High-speed photonic neuromorphic computing using recurrent optical spectrum slicing neural networks

Photonic computing and signal processing are rapidly regaining attention, capitalizing on the saturation of digital electronics capabilities, owed to the ending of the Moore's law era. Optical communications are among the fields in which digital signal processing (DSP) struggles to offer a viable long-term solution, as its power consumption rises significantly in the latest generation optical links. For example, the traditional intensity modulation and direct detection (IM/DD) systems suffer from challenging problems such as the chromatic dispersion induced power fading effect, which necessitates the use of powerful digital equalizers in the 800G and 1.6T optical transceivers. Here, we experimentally validate our previous numerical works in recurrent optical spectrum slicing (ROSS) accelerators for dispersion compensation in IM/DD links. For this, we utilize recurrent filters implemented both through a waveshaper and by exploiting novel programmable photonic platforms. The ROSS accelerators exploit frequency processing in order to mitigate the power fading effect, which hinders the scalability of IM/DD systems. By equalizing even 80 km of 64 Gb/s PAM-4 transmission in C-band, we prove that our system can offer an appealing solution in highly dispersive channels. We employ the simplest digital equalization in the form of a feed-forward equalizer (FFE), avoiding throughput, latency, and complexity restrictions other digital equalizers impose. We achieve to reduce the bit error rate (BER) from 0.11 to less than 1 × 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−2</sup> with the use of only two filter nodes, whereas BER approaches 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-3</sup> when three nodes are incorporated. These results correspond to almost two orders of magnitude BER gain.

Traditional electrical switching network faces the problems of limited bandwidth and large energy consumption. The optoelectronic hybrid data network fully combines the advantages of optical switching and electrical switching, and dynamically adjusts the topology and bandwidth in accordance to the demand, thereby effectively improving network performance and reducing power consumption. A high-performance distributed optoelectronic switching network architecture called DOIN_W was proposed in the study, aiming to solve the problems of the scalability of the optoelectronic switching network architecture and the insufficient flexibility of network resource allocation. The architecture was connected topologically according to the method of 2DTorus, thereby effectively improving the network scale of the photoelectric switching network and supporting the dynamic adjustment of the network scale. The optical switching of DOIN_W adopted the "broadcastselect" method to support different forms of broadcasting communication. A multi-dimensional optical signal switch was designed, and the optical signal broadcast of the switch can reach any of the optical switches in DOIN_W, thereby supporting the direct connection between optical switches. Considering the problems that the multi-broadcast service of DOIN_W was unable to access freely, stateful optical signal communication was hard to maintain and the single path weight resulted in lack priority of communication, the OpenFlow network protocol of the cloud computing network was introduced in the DOIN_W architecture. The anycast mechanism of DOIN_W was optimized based on this protocol. Based on the stream-level network simulator for network performance simulation analysis, the DOIN_W architecture can meet the requirements of network scalability and flexibility. Compared with the OVS architecture and Jellyfish architecture, the DOIN_W architecture can effectively reduce the average flow completion time by more than 30% and reduce average energy consumption by over 25%.

Underwater Optical Wireless Communication (UWOC) becomes an emerging underwater communication technology, with high-data rates over relatively medium transmission ranges. When optical wireless signal transmitted in ocean water channel, it will suffer from drastic scattering and absorption due to water molecules, dissolved particles, air bubbles, and turbulence. Absorption and scattering of the transmitted wireless optical signal in underwater channel led to attenuation in optical signal power. Optical signal attenuation over underwater channel is an aggregate of` different parameters effects that changed frequently, then practical measuring of this attenuation is complicated, difficult, expensive, and time-consuming process. In this work, improved neural network optimized with future search algorithm (FANN) was proposed, as an efficacious solution to obtain an accurate, relabel values of attenuation coefficient in different water types and conditions. The proposed FANN model provides a good much results to the practical measured values. The performance of the proposed FANN model was evaluated using mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE) error indices. The errors in attenuation coefficient values obtained by the proposed FANN model had been calculated and its values are very acceptable which are lie lower than 10-4. The performance of the proposed FANN model shows excellent results which indicate the superior performance of the proposed FANN model.

High-throughput real-time optical integrators are of great importance for applications that require ultrafast optical information processing, such as real-time phase reconstruction of ultrashort optical pulses. In many of these applications, integration of wide optical bandwidth signals is required. Unfortunately, conventional all-optical integrators based on passive devices are usually sensitive to the wavelength and bandwidth of the optical carrier. Here, we propose and demonstrate a passive all-optical intensity integrator whose operation is independent of the optical signal wavelength and bandwidth. The integrator is implemented based on modal dispersion in a multimode waveguide. By controlling the launch conditions of the input beam, the device produces a rectangular temporal impulse response. Consequently, a temporal intensity integration of an arbitrary optical waveform input is performed within the rectangular time window. The key advantage of this device is that the integration operation can be performed independent of the input signal wavelength and optical carrier bandwidth. This is preferred in many applications where optical signals of different wavelengths are involved. Moreover, thanks to the use of a relatively short length of multimode waveguide, lower system latency is achieved compared to the systems using long dispersive fibers. To illustrate the versatility of the optical integrator, we demonstrate temporal intensity integration of optical waveforms with different wavelengths and optical carrier bandwidths. Finally, we use this device to perform high-throughput, single-shot, real-time optical phase reconstruction of phase-modulated signals at telecommunications bit rates.

Multiple parallel channels are ubiquitous in optical communications, with spatial division multiplexing (separate physical paths) and wavelength division multiplexing (separate optical wavelengths) being the most common forms. Here, we investigate the viability of polarization division multiplexing, the separation of distinct parallel optical communication channels through the polarization properties of light. Two or more linearly polarized optical signals (at different polarization angles) are transmitted through a common medium, filtered using aluminum nanowire optical filters fabricated on-chip, and received using individual silicon photodetectors (one per channel). The entire receiver (including optics) is compatible with standard CMOS fabrication processes. The filter model is based upon an input optical signal formed as the sum of the Stokes vectors for each individual channel, transformed by the Mueller matrix that models the filter proper, resulting in an output optical signal that impinges on each photodiode. The results show that two- and three-channel systems can operate with a fixed-threshold comparator in the receiver circuit, but four-channel systems (and larger) will require channel coding of some form. For example, in the four-channel system, 10 of 16 distinct bit patterns are separable by the receiver. The model supports investigation of the range of variability tolerable in the fabrication of the on-chip polarization filters.

Photonic integration is a critical emerging trend in the development of photonic devices. Its significance parallels that of the microelectronic integrated circuits (IC) albeit it has developed slower due to its diversity and often contrasting requirements. Photonic integrated circuits (PIC) is basically made up of small but important building blocks combined to realize even more complex circuits so that it may perform wider range of functionalities on optical signals used in various applications, such as optical communication, instrumentation, sensing, and signal processing.

Deep neural networks (DNNs) have been applied to recover signals in optical communication systems and have shown competence of mitigating linear and nonlinear distortions. However, as the data throughput increases, the heavy computational cost of DNNs impedes them from rapid and power-efficient processing. In this paper, we propose an optical communication signal recovery technology based on a photonic convolutional processor, which is realized by dispersion delay unit and wavelength division multiplexing. Based on the photonic convolutional processor, we implement an optoelectronic convolutional neural network (OECNN) for signal post-equalization and experimentally demonstrate on 16QAM and 32QAM of an optical wireless communication system. With system parameters optimization, we verify that the OECNN can achieve accurate signal recovery where the bit error ratio (BER) is below the 7% forward error correction threshold of 3.8×10-3 at 2Gbps. With adding the OECNN-based nonlinear compensation, compared with only linear compensation, we improve the quality (Q) factor by 3.35 dB at 16QAM and 3.30 dB at 32QAM, which is comparable to that of an electronic neural network. This work proves that the photonic implementation of DNN is promising to provide a fast and power-efficient solution for optical communication signal processing.

As human society to the information in the process of moving and growing demand for bandwidth communications capacity, the optical of new modulation formats increasingly attention and quickly play an important role in optical communications. How can the system bit error rate within a certain degree of stability while still maintaining high-speed long-distance dispersal system, has been a popular issue is the optical communications industry. Starting from the optical modulation format herein, the generation process of the system introduced various optical signal modulation format, the optical signal through the optical fiber was studied and the performance of the simulation, on the basis of the design of advanced optical modulation formats in an optical fiber communication system .

Most modern digital receivers sample the received RF signal at an intermediate frequency (IF), then downconvert to the baseband in the digital domain. The baseband samples are subsequently correlated with a time-limited square pulse, which is readily implemented as an integrator. Since correlation is performed at the sampling frequency, the receiver correlates at an unnecessarily high frequency. This high frequency results in significant power dissipation especially if parallel correlators are employed. By downsampling after appropriate filtering to slightly above the signal Nyquist rate, correlation is achieved at a fraction of the sampling frequency, resulting in corresponding reduction in power consumption. Furthermore, by resampling to an integer number of samples per chip, architectural changes in the correlators are possible, allowing even more reduction in power consumption. In this paper, a resampling correlator with minimal hardware overhead that outperforms the conventional correlator is presented.

This paper reviews opportunities of optical signal form conversion as typified by time-space conversion in optical signal processing. Several examples of typical ultra-fast optical signal processing using optical signal form conversion are described and their applications are introduced in respect to photonic networks, ultra-fast measurement, and so on.

Linearly chirped waveform generation with large time-bandwidth product using sweeping laser and dual-polarization modulator

Machine learning (ML) and neuromorphic computing have been enforcing problem-solving in many applications. Such approaches found fertile ground in optical communications, a technological field that is very demanding in terms of computational speed and complexity. The latest breakthroughs are strongly supported by advanced signal processing, implemented in the digital domain. Algorithms of different levels of complexity aim at improving data recovery, expanding the reach of transmission, validating the integrity of the optical network operation, and monitoring data transfer faults. Lately, the concept of reservoir computing (RC) inspired hardware implementations in photonics that may offer revolutionary solutions in this field. In a brief introduction, I discuss some of the established digital signal processing (DSP) techniques and some new approaches based on ML and neural network (NN) architectures. In the main part, I review the latest neuromorphic computing proposals that specifically apply to photonic hardware and give new perspectives on addressing signal processing in optical communications. I discuss the fundamental topologies in photonic feed-forward and recurrent network implementations. Finally, I review the photonic topologies that were initially tested for channel equalization benchmark tasks, and then in fiber transmission systems, for optical header recognition, data recovery, and modulation format identification.

The rapid progress toward telecommunication by optical fibers and optical information processing using integrated optics has raised considerable scientific interest in appropriate photodiodes for these applications1–4. While there is universal appreciation that silicon photodiodes have the best performance characteristics of any available photodiode, other features intervene, notably the properties of the transmission medium5, which give a considerable advantage to optical signal transmission at wavelengths in the infra-red where silicon is not sensitive. It is difficult to appreciate the true performance of a photodiode without also considering the conditions under which it must operate. A detector can be characterized by three properties: spectral response, rise-time, and noise. In even an average photodiode, the quantum efficiency and rise-time are usually quite good and near the maximum possible. Noise is the crucial feature in determining photodiode performance, because detection in optical communication is concerned with very weak signals. In this case the detector sensitivity and thus the system performance is always limited by the noise. Noise arises from three different origins: shot noise in the optical signal, shot noise in the detector and excess noise generated by avalanche gain in the detector, and amplifier thermal noise.

PurposeOsteoarthritis (OA) is a degenerative disease starting with key molecular events that ultimately lead to the breakdown of the cartilage. The purpose of this study is to use two imaging methods that are sensitive to molecular and macromolecular changes in OA to better characterize the disease process in human osteoarthritic cartilage.ProceduresHuman femoral condyles were collected from patients diagnosed with severe OA during total knee replacement surgeries. T1ρ and T2 magnetic resonance measurements were obtained using a 3-Tesla whole body scanner to assess macromolecular changes in the damaged cartilage matrix. Optical imaging was performed on specimens treated with MMPSense 680 to assess the matrix metalloproteinase (MMP) activity. A linear regression model was used to assess the correlation of MMP optical data with T1ρ magnetic resonance (MR) measurements. Slices from a representative specimen were removed from regions with high and low optical signals for subsequent histological analysis.ResultsAll specimens exhibit high T1ρ and T2 measurements in the range of 48–75 ms and 36–69 ms, respectively. They also show intense photon signals (0.376 to 7.89 × 10−4 cm2) from the activated MMPSense 680 probe, indicative of high MMP activity. The analysis of variance test of the regression model indicates a positive correlation between the MMP optical signal and T1ρ measurements (R2 = 0.8936, P = 0.0044). Histological data also confirmed that regions with high MMP optical signal and intense T1ρ relaxation exhibit severe clefting, abnormal tidemarks, and irregular cellularity.ConclusionsThe high T1ρ and T2 measurements suggest that there is a severe loss of proteoglycans with high water mobility in the damaged cartilage. The intense optical signals found in these specimens indicate the presence of active MMPs, and the positive correlation with T1ρ measurements implicates MMP’s involvement in OA progression, characterized by a severe loss of proteoglycans in the cartilage matrix. The bimodal approach using optical and MR imaging may provide key molecular and macromolecular information of the disease pathway, offering insights toward the development of new tools for the early detection, treatment, and/or prevention of OA.

We study the computational capabilities of a biologically inspired neural model where the synaptic weights, the connectivity pattern, and the number of neurons can evolve over time rather than stay static. Our study focuses on the mere concept of plasticity of the model so that the nature of the updates is assumed to be not constrained. In this context, we show that the so-called plastic recurrent neural networks (RNNs) are capable of the precise super-Turing computational power--as the static analog neural networks--irrespective of whether their synaptic weights are modeled by rational or real numbers, and moreover, irrespective of whether their patterns of plasticity are restricted to bi-valued updates or expressed by any other more general form of updating. Consequently, the incorporation of only bi-valued plastic capabilities in a basic model of RNNs suffices to break the Turing barrier and achieve the super-Turing level of computation. The consideration of more general mechanisms of architectural plasticity or of real synaptic weights does not further increase the capabilities of the networks. These results support the claim that the general mechanism of plasticity is crucially involved in the computational and dynamical capabilities of biological neural networks. They further show that the super-Turing level of computation reflects in a suitable way the capabilities of brain-like models of computation.