CN104065607A - Method for realizing multi-address channel effective order estimation based on differential evolution algorithm - Google Patents

Abstract

The invention discloses an improved channel effective order and channel impulse response combined estimation method based on a noise channel specific to the unlink multi-address channel of a multi-user OFDM/SDMA (Orthogonal Frequency Division Multiplexing/Space Division Multiple Address) system. According to the method, a framework for performing search channel effective order and channel impulse response combined estimation based on a differential evolution algorithm by taking maximum likelihood as a target function. In the method, an Akaike information criterion is introduced as an evaluation function with an optimal search order in order to increase the estimation accuracies of a channel effective order and channel impulse response. The effectiveness and reliability of the method are verified through simulation. As proved by a result, by introducing the Akaike information criterion, the performance of a time-domain maximum likelihood estimator is improved while the effective channel order estimation error is effectively reduced.

Description

A kind of method that realizes the estimation of access channel effective order based on differential evolution algorithm

Technical field

The present invention relates to the communications field.More specifically, relate to a kind of method that realizes the estimation of access channel effective order based on differential evolution algorithm.

Background technology

Multi-user OFDM/SDMA communication mechanism has been integrated multiple-input and multiple-output (Multiple-Input Multiple-Output, MIMO) and OFDM (Orthogonal Frequency Division Multiplexing, OFDM) two kinds of technical advantages, aspect anti-multipath fading and raising bandwidth efficiency and reliability, there is great potential, it is the key technology in future wireless field, space division multiple access access (Space Division Multiple Access based on MIMO, SDMA) technology is as the effective means that improves mobile communication system capacity, become the study hotspot in this field in recent years.Channel condition information is that MIMO-OFDM system multi-user detects, the necessary condition of decoding etc. when empty accurately, so the accuracy of channel estimating will affect the overall performance of system.

Document 1 " Channel estimation for OFDM systems with transmitter diversity in mobile wirelesschannels " (Li Y, Seshadri N, Ariyavisitakul S, IEEE Journal on Selected Areas inCommunications, 1999, 17 (3): 461-471), document 2 " Two EM-type channel estimation algorithm forOFDM with transmitter diversity " (Xie Y Z, Georghiades C N, IEEE Transactions on WirelessCommunications, 2003, 51 (1): 106-115), and document 3 " Joint decision-directed channel andnoise-variance estimation for MIMO OFDM/SDMA systems based on expectation-conditionalmaximization " (Zhang J K, Hanzo L, Mu X M, IEEE Transactions on Vehicular Technology, 2011, 60 (5): 2139-2151) provided the channel estimation methods based on time domain, conventionally by Channel Modeling, be limited long impulse response filter, ignore the channel exponent number that amplitude is less and can reduce system-computed complexity and the precision that improves channel estimating.Document 4 " Onthe robustness of the linear prediction method for blind channel identification with respect toeffective channel undermodeling/overmodeling " (Liavas A P, Regalia P A, Delmas J P, IEEETransactions on Signal Processing, 2000, 48 (5): 1477-1481) proof, the deviation of supposing channel effective order and actual effective order can affect channel estimating performance, estimation of deviation precision is higher, robustness to noise and interchannel interference is better.Document 5 " A robust maximum likelihood channel estimator for OFDM systems " (Wang Z, MathewG, Xin Y, et al., Wireless Communications and Networking Conference, Hongkong, China:IEEE, 2007:169-174) pointed out that channel effective order is relevant with actual channel exponent number with signal to noise ratio, therefore should adaptively selected channel effective order, make it approach optimal value.In ofdm system, one that addresses this problem is exactly to suppose that channel effective order is for guaranteeing not produce the channel exponent number maximum of subcarrier and intersymbol interference compared with simple scheme, and OFDM Cyclic Prefix (Cyclic Prefix, CP) length adds 1.Although this hypothesis has been simplified problem, in the situation that actual channel exponent number is lower, the remarkable decline of channel estimator performance will be caused.Document 6 " Joint time domain channel and channel lengthestimation for OFDM system " (Jia M, Wang Z Y, Gu X M, Communications, Computers andSignal Processing, Victoria, Canada:IEEE, 2007:605-608) from channel maximum order, reverse search, ignore the channel tap that amplitude is less than certain threshold value and determine channel effective order, but the choose sensitivity of this algorithm to threshold value.Application akaike information criterion (Akaike ' s Information Criterion, AIC) determine when channel is imitated exponent number, do not need to carry out threshold design, with minimization AIC value, determine channel effective order.

At document 7 " Joint symbol timing and channel estimation for OFDM based WLANs " (LarrssonE G, Liu G Q, Li J, et al., IEEE Communications Letters, 2001,5 (8): 325-327), AIC is applied to be intended in channel estimating the information loss that minimum channel estimator causes.Directly application AIC criterion carries out the estimation of channel effective order, need to carry out repeatedly the calculating of channel estimating, FFT/IFFT computing and Euclidean distance, has limited its application in practical communication.Seek a kind of algorithm structure simple, amount of calculation is little relatively for this reason, and practical application easily channel Method of determining the optimum is necessary.Document 8 " Efficient OFDM channel estimation via an information criterion " (Tomasoni A, Gatti D, Bellini S, et al., IEEE Transactions on Wireless Communications, 2013,12 (3): 1352-1362) studied based on AIC single-input single-output ofdm system channel exponent number algorithm for estimating, and build Levinson recursive algorithm reduction operand, there is higher robustness.In multi-user OFDM/SDMA system, a plurality of mobile subscribers that suppose link property inequality combine formation input, each user's transmitted signal experiences separate channel, the channel effective order of link is different, they are equivalent to a unified mixed channel matrix after reception antenna place aliasing, and the channel effective order that now must simultaneously obtain all users just can make ML channel estimator reach best estimated performance.Therefore, how access channel effective order being carried out effectively simultaneously and estimate is fast a significant challenge.

Summary of the invention

The object of the invention is to propose a kind of method that realizes the estimation of access channel effective order based on differential evolution algorithm.The present invention is on the basis of existing technology, for multi-user OFDM/SDMA system, take differential evolution (DifferentialEvolution, DE) algorithm as auxiliary, has proposed the channel effective order parallel search method based on AIC criterion.Multiuser channel effective order and channel impulse response (the Channel impulse response of structure based on AIC criterion, CIR) target function is as the fitness function of DE algorithm, then utilize DE at access channel spatial parallelism search effective order, and carry out CIR estimation.Simulation result shows, the method can better be estimated CIR effective order, improves time domain ML channel estimator performance.

A kind of method that realizes the estimation of access channel effective order based on differential evolution algorithm of the present invention, comprises the steps:

1, initialization

Utilize N _pindividual dimension is the channel effective order vector of U initialization of population as every generation obtains formula (1), and U is number of users, and t represents evolutionary generation, and i is label individual in population, and each individuality in population produces at random in solution space,

Wherein, for on round operation, B _uand B _lrepresent respectively the Searching Resolution Space upper bound and lower bound, B _l=1, B _u=CP, rand (0,1) is illustrated in (0,1) interval interior equally distributed random number;

2, variation

Choose at random two individualities, and carry out the 3rd vector of choosing at random of disturbance with their difference, carry out judgement boundary condition sub-step I and Step II, obtain variation individual

If I. $v{L}_{\mathrm{ij}}^{t+1}<B_L,$ Order $v^1{L}_{\mathrm{ij}}^{t+1}=2B_L-v{L}_{\mathrm{ij}}^{t+1},$ If $v^1{L}_{\mathrm{ij}}^{t+1}>B_U,v^1{L}_{\mathrm{ij}}^{t+1}=B_U;v{L}_{\mathrm{ij}}^{t+1}=v^1{L}_{\mathrm{ij}}^{t+1};$

If II. $v{L}_{\mathrm{ij}}^{t+1}<B_U,$ Order $v^2{L}_{\mathrm{ij}}^{t+1}=2B_U-v{L}_{\mathrm{ij}}^{t+1},$ If $v^2{L}_{\mathrm{ij}}^{t+1}<B_L,v^2{L}_{\mathrm{ij}}^{t+1}=B_L;v{L}_{\mathrm{ij}}^{t+1}=v^2{L}_{\mathrm{ij}}^{t+1};$

Wherein, mutagenic factor F is uniformly distributed in interval (0,1), r ₁, r ₂, r ₃∈ { 1,2, L, N _pdifferent, and not identical with label i yet;

3, intersect

It is individual that the interlace operation of through type (10) generates test increase the diversity of population, rand _j(0,1) ∈ [0,1] is equally distributed random number,

$u{L}_{\mathrm{ij}}^{t+1}=\left\{\begin{array}{cc}v{L}_{\mathrm{ij}}^{t+1}& \mathrm{ran}{d}_j\left(\mathrm{0,1}\right)\&le;\mathrm{CRorj}=\mathrm{randn}\left(i\right)\\ L_{\mathrm{ij}}^t& \mathrm{ran}{d}_j\left(\mathrm{0,1}\right)>\mathrm{CRandj}\&NotEqual;\mathrm{randn}\left(i\right)\end{array}\right.---\left(10\right)$

j＝1,2,LU

Wherein, CR is for intersecting the factor, and randn (i) ∈ [1,2, K, U], is the random dimension variable index of selecting; And

4, select

By formula (11), select operation, generate t+1 for individuality enter of future generation evolution,

$L_i^{t+1}=\left\{\begin{array}{cc}u{L}_i^{t+1}& \mathrm{AIC}\left(u{L}_i^{t+1}\right)<\mathrm{AIC}\left(L_i^t\right)\\ L_i^t& \mathrm{AIC}\left(u{L}_i^{t+1}\right)\&GreaterEqual;\mathrm{AIC}\left(L_i^t\right)\end{array}\right.---\left(11\right)$

The said process that iterates, until meet DE algorithm end condition, optimum during termination be channel effective order estimated value while calculating AIC with corresponding CIR estimated value, the ML that is CIR estimates.

Method of the present invention is through emulation, and result shows, the method can effectively be estimated channel effective order, improves the ML precision of channel estimation of system, improves estimator performance, reduces the bit error rate of system.Compare BER=10 with the scheme of tradition based on fixed channel exponent number ^-4time, the algorithm of carrying approximately can obtain the performance gain of 3dB herein.

Accompanying drawing explanation

Fig. 1 is different E _b/ N ₀under the analogous diagram of channel effective order.

Fig. 2 is the comparison diagram of DE searching algorithm performance under different signal to noise ratios;

Fig. 3 launches the comparison diagram of user NMSE performance under different Method of determining the optimums.

Fig. 4 is the comparison diagram of multi-user OFDM/SDMA system performance of BER curve under different Method of determining the optimums.

Embodiment

In order to solve problem of the present invention, model system model.Suppose that U single antenna user combines formation input, place, base station is used multi-user OFDM/SDMA up-line system of Q root reception antenna.S, place of q root reception antenna OFDM symbol is Y _q[s], q=1, K, Q can be expressed as the stack that different user receives signal and AWGN, that is:

$Y_q\left[s\right]=\underset{u=1}{\overset{U}{\mathrm{\&Sigma;}}}{X}^u\left[s\right]F_q^u{h}_q^u\left[s\right]+W_q\left[s\right]---\left(1\right)$

$0\&le;k\&le;-\mathrm{1,0}\&le;l\&le;l_q^u-1$

From (1) formula, can find out, dimension along with multiple access user's channel effective order changes, so the LS channel estimation method that document 1 proposes has certain limitation.A kind of broad sense expectation-maximization algorithm (SAGE) of space-alternating is proposed in document 2, can be the right Parameter Estimation Problems of some single transmit-reception antennas many antenna transmissions channel decomposing, avoided the matrix inversion operation in LS estimation procedure in document 1, the LS that finally obtains channel impulse response separates.Therefore the present invention will adopt SAGE algorithm to estimate the access channel of channel effective order inequality.

Under AWGN situation, the ML of time domain CIR estimates to be expressed as

$\begin{array}{c}\hat{h}=\\ \underset{h_q^u,(u=1,K,U)}{\arg \max }\frac{1}{{\left(\mathrm{\&pi;}{\mathrm{\&sigma;}}_n^2\right)}^K}\mathrm{gexp}(-\frac{1}{{\mathrm{\&sigma;}}_n^2}{\left|\right|Y_q\left[s\right]-\underset{u=1}{\overset{U}{\mathrm{\&Sigma;}}}{X}^u\left[s\right]F_{K\&times;l_q^u}{h}_{q(l_q^u\&times;1)}^u\left[s\right]\left|\right|}^2)\end{array}---\left(3\right)$

(3) formula that the present invention is based on proposes a kind of new access channel effective order and CIR combined estimation method, the basic thought of method is to utilize DE parallel search channel effective order Joint iteration to carry out CIR estimation, and in DE iterative process, introduce AIC criterion and carry out determining rank, finally show that the ML of access channel effective order and CIR estimates.

Chi Chi great time (H.Akaike) is at document 9 " A new look at the statistical model identification " (AkaikeHirotugu, IEEE Transactions on Automatic Control, 1974,19 (6): 716-723), proposed AIC criterion, for selecting best model at finite aggregate.

$\mathrm{AIC}=-2\ln f\left(y\right|\widehat{\mathrm{\&theta;}}\left(y\right))+2p---\left(4\right)$

for the probability density function of solve for parameter, for the ML estimated value of parameter vector θ (y), p is independent parameter number in θ (y).

At reception antenna q place, according to document [9], the channel effective order estimation problem that can set up multi-user system is:

$\begin{array}{c}\mathrm{AIC}(l_1,l_2,K,l_U)=\\ \ln \left(\frac{1}{M}\underset{m=1}{\overset{M}{\mathrm{\&Sigma;}}}{\mathrm{\&sigma;}}_m^2(l_1,l_{2}K,l_U)\right)+\frac{2}{K}\underset{u=1}{\overset{U}{\mathrm{\&Sigma;}}}{l}_u\end{array}---\left(5\right)$

$(l_1,l_2,K,l_U)=\underset{1\&le;l_u\&le;\mathrm{CP},(u=1,\mathrm{KU})}{\arg \min }\mathrm{AIC}(l_1,l_2,K,l_U)---\left(7\right)$

Wherein, M is the sequence number that transmitting terminal is done pilot tone. it is the residual sum of squares (RSS) of m OFDM symbol of receiving terminal.When access channel effective order is carried out to ergodic search, need to carry out CIR maximal possibility estimation CP^U time, wherein CP is the circulating prefix-length of OFDM symbol.Amount of calculation is large, and complicated operation has limited its application in practice.

A kind of method that realizes the estimation of access channel effective order based on differential evolution algorithm of the present invention, comprises the steps:

1, initialization

Utilize N _pindividual dimension is the channel effective order vector of U initialization of population as every generation obtains formula (1), and U is number of users, and t represents evolutionary generation, and i is label individual in population, and each individuality in population produces at random in solution space,

Wherein, for on round operation, B _uand B _lrepresent respectively the Searching Resolution Space upper bound and lower bound, B _l=1, B _u=CP, rand (0,1) is illustrated in (0,1) interval interior equally distributed random number;

2, variation

Choose at random two individualities, and carry out the 3rd vector of choosing at random of disturbance with their difference, carry out judgement boundary condition sub-step I and Step II, obtain variation individual

If I. $v{L}_{\mathrm{ij}}^{t+1}<B_L,$ Order $v^1{L}_{\mathrm{ij}}^{t+1}=2B_L-v{L}_{\mathrm{ij}}^{t+1},$ If $v^1{L}_{\mathrm{ij}}^{t+1}>B_U,v^1{L}_{\mathrm{ij}}^{t+1}=B_U;v{L}_{\mathrm{ij}}^{t+1}=v^1{L}_{\mathrm{ij}}^{t+1};$

If II. $v{L}_{\mathrm{ij}}^{t+1}<B_U,$ Order $v^2{L}_{\mathrm{ij}}^{t+1}=2B_U-v{L}_{\mathrm{ij}}^{t+1},$ If $v^2{L}_{\mathrm{ij}}^{t+1}<B_L,v^2{L}_{\mathrm{ij}}^{t+1}=B_L;v{L}_{\mathrm{ij}}^{t+1}=v^2{L}_{\mathrm{ij}}^{t+1};$

Wherein, mutagenic factor F is uniformly distributed in interval (0,1), r ₁, r ₂, r ₃∈ { 1,2, L, N _pdifferent, and not identical with label i yet;

3, intersect

It is individual that the interlace operation of through type (10) generates test increase the diversity of population, rand _j(0,1) ∈ [0,1] is equally distributed random number,

$u{L}_{\mathrm{ij}}^{t+1}=\left\{\begin{array}{cc}v{L}_{\mathrm{ij}}^{t+1}& \mathrm{ran}{d}_j\left(\mathrm{0,1}\right)\&le;\mathrm{CRorj}=\mathrm{randn}\left(i\right)\\ L_{\mathrm{ij}}^t& \mathrm{ran}{d}_j\left(\mathrm{0,1}\right)>\mathrm{CRandj}\&NotEqual;\mathrm{randn}\left(i\right)\end{array}\right.---\left(10\right)$

j＝1,2,LU

Wherein, CR is for intersecting the factor, and randn (i) ∈ [1,2, K, U], is the random dimension variable index of selecting; And

4, select

By formula (11), select operation, generate t+1 for individuality enter of future generation evolution,

$L_i^{t+1}=\left\{\begin{array}{cc}u{L}_i^{t+1}& \mathrm{AIC}\left(u{L}_i^{t+1}\right)<\mathrm{AIC}\left(L_i^t\right)\\ L_i^t& \mathrm{AIC}\left(u{L}_i^{t+1}\right)\&GreaterEqual;\mathrm{AIC}\left(L_i^t\right)\end{array}\right.---\left(11\right)$

The said process that iterates, until meet DE algorithm end condition, optimum during termination be channel effective order estimated value while calculating AIC with corresponding CIR estimated value, the ML that is CIR estimates.

For the estimated performance of the assessment algorithm of carrying, examine 2 * 2OFDM/SDMA system herein.With reference to the parameter setting of IEEE802.11n WLAN, the subcarrier number K=64 that each user adopts, circulating prefix-length CP=16.The first two OFDM symbol (M=2) of every frame is pilot frequency sequence, for search channel effective order and CIR, estimates.In actual communication system, different users can adopt different modulation systems, in order to simplify all users of setting, adopts 4-QAM modulation.Channel model adopts multi-path Fading Channel model, and index decline is obeyed in the amplitude fading in each footpath, and phase deviation is uniformly distributed in [0,2 π].In DE operation, population scale N _p=20, mutagenic factor F=0.1, intersection factor CR=0.1.Definition normalized mean squared error function (Normalized Mean Square Error, NMSE) expression formula is

$\mathrm{NMSE}=\frac{E\left\{\underset{k=0}{\overset{K}{\mathrm{\&Sigma;}}}{|{\hat{H}}_{\mathrm{ML}}\left(k\right)-H\left(k\right)|}^2\right\}}{E\left\{\underset{k=0}{\overset{K}{\mathrm{\&Sigma;}}}{\left|H\left(k\right)\right|}^2\right\}}---\left(12\right)$

Wherein, and H (k) represents respectively channel frequency domain transfer function maximum likelihood estimator and actual value.

Fig. 1 is respectively L for supposing two users' transmitting antenna to reception antenna 1 place's actual channel exponent number ₁=5, L ₂=9 o'clock, the AIC performance curved surface under different signal to noise ratios, ordinate is the AIC (L under the combination of different channels exponent number ₁, L ₂) value.In curved surface, make AIC minimum ( ) be channel effective order estimated value.Signal to noise ratio E _b/ N ₀=0dB, E _b/ N ₀=10dB, E _b/ N ₀=20dB and E _b/ N ₀during=30dB, channel effective order is respectively (4,4), (5,7), (5,9) and (5,9).As can be seen from the figure, when hypothesis channel exponent number is less than channel effective order, the marked change of AIC value, on the contrary AIC value tends towards stability, and particularly outstanding under high s/n ratio.When this is because supposes that channel exponent number is less than channel effective order, part active link is left in the basket, and channel estimator can not recover channel condition information accurately.

Fig. 2 is DE searching algorithm performance curve, and ordinate represents the probability that DE algorithm estimation multi-user CIR effective order is these rank.As can be seen from the figure,, along with the increase of signal to noise ratio, channel effective order converges on actual channel exponent number, at E _b/ N ₀during=30dB, converge to the probability nearly 90% of actual channel exponent number.This is because of the increase along with signal to noise ratio, and the probability that each link signal power of channel is greater than noise power increases, thereby makes them become efficient channel exponent number.The transmission user with different channels exponent number, under identical transmitting signal to noise ratio, the probability that the signal power of each link of subscriber channel that channel exponent number is less is greater than noise power is larger, and actual channel exponent number more easily becomes effective order, and the probability being therefore detected is larger.

Fig. 3 is under different signal to noise ratios, uses this paper algorithm of carrying to carry out the NMSE performance curve of channel estimating, and user's actual channel exponent number produces at random.By relatively finding out, during low signal-to-noise ratio, based on institute's algorithm of carrying estimation herein, obtain channel effective order, channel estimating NMSE best performance, along with the increase of signal to noise ratio, the NMSE performance of convergence based under actual channel exponent number.Compare with adopting the scheme of fixed channel exponent number, herein the algorithm of carrying can obtain the performance boost of about 5dB.This is because the algorithm of carrying be take channel effective order as benchmark herein, when low signal-to-noise ratio, ignores the actual channel link that noise power is larger, can suppress better noise jamming, improves the NMSE performance of channel estimating.When high s/n ratio, channel effective order converges on actual channel exponent number, and channel estimating performance converges on the NMSE performance under actual channel exponent number.Fig. 4 has provided under above-mentioned three kinds of schemes, the comparison of system bit error rate (Bit Error Rate, BER) performance.From simulation result, can find out, herein the algorithm of carrying can effectively improve the channel estimating performance of system.BER=10 ^-4time, compare with the scheme of fixed channel exponent number, the algorithm of carrying approximately can obtain the performance gain of 3dB herein.

Claims (1)

1. based on differential evolution algorithm, realize the method that access channel effective order is estimated, comprise the steps:

(1) initialization

Utilize N _pindividual dimension is the channel effective order vector of U initialization of population as every generation obtains formula (1), and U is number of users, and t represents evolutionary generation, and i is label individual in population, and each individuality in population produces at random in solution space,

Wherein, for on round operation, B _uand B _lrepresent respectively the Searching Resolution Space upper bound and lower bound, B _l=1, B _u=CP, rand (0,1) is illustrated in (0,1) interval interior equally distributed random number;

(2) variation

Choose at random two individualities, and carry out the 3rd vector of choosing at random of disturbance with their difference, carry out judgement boundary condition sub-step I and Step II, obtain variation individual

If I. order if

If II. order if

Wherein, mutagenic factor F is uniformly distributed in interval (0,1), r ₁, r ₂, r ₃∈ 1,2 ..., N _pdifferent, and not identical with label i yet;

(3) intersect

It is individual that the interlace operation of through type (10) generates test increase the diversity of population, rand _j(0,1) ∈ [0,1] is equally distributed random number,

j＝1,2,…U

Wherein, CR is the intersection factor, and randn (i) ∈ [1,2 ..., U], be the random dimension variable index of selecting; And

(4) select

By formula (11), select operation, generate t+1 for individuality enter of future generation evolution,

The said process that iterates, until meet DE algorithm end condition, optimum during termination be channel effective order estimated value while calculating AIC with corresponding CIR estimated value, the ML that is CIR estimates.