WO2018191762A1

WO2018191762A1 - Method for decoding multi-user signals in massive mimo system

Info

Publication number: WO2018191762A1
Application number: PCT/VN2018/000004
Authority: WO
Inventors: Hoang Anh NGO; Minh Hai TRAN
Original assignee: Military Industry - Telecommunication Group (viettel)
Current assignee: Military Industry - Telecommunication Group (viettel)
Priority date: 2017-04-13
Filing date: 2018-04-11
Publication date: 2018-10-18
Anticipated expiration: 2019-10-13

Abstract

The present disclosure relates to a method for decoding multi-user signals in the Massive MIMO system without the use of matrix inversion including (i) exploiting the use of a matched filter to improve the signal-to-noise ratio (SNR); (ii) initializing an estimated value for each user; and (iii) removing interference/noise and improving the signal of each user.

Description

METHOD FOR DECODING MULTI-USER SIGNALS IN MASSIVE MIMO SYSTEM

Technical field of the invention

The present disclosure relates to designing a base station of the fifth-generation (5G) wireless communication system using a plurality of antennas, which is typically referred to as Massive MIMO (multiple-input multiple-output) technique. Specifically, the present disclosure relates to a method for decoding multi-user signals in the Massive MIMO system without the use of matrix inversion.

Background of the invention

In order to enhance data transmission rate between users and a base station of the 5G wireless communication system, a plurality of antennas can be integrated into the base station and the Massive MIMO technique is used. For instance, a base station using the Massive MIMO with 64, 128, or 256 antennas can provide a data transmission rate 10 times higher than typically provided by a 4G - LTE (Fourth- Generation - Long Term Evolution) system.

One of the challenges in the use of the Massive MIMO is that the extensive use of matrix inversion operations is required to separate and to recover users' signals. For example, a system serving 16 concurrent users, having an 128-antenna base station and using OFDM (Orthogonal Frequency Division Modulation) with 1200 subcarriers for data transmission, must process approximately 2.4 million matrix inversion operations per second. For such a demanding task, implementations on ICs (Integrated Circuits) or FPGA (Field Programmable Gate Array) that directly compute the inverse matrices are complicated and hardware resource consuming, and have large processing delays.

The current methods to detect and recover multi-user signals are divided into three main categories:

- Direct approaches that compute the inverse matrices using QR decomposition algorithm [1, 2] or Cholesky decomposition algorithm [3].

- Inverse matrix approximation methods with polynomial expansion or Neumann series [4]. - Iterative techniques to solve linear equation systems, such as CG (Conjugate Gradient) [5] or CD (Coordinate Descent) [6].

However, the methods described above have certain following disadvantages:

- The direct matrix inversion techniques using QR decomposition algorithm [1, 2] or Cholesky decomposition algorithm [3] have high complexity as well as high processing latency.

- The polynomial expansion and first/second-order Neumann series methods are insufficient in term of accuracy; while third-order Neumann series [4] could provide high accuracy but with the complexity approximately equivalent to that of the direct matrix inversion methods.

- The iterative methods CG [5] and CD [6] require excessive number of multiplying operations for each iteration since the required number of multiplication is proportional to the number of base station antennas (64 or 128). Hence, this approach is highly resource demanding for FPGA implementations.

Therefore, there is a demand for a simple and computationally efficient method for decoding multi-user signals without the use of matrix inversion.

Summary of the invention

The purpose of the present disclosure is to detect and recover multi-user signals in the Massive MIMO system with low complexity and low processing delay, without the use of matrix inversion, and with easy FPGA implementation. In order to achieve this goal, the present disclosure provides a method for decoding multi-user signals in the Massive MIMO system without the use of matrix inversion, the method includes the following steps: i) exploiting the use of a matched filter to improve the signal-to-noise ratio (SNR); ii) initializing an estimated value for each user; iii) removing interference/noise and improving the signal of each user. characterized in that, in the step iii), a processing structure comprising K identical and sequential processing units is provided to effectively eliminate multi-user interference, and the improved signal of user k will be used to remove its interfering effect on the signal of user k+l .

Brief description of the drawings

Figure 1 is the schematic diagram illustrating a method for signal processing at a base station.

Figure 2 is the schematic diagram illustrating a method for decoding the multiuser signal in Massive MIMO system without using matrix inversion.

Figure 3 is the diagram illustrating the sequential structure of the interference cancellation.

Detailed description of the invention

On the base station including a plurality of antennas, the received signal on each of its antennas is an additive sum of the signals of all users which is already distorted by the wireless channel. The general principle of detecting and recovering the multiuser signals is described as follows:

- Combining received signals from the antennas of the base station in order to enhance the quality of the signal of the intended user. This step is performed by using a matched filter. The base station uses L receiving antennas to receive the signal transmitted from user k. Assuming that the channel transfer functions of the wireless transmission channel between the user to each of antennas are estimated at the base station, the SN of user k can be boosted by multiplying received signal at each of antennas with the conjugate of the transfer function of the corresponding channel and then summing the results. This technique will be described further later.

- Filtering out the interference on the desired signal of user k caused by the signals from other users by using an iterative approach. By using a sequential structure, the signal quality for each user is enhanced sequentially through each iteration. - A structure of K sequential and identical processing units has low complexity, hence simplifying hardware implementation.

- Through calculation of Gramma matrix, only one multiplying operation is required for each iteration that is significantly small compared to the CG method [5] and the CD method [6].

Figure 1 is the schematic diagram illustrating the structure of signal processing at a base station serving K users and using OFDM Massive MIMO. Data of each user, represented by a raw binary sequence, is first encoded by the channel coding module using, for instance, Turbo code and then modulated with QAM (quadrature amplitude modulation) as in modules 100-1, 100-2, 100-K. The outputs of the QAM modulator are IQ samples denoted as S_lt S₂, ... , S_K. After the OFDM modulation (modules 101-1, 101-2, 101-K), the signal of each user is then transmitted using an antenna as in module 102. The time-domain signals over antennas of users are denoted as s^, s SK in module 101. These OFDM modulated signals propagate in wireless channels and then are captured by L antennas of the base station.

On the uplink path, the base station uses L antennas to receive and detect signals of K users whereas on the downlink path, it uses these L antennas to simultaneously transmit signals to K users. In module 104 at the base station, the received signals in the time domain are denoted as yi, ₂,—, yi - Upon OFDM demodulation, the received signals at the predefined pilot subcarriers are used for the channel estimation modules 107-1, 107-2, 107-L. The outputs of these channel estimators, which are the channel transfer function vectors H_i, H₂,—, H_L, together with the OFDM demodulated signals at data-carrying subcarriers (denoted as vector Y), are fed into the Massive MIMO multi-user detection module 107. The Massive MIMO multi-user detection module is responsible for separating the corresponding signal for each user and recovering it from distortion and noise. The outputs of this Massive MIMO multi-user detector, denoted as complex samples X , X₂, - , κ^ ^wiH be converted back to the original binary data sequences of users by the QAM demodulation module. A significant challenge in designing the uplink Massive MIMO system is that the multi-user detection module must be able to compute an excessive number of matrix inversion operations within a short period of time. For instance, consider an OFDM Massive MIMO system serving 8 concurrent users with 128 antennas at the base station. Assuming this OFDM system employs 1200 subcarriers to transmit data with the FFT IFFT size of 2048 and a sampling rate of 30.72 Msps (Mega samples per second), its multi-user detection module would demand for a computing power of 2.4 Mmps (Million matrices per second). Implementations on FPGA that could perform 2.4 Mmps would be overly complicated and extremely resource consuming.

A method for decoding multi-user signals in Massive MIMO system using matrix inversion is described as follows:

The received signals after OFDM demodulation at the base station can be expressed as the following equation:

Y = H. S + N (1)

Y G C^{Lx l}, H G C^LxK/S G C^{Kx l}, N G C^{Lx l} (2)

H represents the channel transfer function matrix between K users and L antennas of the base station. N represents the additive Gaussian white noise matrix at the antennas of the base station. The original signals of K users can be recovered using the ZF (Zero-Forcing) method or the MMSE (Minimum Mean Square Error) method as follows:

S_ZF = (HHHrWY (3) S_MMSE = (H^H H + a²I)-¹H^HY (4)

Where σ denotes the white noise variance at the receiving antennas (σ² = E[N^HN]). Both equations (3) and (4) require computation of an inverse matrix, either (H^HH)^-1 or {H^HH + σ²/)^-1. This matrix inversion operation is required for each of 1200 OFDM subcarriers. Hence, the multi-user detection module must perform an excessive number of matrix inversion operations, as a result, it is not suitable for hardware implementation. Figure 2 illustrates the method for decoding multi-user signals in the Massive MIMO system without the use of matrix inversion. This method is described as follows:

Received signal at each antenna of the base station is a combination of the signals of all K users so that equation (1) can be rewritten as:

Y = SiHi + S₂H₂ + - + S_KH_K + N (5)

Hk = [Hik, H₂k, - , H_LK]^T (6) where H_k denotes the channel transfer function vector between user k and L antennas of the base station, and H_lk specifically denotes the channel transfer function between user k and antenna / of the base station.

• Step 1: Matched filter (the module 202).

The inputs of the module 202 are the OFDM demodulated signals from L branches of the base station (Y_lt Y₂,—, Υι) and the estimated channel transfer function H. The outputs produced by module 202 are denoted as b_r, b₂, b_K. A matched filter is used to improve the SNR for each user. Employing the matched filter, the SNR of each user is maximized as follows. b = H^HY (7)

For instance, with user k:

Y_k = | |H_k| |₂S_k + 2_J HJH₁S₁ .+ Hj N

i=l

i≠k

When the number of antennas of the base station keeps increasing and the channel transfer function vectors H_u and H_v of two arbitrary users u and v respectively are uncorrected, the interference components ∑f_{=i i≠k} H_k ^* HiSi will gradually reach zero (0). Therefore, the equation (7) can be used to improve the received SNR for each user.

• Step 2: Pre-processor (the module 203).

The pre-processor module 203 creates the initial estimated value for the signal of each user as follows: (9)

(10)

D and E are calculated from Gramma matrix as follows:

G = H^HH (1 1)

G = D + E (12)

Where D is the main diagonal of the matrix G.

• Step 3: Interference Canceller (the module 204).

After the signal of each user is coarsely estimated by the pre-processor module 203, the interference cancellation module 204 performs the interference cancellation as follows:

(13)

§W =

D

In order to reduce the complexity of the interference cancellation module 204 as well as to increase the effectiveness of the interference cancelling process, a sequential processing structure is designed as illustrated in Figure 3. The interference cancellation module contains K sequential processing units. After the signal of user 1 is improved, it serves as an input to enhance the quality of the signal of user 2. This process continues till user K. Each of these processing units accepts a vector t_Kxl and a coefficient b_k as inputs and then updates the signal of user and vector t as follows:

E_k is the k^th column of matrix E. All of these processing units have identical structures that greatly simplifies the uplink processing design. On the other hand, the pipelining technique is applicable for this method to further increase the throughput of such a sequential processing structure. Number of computations approximately needed in a method for decoding multiuser signals in Massive MIMO system that uses matrix inversion is shown in Table 1 below:

Table 1. The number of computations needed with matrix inversion approach

The number of computations approximately needed in a method for decoding multi-user signals in Massive MIMO system without the use of matrix inversion in accordance with the present disclosure is shown in Table 2 below:

Table 2. The number of computations needed when using the proposed method

(without matrix inversion)

Interference i(2 -bk - tk K K² 3K cancellation

t - _{t +} (i« - i«) ¾

Total 2K K²L ,

— + K² + KL — + KL + M

Tables 1 and 2 show that the method for decoding multi-user signals in Massive MIMO system without using matrix inversion only uses (— + K² + KL) multiplications, whereas the method for decoding multi-user signals in Massive MIMO system with matrix inversion requires (K³ +— + KL) multiplications.

Examples of implementing the invention

In order to confirm the effectiveness of the proposed method, an FPGA implementation using Xilinx System Generator was carried out. System parameters and results are shown in Table 3.

Table 3. Parameters of the testing system and output

DSP48 usage 2368

Note: For a technique that directly computes inverse matrices, DSP48 usage is 4352.

Effects achieved by the invention

The method for decoding multi-user signals in the Massive MIMO system without the use of matrix inversion achieves low complexity and requires less multiplications (with a lower complexity order) than the direct inversion approach. The proposed method also achieves higher accuracy as compared to those approaches using polynomial expansion, Neumann series, CG, or CD thanks to the sequential interference cancelling structure. On the other hand, this method enables pipelining implementations to support high-throughput processing. Lastly, the sequential structure of K identical processing units provides high efficiency and ease of hardware implementation.

Claims

1. A method for decoding multi-user signals in the Massive MIMO system without the use of matrix inversion, the method includes the following steps: i) exploiting the use of a matched filter to improve the signal-to-noise ratio (SNR); ii) initializing an estimated value for each user; and iii) removing interference/noise and improving each user's signal; characterized in that, in the step iii), a processing structure of K identical and sequential processing units is provided to effectively eliminate multi-user interference, the improved signal of user k will be used to remove its interfering effect on the signal of user k+\ .