CN105656538A - Low-complexity belief propagation detection algorithm for large-scale MIMO system - Google Patents

Abstract

The invention discloses a low-complexity belief propagation detection algorithm for a large-scale MIMO system. A corresponding factor graph is built by utilizing an equivalent real number field model, and complex number field operation is translated into real number field operation, thereby implementing BP-based iteration detection; wherein the factor graph is used for representing a dependency relation between a receiving signal and a transmitting signal, the transmitting signal is utilized as a signal node, and the receiving signal is utilized as an observation node; each signal node updates prior information according to posterior information obtained from the observation node, and then transmits the updated prior information to all observation nodes connected with the signal node; each observation node calculates updates posterior information according to prior information obtained from the signal node, and then transmits the updated posterior information to a signal node connected with the observation node. According to the low-complexity belief propagation detection algorithm for the large-scale MIMO system, a symbol-based large-scale MIMO detection algorithm is implemented; high-dimensional matrix inversion is avoided, and the low-complexity belief propagation detection algorithm can be greatly suitable for application scenarios of the large-scale MIMO.

Description

A kind of low complex degree belief propagation detection algorithm of extensive mimo system

Technical field

The invention belongs to next generation wireless communication technical field, apply in extensive multiple antennas (MIMO) system of any antenna configurations and order of modulation, the high performance detector of uplink base station receiver.

Background technology

The modernization development of wireless communication technology starts from the nineties in 20th century, and its development speed is always in upward status, and its development scale is in continuous expansion, and the scope used is also more and more extensive. Development along with computer networking technology, what wireless communication technology was positive utilizes information technology to promote developing of self, the scientific and technological content and the technical merit that make wireless communication technology are obtained for very big lifting, enable wireless communication technology better to apply to more field, the function diversification that enters society, network integration, service synthesization developmental stage. Mobile communication develops towards the direction of two-forty, high power capacity, spectral efficient and low-power consumption, constantly meets the growing data of people and video requirement. According to major carrier and authority's advisory organization prediction: mobile broadband service flow will increase by 1000 times at coming 10 years. Existing 4G technology still cannot meet the demand in future in transfer rate and resource utilization etc., and its wireless coverage and Consumer's Experience also need to be improved further. Countries in the world are while promoting 4G industrialization work, and the 5th third-generation mobile communication technology (5G) has had become as the study hotspot of domestic and international wireless communication field.

Multiaerial system, by transmitting multichannel data in identical frequency band concurrently, substantially increases power system capacity and data rate. MIMO, with its higher spectrum efficiency and connection reliability, receives much concern in research and industrial circle. Along with the continuous growth of future mobile communications demand, the antenna scale that system needs is also increasing, and extensive mimo system arises at the historic moment. Radio Transmission Technology based on extensive MIMO makes spectrum efficiency and power efficiency promote a magnitude on the basis of 4G again, will become the key technology of next generation mobile communication. At present, the study hotspot of the communications field has been become for the detection algorithm of extensive MIMO.

S.Haykin, M.Sellathurai, Y.deJong, andT.Willink, " Turbo-MIMOforwirelesscommunications; " IEEECommun.Mag., vol.42, no.10, pp.48 53, Oct.2004. a kind of method disclosing real-valued decomposition, the method that specifically complex field is decomposed into real number field.

Desirable MAP detection, its computation complexity is in exponential increase along with the increase of transmitting antenna number, is not suitable for extensive MIMO scene. In pertinent literature about MIMO detection, there is a large amount of correlational study for low complex degree detection algorithm. The detection method of approximate ideal, K-Best and globular decoding (SD) algorithm, its performance and complexity such as soft output still suffer from challenging under extensive MIMO scene. Based on least mean-square error (MMSE) detection that Neumann progression is approximate, greatly reduce the computational complexity of detection. But, approximate error is relevant with the configuration of antenna number, thus limiting its range of application. Therefore, under the detection performance premise ensureing approximate ideal, the extensive MIMO detection algorithm designing a kind of low complex degree is particularly important.

Summary of the invention

Goal of the invention: for solving signal detection difficult point in extensive mimo system, the present invention provides the low complex degree belief propagation detection algorithm of a kind of extensive mimo system, the present invention suitable in extensive mimo system, the signal detection of uplink base station receiver. The BP detection algorithm based on symbol of extensive mimo system, its detection performance is better than MMSE detection algorithm. Additionally, transmitting antenna number square with ratio (TSRR) M of reception antenna number²When/N is very big, MMSE detection algorithm is no longer applicable, and the BP algorithm that the present invention proposes is appointed so applicable, and avoids higher dimensional matrix inversion operation such that it is able to be well adapted for the scene of extensive MIMO.

Technical scheme: for achieving the above object, the technical solution used in the present invention is:

The low complex degree belief propagation detection algorithm of a kind of extensive mimo system, in extensive mimo system, uses the real number model of its equivalence to build corresponding factor graph, complex field computing is converted into real number field computing, it is achieved based on the iterative detection of BP; Wherein, factor graph is used for representing reception signal and launching dependence between signal, will launch signal as signal node, and receive signal as observer nodes; Each signal node updates prior information according to the posterior information obtained from observer nodes, is then transferred to all observer nodes being attached thereto; Each observer nodes calculates posterior information according to the prior information from signal node, is then passed back to the signal node being attached thereto; Thus, complete renewal and the transmittance process of a message; After the iteration more than once, observer nodes output Soft Inform ation is used for estimating to launch symbol;

In the real number system of Q rank rectangle QAM modulation, transmitting signal is taken from and is sized toPAM assemble of symbol ��; At signal node place, the vector constituted with the logarithm of the ratio of element prior probability each in set omega represents message to be passed, carrys out more new information according to sum-product algorithm; At observer nodes place, the vector constituted with posteriority log-likelihood ratio represents message to be passed; Utilize the BP algorithm on full limit by noise and interference sum are approximately Gaussian distribution, derive information updating criterion; After the iteration more than once, the message of observer nodes obtain the Soft Inform ation of output, obtain launching the estimation of signal through judgement.

Preferred: the BP algorithm on full limit is simplified, obtain monolateral BP algorithm.

Preferred: the method that complex field computing is converted into real number field computing: in the up-link of extensive mimo system, it is assumed that mobile terminal transmitting antenna number is M, base station reception antenna number is N; Recurrence directive amount isLaunching signal and carried out quadrature amplitude modulation QAM, and constellation sizes is | | �� | |=Q, multiple connection is received vector and isIn equivalence real number system, signal node x_iTo observer nodes r_jThe massage set of transmission is designated asObserver nodes r_jTo signal node x_iThe massage set of transmission is designated asIterations during judgement output is I; For the mimo system of complex field,It is represented by:

$\tilde{r}=\tilde{H}\tilde{x}+\tilde{n};$

Wherein,For independent identically distributed Rayleigh fading channel, it is 0 that the real part of each element and imaginary part all obey average, and variance is the Gauss distribution of 1;Represent the additive white Gaussian noise of channel, and ${\tilde{n}}_i~CN(0,{\mathrm{\σ}}^2);$

By the method for real-valued decomposition, by complex fieldBeing decomposed into real number model is:

R=Hx+n

Wherein, r=[r₁,r₂,...,r_2N]^T, x=[x₁,x₂,...,x_2M] and channel matrix H_2N��2MIt it is all real number;

Mapping relations between plural number and real number model are as follows:

Wherein,WithRepresent real part and imaginary part, the then x of plural number * respectively_i�� ��,

$\mathrm{\Ω}=\{(-\sqrt{Q}+1),...,-1,+1,...,(+\sqrt{Q}-1)\};$

�� be complex value constellation point �� in the same direction or the assemble of symbol of quadrature component.

Preferred: the operating procedure of described iterative detection:

Step 1, initializes: ${\mathrm{\α}}_{i,j}^{\left(0\right)}=0,{\mathrm{\β}}_{i,j}^{\left(0\right)}=0;$

Step 2, the information updating of observer nodes:

${\mathrm{\β}}_{j,i}^{\left(l\right)}\left(s_k\right)=\frac{2h_{j,i}(r_j-{\mathrm{\μ}}_{z_{j,i}})(s_k-s_0)-h_{j,i}^2(s_k^2-s_0^2)}{2{\mathrm{\σ}}_{z_{j,i}}^2};$

Wherein, ${\mathrm{\μ}}_{z_{j,i}}=\underset{k=1}{\overset{2M}{\Σ}}{h}_{j,k}E\left\{x_k\right\}-h_{j,i}E\left\{x_i\right\}={\mathrm{\μ}}_{z_j}-h_{j,i}E\left\{x_i\right\};$

${\mathrm{\σ}}_{z_{j,i}}^2=\underset{k=1}{\overset{2M}{\Σ}}{h}_{j,k}^{2}Var\left\{x_k\right\}-h_{j,i}^{2}Var\left\{x_i\right\}={\mathrm{\σ}}_{z_j}^2-h_{j,i}^{2}Var\left\{x_i\right\};$

Step 3, the information updating of signal node:

${\mathrm{\α}}_{i,j}^{\left(l\right)}\left(s_k\right)=\underset{t=1,t\≠j}{\overset{2N}{\Σ}}{\mathrm{\β}}_{t,i}^{(l-1)}\left(s_k\right);$

$p_{i,j}^{\left(l\right)}(x_i=s_k)=\frac{\exp \left({\mathrm{\α}}_{i,j}^{\left(l\right)}\right(s_k))}{1+\underset{m=1}{\overset{\sqrt{Q}-1}{\Σ}}\exp \left({\mathrm{\α}}_{i,j}^{\left(l\right)}\right(s_m\left)\right)};$

Step 4, output judgement:

$\mathrm{\γ}\left(s_k\right)=\underset{t=1}{\overset{2N}{\Σ}}{\mathrm{\β}}_{t,i}^{\left(L\right)},k=1,\mathrm{...},\sqrt{Q}-1;$

${\hat{x}}_i=\underset{s_k\∈s}{\arg \max }\left\{{\mathrm{\γ}}_i\left(s_k\right)\right\};$

Wherein, $s=\[-\sqrt{Q}+1,...,-1,+1,...,+\sqrt{Q}-1\].$

Beneficial effect: the low complex degree belief propagation detection algorithm of extensive mimo system provided by the invention, compared to existing technology, has the advantages that

Compared with existing MIMO detector, the present invention is applicable to any antenna configurations and the extensive MIMO scene of any order of modulation. Use the equivalent real number model of mimo system, the computing of complex field is converted to real number field computing, it is achieved that based on the extensive MIMO detection algorithm of symbol. The performance of this detection algorithm is better than MMSE detection, and avoids higher dimensional matrix and invert, it is possible to be perfectly suitable for the application scenarios of extensive MIMO. Meanwhile, detection performance is further improved along with the increase of antenna scale, and the theoretical behavior with extensive mimo system is consistent.

Accompanying drawing explanation

Fig. 1. the factor graph of extensive mimo system in real number field.

Fig. 2. information transitive graph between signal node and observer nodes in real number field.

Fig. 3. simulation result contrast (4-QAM) of different detection methods.

Fig. 4. the BP detection algorithm of proposition simulation result (4-QAM) under different antennae configures.

Fig. 5. the simulation result (16-QAM) of strict and approximate MMSE detection.

Fig. 6. the BP detection algorithm of the proposition simulation result (16-QAM) when different TSRR.

Detailed description of the invention

Below in conjunction with the drawings and specific embodiments, it is further elucidated with the present invention, it should be understood that these examples are merely to illustrate the present invention rather than restriction the scope of the present invention, after having read the present invention, the amendment of the various equivalent form of values of the present invention is all fallen within the application claims limited range by those skilled in the art.

The low complex degree belief propagation detection algorithm of a kind of extensive mimo system, in extensive mimo system, uses the real number model of its equivalence to build corresponding factor graph, complex field computing is converted into real number field computing, it is achieved based on the iterative detection of BP; Wherein, factor graph is used for representing reception signal and launching dependence between signal, will launch signal as signal node, and receive signal as observer nodes;Each signal node updates prior information according to the posterior information obtained from observer nodes, is then transferred to all observer nodes being attached thereto; Each observer nodes calculates posterior information according to the prior information from signal node, is then passed back to the signal node being attached thereto; Thus, complete renewal and the transmittance process of a message; After the iteration more than once, observer nodes output Soft Inform ation is used for estimating to launch symbol;

In the real number system of Q rank rectangle QAM modulation, transmitting signal is taken from and is sized toPAM assemble of symbol ��; At signal node place, the vector constituted with the logarithm of the ratio of element prior probability each in set omega represents message to be passed, carrys out more new information according to sum-product algorithm; At observer nodes place, the vector constituted with posteriority log-likelihood ratio represents message to be passed; Utilize the BP algorithm on full limit by noise and interference sum are approximately Gaussian distribution, derive information updating criterion; After the iteration more than once, the message of observer nodes obtain the Soft Inform ation of output, obtain launching the estimation of signal through judgement.

The present embodiment, in extensive mimo system, uses the real number model of its equivalence to build corresponding factor graph, complex field computing is converted into real number field computing, it is achieved based on the iterative detection of BP. Wherein, launch signal and be called " signal node ", receive signal and be called " observer nodes ". First, each signal node updates prior information according to the posterior information obtained from observer nodes, is then transferred to all observer nodes being attached thereto. Secondly, each observer nodes calculates posterior information according to the prior information from signal node, is then passed back to the signal node being attached thereto. Thus, complete renewal and the transmittance process of a message. After several times iteration, observer nodes output Soft Inform ation is used for estimating to launch symbol.

In the real number system of Q rank rectangle QAM modulation, transmitting signal is taken from and is sized toPAM assemble of symbol ��. At signal node place, the vector constituted with the logarithm of the ratio of element prior probability each in PAM assemble of symbol �� represents message to be passed, carrys out more new information according to sum-product algorithm (SPA). At observer nodes place, the vector constituted with posteriority log-likelihood ratio represents message to be passed. By noise and interference sum (NPI) are approximately Gaussian distribution, derive information updating criterion, be called the BP algorithm (FE-BP) on full limit. In order to reduce the complexity of information updating, the information updating criterion on full limit is simplified, obtain monolateral BP algorithm (SE-BP). After enough successive ignition, the message of observer nodes obtain the Soft Inform ation of output, obtain launching the estimation of signal through judgement.

Scalar x represents, vector or matrix x represent. In the up-link of extensive mimo system, it is assumed that mobile terminal transmitting antenna number is M, base station reception antenna number is N. Recurrence directive amount isLaunching signal and carried out quadrature amplitude modulation (QAM), and constellation sizes is | | �� | |=Q, multiple connection is received vector and isIn equivalence real number system, signal node x_iTo observer nodes r_jThe massage set of transmission is designated asObserver nodes r_jTo signal node x_iThe massage set of transmission is designated asIterations during judgement output is I.

1. equivalence real number model

For the mimo system of complex field,It is represented by:

$\tilde{r}=\tilde{H}\tilde{x}+\tilde{n}---\left(1\right)$

Wherein,For independent identically distributed Rayleigh fading channel, each element

Real part and imaginary part all to obey average be 0, variance is the Gauss distribution of 1;Represent channel

Additive white Gaussian noise, and ${\tilde{n}}_i~CN(0,{\mathrm{\σ}}^2).$

The method using real-valued decomposition, the real number model of formula (1) is represented by:

R=Hx+n (2)

Wherein, real reception vector r=[r₁,r₂,...,r_2N]^T, launch vector x=[x in fact₁,x₂,...,x_2M] and real channel matrix H_2N��2M. Mapping relations between plural number and real number model are as follows:

Wherein,WithRepresent real part and imaginary part, the then x of plural number * respectively_i�� ��,

$\mathrm{\Ω}=\{(-\sqrt{Q}+1),...,-1,+1,...,(+\sqrt{Q}-1)\}---\left(4\right)$

�� be complex value constellation point �� in the same direction or the assemble of symbol of quadrature component,Represent that N ties up complex number space,Represent that 2N ties up real number space. Therefore, modulation constellation is the real mimo system that the multiple mimo system of ��, N �� M is equivalent to that modulation constellation is ��, 2N �� 2M.

2. the BP detection algorithm of real number field

Below in conjunction with accompanying drawing, the low complex degree BP detection algorithm in the extensive mimo system propose the present invention is described in detail:

Fig. 1, based on the real number model of mimo system, represents reception signal with factor graph and launches dependence between signal. Fig. 2, based on the factor graph of real number field, describes the message transmission between signal node and observer nodes in BP iterative detection process. For SE-BP, do not consider further that in the information updating of observer nodesRepresent with light black in the drawings. In the l time iteration, the prior information of signal node isThe posteriority message of observer nodes is

The operating procedure of iterative detection of the present invention is as follows:

1) initialize: ${\mathrm{\α}}_{i,j}^{\left(0\right)}=0,{\mathrm{\β}}_{i,j}^{\left(0\right)}=0---\left(5\right)$

2) information updating of observer nodes:

${\mathrm{\β}}_{j,i}^{\left(l\right)}\left(s_k\right)=\frac{2h_{j,i}(r_j-{\mathrm{\μ}}_{z_{j,i}})(s_k-s_0)-h_{j,i}^2(s_k^2-s_0^2)}{2{\mathrm{\σ}}_{z_{j,i}}^2}---\left(6\right)$

Wherein,

${\mathrm{\μ}}_{z_{j,i}}=\underset{k=1}{\overset{2M}{\Σ}}{h}_{j,k}E\left\{x_k\right\}-h_{j,i}E\left\{x_i\right\}={\mathrm{\μ}}_{z_j}-h_{j,i}E\left\{x_i\right\}---\left(7\right)$

${\mathrm{\σ}}_{z_{j,i}}^2=\underset{k=1}{\overset{2M}{\Σ}}{h}_{j,k}^{2}Var\left\{x_k\right\}-h_{j,i}^{2}Var\left\{x_i\right\}={\mathrm{\σ}}_{z_j}^2-h_{j,i}^{2}Var\left\{x_i\right\}---\left(8\right)$

3) information updating of signal node:

${\mathrm{\α}}_{i,j}^{\left(l\right)}\left(s_k\right)=\underset{t=1,t\≠j}{\overset{2N}{\Σ}}{\mathrm{\β}}_{t,i}^{(l-1)}\left(s_k\right)---\left(9\right)$

$p_{i,j}^{\left(l\right)}(x_i=s_k)=\frac{\exp \left({\mathrm{\α}}_{i,j}^{\left(l\right)}\right(s_k))}{1+\underset{m=1}{\overset{\sqrt{Q}-1}{\Σ}}\exp \left({\mathrm{\α}}_{i,j}^{\left(l\right)}\right(s_m\left)\right)}---\left(10\right)$

4) output judgement:

$\mathrm{\γ}\left(s_k\right)=\underset{t=1}{\overset{2N}{\Σ}}{\mathrm{\β}}_{t,i}^{\left(L\right)},k=1,\mathrm{...},\sqrt{Q}-1---\left(11\right)$

${\hat{x}}_i=\underset{s_k\∈s}{\arg \max }\left\{{\mathrm{\γ}}_i\left(s_k\right)\right\}---\left(12\right)$

Wherein, L represents total iterations, and

$s=\[-\sqrt{Q}+1,...,-1,+1,...,+\sqrt{Q}-1\]---\left(13\right)$

Wherein, in the l time iteration, the prior information of signal node is expressed asThe posteriority message table of observer nodes is shown as ${\mathrm{\β}}_{r_j\→x_i}^{\left(l\right)},s=\[-\sqrt{Q}+1,...,-1,+1,...,+\sqrt{Q}-1\];$ h_j,iRepresent the element of channel matrix H jth row, i row, r_jRepresent the jth element receiving vector r, s_kRepresent the kth element of vector s, �� { x_kRepresent the average of kth element, Var{x in vector x_kRepresent x_kVariance,Represent in the l time iteration, from signal node x_iIt is delivered to observer nodes r_jAbout s_kPrior information,Represent in the l time iteration, from signal node x_iIt is delivered to observer nodes r_jAbout s_kPrior probability, represent judgement output Soft Inform ation,Represent and launch signal x_iJudgement output.

3. simulation result

For Rayleigh fading channel, for different antennae configuration and modulation system, simulation result is as seen in figures 3-6. In all emulation, the iterations of BP detection is all set to 7, and is left out chnnel coding.

For mimo system, the real number field BP detection of improvement, bit error rate (BER) performance comparison of the SISO channel of general complex field SE-BP detection and linear MMSE detection and single-input single-output is as shown in Figure 3. From the figure 3, it may be seen that the BER performance of the BP detection improved and general SE-BP detection is far superior to MMSE detection, avoids higher dimensional matrix simultaneously and invert. Additionally, the BP improved detects performance is better than general SE-BP detection performance. Specifically, at BER=10^-2Time, signal to noise ratio improves about 1dB.

Fig. 4 gives under different antennae configuration, the simulation result of the BP detection algorithm of improvement. Antenna configurations three kinds different respectively M=N=16,32 and 64. Observe it can be seen that the performance improving BP detection algorithm promotes along with the increase of number of antennas, constantly approach the detection performance of AWGNSISO channel. This phenomenon is consistent with the theory analysis of extensive mimo system. It is therefore proposed that BP detection algorithm will be well suited for extensive mimo system.

For 16-QAM modulating system, the performance comparison that the MMSE that strict MMSE detects and is similar to based on Neumann progression detects is as shown in Figure 5. Wherein, variable k represent k rank Neumann progression be similar to. As TSRR less (M=8, N=128), based on the MMSE detection that Neumann progression is approximate, there is the BER performance almost consistent with strict MMSE detection. But, when TSRR increases (M=8, N=64), even if adopting higher approximate exponent number to be also unable to reach the performance of strict MMSE detection. It will be appreciated from fig. 6 that as TSRR=16 or 2, it is proposed to BP detection algorithm be superior to MMSE detection, and approach the detection performance of AWGNSISO channel. It follows that for the mimo system with bigger or less TSRR, it is proposed to BP detection algorithm be all suitable for.

Therefore, for extensive mimo system, no matter adopt which kind of modulation system and antenna configurations, it is proposed to the real number field BP detection algorithm based on symbol all effective, and detection performance is better than MMSE detection. Meanwhile, this algorithm avoids matrix inversion, has good application prospect in extensive mimo system.

4. analysis of complexity

The complexity of the BP detection algorithm proposed includes the renewal (formula (6)��(8)) of observer nodes place posterior information, and the renewal of signal node place prior information (formula (9)��(10)), represent with POST.UP and PRI.UP respectively. The complexity of different BP detection algorithms is more as shown in table 1. Complexity according to table 1, FE-BP is exponentially increased along with the increase of M, is similar to maximum a posteriori probability (MAP) detection. Intuitively, it is proposed to BP detection have with general SE-BP detection there is identical complexity magnitude. It practice, general SE-BP is definition computing in real number field, and the BP detection definition proposed is in real number field. It is therefore proposed that BP detection algorithm be more beneficial for hardware realize, especially for high order modulation. For the mimo system of M=8, N=32 and 256-QAM modulation, compare general SE-BP detection, it is proposed to the complexity of BP detection algorithm reduce 75%.

The complexity of the different BP detection algorithm of table 1 compares

Compared with existing extensive MIMO detection algorithm, the main contributions of the present invention is in that: uses the equivalent real number model of mimo system, the computing of complex field is converted to real number field computing, it is achieved that the extensive MIMO detection algorithm of low complex degree. For the extensive mimo system of any antenna configurations and order of modulation, the performance of this detection algorithm is better than MMSE detection, avoids higher dimensional matrix simultaneously and inverts. Additionally, detection performance is further improved along with the increase of antenna scale, it is possible to be perfectly suitable for the application scenarios of extensive MIMO. In High Order Modulation System, the computation complexity of this algorithm is lower than general complex field BP detection algorithm. For in 4-QAM modulating system, BER is 10^-2Time, this algorithm compares general complex field BP detection algorithm, and signal to noise ratio improves about 1dB.

The present invention is based on the factor graph of real number field, it is proposed that a kind of low complex degree detection algorithm based on belief propagation (BP) for extensive multiple antennas (MIMO) system. This algorithm, based on real number field computing, reduces the search volume of detection, detects advantageously than the BP of general complex field in High Order Modulation System. Fig. 1, based on the real number model of mimo system, represents reception signal with factor graph and launches dependence between signal. Fig. 2, based on the factor graph of real number field, describes the message transmission between signal node and observer nodes in BP iterative detection process.Wherein, launch signal and be called " signal node ", receive signal and be called " observer nodes ". First, each signal node updates prior information according to the posterior information obtained from observer nodes, is then transferred to all observer nodes being attached thereto. Secondly, each observer nodes calculates posterior information according to the prior information from signal node, is then passed back to the signal node being attached thereto. Thus, complete renewal and the transmittance process of a message. After enough successive ignition, the message according to observer nodes place, calculate output Soft Inform ation and be used for estimating to launch symbol.

The present invention is suitable in the extensive mimo system of any antenna configurations and order of modulation, and the base station end receiver of up-link designs. The present invention uses the equivalent real number model of extensive mimo system, and the computing of complex field is converted to real number field computing, it is achieved that the BP detection algorithm based on symbol of low complex degree. For the extensive mimo system of any antenna configurations and order of modulation, the performance of this detection algorithm is better than MMSE detection, avoids higher dimensional matrix simultaneously and inverts. The detection performance of this algorithm is better than the BP detection of general complex field, and in High Order Modulation System, complexity detects also below the BP of complex field. Additionally, detection performance is further improved along with the increase of antenna scale, it is possible to be perfectly suitable for the application scenarios of extensive MIMO.

The present invention, in conjunction with existing belief propagation (BP) Iterative detection algorithm, devises a kind of real number field detection algorithm based on symbol. The present invention suitable in extensive mimo system, the signal detection of uplink base station receiver. The present invention, based on the equivalent real number model of extensive mimo system, devises the BP detection algorithm based on symbol suitable in extensive mimo system, and its detection performance is better than MMSE detection algorithm. Additionally, transmitting antenna number square with ratio (TSRR) M of reception antenna number²When/N is very big, MMSE detection algorithm is no longer applicable, and the BP algorithm that the present invention proposes is appointed so applicable, and avoids higher dimensional matrix inversion operation such that it is able to be well adapted for the scene of extensive MIMO.

The above is only the preferred embodiment of the present invention; it is noted that, for those skilled in the art; under the premise without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications also should be regarded as protection scope of the present invention.

Claims (5)

1. the low complex degree belief propagation detection algorithm of an extensive mimo system, it is characterized in that: in extensive mimo system, use the real number model of its equivalence to build corresponding factor graph, complex field computing is converted into real number field computing, it is achieved based on the iterative detection of BP; Wherein, factor graph is used for representing reception signal and launching dependence between signal, will launch signal as signal node, and receive signal as observer nodes; Each signal node updates prior information according to the posterior information obtained from observer nodes, is then transferred to all observer nodes being attached thereto; Each observer nodes calculates posterior information according to the prior information from signal node, is then passed back to the signal node being attached thereto; Thus, complete renewal and the transmittance process of a message; After the iteration more than once, observer nodes output Soft Inform ation is used for estimating to launch symbol.

2. the low complex degree belief propagation detection algorithm of extensive mimo system according to claim 1, it is characterized in that: at signal node place, the vector constituted with the logarithm of the ratio of element prior probability each in PAM assemble of symbol �� represents message to be passed, carrys out more new information according to sum-product algorithm;Wherein, PAM assemble of symbol �� represents the set launching signal in the real number system of Q rank rectangle QAM modulation, and this PAM assemble of symbol �� is sized to

At observer nodes place, the vector constituted with posteriority log-likelihood ratio represents message to be passed; Utilize the BP algorithm on full limit by noise and interference sum are approximately Gaussian distribution, derive information updating criterion.

3. the low complex degree belief propagation detection algorithm of extensive mimo system according to claim 2, it is characterized in that: to the BP algorithm on full limit by replacement criteria is simplified, obtain monolateral BP algorithm, utilize the BP algorithmic derivation outbound message replacement criteria that this is monolateral.

4. the low complex degree belief propagation detection algorithm according to the arbitrary described extensive mimo system of claim 1-3, it is characterized in that: the described method that complex field computing is converted into real number field computing: in the up-link of extensive mimo system, assuming that mobile terminal transmitting antenna number is M, base station reception antenna number is N; Recurrence directive amount isLaunching signal and carried out quadrature amplitude modulation QAM, and constellation sizes is | | �� | |=Q, multiple connection is received vector and isIn equivalence real number system, signal node x_iTo observer nodes r_jThe massage set of transmission is designated asObserver nodes r_jTo signal node x_iThe massage set of transmission is designated asIterations during judgement output is I; For the mimo system of complex field,It is represented by:

$\tilde{r}=\tilde{H}\tilde{x}+\tilde{n};$

Wherein,For independent identically distributed Rayleigh fading channel, it is 0 that the real part of each element and imaginary part all obey average, and variance is the Gauss distribution of 1;Represent the additive white Gaussian noise of channel, and

By the method for real-valued decomposition, by complex fieldBeing decomposed into real number model is:

R=Hx+n

Wherein, real reception vector r=[r₁,r₂,...,r_2N]^T, launch vector x=[x in fact₁,x₂,...,x_2M] and real channel matrix H_2N��2M;

Mapping relations between plural number and real number model are as follows:

Wherein,WithRepresent real part and imaginary part, the then x of plural number * respectively_i�� ��,

$\mathrm{\Ω}=\{(-\sqrt{Q}+1),...,-1,+1,...,(+\sqrt{Q}-1)\};$ �� be complex value constellation point �� in the same direction or the assemble of symbol of quadrature component,Represent that N ties up complex number space,Represent that 2N ties up real number space.

5. the low complex degree belief propagation detection algorithm of extensive mimo system according to claim 4, it is characterised in that: the operating procedure of described iterative detection:

Step 1, initializes: ${\mathrm{\α}}_{i,j}^{\left(0\right)}=0,{\mathrm{\β}}_{i,j}^{\left(0\right)}=0;$

Step 2, the information updating of observer nodes:

${\mathrm{\β}}_{j,i}^{\left(l\right)}\left(s_k\right)=\frac{2h_{j,i}(r_j-{\mathrm{\μ}}_{z_{j,i}})(s_k-s_0)-h_{j,i}^2(s_k^2-s_0^2)}{2{\mathrm{\σ}}_{z_{j,i}}^2};$

Wherein, ${\mathrm{\μ}}_{z_{j,i}}=\underset{k=1}{\overset{2M}{\Σ}}{h}_{j,k}E\left\{x_k\right\}-h_{j,i}E\left\{x_i\right\}={\mathrm{\μ}}_{z_j}-h_{j,i}E\left\{x_i\right\};$

${\mathrm{\σ}}_{z_{j,i}}^2=\underset{k=1}{\overset{2M}{\mathrm{\Σ}}}{h}_{j,k}^{2}Var\left\{x_k\right\}-h_{j,i}^{2}Var\left\{x_i\right\}={\mathrm{\σ}}_{z_j}^2-h_{j,i}^{2}Var\left\{x_i\right\};$

Step 3, the information updating of signal node:

${\mathrm{\α}}_{i,j}^{\left(l\right)}\left(s_k\right)=\underset{t=1,t\≠j}{\overset{2N}{\Σ}}{\mathrm{\β}}_{t,i}^{(l-1)}\left(s_k\right);$

$p_{i,j}^{\left(l\right)}(x_i=s_k)=\frac{\exp \left({\mathrm{\α}}_{i,j}^{\left(l\right)}\right(s_k\left)\right)}{1+\underset{m=1}{\overset{\sqrt{Q}-1}{\Σ}}\exp \left({\mathrm{\α}}_{i,j}^{\left(l\right)}\right(s_m\left)\right)};$

Step 4, output judgement:

$\mathrm{\γ}\left(s_k\right)=\underset{t=1}{\overset{2N}{\Σ}}{\mathrm{\β}}_{t,i}^{\left(L\right)},k=1,...,\sqrt{Q}-1;$

${\hat{x}}_i=\underset{s_k\∈s}{\mathrm{argmax}}\left\{{\mathrm{\γ}}_i\left(s_k\right)\right\};$

Wherein, in the l time iteration, the prior information of signal node is expressed asThe posteriority message table of observer nodes is shown as $s=[-\sqrt{Q}+1,...,-1,+1,...,+\sqrt{Q}-1];$ h_j,iRepresent the element of channel matrix H jth row, i row, r_jRepresent the jth element receiving vector r, s_kRepresent the kth element of vector s, �� { x_kRepresent the average of kth element, Var{x in vector x_kRepresent x_kVariance,Represent in the l time iteration, from signal node x_iIt is delivered to observer nodes r_jAbout s_kPrior information,Represent in the l time iteration, from signal node x_iIt is delivered to observer nodes r_jAbout s_kPrior probability, represent judgement output Soft Inform ation,Represent and launch signal x_iJudgement output.