High spectral purity, the property of an oscillating wave to exhibit very low phase and amplitude fluctuations, is often a highly desirable feature of an electromagnetic oscillator. For example, it enables ultra-precise atomic and molecular spectroscopy, an increased data transmission rate in wireless communication, and more precise synchronization and time-measuring capabilities for applications such as remote sensing, radar, radio-astronomy and particle accelerators.

In the quest for ever higher spectral purity oscillators that operate across the electromagnetic spectrum, attention is now turning to optical frequency combs. These combs, which produce a series of precisely spaced, narrow emission lines in frequency space, have been shown to be a tremendously useful tool. In particular, they make it possible to transfer the spectral purity of radiation from a given part of the optical spectrum (typically the visible or near-infrared, where very low-phase-noise oscillators are routinely produced) to other spectral domains such as the radio-frequency1, microwave2, mid-infrared3 and extreme ultraviolet4 regions. The result is that combs enable the realization of record-low-noise oscillators throughout the electromagnetic spectrum.

Optical frequency combs can be made in several ways but for ultra-compact applications, a relatively new comb technology based on integrated photonic chips, called dispersive Kerr solitons (DKS comb), holds big promise. The reason is that the comb-generating device itself is only a few hundred micrometres in size, compared to the several tens of centimetres of classic comb systems5. DKS combs have already been used to generate low-phase-noise signals in the microwave domain6,7 but this technology exhibits substantial new challenges as well as opportunities.

Now, writing in this issue of Nature Photonics, Tetsumoto et al. report the use of a photonic-chip-generated optical frequency comb to transfer the spectral purity of an oscillator at 3.6 THz to a 300 GHz carrier in the millimetre-wave domain, resulting in a record low-phase-noise signal at this frequency8. This achievement is a major milestone towards producing ultra-compact, ultra-low-noise, continuous-wave sources in this spectral domain for scientific, technological, civilian and military applications alike.

Although Tetsumoto et al. start with an ‘oscillator’ at 3.6 THz, it must be understood that this is a ‘virtual’ 3.6 THz signal (no actual pure THz wave is generated in this experiment): it is in fact the frequency difference between two closely spaced infrared lasers operating at 192.4 and 196 THz respectively (Fig. 1). These two near-infrared lasers, however, exhibit low fluctuations of their relative phase, because they are both generated in the same stabilized fibre-ring resonator and thus any optical length fluctuations are common to both lasers.

Fig. 1: Simplified schematic of the experiment presented by Tetsumoto and colleagues.
figure 1

The fibre spool resonator, appropriately tuned and pumped, provides two near-infrared beams whose frequencies (ν) are separated by 3.6 GHz (top panel). The silicon nitride resonator, pumped appropriately, provides a DKS comb with 300 GHz reference that generates a millimetre wave continuous wave source after photodetection in a uni-travelling-carrier photodiode (UTC-PD) (bottom panel). By combining and mixing of the beatnote signals between the reference radiations and the DKS comb, an error signal is generated and used (via a proportional integrator derivative (PID) controller) to lock the DKS comb repetition rate by amplitude control of its laser pump. The result is a very low-noise 300 GHz signal source. The amplitude control of the DKS comb pump is realized with a single side band modulator device that simultaneously allows control of the amplitude and frequency of the pump.

Full size image

Using these two 3.6 THz-separated lasers as a reference, the authors conveniently realize the servo-locking of a 300 GHz repetition rate DKS comb. Appropriate photodetection of the beatnote signals between each of these two lasers and the nearest optical modes of the DKS comb allows, after nonlinear mixing, generation of an error signal suitable to lock the DKS comb. Once the servo-loop is engaged, the repetition rate of the DKS comb follows the 3.6 THz virtual reference, but divided down to 300 GHz.

This frequency division process is traditionally the method by which optical frequency combs allow the realization of very low-phase-noise sources, as the phase noise of the reference is divided down in the process by as much as the carrier frequency. Although the best possible result is likely to be obtained when dividing an ultra-stable reference in the near-infrared domain, the method followed by Tetsumoto et al. is attractive as it offers a substantial simplification of the apparatus. For example, dealing with the carrier envelop offset frequency of the comb is not necessary. The record low phase noise eventually realized at 300 GHz — already challenging to achieve and characterize — shows clearly that the cost in terms of ultimate performance induced by this approach is largely offset by the increased simplicity of the experimental demonstration.

Once the phase-locking of the DKS comb is realized, the 300 GHz signal is still virtual, in that it is encoded in the repetition rate of the optical pulses from the comb. It is then necessary to generate from the virtual signal a ‘real’ 300 GHz electromagnetic wave and, most importantly, characterize its phase noise. The use of a special, very high bandwidth, uni-travelling-carrier photodiode (UTC-PD) makes the first part possible, as it allows generation of the 300 GHz wave by direct photodetection of the pulse train emitted by the DKS comb.

The phase noise characterization at the required level is particularly challenging in this frequency range and Tetsumoto et al. combine three different techniques to make it possible. This approach allows different parts of the Fourier frequencies spectrum to be targeted with the most appropriate of the three techniques (that is, the one that exhibits the lowest noise floor readout). It also increases the general trust in the measurement, since, as expected, the different techniques give the same results at the few places in the Fourier spectrum where they are not limited by their readout noise.

These three techniques rely on comparing the DKS-comb-generated 300 GHz signal with a signal from a classic optoelectronic source, realized by multiplying the frequency of a low noise microwave source at 10 GHz (a dielectric resonant oscillator (DRO)) to 300 GHz by high-modulation-index optical phase modulation of a mid-infrared optical carrier. Depending on the part of the Fourier spectrum of interest, this classic source can be free-running, and used directly as a low-noise reference (for high Fourier frequencies) or phase-locked to the 300 GHz, so that the 10 GHz DRO signal itself copies the spectral purity of the 300 GHz source (but divided by 30), and can be used for analysing its phase noise with conventional instrumentation (for low Fourier frequencies). However, in the intermediate frequency range (8–50 kHz), the authors had to use a more exotic approach, where the DRO signal is analysed by a low-noise homemade two-wave delay-line interferometer, which is, roughly speaking, a special and highly optimized implementation of the self-interferometry technique with a long delay line.

The final phase noise estimation, realized by stitching the results from the three measurement methods, exhibits a record low phase (or equivalently timing) noise for a 300 GHz source, with a value substantially below 10 as Hz–1/2 from Fourier frequencies above 10 kHz from the carrier.

This achievement is an important milestone as it demonstrates the usefulness of the ultra-compact DKS comb approach for the generation of low-phase-noise signals at the higher end of the millimetre-wave band (close to the terahertz band). In this region, classic optical frequency comb systems with a repetition rate typically in the radio-frequency domain are difficult or impossible to use, even with the help of pulse–interleaver-based frequency multiplication systems9.

Furthermore, the integrated-photonic nature of the DKS comb holds promises for the ultimate goal of development of a fully integrated, compact source of a low-phase-noise signal in the millimetre range. Of course, there is still a long way to go between the work of Tetsumoto et al. and a fully integrated-photonic turn-key system that is easily field-deployable in applications such as a future communications system, for example. Beyond the DKS comb itself, many other components will have to be integrated as well. However, this latest achievement is certainly paving the way towards such a bright future where low-noise sources throughout the millimetre range could be available in a compact and easy-to-use manner.