CN104168038A - Superheterodyne signal receiving system based on non-linear real-time elimination - Google Patents

Abstract

The invention provides a superheterodyne signal receiving system based on non-linear real-time elimination. The superheterodyne signal receiving system can eliminate signals received by a receiver in a non-linear mode in real time, and is high in receiving sensitivity and effective for the input signals of all types. According to the signal receiving system, a non-linear real-time eliminating unit is additionally arranged in a traditional superheterodyne digital receiver and comprises an analog-digital converter, a data caching module, a non-linear compensation model kernel coefficient calculation module and a non-linear distortion elimination operation module, wherein the data caching module stores the digital signals in a short time mode and provides sample data for a non-linear elimination algorithm, the non-linear compensation model kernel coefficient calculation module obtains a non-linear compensation model kernel coefficient through calculation according to the characteristics of the digital signals to be compensated, and the non-linear distortion elimination operation module uses the non-linear compensation model kernel coefficient for performing non-linear distortion real-time elimination operation on the digital signals to be compensated and obtains the compensated digital signals.

Description

Superhet receiving system based on non-linear real-time elimination

Technical field

The present invention relates to a kind of receiving system, be specifically related to a kind of superhet receiving system based on non-linear real-time elimination, belong to signal of communication processing technology field.

Background technology

Digital Signal Processing is the key technology of wireless communication system.Since entering 21 century, along with the continuous increase of people to spectral bandwidth demand, narrow-band digital processing mode cannot meet more and more higher technical need, wideband digital receiver technology starts to come into one's own, and is widely used in the technical fields such as survey of deep space, electronic warfare, radio frequency control.These application demands all require used superhet digital receiver to possess wide input bandwidth coverage and large dynamic range (SFDR).

Can improve receiver dynamic range, the method that can take into account again power efficiency and bandwidth is to adopt certain linearization technique, for example, and widely used feed-forward technique, feedback technique etc. in transmitter." digital adaptation Feed Forward Power Amplifier linearisation research " (microwave journal, 19 (1): 80-84, in March, 2003) by increasing the distortion component in extra circuit extraction system output signal to be compensated, after being amplified, from output signal, deduct, to obtain the output signal of high linearity, but this technology makes circuit structure become very complicated, and amplitude, phase place and time delay in the carrier cancellation loop error of excuse me, but I must be leaving now causes distortion component to be difficult to be offset completely." the power amplifier linearization method based on state feedback " (microwave journal, 27 (4): 329-332, in April, 2010) the radio frequency output signal of system is directly fed back to input, by feeding back to reach the object of the inhibition to nonlinear distortion, but due to feedback loop increase delay deterioration the stability of system works, make the direct feedback technique of radio frequency be only applicable to the linearisation of narrowband systems.

Along with volume is little, at a high speed, appearance and the progress of low power consumption digital signal processor (DSP), self adaptation base band pre-distortion technology becomes a up-and-coming technology that overcomes nonlinear distortion." a kind of new for there being the look-up table predistorter of memory power amplifier " (electronic letters, vol, 36 (9): 1728-1732, in September, 2008) adopt the method for look-up table, have realized the compensation to memoryless nonlinear system.But along with the continuous increase of signal transmission bandwidth, the memory effect that the distortion of digital received front end nonlinearity has is more and more obvious, loop up table is helpless in this case.Randomized jitter injection technique can be upset the correlation of error term and input signal, " improve the SFDR index of ADC with random fluctuation signal method " (nuclear electronics and Detection Techniques, 21 (2): 110-114, March calendar year 2001) use this technology that error is distributed in all frequency components to reach the object that suppresses nonlinear distortion equably in the mode of similar white noise.But the linear compensation of analog to digital converter (ADC) has only been considered in the research of this direction at present, do not consider the nonlinear distortion of digital receiver other parts.

In sum, the shortcoming that prior art exists is as follows:

A, cannot ensure all types of signals all effective.Above-mentioned technology all requires some feature of known pending signal, and according to the feature of receiver and application background, and the feature that receives signal is unknown and urgently crack often.

B, high for operating frequency, the receiver system that signal bandwidth is large, what above-mentioned technology cannot be real-time eliminates non-linear.

C, above-mentioned technology are only applicable to radio-frequency front-end or ADC, and can not ensure has compensation effect to comprising digital received front end and ADC in the nonlinear distortion of interior whole digital receiver.

Summary of the invention

In view of this, the invention provides a kind of superhet receiving system based on non-linear real-time elimination, this system is all effective to all types of signals, and can nonlinear distortion eliminate in real time, strong to the detectability of small-signal.

This superhet receiving system, comprises the receiver being made up of superhet receiver rf front-end and analog to digital converter; Described heterodyne system receiver rf front-end receives the radiofrequency signal of input, and to the radiofrequency signal receiving amplify, down-conversion and filtering is transferred to analog to digital converter with the form of analog signal after processing; The analog signal receiving is converted to digital signal by described analog to digital converter.

In addition, also comprise by data cache module, nonlinear compensation model core coefficients calculation block and nonlinear distortion and eliminate the non-linear real-time elimination unit that computing module forms;

The digital signal after conversion is sent to non-linear real-time elimination unit by described analog to digital converter, and this digital signal is digital signal to be compensated; The process that described non-linear real-time unit carries out non-linear real-time elimination to digital signal to be compensated is:

(1) described data cache module is temporarily stored digital signal to be compensated, for nonlinear compensation model core coefficients calculation block provides the sample data of preseting length;

(2) described nonlinear compensation model core coefficients calculation block is carried out the calculating of nonlinear compensation core coefficient, is specially:

(201) successively the sample data in data cache module is carried out to discrete Fourier transform (DFT), then search out and be positioned at the signal that the first Nyquist frequency band and power spectrum are greater than predefined power spectrum threshold value, this signal was threshold signal, recorded number and the centre frequency thereof of threshold signal;

(202) according to number and the centre frequency thereof of crossing threshold signal, calculated harmonic wave or the intermodulation frequency value of threshold signal, the harmonic wave calculating or intermodulation frequency value are the frequency information of receiver nonlinear distortion component; The top step number of the harmonic wave calculating or intermodulation frequency value equates with the exponent number of nonlinear compensation model;

(203) according to many passbands of frequency information configuration integrate frequency domain filter of nonlinear distortion component, this many passbands frequency domain filter is column vector g, and its dimension is B × 1; Wherein the numerical value of B/2 is more than or equal to the frequency number of nonlinear distortion component;

(204) the many passbands frequency domain filter that adopts step (203) to generate carries out filtering to the digital signal to be compensated after discrete Fourier transform (DFT) in to frequency domain, and calculates the power of nonlinear distortion component; The power of described nonlinear distortion component is the quadratic sum of many passbands frequency domain filter output data;

(205) using the mean-square value of nonlinear distortion power as iterative target function, taking " nonlinear distortion power minimum " as criterion, the adaptive iteration that adopts steepest descent method to carry out nonlinear compensation core coefficient calculates, and obtains nonlinear compensation model core coefficient vector;

(3) described nonlinear distortion is eliminated the nonlinear compensation model core coefficient vector that computing module uses step (2) to obtain, and digital signal to be compensated is carried out to nonlinear distortion and eliminate in real time, the digital signal after being compensated.

Adopt asymmetric discrete Volterra progression as nonlinear compensation model, digital signal to be compensated to be compensated,, in described step (204), the power of nonlinear distortion component is:

$\begin{array}{c}{s}_f^2(k,\mathrm{\ω})={[y\left(k\right)-V^T\left(k\right)\mathrm{\ω}]}^T{\mathrm{gg}}^T[y\left(k\right)-V^T\left(k\right)\mathrm{\ω}]\\ =y^T\left(k\right){\mathrm{gg}}^{T}y\left(k\right)-y^T\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\mathrm{\ω}\\ -{\mathrm{\ω}}^{T}V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)+{\mathrm{\ω}}^{T}V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\mathrm{\ω}\end{array}---\left(1\right)$

Wherein: s _f(k, ω) is the output data of many passbands frequency domain filter; ω is nonlinear compensation model core coefficient vector; T represents matrix transpose computing; K is discrete time point;

y(k)＝[y(k-B+1) y(k-B+2) … y(k-1) y(k)] ^T,

V(k)＝[v(k-B+1) v(k-B+2) … v(k-1) v(k)] ^T,

V (k) represents to use y (k) according to the memory nonlinear column vector of each order of nonlinear compensation Construction of A Model,

v(k)＝[y ²(k) y(k)y(k-1) … y ²(k-N _d+1) y ³(k) y ²(k)y(k-1)… y ^D(k-N _d+1)] ^T,

N _dfor the corresponding memory depth of d rank Volterra core, 2≤d≤D; D is the exponent number of nonlinear compensation model;

In described step (205), target function J (ω) is:

$\begin{array}{c}J\left(\mathrm{\ω}\right)=E\left[s_f^2(k,\mathrm{\ω})\right]\\ =E\left[y^T\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\right]-E\left[y^T\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\right]\mathrm{\ω}\\ -{\mathrm{\ω}}^{T}E\left[V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\right]+{\mathrm{\ω}}^{T}E\left[V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\right]\mathrm{\ω}\end{array}---\left(2\right)$

The steepest decline iterative formula of nonlinear compensation model core coefficient:

ω(i)＝ω(i-1)-Γ×{E[V(k)gg ^TV ^T(k)]ω(i-1)-E[V(k)gg ^Ty(k)]} (3)

In formula (3), Γ is that N × N ties up diagonal matrix, for upgrading the fixing iteration step length μ of nonlinear compensation model core coefficient _d:

$\mathrm{\Γ}=\left[\begin{array}{ccccc}{\mathrm{\Λ}}_2& 0& ...& 0& 0\\ ...& ...& ...& ...& ...\\ 0& ...& {\mathrm{\Λ}}_d& ...& 0\\ ...& ...& ...& ...& ...\\ 0& 0& ...& 0& {\mathrm{\Λ}}_D\end{array}\right],{\mathrm{\Λ}}_d=\left[\begin{array}{ccccc}{\mathrm{\μ}}_d& 0& ...& 0& 0\\ ...& ...& ...& ...& ...\\ 0& ...& {\mathrm{\μ}}_d& ...& 0\\ ...& ...& ...& ...& ...\\ 0& 0& ...& 0& {\mathrm{\μ}}_d\end{array}\right]$

Wherein, Λ _drepresent the fixing iteration step length for upgrading d rank nonlinear compensation model core coefficient, Λ _dfor dimension diagonal angle square formation;

By average the finite length sequence that to ask mathematic expectaion to be converted to length be K, formula (3) is:

$\mathrm{\ω}\left(i\right)=\mathrm{\ω}(i-1)-\mathrm{\Γ}\×\left\{\right[\underset{k=k_b(i-1)}{\overset{k_e(i-1)}{\mathrm{\Σ}}}V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)]\·\mathrm{\ω}(i-1)-\underset{k=k_b(i-1)}{\overset{k_e(i-1)}{\mathrm{\Σ}}}V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\}---\left(4\right)$

Wherein, k _band k (i) _e(i) represent respectively starting point and the terminating point of the i time iteration data used; Can calculate nonlinear compensation model core coefficient vector ω by formula (4);

In described step (3), use asymmetric discrete Volterra progression to carry out nonlinear distortion while eliminating in real time to the digital signal to be compensated, the digital signal s (k) after compensation is:

s(k)＝y(k)-v ^T(k)×ω (5)

Wherein, the dimension of v (k) and ω is N × 1,

Described nonlinear compensation model core coefficients calculation block is carried out fast fourier transform to the bandpass modulation signal in digital signal to be compensated and is conciliate allocation and transportation calculation, obtain power spectrum data and the planisphere data of digital signal to be compensated, and send to host computer by data interface module; Described nonlinear distortion is eliminated computing module the bandpass modulation signal in the digital signal after compensating is carried out to fast fourier transform reconciliation allocation and transportation calculation, the power spectrum data of the digital signal after being compensated and planisphere data, and send to host computer by data interface module.

Described iteration step length μ _dshould meet:

$0<{\mathrm{\&mu;}}_d<\frac{1}{\mathrm{tr}\left[R\left(k\right)\right]}$ Or $0<{\mathrm{\&mu;}}_d<\frac{1}{(N+1)P_{\mathrm{in}}}$

Wherein, tr[R (k)] be the mark of autocorrelation matrix, P _infor the power of signal V (k) g.

Beneficial effect:

(1) in non-linear real-time elimination unit, according to the humorous intermodulation frequency value that involves of crossing threshold signal, use many passbands frequency domain filter to obtain nonlinear distortion component, power using nonlinear distortion component calculates nonlinear compensation model core coefficient as target function, make this system all effective to all types of input signals, do not need to predict the prior informations such as frequency, amplitude distribution and the modulation system of input signal.

(2) dynamic range performance of receiver is promoted significantly, strengthened the detectability of receiver to small-signal; Receiver the in the situation that of large and small signal coexistence, to the detectability of small-signal be improved significantly.

(3) process of nonlinear distortion being eliminated is carried out in real time, and the operation of algorithm does not affect the normal use of receiver.

Brief description of the drawings

Fig. 1 is this system composition and signal flow graph;

Fig. 2 is the flow chart of non-linear real-time elimination;

Fig. 3 is the effect contrast figure that antenna receiving signal carries out nonlinear compensation.

Embodiment

Below in conjunction with the accompanying drawing embodiment that develops simultaneously, describe the present invention.

The present embodiment provides a kind of superhet receiving system based on non-linear real-time elimination, this receiving system is strong to the detectability of small-signal, the process that nonlinear distortion is eliminated is carried out in real time, and the operation of algorithm does not affect the normal use of receiver; Simultaneously all effective to all types of input signals.

The concrete composition of this system as shown in Figure 1, comprises receiver, non-linear real-time elimination unit and data interface module; Wherein receiver comprises superhet receiver rf front-end and high-speed AD converter (ADC); Non-linear real-time elimination unit comprises data cache module, nonlinear compensation model core coefficients calculation block and nonlinear distortion elimination computing module.

Adopt this system to receive signal process that line nonlinearity eliminates in real time of going forward side by side to be:

(1) superhet receiver rf front-end receives the radiofrequency signal of input, and to the radiofrequency signal receiving amplify, down-conversion and filtering processing, then radiofrequency signal after treatment is transferred to high-speed AD converter with the form of analog signal.In the present embodiment, superhet receiver rf front-end comprises preselection filter, low noise amplifier, low-converter and intermediate-frequency channel filter.Wherein preselection filter is for unwanted signal and the interference of filtering input radio frequency signal; Low noise amplifier is for the radiofrequency signal after amplification filtering; Low-converter is for downconverting to high-frequency signal within the scope of the working frequency range of analog to digital converter; Intermediate-frequency channel filter is for the unwanted signal outside the passage producing after filtering down-conversion.

(2) high-speed AD converter carries out bandpass sampling to the analog signal receiving, and obtains digital signal to be compensated; Then give the data cache module in non-linear real-time elimination unit by digital signal to be compensated by High speed rear panel bus transfer.

(3) data cache module is stored in short-term to the digital signal to be compensated receiving, and for Nonlinear elimination provides the sample data that n length is M, n and M are the integer that is greater than 0.The function of this module realizes by High speed asynchronous FIFO (data input into/output from cache device).

(4) nonlinear compensation model core coefficients calculation block is calculated nonlinear compensation model core coefficient, and the nonlinear compensation model core coefficient of calculating is transferred to nonlinear distortion elimination computing module.Nonlinear compensation model core coefficients calculation block is carried out fast fourier transform reconciliation allocation and transportation calculation to the bandpass modulation signal in digital signal to be compensated simultaneously, obtain power spectrum data and the planisphere data of digital signal to be compensated, and be transferred to data interface module.The function of nonlinear compensation model core coefficients calculation block completes in DSP.

In the present embodiment, adopt asymmetric discrete Volterra progression as nonlinear compensation model, the nonlinear distortion of receiver to be compensated, the computational process of nonlinear compensation model core coefficient is:

(401) searched for threshold signal number and centre frequency thereof:

Nonlinear compensation model core coefficients calculation block is carried out discrete Fourier transform (DFT) (DFT) to the sample data of the n in data cache module successively, search out be positioned at the first Nyquist frequency band and power spectrum be greater than the signal (cross threshold signal) of predefined power spectrum threshold value, be strong signal by this signal decision, record number and the centre frequency thereof of strong signal.

(402) frequency information of calculating nonlinear distortion component:

According to number and the centre frequency thereof of the strong signal of adjudicating in step (401), calculate harmonic wave or the intermodulation frequency value of strong signal, the harmonic wave calculating or intermodulation frequency value are the frequency information of receiver nonlinear distortion component.The top step number of the harmonic wave that need to calculate or intermodulation frequency value equates with the exponent number D of nonlinear compensation model.

The strong signal of being adjudicated taking step (401) is as constant amplitude two-frequency signal x (t) is as example, and the centre frequency of supposing x (t) is f ₁and f ₂, the local frequency of superhet receiver rf front-end is f _l, and have f _l>f ₁, f _l>f ₂, the harmonic wave of signal x (t) or intermodulation frequency value can obtain by formula (1):

f _J＝mf _L+p(f _L-f ₁)+q(f _L-f ₂)，3≤m+p+q≤D (1)

Wherein: f _jfor harmonic wave or the intermodulation frequency value of signal x (t), f _j>0, m, p, q are integer.M, p, q represent harmonic wave or the intermodulation frequency value of signal x (t) under different rank, work as m=1, p=1, q=0 or m=1, p=0, when q=1, the harmonic wave calculating or intermodulation frequency value are the frequency values of the intermediate-freuqncy signal after down-conversion, are not designated as the frequency information of nonlinear distortion component.

(403) generate many passbands frequency domain filter:

The frequency information of the nonlinear distortion component obtaining according to step (402), configuration integrate is passband frequency domain filter more than one, this many passbands frequency domain filter is column vector g, and its dimension is B × 1, and the numerical value of B/2 is greater than or equals the frequency number of nonlinear distortion component.

Step according to many passbands of frequency sampling method configuration integrate frequency domain filter is:

The frequency information of the nonlinear distortion component first obtaining according to step (402) obtains the amplitude-frequency response function of many passbands frequency domain filter, then this amplitude-frequency response function is carried out to inverse discrete Fourier transformer inverse-discrete; Finally the amplitude-frequency response function after inverse discrete Fourier transformer inverse-discrete and window function being multiplied each other, to obtain many passbands frequency domain filter be column vector g.Wherein window function used is the window function that B point extracts, and generally extracts based on breathing out bright (Hamming) window, and non-extraction point is null value.

(404) calculate nonlinear distortion power:

Utilize the to be compensated digital signal of many passbands frequency domain filter after frequency domain is to discrete Fourier transform (DFT) that step (403) generates to carry out many passbands frequency domain filtering, and calculate the power of nonlinear distortion component.The power of nonlinear distortion component is the output data s of many passbands frequency domain filter _fthe quadratic sum of (k, ω), that is:

$\begin{array}{c}{s}_f^2(k,\mathrm{\&omega;})={[y\left(k\right)-V^T\left(k\right)\mathrm{\&omega;}]}^T{\mathrm{gg}}^T[y\left(k\right)-V^T\left(k\right)\mathrm{\&omega;}]\\ =y^T\left(k\right){\mathrm{gg}}^{T}y\left(k\right)-y^T\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\mathrm{\&omega;}\\ -{\mathrm{\&omega;}}^{T}V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)+{\mathrm{\&omega;}}^{T}V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\mathrm{\&omega;}\end{array}---\left(2\right)$

Wherein: ω is the core coefficient vector of nonlinear compensation model, in the present embodiment, be Volterra core vector;

y(k)＝[y(k-B+1)、y(k-B+2)…y(k-1)、y(k)] ^T；

V(k)＝[v(k-B+1)、v(k-B+2)…v(k-1)、v(k)] ^T；

V (k) represents to use y (k) according to the memory nonlinear column vector of each order of nonlinear compensation Construction of A Model, V (k)=[y ²(k), y (k) y (k-1) ... y ²(k-N _d+ 1), y ³(k), y ₂(k) y (k-1) ... y ^d(k-N _d+ 1)] ^t;

N _dfor the corresponding memory depth of d (2≤d≤D) rank Volterra core, T represents matrix transpose computing, and k is discrete time point.

(405) nonlinear compensation model core coefficient vector adaptive iteration calculates:

The nonlinear distortion power that calculation procedure (404) obtains mean-square value, using this mean-square value as iterative target function, taking nonlinear distortion power minimum as criterion, the adaptive iteration that adopts steepest descent method to carry out Volterra core vector ω calculates.Target function J (ω) is:

$\begin{array}{c}J\left(\mathrm{\&omega;}\right)=E\left[s_f^2(k,\mathrm{\&omega;})\right]\\ =E\left[y^T\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\right]-E\left[y^T\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\right]\mathrm{\&omega;}\\ -{\mathrm{\&omega;}}^{T}E\left[V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\right]+{\mathrm{\&omega;}}^{T}E\left[V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\right]\mathrm{\&omega;}\end{array}---\left(3\right)$

Above-mentioned target function J (ω) is that its performance curved surface is hyperparaboloid, has unique globally optimal solution about the quadratic function of Volterra core vector ω.The gradient ▽ [J (ω)] of J (ω) adopts formula (4) to calculate:

▽[J(ω)]＝-2p(k)+2R(k)×ω (4)

Wherein: R (k) is autocorrelation matrix, p (k) is cross-correlation vector, and it is defined as:

$\left\{\begin{array}{c}R\left(k\right)=E[V\left(k\right)\&times;g\&times;g^T\&times;V^T\left(k\right)]\\ p\left(k\right)=E[V\left(k\right)\&times;g\&times;g^T\&times;y\left(k\right)]\end{array}\right.---\left(5\right)$

Make formula (4) equal zero, can obtain the steepest decline iterative formula of optimum Volterra core vector ω:

ω(i)＝ω(i-1)-Γ×{E[V(k)gg ^TV ^T(k)]ω(i-1)-E[V(k)gg ^Ty(k)]} (6)

Wherein: Γ is that N × N ties up diagonal matrix, for upgrading the fixing iteration step length μ of Volterra core vector ω _d;

$\mathrm{\&Gamma;}=\left[\begin{array}{ccccc}{\mathrm{\&Lambda;}}_2& 0& ...& 0& 0\\ ...& ...& ...& ...& ...\\ 0& ...& {\mathrm{\&Lambda;}}_d& ...& 0\\ ...& ...& ...& ...& ...\\ 0& 0& ...& 0& {\mathrm{\&Lambda;}}_D\end{array}\right],{\mathrm{\&Lambda;}}_d=\left[\begin{array}{ccccc}{\mathrm{\&mu;}}_d& 0& ...& 0& 0\\ ...& ...& ...& ...& ...\\ 0& ...& {\mathrm{\&mu;}}_d& ...& 0\\ ...& ...& ...& ...& ...\\ 0& 0& ...& 0& {\mathrm{\&mu;}}_d\end{array}\right]---\left(7\right)$

Wherein: Λ d is the fixing iteration step length for upgrading d rank Volterra core vector, and Λ d is dimension diagonal angle square formation.For ensureing the convergence of steepest descent method, iteration step length μ _dmust meet following restrictive condition:

$0<{\mathrm{\&mu;}}_d<\frac{1}{\mathrm{tr}\left[R\left(k\right)\right]}$ Or $0<{\mathrm{\&mu;}}_d<\frac{1}{(N+1)P_{\mathrm{in}}}---\left(8\right)$

Wherein, tr[R (k)] be the mark of autocorrelation matrix, P _infor the power of signal V (k) g.

In practical application, can be by average the finite length sequence that to ask mathematic expectaion to be converted to length be K, formula (6) is like this:

$\mathrm{\&omega;}\left(i\right)=\mathrm{\&omega;}(i-1)-\mathrm{\&Gamma;}\&times;\left\{\right[\underset{k=k_b(i-1)}{\overset{k_e(i-1)}{\mathrm{\&Sigma;}}}V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)]\&CenterDot;\mathrm{\&omega;}(i-1)-\underset{k=k_b(i-1)}{\overset{k_e(i-1)}{\mathrm{\&Sigma;}}}V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\}---\left(9\right)$

Wherein, k _band k (i) _e(i) represent respectively starting point and the terminating point of the i time iteration data used.Just can calculate Volterra core vector ω, i.e. nonlinear compensation model core coefficient vector by formula (9).

(5) nonlinear distortion is eliminated the nonlinear compensation model core coefficient that computing module utilization receives, digital signal to be compensated is carried out to nonlinear distortion and eliminate in real time computing, digital signal s (k) after being compensated, and be transferred to data interface module.This module is carried out fast fourier transform reconciliation allocation and transportation calculation to the bandpass modulation signal in the digital signal to after compensation simultaneously, the power spectrum data of the digital signal after being compensated and planisphere data, and be transferred to data interface module.The function that nonlinear distortion is eliminated computing module realizes in FPGA.

Utilizing nonlinear compensation model core coefficient to treat compensated digital signal carries out nonlinear distortion and eliminates in real time the process of computing and be:

While using asymmetric discrete Volterra progression to compensate the nonlinear distortion of superhet digital receiver, memory nonlinear Xiang Buhan linear term and second order term, the digital signal s (k) after compensation is:

$s\left(k\right)=y\left(k\right)-\underset{d=3}{\overset{D}{\mathrm{\&Sigma;}}}(\underset{r_1=0}{\overset{N_d-1}{\mathrm{\&Sigma;}}}\&CenterDot;\&CenterDot;\&CenterDot;\underset{r_d=r_{d-1}}{\overset{N_d-1}{\mathrm{\&Sigma;}}}h(r_1,r_2,...,r_d)\underset{j=1}{\overset{d}{\mathrm{\&Pi;}}}y(k-r_j)),3\&le;d\&le;D---\left(10\right)$

Wherein, h (r ₁, r ₂..., r _d) expression d rank time domain Volterra core coefficient.

Volterra core vector ω can be expressed as

ω＝[h(0，0，0) h(0，0，1)...h(N ₃-1，N ₃-1，N ₃-1)h(0，...，0) h(0，...，1)...h(N _D-1，...，N _D-1)] ^T （11）

In conjunction with the expression formula of v (k), formula (10) can be rewritten as:

s(k)＝y(k)-v ^T(k)×ω (12)

Wherein, the dimension of v (k) and ω is N × 1.N can calculate with following formula:

$N=\underset{d=3}{\overset{D}{\mathrm{\&Sigma;}}}\frac{(N_d+d-1)!}{(N_d-1)!d!}---\left(13\right)$

Use form parallel carry out multiplication and the accumulating operation of formula (12) with streamline, just can nonlinear distortion be realized in real time and being eliminated.

(6) data interface module by the digital signal after digital signal to be compensated and power spectrum data thereof and planisphere data, compensation and power spectrum data thereof and planisphere real-time data transmission to host computer, so that observe non-linear real-time eradicating efficacy.The function of data interface module realizes in high-speed PCI-X bridger.

Fig. 3 is the effect contrast figure that antenna receiving signal carries out nonlinear compensation.Antenna receiving signal is carried out after filtering, amplification, down-conversion and analog-to-digital conversion, digital signal to be compensated is that (centre frequency is 18.7MHz with 8PSK signal by three frequency constant amplitude sinusoid signals (frequency values is respectively 12.2MHz, 15.3MHz, 21.8MHz), bandwidth is 1MHz, rolloff-factor α=0.5 of radical sign raised cosine filter) superposition forms, and the constant amplitude three frequently power spectrum of signals and 8PSK signal differs about 65dB.

From the power spectrum of compensation front signal, due to the nonlinear distortion of superhet receiver rf front-end and high-speed AD converter generation, cause 8PSK signal by the intermodulation component severe jamming of multifrequency sinusoid.Now directly this small-signal is carried out to the planisphere that frequency-selecting, filtering, demodulation obtain very messy.Use native system carry out after non-linear real-time elimination, compensation after signal dynamic range (SFDR) by compensation before 58.54dBFS bring up to 78.28dBFS, performance improvement nearly 20dB, nonlinear distortion is effectively suppressed.The Signal to Interference plus Noise Ratio (SINR) of 8PSK signal is greatly improved, and the signal constellation (in digital modulation) figure that final demodulation obtains is rule very, shows that receiver is the in the situation that of large and small signal coexistence, to the detectability of small-signal be improved significantly.

In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment of doing, be equal to replacement, improvement etc., within all should being included in protection scope of the present invention.

Claims (4)

1. the superhet receiving system based on non-linear real-time elimination, comprises the receiver being made up of superhet receiver rf front-end and analog to digital converter; Described heterodyne system receiver rf front-end receives the radiofrequency signal of input, and to the radiofrequency signal receiving amplify, down-conversion and filtering is transferred to analog to digital converter with the form of analog signal after processing; The analog signal receiving is converted to digital signal by described analog to digital converter;

It is characterized in that, also comprise by data cache module, nonlinear compensation model core coefficients calculation block and nonlinear distortion and eliminate the non-linear real-time elimination unit that computing module forms;

The digital signal after conversion is sent to non-linear real-time elimination unit by described analog to digital converter, and this digital signal is digital signal to be compensated; The process that described non-linear real-time unit carries out non-linear real-time elimination to digital signal to be compensated is:

(1) described data cache module is temporarily stored digital signal to be compensated, for nonlinear compensation model core coefficients calculation block provides the sample data of preseting length;

(2) described nonlinear compensation model core coefficients calculation block is carried out the calculating of nonlinear compensation core coefficient, is specially:

(201) successively the sample data in data cache module is carried out to discrete Fourier transform (DFT), then search out and be positioned at the signal that the first Nyquist frequency band and power spectrum are greater than predefined power spectrum threshold value, this signal was threshold signal, recorded number and the centre frequency thereof of threshold signal;

(202) according to number and the centre frequency thereof of crossing threshold signal, calculated harmonic wave or the intermodulation frequency value of threshold signal, the harmonic wave calculating or intermodulation frequency value are the frequency information of receiver nonlinear distortion component; The top step number of the harmonic wave calculating or intermodulation frequency value equates with the exponent number of nonlinear compensation model;

(203) according to many passbands of frequency information configuration integrate frequency domain filter of nonlinear distortion component, this many passbands frequency domain filter is column vector g, and its dimension is B × 1; Wherein the numerical value of B/2 is more than or equal to the frequency number of nonlinear distortion component;

(204) the many passbands frequency domain filter that adopts step (203) to generate carries out filtering to the digital signal to be compensated after discrete Fourier transform (DFT) in to frequency domain, and calculates the power of nonlinear distortion component; The power of described nonlinear distortion component is the quadratic sum of many passbands frequency domain filter output data;

(205) using the mean-square value of nonlinear distortion power as iterative target function, taking " nonlinear distortion power minimum " as criterion, the adaptive iteration that adopts steepest descent method to carry out nonlinear compensation core coefficient calculates, and obtains nonlinear compensation model core coefficient vector;

(3) described nonlinear distortion is eliminated the nonlinear compensation model core coefficient vector that computing module uses step (2) to obtain, and digital signal to be compensated is carried out to nonlinear distortion and eliminate in real time, the digital signal after being compensated.

2. the superhet receiving system based on non-linear real-time elimination as claimed in claim 1, it is characterized in that, adopt asymmetric discrete Volterra progression as nonlinear compensation model, digital signal to be compensated to be compensated,, in described step (204), the power of nonlinear distortion component is:

$\begin{array}{c}{s}_f^2(k,\mathrm{\&omega;})={[y\left(k\right)-V^T\left(k\right)\mathrm{\&omega;}]}^T{\mathrm{gg}}^T[y\left(k\right)-V^T\left(k\right)\mathrm{\&omega;}]\\ =y^T\left(k\right){\mathrm{gg}}^{T}y\left(k\right)-y^T\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\mathrm{\&omega;}\\ -{\mathrm{\&omega;}}^{T}V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)+{\mathrm{\&omega;}}^{T}V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\mathrm{\&omega;}\end{array}---\left(1\right)$

Wherein: s _f(k, ω) is the output data of many passbands frequency domain filter; ω is nonlinear compensation model core coefficient vector; T represents matrix transpose computing; K is discrete time point;

y(k)＝[y(k-B+1) y(k-B+2) … y(k-1) y(k)] ^T,

V(k)＝[v(k-B+1) v(k-B+2) … v(k-1) v(k)] ^T,

V (k) represents to use y (k) according to the memory nonlinear column vector of each order of nonlinear compensation Construction of A Model,

v(k)＝[y ²(k) y(k)y(k-1) … y ²(k-N _d+1) y ³(k) y ²(k)y(k-1) … y ^D(k-N _d+1)] ^T,

N _dfor the corresponding memory depth of d rank Volterra core, 2≤d≤D; D is the exponent number of nonlinear compensation model;

In described step (205), target function J (ω) is:

$\begin{array}{c}J\left(\mathrm{\&omega;}\right)=E\left[s_f^2(k,\mathrm{\&omega;})\right]\\ =E\left[y^T\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\right]-E\left[y^T\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\right]\mathrm{\&omega;}\\ -{\mathrm{\&omega;}}^{T}E\left[V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\right]+{\mathrm{\&omega;}}^{T}E\left[V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)\right]\mathrm{\&omega;}\end{array}---\left(2\right)$

The steepest decline iterative formula of nonlinear compensation model core coefficient:

ω(i)＝ω(i-1)-Γ×{E[V(k)gg ^TV ^T(k)]ω(i-1)-E[V(k)gg ^Ty(k)]} (3)

In formula (3), Γ is that N × N ties up diagonal matrix, for upgrading the fixing iteration step length μ of nonlinear compensation model core coefficient _d:

$\mathrm{\&Gamma;}=\left[\begin{array}{ccccc}{\mathrm{\&Lambda;}}_2& 0& ...& 0& 0\\ ...& ...& ...& ...& ...\\ 0& ...& {\mathrm{\&Lambda;}}_d& ...& 0\\ ...& ...& ...& ...& ...\\ 0& 0& ...& 0& {\mathrm{\&Lambda;}}_D\end{array}\right],{\mathrm{\&Lambda;}}_d=\left[\begin{array}{ccccc}{\mathrm{\&mu;}}_d& 0& ...& 0& 0\\ ...& ...& ...& ...& ...\\ 0& ...& {\mathrm{\&mu;}}_d& ...& 0\\ ...& ...& ...& ...& ...\\ 0& 0& ...& 0& {\mathrm{\&mu;}}_d\end{array}\right]$

Wherein, Λ _drepresent the fixing iteration step length for upgrading d rank nonlinear compensation model core coefficient, Λ _dfor dimension diagonal angle square formation;

By average the finite length sequence that to ask mathematic expectaion to be converted to length be K, formula (3)

$\mathrm{\&omega;}\left(i\right)=\mathrm{\&omega;}(i-1)-\mathrm{\&Gamma;}\&times;\left\{\right[\underset{k=k_b(i-1)}{\overset{k_e(i-1)}{\mathrm{\&Sigma;}}}V\left(k\right){\mathrm{gg}}^T{V}^T\left(k\right)]\&CenterDot;\mathrm{\&omega;}(i-1)-\underset{k=k_b(i-1)}{\overset{k_e(i-1)}{\mathrm{\&Sigma;}}}V\left(k\right){\mathrm{gg}}^{T}y\left(k\right)\}---\left(4\right)$

Wherein, k _band k (i) _e(i) represent respectively starting point and the terminating point of the i time iteration data used; Can calculate nonlinear compensation model core coefficient vector ω by formula (4);

In described step (3), use asymmetric discrete Volterra progression to carry out nonlinear distortion while eliminating in real time to the digital signal to be compensated, the digital signal s (k) after compensation is:

s(k)＝y(k)-v ^T(k)×ω (5)

Wherein, the dimension of v (k) and ω is N × 1,

3. the superhet receiving system based on non-linear real-time elimination as claimed in claim 1, it is characterized in that, described nonlinear compensation model core coefficients calculation block is carried out fast fourier transform to the bandpass modulation signal in digital signal to be compensated and is conciliate allocation and transportation calculation, obtain power spectrum data and the planisphere data of digital signal to be compensated, and send to host computer by data interface module; Described nonlinear distortion is eliminated computing module the bandpass modulation signal in the digital signal after compensating is carried out to fast fourier transform reconciliation allocation and transportation calculation, the power spectrum data of the digital signal after being compensated and planisphere data, and send to host computer by data interface module.

4. the superhet receiving system based on non-linear real-time elimination as claimed in claim 2, is characterized in that, described iteration step length μ _dshould meet:

$0<{\mathrm{\&mu;}}_d<\frac{1}{\mathrm{tr}\left[R\left(k\right)\right]}$ Or $0<{\mathrm{\&mu;}}_d<\frac{1}{(N+1)P_{\mathrm{in}}}$

Wherein, tr[R (k)] be the mark of autocorrelation matrix, P _infor the power of signal V (k) g.