Index


Figures


Tables

Kim , Kang , and Kim: Phase-Shifted DMRS-Aided Automatic Modulation Classification for PDSCH in 5G New Radio

Jeongseok Kim♦ , Byeong-Gwon Kang* and Taehyoung Kim°

Phase-Shifted DMRS-Aided Automatic Modulation Classification for PDSCH in 5G New Radio

Abstract: In this letter, a novel automatic modulation classification (AMC) scheme is developed for physical downlink shared channel (PDSCH) in 5G New Radio (NR). We design a convolutional neural network (CNN) to classify modulation types of the received PDSCH. To improve the classification accuracy, an enhanced demodulation reference signal (DMRS) structure is proposed where the phase of the DMRS is shifted depending on the modulation types. Simulation results verify that the proposed AMC scheme achieves 31.5% gain compared to the legacy scheme in terms of classification accuracy.

Keywords: AMC , PDSCH , 5G NR , CNN , DMRS

1. Introduction

With the large-scale commercial rollout of 5G and a new breakthroughs for 6G, mobile communication is developing toward intelligent communication[1]. Automatic modulation classification (AMC) for intelligent communication has got significant attention due to its wide range of applications in wireless communication[2,3]. An intelligent receiver can preprocess the received signal to identify the modulation type of the transmitted signal without prior information. Consequently, the control signal overhead can be reduced. Minimizing control signal overhead is particularly important for low complexity and battery- limited devices since it contributes to improve the reliability of physical downlink control channel (PDCCH) reception and reduces the power consumption associated with PDCCH blind decoding of the terminal[4]. Recently, deep learning-based AMC techniques have been widely studied[2,3]. Compared to the traditional maximum likelihood method, deep learning approach shows robust classification performance under fading channels and low signal- to-noise ratio (SNR) environments[2].

In 5G NR systems, the data channel, i.e., physical downlink shared channel (PDSCH), is transmitted with its demodulation reference signal (DMRS). Based on the current specification, the data symbols of a PDSCH can be modulated with one of modulation types among quadrature phase shift keying (QPSK), 16-quadrature amplitude modulation (QAM), 64-QAM, and 256-QAM. On the other hand, DMRS symbols are modulated with QPSK[5]. When AMC is applied to the NR PDSCH including DMRS, the classification accuracy is degraded due to the presence of the fixed QPSK symbols of DMRS.

In this paper, we develop a novel AMC method based on phase-shifted DMRS for NR PDSCH. To this end, we design a low complexity convolutional neural network (CNN) structure to classify one of four modulation types. In addition, we propose an enhanced DMRS structure where the phase of DMRS is associated with the modulation type of data. The modulation order-specific DMRS derives positive impacts on AMC performance, leading to improved classification accuracy.

Ⅱ. System Model

We consider NR PDSCH structure based on 5G NR specification[5]. A PDSCH can be transmitted in a slot which consists of 14 orthogonal frequency division multiplexing (OFDM) symbols. In frequency do-main, a PDSCH is allocated to multiple physical resource blocks (PRBs) consisting of 12 subcarriers. The data bits are modulated to complex-valued symbols based on one of modulation types among QPSK, 16-QAM, 64-QAM, and 256-QAM.

The DMRS for PDSCH is modulated by QPSK as

(1)
[TeX:] $$r(n)=\frac{1}{\sqrt{2}}(1-2 c(2 n))+j \frac{1}{\sqrt{2}}(1-2 c(2 n+1)),$$

where the pseudo-random sequence c(i) is defined in clauses 5.2.1 and 7.4.1.1 in TS 38.211[5]. The DMRS is mapped to resource element (RE) (k, l) according to the following equation:

(2)
[TeX:] $$\alpha_{k, l}=\beta w_f\left(k^{\prime}\right) w_t\left(l^{\prime}\right) r\left(2 n+k^{\prime}\right),$$

where k is subcarrier index, l is OFDM symbol index, β is a scaling factor, [TeX:] $$k^{\prime}=0,1, l=1+1, k=4 n+2 k^{\prime}+\Delta$$ for configuration type 1, and [TeX:] $$k=6 n+k^{\prime}+\Delta$$ for configuration type 2[5].

The received signal on (k, l), [TeX:] $$x_{k, l},$$ is given by

(3)
[TeX:] $$x_{k, l}=\rho \cdot c_{k, l} \cdot s_{k, l}^{(q)}+n_{k, l},$$

where ρ is transmit SNR, [TeX:] $$c_{k, l}$$ is channel impulse response, [TeX:] $$s_{k, l}^{(q)}$$ is the transmitted symbol modulated by q-th modulation type where q = 0, 1, 2, 3, and [TeX:] $$n_{k, l}$$ is additive white Gaussian noise (AWGN) with zero mean and unit variance.

Ⅲ. Proposed AMC scheme

The classifier is designed to provide the probability of a given type of predictive modulation as follows:

(4)
[TeX:] $$\operatorname{Pr}\left(s_{k, l} \in M(q) \mid x_{k, l} \text { for } k, l \in \Omega_{p d s c h}\right),$$

where M(q) represents the q-th modulation type and [TeX:] $$\Omega_{p d s c h}$$ is a set of REs where the PDSCH is received.

The CNN architecture for AMC of PDSCH is shown in Fig. 1. The designed network uses real-valued raw data as the input. The received signal consists of in-phase and the quadrature components, and the length of the signal is set to [TeX:] $$N_{\text {sym }} N_{\text {sc }}, \text { where } N_{\text {sym }}$$ is the number of symbols and [TeX:] $$N_{\mathrm{sc}}$$ is the number of subcarriers. In Fig. 1, "Conv" means the convolution layer, "BN" represents a batch normalization layer that normalizes the input to a standard normal distribution, “ReLU" means rectification linear activation. We set 2 × 1 and 16 × 1 pooling kernel sizes for the maximum pooling layer and the average pooling layer, respectively. “FC” is a fully connected layer and "SoftMax" is a softmax layer which is an activation function of a fully connected layer. Finally, the output is a classified modulation type. The proposed CNN is designed to have much lower complexity compared to the existing networks such as TRNN[2] and ResNet[3]. The number of training parameters of CNN, TRNN, and ResNet are 96.7k, 223.8k, and 121.8k.

Fig. 2 illustrates the proposed phase-shifted DMRS structure. As shown in Fig. 2, the phase for QPSK constellation of DMRS is determined by the modulation type of data. Specifically, the proposed DMRS is modulated based on the following equation:

(5)
[TeX:] $$r^{\prime}(n)=e^{j \theta(q)}\left(\frac{1}{\sqrt{2}}(1-2 c(2 n))+j \frac{1}{\sqrt{2}}(1-2 c(2 n+1))\right),$$

Fig. 1.

CNN architecture of AMC for NR PDSCH.
1.png

Fig. 2.

Proposed phase-shifted DMRS structure.
2.png

where [TeX:] $$\theta(q)=(\mathrm{q}-1) \cdot \pi / 16$$ is the phase shift value for q-th modulation type. Although DMRS still uses QPSK modulation, the additional phase information associated with modulation type can contribute to extract feature for the specific modulation type.

Ⅳ. Simulation Results

In this section, we provide simulation results to evaluate the classification performance for the proposed AMC scheme. Table 1 describes detailed simulation parameters. For training and test, we generate PDSCHs containing data symbols which is randomly modulated by one of QPSK, 16-QAM, 64-QAM, and 256-QAM. DMRSs are generated based on the fixed parameters given in Table 1 and inserted to the REs corresponding to eq. (2). The generated physical frame is transmitted via fading channel with AWGN. Then, it is input to the deep learning network in frequency domain after FFT.

Table 1.

Simulation Parameters
Parameter Value
Channel model TDL-A
Carrier frequency 6 GHz
Delay spread [TeX:] $$10^{-8.3}$$ sec
Velocity 30 km/h
Subcarrier spacing 30 kHz
Number of PDSCH symbols 12 symbols
Number of RB 10 RBs
DMRS configuration type 2
DMRS length 2
DMRS additional position 1
DMRS scrambling ID 10
Number of training data 480,000
Number of validation data 120,000
Number of test data 200,000
Epoch 25
Mini batch size 1024

Fig. 3 illustrates the classification accuracy depending on the transmit SNR. The proposed AMC with enhanced DMRS achieves 96 % accuracy, while that with legacy DMRS shows 73% accuracy at 20 dB SNR, i.e., 31.5 % performance gain is achieved. In addition, the proposed CNN shows similar classification accuracy compared to the legacy TRNN[2] and ResNet[3] although the CNN has much lower complexity. Therefore, the proposed

AMC scheme is effective for realizing the real-time AMC technique for NR PDSCH.

Fig. 3.

Classification accuracy for the proposed AMC.
3.png

Ⅴ. Conclusion

In this letter, a novel CNN-based AMC method is proposed for NR PDSCH. We develop a low complexity CNN to classify modulation types of the received PDSCH. In addition, an enhanced DMRS structure which applies different phase shift value according to the modulation order is proposed. Based on the simulation, it is verified that the proposed AMC scheme achieves improved classification accuracy compared to legacy scheme.

References

  • 1 C.-X. Wang, et al., "On the road to 6G: visions, requirements, key technologies, and testbeds," IEEE Commun. Surv. Tut., vol. 25, no. 2, pp. 905-974, Secondquarter 2023.doi:[[[10.1109/COMST.2023.3249835]]]
  • 2 L. Zhang, et al., "Real-time OFDM signal modulation classification based on deep learning and software-defined radio," IEEE Commun. Lett., vol. 25, no. 9, pp. 2988-2992, Fig. 3. Classification accuracy for the proposed AMC. Parameter Value Channel model TDL-A Carrier frequency 6 GHz Delay spread 10 -8.3 sec Velocity 30 km/h Subcarrier spacing 30 kHz Number of PDSCH symbols 12 symbols Number of RB 10 RBs DMRS configuration type 2 DMRS length 2 DMRS additional position 1 DMRS scrambling ID 10 Number of training data 480,000 Number of validation data 120,000 Number of test data 200,000 Epoch 25 Mini batch size 1024 Table 1. Simulation Parameters 1561 Sep. 2021.doi:[[[10.1109/LCOMM.2021.3093451]]]
  • 3 T. J. O’Shea, T. Roy, and T. C. Clancy, "Over-the-air deep learning based radio signal classification," IEEE J. Sel. Top. Sig. Proc., vol. 12, no. 1, pp. 168-179, Feb. 2018. (https://doi.org/10.1109/JSTSP.2018.2797022)doi:[[[10.1109/JSTSP.2018.2797022]]]
  • 4 T. Kim, Y. Kim, M. Jung, and H. Son, "Intelligent partial sensing based autonomous resource allocation for NR V2X," IEEE Internet of Things J., vol. 11, no. 2, pp. 3144-3160, Jan. 2024. (https://doi.org/10.1109/JIOT.2023.3295024)doi:[[[10.1109/JIOT.2023.3295024]]]
  • 5 3GPP TS 38.211, "NR; Physical channels and modulation," V17.2.0, Release 17, Jun. 2022.custom:[[[-]]]

Statistics


Related Articles

Deep Learning-Based Approaches for Nucleus Segmentation
D. C. Bui and M. Yoo
StarGAN v2 기반 패션 아이템 다중 도메인 변환을 통한 이미지 생성 모델 구현
H. Jang, B. Son, J. Lee
Shallow CNN을 활용한 주가 예측 방법론
Y. Cho, E. Kim, H. Shin, Y. Choi
5G NR-Unlicensed의 물리 계층 표준 규격과 기술 분석
S. Jung, I. Cho, S. Choi
인공자기도체를 이용한 소형 대수 주기 다이폴 배열 안테나
O. Kwon and K. C. Hwang
5G NR PRACH 프리앰블을 이용한 낮은 복잡도의 왕복 시간 추정기법
S. Choi, J. Kim, S. Back, W. Oh
5G NR 시스템을 위한 자기상관 기반 주파수 오프셋 추정 기법
M. So and W. Oh
낮은 복잡도의 5G NR PSS 검출 기법
M. So and W. Oh
도심 환경에서 5G NR V2X의 LDM 기반 분산혼잡제어 성능 분석
S. Ji, G. Jung, C. Mun
5G NR PBCH 데이터의 부반송파 매핑 방법 수정을 통한 복호 성능 개선 기법
J. Kim, S. Back, S. Choi, W. Oh

Cite this article

IEEE Style
J. Kim, B. Kang, T. Kim, "Phase-Shifted DMRS-Aided Automatic Modulation Classification for PDSCH in 5G New Radio," The Journal of Korean Institute of Communications and Information Sciences, vol. 49, no. 11, pp. 1558-1561, 2024. DOI: 10.7840/kics.2024.49.11.1558.


ACM Style
Jeongseok Kim, Byeong-Gwon Kang, and Taehyoung Kim. 2024. Phase-Shifted DMRS-Aided Automatic Modulation Classification for PDSCH in 5G New Radio. The Journal of Korean Institute of Communications and Information Sciences, 49, 11, (2024), 1558-1561. DOI: 10.7840/kics.2024.49.11.1558.


KICS Style
Jeongseok Kim, Byeong-Gwon Kang, Taehyoung Kim, "Phase-Shifted DMRS-Aided Automatic Modulation Classification for PDSCH in 5G New Radio," The Journal of Korean Institute of Communications and Information Sciences, vol. 49, no. 11, pp. 1558-1561, 11. 2024. (https://doi.org/10.7840/kics.2024.49.11.1558)