally 1~2 times of 4.

[0075] Combined with Figure 3 As shown, according to the oversampling multiple , get the number of OS-ELM input layers ,Right now:

[0076] ;

[0077] Said is the symbol span, which is set by engineering experience and is generally 10~20.

[0078] According to the number of input layers , get the number of hidden layers of OS-ELM ,Right now:

[0079] ;

[0080] According to the number of hidden layers , get the number of OS-ELM output layers ,Right now:

[0081] ;

[0082] Combined with Figure 4 As shown, according to some preferred embodiments of the present invention, the decision unit includes a convolutional layer and a fully connected layer, which can be constructed according to the following process:

[0083] According to the number of output layers of OS-ELM , the receptive field of the convolutional layer of the decision unit and the number of convolution kernels It is expressed as:

[0084] ;

[0085] According to the output dimension of the convolutional layer, determine the network architecture of the fully connected layer, that is, the number of hidden layers in the fully connected layer. It is expressed as:

[0086] ;

[0087] The output layer of the fully connected layer is a single neuron, that is, the output of the decision unit is a symbol decision soft value.

[0088] According to some preferred embodiments of the present invention, the working process of the OS-ELM equalization unit is given by the following process:

[0089] remember The baseband signal after DDC, the synchronous preprocessing module will output a timing signal of 1 times the symbol clock as well as Times symbol clock The synchronization signal ; expressed as:

[0090] ;

[0091] ;

[0092] ;

[0093] Right now It is a timing signal of 1 times symbol clock; for times the symbol clock, and It is a multiple of 4, obtained from engineering experience. It is a timing synchronization signal;

[0094] remember OS-ELM equalization unit input vector at the rising edge triggering moment It is expressed as:

[0095] ;

[0096] remember is the input weight matrix of OS-ELM, is the hidden layer bias vector of OS-ELM, is the output weight vector of OS-ELM, then the equalization output result of OS-ELM equalization unit is It is expressed as:

[0097] ;

[0098] Among them, the is the hidden layer activation function of the OS-ELM equalization unit, which is any continuous function (such as sigmoid, tanh, softmax, relu, etc.), determined by engineering experience;

[0099] It should be noted that since there may be an inter-symbol interference (ISI) residual signal in the equalization output result, the inter-symbol interference residual signal vector can be extracted after the equalization output result is cached by the buffer, and then the signal vector is processed by the decision unit. Finally, the interference residual signal in the equalization output result can be removed after combined training by the online training unit, so that the accuracy of the equalization processing result is improved.

[0100] According to some preferred embodiments of the present invention, the working process of the decision unit is given by the following process:

[0101] According to the equalization output results of OS-ELM equalization unit , the equalized output vector of the decision unit input is obtained through the buffer Buffer , which is the inter-symbol interference residual signal vector, namely:

[0102] ;

[0103] According to the equalization output vector , the decision unit CNN layer output and The calculation formula is expressed as:

[0104] ;

[0105] Among them, the and They are respectively the real part operation and the imaginary part operation; It is the first The convolution kernel weights of the convolution kernels, and and Constitute the demodulation symbols.

[0106] Accordingly, the symbol decision soft value of the decision unit , expressed as:

[0107] ;

[0108] in, is the symbol decision soft value, and They are respectively represented as the real and imaginary parts of the symbol decision soft value; is the output weight vector of the full connection for real part decision; is the fully connected input weight matrix of the real part decision; is the fully connected hidden layer bias vector for real part decision; is the output weight vector of the full connection for imaginary part judgment; is the fully connected input weight matrix of the imaginary part decision; is the fully connected hidden layer bias vector for imaginary part decision;

[0109] and are the real and imaginary part of the convolutional layer output vectors, expressed as:

[0110] ;

[0111] Combined with Figure 5 As shown, according to some preferred embodiments of the present invention, the working process of the online training unit is given by the following process:

[0112] According to the timing signal , and symbol decision soft value , after being processed by the cache module in the online training unit, the online label vector of OS-ELM is obtained ,Right now:

[0113] ;

[0114] According to the timing signal , and the balanced input vector, after being processed by the cache module in the online training unit, the hidden layer output matrix is ​​obtained , expressed as:

[0115] ;

[0116] The hidden layer output matrix updated last time is , then the output weight update method of OS-ELM is expressed by the following calculation formula, namely:

[0117] ;

[0118] thus, It is updated to the OS-ELM equalization unit to realize the adaptive real-time update function of the equalizer.

[0119] Compared with the existing blind equalization technology based on nonlinear / linear modeling, the algorithm has better noise immunity and nonlinear distortion equalization performance. First, the baseband signal after digital down conversion (Digital down convert, DDC) is synchronized and preprocessed (including timing / carrier synchronization) to output the timing synchronization signal of 1 times the symbol clock and the timing synchronization signal of times the symbol clock. Then, the online sequential-extreme learning machine (OS-LEM) equalization unit is used as an equalizer to The timing synchronization signal of times the symbol clock is used for linear / nonlinear channel equalization; then, the decision unit outputs the symbol decision soft value according to the ISI residual signal vector in the equalization output result cached in the buffer; finally, the weight coefficient of OS-ELM is updated online through the online training unit, and the OS-ELM equalization unit outputs the equalization result.

[0120] In addition, compared with the traditional blind equalization algorithm, the present invention has better resistance to nonlinear distortion and noise performance. Compared with the existing blind equalization method based on the nonlinear distortion model, the modeling complexity of the nonlinear distortion model is removed. Compared with the existing nonlinear equalization technology based on machine learning, the equalization network architecture and weights are simpler. At the same time, the decision unit is improved by the convolutional neural network (CNN), thereby outputting a modulation symbol decision result with a higher accuracy.

[0121] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A power amplifier nonlinear correction algorithm suitable for broadband high-order modulated signals, characterized in that: include: After acquiring the baseband signal and synchronous preprocessing, a timing signal with 1 times the symbol clock is generated. A timing synchronization signal that is times the symbol clock; generating an equalization input vector according to the timing signal and the timing synchronization signal; Construct an OS-ELM equalization unit to perform linear / nonlinear channel equalization on the equalization input vector and generate an equalization output result; After the equalization output result is processed by the buffer, the equalization output vector is extracted; A decision unit is constructed to generate a symbol decision soft value according to the timing signal and the equalization output vector; An online training unit is constructed to determine soft values ​​according to the timing signal and the symbol, generate an online label vector, and generate a hidden layer output matrix according to the timing signal and the equalized input vector; Finally, the weight matrix is ​​generated by combining the online label vector, the hidden layer output matrix and the hidden layer output matrix of the historical samples; Use the weight matrix to update the equalization output results.

2. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulated signals according to claim 1, characterized in that: The calculation formula of the timing signal and the timing synchronization signal is: ； ； ； in, is the baseband signal after DDC, It is a timing signal of 1 times symbol clock; for times the symbol clock, and It is a multiple of 4, obtained from engineering experience. It is the timing synchronization signal.

3. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 2, characterized in that: The calculation formula of the balanced input vector is: ； in, Input vector to the equalization module.

4. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 3, characterized in that: The OS-ELM equalization unit includes an input layer, a hidden layer and an output layer.

5. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 4, characterized in that: The calculation formula of the equalization output result is: ； in is the input weight matrix of the OS-ELM equalization unit, is the hidden layer bias vector of the OS-ELM equalization unit, is the output weight vector of the OS-ELM equalization unit, is the hidden layer activation function of the OS-ELM equalization unit, It is the equalization output result of OS-ELM equalization unit.

6. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 5, characterized in that: The calculation formula of the balanced output vector is: ； is the balanced output vector.

7. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 6, characterized in that: The decision unit includes a convolutional layer and a fully connected layer.

8. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 7, characterized in that: The calculation formula of the symbol decision soft value is: ； in, is the symbol decision soft value, and They are respectively represented as the real and imaginary parts of the symbol decision soft value; is the output weight vector of the full connection for real part decision; is the fully connected input weight matrix of the real part decision; is the fully connected hidden layer bias vector for real part decision; is the output weight vector of the full connection for imaginary part judgment; is the fully connected input weight matrix of the imaginary part decision; is the fully connected hidden layer bias vector for imaginary part decision; and are the real and imaginary part convolution layer output vectors respectively, and the calculation formula is; ； According to the balanced output vector , the output of the decision unit CNN layer and The calculation formula is expressed as: ； Among them, the and They are respectively the real part operation and the imaginary part operation; It is the first The convolution kernel weights of the convolution kernels, and and Constitute the demodulation symbols.

9. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 8, characterized in that: The online label vector The calculation formula is: ； The hidden layer output matrix The calculation formula is: 。 10. The power amplifier nonlinear correction algorithm applicable to broadband high-order modulation signals according to claim 9, characterized in that: The calculation formula of the weight matrix is: 。

Images