深度学习辅助的超奈奎斯特速率光空间脉冲位置调制
As an innovative multiple-input-multiple-output (MIMO) technology, optical spatial modulation (OSM) resolves antenna interference and synchronization challenges in MIMO systems by selecting a single antenna to carry information and collectively transmits the antenna index as additional information. However, existing OSM research predominantly adheres to the orthogonal transmission criterion, and imposes limitations on enhancing the transmission rate of the system although the research is effective in avoiding inter-symbol interference. To this end, the introduction of non-orthogonal transmission via Faster-Than-Nyquist (FTN) technology compresses symbol intervals during pulse shaping, enabling an increase in transmission rate within the same bandwidth per unit time. As a result, we propose a novel Faster-Than-Nyquist rate optical spatial pulse position modulation scheme that combines OSM with FTN to further enhance the transmission rate and spectrum efficiency of the system. Additionally, in response to the highly complex receiver issue, a multiclassification neural network (MNN) decoder is proposed to significantly reduce computational complexity and achieve approximate optimal detection.
At the transmitting end, the input binary bit stream is divided into two groups of data blocks after serial/parallel transformations. The first group of data blocks is mapped to the index of the selected lasers for each symbol period, while the second group is mapped to pulse position modulation (PPM) symbols. An FTN shaping filter is employed to compress the PPM symbols. Then, the compressed PPM-FTN signals are loaded onto the chosen lasers for transmission. The signal traverses the Gamma-Gamma channel, and it is received by photodetectors (PDs) and converted into an electrical signal for further signal processing at the receiving end. Initially, downsampling is conducted to obtain a signal with the same dimensionality as the input signal. The downsampled signal is then classified based on its effective features, with each class being assigned the corresponding label. Subsequently, different samples with varying signal-to-noise ratios (SNRs), along with their associated label values, are utilized as input and output for offline training of a neural network model. The objective is to achieve optimal decoding accuracy by defining average loss and learning rate parameters to construct an MNN, which helps determine the number of hidden layers and neurons. Finally, the well-constructed MNN is employed for online signal detection. Then, inverse mapping is conducted on output label values from the decoder to recover the corresponding modulation symbols and laser index.
Monte Carlo simulations are conducted to evaluate the proposed scheme in a Gamma-Gamma channel. We first derive an upper bound of the average bit error rate (ABER) of the system and provide a comparison of the simulated BER with the ABER in Fig. 3. The results show that the two curves asymptotically coincide at high SNRs, which demonstrates the correctness of the derived ABER. Then, an analysis is performed on the influence of various parameters such as the number of lasers, the number of detectors, and modulation order on the error performance of the OSPPM-FTN system. The findings reveal that an increase in these parameters can enhance both the transmission rate and BER performance of the system, despite at varying costs. Furthermore, in Fig. 5, we compare the transmission rate, spectrum efficiency, and BER performance of the proposed system with traditional OSPPM. The results indicate that under the acceleration factor of 0.9, compared to the OSPPM system, the proposed system shows a 17% increase in spectrum efficiency and a 5.5% increase in transmission rate with only 1 dB SNR lossy. As the acceleration factor decreases from 0.9 to 0.7, the spectrum efficiency and transmission rate of the OSPPM-FTN system rise by 73% and 21.5% respectively. Thus, the proposed scheme demonstrates a significant improvement in both transmission rate and spectrum efficiency with the reduction of the acceleration factor. Through the comparison with the maximum likelihood (ML) algorithm, Figs. 7 and 8 illustrate the computational complexity reduction and BER performance of the proposed MNN decoder. The results show that the MNN decoder achieves near-optimal decoding performance, and as the detectors increases, the computational complexity of the MNN decoder is significantly lower than that of ML. For instance, when there are 8 or 16 PDs, our decoder can reduce computational complexity by 69.75% and 89.95% respectively.
A Faster-Than-Nyquis rate optical spatial pulse position modulation scheme is proposed by combining optical spatial pulse position modulation with the FTN technique, which effectively improves the transmission rate and spectrum efficiency of the system. Compared to traditional optical spatial modulation, simulation results show that the proposed scheme achieves a significant improvement in transmission rate and spectrum efficiency with the decreasing acceleration factor. Simultaneously, increasing the modulation order, the number of lasers, and the number of detectors can improve the transmission rate and error performance of the system. However, the cost associated with each parameter varies, and the selection of these parameters should be contingent on specific circumstances. Additionally, the MNN decoder proposed for the OSPPM-FTN scheme achieves near-optimal decoding performance while substantially reducing computational complexity. It is noteworthy that this advantage is particularly pronounced in large-scale MIMO systems.
1 引言
光空间调制(OSM)[1]是由Mesleh等提出的一种新型多输入多输出(MIMO)[2-3]技术,通过每时隙选择单根天线加载信息并将天线索引序号作为隐含信息共同传输,既解决了MIMO中存在的天线干扰和同步难的问题,又提高了系统的传输速率和频谱效率,该技术一经提出便得到了快速发展。文献[4]将OSM与脉冲位置-幅度调制相结合,提出了高传输速率和频谱效率的光空间脉冲位置-幅度联合调制。文献[5]建立了Gamma-Gamma信道下的OSM系统模型,推导了平均符号错误概率的闭合表达式。文献[6]提出了基于正交频分复用的二维广义正交OSM方案,在不损失可靠性的同时提高了系统的频谱效率和功率效率。文献[7]利用分层空间结构提出了高传输速率的标记型多层光空间脉冲位置调制(OSPPM)。上述研究都遵循正交化传输准则,通过在光脉冲间保持正交性来有效避免码间串扰,但其正交性限制了系统可传输的最高信息速率。为突破上述限制,亟须将一种非正交技术与OSM结合以进一步提升系统的传输速率和频谱效率。
超奈奎斯特(FTN)技术作为非正交传输技术的代表,其最高传输速率突破了奈奎斯特速率的限制,通过在脉冲成型时压缩符号间隔,实现相同带宽下单位时间内传输速率的提升[8-9]。文献[10]提出了高于奈奎斯特速率的FTN-脉冲幅度调制方案。文献[11]将FTN技术与MIMO相结合,得到了频率选择衰落信道中MIMO-FTN系统的最高可达速率。上述研究证明了MIMO-FTN系统在提升传输速率和频谱效率方面的优势,而OSM作为一种新型的MIMO技术,将其与FTN结合对于提升传输速率和频谱效率有着重要的研究意义。为此,本文将FTN技术引入OSM中,结合具有抗干扰能力强且功率效率高的脉冲位置调制(PPM),提出一种超奈奎斯特速率的光空间脉冲位置调制(OSPPM-FTN)方案,以进一步提升系统的传输速率和频谱效率。
另一方面,由于空间调制信号检测的复杂性,在接收端探索一种性能近似最优且复杂度较低的信号检测算法也尤为重要。目前,深度学习理论已被广泛应用于图像处理、无线通信[12-15]等领域,其出色的学习能力和分类优势为信号检测提供了新思路。文献[16]针对广义空间调制提出了一种基于块-深度神经网络架构的检测器,可克服有源天线间信道干扰对信号检测的影响,实现近似最优的检测性能。文献[17]提出了一种基于正交频分复用-广义OSM的深度神经网络辅助检测算法,有效解决了最大似然和最大比合并检测时误差传播和噪声放大等问题。文献[18]针对大规模MIMO系统提出了基于神经网络的近似消息传递检测算法,其收敛速度更快且计算复杂度更低。上述研究证明将深度学习方法应用于信号检测领域具有可行性且检测性能良好。鉴于此,根据所提OSPPM-FTN方案接收信号的特征,提出了一种多分类神经网络(MNN)译码器,其在实现近似最优检测的同时大幅降低了计算复杂度。
2 OSPPM-FTN系统模型
对于一个有
表 1. OSPPM-FTN系统映射表
Table 1. Mapping table of OSPPM-FTN system
|
具体地,OSPPM-FTN系统的发端映射方案包括两部分:空间域LD序号索引模块和信号域映射模块。其中,信号域映射又包含PPM符号映射和FTN成型,详细过程如下。
对于LD序号索引模块,在每个符号周期内仅选择一个LD发送信息,则被选LD索引号的映射关系可表示为
式中:
在信号域映射模块中,首先将
式中:
其次,将映射后的PPM符号进行FTN成型。先将PPM符号流转换为行向量,即
式中:
将成型后的
式中:
为了更清楚地表征映射过程,
发送信号
式中:
式中:
特别地,在接收端进行信号检测时,先对
式中:
3 误码性能分析
根据OSPPM-FTN系统发端映射方案的特点,可能发生的错误类型可归纳为3种情况:1)LD索引号估计正确,调制符号译码错误,此时造成的平均误码率(ABER)记为
值得注意的是,所提系统中FTN成型时引入的码间串扰
式中:
式中:
对于第一类错误,
式中:
式中:
式中:
对于第二类错误,
式中:
式中:
对于第三类错误,
同理,用MGF方法求得
式中:
将式(
式(18)体现了系统ABER与SNR的关系,即
由式(
4 低复杂度信号译码器
虽然ML算法具备最优译码性能,但其穷搜索过程会导致计算复杂度偏高,在实际通信中难以实现,尤其是在大规模或高阶调制的OSPPM-FTN系统中计算复杂度急剧增加。近几年,有学者用深度学习方法来解决MIMO系统中的信号检测问题,将传统译码算法中求解信号间最小欧氏距离的问题转化为基于信号有效特征的分类问题,这样可大幅降低计算复杂度且译码性能逼近最优[23-24]。因此,根据OSPPM-FTN系统接收信号特征和深度学习的分类特性,提出一种MNN译码器。
对于OSPPM-FTN系统,接收端译码的目的是从接收信号中恢复出PPM符号和LD索引号。当调制阶数和LD数目固定时,接收信号可分为多类包含不同有效特征的信号,将其转化为多分类问题来处理。具体地,首先将接收信号分为
MNN译码器包含1个输入层、3个隐藏层、1个输出层和1个分类层。
式中:
在MNN译码器中,以每批次训练的平均损失
式中:l为网络中隐藏层和输出层的参数;
获得训练批次的平均损失后,为了使预测值更加准确,需要通过交叉熵损失函数不断调整网络参数以确定最优参数值。该过程通过随机梯度下降(SGD)优化实现,即以当前梯度值的学习率为基础减去或增加旧的权值。SGD通过梯度值随时间的迭代进行更新,迭代过程如下:
式中:
5 仿真结果与分析
为了验证上述理论的正确性,采用蒙特卡罗方法对所提方案在Gamma-Gamma信道下进行仿真。假设接收端已知信道状态信息,仿真中若无特殊说明,各参数的取值分别为
5.1 ML时OSPPM-FTN性能分析
图 3. OSPPM-FTN系统BER的理论上界和仿真性能对比
Fig. 3. Comparison of theoretical upper bound and simulation performance of BER in OSPPM-FTN system
图 4. 不同参数时OSPPM-FTN系统的误码性能
Fig. 4. BER performance of OSPPM-FTN system with different parameters
表 3. 不同方案的频谱效率和传输速率
Table 3. Spectral efficiency and transmission rate of different schemes
|
图 5. OSPPM-FTN和OSPPM系统的性能对比。(a)OSPPM和OSPPM-FTN系统在不同 值下的误码性能;(b)OSPPM和OSPPM-FTN系统在不同 值下的传输速率和频谱效率
Fig. 5. Performance comparison of OSPPM and OSPPM-FTN systems. (a) BER performance of OSPPM and OSPPM-FTN systems with different ; (b) transmission rate and spectral efficiency of OSPPM and OSPPM-FTN systems with different
5.2 MNN译码器性能分析
MNN译码器的参数如
表 4. MNN译码器参数
Table 4. Parameters of MNN decoder
|
首先,讨论MNN译码器的隐藏层数量的选取问题,该参数决定了整个网络对输入信号特征提取的能力。隐藏层数量过多会导致计算复杂度显著增加和过拟合现象,而隐藏层数量较少会造成欠拟合。因此,
表 5. 隐藏层数量与译码精度之间的关系
Table 5. Relationship between hidden layer number and decoding accuracy
|
其次,分析模型离线训练时学习率的取值以及平均损失与训练轮次的关系。
表 6. 学习率与译码精度之间的关系
Table 6. Relationship between learning rate and decoding accuracy
|
图 6. MNN译码器的平均损失与训练轮次的关系
Fig. 6. Relationship between loss and training rounds of MNN decoder
最后,对比了MNN译码器和ML的计算复杂度,如
表 7. 不同译码器的计算复杂度
Table 7. Computational complexity of different decoders
|
图 8. 不同湍流下ML与MNN的BER
Fig. 8. BER performance of ML and MNN decoders under different turbulence
表 2. 湍流模型参数
Table 2. Turbulence model parameters
|
6 结论
将OSPPM与FTN技术结合,提出了一种OSPPM-FTN方案,该方案有效提升了系统的传输速率和频谱效率。研究结果表明,相比于传统OSM,所提方案的传输速率和频谱效率随着加速因子的减小而显著提升。同时,提高调制阶数、LD和探测器数目均可改善系统的传输速率和误码性能,但其代价也不尽相同,如何选取各参数需视具体情况而定。另外,针对发端方案提出的MNN译码器在取得近最优译码性能的同时大幅降低了计算复杂度,尤其在大规模MIMO系统中,MNN译码器的优势更加明显。
[1] Mesleh R, Elgala H, Haas H. Optical spatial modulation[J]. Journal of Optical Communications and Networking, 2011, 3(3): 234-244.
[2] Wang Y P, Chen X M, Pei H L, et al. MIMO performance enhancement of MIMO arrays using PCS-based near-field optimization technique[J]. Science China Information Sciences, 2023, 66(6): 162302.
[3] 赵黎, 王昊, 张峰. 基于光广义空间调制的VLC-MIMO系统研究[J]. 中国激光, 2022, 49(23): 2306001.
[4] Özbilgin T, Koca M. Optical spatial modulation over atmospheric turbulence channels[J]. Journal of Lightwave Technology, 2015, 33(11): 2313-2323.
[5] Geng C, Yu S Y, Lu G Y, et al. On the correlated Gamma-Gamma distribution and coherent detection performance of optical spatial modulation[J]. Optik, 2023, 276: 170622.
[6] Qiu Y, Xu X H, Shen M, et al. OFDM-based 2D generalized optical quadrature spatial modulation for MIMO-OWC systems[J]. Optical Engineering, 2022, 61(12): 126103.
[7] 王惠琴, 叶归清, 彭清斌, 等. 标记型多层光空间脉冲位置调制[J]. 光学学报, 2022, 42(14): 1406003.
[8] Mazo J E. Faster-than-Nyquist signaling[J]. Bell System Technical Journal, 1975, 54(8): 1451-1462.
[9] Mohammadkarimi M, Schober R, Wong V W S. Channel coding rate for finite blocklength faster-than-Nyquist signaling[J]. IEEE Communications Letters, 2021, 25(1): 64-68.
[10] Che D, Chen X. Higher-order modulation vs faster-than-Nyquist PAM-4 for datacenter IM-DD optics: an AIR comparison under practical bandwidth limits[J]. Journal of Lightwave Technology, 2022, 40(10): 3347-3357.
[11] Wang J T. Time-space MIMO system with faster-than-Nyquist signaling in frequency-selective fading channel[J]. Multidimensional Systems and Signal Processing, 2021, 32(3): 1027-1039.
[12] Baek M S, Kwak S, Jung J Y, et al. Implementation methodologies of deep learning-based signal detection for conventional MIMO transmitters[J]. IEEE Transactions on Broadcasting, 2019, 65(3): 636-642.
[13] 张桐浩, 王旭东, 吴楠. 基于CNN的QAM-PPM混合调制端到端通信系统[J]. 激光与光电子学进展, 2022, 59(17): 1706001.
[15] 石佳, 黄爱萍, 陶林伟. 深度学习辅助水下光通信信道估计和信号检测[J]. 中国激光, 2022, 49(17): 1706004.
[16] Albinsaid H, Singh K, Biswas S, et al. Block deep neural network-based signal detector for generalized spatial modulation[J]. IEEE Communications Letters, 2020, 24(12): 2775-2779.
[17] Chen C, Zeng L, Zhong X, et al. Deep learning-aided OFDM-based generalized optical quadrature spatial modulation[J]. IEEE Photonics Journal, 2022, 14(1): 7302306.
[18] Yang Y, Chen S P, Gao X Q. Deep learning-based AMP for massive MIMO detection[J]. China Communications, 2022, 19(10): 69-77.
[19] Chen D, Hui J X. Parameter estimation of Gamma-Gamma fading channel in free space optical communication[J]. Optics Communications, 2021, 488: 126830.
[20] Yu S Y, Geng C, Zhong J, et al. Performance analysis of optical spatial modulation over a correlated gamma-gamma turbulence channel[J]. Applied Optics, 2022, 61(8): 2025-2035.
[21] Gupta J, Goel A. Piece-wise approximation for Gaussian Q-function and its applications[J]. Sādhanā, 2022, 47(3): 169.
[23] Shankar R, Chaudhary B P, Mehraj H, et al. Impact of node mobility on the DL based uplink and downlink MIMO-NOMA network[J]. International Journal of Information Technology, 2023, 15(6): 3391-3404.
[24] Ro J H, Ha J G, Lee W S, et al. Improved MIMO signal detection based on DNN in MIMO-OFDM system[J]. Computers, Materials & Continua, 2022, 70(2): 3625-3636.
[25] Wang T J, Yang F, Song J. Deep learning-based detection scheme for visible light communication with generalized spatial modulation[J]. Optics Express, 2020, 28(20): 28906-28915.
Article Outline
张悦, 叶翔文, 曹明华, 王惠琴. 深度学习辅助的超奈奎斯特速率光空间脉冲位置调制[J]. 光学学报, 2024, 44(5): 0506003. Yue Zhang, Xiangwen Ye, Minghua Cao, Huiqin Wang. Deep Learning-Aided Faster-Than-Nyquist Rate Optical Spatial Pulse Position Modulation[J]. Acta Optica Sinica, 2024, 44(5): 0506003.