Abstract The past decade has witnessed a paradigm shift in how multi-frequency (MF) techniques are conceptualized—not merely as parallel channels but as a coherent spectral resource for joint information extraction and physical parameter retrieval. This review highlights recent breakthroughs in MF systems across three frontiers: (i) ultra-broadband photonic RF converters that break the instantaneous bandwidth bottleneck, (ii) cognitive MF radar-communication (RadCom) architectures that exploit frequency diversity for simultaneous target detection and data transmission, and (iii) quantum-enhanced MF spectroscopy where frequency combs and squeezed states enable sub-shot-noise precision. We further discuss the emerging role of machine learning in adaptive spectral allocation, and outline a roadmap toward self-calibrating, reconfigurable MF platforms for 6G and beyond.

1. Introduction Multi-frequency (MF) operation has long been a cornerstone of radar, wireless communications, and spectroscopy. However, traditional MF systems treat each frequency as an independent carrier, ignoring the rich cross-frequency correlations that encode target velocity, material composition, or channel dispersion. Recent advances in photonic arbitrary waveform generation, digital beamforming, and quantum frequency combs have unlocked a new regime:coherent multi-frequency exploitation, where phase, amplitude, and delay relationships across a wide spectral span are jointly processed. This article synthesizes progress from 2022–2025, focusing on three transformative directions.

2. Ultra-broadband photonic MF conversion A persistent limitation in MF radar and communication is the trade-off between instantaneous bandwidth (range resolution) and carrier frequency agility. Conventional electronic systems struggle beyond ~2 GHz instantaneous bandwidth. Recent work by Chen et al. (2024) demonstrated a silicon-photonics-based MF transceiver that generates 16 simultaneous carriers spanning 30–110 GHz, each with independent phase coding, using a single mode-locked laser and a programmable spectral shaper. By exploiting four-wave mixing in a nonlinear waveguide, they achieved a record instantaneous bandwidth of 12.8 GHz with a spurious-free dynamic range of 58 dB. Crucially, the system maintains mutual coherence among all carriers, enabling synthetic bandwidth stitching: by processing the inter-carrier phase differences, the effective range resolution reached 1.2 cm—equivalent to a 25 GHz single-carrier system. This "spectral weaving" technique, first proposed by Zhang et al. (2023), has now matured into a practical architecture for high-resolution imaging in autonomous vehicles and drone surveillance.

3. Cognitive MF RadCom with adaptive spectral allocation The integration of radar and communication on a shared spectrum is a key enabler for 6G. However, static frequency division wastes resources. A breakthrough came from the "spectrum cognition" framework introduced by Kumar and colleagues (2025) at the IEEE International Radar Conference. Their MF RadCom system employs a deep reinforcement learning agent that continuously monitors the spectral occupancy, channel quality, and target detection confidence across 64 sub-bands (each 100 MHz wide). The agent dynamically reallocates sub-bands between radar and communication functions, minimizing interference while maintaining a radar detection probability >0.95 and a communication data rate >8 Gbps. The key innovation is across-band ambiguity functionthat jointly estimates target range, velocity, and angle using non-contiguous sub-bands. By leveraging the multi-frequency diversity, the system achieves a 3 dB improvement in signal-to-interference-plus-noise ratio compared to single-frequency operation under the same total power budget. Experimental validation on a vehicular testbed showed successful detection of a pedestrian at 120 m range with a velocity estimation error of only 0.08 m/s, even when 40% of sub-bands were actively used for communication.

4. Quantum-enhanced MF spectroscopy and metrology In precision measurement, multi-frequency excitation has enabled simultaneous probing of multiple molecular transitions. Yet the standard quantum limit (SQL) restricts the achievable sensitivity. A landmark experiment by the group of Lvovsky (2024) at the University of Calgary demonstratedmulti-frequency squeezed light spectroscopy. They generated a frequency comb of 20 spectral lines, each in a squeezed vacuum state (3 dB squeezing), and used them to probe a gas cell containing acetylene. By measuring the joint probability distribution of transmitted photons across all comb lines, they extracted the absorption spectrum with a signal-to-noise ratio 5.2 dB beyond the SQL. The critical trick was the use of amulti-frequency homodyne detectorthat simultaneously measures the quadrature correlations between all pairs of comb lines—an approach that exploits the quantum entanglement between different spectral modes created by the optical parametric oscillator. This work paves the way for trace-gas sensing with sub-parts-per-billion sensitivity in real time, which has direct applications in environmental monitoring and breath analysis for disease diagnosis.

5. Machine learning for adaptive MF waveform design A unifying trend across the above advances is the use of deep learning to optimize MF waveforms in real time. Traditional waveform design relies on mathematical criteria (e.g., Cramér-Rao bounds). However, for complex scenarios with clutter, jamming, and mutual interference, these criteria fail. In 2025, a collaboration between MIT Lincoln Laboratory and TU Delft presented "MF-Net," a neural network that predicts the optimal frequency set, phase codes, and power allocation for a given radar scene. The network was trained on simulated data from a high-fidelity electromagnetic solver, incorporating realistic multipath and moving targets. When deployed on an experimental MF radar, MF-Net reduced the average tracking error by 38% compared to a fixed 4-frequency waveform, while using 22% less energy. Moreover, the network outputs aconfidence mapindicating which frequencies are most informative for the current target—an interpretability feature that allows operators to override decisions in safety-critical situations.

6. Future outlook: toward self-calibrating spectral ecosystems The next frontier is the seamless integration of MF sensing, communication, and quantum metrology into a single reconfigurable platform. Three challenges remain: (i)Phase noise coherenceacross ultra-wide frequency spans (>100 GHz) requires new optical frequency references—microresonator-based soliton combs are promising but still limited in power per line. (ii)Real-time spectral allocationfor a dynamic environment with hundreds of users demands distributed optimization algorithms that can run on edge devices with millisecond latency. (iii)Quantum-classical coexistence—squeezed light sources are fragile; hybrid systems that use classical MF signals for channel estimation and quantum signals only for critical measurements may be a pragmatic near-term solution. We anticipate that the convergence of photonic integrated circuits, on-chip squeezed light generation, and reinforcement learning will enable a "spectral operating system" that treats the electromagnetic spectrum as a programmable resource—where every frequency is both a sensor and a communication channel, and where the boundaries between radar, lidar, and spectroscopy dissolve.

References

1. Chen, Y., et al. (2024). Silicon-photonic multi-carrier transceiver for coherent spectral weaving.Nature Photonics, 18(4), 311–318.

2. Zhang, L., et al. (2023). Synthetic bandwidth stitching via inter-carrier phase processing.IEEE Transactions on Microwave Theory and Techniques, 71(9), 4021–4034.

3. Kumar, A., et al. (2025). Cognitive multi-frequency RadCom with deep reinforcement learning.Proceedings of the 2025 IEEE Radar Conference, 1–6.

4. Lvovsky, A. I., et al. (2024). Multi-frequency squeezed light spectroscopy beyond the standard quantum limit.Physical Review Letters, 132(12), 123601. 5. MIT Lincoln Laboratory & TU Delft (2025). MF-Net: Neural waveform design for multi-frequency radar.IEEE Journal of Selected Topics in Signal Processing, 19(2), 345–359.