Zhipei Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zhipei Li has a total of 163 co-authors. Here are the top co-authors by number of shared publications:
Xiangjun Xin
Beijing University of Posts and Telecommunications (BUPT)
28 shared publications- Ze Dong
Beijing Institute of Technology
24 shared publications Ran Gao
National Key Laboratory of Science and Technology on Space-Born Intelligent Information Processing
23 shared publications- Dong Guo
School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China
16 shared publications Huan Chang
16 shared publications- Junyuan Song
Beijing Institute of Technology
15 shared publications
Latest Publications
- 26 October 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (22) , 42154
Photonic-aided wideband mm-wave transmission with scalable subband-selective reception based on AGFB-OFDM
As the bandwidth of photonic-aided millimeter-wave (mm-wave) transmission continues to expand, a growing mismatch emerges between the wideband capability of the photonic front end and the limited digital processing capability at the receiver. To address this mismatch while supporting flexible spectral access and heterogeneous services, this paper proposes a photonic-aided wideband mm-wave transmission architecture with scalable subband-selective reception based on adaptive generalized filter bank orthogonal frequency-division multiplexing (AGFB-OFDM). AGFB-OFDM establishes an interface between wideband transmission and receiver processing through structured subbands with well-defined boundaries, generated using a two-stage inverse fast Fourier transform (IFFT) and adaptive polyphase filtering. At the receiver, selective subband extraction enables the processing granularity to be matched to the required spectral resources. Simulations with 8, 10, and 12 subbands demonstrate the scalability of the proposed scheme, with comparable or lower average bit error rate (BER) than conventional full-band reception at a given optical signal-to-noise ratio (OSNR). Experimentally, a 6-GHz signal comprising 10 subbands is transmitted over a 10-km standard single-mode fiber (SMF) and a 1.5-m wireless link at 26 GHz, achieving an aggregate data rate of 22.5 Gbit/s, a spectral utilization of 93.75%, and a spectral efficiency of 3.75 bit/s/Hz. At a photodiode (PD) input power of −6 dBm, all subbands achieve BERs below the hard-decision forward error correction (HD-FEC) threshold. The receiver can process each selected subband using a 512-point FFT instead of a monolithic 5120-point FFT for the entire aggregated bandwidth, demonstrating flexible receiver processing granularity. Furthermore, heterogeneous 16QAM/QPSK transmission across different subbands demonstrates the capability to support heterogeneous services.
- 26 October 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (22) , 42154
Photonic-aided wideband mm-wave transmission with scalable subband-selective reception based on AGFB-OFDM
As the bandwidth of photonic-aided millimeter-wave (mm-wave) transmission continues to expand, a growing mismatch emerges between the wideband capability of the photonic front end and the limited digital processing capability at the receiver. To address this mismatch while supporting flexible spectral access and heterogeneous services, this paper proposes a photonic-aided wideband mm-wave transmission architecture with scalable subband-selective reception based on adaptive generalized filter bank orthogonal frequency-division multiplexing (AGFB-OFDM). AGFB-OFDM establishes an interface between wideband transmission and receiver processing through structured subbands with well-defined boundaries, generated using a two-stage inverse fast Fourier transform (IFFT) and adaptive polyphase filtering. At the receiver, selective subband extraction enables the processing granularity to be matched to the required spectral resources. Simulations with 8, 10, and 12 subbands demonstrate the scalability of the proposed scheme, with comparable or lower average bit error rate (BER) than conventional full-band reception at a given optical signal-to-noise ratio (OSNR). Experimentally, a 6-GHz signal comprising 10 subbands is transmitted over a 10-km standard single-mode fiber (SMF) and a 1.5-m wireless link at 26 GHz, achieving an aggregate data rate of 22.5 Gbit/s, a spectral utilization of 93.75%, and a spectral efficiency of 3.75 bit/s/Hz. At a photodiode (PD) input power of −6 dBm, all subbands achieve BERs below the hard-decision forward error correction (HD-FEC) threshold. The receiver can process each selected subband using a 512-point FFT instead of a monolithic 5120-point FFT for the entire aggregated bandwidth, demonstrating flexible receiver processing granularity. Furthermore, heterogeneous 16QAM/QPSK transmission across different subbands demonstrates the capability to support heterogeneous services.
- 21 October 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (22) , 41412
Investigation of multipath interference impairments and a FEC-aided equalization solution in IM/DD PAM4 interconnects
This work investigates multipath interference (MPI) impairments and proposes a forward error correction (FEC) aided MPI equalization (FAME) solution for intensity-modulation direct-detection (IM/DD) PAM4 interconnects. We comprehensively analyze the impact of laser linewidth, signal-to-interference ratio (SIR), and reflection delay on MPI generation through both theoretical models and experiments. To mitigate these impairments, a single-iteration frame is integrated by a FEC decoder and a simplified error-magnitude-dependent (EMD) equalizer. The decoded bits from the FEC decoder serve as reference symbols for the convergence of the equalizer, and the equalized symbols are fed back into the FEC decoder, establishing a closed-loop iterative process. This iterative mechanism enhances reference symbol fidelity compared to classical decision-directed schemes, effectively mitigating error propagation. Experimental validation via 60 GBaud PAM4 transmission over 2 km of standard single-mode fiber demonstrates up to 2.2 dB SIR tolerance improvement at the 2.4 × 10 −4 FEC threshold.
- 21 October 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (22) , 41412
Investigation of multipath interference impairments and a FEC-aided equalization solution in IM/DD PAM4 interconnects
This work investigates multipath interference (MPI) impairments and proposes a forward error correction (FEC) aided MPI equalization (FAME) solution for intensity-modulation direct-detection (IM/DD) PAM4 interconnects. We comprehensively analyze the impact of laser linewidth, signal-to-interference ratio (SIR), and reflection delay on MPI generation through both theoretical models and experiments. To mitigate these impairments, a single-iteration frame is integrated by a FEC decoder and a simplified error-magnitude-dependent (EMD) equalizer. The decoded bits from the FEC decoder serve as reference symbols for the convergence of the equalizer, and the equalized symbols are fed back into the FEC decoder, establishing a closed-loop iterative process. This iterative mechanism enhances reference symbol fidelity compared to classical decision-directed schemes, effectively mitigating error propagation. Experimental validation via 60 GBaud PAM4 transmission over 2 km of standard single-mode fiber demonstrates up to 2.2 dB SIR tolerance improvement at the 2.4 × 10 −4 FEC threshold.
- 20 October 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (22) , 41227
Real-time adaptive optics compensation for turbulence-impaired OAM-multiplexed optical communications
Orbital angular momentum (OAM) multiplexing provides additional spatial degrees of freedom for free-space optical communication (FSOC), but atmospheric turbulence induces wavefront distortion, desired-mode power loss, and inter-modal coupling, thereby degrading the received signal quality and bit error rate (BER). Wavefront-sensorless adaptive optics (AO) based on stochastic parallel gradient descent (SPGD) enables model-free compensation for turbulence-induced distortions. However, conventional SPGD often converges slowly and exhibits steady-state oscillations in high-dimensional control, prolonging insufficient wavefront correction and limiting the effectiveness of wavefront-distortion compensation. In this paper, we design a modified SPGD algorithm, called adaptive-decay and cosine-annealed Adam-SPGD (ACA-SPGD), to accelerate initial convergence and suppress steady-state fluctuations. ACA-SPGD dynamically adjusts the first-order momentum decay factor and employs a cosine-annealed step-size schedule to balance rapid optimization and stable convergence. To enable low-latency closed-loop operation and real-time AO compensation, we further implement a bandwidth-aware parallel pipelined architecture (BAPPA) with dual-channel streaming evaluation on an FPGA to execute ACA-SPGD and generate the 137-element DM control-voltage vector. Both the numerical and experimental results show that the proposed ACA-SPGD algorithm exhibits faster and more stable convergence performance than conventional SPGD and Adam-SPGD. The ACA-SPGD-based AO scheme effectively compensates for turbulence-induced wavefront distortions, thereby improving the communication performance of the OAM-FSOC system. For the present implementations, the runtime ratio (MATLAB/FPGA) is approximately 1283 ×, while the FPGA computational kernel maintains a computational error on the order of 10 −5 relative to the MATLAB reference implementation.
- 20 October 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (22) , 41227
Real-time adaptive optics compensation for turbulence-impaired OAM-multiplexed optical communications
Orbital angular momentum (OAM) multiplexing provides additional spatial degrees of freedom for free-space optical communication (FSOC), but atmospheric turbulence induces wavefront distortion, desired-mode power loss, and inter-modal coupling, thereby degrading the received signal quality and bit error rate (BER). Wavefront-sensorless adaptive optics (AO) based on stochastic parallel gradient descent (SPGD) enables model-free compensation for turbulence-induced distortions. However, conventional SPGD often converges slowly and exhibits steady-state oscillations in high-dimensional control, prolonging insufficient wavefront correction and limiting the effectiveness of wavefront-distortion compensation. In this paper, we design a modified SPGD algorithm, called adaptive-decay and cosine-annealed Adam-SPGD (ACA-SPGD), to accelerate initial convergence and suppress steady-state fluctuations. ACA-SPGD dynamically adjusts the first-order momentum decay factor and employs a cosine-annealed step-size schedule to balance rapid optimization and stable convergence. To enable low-latency closed-loop operation and real-time AO compensation, we further implement a bandwidth-aware parallel pipelined architecture (BAPPA) with dual-channel streaming evaluation on an FPGA to execute ACA-SPGD and generate the 137-element DM control-voltage vector. Both the numerical and experimental results show that the proposed ACA-SPGD algorithm exhibits faster and more stable convergence performance than conventional SPGD and Adam-SPGD. The ACA-SPGD-based AO scheme effectively compensates for turbulence-induced wavefront distortions, thereby improving the communication performance of the OAM-FSOC system. For the present implementations, the runtime ratio (MATLAB/FPGA) is approximately 1283 ×, while the FPGA computational kernel maintains a computational error on the order of 10 −5 relative to the MATLAB reference implementation.
- 14 September 2026
- Published by Optica Publishing Group in Optics Letters
- Vol. 51 (18) , 5249-5252
Robust Quinn-based hierarchical compensation for linewidth-induced impairment mitigation in RoF-OTFS ISAC system
Orthogonal time-frequency space (OTFS) is a promising 6G integrated sensing and communication (ISAC) waveform, with radio-over-fiber (RoF) serving as its essential distribution backbone. However, laser linewidth-induced phase noise hinders its integration by shifting signals off the discrete delay-Doppler grid. We propose a hierarchical compensation framework using Quinn’s estimator to precisely rectify these fractional offsets across multiple domains. Simulations verify that the proposed scheme significantly enhances the laser linewidth tolerance of the OTFS waveform; specifically, it maintains the BER below the HD-FEC threshold for linewidths up to 140 kHz using QPSK and 10 kHz using 16-QAM, outperforming uncompensated systems. Experimental results over a 20 km fiber and 1 m wireless link with a 2 GHz bandwidth achieve a 4 Gbps peak rate, successfully surpassing the HD-FEC threshold. This approach enables the use of cost-effective lasers, providing a robust solution for high-capacity, photonic-assisted 6G ISAC networks.
- 14 September 2026
- Published by Optica Publishing Group in Optics Letters
- Vol. 51 (18) , 5249-5252
Robust Quinn-based hierarchical compensation for linewidth-induced impairment mitigation in RoF-OTFS ISAC system
Orthogonal time-frequency space (OTFS) is a promising 6G integrated sensing and communication (ISAC) waveform, with radio-over-fiber (RoF) serving as its essential distribution backbone. However, laser linewidth-induced phase noise hinders its integration by shifting signals off the discrete delay-Doppler grid. We propose a hierarchical compensation framework using Quinn’s estimator to precisely rectify these fractional offsets across multiple domains. Simulations verify that the proposed scheme significantly enhances the laser linewidth tolerance of the OTFS waveform; specifically, it maintains the BER below the HD-FEC threshold for linewidths up to 140 kHz using QPSK and 10 kHz using 16-QAM, outperforming uncompensated systems. Experimental results over a 20 km fiber and 1 m wireless link with a 2 GHz bandwidth achieve a 4 Gbps peak rate, successfully surpassing the HD-FEC threshold. This approach enables the use of cost-effective lasers, providing a robust solution for high-capacity, photonic-assisted 6G ISAC networks.
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 44 (17) , 7421-7427
Ultra-High Bandwidth Efficient Implementation Employing Geometric Shaped Symbol Division Multiplexing in Optical DMT Interconnection
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 44 (17) , 7421-7427
Ultra-High Bandwidth Efficient Implementation Employing Geometric Shaped Symbol Division Multiplexing in Optical DMT Interconnection
- 31 August 2026
- Published by Optica Publishing Group in Optics Continuum
- Vol. 5 (9) , 3035
Robust enhanced low-complexity Kullback-Leibler divergence-based carrier phase recovery under probabilistic shaping conditions for a coherent optical fiber systems
We propose what we believe to be a novel weight-balanced Kullback–Leibler divergence algorithm based on squared symbols (WB-KLSQ) to mitigate the deterioration in carrier phase recovery (CPR) performance of the conventional KLSQ algorithm under strong probabilistic shaping (PS) conditions for coherent optical systems, particularly when the length of the a posteriori training sequence (TS) and the size of the look-up table (LUT) are limited. The proposed WB-KLSQ scheme effectively mitigates the mismatch between the LUT and the actual received symbol distribution by extending the TS with a known Maxwell–Boltzmann (MB) distribution, thereby maximizing the CPR performance of the KLSQ algorithm. Simulation results demonstrate that the proposed WB-KLSQ scheme can improve the total power budget by 1.1 dB under different launched optical powers (LOPs) and achieve a signal-to-noise ratio (SNR) tolerance of 0.3 dB under different SNRs, with only a 21.76% increase in computational complexity compared with the conventional KLSQ algorithm.
- 31 August 2026
- Published by Optica Publishing Group in Optics Continuum
- Vol. 5 (9) , 3035
Robust enhanced low-complexity Kullback-Leibler divergence-based carrier phase recovery under probabilistic shaping conditions for a coherent optical fiber systems
We propose what we believe to be a novel weight-balanced Kullback–Leibler divergence algorithm based on squared symbols (WB-KLSQ) to mitigate the deterioration in carrier phase recovery (CPR) performance of the conventional KLSQ algorithm under strong probabilistic shaping (PS) conditions for coherent optical systems, particularly when the length of the a posteriori training sequence (TS) and the size of the look-up table (LUT) are limited. The proposed WB-KLSQ scheme effectively mitigates the mismatch between the LUT and the actual received symbol distribution by extending the TS with a known Maxwell–Boltzmann (MB) distribution, thereby maximizing the CPR performance of the KLSQ algorithm. Simulation results demonstrate that the proposed WB-KLSQ scheme can improve the total power budget by 1.1 dB under different launched optical powers (LOPs) and achieve a signal-to-noise ratio (SNR) tolerance of 0.3 dB under different SNRs, with only a 21.76% increase in computational complexity compared with the conventional KLSQ algorithm.
- 22 June 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (13) , 23968-23982
Joint channel coding and fiber nonlinearity compensation with adaptive perturbation estimation in coherent optical transmission systems
A joint channel coding and fiber nonlinearity compensation scheme based on adaptive perturbation estimation is proposed for coherent optical transmission systems. The proposed PB-JCNC framework combines perturbation-based nonlinear modeling with iterative soft-information exchange from the forward error correction (FEC) decoder, enabling accurate estimation of nonlinear distortion without requiring precise knowledge of transmission link parameters. Simulation and experimental results demonstrate the effectiveness of the proposed method for both PDM-16QAM and PDM-64QAM signals. For 400 km PDM-64QAM transmission, a pre-FEC Q-factor improvement of up to 2.03 dB (simulation) and 1.46 dB (experiment) is achieved, along with a post-FEC launch power tolerance gain of up to 0.38 dB. For 1000 km PDM-16QAM transmission, consistent performance improvements are also observed. The results indicate that the PB-JCNC highlights the critical role of soft-information reliability in iterative nonlinear compensation.
- 22 June 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (13) , 23968-23982
Joint channel coding and fiber nonlinearity compensation with adaptive perturbation estimation in coherent optical transmission systems
A joint channel coding and fiber nonlinearity compensation scheme based on adaptive perturbation estimation is proposed for coherent optical transmission systems. The proposed PB-JCNC framework combines perturbation-based nonlinear modeling with iterative soft-information exchange from the forward error correction (FEC) decoder, enabling accurate estimation of nonlinear distortion without requiring precise knowledge of transmission link parameters. Simulation and experimental results demonstrate the effectiveness of the proposed method for both PDM-16QAM and PDM-64QAM signals. For 400 km PDM-64QAM transmission, a pre-FEC Q-factor improvement of up to 2.03 dB (simulation) and 1.46 dB (experiment) is achieved, along with a post-FEC launch power tolerance gain of up to 0.38 dB. For 1000 km PDM-16QAM transmission, consistent performance improvements are also observed. The results indicate that the PB-JCNC highlights the critical role of soft-information reliability in iterative nonlinear compensation.
- 29 April 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (9) , 16970-16982
Experimental demonstration of 10Gbaud 64QAM SDM self-homodyne coherent optical transmission in U band with low-complexity sparsity-aware volterra equalization algorithm
In this paper, we design and implement a U-band 10GBaud 64QAM space-division multiplexing (SDM) transmission system based on a multi-core fiber (MCF) with self-homodyne coherent detection (SHCD) architecture, achieving a potential aggregate gross capacity of 39.6 Tbit/s. To address the prohibitive computational overhead and asymmetric nonlinearities inherent in this high-order modulation configuration, we propose a sparsity-aware real-valued intensity-dependent volterra (SA-RIV) equalizer optimized by the zero-attracting recursive least squares (ZA-RLS) algorithm. By decoupling In-phase and Quadrature (I/Q) processing into independent real-valued paths and employing a zero-attracting mechanism to prune redundant coefficients, the proposed method achieves significant complexity reduction. The experimental results show that the proposed SA-RIV provides a significant sensitivity gain of 2.42 dB while maintaining exceptional robustness against modulator bias instability, with all channels operating reliably within the 15% soft-decision forward error correction (SD-FEC) threshold. The SA-RIV achieves a 90% reduction in computational complexity compared to traditional volterra implementations.
- 29 April 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (9) , 16970-16982
Experimental demonstration of 10Gbaud 64QAM SDM self-homodyne coherent optical transmission in U band with low-complexity sparsity-aware volterra equalization algorithm
In this paper, we design and implement a U-band 10GBaud 64QAM space-division multiplexing (SDM) transmission system based on a multi-core fiber (MCF) with self-homodyne coherent detection (SHCD) architecture, achieving a potential aggregate gross capacity of 39.6 Tbit/s. To address the prohibitive computational overhead and asymmetric nonlinearities inherent in this high-order modulation configuration, we propose a sparsity-aware real-valued intensity-dependent volterra (SA-RIV) equalizer optimized by the zero-attracting recursive least squares (ZA-RLS) algorithm. By decoupling In-phase and Quadrature (I/Q) processing into independent real-valued paths and employing a zero-attracting mechanism to prune redundant coefficients, the proposed method achieves significant complexity reduction. The experimental results show that the proposed SA-RIV provides a significant sensitivity gain of 2.42 dB while maintaining exceptional robustness against modulator bias instability, with all channels operating reliably within the 15% soft-decision forward error correction (SD-FEC) threshold. The SA-RIV achieves a 90% reduction in computational complexity compared to traditional volterra implementations.
- 27 April 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (9) , 16599-16608
4D modulation scheme employing concatenated multi-level coding for short-reach coherent optical communication systems
We propose and experimentally demonstrate a four-dimensional (4D) modulation scheme employing concatenated multi-level coded (MLC) for short-reach coherent optical communication systems. Different from conventional set partitioning (SP) quadrature amplitude modulation assisted with Exclusive OR parity bits, the polar-based eight-level MLC architecture is proposed to provide inner- and inter-polarization protection by encoding all significant bits across the 4D dual-polarization 16-ary quadrature amplitude modulation (4D-DP-16QAM), offering enhanced system bandwidth efficiency. An outer concatenated encoder is further introduced to strengthen polarization-domain interdependence, which effectively boosts cross-polarization correlation and mitigates the error floor. For 4D decoding, a multi-stage decoding (MSD) is utilized to facilitate efficient soft-information exchange between polarization branches, resulting in improved robustness against inter-polarization impairments. Experimental results demonstrate that under a 60 GBaud 4D-DP-16QAM communication system over standard single-mode fiber, the proposed scheme achieves a 0.77 dB improvement in optical signal-to-noise ratio (OSNR) sensitivity and exhibits superior robustness to both polarization-dependent loss (PDL) and in-phase and quadrature (IQ) skew, compared with a conventional coherent optical DP-16QAM communication system.
- 27 April 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (9) , 16599-16608
4D modulation scheme employing concatenated multi-level coding for short-reach coherent optical communication systems
We propose and experimentally demonstrate a four-dimensional (4D) modulation scheme employing concatenated multi-level coded (MLC) for short-reach coherent optical communication systems. Different from conventional set partitioning (SP) quadrature amplitude modulation assisted with Exclusive OR parity bits, the polar-based eight-level MLC architecture is proposed to provide inner- and inter-polarization protection by encoding all significant bits across the 4D dual-polarization 16-ary quadrature amplitude modulation (4D-DP-16QAM), offering enhanced system bandwidth efficiency. An outer concatenated encoder is further introduced to strengthen polarization-domain interdependence, which effectively boosts cross-polarization correlation and mitigates the error floor. For 4D decoding, a multi-stage decoding (MSD) is utilized to facilitate efficient soft-information exchange between polarization branches, resulting in improved robustness against inter-polarization impairments. Experimental results demonstrate that under a 60 GBaud 4D-DP-16QAM communication system over standard single-mode fiber, the proposed scheme achieves a 0.77 dB improvement in optical signal-to-noise ratio (OSNR) sensitivity and exhibits superior robustness to both polarization-dependent loss (PDL) and in-phase and quadrature (IQ) skew, compared with a conventional coherent optical DP-16QAM communication system.
- 7 April 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (8) , 14044
Polar coded symbol division multiplexing scheme for bandwidth efficient short-reach optical interconnection
To mitigate the impairments by system bandwidth limitation, a joint polar coded pulse amplitude modulation (PAM) scheme based on symbol division multiplexing (SDM) is proposed and experimentally demonstrated in an intensity modulation and directly detected optical interconnection. The SDM-PAM4 employs a polar coded many-to-one mapping scheme to enhance information capacity per symbol, thus improving the bandwidth efficiency and reducing the symbol rate requirement while maintaining the same data rate. To realize the demultiplexing SDM (De-SDM) without iterative calculation, a systematic polar encoder with the arrangement of ambiguous bits in specific parity bits is proposed to obtain optimal system performance in simultaneous De-SDM and polar decoding. Experimental results show that the proposed scheme improves receiver power sensitivity by 2.5 dB compared with conventional PAM4 under constrained system bandwidth, while achieving the same data rate over 2-km standard single-mode fiber transmission. Benefiting from the short code length and flexible rates, the polar coded SDM is highly promising for bandwidth limited optical interconnection.
- 7 April 2026
- Published by Optica Publishing Group in Optics Express
- Vol. 34 (8) , 14044
Polar coded symbol division multiplexing scheme for bandwidth efficient short-reach optical interconnection
To mitigate the impairments by system bandwidth limitation, a joint polar coded pulse amplitude modulation (PAM) scheme based on symbol division multiplexing (SDM) is proposed and experimentally demonstrated in an intensity modulation and directly detected optical interconnection. The SDM-PAM4 employs a polar coded many-to-one mapping scheme to enhance information capacity per symbol, thus improving the bandwidth efficiency and reducing the symbol rate requirement while maintaining the same data rate. To realize the demultiplexing SDM (De-SDM) without iterative calculation, a systematic polar encoder with the arrangement of ambiguous bits in specific parity bits is proposed to obtain optimal system performance in simultaneous De-SDM and polar decoding. Experimental results show that the proposed scheme improves receiver power sensitivity by 2.5 dB compared with conventional PAM4 under constrained system bandwidth, while achieving the same data rate over 2-km standard single-mode fiber transmission. Benefiting from the short code length and flexible rates, the polar coded SDM is highly promising for bandwidth limited optical interconnection.