A Digital Enhancement Method for a Power Amplifier Suitable for 5G Signals

Abstract

The present invention provides a digital enhancement method for a power amplifier applicable to 5G signals, belonging to the field of wireless communication technologies. The method of the present invention sets up a digital enhancement module to process the generated OFDM sequence source signal, and performs time-domain clipping on the source signal to reduce the peak-to-average power ratio of the signal; cascades the Volterra model and the linear time-invariant system LTI to implement PA modeling; for the non-linear part, selects the short-memory GMP model, and the LTI is implemented through a filter; constructs a low-complexity compensation module to reduce the memory effect of the diagonal term, calculates the GMP model coefficients using the input signal and the feedback signal of the PA and stores them in a look-up table; obtains the final predistortion signal by multiplying the input signal by the corresponding coefficients in the look-up table and then summing. The present invention can implement dynamic configuration compensation based on a look-up table for multi-phase branches in hardware, greatly improving the flexibility of compensation while reducing the broadband compensation complexity, and having practical, efficient and high precision.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication, and specifically relates to a digital enhancement method for 5G signals for a broadband amplifier of hundreds of megabits. Background Art

[0002] With the development of wireless communication systems and the rapid increase in user demand, the need for high-throughput data transmission has led to the use of complex modulation and demodulation technologies to improve spectrum utilization. For example, the fifth-generation (5G) mobile communication system uses orthogonal frequency division multiplexing (OFDM) technology. However, while improving spectrum utilization, the signals in the system have a high peak-to-average power ratio (PAPR), which will cause nonlinear distortion and adjacent channel leakage when passing through a radio frequency power amplifier (PA), thereby reducing the linearity of the system and bringing interference.

[0003] The radio frequency power amplifier is an indispensable part of a wireless communication system. However, when the signal transmission power does not meet the channel requirements, in order to ensure normal transmission, a common method is to back off the input of the PA, which results in a decrease in the power amplifier efficiency. Digital predistortion compensates for the opposite distortion characteristics by preprocessing the input signal. After the signal is input into the power amplifier, the two distortions cancel each other out, realizing the linearized output of the power amplifier. Therefore, the linearization technology of broadband power amplifiers is an important technology in current wireless communication systems and has important practical significance.

[0004] In order to better study the linearization technology of broadband power amplifiers, accurate power amplifier modeling is required. The concise memory polynomial model has achieved a good balance between performance and complexity and is most widely used in the behavioral modeling of power amplifiers. However, the power amplifier under the excitation of large-bandwidth and high-peak-to-average ratio signals exhibits more complex nonlinear characteristics. The traditional memory polynomial model only contains diagonal term coefficients and has limited modeling accuracy for complex memory effects. After the traditional generalized memory polynomial model introduces cross terms, the number of coefficients is numerous, and high-complexity models require more hardware resources.

[0005] In practical applications, the compensation in the main path is usually implemented by means of a display look-up table (LUT). Therefore, to ensure good linearity of the power amplifier under broadband signal excitation and low resource consumption of the hardware, it is necessary to implement a low-complexity digital enhancement technology based on LUT under high-precision compensation for PA nonlinearity. Summary of the Invention

[0006] In view of the problem of large resource occupation in the compensation scheme based on look-up table commonly used for power amplifiers in broadband communication systems, and in order to meet the need of reducing the complexity of broadband predistortion compensation while maintaining good linearity and efficiency of the power amplifier, the present invention proposes a digital enhancement method for a hundred-megabit broadband amplifier applicable to 5G signals. Through the polyphase cascade structure of a short-memory nonlinear system and a long-memory LTI (linear time-invariant system), the memory depth of the diagonal terms is effectively reduced, and dynamic configuration compensation based on look-up tables for polyphase branches is realized in hardware, greatly improving the flexibility of compensation.

[0007] A digital enhancement method for a power amplifier applicable to 5G signals provided by the present invention includes the following steps:

[0008] (1) A signal source module generates an OFDM signal x n ; the digital enhancement module sets the crest factor clipping rate, adjusts the peak-to-average power ratio of the OFDM signal, and generates a clipped signal v k ; the digital enhancement module interpolates v k to generate a polyphase signal for transmission, which passes through a power amplifier and is sampled for feedback signals;

[0009] (2) Construct a PA model, and use the Volterra series to represent the nonlinearity in the signal transmission process; after low-pass filtering and downsampling, the equivalent baseband output structure of the PA model consists of two parts: a nonlinear system involving the current signal and previous signals, and a linear time-invariant system;

[0010] (3) Construct a compensation module in the digital enhancement module, including:

[0011] (1) Represent the nonlinear system using a GMP (generalized memory polynomial) model. Let the diagonal term, lag term, and lead term basis function matrices of this model be Φ a , Φ b , Φ c respectively, and the corresponding model coefficient matrices be W a , W b , W c ; the diagonal term model coefficient a a is in W kl , the lag term model coefficient b b is in W klm , and the lead term model coefficient c c is in W klm ;

[0012] (2) Take the low-memory part of the diagonal term basis function matrix Φ a to form a matrix Φ a ', and the convolution output matrix Ψ a ' of Φ a, together with Φ b and Φ c to form a new basis function matrix Ψ, Ψ = [Ψ a Φ b Φ c ; The convolution kernel H is implemented by configuring a FIR filter bank;

[0013] (3) Calculate the model coefficient matrix W by the least squares method, W = (Ψ Η Ψ) -1 Ψ Η X n ; where, W = [W a W b W c T , the superscript T represents the transpose, H represents the conjugate matrix, -1 represents the inverse matrix, and X n represents the input multi-phase signal;

[0014] (4) Establish a look-up table for the model coefficients and implement it in hardware, including:

[0015] Express the output of the GMP model as follows:

[0016]

[0017] where, K and L a represent the non-linear order and memory depth of the diagonal term respectively, M b and L b represent the lag depth and memory depth of the lag term respectively, M c and L c represent the lead depth and memory depth of the lead term respectively; |x(n - l)| represents the envelope of the input signal x(n - l); * represents the convolution operation, and h k (n) represents the unit impulse response of the (k + 1)-th filter in the configured FIR filter bank; the model coefficients in the above formula are calculated in advance to establish a look-up table;

[0018] Multiply the input signal x(n - l) by the model coefficients obtained from the corresponding look-up table index and sum them to obtain the final pre-distorted signal y(n).

[0019] ​The advantages and positive effects of the method of the present invention are as follows: The present invention realizes a digital enhancement method for a hundred-megabit broadband amplifier. By adopting a polyphase cascade structure and cascading a short-memory nonlinear system and a long-memory LTI (linear time-invariant system), the memory depth of the diagonal terms in the existing power amplifier model is effectively reduced. The dynamic configuration compensation based on a look-up table for the polyphase branches can be implemented in hardware, which greatly improves the flexibility of compensation while reducing the complexity of broadband compensation. The method of the present invention can reduce the hardware resources required to achieve broadband power amplifier linearization, has a higher configuration flexibility, reduces the implementation cost in actual engineering, maintains the good linearity and efficiency of the power amplifier, and has practical high efficiency and high precision. Description of the Drawings

[0020] Figure 1 It is a specific implementation flowchart of the digital enhancement method for the power amplifier of the present invention;

[0021] Figure 2 It is an inverse model DPD (digital predistortion) structure;

[0022] Figure 3 It is the corresponding normalized power spectrum diagram of the PA modeling output signal and the actual output signal of the PA of the present invention;

[0023] Figure 4 It is a schematic diagram of the implementation structure of the cascade structure model based on LUT constructed by the present invention;

[0024] Figure 5 It is the normalized power spectrum diagram of the digital enhancement signal proposed by the present invention and the original PA output signal. Detailed Embodiment

[0025] The following combines specific embodiments and drawings to describe the implementation manner of the present invention in detail and clearly.

[0026] A digital enhancement method for a power amplifier applicable to 5G signals implemented in an embodiment of the present invention, as Figure 1 shown, Figure 1 The structure of the cascade unit in Figure 4 is shown. The implementation of each step of the method of the present invention will be described separately below.

[0027] Step 1: The signal source module modulates the data symbol X k by quadrature phase shift keying (QPSK) or quadrature amplitude modulation (QAM), and generates an OFDM signal x n through inverse fast Fourier transform (IFFT) and oversampling.

[0028]

[0029] In the formula, X kThe data symbol modulated by quadrature phase shift keying (QPSK) or quadrature amplitude modulation (QAM) for the k-th subcarrier, N is the number of subcarriers, J is the oversampling rate, and j is the imaginary unit.

[0030] Step 2: The digital enhancement module sets the clipping rate CR of crest factor reduction (CFR), and clips the OFDM signal x n to obtain x n in the time domain, and reduces the PAPR of the OFDM signal through frequency domain clipping and filtering (CF) to generate v k The proportionality coefficient β is pre-calculated and stored in a table.

[0031]

[0032]

[0033]

[0034] In the formula, f n is the time domain clipping noise, and f n is obtained as F through fast Fourier transform (FFT) k F k is the frequency domain clipping noise, is the filtered frequency domain clipping noise, β is the proportionality coefficient, and V k is the frequency domain clipping signal. The inverse Fourier transform (IFFT) is performed on V k to obtain v k .

[0035] Step 3: The following parameters can be dynamically configured as needed:

[0036] (1) The signal clipping rate CR can be configured to adjust the PAPR of the OFDM signal as required;

[0037] (2) According to the bandwidth BW and the corresponding rate SR of the source signal, the appropriate number of polyphase branches M is set, and the delay parameters corresponding to each phase can be configured to achieve the sum after adding the parameters corresponding to the delays of each phase. The value of M is as follows:

[0038]

[0039] Among them, p is the number of half-band filters in the interpolation module, SR is the signal rate, f clk is the hardware clock frequency, and ceil(·) is rounding up.

[0040] (3) Calculate the pre-distortion (DPD) coefficient according to the source signal and the feedback signal, and the pre-distortion (DPD) coefficient stored in the corresponding main look-up table and spare look-up table can be updated.

[0041] Step 4: v kInterpolate to generate a multi-phase signal for transmission, pass it through a power amplifier (PA), and then sample it. Use the source signal v k and the sampled signal u k , as Figure 2 shown, adopt a model inversion structure to extract coefficients, construct a PA model based on a low-complexity cascaded model and train it. Any non-linearity during transmission can be represented by the Volterra series. Its non-linear system represents the entire RF signal chain. For ease of explanation, the PA is modeled as a second-order non-linear system with memory y(t). Figure 2 Among them, y(n) is the discrete signal after sampling y(t), y(t) is the output of the PA, x(n) is the discrete signal corresponding to x(t), and x(t) is the input of the PA.

[0042]

[0043]

[0044] In the formula, g(t) is the continuous signal after v k modulation, p(t) is a zero-order hold, T is the sampling duration, t, τ1, τ2 all represent corresponding moments, and 0 < τ1 ≤ τ2 ≤ T, w c represents the carrier frequency; k2(τ1, τ2) represents the second-order kernel of the Volterra series model.

[0045]

[0046]

[0047]

[0048] Among them, z(t) is the component generated corresponding to the second-order non-linear system, p1(t) is the pulse from τ2 to T + τ1, and p2(t) is the pulse from τ1 to τ2. Z(jw) represents the frequency-domain transformation of z(t), w represents the frequency, δ represents the impulse response, H 1k (jw), H 2k (jw) respectively correspond to the frequency response characteristics of the non-linear terms. LTI represents a linear time-invariant system, and NL represents a non-linear system. After low-pass filtering and downsampling, the above formula shows that the structure of the equivalent baseband output of the PA second-order non-linear system consists of two parts: a non-linear conversion involving the current and previous samples, represented by v k 2 , v k v k+1 and labeled as NL, and the other is a linear time-invariant system, labeled as LTI.

[0049] Accurate modeling of PA nonlinearity can be achieved by cascading the Volterra model and the LTI system. The power spectra of the modeled output signal and the actual output signal of the PA are as Figure 3 shown. The NMSE (Normalized Mean Square Error) performance can reach -36 dB, enabling high-precision modeling of the PA.

[0050] Step 5: From the above derivation, the nonlinear part of the PA model selects the short-memory GMP model, and the long-memory LTI is implemented through a filter. The extraction process of the coefficient matrix W of the model is described below. In addition, the same cascaded model is also used to construct a low-complexity compensation module. The model used in this compensation module is the same as the PA model. The parameter extraction process is the same as the above derivation except that the input and output need to be swapped, that is, the output of the PA is used as the input and the source signal is used as the output. The specific hardware structure is as Figure 4 shown. The short-memory GMP model coefficients are stored in the lookup table, and the long-memory LTI is implemented by a polyphase filter bank.

[0051] The GMP model is expressed as follows:

[0052]

[0053] where x(n) represents the input signal and y(n) represents the output signal; a kl , b klm , c klm represent the diagonal term, lag term, and lead term model coefficients respectively, K a , L a represent the nonlinear order and memory depth of the diagonal term respectively, K b , M b , L b represent the nonlinear order, lag depth, and memory depth of the lag term respectively, K c , M c , L c represent the nonlinear order, lead depth, and memory depth of the lead term respectively; |x(n - l)| represents the signal envelope.

[0054] Rewrite the above equation in matrix form: where:

[0055]

[0056]

[0057]

[0058] In the equation, Φ a , Φ b , Φ c are the diagonal term, lag term, and lead term basis function matrices respectively, Wa , W b , W c represents model parameters. Considering that the memory depth of the cross terms is relatively low, only the memory effect of the diagonal terms is reduced, and the low-memory part of the Φ a matrix is written as Φ a ', whose dimension is much lower than that of Φ a .

[0059] Ψ a = H * Φ a '

[0060] where Φ a ' is the column matrix of low-memory basis functions for diagonal terms, H is the convolution kernel of M1×N1, M1 is the filter order, N1 is the number of filters, which can be implemented by a configurable FIR filter bank. The matrix Ψ a is the convolution output matrix of the Φ a ' column matrix and the convolution kernel H. Combining Ψ a with Φ b , Φ c forms a new basis function matrix, Ψ = [Ψ a Φ b Φ c .

[0061] The model coefficient matrix is calculated by the LS (least squares) algorithm:

[0062] W = (Ψ Η Ψ) -1 Ψ Η X n

[0063] where the superscript H represents the conjugate matrix and the superscript -1 represents the inverse matrix. X n represents the input polyphase signal.

[0064] For example, in Figure 1 , the generated polyphase signal X n-L …X n is input into the cascade unit. L is the number of cascade units, and the polyphase signal X n = [X[nM + 1], X[nM + 2]…X[(n + 1)M]]. As described in Step 3, the number of branches M of the polyphase signal can be set according to the bandwidth and rate. For example, in Figure 4As shown, the indexing configuration and cross-indexing configuration are set in the cascade unit of the present invention, both of which are configurable signal paths. The input signal amplitude can be used as an index to input to the corresponding look-up table to query the model coefficients. Among them, the coefficient index of the diagonal term only involves the current moment. The amplitude of the current signal indexes the look-up table corresponding to the signal at this moment through the signal path of the indexing configuration. The coefficient indexes of the lag term and the lead term involve the cross of signals at different moments, and the cross-signal path of the cross-indexing configuration is used to index the look-up tables at different moments. The configurable delay is to delay the multi-phase signal, so as to multiply the multi-phase signal and the compensation coefficient at the corresponding moment.

[0065] The multi-phase input signal queries the model coefficients in the corresponding look-up table according to the signal amplitude. In the indexing configuration, each phase signal finds the corresponding model coefficient from the look-up table, multiplies the signal with the model coefficient at the configured delay moment, and then outputs after filtering by the multi-phase filter bank. The multi-phase filter bank is composed of the filters used in the reduction of the diagonal term basis function mentioned above; in the cross-indexing configuration, the input multi-phase signal finds the corresponding model coefficient from the corresponding look-up table, multiplies the signal with the model coefficient at the configured delay moment, and the output is added to the output of the indexing configuration for the corresponding phase signal after a fixed delay to obtain the compensation signal Y n =[Y[nM + 1], Y[nM + 2]…Y[(n + 1)M]].

[0066] Step six: Calculate the model coefficients according to the input and output signals of the power amplifier, and the model coefficients establish the LUT for hardware implementation.

[0067] First, perform the following equation transformation on the model coefficients:

[0068]

[0069]

[0070]

[0071] Among them, K represents the order of the diagonal term after reducing the order.

[0072] Store the product sum of the non-linear power part in a table, use the look-up table operation to avoid complex complex multiplication operations, save DSP resources, and effectively reduce the hardware implementation difficulty compared with direct calculation. After that, only need to calculate the amplitude of the signal through the cordic algorithm (coordinate rotation digital calculation algorithm), and index the corresponding model parameters in the storage. At the same time, after extracting the model parameters, in order to reduce the calculation amount in the convolution process during specific implementation, consider the linear transformation, Ψ a *W a is equivalent to Φ a '*W a After convolution, the final implementation expression is obtained:

[0073]

[0074] where h k (n) represents the unit impulse response of the (k + 1)-th filter in the polyphase filter bank, where k = 0, 1, 2, … K. According to this formula, the input signal x(n - l) is multiplied by the table value obtained from the corresponding lookup table index and then summed to obtain the final predistortion signal, realizing a low-complexity digital enhancement module.

[0075] As Figure 5 shown, the effectiveness of the method of the present invention is proved by the experimental platform. The center frequency of the power amplifier of the 5G micro base station is 3.5 GHz, and the saturation power is 24 dbm. The present invention collects 3000 groups of 100-MHz OFDM signals. The method of the present invention can reduce the model coefficients while ensuring the model accuracy, and the NMSE (Normalized Mean Square Error) performance can reach -36 dB. When the signal bandwidth is 100 M, the ACLR (Adjacent Channel Leakage Ratio) of the power amplifier can be improved by more than 15 dB, and the ACPR (Adjacent Channel Power Ratio) can reach below -45 dBc, meeting the requirements of high efficiency and high precision that are practical and feasible in actual engineering.

Claims

1. A digital enhancement method for a power amplifier applicable to 5G signals, characterized in that, Set up a digital enhancement module to process the OFDM signal x generated by the signal source module n The method includes the following steps: (1) A crest factor clipping rate is preset in the digital enhancement module to adjust the peak-to-average power ratio of the OFDM signal and generate a clipped signal v k , for v k interpolation is performed to generate a multi-phase signal for transmission, which passes through a power amplifier and the feedback signal is sampled; where OFDM represents orthogonal frequency division multiplexing; (2) Build a PA model and train the model. Represent the nonlinearity in the signal transmission process using the Volterra series. The structure of the equivalent baseband output of the PA model consists of two parts: a nonlinear system involving the current signal and previous signals, and a linear time-invariant system; PA represents the power amplifier. (3) Build a compensation module in the digital enhancement module, including: (1) The GMP model is used to represent the nonlinear system. Let the diagonal term, lag term, and lead term basis function matrices of the model be Φ a , Φ b , Φ c respectively, and the corresponding model coefficient matrices be W a , W b , W c ; The diagonal term model coefficient a kl is in W a , the lag term model coefficient b klm is in W b , and the lead term model coefficient c klm is in W c ; (2) Take the low-memory part of the diagonal-term basis function matrix Φ a to form the matrix Φ a '. Convolve Φ a ' with the convolution kernel H to obtain the convolution output matrix Ψ a . Combine Ψ b with Φ c to form a new basis function matrix Ψ; the convolution kernel H is implemented by configuring an FIR filter bank; (3) Calculate the model coefficient matrix W by the least squares method, W = (Ψ Η Ψ) -1 Ψ Η X n ; where W = [W a W b W c T , the superscript T represents the transpose, H represents the conjugate matrix, -1 represents the inverse matrix, and X n represents the input polyphase signal;​ (4) Establish a look-up table for the model coefficients and implement it in hardware, including: Represent the output of the GMP model as follows: where K and L a represent the non - linear order and memory depth of the diagonal terms respectively, and M b , L b represent the lag depth and memory depth of the lag terms respectively, and M c , L c represent the lead depth and memory depth of the lead terms respectively; |x(n - l)| represents the envelope of the input signal x(n - l); * represents the convolution operation, and h k (n) represents the unit impulse response of the (k + 1)-th filter in the FIR filter bank; The model coefficients in the above formula Calculate the model coefficients in advance and establish a lookup table; Multiply the input signal x(n - l) by the model coefficients obtained from the corresponding look-up table index and sum them to obtain the final predistorted signal y(n).

2. The method according to claim 1, wherein In the step (i) described above, a crest factor clipping rate CR is set, and the OFDM signal x n is obtained by time-domain clipping The peak-to-average power ratio of the OFDM signal is reduced by frequency-domain clipping and filtering as follows: where, f n is the time-domain clipping noise, and for f n the frequency-domain clipping noise F is obtained through the fast Fourier transform FFT k , is the frequency-domain clipped noise after filtering; N is the number of subcarriers, J is the oversampling rate; X k is the data symbol modulated by the k-th subcarrier; β is the proportionality coefficient; V k is the frequency-domain clipping signal, and for V k the inverse Fourier transform IFFT is performed to obtain v k .

3. The method according to claim 1, characterized in that, In step (3) above, the diagonal term, lag term, and lead term basis function matrices of the GMP model are as follows: Among them, K a , L a respectively represent the non - linear order and memory depth of the diagonal term, K b , M b , L b respectively represent the non - linear order, lag depth, and memory depth of the lag term, K c , M c , L c respectively represent the non - linear order, lead depth, and memory depth of the lead term.

4. The method according to claim 1 or 2 or 3, characterized in that, For the digital enhancement module, set the number of polyphase branches M according to the bandwidth and signal rate of the source signal, and configure the delay parameters for each phase.

5. The method according to claim 1 or 2 or 3, characterized in that, The digital enhancement module calculates a predistortion coefficient based on the source signal v of the input PA k and the sampled feedback signal, and updates the look-up table.

6. The method according to claim 1 or 3, characterized in that, In step (3) above, in the polyphase signal input cascade unit generated, an index configuration and a cross-index configuration are set in the cascade unit. In the index configuration and cross-index configuration, each phase signal finds the corresponding model coefficients from the look-up table, and the signal is multiplied by the model coefficients at the configured delay time; the output signal multiplied by the index configuration is input to the FIR filter bank, and the output signal multiplied by the cross-index configuration is delayed by a fixed time and then summed with the signal output from the FIR filter bank to obtain the compensation signal.

Images