The maximum likelihood detecting method of low complex degree and device in the communication system
Technical field
The present invention relates to a kind of maximum likelihood detecting method and device thereof, relate in particular to a kind of maximum likelihood detecting method and device that reduces complexity.
Background technology
The defeated MIMO-OFDM of multi-antenna transmitting can greatly increase the capacity of traditional ofdm system and improve the reliability of transmission.Yet the receiving algorithm that the acquisition of diversity and spatial multiplexing gain depends on, the optimum receiving algorithm-maximum likelihood receiving algorithm in the MIMO-OFDM system can be obtained optimum performance.Yet the maximum likelihood receiving algorithm has high complexity, in the detection of mimo system, the exponent number of its complexity and antenna number, symbol-modulated is relevant and be exponential increase with number of antennas, makes this algorithm also can't directly use at the practical MIMO system of median size number of antennas.Lower based on linear non-linear receiving algorithm complexities such as OSIC, the PIC of Interference Cancellation and ZF, MMSE, DFE, it is bigger that but performance and maximum likelihood receiving algorithm are compared gap, especially can not obtain satisfactory performance under the many situations of high order modulation, number of antennas.In order to solve the application bottleneck of optimum receiving algorithm in mimo system, some maximum likelihood receiving algorithms that reduce complexity successively are suggested.Globular decoding (Sphere decoding) algorithm is a kind of symbol detection algorithm of low complex degree, by the dynamic adjustment centre of sphere and radius, search has the symbol sebolic addressing of minimum Eustachian distance in the subspace of maximum likelihood search volume, thereby obtains the performance of approximate maximum likelihood.Spherical algorithm is based on the subspace search, because optimum symbol sebolic addressing might be also might be for the first time to obtain in searching for the last time, therefore the complexity of spherical algorithm is unfixing, and maximum complexity is identical with the maximum likelihood exhaustive search, and especially complexity is still very high when low signal-to-noise ratio; In addition, do not exist simple algorithm to determine spherical radius, often utilize empirical value or experiment value to determine radius.In order to obtain the subsequent channel needed bit soft information of decoding, must obtain a plurality of estimated sequences by (List-Sphere) spherical algorithm and generate log-likelihood ratio.
Consider N
_{T}* N
_{R}The MIMO-OFDM system of configuration, number of sub carrier wave is K, the length N of Cyclic Prefix
_{g}, as shown in Figure 1: in the MIMO-OFDM system, signal sends on the subcarrier of a plurality of antennas, all experiences falt fading channel on each subcarrier, receiver can be on each subcarrier independent detection to recover original information bits.When considering the reception of signal on the receiving terminal n moment k subcarrier, take by force sequence number n and k, many antennas baseband signal can be expressed as
$y=\sqrt{\frac{1}{{N}_{T}}}\mathrm{Hx}+z---\left(1\right)$
Wherein
$y={[{y}_{k}^{1},{y}_{k}^{2},L,{y}_{k}^{{N}_{R}}]}^{T},$ y
_{k} ^{1}, 1≤i≤N
_{T}Be i the received signal on the reception antenna subcarrier k, x is the many antenna transmission signal phasor on the subcarrier k
$x={[{x}_{k}^{1},{x}_{k}^{2},L,{x}_{k}^{{N}_{T}}]}^{T},$ Dimension is N
_{T}* 1, its element is M=2
^{Q}Contrast system symbol
${x}_{k}^{i}\∈S,1\≤i\≤{N}_{T},$ Q is the bit number in each symbol, modulation symbol collection S={ α
_{1}, α
_{1}, L, α
_{M}Through normalization, its average and average symbol energy are respectively
${\mathrm{\Σ}}_{i=1}^{M}{\mathrm{\α}}_{i}=0;$
${E}_{x}=\left({\mathrm{\Σ}}_{i=1}^{M}{\left|{\mathrm{\α}}_{i}\right|}^{2}\right)/M=1$
(1) in
Be the transmitting terminal power normalization factor, make system not change transmitted power because of the increase and decrease of number of antennas.H is N
_{R}* N
_{T}The complex channel matrix of dimension can utilize pilot tone or training sequence to estimate by least square method (LS), least mean-square error methods such as (MMSE) at receiver, at the complete known H of this supposition receiver.Z is the additive white noise vector, and its each element is obeyed the multiple Gaussian Profile of independent identically distributed zero-mean, and variance is σ
_{z} ^{2}
When the distribution situation of many antenna transmission signal when receiving terminal is unknown, optimum receiving algorithm is the maximum likelihood detection algorithm.When receiver adopted the maximum likelihood detection algorithm on each subcarrier, y obeyed the multivariable normal distribution.The bit vectors of supposing symbolic vector x correspondence is
$b={[{b}_{1},{b}_{2},L,{b}_{{N}_{T}}]}^{T},$ N
_{T}Q * 1 dimension, wherein b
_{i}=[b
_{I1}, b
_{I2}, L, b
_{IQ}]
^{T}Be symbol x
_{i}Corresponding bit, then bit b among the b
_{Ij}, 1≤i≤N
_{T}, the log-likelihood ratio of 1≤j≤Q is
$L\left({b}_{\mathrm{ij}}\right)=\mathrm{ln}\left(\frac{\underset{b\∈{B}_{\mathrm{ij}.1}}{\mathrm{\Σ}}\mathrm{Pr}\left(y\right|H,x)}{\underset{b\∈{B}_{\mathrm{ij}.0}}{\mathrm{\Σ}}\mathrm{Pr}\left(y\right|H,x)}\right)---\left(3\right)$
B wherein
_{Ij, 1}, B
_{Ij, 0}Be 2
^{NTQ-1}The set that individual bit vectors b forms, vector b are wherein all satisfied
B
_{ij，1}＝{b|b
_{ij}＝1}B
_{ij，0}＝{b|b
_{ij}＝0}，1≤i≤N
_{T}，1≤j≤Q (4)
(3) the direct calculating of formula relates to the computing of exponential sum logarithm, and is relatively more difficult when actual hardware is realized, adopts MaxLog to be similar to usually and simplify log-likelihood calculations, and (3) can be write as at this moment
$L\left({b}_{\mathrm{ij}}\right)\≈\frac{1}{{\mathrm{\σ}}_{z}^{2}}(\underset{b\∈{B}_{\mathrm{ij},0}}{\mathrm{min}}\left(D\right)-\underset{b\∈{B}_{\mathrm{ij}.1}}{\mathrm{min}}\left(D\right));---\left(5\right)$
$D={\left|\right|y-\sqrt{\frac{1}{{N}_{T}}}\mathrm{Hx}\left|\right|}^{2}$
,, when calculating the log-likelihood ratio information of each bit, need in all possible sequence b, search for even utilize MaxLog to simplify computing as can be seen by (5).Because noise variance is constant in formula, gets maximum respectively and is equivalent to search finding.By 2
^{NTQ-1}The all possible bit sequence b of search in the set of individual sequence, thus can obtain to satisfy the symbol sebolic addressing x of (5), because the length of sequence b is N
_{T}Q, the complexity that soft demodulation maximum likelihood detects increases with the length exponentially of bit vectors b, and when antenna number and modulation order number average greatly the time, the time delay of search and computational complexity are very high.Realize that difficulty is owing to the set or the space of search finding are bigger, reducing the direct method of complexity is exactly the scope that reduces the search volume, searches in a less relatively set, and efficient will improve greatly.Can at first get rid of by (5) and to make the bigger sequence of formula D value, thereby obtain the subclass B ' of B, wherein must comprise the maximum likelihood that sends bit sequence b and separate,
${\hat{b}}_{\mathrm{ML}}=\underset{b\∈B}{\mathrm{arg}\mathrm{min}}D---\left(6\right)$
Be a unique solution, can not be directly used in and calculate bit likelihood ratio (5), so B ' can think to comprise the subclass of the B that maximum likelihood estimates, can not influence the computational accuracy of likelihood ratio in set B ' middle search.Therefore (5) formula can be reduced to
$L\left({b}_{\mathrm{ij}}\right)\≈\underset{b\∈{B}_{\mathrm{ij}.1}^{\′}}{\mathrm{max}}\{-D/{\mathrm{\σ}}_{z}^{2}\}-\underset{b\∈{B}_{\mathrm{ij}.0}^{\′}}{\mathrm{max}}\{-D/{\mathrm{\σ}}_{z}^{2}\}---\left(7\right)$
For determining B ', can at first investigate maximum likelihood and separate
Maximum likelihood is separated (6) to be needed equally by search finding, and for reducing the computational complexity that maximum likelihood detects, channel matrix H can be carried out matrix decomposition
H＝QR (8)
Wherein Q is a unitary matrice, Q
^{H}Q=I, R are upper triangular matrix, and 2 norm computings in (7) can be written as
$D={\left|\right|{Q}^{H}y-\sqrt{\frac{1}{{N}_{T}}}\mathrm{Rx}\left|\right|}^{2}+{\left|\right|(I-{Q}^{H}Q)y\left|\right|}^{2}---\left(9\right)$
Second of equation the right is the estimation that constant does not influence x, can omit when search.Order
$E\left({n}^{H}Q{Q}^{H}n\right)={\mathrm{\σ}}_{n}^{2}{I}_{{N}_{T}},$ Then optimum search is equivalent to
Because the R matrix is a upper triangular matrix, can utilize these characteristics to reduce computing, wherein
Only comprise x
_{NT}Influence, be not subjected to other antenna transmission signal x
_{i}, 1≤i≤N
_{T}Interference,
In not only comprise from N
_{T}-1 antenna x
_{NT-1}Signal also comprise x
_{NT}Interference signal, and can and the like to until
Wherein
Comprise signal from all transmitting antennas.
By second equation of (10), optimal bit or symbol sebolic addressing are equivalent to and make N
_{T}Individual numerical value and minimum also are N
_{T}Individual tolerance and M
_{NT}Minimum,
Part and can be write as recursive form simultaneously
By (12), part and M
_{i}Only depend on symbol x
_{m}, m=i, i+1, L, N
_{T}And a preceding part and M
_{I-1}M wherein
_{NT}Corresponding to the tolerance of whole symbol sebolic addressing x and.
The maximum likelihood sequencal estimation need be N for Q length at status number
_{T}The state transitions grid chart in search for, obtain correspondence metric and M
_{NT}Minimum symbol sebolic addressing is estimated
Simplify the calculating of path metric except utilizing (8) (12), can also wait the status number that reduces search with Viterbi (Viterbi) algorithm, sequence search, globular decoding (Sphere Decoding) algorithm, thereby reduce the complexity and and the operation time of search.Shown in (7), calculating bit soft information need be comprised
Sequence sets, above-mentioned algorithm all can only produce a maximum likelihood sequencal estimation, need to adjust to produce a plurality of sequencal estimations such as List globular decoding (List Sphere Decoding) algorithm, storehouse (Stack) algorithm, M algorithm etc.Yet these algorithms all are based on the search and the computing of symbol sebolic addressing, below propose a kind of simple algorithm (L-MLD) based on linearity test and calculate bit soft information, and approach the maximum likelihood performance.
Summary of the invention
Technical problem to be solved by this invention provides a kind of maximum likelihood detecting method and device of low complex degree, can approach the performance that maximum likelihood detects, and shortcut calculation.
In order to solve the problems of the technologies described above, the technical solution adopted in the present invention is:
A kind of maximum likelihood detecting method of low complex degree is provided, comprises the steps:
Step 1, carry out linearity test, obtain initial valuation of symbol and bits likelihood information in use successively;
Step 2, according to result of linear detection, generate the space of a reduction, the space of this reduction is the subspace of the total space;
Step 3, in the space of reduction, search for, obtain updated bits likelihood information in use, be used for decoding and handle.
Preferably, the method in the space of a reduction of described generation is: carry out declaring than ultrahard the Hamming subspace that obtains reducing according to the value of declaring firmly according to result of linear detection.
Correspondingly, the present invention also provides a kind of maximum likelihood checkout gear of low complex degree, comprising:
The linearity test device, carry out linearity test, obtain initial valuation of symbol and bits likelihood information in use successively;
The subspace generating apparatus, according to result of linear detection, generate the space of a reduction, the space of this reduction is the subspace of the total space;
The likelihood information updating device, in the space of reduction, search for, obtain updated bits likelihood information in use, be used for decoding and handle.
Preferably, described Hamming space generating apparatus comprises:
Declare device, be used for carrying out declaring than ultrahard, the Hamming subspace that obtains reducing according to the value of declaring firmly than ultrahard according to result of linear detection;
Hamming space generating apparatus, be used for the Hamming subspace that obtains reducing according to the value of declaring firmly.
The present invention can obtain extensive use at wireless broadband communication and moving communicating field, especially in the 4th third-generation mobile communication, multi-aerial transmission system, vast potential for future development will be arranged, the relative total space in Hamming subspace that the method according to this invention generates greatly reduces and has very high reliability, and the present invention can reduce the complexity of the maximum likelihood sequence search algorithm of traditional soft output effectively also can approach the maximum likelihood performance; Have the multinomial complexity, the sequence error vector can effectively utilize the bit XOR and generate in advance.The present invention can simplify the multi-aerial receiver algorithm, improves receiver performance.
Description of drawings
Fig. 1 is a MIMO-OFDM transmitter base band theory diagram.
Fig. 2 is the FB(flow block) of maximum likelihood detecting method of the present invention.
Fig. 3 is the channel delay power spectrum.
Fig. 4 is the QPSK modulation, and the OSIC performance of the MIMO-OFDM system of different antennae configuration relatively.
Fig. 5 is the QPSK modulation, and the ML performance of the MIMO-OFDM system during the different antennae configuration relatively.
Fig. 6 is the 16QAM modulation, and the OSIC performance of the MIMO-OFDM system during the different antennae configuration relatively.
Fig. 7 is the 16QAM modulation, and the ML performance of the MIMO-OFDM system during the different antennae configuration relatively.
Fig. 8 is the BER performance of QPSK4 * 4MIMO-OFDM system algorithms of different.
Fig. 9 is the BER performance of 16QAM4 * 4MIMO-OFDM system algorithms of different.
Figure 10 is the BER performance of 64QAM4 * 4MIMO-OFDM system algorithms of different.
Embodiment
As shown in Figure 2, the maximum likelihood detecting method of low complex degree of the present invention comprises the steps:
Step 101, carry out linearity test, obtain initial valuation of symbol and bits likelihood information in use successively;
Step 102, according to result of linear detection, generate the space of a reduction, the space of this reduction is the subspace of the total space; Preferably, the method in the space of a reduction of described generation is: carry out declaring than ultrahard the Hamming subspace that obtains reducing according to the value of declaring firmly according to result of linear detection.
Step 103, in the space of reduction, search for, obtain updated bits likelihood information in use, be used for decoding and handle.
In this specific embodiment, concrete maximum likelihood detecting method is as follows:
For avoiding the high complexity of maximum likelihood search, can obtain N by the linearity test receiver earlier
_{T}The estimation of individual transmission symbol, this moment, bits likelihood information in use need only be calculated in single symbol.According to (1), the Linear Estimation that sends signal is
$\hat{x}=\mathrm{Wy}---\left(13\right)$
When adopting ZF (ZF) detection algorithm and least mean-square error (MMSE) detection algorithm, be respectively
${W}_{\mathrm{ZF}}=\sqrt{{N}_{T}}{\left({H}^{H}H\right)}^{-1}{H}^{H}---\left(14\right)$
${W}_{\mathrm{MMSE}}=\sqrt{{N}_{T}}{({H}^{H}H+{N}_{T}{\mathrm{\σ}}_{n}^{2}I)}^{-1}{H}^{H}$
Being estimated as of i antenna transmission signal
${\hat{x}}_{i}=\sqrt{\frac{1}{{N}_{T}}}({W}_{i}{h}_{i}{x}_{i}+\underset{j\≠i}{\mathrm{\Σ}}{W}_{i}{h}_{j}{x}_{j}+{W}_{i}z)---\left(15\right)$
W wherein
_{i}The i of expression linearity test weight coefficient matrix W is capable, h
_{i}Be matrix
$H=[{h}_{1},{h}_{2},L,{h}_{{N}_{T}}]$ I row, first is x
_{i}The Linear Estimation desired value, second is that other antenna transmission signals are to x
_{i}Interference, last is an equivalent noise.Because linear filter W influences the distribution of noise, send symbol energy and be normalized to 1, equivalent noise variance and interference signal variance are respectively
${\mathrm{\σ}}_{I}^{2}=\underset{j\≠i}{\mathrm{\Σ}}{\left|{W}_{i}{h}_{j}\right|}^{2}/{N}_{T}$
Consider to disturb and The noise symbol x
_{i}J bit b
_{Ij}The soft information of likelihood be
When adopting ZF detection and MMSE to detect, simplifying, can be reduced to respectively
$L\left({b}_{\mathrm{ij}}\right)\≈\underset{x\∈{S}_{j.1}}{\mathrm{max}}\{-\frac{{|\sqrt{{N}_{T}}{\hat{x}}_{i}-x|}^{2}}{{\left|\right|{W}_{i}\left|\right|}^{2}{\mathrm{\σ}}_{z}^{2}}\}-\underset{x\∈{S}_{j.0}}{\mathrm{max}}\{-\frac{{|\sqrt{{N}_{T}}{\hat{x}}_{i}-x|}^{2}}{{\left|\right|{W}_{i}\left|\right|}^{2}{\mathrm{\σ}}_{z}^{2}}\}$
$L\left({b}_{\mathrm{ij}}\right)\≈\underset{x\∈{S}_{j.1}}{\mathrm{max}}\{-\frac{{|\sqrt{{N}_{T}}{\hat{x}}_{i}-{W}_{i}{h}_{i}x|}^{2}}{(\underset{j\≠i}{\mathrm{\Σ}}{\left|{W}_{i}{h}_{j}\right|}^{2}+{\left|\right|{W}_{i}\left|\right|}^{2}{\mathrm{\σ}}_{z}^{2})}\}-\underset{x\∈{S}_{j.0}}{\mathrm{max}}\{-\frac{{|\sqrt{{N}_{T}}{\hat{x}}_{i}-{W}_{i}{h}_{i}x|}^{2}}{(\underset{j\≠i}{\mathrm{\Σ}}{\left|{W}_{i}{h}_{j}\right|}^{2}+{\left|\right|{W}_{i}\left|\right|}^{2}{\mathrm{\σ}}_{z}^{2})}\}---\left(18\right)$
Formula (18) obtains the bits likelihood information in use of linear detector output respectively, this separates and is suboptimal solution, for improving performance, can on this bits likelihood information in use basis, further search for and obtain a subclass B ' who dwindles the B in space, thereby the formula of utilization (7) obtains more excellent sequence bits likelihood information.
Suppose by formula (18) or the bits likelihood information in use that obtains sequence and be
${L}_{0}=[{L}_{1}^{0},{L}_{2}^{0},L,{L}_{i}^{0},L,{L}_{{\mathrm{QN}}_{T}}^{0}],$ Corresponding ratio ultrahard value of declaring is
$\stackrel{\‾}{b}=[{\stackrel{\‾}{b}}_{1},{\stackrel{\‾}{b}}_{2},L,{\stackrel{\‾}{b}}_{i},L,{\stackrel{\‾}{b}}_{{\mathrm{QN}}_{T}}],$ Wherein
${\stackrel{\‾}{b}}_{i}=\left\{\begin{array}{cc}1& {L}_{i}^{0}\≥0\\ 0& {L}_{i}^{0}<0\end{array}\right.---\left(19\right)$
L
_{i} ^{0}The initial soft information that obtains for linearity test,
Declare the ratio ultrahard value of declaring that obtains, N for soft
_{T}Be the transmitting antenna number, the bit number of Q for comprising in each many antenna transmission symbol.Utilize subspace method, the subclass B ' of B may be defined as all with
Hamming distance is less than the set of the bit vectors formation of p, and the element number in this set is much smaller than 2
^{QNT}Bit sequence b
_{i}, b
_{j}Between Hamming distance be defined as
d
_{H}(b
_{i}，b
_{j})＝|{m：b
_{im}≠b
_{jm}，1≤m≤QN
_{T}}| (20)
b
_{Im}Be sequence b
_{i}M position bit.
Wherein | A| represents the radix (cardinality) of set A.This moment set B ' can be write as
${B}^{\′}=\{b\∈B|{d}_{H}(b,\stackrel{\‾}{b})\≤p\}$
B ' can utilize the XOR of bit vectors to realize easily, can make up a sub spaces B ' who is obtained by linearity test thus, and search must obtain the bit likelihood soft information more excellent than formula (17) in the space that this reduces.
$L\left({b}_{\mathrm{ij}}\right)\≈\underset{b\∈{B}_{\mathrm{ij}.1}^{\′}}{\mathrm{max}}\{-{\left|\right|y-\sqrt{\frac{1}{{N}_{T}}}\mathrm{Hx}\left|\right|}^{2}/{\mathrm{\σ}}_{n}^{2}\}-\underset{b\∈{b}_{\mathrm{ij}.0}^{\′}}{\mathrm{max}}\{-{\left|\right|y-\sqrt{\frac{1}{{N}_{T}}}\mathrm{Hx}\left|\right|}^{2}/{\mathrm{\σ}}_{n}^{2}\}---\left(22\right)$
Wherein, L (b
_{Ij}) be i transmitting antenna j bit b
_{Ij}Likelihood information, y is many antennas baseband receiving signals, H is an equivalent channel matrix, x is many antennas baseband transmit signals, σ
_{n} ^{2}Be noise variance.The value of p is big, and the space of set B ' open is just big more, and then the scope of formula (22) search is wide more, and the likelihood information of trying to achieve is reliable more, and complexity is high more, and its limiting case is the maximum likelihood search of total space B.
Correspondingly, the present invention also provides a kind of maximum likelihood checkout gear of low complex degree, comprising:
The linearity test device, carry out linearity test, obtain initial valuation of symbol and bits likelihood information in use successively;
The subspace generating apparatus, according to result of linear detection, generate the space of a reduction, the space of this reduction is the subspace of the total space;
The likelihood information updating device, in the space of reduction, search for, obtain updated bits likelihood information in use, be used for decoding and handle.
Preferably, described Hamming space generating apparatus comprises:
Declare device, be used for carrying out declaring than ultrahard, the Hamming subspace that obtains reducing according to the value of declaring firmly than ultrahard according to result of linear detection;
Hamming space generating apparatus, be used for the Hamming subspace that obtains reducing according to the value of declaring firmly.
Detector of the present invention is divided into two parts to carry out, at first utilize the linearity test device to get the bits of original likelihood information, expansion obtains the solution space B ' of a reduction thus, thereby in this space search bits likelihood information in use, its complexity depends on two-part complexity respectively.Formula (14) needs matrix inversion operation, and complexity is o (N
_{T} ^{3}), formula (21) can effectively utilize the bit XOR does not need main complexity, calculates the size that the required bit sequence number of soft information depends primarily on p, and p gets smaller value and can satisfy computational accuracy in practice.Table 1 is that MLD, List-Sphere and the L-MLD complexity that proposes algorithm compare.
Table 1MIMO-OFDM system detector complexity relatively
The complex multiplication complex addition
MLD N
_{T}N
_{R}Q
^{NT} N
_{T}N
_{R}Q
^{NT}
L-MLD
${N}_{T}{N}_{R}\underset{i=0}{\overset{p}{\mathrm{\Σ}}}\left(\begin{array}{c}Q{N}_{T}\\ i\end{array}\right)+{N}_{T}^{3}$ ${N}_{T}{N}_{R}\underset{i=0}{\overset{p}{\mathrm{\Σ}}}\left(\begin{array}{c}{\mathrm{QN}}_{T}\\ i\end{array}\right)+{N}_{T}^{3}$
List-Sphere N
_{T}N
_{R}P+N
_{T} ^{3} N
_{T}N
_{R}P+N
_{T} ^{3}
As can be seen from the table when the P value hour, algorithm complex is much smaller than MLD, and is similar to the List-Sphere algorithm complex, is the multinomial complexity.
Provide in the MIMO-OFDM system performance and comparison thereof with the lower part based on the L-MLD algorithm.The performance that the algorithm of carrying is superior to traditional receiving algorithm and approaches maximum likelihood.Be the verification system overall performance, the MIMO-OFDM system parameters is as shown in table 2, and adopts the M.1225 outdoor channel PA channel model of middle definition of ITU R.
Fig. 3, Fig. 4, Fig. 5 are respectively OSIC and the comparison of ML detector BER performance in the MIMO-OFDM system when not adding chnnel coding, and modulation system is QPSK and 16QAM, and antenna configurations is respectively 2 * 2,2 * 4,4 * 4,4 * 2.In mimo system, increase the reception antenna number as seen from the figure and can obtain receive diversity raising systematic function, 2 * 4 system all is better than 2 * 2 performance when OSIC and ML detection, 4 * 4 throughput of system is than 2 * 2 system's height, and the BER performance is better than 2 * 2 system too when high s/n ratio.When adopting high order modulation, the performance of OSIC worsens rapidly, in addition when accepting number of antennas and be less than the transmitting antenna number, tradition OSIC detector can not be worked, ML detected and still can obtain better performance this moment, still have good performance and the irreplaceable effect of traditional detection though ML detection algorithm complexity height is described thus, the ML detector that reduces complexity has good application prospect equally.
Table 2MIMO-OFDM system emulation parameter
Parameter value
System bandwidth B 20MHz
Number of sub carrier wave K 2048
Effective number of sub carrier wave N
_{A}1536
Subcarrier spacing Δ f 12.207kHz
Signal duration T
_{U}81.92 μ s
The time T that CP takies
_{CP}18.08 μ s
OFDM symbol period T
_{S}100 μ s
Chnnel coding Turbo code, production (13,15) 8
Channel-decoding MaxLogMap, 8 iteration
Code check 1/3
Modulation system QPSK, 16QAM, 64QAM
Many antenna configurations 2 * 2,4 * 4,2 * 4,4 * 2
Fig. 6, Fig. 7, Fig. 8 are that modulation system adopts QPSK, 16QAM and 64QAM respectively in 4 * 4 MIMO-OFDM systems, and when chnnel coding was Turbo Code, the performance of BER of the algorithm of proposition and MMSE OSIC, List-Sphere, ML detection algorithm relatively.The algorithm of Ti Chuing can obtain relatively the gain based on the OSIC algorithm 1dB of MMSE when p=2 as seen from the figure, performance is further enhanced when increasing the value of p, the error vector number that generates during p=5 among Fig. 6 is that 219 its complexities are approximate identical with the List-Sphere algorithm complex of L=256, both performance differences are very little, approximately 0.5dB approaches the performance of ML, compares the gain that 1.9dB is arranged with MMSE OSIC.Among Fig. 8 during high order modulation, the error vector number sharp increase that generates when increasing the value of p, for reducing computational complexity, set maximum number and be 1024 identical with alternative sequence number in the List-Sphere algorithm, propose algorithm as can be seen by emulation and have identical performance with the List-Sphere algorithm, with the only poor 0.7dB of the performance of ML, 1dB, compare the gain that 2.1dB, 2.7dB are arranged respectively with MMSE OSIC.