arXiv subjects

Tianhua Xu

Publications and source records attributed to Tianhua Xu.

At least 19 recordsLinked to original sources

arXiv2026

ESCROW: Guarded and Dual-Objective Continual Maintenance for Agents in Policy-Governed Enterprise Workflows

LLM agents increasingly run policy-bound enterprise workflows, where they must apply rules consistently and stay auditable. Deploying such an agent is the start of its long-term maintenance cycle: it must adapt to a stream of operational signals, yet reliably turning these sparse, unlabeled signals into reusable skill revisions is hard, and a careless update can trade one task category's accuracy for the overall gain, revive a resolved failure, or land at an undeployable cost. We present ESCROW, a post-deployment maintenance framework that updates an agent's external, reviewable skills under a Strict Update Boundary: the LLM proposes candidate revisions, but only an empirically evaluated version is deployed. It combines distributed diagnosis with consensus, a per-category non-regression guard, cross-cycle anti-regression, and accuracy--cost Pareto search, emitting a versioned, auditable diff per change. In real production on our internal financial document-auditing system, it attains the strongest evaluated accuracy--cost trade-off among baselines, with a transfer probe on public $τ$-bench.

cs.AI↗

arXiv2026

Evolving Excellence: Automated Optimization of LLM-based Agents

Agentic AI systems built on large language models (LLMs) offer significant potential for automating complex workflows, from software development to customer support. However, LLM agents often underperform due to suboptimal configurations; poorly tuned prompts, tool descriptions, and parameters that typically require weeks of manual refinement. Existing optimization methods either are too complex for general use or treat components in isolation, missing critical interdependencies. We present ARTEMIS, a no-code evolutionary optimization platform that jointly optimizes agent configurations through semantically-aware genetic operators. Given only a benchmark script and natural language goals, ARTEMIS automatically discovers configurable components, extracts performance signals from execution logs, and evolves configurations without requiring architectural modifications. We evaluate ARTEMIS on four representative agent systems: the \emph{ALE Agent} for competitive programming on AtCoder Heuristic Contest, achieving a \textbf{$13.6\%$ improvement} in acceptance rate; the \emph{Mini-SWE Agent} for code optimization on SWE-Perf, with a statistically significant \textbf{10.1\% performance gain}; and the \emph{CrewAI Agent} for cost and mathematical reasoning on Math Odyssey, achieving a statistically significant \textbf{$36.9\%$ reduction} in the number of tokens required for evaluation. We also evaluate the \emph{MathTales-Teacher Agent} powered by a smaller open-source model (Qwen2.5-7B) on GSM8K primary-level mathematics problems, achieving a \textbf{22\% accuracy improvement} and demonstrating that ARTEMIS can optimize agents based on both commercial and local models.

cs.SE↗

arXiv2026

CARLA-Round: A Multi-Factor Simulation Dataset for Roundabout Trajectory Prediction

Accurate trajectory prediction of vehicles at roundabouts is critical for reducing traffic accidents, yet it remains highly challenging due to their circular road geometry, continuous merging and yielding interactions, and absence of traffic signals. Developing accurate prediction algorithms relies on reliable, multimodal, and realistic datasets; however, such datasets for roundabout scenarios are scarce, as real-world data collection is often limited by incomplete observations and entangled factors that are difficult to isolate. We present CARLA-Round, a systematically designed simulation dataset for roundabout trajectory prediction. The dataset varies weather conditions (five types) and traffic density levels (spanning Level-of-Service A-E) in a structured manner, resulting in 25 controlled scenarios. Each scenario incorporates realistic mixtures of driving behaviors and provides explicit annotations that are largely absent from existing datasets. Unlike randomly sampled simulation data, this structured design enables precise analysis of how different conditions influence trajectory prediction performance. Validation experiments using standard baselines (LSTM, GCN, GRU+GCN) reveal traffic density dominates prediction difficulty with strong monotonic effects, while weather shows non-linear impacts. The best model achieves 0.312m ADE on real-world rounD dataset, demonstrating effective sim-to-real transfer. This systematic approach quantifies factor impacts impossible to isolate in confounded real-world datasets. Our CARLA-Round dataset is available at https://github.com/Rebecca689/CARLA-Round.

cs.CV↗

arXiv2024

EasyECR: A Library for Easy Implementation and Evaluation of Event Coreference Resolution Models

Event Coreference Resolution (ECR) is the task of clustering event mentions that refer to the same real-world event. Despite significant advancements, ECR research faces two main challenges: limited generalizability across domains due to narrow dataset evaluations, and difficulties in comparing models within diverse ECR pipelines. To address these issues, we develop EasyECR, the first open-source library designed to standardize data structures and abstract ECR pipelines for easy implementation and fair evaluation. More specifically, EasyECR integrates seven representative pipelines and ten popular benchmark datasets, enabling model evaluations on various datasets and promoting the development of robust ECR pipelines. By conducting extensive evaluation via our EasyECR, we find that, \lowercase\expandafter{\romannumeral1}) the representative ECR pipelines cannot generalize across multiple datasets, hence evaluating ECR pipelines on multiple datasets is necessary, \lowercase\expandafter{\romannumeral2}) all models in ECR pipelines have a great effect on pipeline performance, therefore, when one model in ECR pipelines are compared, it is essential to ensure that the other models remain consistent. Additionally, reproducing ECR results is not trivial, and the developed library can help reduce this discrepancy. The experimental results provide valuable baselines for future research.

cs.AI↗

arXiv2024

A Low-Cost Multi-Band Waveform Security Framework in Resource-Constrained Communications

Traditional physical layer secure beamforming is achieved via precoding before signal transmission using channel state information (CSI). However, imperfect CSI will compromise the performance with imperfect beamforming and potential information leakage. In addition, multiple RF chains and antennas are needed to support the narrow beam generation, which complicates hardware implementation and is not suitable for resource-constrained Internet-of-Things (IoT) devices. Moreover, with the advancement of hardware and artificial intelligence (AI), low-cost and intelligent eavesdropping to wireless communications is becoming increasingly detrimental. In this paper, we propose a multi-carrier based multi-band waveform-defined security (WDS) framework, independent from CSI and RF chains, to defend against AI eavesdropping. Ideally, the continuous variations of sub-band structures lead to an infinite number of spectral features, which can potentially prevent brute-force eavesdropping. Sub-band spectral pattern information is efficiently constructed at legitimate users via a proposed chaotic sequence generator. A novel security metric, termed signal classification accuracy (SCA), is used to evaluate the security robustness under AI eavesdropping. Communication error probability and complexity are also investigated to show the reliability and practical capability of the proposed framework. Finally, compared to traditional secure beamforming techniques, the proposed multi-band WDS framework reduces power consumption by up to six times.

eess.SP↗

arXiv2019

Information Rate in Ultra-Wideband Optical Fiber Communication Systems Accounting for High-Order Dispersion

The effect of Kerr-induced optical fiber nonlinearities in C-band (5 THz) EDFA and C+L-band (12.5 THz) Raman-amplified optical communication systems has been studied considering the impact of third-order fiber dispersion. The performance of digital nonlinearity compensation with single channel and 250-GHz bandwidth in both EDFA and Raman amplified systems has been investigated, respectively. The achievable information rates (AIRs) and optimum code rates in each individual transmission channel have been evaluated for the DP-64QAM, the DP-256QAM and the DP-1024QAM modulation formats, both with and without the use of the probabilistic shaping technique. It is found that, for all considered modulation formats, the signal-to-noise ratios, AIRs and code rates exhibit significantly asymmetric behavior about the central channel due to the presence of the third-order dispersion. This provides a new insight that the forward error correction schemes have to be optimized asymmetrically, on a per-channel basis, to maximize the overall throughput.

eess.SP↗

arXiv2018

Dynamic physical layer equalization in optical communication networks

In optical transport networks, signal lightpaths between two terminal nodes can be different due to current network conditions. Thus the transmission distance and accumulated dispersion in the lightpath cannot be predicted. Therefore, the adaptive compensation of dynamic dispersion is necessary in such networks to enable flexible routing and switching. In this paper, we present a detailed analysis on the adaptive dispersion compensation using the least-mean-square (LMS) algorithm in coherent optical communication networks. It is found that the variable-step-size LMS equalizer can achieve the same performance with a lower complexity, compared to the traditional LMS algorithm.

cs.IT↗

arXiv2017

Digital Nonlinearity Compensation in High-Capacity Optical Fibre Communication Systems: Performance and Optimisation

Meeting the ever-growing information rate demands has become of utmost importance for optical communication systems. However, it has proven to be a challenging task due to the presence of Kerr effects, which have largely been regarded as a major bottleneck for enhancing the achievable information rates in modern optical communications. In this work, the optimisation and performance of digital nonlinearity compensation are discussed for maximising the achievable information rates in spectrally-efficient optical fibre communication systems. It is found that, for any given target information rate, there exists a trade-off between modulation format and compensated bandwidth to reduce the computational complexity requirement of digital nonlinearity compensation.

eess.SP↗

arXiv2017

Digital Signal Processing for Optical Communications and Networks I: Linear Compensation

The achievable information rates of optical communication networks have been widely increased over the past four decades with the introduction and development of optical amplifiers, coherent detection, advanced modulation formats, and digital signal processing techniques. These developments promoted the revolution of optical communication systems and the growth of Internet, towards the direction of high-capacity and long-distance transmissions. The performance of long-haul high-capacity optical fiber communication systems is significantly degraded by transmission impairments, such as chromatic dispersion, polarization mode dispersion, laser phase noise and Kerr fiber nonlinearities. With the entire capture of the amplitude and phase of the signals using coherent optical detection, the powerful compensation and effective mitigation of the transmission impairments can be implemented using the digital signal processing in electrical domain. This becomes one of the most promising techniques for next-generation optical communication networks to achieve a performance close to the Shannon capacity limit. This chapter will focus on the introduction and investigation of digital signal processing employed for channel impairments compensation based on the coherent detection of optical signals, to provide a roadmap for the design and implementation of realtime optical fiber communication systems.

cs.IT↗

arXiv2017

Carrier Phase Estimation in Dispersion-Unmanaged Optical Transmission Systems

The study on carrier phase estimation (CPE) approaches, involving a one-tap normalized least-mean-square (NLMS) algorithm, a block-wise average algorithm, and a Viterbi-Viterbi algorithm has been carried out in the long-haul high-capacity dispersion-unmanaged coherent optical systems. The close-form expressions and analytical predictions for bit-error-rate behaviors in these CPE methods have been analyzed by considering both the laser phase noise and the equalization enhanced phase noise. It is found that the Viterbi-Viterbi algorithm outperforms the one-tap NLMS and the block-wise average algorithms for a small phase noise variance (or effective phase noise variance), while the three CPE methods converge to a similar performance for a large phase noise variance (or effective phase noise variance). In addition, the differences between the three CPE approaches become smaller for higher-level modulation formats.

cs.IT↗

arXiv2017

Analysis of chromatic dispersion compensation and carrier phase recovery in long-haul optical transmission system influenced by equalization enhanced phase noise

The performance of long-haul coherent optical fiber transmission system is significantly affected by the equalization enhanced phase noise (EEPN), due to the interaction between the electronic dispersion compensation (EDC) and the laser phase noise. In this paper, we present a comprehensive study on different chromatic dispersion (CD) compensation and carrier phase recovery (CPR) approaches, in the n-level phase shift keying (n-PSK) and the n-level quadrature amplitude modulation (n-QAM) coherent optical transmission systems, considering the impacts of EEPN. Four CD compensation methods are considered: the time-domain equalization (TDE), the frequency-domain equalization (FDE), the least mean square (LMS) adaptive equalization are applied for EDC, and the dispersion compensating fiber (DCF) is employed for optical dispersion compensation (ODC). Meanwhile, three carrier phase recovery methods are also involved: a one-tap normalized least mean square (NLMS) algorithm, a block-wise average (BWA) algorithm, and a Viterbi-Viterbi (VV) algorithm. Numerical simulations have been carried out in a 28-Gbaud dual-polarization quadrature phase shift keying (DP-QPSK) coherent transmission system, and the results indicate that the origin of EEPN depends on the choice of chromatic dispersion compensation methods, and the effects of EEPN also behave moderately different in accordance to different carrier phase recovery scenarios.

cs.IT↗

arXiv2017

Achievable information rates estimates in optically-amplified transmission systems using nonlinearity compensation and probabilistic shaping

Achievable information rates (AIRs) of wideband optical communication systems using ~40 nm (~5 THz) EDFA and ~100 nm (~12.5 THz) distributed Raman amplification are estimated based on a first-order perturbation analysis. The AIRs of each individual channel have been evaluated for DP-64QAM, DP-256QAM, and DP-1024QAM modulation formats. The impact of full-field nonlinear compensation (FF-NLC) and probabilistically shaped constellations using a Maxwell-Boltzmann distribution were studied and compared to electronic dispersion compensation. It is found that a probabilistically shaped DP-1024QAM constellation combined with FF-NLC yields AIRs of ~75 Tbit/s for the EDFA scheme and ~223 Tbit/s for the Raman amplification scheme over 2000 km standard single mode fibre transmission.

cs.IT↗

arXiv2016

Close-Form Expression of One-Tap Normalized LMS Carrier Phase Recovery in Optical Communication Systems

The performance of long-haul high speed coherent optical fiber communication systems is significantly degraded by the laser phase noise and the equalization enhanced phase noise (EEPN). In this paper, the analysis of the one-tap normalized least-mean-square (LMS) carrier phase recovery (CPR) is carried out and the close-form expression is investigated for quadrature phase shift keying (QPSK) coherent optical fiber communication systems, in compensating both laser phase noise and equalization enhanced phase noise. Numerical simulations have also been implemented to verify the theoretical analysis. It is found that the one-tap normalized least-mean-square algorithm gives the same analytical expression for predicting CPR bit-error-rate (BER) floors as the traditional differential carrier phase recovery, when both the laser phase noise and the equalization enhanced phase noise are taken into account.

physics.optics↗

arXiv2016

Analytical Investigations on Carrier Phase Recovery in Dispersion-Unmanaged n-PSK Coherent Optical Communication Systems

Using coherent optical detection and digital signal processing, laser phase noise and equalization enhanced phase noise can be effectively mitigated using the feed-forward and feed-back carrier phase recovery approaches. In this paper, theoretical analyses of feed-back and feed-forward carrier phase recovery methods have been carried out in the long-haul high-speed n-level phase shift keying (n-PSK) optical fiber communication systems, involving a one-tap normalized least-mean-square (LMS) algorithm, a block-wise average algorithm, and a Viterbi-Viterbi algorithm. The analytical expressions for evaluating the estimated carrier phase and for predicting the bit-error-rate (BER) performance (such as the BER floors) have been presented and discussed in the n-PSK coherent optical transmission systems by considering both the laser phase noise and the equalization enhanced phase noise. The results indicate that the Viterbi-Viterbi carrier phase recovery algorithm outperforms the one-tap normalized LMS and the block-wise average algorithms for small phase noise variance (or effective phase noise variance), while the one-tap normalized LMS algorithm shows a better performance than the other two algorithms for large phase noise variance (or effective phase noise variance). In addition, the one-tap normalized LMS algorithm is more sensitive to the level of modulation formats.

cs.IT↗

arXiv2016

Analytical Estimation in Differential Optical Transmission Systems Influenced by Equalization Enhanced Phase Noise

An analytical model is presented for assessing the performance of the bit-error-rate (BER) in the differential m-level phase shift keying (m-PSK) transmission systems, where the influence of equalization enhanced phase noise (EEPN) has been considered. Theoretical analysis has been carried out in differential quadrature phase shift keying (DQPSK), differential 8-PSK (D8PSK), and differential 16-PSK (D16PSK) coherent optical transmission systems. The influence of EEPN on the BER performance, in term of signal-to-noise ratio (SNR), are investigated for different fiber dispersion, LO laser linewidths, symbol rates, and modulation formats. Our analytical model achieves a good agreement with previously reported EEPN induced BER floors, and can give an accurate prediction for the DQPSK system, and a leading-order approximation for the D8PSK and the D16PSK systems.

cs.IT↗

arXiv2016

Phase Noise Influence in Long-range Coherent Optical OFDM Systems with Delay Detection, IFFT Multiplexing and FFT Demodulation

We present a study of the influence of dispersion induced phase noise for CO-OFDM systems using FFT multiplexing/IFFT demultiplexing techniques (software based). The software based system provides a method for a rigorous evaluation of the phase noise variance caused by Common Phase Error (CPE) and Inter-Carrier Interference (ICI) including - for the first time to our knowledge - in explicit form the effect of equalization enhanced phase noise (EEPN). This, in turns, leads to an analytic BER specification. Numerical results focus on a CO-OFDM system with 10-25 GS/s QPSK channel modulation. A worst case constellation configuration is identified for the phase noise influence and the resulting BER is compared to the BER of a conventional single channel QPSK system with the same capacity as the CO-OFDM implementation. Results are evaluated as a function of transmission distance. For both types of systems, the phase noise variance increases significantly with increasing transmission distance. For a total capacity of 400 (1000) Gbit/s, the transmission distance to have the BER < 10^-2 for the worst case CO-OFDM design is less than 800 and 460 km, respectively, whereas for a single channel QPSK system it is less than 1400 and 560 km.

cs.IT↗

arXiv2016

Digital Adaptive Carrier Phase Estimation in Multi-Level Phase Shift Keying Coherent Optical Communication Systems

The analysis of adaptive carrier phase estimation is investigated in long-haul high speed n-level phase shift keying (n-PSK) optical fiber communication systems based on the one-tap normalized least-mean-square (LMS) algorithm. The close-form expressions for the estimated carrier phase and the bit-error-rate floor have been derived in the n-PSK coherent optical transmission systems. The results show that the one-tap normalized LMS algorithm performs pretty well in the carrier phase estimation, but will be less effective with the increment of modulation levels, in the compensation of both intrinsic laser phase noise and equalization enhanced phase noise.

cs.IT↗

arXiv2016

Phase Noise Influence in Coherent Optical OFDM Systems with RF Pilot Tone: Digital IFFT Multiplexing and FFT Demodulation

We present a comparative study of the influence of dispersion induced phase noise for CO-OFDM systems using Tx channel multiplexing and Rx matched filter (analogue hardware based); and FFT multiplexing/IFFT demultiplexing techniques (software based). An RF carrier pilot tone is used to mitigate the phase noise influence. From the analysis, it appears that the phase noise influence for the two OFDM implementations is very similar. The software based system provides a method for a rigorous evaluation of the phase noise variance caused by Common Phase Error (CPE) and Inter-Carrier Interference (ICI) and this, in turns, leads to a BER specification. Numerical results focus on a CO-OFDM system with 1GS/s QPSK channel modulation. Worst case BER results are evaluated and compared to the BER of a QPSK system with the same capacity as the OFDM implementation. Results are evaluated as a function of transmission distance, and for the QPSK system the influence of equalization enhanced phase noise (EEPN) is included. For both types of systems, the phase noise variance increases significantly with increasing transmission distance. An important and novel observation is that the two types of systems have very closely the same BER as a function of transmission distance for the same capacity. For the high capacity QPSK implementation, the increase in BER is due to EEPN, whereas for the OFDM approach it is due to the dispersion caused walk-off of the RF pilot tone relative to the OFDM signal channels. For a total capacity of 400 Gb/s, the transmission distance to have the BER < 10-4 is less than 277 km.

cs.IT↗