1. Introduction

Microwave photonic technology (MPT) has attracted intensive interests in recent years, which is highly desirable in many important applications such as satellite communications [1], surveillance [2], radar [3–5], astronomy [6] and so on. In the most of the applications, photonic microwave signal generation is a key requirement. Therefore, it has become a widely investigated area.

In the past few decades, various photonic methods for microwave signal generation have been proposed and demonstrated [7–10]. Among these approaches, photonic microwave oscillator has drawn great attentions because it can generate microwave signal with extremely high spectral purity [11–13]. Generally, the configuration of a photonic microwave oscillator is a positive feedback loop with a frequency selection module. The typical examples are optoelectronic oscillators (OEOs) [14,15] or coupled optoelectronic oscillators (COEOs) [16–18], in which a long fiber loop or active fiber loop serves as a high-Q resonator. Since they are optoelectronic hybrid systems, the oscillation signal has to be converted from optical domain to electrical domain by a photodetector (PD) within the loop. Then the oscillation frequency is selected by an electrical bandpass filter (EBPF), and fed back to optical domain again by an electro-optic modulator. One fact is that the involved electrical microwave components increase the cost of the system and limit the operation bandwidth, especially working in high-frequency range. In addition, EBPFs or modulators also make the system bulky and non-integrated. In order to take more advantages of the photonic technology and avoid the use of electrical microwave devices, many significant works have been reported. For instance, the function of EBPF can be implemented by a Fabry-Perot (F-P) filter [19], but a stable light source has to be employed to align one of the filter windows and the tunability is not available due to the fixed free spectral range (FSR) of F-P filter. Brillouin-selective sideband amplification (BSSA) [20–23] is also a good choice for oscillation frequency selecting in optical domain. However, the stability of the oscillation frequency is determined by the relative stability between signal light and pump light, which means that high performance light sources have to be employed. Since optical injection locking mechanism in semiconductor laser diode presents a narrow gain spectrum with large rejection ratio, it provides a microwave frequency selection method in photonic way and is applied in OEOs. In this case, the oscillation frequency can be tuned by varying the wavelength difference between master laser and slave laser with a tuning range from 6.41 GHz to 10.85 GHz [24]. However, as the same difficulty as the case of BSSA, a stable oscillation is still relied on the wavelength stability between two laser diodes. Although these photonic microwave frequency selection based on OEOs or COEOs are successfully demonstrated, they are essentially optoelectronic hybrid systems due to the indispensable links of photoelectric conversion and electro-optic modulation.

In order to completely eliminate the use of electrical microwave components, an all-optical microwave oscillator has been reported [25]. In this system, the oscillation frequency is selected by BSSA, and the envelope detection and all-optical feedback modulation are implemented by a semiconductor optical amplifier (SOA). This all-optical microwave oscillator extends the concept of photonic microwave signal generation, but there are still some drawbacks. Firstly, a single-mode microwave signal is not realized due to the limited capability of mode selection by two coupled loops [26–28]. For single-mode oscillation, an assisted active ring resonance cavity should be introduced [29]. Secondly, under the condition of self-pumped stimulated Brillouin scattering (SBS) in optical fiber, the frequency tuning flexibility is almost lost [20,21,30]. Although this problem can be solved by using an independent pump light, the high stability requirement on both signal light and pump light cannot be easily reached in practice. Compared with the schemes mentioned above, if both of the microwave frequency selection and feedback modulation are implemented in optical domain, an all-optical microwave oscillator is able to be obtained. For practicability, single-mode oscillation, larger tuning range and lower performance requirement on the involved light sources are expected.

In this paper, we propose and experimentally demonstrate a tunable all-optical coupled microwave oscillator without optical-electrical-optical conversion or any electrical microwave components. In this system, a fiber ring laser (FRL) serves as a multi-longitudinal-mode light source with specific cavity mode spacing. After transferring the characteristics of cavity modes on the center wavelength of the master laser by optical-optical modulation in a SOA, the light field with copied cavity modes are injected into a distributed feedback laser diode (DFB-LD), which acts as a slave laser. Once one of the cavity modes falls into the locking range of the DFB-LD, the cavity mode can be selected and amplified, which contributes a certain microwave frequency by beating with the optical carrier from the master laser. After the optical-optical feedback modulation by another SOA within the fiber ring laser, the beating signal is introduced into the FRL again. Thus the microwave oscillation loop is closed and a photonic microwave signal is generated consequently. Because of the mode pulling effect [31] in the injection locking process, the oscillation frequency instability caused by wavelength drift on both master laser and slave laser is greatly suppressed. The oscillation frequency can be easily tuned by simply adjusting the center wavelength of the master laser. In the experiment, good quality single-mode microwave signals with frequency tuning range from 6.93 to 25.54 GHz are demonstrated. The measured oscillation frequency drift within one hour is 78 kHz at the operation frequency of 14.23 GHz, and no mode hopping happens. The phase noise of the generated microwave signals are also investigated.

2. Operation principle

The schematic diagram of the proposed all-optical coupled microwave oscillator is shown in Fig. 1. It mainly consists of two parts: a FRL and a frequency selection module. When the FRL works alone, a multi-longitudinal-mode optical field with a center wavelength of ${\lambda _1}$ can be excited, and the mode spacing is determined by the ring cavity length. It is well known that the FSR of an optical resonator can be expressed as $FSR = c/({\textrm{n}_\textrm{g}} \cdot L)$, where L is the cavity length and $c/{n_g}$ is the speed of light in the cavity. Therefore, for fixed c and ${n_g}$, the FSR is related to the cavity length. Since a fiber loop usually has long cavity length, a small FSR is expected. According to the Vernier effect [26,27], when a short fiber ring loop formed by a 3-dB optical coupler (OC) is coupled with the main cavity, the effective FSR can be expressed as $f = pFS{R_1} = qFS{R_2}$, where p and q are integers and $FS{R_1}$ and $FS{R_2}$ are the FSR for each cavity. Obviously, the effective FSR is enlarged by coupling multiple cavities. In general, the optical filter in the fiber ring laser has much larger bandwidth than the value of f, which means that the fiber ring laser outputs an optical field with large number of longitudinal modes and the mode hopping easily happens around the center wavelength. Therefore, the fiber ring laser has a jumped dominant mode and cannot be directly used as a stable optical source. But even so, it may act two important roles in our system. At first, because the mode spacing provides a certain microwave frequency scale, it is helpful for stable microwave frequency selection in optical domain. Secondly, as an active cavity, the configuration of the fiber ring laser can be regarded as a high-Q optical resonator. It is the key requirement for photonic microwave oscillator. In Figs. 2(a) and 2(b), the corresponding mode characteristics of the fiber ring laser with the cases of one and two cavities are presented.

 

Fig. 1. The configuration of the proposed all-optical coupled microwave oscillator. ODL: optical delay line, OBPF: optical bandpass filter, EDFA: erbium-doped fiber amplifier, OC: optical coupler, PC: polarization controller, ISO: isolator, SOA: semiconductor optical amplifier, Cir: circulator, ECL: external cavity laser, ATT: attenuator, DFB-LD: distributed-feedback laser diode, SMF: single-mode fiber, PD: photodetector, OSA: optical spectrum analyzer, ESA: electrical spectrum analyzer, Osc: oscilloscope.

Download Full Size | PDF

 

Fig. 2. The mode characteristics of FRL and the principle of frequency selection process. (a) The cavity modes of single-loop FRL. (b) The cavity modes of dual-loop FRL. (c) After transferring the cavity modes on the center wavelength of ECL. (d) Selective amplification of one cavity mode by optical injection locking.

Download Full Size | PDF

In the frequency selection module, the master laser, which is a tunable external cavity laser (ECL), emits a single-mode optical field with a wavelength of ${\lambda _2}$. When it is launched into SOA2 in the opposite direction to the FRL, the peaks of the envelope of beat signal among cavity modes consume almost of the carriers in SOA2. Consequently, the optical gain for continuous wave (${\lambda _2}$) is relatively weak in the corresponding time slot. On the contrary, the hollows of the envelope may not consume the carriers in SOA2. As a result, the optical gain for continuous wave is relatively large. According to this process, the envelope of beat signal leads to the carrier density variation in SOA2, and consequently the injected continuous wave suffers optical-optical modulation called cross-gain modulation (XGM). In other words, the optical-optical modulation is realized by the gain saturation mechanism in SOA2. The result is that the cavity modes from the fiber ring laser can be transferred on the center wavelength of the master laser. This procedure is able to ignore the mode hopping on the output of the FRL, and the frequency scale ($f$) from the FRL is inherited by the master laser. The result is schematically shown by Fig. 2(c). In the following injection locking process, there is a DFB-LD (slave laser) with a wavelength of ${\lambda _3} \pm \frac{1}{2}\Delta f$, where the $\Delta f$ denotes the wavelength drift of the DFB-LD. Suppose $\Delta f < f$ and ${\lambda _3}\textrm{ - }{\lambda _2}$ roughly equals to $nf$, where n is an integer, only one transferred cavity mode can fall into the locking range, and a frequency of $nf$ is presented by the wavelength difference between ${\lambda _2}$ and ${\lambda _3}$. It is worth emphasizing that the injection locking process exhibits mode pulling effect, which means that the injection locking can be always occurred if one transferred cavity mode falls into the locking range of the DFB-LD. In other words, a small wavelength detuning between injecting mode and ${\lambda _3}$ is permitted. Mathematically, let us suppose all the wavelength instability comes from the DFB-LD, and the total relative wavelength instability can be written as $\Delta f$. Then, the tolerance for frequency selection is expressed as $\Delta f < f$, which is shown by Fig. 2(d). Therefore, the critical stability requirement on master laser and slave laser is relaxed.

Although the beating signal between ${\lambda _2}$ and ${\lambda _3}$ may contribute a microwave signal with frequency of $nf$, the signal quality is not satisfactory, because of no phase relationship between ${\lambda _2}$ and ${\lambda _3}$ . In order to build up the phase relationship, the beating signal is fed back to the FRL through SOA1. Thanks to the carrier dynamic properties of SOA1, the envelope of beating signal between ${\lambda _2}$ and ${\lambda _3}$ leads to carrier density variation in SOA1 and the feedback modulation is implemented. As a result, a closed microwave resonator is formed, in which a photonic microwave signal with frequency of $nf$ can be selectively enhanced and consequently generated. By adjusting the wavelength of the master laser, the oscillation frequency can be coarsely tuned with a step of f. Fine tuning can be implemented by slightly changing the cavity length of the FRL.

3. Experiments and results

To verify the feasibility of the proposed scheme, the experiments based on the setup shown in Fig. 1 are carried out. Firstly, we investigate the characteristics of the single-loop FRL by disconnecting point A, B and E. In this case, a tunable optical delay line (ODL), an optical bandpass filter (OBPF), an erbium-doped fiber amplifier (EDFA), a SOA1 (Thorlabs Quantum Electronics, SOA-17415-11450.29.C01), a polarization controller (PC1), a circulator (Cir1) and two optical isolators (ISOs) are linked in sequence to form a loop cavity. Here, The ODL is used to adjust the cavity length. The OBPF with a bandwidth of 0.5nm is centered at 1548.4 nm, which decides the lasing wavelength of the FRL. There are two ISOs placed in the cavity to ensure unidirectional operation and the polarization state of the optical field is controlled by PC1. Although both of the EDFA and SOA1 can provide the required optical gain for the FRL, the SOA1 is more importantly utilized as optical-optical modulation device for the followed feedback process. Once the oscillation threshold is reached, an excited optical field with numerous longitudinal modes is circled in the optical resonator and output from a 3-dB optical coupler (OC2). To observe the mode characteristics, the beating signals among longitudinal modes are converted into electrical domain at point B by a photodetector with a bandwidth of 50 GHz, which are analyzed by an electrical spectrum analyzer (ESA, Agilent N9010A).

As shown by Fig. 3(a), the beating modes present a set of stable sharp peaks with an interval of 3.2 MHz, which indicates the optical resonator has a high-Q factor and a FSR of 3.2 MHz. Considering the bandwidth of the OBPF, mode hopping among different longitudinal modes may frequently happens. In order to match the tolerance of the following frequency selection process, a 0.59-m fiber ring loop is inserted into the main cavity by connecting point A of the 3-dB optical coupler (OC1) to increase the FSR. It can be seen from Fig. 3(b) that the dominant beating frequency is shifted to 364.8 MHz, 114 times of the former case, due to the Vernier effect. The inset gives a zoom in view of the first peak. One can find that the residual modes are effectively suppressed by a ratio of 27.8 dB. It needs to be pointed out that although larger mode spacing reduces the number of cavity modes in FRL, the mode hopping of the FRL may not be vanished in this case. Therefore, it is necessary to transfer the mode characteristics from the FRL to an ECL.

 

Fig. 3. The FSR measurement of the FRL. (a) Single-loop FRL. (b) Dual-loop FRL. Insert: a zoom in view of the dominant beating frequency.

Download Full Size | PDF

Before starting the microwave frequency selection, the relative wavelength drift between ECL and DFB-LD is tested through an optical heterodyning system as shown in Fig. 4(a). The tunable ECL outputs continuous wave (CW) with a linewidth of 100 kHz, whose wavelength is set at 1550.30 nm. The DFB-LD with threshold current of 17 mA is biased at 100 mA, and the temperature is maintained at 25°C by a thermoelectric cooler (TEC). Under the free running condition, it emits a CW with a linewidth of 2 MHz and a wavelength of 1550.41 nm, corresponding to 14.23-GHz frequency detuning from the ECL. The output optical power of the DFB-LD is 10.96 dBm. When these two optical fields are frequency mixing in a PD, a random frequency-hopping beat signal can be observed by the ESA. Using the “Max Hold” function of the ESA, the frequencies of the beat signal are continuously recorded in one hour and the variation trace is given by Fig. 4(b). It is clear that the relative wavelength drift range is 248 MHz (corresponding to the wavelength drift of the DFB-LD: $\Delta f$). This value is too large to be accepted if the beat frequency serves as a reference of a microwave oscillator. Fortunately, it is smaller than the FSR of the dual-loop FRL. Therefore, if the cavity modes carried by ECL are injected into the DFB-LD, one and only one cavity mode can fall into the injection locking range. Combining with the mode pulling effect, the beating frequency between master laser and slave laser is fixed at $nf$.

 

Fig. 4. The schematic diagram and experimental results of the optical heterodyning test. (a) Schematic diagram of the optical heterodyning system. ECL: external cavity laser, DFB-LD: distributed-feedback laser diode, PC: polarization controller, OC: optical coupler, PD: photodetector, ESA: electrical spectrum analyzer. (b) The measurement of Long-term wavelength drift within 1 hour.

Download Full Size | PDF

By connecting point B, the output from the FRL with an optical power of 0.2 dBm is launched into the frequency selection module, in which the mode transfer is implemented at first. In this case, the ECL is tuned at 1550.36 nm with an output optical power of 1.3 dBm, which propagates in the opposite direction to the multi-longitudinal-mode optical field from the FRL. The PC2 and PC3 are used to align the polarization state of the two light fields, and the ISO3 prevents the light field from FRL injecting into the ECL. When two optical fields encounter in SOA2, the mode characteristics from FRL is transferred on the CW from ECL due to the optical-optical modulation effect. Through the port 3 of the Cir 2, the master laser is ready for injection locking process, and the corresponding electrical spectrum at point C is shown by Fig. 5. The same as the original case, a set of beating modes with frequency spacing of 364.8 MHz can be clearly observed. It indicates that the precise frequency scale from the FRL is successfully transferred, and the mode hopping is excluded.

 

Fig. 5. The measured RF spectrum after transferring the cavity modes on the center wavelength of the ECL.

Download Full Size | PDF

After controlling the power and polarization state of the master laser by a tunable attenuator (ATT) and PC4, the optical field is injected into the DFB-LD through the port 2 of the Cir3. Here, the DFB-LD keeps the same operation condition as the optical heterodyning test, and the injecting power of the master laser is 1.2 dBm. At point D, the optical spectra are monitored by an optical spectrum analyzer (OSA) (YOKOGAWA AQ6370C), as shown by Fig. 6(a). In this figure, the red dash line and black dot line present the optical spectra of ECL and free-running DFB-LD. The optical spectrum after injection locking is plotted by the blue solid line, where the lasing wavelength of DFB-LD is red-shifted from 1550.41 nm to 1550.47 nm and the spectrum is broadened by the four wave mixing (FWM) effect in DFB-LD [32,33]. In this case, the spectral lines have an interval of 0.11 nm. From the corresponding beat signal given by Fig. 6(b), a noise-like envelope centered at 14.23 GHz is produced by directly beating between master laser and free-running slave laser, because the transferred cavity modes could not cover such large frequency span and consequently no cavity mode locates at the locking range of the DFB-LD. Therefore, the feedback modulation is particularly necessary to excite a mode within the locking range and establish the phase relationship among cavity modes.

 

Fig. 6. The measured optical spectra and the corresponding RF spectrum without feedback modulation. (a) The optical spectra before and after injection locking. (b) The corresponding RF spectrum without feedback modulation.

Download Full Size | PDF

To close the whole microwave oscillator, the point E is connected and the beat signal is fed back to FRL through port 1 of the Cir1. In this case, the envelope of beat signal is responded by SOA1 and the cavity modes in FRL are modulated. The 1-km optical fiber is inserted into the feedback branch due to longer feedback loop leading higher Q factor. In this case, although there are three resonators involved in the microwave signal oscillation, the effective FSR is still determined by the fiber ring laser. In terms of system parameters, the cavity mode spacing of the fiber ring laser is 364.8 MHz. The cavity length of the third resonator is approximately 1 km, which has the cavity mode spacing of 200 kHz. This means that nearly 1867 cavity modes are distributed between two adjacent cavity modes of the fiber ring laser. Therefore, the cavity modes of the long cavity can easily match the cavity modes of the fiber ring laser. The effective FSR can be still expressed as $f = pFS{R_1} = qFS{R_2}$. By fine tuning the ODL in the FRL, a photonic microwave signal is obtained immediately, which is measured at point D and shown by Figs. 7(a)–7(d). It is not shown here, the measured optical spectrum with and without feedback modulation has no visible change. From Fig. 7(a), only one microwave signal with a frequency of 14.23 GHz appeared in the full measurement range (0–26.5 GHz) of the ESA, and the side-modes are suppressed by the gain competition in the resonator. The close-up view with a SPAN of 1 GHz is shown by Fig. 7(b), where no side-modes at frequency offset of 364.8 MHz from the carrier can be found. Keeping the same observation window, the inset gives the frequency variation recorded by using “Max Hold” function for one hour. It indicates no mode hopping occurs. In addition, the detailed spectrum and waveform are shown by Fig. 7(c). There is no 200-kHz side-modes noise around the generated 14.36-GHz microwave signal. It can be concluded that the FRL acts as a high-Q optical resonator. Once the feedback loop is closed and the oscillation is established, the high-Q optical resonator can strongly suppress the non-public cavity modes in each resonator. Both of the results prove that a stable and good quality microwave signal is successfully generated. To further illustrate that the mode pulling effect is able to effectively overcome the relative wavelength drift between master laser and slave laser, a long-term frequency stability test in a frequency window of 500 kHz is taken, which is shown by Fig. 7(d). One can read that the frequency drift range is only 76 kHz within one hour. Compared with the free-running case in Fig. 4(b), the frequency stability is greatly improved.

 

Fig. 7. The measured RF spectra, waveform and frequency stability of the generated microwave signal. (a) The overall view of the RF spectrum. (b) The RF spectrum of the generated 14.23-GHz microwave signal with a SPAN of 1 GHz. Inset: the corresponding long-term frequency drift within 1-hour measurement. (c) Detailed RF spectrum. Inset: the corresponding waveform. (d) Long-term frequency drift within 1-hour measurement with a span of 500 kHz.

Download Full Size | PDF

Furthermore, the proposed all-optical microwave oscillator exhibits a large frequency tunability by simply changing the center wavelength of the ECL. Figure 8 shows the generated microwave signals with frequency tuning range from 6.93 GHz to 25.54 GHz. The two spectral lines around 7 GHz are 6.93 GHz and 7.29 GHz, respectively, which show the frequency difference of 364.8 MHz. The lower frequency bound on the microwave tuning range is determined by the mode pulling effect. Because the optical field from ECL is much stronger than that of the transferred cavity modes, when the wavelength of the master laser is close to the free-running mode of DFB-LD, strong optical injection makes the mode of DFB-LD shifted and coincided with the master laser. As a result, the transferred cavity mode can not be selectively amplified, which leads to the frequency selection mechanism failure. In principle, the frequency can be even higher, but it is already closed to the measurement maximum limit of the ESA. Because the FSR of the fiber loop cavity can be regarded as the frequency reference, the frequencies of generated microwave signals must be an integer multiples of the cavity mode spacing. On the one hand, it makes the oscillator with a tuning step of 364.8 MHz, on the other hand, it relaxes the restriction on wavelength tuning precision and stability since the mode pulling effect can force the DFB-LD automatically tracking the cavity mode within its locking range. Meanwhile, by keeping the master laser and slave laser unchanged, fine tuning can be implemented by slightly adjusting the optical delay line in the fiber ring laser. Without mode hopping, the measured fine tuning range is approximately 62 kHz in the experiment.

 

Fig. 8. Tunability of the proposed system with tuning range from 6.93 to 25.54 GHz.

Download Full Size | PDF

Finally, the single-sideband (SSB) phase noise performances of the generated microwave signals are investigated. Figure 9 shows the phase noise curves under different operation frequencies. It can be found that the phase noise performances are very closed in all cases, and the values are approximately −95 dBc /Hz @ 10 kHz offset from the carriers. This result reveals the phase noise performance is independent on operation frequency, similar to OEO systems. It needs to be noticed that the phase noise performance near the carrier is not satisfactory. One reason is that the amplified spontaneous emission (ASE) of SOA and the relative intensity noise (RIN) of DFB-LD may be introduced into the generated microwave signals. The other one is that the injection locking process is a polarization sensitive process, which is easily affected by the environmental fluctuations. Therefore, the slight timing jitter or frequency drift caused by the all-fiber structure can be converted into the phase noise. Even so, these problems should be solved by developing waveguide structure, keeping the system in isolated environment without mechanical jitter and temperature fluctuation, or employing polarization maintaining optical fibers.

 

Fig. 9. SSB phase noise performances at different microwave frequencies.

Download Full Size | PDF

From all the experimental results presented above, the proposed all-optical coupled microwave oscillator can generate a stable single-mode microwave signal with large tuning range. The prominent significance of the scheme is that the whole system has no photoelectric conversion or electro-optic modulation. In this scheme, the FRL acts as a high-Q active resonator and provides a set of frequency benchmarks for the oscillator. For the frequency selection mechanism, the cavity modes carried by the master laser can be directly selected and amplified via injection locking process. More importantly, the lasing mode of the slave laser has to automatically follow the injected cavity mode within locking range due to mode pulling effect. This process can effectively eliminates the affects from the wavelength instability and realizes a self-adaptation oscillation frequency locking. Such advantages cannot be obtained by using BSSA or high-Q F-P filter.

4. Summary

In conclusion, an all-optical coupled microwave oscillator with good performance has been proposed and experimentally demonstrated. In the system, the FRL not only provides stable and specific frequency references but also acts as an active resonator. The oscillation frequency selection is achieved by mode transfer and injection locking process. After optical envelope detection and feedback modulation by a SOA, a good quality and high stability single-mode microwave signal is successfully generated by a pure optical oscillation. Thanks to the mode pulling effect in the injection locking process, the frequency instability caused by wavelength drift on both master laser and slave laser can be effectively eliminated. By simply adjusting the wavelength of ECL, a frequency tuning range from 6.93 GHz to 25.54 GHz is realized. The whole system has no any electrical microwave components as well as photoelectric conversion, which can be applied in future all-optical information systems.