Research & Publications

Publications

2022

1. Journal

   Legacy-Agnostic Band Integration Approach: A Cost-Effective and Seamless C+ L-Band Optical Network Migration/Sec Frame-Based Phase Locked Loop on GPU

   MH. Shammakhi, F. Arpanaei, H. Beyranvand, H. Sheikhzadeh

   Journal of Lightwave Technology, 2024

   Abstract Link

   This paper addresses the escalating demand for enhanced data transmission rates and increased network capacity in elastic optical networks. We propose a novel legacy-agnostic band integration approach for cost-effective and seamless C+L-band optical network migration, focusing on a “pay-as-you-grow” (PaYG) strategy to optimize capital expenditure (CAPEX). This approach leverages advanced modeling techniques, including the enhanced generalized Gaussian noise (EGGN) model, to account for inter-channel stimulated Raman scattering (ISRS) and other nonlinear effects, ensuring optimal quality of transmission (QoT). Through extensive simulations on real-world network topologies, we compare the performance of various PaYG algorithms against a traditional Day-One (DO) approach, demonstrating significant gains in throughput and CAPEX efficiency. Notably, the proposed PaYG-WAR (PaYG-With All Reconfigurations) outperforms other strategies, maximizing throughput while minimizing the deployment of costly infrastructure components. Our results provide valuable insights for telecom operators seeking a financially viable and practical solution for migrating optical networks to C+L-band operation. Specifically, PaYG-WAR achieves 20% higher throughput compared to DO and PaYG-WoR (PaYG-Without Reconfigurations) and 17% more than PaYG-WFR (PaYG-With Full Reconfigurations), while requiring 25% less CAPEX than DO-CHB (DO-Channel Based). These findings underscore the potential of PaYG-WAR in facilitating efficient, cost-effective network upgrades to meet next-generation communication demands.

2. Journal

   Gb/Sec Frame-Based Phase Locked Loop on GPU

   M.H. Shammakhi, N. Kianersi, P. Haji Faraji, M. Hoseinzadeh

   In Elsevier PARCO, 2022

   Abstract Link

   This paper addresses the challenge of extracting bits from wideband signals with different modulations in telecommunication networks. Typically performed by FPGAs and GPUs, the sequential nature of Timing Recovery and Phase Locked Loop (PLL) processes limits processing rates on the GPU. The paper introduces a frame-based method that enables parallel processing, boosting the processing rate to several GSample/sec. Unlike conventional PLL algorithms, the proposed algorithm accurately recovers phase and frequency offsets from the first output, without requiring a locked time. Each subframe is processed in parallel, and a separate kernel compensates for phase gaps between subframes at a high rate. The approach presented is significantly faster than the classic PLL using the Costas loop and achieves comparable or better error vector magnitude values.

3. Journal

   GPU-Based Parallel Algorithm for Wideband Signal Timing Recovery

   M.H. Shammakhi, P. Haji Faraji, M. Mohammadi, M. Hoseinzadeh

   TURCOMAT, 2022

   Abstract Link

   This paper introduces a new implementation of GPU-based symbol timing recovery in a coherent receiver. The proposed method is a parallel version of Gardner's method, which minimizes timing errors. Unlike Gardner's method that relies on sequential processing and feedback error, the proposed method uses a parallel timing recovery structure to enable fast error detection. The interpolation filter coefficients are calculated beforehand to detect timing errors. By comparing different parallel techniques on various GPUs, the method achieves significant improvements in processing speed, up to 100 times faster than Gardner's method. Performance evaluations demonstrate high-rate timing recovery (50 Msymb/sec on GTX 1050 Ti) through optimized GPU implementation.

2016

1. Conference

   A Digital Twin-Driven Framework for Cost-Effective Upgrading from C-Band to Multi-Band Optical Networks

   F. Arpanaei; M.H. Shammakhi; H. Beyranvand; H. Sheikhzadeh; M. Ranjbar Zefreh; S. Yan

   In IEEE ICTON, Barcelona, Spain, 2025

   Abstract Link

   The transition from C-band to ultra-wideband (C+L+S-band) optical networks is essential to meet escalating traffic demands but poses significant technical and economic challenges. This paper presents a digital twin (DT)-driven framework to evaluate and guide Pay-as-You-Grow (PaYG) migration strategies, enabling flexible and low-risk spectrum expansion. The proposed optical network digital twin (ONDT) accurately models inter-channel stimulated Raman scattering (ISRS), generalized signal-to-noise ratio (GSNR), and dynamic service reconfiguration to assess upgrade impacts in a virtual environment. We evaluate five PaYG strategies on the USB14 ultra-long-haul network using 64 GBaud transceivers and 20 THz total spectrum. Simulation results show that the most aggressive approach, PaYG-WAR (with all reconfigurations), achieves up to 17% higher throughput and 70% lower CAPEX per 100 Gbps of traffic compared to conservative strategies, while minimizing hardware upgrades. The ONDT-guided approach ensures cost-effective, scalable, and non-disruptive multi-band migration, making it a robust tool for future optical backbone evolution.

2. Conference

   Combined mRMR filter and sparse Bayesian classifier for analysis of gene expression data

   M. Soltani, M.H. Shammakhi, S. Khorram, H. Sheikhzadeh

   In IEEE ICSPIS, Tehran, Iran, 2016

   Abstract Link

   This paper addresses the challenge of identifying genes related to target diseases in gene expression microarray analysis. The high-dimensional nature of gene expression data poses a challenge in establishing a reliable method for disease classification. To tackle this, the paper introduces a new framework that combines two stages. Firstly, an information-theoretic forward feature selection algorithm called mRMR is utilized to identify a candidate set of the best features. Then, the RVM classifier, which is well-suited for gene data analysis, is employed. The RVM offers advantages such as providing membership probabilities for each class, which is crucial in diagnosing severe diseases, and a more sparse approach to model fitting, preventing overfitting. Experimental results demonstrate that the proposed algorithm outperforms previous methods in terms of classification accuracy and model sparsity. This framework has the potential to improve the diagnosis and understanding of various disorders based on gene expression data.

3. Conference

   Robust Tracking of Object Dimension Changes Using the Displacement Averaging Method

   V. Ghanbari, M.H. Shammakhi

   In IFIA, Tehran, Iran, 2016

   Abstract Link

   In this article, an accurate and efficient tracking method based on the Displacement Averaging method is proposed. This method utilizes the Displacement Averaging algorithm to find precise object positions and dimensions in consecutive frames. The Displacement Averaging method has achieved significant success in visual tracking due to its parameter-free nature. However, it lacks satisfactory performance when dealing with changing object dimensions. In this paper, an algorithm is introduced to make this method robust against object dimension changes. Firstly, the precise object position is obtained using the Displacement Averaging method. Then, by matching corresponding points of the object in the previous and current frames and applying a linear transformation, the new dimensions of the object are calculated. The proposed algorithm demonstrates good performance in cases where the object dimensions change, and this claim is supported by multiple experiments conducted on various datasets.

4. Conference

   Combined mRMR-MLPSVM scheme for high accuracy and low cost handwritten digits recognition

   M.H. Shammakhi , V. Pourahmadi ,P. Khavari , A. Mirzaei

   In IEEE MVIP, Tehran, Iran, 2015

   Abstract Link

   This paper introduces a new algorithm for accurate handwritten digit recognition with low implementation complexity. The algorithm utilizes the mRMR method to sort and select the best features from the training data. These selected features are then used as input for an MLPSVM classifier, which combines MLP and SVM techniques. The proposed algorithm achieves high accuracy of 96.1 and 98.14 on ORHD and MNIST datasets respectively, while maintaining lower complexity compared to existing methods.