Parallel QRD-MLD for MIMO detection

Abstract

A new detection method based on parallel calculation of multiple QR decomposition on the permutations of channel matrix is invented for multiple input and multiple output (MIMO) system. This method switches the ith column and the last column of channel matrix to construct multiple permutation of channel matrix. Also it switches the ith transmitting data and the last transmitting data. The QR decomposition is performed on these permutations of channel matrix and data. The Maximum likelihood (ML) rate metric of every transmitting data is derived from these QR decompositions. This method achieves better performance than conventional ML detection method. Its complexity is much lower than the other MIMO detection methods. Its detection latency is much lower than other methods.

Description

[0001] Title of in invention: Parallel QRD-MLD for MIMO Detection

[0002] Technical field

[0003] This technology is developed to reduce the complexity and latency of detection method in multiple input and multiple output ( MIMO ) in wireless systems, such as wireless local area network (WLAN), 5G and other wireless communication system .

[0004] Summary of the invention

[0005] In the following, the technical problem of current MIMO detection technology is described. Our solution to the problem is proposed. The advantageous of our invention is analyzed.

[0006] Technical Problem

[0007] MIMO system is developed to increase spatial diversity or throughput for many wireless communication systems, such as 5G, 6G, WLAN. However, the MIMO detect need the complicate detection method to recover signal. Also it results in long detection latency, especially for massive MIMO system. Over the past decades, many detection technologies were developed. Maximum likelihood rate (ML) detection has the best performance. However, its complexity is exponential proportional to QAM constellation number and antenna number, so it is not feasible in application. For example, a Wifi 7 system transmits signal using 4096 QAM modulation over a 16X16 antenna MIMO system. The total calculation number is 409616and then comparison is also huge number. Other methods are classified into two groups. The first group is Linear detection methods, such as zero forcing (ZF), Minimum Mean Square Error (MMSE). The nonlinear method includes an order Successive Interference Cancellation (OSIC), QR decomposition Maximum Likelihood rate detection (QRD-MLD), sphere decoding method.

[0008] The ZF and MMSE are linear detection method. They have a low computation complexity. However, the Performance is not good. The OSIC method recovers the data by eliminating the detected signal from the received signal in a predefined order. The OSIC requires higher computation complexity than ZF and MMSE. But it provides sub-optimal performance of ML. It suffers error propagation, and this dramatically affects its performance. A sphere decoding method achieves performance close to ML method. However, it is difficult to choose the initial value for the radius of the sphere. If the radius is too small, it may not arrive to an optimal solution. If the radius is too large, it results in a large computation complexity.

[0009] The performance of a QRD-MLD method depends on the candidate sets. A large candidate set can achieve a performance close to ML. But it raises the computation complexity. Using a small candidate set, the performance degrades dramatically. Our invention focuses QRD-MLD field, so we will describe conventional QRD MLD method as below.

[0010] A MIMO system with M transmitting antennas and N receiving antennas is described as below. The channel matrix is H, and x denotes the transmitting data. The ML metric is as below.

[0011] For a 4X4 system, after QR decomposition, the ML metric is

[0012] The decoding procedure begins from finding an optimal solution for . The metric is to find a value of which minimizes . Due to the noise the candidate set of must be large enough to achieve a good performance. After decoding the set of , every value of the candidate set of is substituted into to decode . The decoding of and follows the rule abovementioned. To achieve better performance, the candidate set need to be larger. But it increases computation complexity. The decoding of an antenna depends on the result of previous decoded antennas. So this method suffers error propagation. This method also needs iterative calculation do decode the data of different transmitting antennas one after one. When the antenna number increases, the decoding latency increases dramatically.

[0013] As above mentioned, the current detection methods all have their drawbacks, either performance or computation complexity. Also the OSIC, sphere decoding, QRM-MLD result in long detecting latency when the antenna number increase. These methods also suffer error propagation. Therefore, we need to invent a new method which can achieve ML performance with low computation complexity.

[0014] Solution to Problem

[0015] A solution based on parallel computation multiple QR decomposition on the permutation of channel and transmitting data is invented.

[0016] To better explain how this method works, a MIMO system with four transmitting antennas and four receiving antennas is used as the example to describe this invention as follow.

[0017] The equation (1) is a 4X4 MIMO system. The received data at receive antenna 1, 2, 3, and 4 are denoted by . The transmitting data are . The Gaussian noise at the four receiving antennas are denoted by . The channel matrix is where denotes the channel response from the jth transmitting antenna to the / th receiving antenna.

[0018] Step one, the QR decomposition is performed on channel matrix H abovementioned as below.

[0019] Step two, the third and fourth column of channel matrix H are switched, and the transmitting data and are switched, a new matrix is as below

[0020] After QR decomposition, it generates

[0021] Step three, the second and fourth column of channel matrix H, are switched and the transmitting data and are switched as below

[0022] After QR decomposition, we have

[0023] Step four, the first column and fourth channel matrix H are switched, and the transmitting data and are switched. After the QR decomposition, it generates

[0024] From equation (2), (4), (6), and (8), the individual ML metric of are derived as follows, where C is the QAM constellation sets.

[0025] In actual application, the ZF or MMSE is used to decode from as below.

[0026] For ZF decoding, it gets

[0027] For MMSE decoding, it gets

[0028] For a soft decoding method, the log-likelihood-rate (LLR) can be calculated from and . Then The LLR is sent to error correction decoder, such as Viterbi decoder and LDPC decoder, to decode the data.

[0029] The abovementioned method can be extended to any antenna number of M X N following the rule abovementioned. It just needs to switch the / th column and the last column of channel matrix . And then the / th data and last data are switched.

[0030] Advantageous effect of the invention

[0031] Since these QR decompositions can be performed parallel, the decoding of data from different transmitting antenna can be isolated and do not depend each other. There is no error propagation. Since this method does not need iterative search, the latency is much lower than other methods. Its calculation complexity is much lower than other methods. From the simulation result shown in Figure 1, its performance even surpasses a conventional ML method. In Figure 1, a 4X4 MIMO system with 16 QAM modulation is used in simulation. This method shows about 3 dB SNR improvement at BER=1O4when it is compared to conventional ML method.

[0032] Brief description of drawing

Claims

ClaimClaim 1 A new ML method is invented. This method switches the / th column and last column of channel matrix. And then it switches the / th and last transmitting data to construct multiple Y = HX + N . An example of 4X4 MlMO system is as below.Claim 2 Multiple QR decomposition are performed on the equations in claim 1 parallel, it generates multiple QR decomposition belowClaim 3 Individual ML metric is achieved from the equation of Claim 2 as belowClaim 4 After ZF or MMSE decoding on the metric of claim 3, the decode data is achieved as below.For ZF, it getsFor MMSE, it getsClaim 5 the LLR can be calculated from decoded data from Claim 4. Then the LLR is sent to error correction decoder, such as Viterbi or LDPC decoder.Claim 6 the abovementioned method can be applied to any M X N antenna MIMO system.CitationYong Soo Cho, Jaekwon Kim, Won Young Yang, Chuang-Gu Kang, "MIMO-OFDM Wireless Communications with MATLAB ", pp. 319-344, Wiley, 2010.

Images