CN103746731A - Probability calculation-based multiple input multiple output detector and detection method - Google Patents

Abstract

The invention discloses a multi-input multi-output detector based on probability calculation, which includes a matrix QR decomposer, the matrix QR decomposer is respectively connected to two random sequence generators, and the two random sequence generators are respectively connected to a probability complex multiplier , one of the probability complex multipliers is connected to the Gibbs sampling update unit, and the other probability complex multiplier and the Gibbs sampling update unit are simultaneously connected to the log likelihood ratio calculation unit. The multi-input multi-output detector of the present invention uses probability calculation to realize MCMC algorithm, greatly reduces computational complexity, improves the transition probability of Markov chain in MCMC algorithm, and solves the problem of locking under high performance-to-noise ratio. The Gibbs sampling update is performed by using the sliding window generation sequence method, which reduces the length of the probability sequence. Applying the multi-input multi-output detector of the present invention to construct a fully parallel detector can be realized in a fully parallel manner to achieve higher throughput.

Description

Multi-input multi-output detector and detection method based on probability calculation

technical field

The present invention relates to the technical field of wireless communication, in particular to a multiple input multiple output (Multiple Input Multiple Output, MIMO) detector based on probability calculation, and a fully parallel detection system composed of the multiple input multiple output detector, and multiple Enter the detection method for the multiple output detector.

Background technique

The MIMO communication system uses multiple antennas at both ends of the wireless link to receive and transmit signals at the same time, which can fully exploit space resources and double the capacity and reliability of the communication system without increasing spectrum resources and transmission power. sex. Since multiple-input multiple-output (MIMO) technology can provide mobile users with high-quality, high-speed information transmission in a wireless mobile environment, MIMO technology has become the core technology of the fourth generation mobile communication and future wireless/mobile communication.

The current MIMO communication system only detects and restores the received signal to the transmitted signal after conversion, and obtains the estimated value of the transmitted signal. Without further detection of the similarity between the transmitted signal and the received signal, it cannot be directly connected to the decoder for use. In addition, the Monte Carlo Markov Chain (MCMC) algorithm is often used in MIMO communication systems to detect received signals. The MCMC algorithm is a very beneficial mathematical tool. Its basic idea is to construct a Markov chain, make its stationary distribution the posterior distribution of the parameters to be estimated, and generate samples of the posterior distribution through this Markov chain, and Monte Carlo integration is performed based on the samples (effective samples) when the Markov chain reaches a stationary distribution. Since the Gibbs sampling (Gibbs sampling) update in the traditional MCMC algorithm involves a large number of multiplication and addition operations, the traditional MCMC algorithm requires a large amount of calculations to update the conditional probability and sampling data, and the computational complexity is high.

Contents of the invention

The purpose of the present invention is to overcome the disadvantages of the conventional MCMC algorithm-based MIMO detector with a large amount of computation and complex structure existing in the prior art, to provide a multi-input multi-output detector based on probability calculation, and to apply the multi-input multi-output detector. A fully parallel detection system composed of output detectors. The multi-input multi-output detector and the full parallel detection system of the present invention adopt the MCMC algorithm based on probability calculation, which can greatly reduce the computational complexity and simplify the structure of the MIMO detector.

In order to realize the above-mentioned purpose of the invention, the present invention provides the following technical solutions:

The multi-input multi-output detector based on probability calculation includes a matrix QR decomposer, and the matrix QR decomposer is respectively connected to the second random sequence generator and the third random sequence generator; the third random sequence generator is passed through the third random sequence generator Two probability complex multipliers are connected to the Gibbs sampling update unit; the second random sequence generator is connected to the first probability complex multiplier, and the first probability complex multiplier is also connected to the first random sequence generator and the log likelihood ratio calculation unit; the operation result of the first probability complex multiplier is output to the log likelihood ratio calculation unit, and the operation result of the Gibbs sampling update unit is respectively output to the log likelihood ratio calculation unit and the second probability complex multiplier, and the log likelihood The output of the ratio calculation unit is used as the input of the Gibbs sampling update unit.

In the above-mentioned multi-input multi-output detector based on probability calculation, the first random sequence generator, the second random sequence generator and the third random sequence generator have the same structure, and all include an absolute value operation module and a comparator, The absolute value calculation module calculates the absolute value of the input signal of each clock and outputs it to the comparator. The comparator compares the absolute value of the input signal with any random number that satisfies a uniform distribution in the interval [0, m). If the input If the absolute value of the signal is greater than the random number, output 1, otherwise output 0, and obtain a value sequence composed of 1 and 0; corresponding to each bit in the value sequence, if the input signal is less than 0, output 1, otherwise output 0, according to A sequence of values gets a signed sequence consisting of 0s and 1s.

和第二随机序列生成器输出的第二随机序列

经过复数乘法运算后获得第一乘积序列

第三随机序列生成器输出的第三随机序列与Gibbs采样更新单元输出的似然估计序列S经过复数运算后获得第二乘积序列

In the above-mentioned multi-input multi-output detector based on probability calculation, the structure of the first probability complex multiplier and the second probability complex multiplier are the same, and are all composed of four probability real number multipliers, two probability real number adders and a Negative circuit composition; four probability real number multipliers are respectively the first probability real number multiplier, the second probability real number multiplier, the third probability real number multiplier and the fourth probability real number multiplier; the two probability real number adders are respectively the first probability real number multiplier A probability real number adder and a second probability real number adder; the first probability real number multiplier is connected to the first probability real number adder, and the second probability real number multiplier is connected to the first probability real number adder through an inversion circuit; the third probability real number multiplication The device and the fourth probability real number multiplier are respectively connected with the second probability real number adder; the first random sequence output by the first random sequence generator

and the second random sequence output by the second random sequence generator

Obtain the first product sequence after complex multiplication

The third random sequence output by the third random sequence generator The second product sequence is obtained after complex operation with the likelihood estimation sequence S output by the Gibbs sampling update unit

和第二概率复数乘法器输出的第二乘积序列

进行滑窗处理，分别截取第一乘积序列

和第二乘积序列

中时间窗口t₁内的数据，截取的数据与Gibbs采样更新单元输出的似然估计序列S进行对数似然值计算，获取最大似然估计值

$\mathrm{LLR}\left(X_{k,b}^{t_1}\right|Z^{t_1})=\min {\mathrm{\Σ}}_{i=1}^{N_t}{\left|\right|{(Q^{H,t}\·Z^{t_1})}_i-{(R^{t_1}\·S)}_i-R_{i,k}^{t_1}\·S_k^{t_1}\left(x\right)\left|\right|}^2,$ Nt为接收天线数，

为似然估计序列S中的第k个参数。In the above-mentioned multi-input multi-output detector based on probability calculation, the log-likelihood ratio calculation unit outputs the first product sequence of the first probability complex multiplier

and the second product sequence output by the second probabilistic complex multiplier

Carry out sliding window processing, and intercept the first product sequence respectively

and the second product sequence

The data in the middle time window _t1 , the intercepted data and the likelihood estimation sequence S output by the Gibbs sampling update unit perform logarithmic likelihood calculation to obtain the maximum likelihood estimation value

$\mathrm{LLR}\left(x_{k,\mathrm{<figure-callout\; id="1"\; label="b\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">b</figure-callout>}}^{t_{}}\right|Z^{t_1})=\min {\mathrm{\&Sigma;}}_{i=1}^{N_t}{\left|\right|{(Q^{h,t}\&CenterDot;Z^{t_1})}_i-{({\mathrm{<figure-callout\; id="1"\; label="R\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">R</figure-callout>}}^{t_{}}\&Center\; Dot;S)}_i-R_{i,\mathrm{<figure-callout\; id="1"\; label="k\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">k</figure-callout>}}^{t_{}}\&CenterDot;S_k^{t_1}\left(x\right)\left|\right|}^2,$ Nt is the number of receiving antennas,

is the likelihood estimate for the kth parameter in the sequence S.

获得由M个变量组成的参数序列

经过QAM映射后的参数序列

更替似然估计序列S中的参数

获得更新后的似然估计序列，M为时间窗口t₁内截取的数据数。In the above-mentioned multiple-input multiple-output detector based on probability calculation, the Gibbs sampling update unit is based on

Get a parameter sequence consisting of M variables

Parameter sequence after QAM mapping

Replacement Likelihood Estimation of Parameters in Sequence S

The updated likelihood estimation sequence is obtained, and M is the number of intercepted data in the time window _t1 .

子矩阵Q经过另一个随机序列生成器生成第二随机序列

第三随机序列

和第二随机序列广播至多个并行的基于概率计算的多输入多输出检测器。The present invention also provides an all-parallel detection system, including the above-mentioned MIMO detector based on probability calculation, and multiple MIMO detectors based on probability calculation are respectively connected in parallel to the same decoder. Preferably, the full parallel detection system also includes a matrix preprocessing unit, which includes a matrix QR decomposer and two random sequence generators, and the channel matrix H is decomposed into sub-matrix Q and sub-matrix by the matrix QR decomposer. Matrix R, sub-matrix R generates a third random sequence through a random sequence generator

The sub-matrix Q generates a second random sequence through another random sequence generator

third random sequence

and the second random sequence Broadcast to multiple parallel probability-based MIMO detectors.

The present invention also provides a detection method of a multi-input multi-output detector based on probability calculation, comprising the following steps:

并输出至第一概率复数乘法器，子矩阵Q经过矩阵求逆运算得到表征矩阵Q^H，表征矩阵Q^H再经过第二随机序列生成器生成第二随机序列

并输出至第二概率复数乘法器；接收信号Z经过第一随机序列生成器生成第一随机序列

并输出至第一概率复数乘法器；Step 1: The matrix QR decomposer decomposes the channel matrix H into sub-matrices Q and R, and the sub-matrix R generates a third random sequence through the third random sequence generator

And output to the first probability complex multiplier, the sub-matrix Q obtains the representation matrix Q ^H through the matrix inversion operation, and the representation matrix Q ^H generates the second random sequence through the second random sequence generator

And output to the second probability complex multiplier; the received signal Z generates the first random sequence through the first random sequence generator

And output to the first probability complex multiplier;

和第一随机序列进行复数乘法运算，获得第一乘积序列

并输出至对数似然比计算单元；第二概率复数乘法器对第三随机序列和Gibbs采样更新单元输出的似然估计序列S进行复数乘法运算，获得第二乘积序列

Step 2: The first probabilistic complex multiplier against the second random sequence

and the first random sequence Perform complex multiplication to obtain the first product sequence

and output to the logarithmic likelihood ratio calculation unit; the second probability complex multiplier is to the third random sequence Perform complex multiplication with the likelihood estimation sequence S output by the Gibbs sampling update unit to obtain the second product sequence

和第二乘积序列进行滑窗处理，分别截取第一乘积序列

和第二乘积序列

中时间窗口t₁内的数据，截取的数据与Gibbs采样更新单元输出的似然估计序列S进行对数似然值计算，获取最大似然估计值

$\mathrm{LLR}\left(X_{k,b}^{t_1}\right|Z^{t_1})=\min {\mathrm{\&Sigma;}}_{i=1}^{N_t}{\left|\right|{(Q^{H,t}\&CenterDot;Z^{t_1})}_i-{(R^{t_1}\&CenterDot;S)}_i-R_{i,k}^{t_1}\&CenterDot;S_k^{t_1}\left(x\right)\left|\right|}^2;$ Step 3: The log-likelihood ratio calculation unit pairs the first product sequence

and the second product sequence Carry out sliding window processing, and intercept the first product sequence respectively

and the second product sequence

The data in the middle time window _t1 , the intercepted data and the likelihood estimation sequence S output by the Gibbs sampling update unit perform logarithmic likelihood calculation to obtain the maximum likelihood estimation value

$\mathrm{LLR}\left(x_{k,\mathrm{<figure-callout\; id="1"\; label="b\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">b</figure-callout>}}^{t_{}}\right|Z^{t_1})=\min {\mathrm{\&Sigma;}}_{i=1}^{N_t}{\left|\right|{(Q^{h,t}\&Center\; Dot;Z^{t_1})}_i-{({\mathrm{<figure-callout\; id="1"\; label="R\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">R</figure-callout>}}^{t_{}}\&CenterDot;S)}_i-R_{i,\mathrm{<figure-callout\; id="1"\; label="k\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">k</figure-callout>}}^{t_{}}\&Center\; Dot;S_k^{t_1}\left(x\right)\left|\right|}^2;$

更新似然估计序列S：根据

获得由M个变量组成的参数序列

经过QAM映射后的参数序列

更替似然估计序列S中的参数

获得更新后的似然估计序列；经过N_t次更新后获得完成一次迭代的似然估计序列S'， $S^{\&prime;}=[S_1^{\left(t\right)t_1},\&CenterDot;\&CenterDot;\&CenterDot;,S_k^{\left(t\right)t_1},\&CenterDot;\&CenterDot;\&CenterDot;,S_{N_t}^{\left(t\right),t_1}];$ Step 4: The Gibbs sampling update unit calculates the maximum likelihood estimate of the unit output according to the log-likelihood ratio

Update the likelihood estimation sequence S: According to

Get a parameter sequence consisting of M variables

Parameter sequence after QAM mapping

Replacement Likelihood Estimation of Parameters in Sequence S

Obtain the updated likelihood estimation sequence; after N _t updates, obtain the likelihood estimation sequence S' that completes one iteration, $S^{\&prime;}=[S_1^{\left(t\right){\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">t</figure-callout>}}_1},\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;,S_k^{\left(t\right){\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">t</figure-callout>}}_1},\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;,S_{N_t}^{\left(t\right),t_1}];$

Step 5: Perform step 3 and step 4 in a loop, iteratively update the likelihood estimation sequence S' after one iteration until the number of iterations reaches the set value, and obtain the final likelihood estimation sequence S _z ' after the iterative update is completed , and the final likelihood estimation sequence S _z ' is sent to the LLR calculation unit to obtain the final maximum likelihood estimation value and output it.

In the above method, in step 2, in the first probabilistic complex multiplier and the second probabilistic complex multiplier, the first probabilistic real multiplier inputs the real part random sequence of the multiplicand and the real part random sequence of the input multiplier Perform a multiplication operation, and output the operation result to the first probability real number adder; the second probability real number multiplier multiplies the imaginary part random sequence of the input multiplicand and the imaginary part random sequence of the input multiplier, and the operation result is taken After inversion, output to the first probability real number adder; the first probability real number adder sums the two input quantities to obtain the real part random sequence of the output quantity; the third probability real number multiplier inputs the real part random sequence of the multiplicand Perform multiplication with the random sequence of the imaginary part of the input multiplier, and output the operation result to the second probability real number adder; the fourth probability real number multiplier randomizes the random sequence of the imaginary part of the input multiplicand and the real part of the input multiplier The sequence is multiplied, and the operation result is output to the second probabilistic real-complex adder; the second probable real-complex adder sums the two input quantities to obtain a random sequence of the imaginary part of the output quantity.

Compared with prior art, the beneficial effect of the present invention:

The multi-input multi-output detector based on probability calculation of the present invention performs similarity calculation on the restored transmitted signal and received signal to obtain likelihood information, and can be directly connected with a decoder for use.

The multi-input multi-output detector based on probability calculation of the present invention uses probability calculation to realize MCMC algorithm, and the addition operation and multiplication operation involved in Gibbs sampling update can be realized by a small number of simple logic units, which greatly reduces the operational complexity and structural complexity .

In the traditional MCMC algorithm, an additional circuit is needed to simulate a Gaussian white noise signal to increase the transition probability and reduce the locked state. The multi-input multi-output detector of the present invention uses probability calculation to realize the MCMC algorithm, and the probability calculation itself can generate white noise calculation variance, which can improve the transition probability of the Markov chain in the MCMC algorithm, so it can intelligently solve the problem of high SNR (SNR) ) The problem of lower locking avoids additional additional circuits and simplifies the structure of the MIMO detector.

Using the sliding window generation sequence method (SWG) for Gibbs sampling update, the sliding window method only uses the statistical value of part of the probability sequence, not all the probabilities, so its calculation cycle is reduced from the full sequence length to the window length to reduce the calculation cycle , to speed up the operation.

Applying the multi-input multi-output detector of the present invention to construct a fully parallel detection system can realize the construction of a fully parallel detection system in a fully parallel manner, and can achieve a higher throughput rate.

Description of drawings:

Fig. 1 is a structural block diagram of the MIMO detector of the present invention.

Figure 2 is a block diagram of the structure of the random sequence generator in the detector.

Figure 3 is a block diagram of the structure of the probabilistic complex multiplier in the detector.

Figure 4 is a flow chart of signal processing for Gibbs sampling update in the sliding window generation sequence (SWG) method.

Fig. 5 is a block diagram of the composition of the logarithmic likelihood ratio (LLR) calculation unit in the detector.

Fig. 6 is a block diagram of the structure of the all-parallel MIMO detector of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with test examples and specific embodiments. However, it should not be understood that the scope of the above subject matter of the present invention is limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

With reference to Fig. 1, the multi-input multi-output detector based on the calculation of probability that the present embodiment enumerates comprises three random sequence generators, is respectively the first random sequence generator 103, the second random sequence generator 108, the third random sequence generator A matrix QR decomposer 104; two probability complex multipliers, respectively the first probability complex multiplier 112 and the second probability complex multiplier 119; Gibbs sampling update unit 113; log likelihood ratio (LLR) calculation Unit 115.

为布尔型概率序列。用于MIMO信号检测的信道矩阵H经过矩阵QR分解器104完成QR分解，得到子矩阵Q和子矩阵R，其中，Q为正交矩阵，R为上三角矩阵，子矩阵R经过第三随机序列生成器109形成满足概率计算的随机序列

子矩阵Q经过矩阵求逆运算得到表征矩阵Q^H，表征矩阵Q^H再经过第二随机序列生成器108形成满足概率计算的随机序列

随机序列

和信号序列一起被送入第一概率复数乘法器112，得到

随机序列

和Gibbs采样更新单元113输出的似然估计序列S（初始化时，Gibbs采样更新单元没有输出，因此预输入任意值组成的S序列作为似然估计序列的初始序列）一起被送入第二概率复数乘法器119得到

和一起被送入对数似然比计算单元进行最大似然估计值运算，求取到的最大似然估计值传输至Gibbs采样更新单元113，进行蒙特卡洛马尔科夫链算法中的Gibbs采样更新运算，得到更新后的似然估计序列S，更新后的似然估计序列S再传输至对数似然比计算单元，与

和

一起进行最大似然估计值运算。对数似然比计算单元运算得到的最大似然估计值传输至Gibbs采样更新单元进行似然估计序列更新，更新后的似然估计序列再返回至对数似然比计算单元参与最大似然估计值运算，如此迭代更新，直至迭代检测完成后，即迭代次数达到设定值（迭代次数根据实际信道条件而设定，迭代次数越多，获取的似然估计值越精确，通常的，设置迭代次数为24次即可），由LLR计算单元115输出发送信息的最大似然估计值

该最大似然估计值即为多输入多输出检测器的输出。The input signal of the MIMO detector based on probability calculation is the received signal Z received by the receiving antenna and the channel matrix H. The channel matrix H has N _r rows and N _t columns, N _r is the number of transmitting antennas, and N _t is the receiving antenna number. The received signal Z passes through the first random sequence generator 103 to form a signal sequence satisfying probability calculation The meaning of the sequence satisfying the probability calculation is a Boolean probability sequence used for probability calculation, that is to say, the signal sequence

is a Boolean probability sequence. The channel matrix H used for MIMO signal detection is decomposed through the matrix QR decomposer 104 to obtain sub-matrix Q and sub-matrix R, wherein Q is an orthogonal matrix, R is an upper triangular matrix, and sub-matrix R is generated through a third random sequence The device 109 forms a random sequence satisfying the probability calculation

The sub-matrix Q undergoes a matrix inversion operation to obtain the representation matrix Q ^H , and the representation matrix Q ^H passes through the second random sequence generator 108 to form a random sequence that satisfies the probability calculation

random sequence

and signal sequence are sent to the first probability complex multiplier 112 together to obtain

random sequence

Together with the likelihood estimation sequence S output by the Gibbs sampling update unit 113 (at initialization, the Gibbs sampling update unit has no output, so the S sequence composed of pre-input arbitrary values is used as the initial sequence of the likelihood estimation sequence), it is sent into the second probability complex number Multiplier 119 gets

and Together, they are sent to the logarithmic likelihood ratio calculation unit to perform the maximum likelihood estimation value calculation, and the obtained maximum likelihood estimation value is transmitted to the Gibbs sampling update unit 113 for Gibbs sampling update in the Monte Carlo Markov chain algorithm operation to obtain the updated likelihood estimation sequence S, and then transmit the updated likelihood estimation sequence S to the logarithmic likelihood ratio calculation unit, and

and

together for maximum likelihood estimation. The maximum likelihood estimation value obtained by the log likelihood ratio calculation unit is transmitted to the Gibbs sampling update unit for the likelihood estimation sequence update, and the updated likelihood estimation sequence is returned to the log likelihood ratio calculation unit to participate in the maximum likelihood estimation Value operation, such iterative update, until the iterative detection is completed, that is, the number of iterations reaches the set value (the number of iterations is set according to the actual channel conditions, the more iterations, the more accurate the likelihood estimate obtained, usually, set the iteration The number of times is 24 times), and the maximum likelihood estimation value of the transmitted information is output by the LLR calculation unit 115

The maximum likelihood estimate is the output of the multiple-input multiple-output detector.

需要说明的是，针对复数信号，复数信号分别用两路信号x'₀和x″₀表示其实部和虚部，分别针对实部和虚部，获取实部的值序列、符号序列，及获取虚部的值序列、符号序列。Referring to Fig. 2, three random sequence generators (103, 108, 109) have the same structure, the input signal of the first random sequence generator 103 is the received signal Z received by the receiving antenna, the input of the second random sequence generator 108 The signal is a sub-matrix Q obtained through a matrix inversion operation to obtain a representation matrix Q ^H , and the input signal of the third random sequence generator 109 is a sub-matrix R obtained through QR decomposition. In order to facilitate the description of the structure of the random sequence generator, the random sequence generated The input of the device is collectively referred to as the input signal x ₀ . If the input signal is a complex signal, the complex signal is represented by two signals x' ₀ and x″ ₀ , respectively representing the real part and the imaginary part. If the input signal is a real signal, Then, it is represented by a signal (equivalent to its imaginary part being zero). The dynamic range of the input signal _x0 is (-m, m), and the absolute value operation is performed through the absolute value operation module 202 to obtain the absolute value of the input signal _x0 |x ₀ |. For each clock, the absolute value |x ₀ | and a random number that satisfies a uniform distribution in the interval [0, m) are sent to the comparator 205 for comparison, and the comparison output value 0 or 1. The comparison rule is: if the absolute value |x ₀ | is greater than the random number, output 1, otherwise output 0. As the clock advances, the value sequence value(x) composed of 1 and 0 is obtained, and its length is 32, such as The value(x) sequence 010001110011... shown in Figure 2 is obtained from the actual simulation results. Corresponding to each bit in the value sequence value(x), the symbol sequence sign(x) consisting of 0 or 1 is obtained according to the sign rule ), its length is also 32, as shown in Figure 2, the sequence 000000000000.... The sign rule is: if x ₀ is less than 0, then output 1, otherwise output 0. Both value(x) sequence and sign(x) sequence are Boolean type sequence vector, value(x) sequence and sign(x) sequence together constitute a probability random sequence

It should be noted that, for the complex signal, the real part and the imaginary part of the complex signal are respectively represented by two signals x' ₀ and x″ ₀ , and the value sequence and the symbol sequence of the real part are obtained for the real part and the imaginary part respectively, and the acquisition A sequence of values, a sequence of signs, for the imaginary part.

第二概率复数乘法器的两个输入量分别为来自第三随机序列生成器的

及来自Gibbs采样更新单元113的更新后的似然估计序列（第一次进行复数乘法运算时，由于Gibbs采样更新单元113没有输出，因此，第一次进行复数乘法运算时，以给定的随机序列S为输入量）

针对概率复数乘法器的两个输入量，其中一个输入量称为输入乘数，另一个输入量称为输入被乘数，其中输入乘数和输入被乘数均分别由实部随机序列和虚部随机序列构成，输入乘数可表示为a+jb，输入被乘数表示为c+jd，a和c分别表示实部随机序列，b和d分别表示虚部随机序列，j为虚数单位。概率复数乘法器对两个输入量进行乘法运算，得到x+jy=(a+jb)(c+jd)，x为概率复数乘法器输出的实部随机序列，y为概率复数乘法器输出的虚部随机序列。输入被乘数的实部随机序列a和输入乘数的实部随机序列c一起经过概率实数乘法器308进行乘法运算，运算结果输入至概率实数加法器310；输入被乘数的虚部随机序列b和输入乘数的虚部随机序列d一起经过概率实数乘法器309进行乘法运算，运算结果再经过取反电路进行取反，取反后的运算结果再输入至概率实数加法器310；概率实数加法器310输出的x为概率复数乘法器输出的实部随机序列。输入被乘数的实部随机序列a和输入乘数的虚部随机序列d一起经过概率实数乘法器313进行乘法运算，运算结果输入至概率实数加法器314；输入被乘数的虚部随机序列b和输入乘数的实部随机序列c一起经过概率实数乘法器312进行乘法运算，运算结果输入至概率实复数加法器314；概率实数加法器314输出的y为概率复数乘法器输出的虚部随机序列。Referring to Fig. 3, the input quantity of the probability complex multiplier is the Boolean probability sequence generated by the random sequence generator, that is to say, the probability complex multiplier uses the probability calculation method to complete complex multiplication. The probability complex multiplier is composed of four probability real number multipliers (308, 309, 312, 313), two probability real number adders (310, 314) and a negation circuit, and the probability real number multipliers (308, 309, 312, 313) and probabilistic real number adders (310, 314) are realized by simple logic gates. There are two input quantities of the probability complex multiplier, and the two input quantities of the first probability complex multiplier are respectively from the first random sequence generator and from the second random sequence generator

The two inputs of the second probabilistic complex multiplier are respectively from the third random sequence generator

and the updated likelihood estimation sequence from the Gibbs sampling update unit 113 (when the complex multiplication is performed for the first time, since the Gibbs sampling update unit 113 has no output, when the complex multiplication is performed for the first time, the given random Sequence S is the input quantity)

For the two input quantities of the probabilistic complex multiplier, one of the input quantities is called the input multiplier, and the other input quantity is called the input multiplicand, where the input multiplier and the input multiplicand are respectively composed of real random sequences and imaginary The input multiplier can be expressed as a+jb, the input multiplicand can be expressed as c+jd, a and c respectively represent the real part random sequence, b and d represent the imaginary part random sequence, and j is the imaginary unit. The probability complex multiplier multiplies two input quantities to obtain x+jy=(a+jb)(c+jd), x is the real part random sequence output by the probability complex multiplier, and y is the output of the probability complex multiplier The imaginary random sequence. The random sequence a of the real part of the input multiplicand and the random sequence c of the real part of the input multiplier are multiplied through the probability real number multiplier 308, and the result of the operation is input to the probability real number adder 310; the random sequence of the imaginary part of the input multiplicand b and the random sequence d of the imaginary part of the input multiplier are multiplied through the probability real number multiplier 309, and the result of the operation is reversed through the negation circuit, and the result of the negation is then input to the probability real number adder 310; the probability real number The x output by the adder 310 is the real part random sequence output by the probabilistic complex multiplier. The random sequence a of the real part of the input multiplicand and the random sequence d of the imaginary part of the input multiplier are multiplied through the probability real number multiplier 313, and the operation result is input to the probability real number adder 314; the random sequence of the imaginary part of the input multiplicand b and the real part random sequence c of the input multiplier are multiplied through the probability real number multiplier 312, and the operation result is input to the probability real complex number adder 314; the y output by the probability real number adder 314 is the imaginary part output by the probability complex number multiplier random sequence.

基于概率计算，仅需要较少的逻辑运算单元即完成乘积运算，降低了运算复杂度，简化了检测器的结构。In the random sequence generator, the absolute value of the input signal is compared with a random number that satisfies a uniform distribution in the interval [0,m), and a sequence of values consisting of 0 or 1 is obtained. In the probabilistic complex multiplier, two input quantities are multiplied. In the sequence of the two input quantities, the product is 1 only if the two values at the corresponding positions are both 1. The value 0 or 1 in the value sequence is obtained by comparing the absolute value of the input signal with a random number in the interval [0, m), then the probability of obtaining a value of 1 is

Based on the probability calculation, only fewer logic operation units are needed to complete the product operation, which reduces the operation complexity and simplifies the structure of the detector.

For the convenience of description, the two input quantities of the probability real number multiplier (308, 309, 312, 313) (input the real part random sequence a of the multiplicand and the real part random sequence c of the input multiplier, or input the multiplicand The random sequence b of the imaginary part of the input multiplier and the random sequence d of the imaginary part of the input multiplier, or the random sequence a of the real part of the input multiplicand and the random sequence d of the imaginary part of the input multiplier, or the random sequence b of the imaginary part of the input multiplicand and the real part random sequence c) of the input multiplier is expressed as x1 and y1, and the probability real number multiplier performs the following logic operations on the two input random sequences x1 and y1:

sign(z1)=sign(x1)XORsign(y1)

value(z1)=value(x1)ANDvalue(y1)

sign(z1) and value(z1) are the sign vector and the absolute value vector obtained after the two input quantities are operated by the probability real number multiplier respectively.

For the convenience of description, the two input quantities of the probabilistic real number adder (310, 314) are expressed as x2 and y2, and the following logical operations are performed on the two input random sequences x2 and y2:

sign(z2)=(value(x2)ANDsign(y2))OR(value(y2)ANDsign(x2))

value(z2)=(value(x2)XNORvalue(y2))OR(value(x2)AND(sign(y2)XORsign(x2)))

sign(z2) and value(z2) are the sign vector and absolute value vector obtained after the two input quantities are operated by the probabilistic real number adder respectively.

For the convenience of description, the input quantity of the inversion circuit is expressed as x3, and the inversion circuit performs the following logical operation on the input random sequence x3:

sign(z3)=NOT(sign(x3))

value(z3)=value(x3)

sign(z3) and value(z3) are the sign vector and absolute value vector obtained after the input quantity undergoes negation circuit operation respectively.

来自第二概率复数乘法器的

和来自Gibbs采样更新单元更新后的似然估计序列S，三个输入量均以随机序列表征。为了保证以较短的序列获得满足系统性能要求的计算精度，对

和

进行滑窗处理，即是说，以时间长度t₁为窗口，分别选取和序列中时间窗口t₁内截取的数据进行对数似然比计算。The process of log-likelihood ratio calculation unit and Gibbs sampling update will be described below with reference to FIG. 4 and FIG. 5 . Referring to Fig. 5, the three input quantities of the LLR calculation unit are respectively from the first probability complex multiplier

from the second probability complex multiplier

and the updated likelihood estimation sequence S from the Gibbs sampling update unit, the three input quantities are characterized by random sequences. In order to ensure that the calculation accuracy that meets the system performance requirements is obtained in a short sequence, the

and

Carry out sliding window processing, that is to say, take the time length t ₁ as the window, select and The data intercepted in the time window _t1 in the sequence are calculated by logarithmic likelihood ratio.

With reference to Figure 5, the following calculations are performed in the LLR calculation unit:

$\mathrm{<span\; class="notranslate">\; LLR</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="b\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">b</figure-callout></span><figure-callout\; id="1"\; label="b\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">\; </figure-callout>}}^{{\mathrm{<span\; class="notranslate">\; </span>}}_{}}<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}^{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; min</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Q</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}^{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; )</span>}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="R\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">R</figure-callout></span><figure-callout\; id="1"\; label="R\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">\; </figure-callout>}}^{{\mathrm{<span\; class="notranslate">\; </span>}}_{}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; )</span>}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; <figure-callout\; id="1"\; label="k\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">k</figure-callout></span><figure-callout\; id="1"\; label="k\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">\; </figure-callout>}}^{{\mathrm{<span\; class="notranslate">\; </span>}}_{}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}^{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

表示

序列中每个数值与S序列中每个数值相乘，

表示

序列中与S序列中参数

相对应的第k个部分（每一个部分包含的数值个数与参数

中数值个数相同）的每个数值与参数

中每个数值相乘。LLR计算单元由三个子运算单元（501、502、503）组成，其中，子运算单元501完成

的计算工作，具体的，N_t（N_t为接收天线数）个输入量

经过一概率复数加法器进行求和运算，再经过取反电路取反，取反后的值与

一起输入一个概率复数加法器进行求和运算，运算结果作为一个输入量分别输入至M（M为一个时间窗口t₁内截取的数据个数）路概率复数加法器，在各路概率复数加法器中，运算结果与

一起进行加法运算，得到的运算值

再经过取绝对值||·||得到

输出至子运算单元502。子运算单元502完成对Nt个

值进行求和运算，具体的，先求取出Nt个输入量

的平方，求取平方后N_t个输入量经过一个概率实数加法器进行加法运算，得到的

再将求得的和由概率序列转换为二进制数，并输出至存储器。子运算单元501和子运算单元502分别针对每一个似然估计序列S中的参数

求取出其

并将求得的和由概率序列转换为二进制数，输出至存储器存储。子运算单元503完成

的计算，即是说，子运算单元503从存储器中存储的N_t个

中，求取与接收信号偏差最小的值，即最大似然估计值

然后将

输出至Gibbs采样更新单元。In the formula,

express

Each value in the sequence is multiplied by each value in the S sequence,

express

In-sequence and S-sequence parameters

The corresponding kth part (the number of values and parameters contained in each part

The same number of values in the middle) each value and parameter

Multiply each value in . The LLR calculation unit is composed of three sub-operation units (501, 502, 503), among which, the sub-operation unit 501 completes

The calculation work, specifically, N _t (N _t is the number of receiving antennas) input quantities

After a probabilistic complex number adder for summation operation, and then through the inversion circuit for inversion, the value after inversion is the same as

Input a probabilistic complex adder together for summing operation, and the operation result is input as an input to M (M is the number of data intercepted in a time window t ₁ ) probabilistic complex adder, and each probabilistic complex adder , the result of the operation is the same as

Perform addition operations together to obtain the calculated value

Then take the absolute value||·|| to get

output to the sub-operation unit 502. The sub operation unit 502 completes the pair of Nt

Values are summed, specifically, Nt input values are calculated first

The square of , after calculating the square, N _t input quantities are added through a probabilistic real number adder, and the obtained

Then the obtained sum is converted from a probability sequence into a binary number and output to the memory. The sub-operation unit 501 and the sub-operation unit 502 are respectively for each parameter in the likelihood estimation sequence S

Find it

And the calculated sum is converted from a probability sequence into a binary number and output to a memory for storage. The sub operation unit 503 completes

The calculation of , that is to say, the sub-operation unit 503 stores N _t from the memory

Among them, find the value with the smallest deviation from the received signal, that is, the maximum likelihood estimate

Then

Output to Gibbs sample update unit.

进行采样更新，根据

得到似然估计序列S中更新后的

即是说，当来自LLR计算单元的最大似然估计值

小于零时x_k,b=1，当

大于零时x_k,b=-1，当

等于零时，随机的x_k,b=1或x_k,b=-1，evenly的含义是“随机的”。由M个x_k,b组成序列[x_k,1,x_k,2...,x_k,b,...x_k,M]，[x_k,1,x_k,2...,x_k,b,...x_k,M]经过QAM映射后得到

替代原似然估计序列S中的参数S1，原似然估计序列S中的其他参数不变，得到第一次更新后的似然估计序列S，

更新后的似然估计序列输出至第二概率复数乘法单元，与

进行复数乘法运算，得到

在第一次采样更新时，似然估计序列S为任意给定的随机序列，由

组成。更新后的似然估计序列S与经过第二次滑窗处理后的

一起在LLR计算单元中进行似然估计值运算，获取最大似然估计值，并传输至Gibbs采样更新单元。更新后的似然估计序列S在Gibbs采样更新单元中进行第二次Gibbs采样运算，得到更新后的

即是说，第一次更新后的似然估计序列S中的S₂由再次更新后的

替代，得到再次更新后的似然估计序列S，

如此，进行至第N_t次Gibbs采样运算后，得到更新后的

即完成整个似然估计序列S中所有的参数更新，得到完全更新后的似然估计序列S'， $S^{\&prime;}=[S_1^{\left(t\right),t_1,}{S}_2^{\left(t\right),t_1},...,S_k^{\left(t\right),t_1},...,S_{N_t}^{\left(t\right),t_1}],$ 完成一次迭代更新。完成一次迭代更新后，对更新后的似然估计序列S'再次进行迭代更新，直至迭代次数达到设定值，得到最终更新后的似然估计序列S'_z，似然估计序列S'_z再被送至LLR计算单元，求取出最终的最大似然估计值输出。多输入多输出检测器与译码器连接后，多输入多输出检测器将求取的最终的最大似然估计值输出至译码器。The Gibbs sampling update is a classic Monte Carlo Markov chain (MCMC) algorithm. The present invention uses the random sequence truncated by the sliding window sequence generation (SWG) method to complete the Gibbs sampling and update operation by means of probability calculation. Referring to Figure 4, in the Gibbs sampling update unit, according to the maximum likelihood estimation value from the LLR calculation unit

Sample updates are performed according to

Get the updated likelihood estimation sequence S

That is, when the maximum likelihood estimate from the LLR computation unit

When x _k,b =1 is less than zero, when

When x _k,b =-1 is greater than zero, when

When equal to zero, random x _k,b =1 or x _k,b =-1, evenly means "random". A sequence [x _{k ,1} ,x _k,2 ...,x _k,b ,...x _k,M ] consists of M x k,b, [x _k,1 ,x _k,2 ... ,x _k,b ,...x _k,M ] are obtained after QAM mapping

Substituting the parameter S1 in the original likelihood estimation sequence S, and keeping other parameters in the original likelihood estimation sequence S unchanged, the likelihood estimation sequence S after the first update is obtained,

The updated likelihood estimation sequence is output to the second probability complex multiplication unit, and

Perform complex multiplication to get

In the first sampling update, the likelihood estimation sequence S is any given random sequence, given by

composition. The updated likelihood estimation sequence S and after the second sliding window processing

Together, the likelihood estimation value operation is performed in the LLR calculation unit to obtain the maximum likelihood estimation value, and is transmitted to the Gibbs sampling update unit. The updated likelihood estimation sequence S performs the second Gibbs sampling operation in the Gibbs sampling update unit to obtain the updated

That is to say, S ₂ in the likelihood estimation sequence S after the first update is determined by the updated

Instead, get the updated likelihood sequence S again,

In this way, after the N _t Gibbs sampling operation, the updated

That is to complete all the parameter updates in the entire likelihood estimation sequence S, and obtain the completely updated likelihood estimation sequence S', $S^{\&prime;}=[S_1^{\left(t\right),{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">t</figure-callout>}}_1,}{S}_2^{\left(t\right),{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">t</figure-callout>}}_1},...,S_k^{\left(t\right),{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">t</figure-callout>}}_1},...,S_{N_t}^{\left(t\right),t_1}],$ Complete an iterative update. After completing an iterative update, iteratively update the updated likelihood estimation sequence S' again until the number of iterations reaches the set value, and obtain the final updated likelihood estimation sequence S' _z , and then the likelihood estimation sequence S' _z It is sent to the LLR calculation unit to obtain the final maximum likelihood estimation value output. After the MIMO detector is connected to the decoder, the MIMO detector outputs the obtained final maximum likelihood estimation value to the decoder.

和

再广播至L路并行的基于概率计算的多输入多输出检测器。

和Z₁在第一路MIMO检测器607求解出发送信号的最大似然估计值；和Z₂在第二路MIMO检测器604求解出发送信号的最大似然估计值；如此并行完成L路MIMO信号检测，获得提供给译码器608使用的代表一个完整发送信道码码字的最大似然估计值，由译码器608完成信道译码，获得通信所传输的数据信息。全并行检测系统可以处理L路并行的MIMO符号，L由信道编码长度、调制方式、MIMO规模共同决定，使得L路并行检测的结果可以支持一个完整的信道码码字进行并行译码，信道译码由译码器608完成，可以是LDPC、Turbo、卷积码、BHC码、RS码等纠错编码形式。The invention also provides a fully parallel detection system constructed by applying the probability calculation-based MIMO detector of the invention. The full parallel detection system includes multiple probability-based multiple-input multiple-output detectors (604, 606, 607 in the figure), and each probability-based multiple-input multiple-output detector is connected in parallel with a decoder 608 , each multi-input multi-output detector based on probability calculation outputs the obtained final maximum likelihood estimation value to the decoder. Since each MIMO detector based on probability calculation will perform QR decomposition on the channel matrix H, a common matrix preprocessing unit is used to reduce the amount of computation and improve processing efficiency. Referring to Fig. 6, the matrix preprocessing unit 601 is connected to L multiple-input multiple-output detectors based on probability calculation, the matrix preprocessing unit includes a matrix QR decomposer and two random sequence generators, and the channel matrix H is decomposed by the matrix QR decomposer For sub-matrix Q and sub-matrix R, sub-matrix R generates a random sequence through a random sequence generator The sub-matrix Q generates a random sequence through a random sequence generator

and

Then broadcast to L-way parallel MIMO detectors based on probability calculation.

and _Z1 solve the maximum likelihood estimation value of the transmitted signal in the first MIMO detector 607; and Z ₂ solve the maximum likelihood estimation value of the transmitted signal in the second MIMO detector 604; in this way, the L-channel MIMO signal detection is completed in parallel, and the maximum value representing a complete transmitted channel code word provided to the decoder 608 is obtained. For the likelihood estimation value, the decoder 608 completes the channel decoding to obtain the data information transmitted by communication. The full parallel detection system can process L channels of parallel MIMO symbols, and L is determined by the channel coding length, modulation mode, and MIMO scale, so that the results of L channels of parallel detection can support a complete channel code word for parallel decoding, channel decoding The code is completed by the decoder 608, and can be in the form of error correction codes such as LDPC, Turbo, convolutional code, BHC code, and RS code.

The multiple-input multiple-output detector of the present invention can be implemented in the form of an application-specific integrated circuit, or in the form of a programmable logic gate array, or in the form of a programmable general-purpose microprocessor circuit.

The present invention also provides a detection method based on a probability calculation MIMO detector, comprising the following steps:

并输出至第二概率复数乘法器；接收信号Z经过第一随机序列生成器生成信号序列

并输出至第一概率复数乘法器。Step 1: The matrix QR decomposer decomposes the channel matrix H into sub-matrices Q and R, and the sub-matrix R generates a random sequence through the third random sequence generator And output to the first probability complex multiplier, the sub-matrix Q undergoes matrix inversion operation to obtain the representation matrix Q ^H , and the representation matrix Q ^H generates a random sequence through the second random sequence generator

And output to the second probability complex multiplier; the received signal Z generates a signal sequence through the first random sequence generator

And output to the first probability complex multiplier.

进行复数乘法运算，获得

并输出至对数似然比计算单元；第二概率复数乘法器对

和Gibbs采样更新单元输出的更新后的似然估计序列S进行复数乘法运算，获得

首次进行复数乘法运算时，由于Gibbs采样更新单元没有输出，因此，进行复数乘法运算的是任意给出的S序列。Step 2: First Probabilistic Complex Multiplier Pair and

Perform complex multiplication to obtain

And output to the logarithmic likelihood ratio computing unit; The second probability complex multiplier pairs

Perform complex multiplication with the updated likelihood estimation sequence S output by the Gibbs sampling update unit to obtain

When the complex multiplication operation is performed for the first time, since the Gibbs sampling update unit has no output, the complex multiplication operation is an arbitrarily given S sequence.

Specifically, in the first probabilistic complex multiplier and the second probabilistic complex multiplier, the first probabilistic real multiplier multiplies the real part random sequence of the input multiplicand and the real part random sequence of the input multiplier, and performs the operation The result is output to the first probability real number adder; the second probability real number multiplier multiplies the imaginary part random sequence of the input multiplicand and the imaginary part random sequence of the input multiplier, and the operation result is inverted and output to the first probability Real number adder; the first probability real number adder sums the two input quantities to obtain the real part random sequence of the output quantity; the third probability real number multiplier inputs the real part random sequence of the multiplicand and the imaginary part of the input multiplier The random sequence is multiplied, and the result of the operation is output to the second probability real number adder; the fourth probability real number multiplier multiplies the imaginary part random sequence of the input multiplicand and the real part random sequence of the input multiplier, and The operation result is output to the second probabilistic real-complex adder; the second probable real-complex adder sums the two input quantities to obtain a random sequence of the imaginary part of the output quantity.

和进行滑窗处理后，与更新后的似然估计序列S进行 $\mathrm{LLR}\left(X_{k,b}^{t_1}\right|Z^{t_1})=\min {\mathrm{\&Sigma;}}_{i=1}^{N_t}{\left|\right|{(Q^{H,t}\&CenterDot;Z^{t_1})}_i-{(R^{t_1}\&CenterDot;S)}_i-R_{i,k}^{t_1}\&CenterDot;S_k^{t_1}\left(x\right)\left|\right|}^2$ 运算，求取最大似然估计值

并将求取的最大似然估计值

输出至Gibbs采样更新单元；循环执行Gibbs采样更新和似然估计运算，直至迭代次数达到设定值，对数似然比计算单元则直接输出最终得到的最大似然估计值。Step 3: Calculate the log-likelihood ratio for the input

and After the sliding window processing, and the updated likelihood estimation sequence S $\mathrm{LLR}\left(x_{k,\mathrm{<figure-callout\; id="1"\; label="b\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">b</figure-callout>}}^{t_{}}\right|Z^{t_1})=\min {\mathrm{\&Sigma;}}_{i=1}^{N_t}{\left|\right|{(Q^{h,t}\&CenterDot;Z^{t_1})}_i-{({\mathrm{<figure-callout\; id="1"\; label="R\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">R</figure-callout>}}^{t_{}}\&CenterDot;S)}_i-R_{i,\mathrm{<figure-callout\; id="1"\; label="k\; t"\; filenames="HDA0000459684140000012.png"\; state="\{\{state\}\}">k</figure-callout>}}^{t_{}}\&CenterDot;S_k^{t_1}\left(x\right)\left|\right|}^2$ operation to find the maximum likelihood estimate

and obtain the maximum likelihood estimate

The output is sent to the Gibbs sampling update unit; the Gibbs sampling update and likelihood estimation operation are performed cyclically until the number of iterations reaches the set value, and the log-likelihood ratio calculation unit directly outputs the final maximum likelihood estimation value.

得到似然估计序列S中更新后的即是说，当来自LLR计算单元的最大似然估计值

小于零时x_k,b=1，当大于零时x_k,b=-1，当

等于零时，随机的x_k,b=1或x_k,b=-1，evenly的含义是“随机的”。由M个x_k,b组成序列[x_k,1,x_k,2...,x_k,b,...x_k,M]，[x_k,1,x_k,2...,x_k,b,...x_k,M]经过QAM映射后得到

替代原似然估计序列S中的参数S₁，原似然估计序列S中的其他参数不变，得到第一次更新后的似然估计序列S，

更新后的似然估计序列输出至第二概率复数乘法单元，与

进行复数乘法运算。第一次采样更新时，似然估计序列S为任意给定的随机序列，由

组成。更新后的似然估计序列S与经过第二次滑窗处理后的

一起在LLR计算单元中进行似然估计值运算，获取最大似然估计值，并传输至Gibbs采样更新单元。更新后的似然估计序列S在Gibbs采样更新单元中进行第二次Gibbs采样运算，得到更新后的

即是说，第一次更新后的似然估计序列S中的S2由再次更新后的

替代，得到再次更新后的似然估计序列S，

如此，进行至第N_t次Gibbs采样运算后，得到更新后的

即完成整个似然估计序列S中所有的参数更新，得到完全更新后的似然估计序列S'，

完成一次更新迭代。完成一次更新迭代后，对更新后的似然估计序列S'再次进行迭代更新，直至迭代次数达到设定值，得到最终更新后的似然估计序列S'_z，似然估计序列S'_z再被送至LLR计算单元，求取出最终的最大似然估计值输出。多输入多输出检测器与译码器连接后，多输入多输出检测器将求取的最终的最大似然估计值输出至译码器。Step 4: The Gibbs sampling update unit updates the likelihood estimation sequence S according to the maximum likelihood estimation value, according to

Get the updated likelihood estimation sequence S That is, when the maximum likelihood estimate from the LLR computation unit

When x _k,b =1 is less than zero, when When x _k,b =-1 is greater than zero, when

When equal to zero, random x _k,b =1 or x _k,b =-1, evenly means "random". A sequence [x _{k ,1} ,x _k,2 ...,x _k,b ,...x _k,M ] consists of M x k,b, [x _k,1 ,x _k,2 ... ,x _k,b ,...x _k,M ] are obtained after QAM mapping

Substituting the parameter S ₁ in the original likelihood estimation sequence S, and keeping other parameters in the original likelihood estimation sequence S unchanged, the likelihood estimation sequence S after the first update is obtained,

The updated likelihood estimation sequence is output to the second probability complex multiplication unit, and

Perform complex multiplication. When the first sampling is updated, the likelihood estimation sequence S is any given random sequence, given by

composition. The updated likelihood estimation sequence S and after the second sliding window processing

Together, the likelihood estimation value operation is performed in the LLR calculation unit to obtain the maximum likelihood estimation value, and is transmitted to the Gibbs sampling update unit. The updated likelihood estimation sequence S performs the second Gibbs sampling operation in the Gibbs sampling update unit to obtain the updated

That is to say, S2 in the likelihood estimation sequence S after the first update is determined by the updated

Instead, get the updated likelihood sequence S again,

In this way, after the N _t Gibbs sampling operation, the updated

That is to complete all the parameter updates in the entire likelihood estimation sequence S, and obtain the completely updated likelihood estimation sequence S',

Complete an update iteration. After completing an update iteration, iteratively update the updated likelihood estimation sequence S' again until the number of iterations reaches the set value, and obtain the final updated likelihood estimation sequence S' _z , and then the likelihood estimation sequence S' _z It is sent to the LLR calculation unit to obtain the final maximum likelihood estimation value output. After the MIMO detector is connected to the decoder, the MIMO detector outputs the obtained final maximum likelihood estimation value to the decoder.

All features disclosed in this specification, or steps in all methods or processes disclosed, may be combined in any manner, except for mutually exclusive features and/or steps.

Any feature disclosed in this specification (including any appended claims, abstract and drawings), unless expressly stated otherwise, may be replaced by alternative features which are equivalent or serve a similar purpose. That is, unless expressly stated otherwise, each feature is one example only of a series of equivalent or similar features.

Claims (9)

1. The multi-input multi-output detector based on probability calculation, is characterized in that, comprises a matrix QR decomposer, and described matrix QR decomposer connects respectively the second random sequence generator, the 3rd random sequence generator; The 3rd The random sequence generator is connected to the Gibbs sampling update unit through the second probability complex multiplier; the second random sequence generator is connected to the first probability complex multiplier, and the first probability complex multiplier is also connected to the first random sequence generator and The logarithmic likelihood ratio calculation unit; the operation result of the first probability complex multiplier is output to the logarithm likelihood ratio calculation unit, and the operation result of the Gibbs sampling update unit is respectively output to the logarithm likelihood ratio calculation unit and the second probability complex multiplication The output of the logarithmic likelihood ratio calculation unit is used as the input of the Gibbs sampling update unit. 2. the multiple-input multiple-output detector based on probability calculation according to claim 1, is characterized in that, the structure of described first random sequence generator, the second random sequence generator and the 3rd random sequence generator is identical, Both include an absolute value operation module and a comparator. The absolute value operation module calculates the absolute value of the input signal of each clock and outputs it to the comparator. The comparator compares the absolute value of the input signal with any one in the interval [0,m) Uniformly distributed random numbers are compared, if the absolute value of the input signal is greater than the random number, output 1, otherwise output 0, and obtain a value sequence consisting of 1 and 0; corresponding to each bit in the value sequence, if the input signal is less than 0 , then output 1, otherwise output 0, and obtain a symbol sequence consisting of 0 and 1 according to the value sequence. 3. the MIMO detector based on probability calculation according to claim 1, is characterized in that, the structure of described first probability complex multiplier and the second probability complex multiplier is identical, all is multiplied by four probability real numbers The four probability real number multipliers are respectively the first probability real number multiplier, the second probability real number multiplier, the third probability real number multiplier and the fourth probability real number multiplier ; The two probability real number adders are respectively the first probability real number adder and the second probability real number adder; the first probability real number multiplier is connected to the first probability real number adder, and the second probability real number multiplier is connected to the first probability real number multiplier by negating the circuit Probability real number adder; the third probability real number multiplier and the fourth probability real number multiplier are respectively connected with the second probability real number adder; the first random sequence output by the first random sequence generator

and the second random sequence output by the second random sequence generator

Obtain the first product sequence after complex multiplication

The third random sequence output by the third random sequence generator

The second product sequence is obtained after complex operation with the likelihood estimation sequence S output by the Gibbs sampling update unit

4. the multi-input multi-output detector based on probability calculation according to claim 3, is characterized in that, the first product sequence that described logarithmic likelihood ratio calculation unit outputs to the first probability complex multiplier and the second product sequence output by the second probabilistic complex multiplier

Carry out sliding window processing, and intercept the first product sequence respectively

and the second product sequence

The data in the middle time window _t1 , the intercepted data and the likelihood estimation sequence S output by the Gibbs sampling update unit perform logarithmic likelihood calculation to obtain the maximum likelihood estimation value

$\mathrm{LLR}\left(x_{k,b}^{t_1}\right|Z^{t_1})=\min {\mathrm{\&Sigma;}}_{i=1}^{N_t}{\left|\right|{(Q^{h,t}\&Center\; Dot;Z^{t_1})}_i-{(R^{t_1}\&Center\; Dot;S)}_i-R_{i,k}^{t_1}\&Center\; Dot;S_k^{t_1}\left(x\right)\left|\right|}^2,$ N _t is the number of receiving antennas,

is the likelihood estimate for the kth parameter in the sequence S. 5. the multiple-input multiple-output detector based on probability calculation according to claim 1, is characterized in that, described Gibbs sampling update unit according to Get a parameter sequence consisting of M variables

Parameter sequence after QAM mapping

Replacement Likelihood Estimation of Parameters in Sequence S

The updated likelihood estimation sequence is obtained, and M is the number of intercepted data in the time window _t1 . 6. A full-parallel detection system is characterized in that, comprising multiple MIMO detectors based on probability calculations according to claim 1, a plurality of MIMO detectors based on probability calculations are respectively connected with the same interpreter Encoders are connected in parallel. 7. the full parallel detection system according to claim 6, is characterized in that, described full parallel detection system also comprises matrix preprocessing unit, comprises matrix QR decomposer and two random sequence generators in the matrix preprocessing unit , the channel matrix H is decomposed into a sub-matrix Q and a sub-matrix R by a matrix QR decomposer, and the sub-matrix R generates a third random sequence through a random sequence generator

The sub-matrix Q generates a second random sequence through another random sequence generator

third random sequence

and the second random sequence

Broadcast to multiple parallel probability-based MIMO detectors. 8. the detection method of the MIMO detector based on probability calculation according to claim 1, is characterized in that, comprises the following steps: Step 1: The matrix QR decomposer decomposes the channel matrix H into sub-matrices Q and R, and the sub-matrix R generates a third random sequence through the third random sequence generator

And output to the first probability complex multiplier, the sub-matrix Q obtains the representation matrix Q ^H through the matrix inversion operation, and the representation matrix Q ^H generates the second random sequence through the second random sequence generator

And output to the second probability complex multiplier; the received signal Z generates the first random sequence through the first random sequence generator

And output to the first probability complex multiplier; Step 2: The first probabilistic complex multiplier against the second random sequence

and the first random sequence

Perform complex multiplication to obtain the first product sequence

and output to the logarithmic likelihood ratio calculation unit; the second probability complex multiplier is to the third random sequence Perform complex multiplication with the likelihood estimation sequence S output by the Gibbs sampling update unit to obtain the second product sequence

Step 3: The log-likelihood ratio calculation unit pairs the first product sequence and the second product sequence

Carry out sliding window processing, and intercept the first product sequence respectively

and the second product sequence

The data in the middle time window _t1 , the intercepted data and the likelihood estimation sequence S output by the Gibbs sampling update unit perform logarithmic likelihood calculation to obtain the maximum likelihood estimation value

$\mathrm{LLR}\left(x_{k,b}^{t_1}\right|Z^{t_1})=\min {\mathrm{\&Sigma;}}_{i=1}^{N_t}{\left|\right|{(Q^{h,t}\&Center\; Dot;Z^{t_1})}_i-{(R^{t_1}\&CenterDot;S)}_i-R_{i,k}^{t_1}\&Center\; Dot;S_k^{t_1}\left(x\right)\left|\right|}^2;$ Step 4: The Gibbs sampling update unit calculates the maximum likelihood estimate of the unit output according to the log-likelihood ratio

Update the likelihood estimation sequence S: According to

Get a parameter sequence consisting of M variables

Parameter sequence after QAM mapping

Replacement Likelihood Estimation of Parameters in Sequence S

Obtain the updated likelihood estimation sequence; after Nt updates, obtain the likelihood estimation sequence S' that completes one iteration, $S^{\&prime;}=[S_1^{\left(t\right)t_1},\&Center\; Dot;\&Center\; Dot;\&Center\; Dot;,S_k^{\left(t\right)t_1},\&CenterDot;\&Center\; Dot;\&Center\; Dot;,S_{N_t}^{\left(t\right),t_1}];$ Step 5: Perform step 3 and step 4 in a loop, iteratively update the likelihood estimation sequence S' after one iteration until the number of iterations reaches the set value, and obtain the final likelihood estimation sequence Sz' after the iterative update is completed, The final likelihood estimation sequence Sz' is then sent to the LLR calculation unit to obtain and output the final maximum likelihood estimation value. 9. detection method according to claim 8 is characterized in that, in described step 2, in the first probability complex multiplier and the second probability complex multiplier, the first probability real number multiplier is to the real of input multiplicand Part random sequence and the real part random sequence of the input multiplier are multiplied, and the operation result is output to the first probability real number adder; the second probability real number multiplier performs the multiplication operation on the input imaginary part random sequence of the The imaginary part random sequence is multiplied, and the result of the operation is inverted and then output to the first probability real number adder; the first probability real number adder sums the two input quantities to obtain the real part random sequence of the output quantity; the third probability real number The multiplier multiplies the random sequence of the real part of the input multiplicand and the random sequence of the imaginary part of the input multiplier, and outputs the operation result to the second probability real number adder; The imaginary part random sequence and the real part random sequence of the input multiplier are multiplied, and the operation result is output to the second probability real complex number adder; the second probability real number adder sums the two input quantities to obtain the imaginary output quantity a random sequence.

Images