CN1381995A - Method for implementing 8PSK equalizing demodulation in EDGE system - Google Patents

Abstract

A method for equilibrium demodulation of 8PSK for EDGE system features that the gradual convergence is performed several times to choose the sequence with maximal probabal function value as output sequence from the possible sequences. It can make that the improve MLSE receiver can be used in EDGE system.

Description

A kind of 8PSK equalizing demodulation implementation method that is applicable to the EDGE system

The invention belongs to the mobile communication technology field, relate in particular to the 8PSK equalizing demodulation implementation method of a kind of EDGE of being applicable to system.

In gsm mobile communication system, consider serious multipath fading, adopt the equalizing demodulation device of better performances usually.Maximum-likelihood sequence estimation equilibrium (MLSE) is exactly wherein be most widely used a kind of.The application of follow-on Viterbi algorithm in MLSE is improved its performance, and operand also significantly reduces simultaneously.The Viterbi algorithm is the coding method of convolution code that A.J.Viterbi proposed in 1967.This algorithm operation quantity is little, the error rate is low, has caused people's great attention and broad research.Forney was used for balanced the reception with the Viterbi algorithm in 1972, had proposed a kind of structure of attainable maximum likelihood equalizer.Since then, adopt the balanced receiver of Viterbi algorithm to be widely used in the communications field.

In the process of 3-G (Generation Three mobile communication system) development,, EDGE (Enhanced Data rates for Global/GSMEvolution) technology has been proposed at gsm mobile communication system in order to support the higher rate transfer of data.The EDGE technology is to change the GMSK modulation demodulation system into 8PSK modulation efficiently on the basis of GSM technology, cooperates existing GPRS technology, can realize higher data transmission capabilities.At the 8PSK modulation mode of present proposition, if prolong implementation method with the MLSE receiver of original modified model Viterbi, its computation complexity is 8 ^L-1(L is a disperse length, is commonly considered as 5 or 6 in the gsm system), system can't realize at all.

The objective of the invention is,, propose a kind of equalizing demodulation implementation method of the 8PSK modulation that adapts with it, adopt this method, make that computation complexity is reduced to N under the prerequisite that does not influence the equalizing demodulation performance at the EDGE technology in the 3-G (Generation Three mobile communication system) _d* 3 * 2 ^L-1(N _dBe iterations, be generally 1 or 2).That is to say that equalizing demodulation implementation method provided by the invention makes that the MLSE receiver of modified model Viterbi can be used in the EDGE mobile communication system, and amount of calculation is little, system's realizability is strong.

The concrete implementation method of the present invention may further comprise the steps:

A. the digital sampled signal that receives is at first carried out despining;

B. utilize local known training sequence signal and received signal to carry out related operation, seek maximum correlation value and finish synchronously, and obtain the Timing Advance parameter;

C. according to the training sequence of described Timing Advance parameter, received signal and the signal after the despining, adopt related operation to carry out channel estimating, obtain channel parameter;

D. according to described channel parameter, the signal after making it to despining carries out matched filtering, carries out coherent demodulation again according to described result, and judges the symbol after the demodulation;

E. utilize the data after channel parameter and the described step d) coherent demodulation, carry out N _dInferior iteration route searching.

The present invention restrains by multiple gradual, from all possible sequence, select have the maximum likelihood function value sequence as output sequence, relate to modified model Viterbi algorithm, below this is further specified.

Maximization cost function J _n({ r _n): $J_n\left(\right\{r_n\left\}\right)=2\mathrm{Re}\left(\underset{\mathrm{nT}\∈I}{\mathrm{\Σ}}{r_n}^{*}{Z}_n\right)-\underset{\mathrm{nT}\∈I}{\mathrm{\Σ}}\underset{\mathrm{mT}\∈I}{\mathrm{\Σ}}{r_n}^{*}{r}_m{S}_l---\left(1\right)$

Its recursive function is: $J_n\left(\right\{r_n\left\}\right)=J_{n-1}\left(\right\{r_{n-1}\left\}\right)+2\mathrm{Re}\left[{r_n}^{*}(Z_n-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{r}_{n-l}{S}_l)\right]---\left(2\right)$

The parameter in equation (1) and (2) wherein: $Z_n=\underset{t\∈I}{\∫}y\left(t\right)h^{*}(t-\mathrm{nT})\mathrm{dt}---\left(3\right)$ $S_l=\underset{t\∈I}{\∫}{h}^{*}(t-\mathrm{nT})h(t-\mathrm{mT})\mathrm{dt}---\left(4\right)$

Improved Viterbi algorithm has utilized the recurrence relation of equation (2) to obtain the sequence of maximum likelihood function value, and detailed process is as follows:

(1) from node K=L, the length of calculating each state correspondence is the branch tolerance of L, and computing formula is referring to above-mentioned (1) formula.

Store all part paths and part path metric, as the survivor path and the survivor path tolerance of k=L node.

(2) k=k+1 enters all branch tolerance of each state to (k+1) constantly and the survivor path of the previous moment that links to each other with these branches is measured addition, to each state, from N _LSelect a path with maximum metric in the bar access path as the new survivor path tolerance of this state, this paths is just as new survivor path.State S _{I, k+1}Survivor path metric calculation formula as follows: $J_n\left(\right\{r_n\left\}\right)=\underset{\left\{S_{n-1}\right\}\→S_i}{\mathrm{MAX}}\{J_{n-1}\left(\right\{r_{n-1}\left\}\right)+2\mathrm{Re}[{r_n}^{*}(Z_n-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{r}_{n-l}{S}_l)\left]\right\}---\left(5\right)$

(3) if k＜Length (Length is a list entries length) then turns back to (2); Otherwise, stop iterative process, the sequence measurement of respectively surviving, get have maximum metric the survival sequence as Maximum likelihood sequence { r _n, stop.

If directly adopt modified model Viterbi algorithm once to restrain the sequence that finds the maximum likelihood function value, consider symbol r _nEight kinds of states are arranged, the computational complexity N in the said process _LEqual 8 ^L-1, (wherein L is a disperse length), the complexity of this method reaches 8 ^L-1, system is difficult to realize.By research 8PSK sign map figure, adopt the method for multiple gradual convergence, find the sequence of maximum likelihood function value, thereby reduce the complexity of computing greatly.

Below, with reference to accompanying drawing, for those skilled in the art that, from the detailed description to the inventive method, above-mentioned and other purposes of the present invention, feature and advantage will be apparent.

Fig. 1 provides the flow chart of equalizing demodulation implementation method of the 8PSK modulation of the EDGE of being applicable to provided by the invention system;

Fig. 2 represents through postrotational symbol r _i(a _3i, a _3i+1, a _3i+2) mapping graph;

Fig. 3 represents the signal to noise ratio under the channel TU50 and the simulated effect figure of the error rate.

Please refer to shown in Figure 1ly, digital I, Q signal through over-sampling obtains at first carry out despining; Carry out then synchronously, it is relevant to utilize I ', Q ' after known training sequence and the despining to do, and obtains Timing Advance parametric t oa; Then carry out channel estimating again, utilize the toa obtain, training sequence relevant, carry out channel estimating, obtain channel parameter hi and S with I ', the realization of Q ' signal after the despining _LAgain channel parameter hi is passed to matched filter, I ', Q ' signal after making it to despining carry out matched filtering; Its result imports coherent demodulator, carries out coherent demodulation; Carry out N at last _dInferior iteration route searching judgement utilizes parameter S _L, the data after the coherent demodulation, carry out N _dInferior iteration and judgement output.

Respectively each step is elaborated below.

According to GSM05.04 (V8.0.0) agreement, to the regulation of 8PSK modulation mode among the EDGE, the baseband modulation process can be divided into three steps: 1) simple 8 phase modulated; 2) symbol rotation; 3) forming filter C0 (t) filtering.Therefore, after the new modulating vector that obtains through the 1st, 2 steps is the forming filter filtering of C0 (t) by impulse response, just can obtain the baseband modulation signal of 8PSK.The equivalent model of this baseband modulation signal is: $y_k=\underset{i}{\mathrm{\Σ}}{e}^{j2\mathrm{\π}{l}_{k+1}/8}\×e^{j3\mathrm{\π}(k+i)/8}\×h_i+n_k---\left(6\right)$

Wherein: h _iChannel transfer functions;

n _kBe additive Gaussian noise;

l _iBe symbolic parameter.

In order to reduce back level complex operation, at first carry out despining: $y_k^{\′}=y_k{e}^{-j3\mathrm{k\π}/8}=\underset{i}{\mathrm{\Σ}}{e}^{j2\mathrm{\π}{l}_{k+i}}{h}_i^{\′}+n_k^{\′}---\left(7\right)$

Promptly $y_k^{\′}=y_k(\cos \frac{3\mathrm{k\π}}{8}-j\sin \frac{3\mathrm{k\π}}{8})---\left(8\right)$

Launch: $I^{\′}=I\cos \frac{3\mathrm{k\π}}{8}+Q\sin \frac{3\mathrm{k\π}}{8}---\left(9\right)$ $Q^{\′}=Q\cos \frac{3\mathrm{k\π}}{8}-I\sin \frac{3\mathrm{k\π}}{8}---\left(10\right)$

Wherein I, Q are I, the Q signal before the despining; I ', Q ' are I, Q signal after the despining.

Timing in the digital communication is extremely important synchronously, the synchronous whether accurate quality that is directly connected to system's receptivity.The EDGE system utilizes training sequence to finish regularly synchronously.At first, carry out phase mapping (referring to modulating part) to training sequence, obtain the real part TSCr and the imaginary part TSCi of training sequence.TSCr has good autocorrelation performance and their cross correlation, and TSCi is 0 sequence.

Training sequence after the despining also has identical autocorrelation performance, therefore, can carry out related operation with local known training sequence signal and received signal I ', Q ', seeks maximum correlation value and finishes synchronously.Concrete steps are as follows:

Calculate the auto-correlation of reference signal and received signal.Reference signal is the centre 16 bit sampling sequences of the 8PSK baseband signal of 26 bit training sequences generation, be designated as D (n), n=0,1 ..., 15}, correlation function is as follows: $R_j\left(k\right)=\underset{n=0}{\overset{15}{\mathrm{\Σ}}}D\left(n\right)d_j(n+k+N)---\left(11\right)$ { d _j(n) } be I ', Q ' signal after the despining:

d _i(n)＝d(nT _b+jT _b/i+t ₀) j＝0，1，…，i-1 (12)

T wherein _bBe symbol width, t ₀Be initial sampling phase, T _b/ i represents sampling time interval, and i generally is taken as 2 or 4.Select the feasible receiving sequence sample value of carrying out correlation computations of suitable N as far as possible near the middle part of sequence, wherein the value of k should satisfy | k|≤5.

In all j, k, select j ', k ' to make the sampled value of received signal mid portion and the correlation function of local reference signal sample value reach maximum, that is: $\left|R_j\left(k^{\′}\right)\right|=\underset{j,k}{\max }\left|R_j\left(k\right)\right|---\left(13\right)$ At this moment:

toa＝k′×i+j′ (14)

Channel estimating provides filter factor for matched filter, and the parameter of metric calculation is provided for MLSE simultaneously.In the EDGE system, for receiving terminal, 26 symbol training sequences in the middle of each burst of transmission are known, thereby can carry out correlation computations with receiving sequence sample value and known array, estimate the channel response parameter.Because coherence time of channel is much larger than the duration of 1 burst, thereby, can think that the variation of channel in single burst is very little, so the channel estimating of training sequence position is applicable to whole burst.

Channel estimating adopts least square method.Least square method is a kind of simple method, and it seeks the argument sequence { h of error sum of squares ε minimum _iAs estimation to channel impulse response. $\mathrm{\ϵ}=\underset{k=5}{\overset{20}{\mathrm{\Σ}}}|d_{j^{\′}}(k^{\′}+N+k-5)-\underset{n=-L}{\overset{L}{\mathrm{\Σ}}}{D}_{k-n}{h}_i|---\left(15\right)$ $h_i=\frac{1}{16}\underset{k=5}{\overset{20}{\mathrm{\Σ}}}{D_{k-i}d}_{j^{\′}}(k^{\′}+k+N-5)i=-L,-1,\·\·\·L---\left(16\right)$

Wherein, { d _j(n) } be I, Q signal after the despining.N is the initial offset of related operation, is that N in the equation (11) is identical with synchronization module, and j ', k ' try to achieve at synchronization module.{ D _kBe known training sequence, L is the disperse length of channel.

For avoiding when matched filtering and Viterbi are balanced computing excessive, and consider that algorithm performance is no influence substantially, from 5 continuous sampling point h of channel parameter extraction of these sample points _Bt, h _Bt+1, h _Bt+2, h _Bt+3, h _Bt+4, make the energy and the maximum of these 5 sampling points.And they tap coefficients as matched filter. $\mathrm{Max}\_h_j=\underset{i=j}{\overset{j+5}{\mathrm{\Σ}}}\left|h_i\right|j=-L,\·\·\·L-5---\left(17\right)$ In all j, find and make Max_h _jMaximum j ', bt=j ' at this moment.The autocorrelative calculating of filter response:

S ₁=∫ h ^*(t-nT) h (t-mT) dt (18) is through the h after the channel estimation process _iSignal and received signal, carry out the output after the matched filtering: ${\mathrm{mout}}_q=\underset{i=0}{\overset{4}{\mathrm{\Σ}}}{D}_{q-i}(k-i)\×h_{\mathrm{bt}+i}---\left(19\right)$

Wherein q represents matched filter output sequence number.Matched filter output (I+jQ) is carried out the rotation of π/8 earlier, its postrotational graphical diagram (I _d+ jQ _d) be illustrated in fig. 2 shown below, adjudicate according to following criterion:

(1) for Q _dIf signal is Q _d＞0, a then _3i=0, otherwise a _3i=1;

(2) for I _dIf signal is I _d＞0, a then _3i+1=1, otherwise a _3i+1=0;

(3) if I _dCos π/4+Q _dSin π/4 and I _dCos π/4-Q _dSin π/4 are all greater than 0, or all less than 0, then a _3i+2=0, otherwise a _3i+2=1.

Consider each symbol r _i(a _3i, a _3i+1, a _3i+2) eight kinds of states are arranged, so when search, carry out according to following steps:

A, a that goes out with coherent demodulation _3i+1, a _3i+2Substitution r _i(a _3i, a _3i+1, a _3i+2) in, adopt modified model Viterbi algorithm to a _3iSearch for judgement.

Detailed process is described in front, at this moment computational complexity N _LEqual 2 ^L-1, computational complexity is 2 ^L-1{ the r that is declared in utilization _n, rule out a _3i

B, a that goes out with coherent demodulation again _3i+2, a that rules out by the A step _3iSubstitution r _i(a _3i, a _3i+1, a _3i+2) in, to the similar employing modified model of A step Viterbi algorithm to a _3i+1Search for judgement.This step computational complexity also is 2 ^L-1

The a that C, utilization rule out _3iAnd a _3i+1Substitution r _i(a _3i, a _3i+1, a _3i+2) in, similar to the A step to a _3i+2Adjudicate.This step computational complexity also is 2 ^L-1

The a that D, utilization rule out _3i+1And a _3i+2, turn back to the A step and adjudicate.

So recurrence is searched for N _dInferior, its computational complexity is 2 ^L-1* 3 * N _dThrough the emulation explanation, work as N _d=1, the time, just restraining, the error rate of court verdict just satisfies the requirement of regulation.

Fig. 3 provides one for the TU50 channel, its signal to noise ratio and error rate k _SER(N _dThe simulated effect figure of relation=1).When the code requirement signal to noise ratio is 9dB, under the TU50 channel situation, two class bit error rate k _SER=8%.The simulation value k that adopts this method to obtain _SER=3.3506%, reached code requirement.

In addition, become the intersymbol interference that multipath fading causes when this equalizing demodulation implementation method has solved in the EDGE mobile communication system, and showed good performance.Tentatively prove through simulation software: for the ideal communication channel that does not have decline, noise, its two classes bit error rate k _SER=0.051, adopted back level N _dInferior alternative manner, its two classes bit error rate k _SER=0 (N _d=1).

The front provides the description to preferred embodiment, so that any technical staff in this area can use or utilize the present invention.It is conspicuous that the kind of these embodiment is revised those skilled in the art, can be applied to total principle described here each other embodiment and does not use creativeness.Thereby, the embodiment shown in the present invention will be not limited to here, and the wide region of principle that should disclose and new feature according to meeting here.

Claims (18)

1. 8PSK equalizing demodulation implementation method that is applicable to the EDGE system, by the multiple gradual convergence, from all possible sequence, select have the maximum likelihood function value sequence as output sequence.

2. method according to claim 1 is characterized in that, said method comprising the steps of:

A. the digital sampled signal that receives is at first carried out despining;

B. utilize local known training sequence signal and received signal to carry out related operation, seek maximum correlation value and finish synchronously, and obtain the Timing Advance parameter;

C. according to the training sequence of described Timing Advance parameter, received signal and the signal after the despining, adopt related operation to carry out channel estimating, obtain channel parameter;

D. according to described channel parameter, the signal after making it to despining carries out matched filtering, carries out coherent demodulation again according to described result, and judges the symbol after the demodulation;

E. utilize the data after channel parameter and the described step d) coherent demodulation, carry out N _dInferior iteration route searching, described iterations N _dBe 1 or 2.

3. method according to claim 2 is characterized in that, described despining a) step is carried out according to following formula: $y_k^{\′}=y_k{e}^{-j3\mathrm{k\π}/8}=\underset{i}{\mathrm{\Σ}}{e}^{j2\mathrm{\π}{l}_{k+i}}{h}_i^{\′}+n_k^{\′}$

Promptly $y_k^{\′}=y_k(\cos \frac{3\mathrm{k\π}}{8}-j\sin \frac{3\mathrm{k\π}}{8})$

4. method according to claim 2, it is characterized in that, described synchronizing step b) training sequence need carry out phase mapping with it in modulating part, obtain the real part (TSCr) and the imaginary part (TSCi) of training sequence, described real part (TSCr) has good auto-correlation and their cross correlation, and described imaginary part (TSCi) is 0 sequence.

5. method according to claim 2 is characterized in that, described step b) is according to correlation function: $R_j\left(k\right)=\underset{n=0}{\overset{15}{\mathrm{\Σ}}}D\left(n\right)d_j(n+k+N)$

K=0,1,2 ... 15, N is the related operation initial offset.

Calculate the auto-correlation of reference signal and received signal.

6. method according to claim 5 is characterized in that, D (n) is the centre 16 bit sampling sequences of the 8PSK baseband signal of 26 bit training sequences generation.

7. method according to claim 5 is characterized in that, described d _j(n) be the received signal after the despining:

d _j(n)＝d(nT _b+jT _b/i+t ₀) j＝0，1，…，i-1

8. method according to claim 7 is characterized in that, described t ₀Be initial sampling phase, T _bBe symbol width, T _b/ i represents sampling time interval.

9. method according to claim 7 is characterized in that i is generally 2,4.

10. method according to claim 5 is characterized in that, selects suitable related operation initial offset N, makes the receiving sequence sample value of carrying out related operation try one's best near the middle part of described sequence.

11., it is characterized in that according to claim 5 or 7 described methods, in all j, k, select to make the sampled value of received signal mid portion and the correlation function of local reference signal sample value to reach maximum, promptly satisfy: $\left|R_j\left(k^{\′}\right)\right|=\underset{j,k}{\max }\left|R_j\left(k\right)\right|$

12. method according to claim 2 is characterized in that, the channel estimating of step c) adopts least square method, promptly $\mathrm{\ϵ}=\underset{k=5}{\overset{20}{\mathrm{\Σ}}}|d_{j^{\′}}(k^{\′}+N+k-5)-\underset{n=-L}{\overset{L}{\mathrm{\Σ}}}{D}_{k-n}{h}_i|$

Wherein, { D _kBe known training sequence, L is the disperse length of channel.

13. method according to claim 12 is characterized in that, argument sequence h _iFor:

I=-L ,-1 ... L

14. method according to claim 12 is characterized in that, described channel dispersion length L is generally 5 or 6 in gsm system.

15. method according to claim 13 is characterized in that, from L argument sequence of channel parameter continuous drawing, makes the energy and the maximum of a described L sampling point.

16. method according to claim 2 is characterized in that, described determining step d) comprising:

A ') if the Q after the matched filtering _d＞0, a then _3i=0, otherwise a _3i=1;

B ') if the I after the matched filtering _d＞0, a then _3i+1=1, otherwise a _3i+1=0;

C ') if (I _dCos π/4+Q _dSin π/4) with (I _dCos π/4-Q _dSin π/4) all greater than 0, or all less than 0, a then _3i+2=0, otherwise a _3i+2=1.

17. method according to claim 2 is characterized in that, described iterative search step e) comprises:

The a that a ") goes out with coherent demodulation _3i+1, a _3i+2Substitution r _i(a _3i, a _3i+1, a _3i+2) in, adopt modified model Viterbi algorithm to a _3iSearch for judgement;

The a that b ") goes out with coherent demodulation again _3i+2, a that utilizes described step a ") to rule out _3iSubstitution r _i(a _3i, a _3i+1, a _3i+2) in, adopt modified model Viterbi algorithm to a _3i+1Search for judgement;

C ") is according to a that rules out _3iAnd a _3i+1Substitution r _i(a _3i, a _3i+1, a _3i+2) in, employing and a ") same steps as are to a _3i+2Adjudicate;

D ") utilizes a that rules out _3i+1And a _3i+2, return step a ") and adjudicate.

18. method according to claim 2 is characterized in that, described channel parameter comprises filter factor (h _i), and the parameter (S that metric calculation is provided for MLSE _L).

Images