1. Introduction

Since the lightwave synthesized frequency sweeper (LSFS) was first proposed in 1990 [1],[2], its potential as a reference source with linear scanning has been recognized, and several variations of the configuration have since then emerged. Fixed filters were used in early systems to suppress spontaneous emission noise. This limited scanning ranges to around 1 nm (in the 1.5 µm range) [3]. Since then, tunable filters have been incorporated in these systems, and scanning ranges of up to 10 nm have been reported [4]. However, all LSFSs reported in literature known to the authors are based on Erbium doped fiber amplifiers (EDFAs) operating at 1.55 µm.

The results in the present paper focuses on biomedical applications, and we are interested in frequency scanning sources for investigation of retinal features in the eye. For this particular application, it is beneficial to use shorter wavelengths compared those traditionally offered by telecom components, as retinal imaging requires propagation through the vitreous humor. The vitreous humor has a high water content, where absorption and dispersion at 1.55 µm becomes significant. Wavelengths in the visible range are less attractive, since scattering losses in the retina becomes significant. However, a window in the 1–1.1 µm range offers low absorption and dispersion in water [5], [6].

Ytterbium doped fibre amplifiers (YDFAs) can provide gain throughout this wavelength range [7, 8, 9], and it seems obvious to use an YDFA in the LSFS. However, the scanning range of the LSFS is known to be limited by the amplified spontaneous emission noise that builds up in the ring during scanning [3, 10]. When the LSFS is based on EDFA’s, the amplifier noise figure can approach the quantum noise limit. YDFA’s are not able to reach quantum limited noise figures, when pumped in the efficient 975 nm band [11], which could render YDFAs unsuited for this configuration.

The purpose of the investigation presented here is, therefore, to determine, whether it is feasible to build an YDFA-based LSFS, and if so, to validate whether numerical models can predict the behavior of the system. We present here for the first time experimental evidence that YDFA-based LSFS sources are operational. Furthermore, when using traditional timing schemes, we experience unwanted Q-switching behavior. However, we are able to suppress this behavior using a novel timing scheme.

In section 2 of this paper, a short introduction to the general LSFS configuration is given. The setup used for the experiments is described in section 3. In section 4, we present our new timing scheme. Results from the experiments are described in section 5 and compared with numerical modelling in section 6. The results are summarized in section 7.

2. Operation of the LSFS

A principal sketch of the LSFS configuration is shown in Fig.1(a). It consists of an optical ring containing a beamsplitter, an optical amplifier, a frequency shifting element (Δf) and a band-pass filter (BPF). The ring is initially seeded with a pulse from a seed laser. The duration of the seed pulse is comparable to the optical round-trip time in the ring. A fraction of the seed pulse is sent directly to the detector and the remaining part is sent into the ring. In a single round-trip, the pulse is amplified, frequency shifted and filtered. Upon its return to the beamsplitter, a fraction is once again sent to the detector and the remaining part sent on another round-trip in the ring. Losses experienced by the pulse upon a round-trip is compensated by the optical amplifier. The output consists of a series of pulses. Each pulse is a replica of the initial pulse, except that each pulse is frequency shifted Δf relative to its predecessor. Since this is a feedback system, the pulse power at the output of the amplifier will gradually approach the saturated output power, much like in a traditional laser. Initially one might assume, that the pulse circulation would continue until the optical frequency of the pulse was outside the gain spectrum of the amplifier. This is not the case in practice. When a signal is amplified, the amplifier also adds noise to the signal. The generated amplified spontaneous emission (ASE) noise will also circulate in the ring, and more noise is added upon each circulation. The accumulating noise will gradually use more and more of the available saturated amplifier gain until the signal is eventually lost, and a new scanning cycle must be initiated. Our application requires rapid consecutive frequency scans, and a new scanning cycle is initiated immediately after termination of the previous cycle, as shown in Fig. 1(b).

 

Fig. 1. Principal sketch of the LSFS configuration (a), and the associated ideal output signal power and optical frequency (b).

Download Full Size | PDF

By introducing a bandpass filter in the ring, the effective noise bandwidth is reduced, and noise accumulation is, therefore, also reduced. However, in selecting a suitable filter bandwidth one is faced with a trade off: A narrow filter bandwidth reduces noise accumulation, but also the effective range over which the signal can be scanned [4]. By using a tunable filter, this trade off can be overcome - a narrow bandwidth filter then need to have its center frequency follow the optical frequency of the circulating signal pulse over time. Such a system based on an EDFA has been demonstrated with tuning ranges up to 10 nm [4].

3. Experimental setup

The experimental setup is illustrated in Fig. 2. The system was seeded with a single mode semiconductor seed laser (100 mW, 1060 nm). The seed laser was operated in CW mode, and the seed pulse was generated using a 175 MHz acousto-optic modulator capable of rise times down to 20 ns. The YDFA was 8 meters long and had a double-clad structure. The inner cladding diameter was 210 µm and flower shaped. The amplifier was pumped by a 2 W broad area semiconductor laser with center wavelength at 975 nm. A reverse pumping scheme was deployed, and dichroic mirrors combined/separated pump and signal paths. A 200 MHz AOM was simultaneously used as frequency shifter and switch in the ring. The +1 order diffracted beam from the AOM was used in the experiment, i.e., the wavelength was shifted down. The band pass filter (BPF) used in the ring had bandpass center at 1080 nm, and a 3 dB bandwidth of 10 nm. The center wavelength was shifted down to the signal wavelength by adjusting the filter angle. The dichroic mirrors were sensitive to polarization, and fiber based polarization compensators were positioned in the signal path just prior to the mirrors. The round-trip loss in the ring was estimated experimentally to be 13–14dB.

 

Fig. 2. LSFS used for the experiments.

Download Full Size | PDF

The rate at which the frequency is swept is determined by the frequency shift per circulation and the length of the ring. We are ultimately interested in rapid frequency shifts and want the ring to be as short as possible. However, it should be possible to generate suitable seed pulses within the rise-time of the seed AOM. We, therefore, chose a delay fiber consisting of 100 meters single mode fiber, which gave a total round-trip time in the ring of 540 ns. Compared to previous reported experiments with LSFS sources, this is a relative short ring length. YDFAs are extremely sensitive to reflections, and end facets of the amplifier and delay fiber were, therefore, polished at an angle of 12°. Timing of the source, i.e., switching the AOMs, was done with a timing generator.

4. Novel Timing Scheme

A new scanning cycle has to be initiated, when the signal is lost in the noise. Prior to initiating a new cycle the accumulated ASE noise must be terminated. This can be achieved by switching off the AOM in the ring, causing the ring transmission to drop to zero. In a previously reported timing scheme [3], the ring was reset by turning the ring AOM off between scanning cycles. However, when re-initiating a scanning cycle, the ring AOM was turned on just prior to injecting a seed pulse, as shown in the timing scheme in Fig. 3(b). Using this timing scheme in our experiment caused the ring to exhibit Q-switching behavior, when the ring AOM was turned on. The spike caused depletion of the amplifier inversion and made it cumbersome to maintain a circulating pulse in the ring afterwards. Furthermore, the spike was a risk for optical components in the system. Our novel timing scheme, developed to overcome this problem, is shown in Fig 3(a).

 

Fig. 3. Timing diagram for the novel Q-switch suppressing timing scheme (a) and the traditional timing scheme (b).

Download Full Size | PDF

When a scanning cycle was completed, the ring AOM was turned off, hence terminating the accumulated noise. A new scanning cycle was initiated by turning the seed AOM on while the ring AOM was off. This results in a pre-seeding of the amplifier. After the pre-seeding period the ring AOM was turned on, and after an additional delay, the seed AOM was turned off. The pre-seed duration was in the experiments chosen to be 2 ms, well after the signal has been lost in the noise. The time between turning the ring AOM on and turning the seed AOM off determined the duration of the pulse circulating in the ring. The times shown in Fig. 3(a) are related through:

where τ_pre-seed , τ_pulse and τ_{SR AOM} is the pre-seed delay, duration of the circulating pulse and the propagation time from the seed AOM to the ring AOM, respectively. τ_{SR AOM} has to be subtracted in t ₂, in order to compensate for the delay between the seed AOM and the ring AOM.

The advantage of the pre-seeding scheme is gain clamping of the amplifier; it is brought to the inversion level it will stay at, while a pulse is scanned in the ring. Hence, when turning the ring AOM on, no Q-switching behavior is seen. The Q-switching behavior has not been described in previous reports on the LSFS system. We believe that the comparatively short ring length used in our experiments is the reason why this behavior is so pronounced. However, increasing the ring length would decrease the scanning speed.

 

Fig. 4. AC (a) and DC (b) coupled detector signals measured at the output port of the source.

Download Full Size | PDF

5. Experimental results

A typical signal recorded at the output port is shown in Fig. 4, where the output detector signal is shown as a function of time after injecting the seed pulse. The signal is AC-coupled by the detector hardware, the AC-coupled detector signal shown in plot (a) and DC in plot (b). In order to determine how noise accumulates during a scan, the seed pulses were made slightly shorter than the round-trip time in the ring (τ_ring > τ_pulse ). By observing the modulation depth, here defined as the pulse power relative to the noise floor between pulses, the noise accumulation was measured. A zoom of the individual pulses is shown in Fig.5, and the principle in deriving the modulation depth is illustrated in Fig.6. The modulation depth derived by post-processing the detector signal is shown in Fig.7. The derived modulation depth shows three domains typical for the LSFS; a relaxation domain, a stable domain and the final domain where noise rapidly increases and the signal pulse is eventually lost.

In the relaxation domain, a slowly varying oscillation is seen in the modulation depth, and is also evident in the AC-coupled detector signal shown in Fig.4(a). These oscillations are caused by the interaction between the circulating field and the population inversion in the amplifier. They can be reduced by careful optimization of the seed power, but cannot be avoided altogether. Compared to previous reports on the LSFS, the relaxation oscillation seen in our setup is more pronounced. We believe that this is caused by our relative short ring length, since it has been reported that increasing the ring length also reduces the effect of relaxation oscillations [3]. In attempts to further increase the scanning speed by reducing the length of the ring, relaxation oscillations must be reduced. Automatic gain control can be used to control the relaxation oscillations, for example with a feedback circuit controlling either pump power of the gain medium or RF power to the ring AOM.

 

Fig. 5. Zoom of the AC detector signal showing individual pulses. (Movie of the pulse train, 1.2 Mb).

Download Full Size | PDF

 

Fig. 6. Principle in deriving the modulation depth.

Download Full Size | PDF

After the relaxation oscillations have ceased, the pulse amplitude and modulation depth settles at a constant level. The modulation depth remains constant until the signal pulse starts to experience attenuation from the bandpass filter, as its optical frequency is scanned towards the filter edge. At this final domain, the signal amplitude drops rapidly because losses for the signal increases, while the ASE noise, that can exist anywhere within the filter bandwidth, gains strength. The circulating pulse is frequency shifted upon each roundtrip in the ring by the AOM. The total frequency shift is determined by multiplying the total number of roundtrips with the AOM frequency shift. The number of roundtrips is determined by dividing the time in which the pulse can be maintained by the roundtrip time in the ring. The pulse could be maintained for ~0.8 ms, which corresponds to 1480 roundtrips and a total frequency shift of 296 GHz. The corresponding wavelength shift is 1.1 nm centered at 1060 nm. The recorded signal from the DC-port of the detector is shown in Fig.4(b). The relaxation oscillation is clearly evident in the signal, but more importantly, even after the modulation depth in the AC signal has dropped to zero the DC signal remains constant until the ring AOM is turned off at t=1.5 ms. Neglecting relaxation oscillations, this is a sign of the constant saturated output power of the system. Initially, the output power is used by the circulating signal, but is gradually handed over to the accumulating noise. The noise continues to exist until the ring AOM is switched off. The linewidth of the individual pulses is expected to be comparable to that of the seed source. In this study the linewidth of the individual pulses was not measured.

 

Fig. 7. Modulation depth derived from the recorded AC-coupled detector signal, and normalized to the stable domain.

Download Full Size | PDF

6. Model comparison

Prediction of the LSFS scanning behavior using numerical models is important, especially in the design of tunable filters, which we expect will increase the scanning range. We implemented a concatenated amplifier model, similar to that proposed in [3], for comparisons with our experiments. Good agreement would imply, that the numerical model is suited for further system optimization, e.g., in selection of parameters for a tunable filter. The model assumes a constant saturated output power of the amplifier and the ring with the constant power shared by signal and noise [4],[10]

where P_sat is the saturated output power of the ring, $P_s^i$ the signal and $P_n^i$ the total noise power of the i’th ring circulation. The equation that propagates the signal from one circulation to the next is given by

where G_i is the gain in the i’th circulation and T the passive ring loss at signal frequency ${\nu }_s^i$ for the i’th circulation. The passive ring loss includes coupling losses and the transmission function of the bandpass filter. The noise power covers a broad spectrum and has to be resolved in a number of spectral slots. In the model used in [4] and [10] the spectral slots was spaced by Δf. To reduce calculation time, we have increased the width of the spectral slots while ensuring that it did not affect the outcome. Using this method, each spectral slot $p_n^i$ is brought from one circulation to the next using

where ${\nu }_n^j$ and $p_n^i$ [${\nu }_n^j$ ] is the center frequency and power of the j’th noise slot respectively. Δν_n is the with of the noise slot, n_sp the spontaneous emission factor and h Planck’s constant. The spontaneous emission factor is related to the noise figure (NF) through NF=2n_sp . The sign of the second term depends upon whether the frequency is scanned up or down. The total noise power is the sum over all spectral slots

The constant power approximation is then invoked through the gain seen by signal and noise in the i’th circulation [10], [4]

These are the general equations governing the model, which are solved for each circulation in the ring. The calculation is initiated assuming P_sat =$P_s^1$ , $P_n^1$ =0. It has been assumed that the amplifier does not exhibit any spectral variations in its saturated output power, which is valid for the scanning ranges considered here. However, the model does not directly include the population inversion of the amplifier, and, therefore, it cannot predict the behavior of relaxation oscillations. With this model and parameters corresponding to the experiment, we have pre-

Table 1. Parameters used to generate the curves in Fig. 8.

View Table

dicted how the signal and noise power develops through a scanning cycle. The parameters used for the calculation are shown i Table 1. The parameter Δν _filt describes how much the center frequency of the filter deviates from the optical frequency of the seed source.

As illustrated in Fig. 8, the numerical experiment shows that the scanning range can be prolonged by shifting the center frequency of the filter in the direction in which the signal is scanned. This is consistent with experiments, where the center frequency of the filter initially was set for maximum transmission of the initial seed pulse by angular tuning of the thin film filter. By gradually shifting the center frequency towards a shorter wavelength, the scanning range increased. In Fig. 8, the signal power development is shown for various values of Δν _filt , where the y-axis is the output power normalized to the saturated output power of the system. However, as the offset becomes larger and larger a dip in the signal power starts developing in the time interval 0.2–0.6 ms, and the dip starts to become significant for offsets above 160 GHz. In Fig. 9, the normalized signal development predicted by the model is compared with the normalized modulation depth from the experiment. The modulation depth from the experiment is normalized to the stable domain. In both cases the normalized curves correspond to normalizing to the saturated output power. Neglecting relaxation oscillations, both experiment and model predict a stable output up to around 0.9 ms. Hereafter, the signal is rapidly lost due to noise accumulation. The filter bandwidth was controlled experimentally. Varying modelling parameters within the experimental uncertainty did not affect the numerical outcome significantly.

 

Fig. 8. Modulation depth/signal development predicted by the concatenated numerical amplifier model. Model parameters are shown in Table 1. (Movie showing development in signal and noise spectral content over time, 3.3 Mb)

Download Full Size | PDF

 

Fig. 9. Comparison between the experimentally derived modulation depth and the modulation depth predicted by the numerical model. The filter offset parameter Δν _{f ilt} has been set to 160 GHz; other parameters as in Table 1.

Download Full Size | PDF

7. Conclusion

A frequency shifting ring source operating in the 1–1.1 µm range has been demonstrated. We have, for what we believe to be the first time, shown that the YDFA successfully may be used in the LSFS configuration in spite of a noise figure above the fundamental 3 dB limit. Furthermore, the experiments showed a qualitative behavior similar to EDFA-based systems. The behavior observed experimentally could be predicted using a concatenated numerical amplifier model. An unwanted Q-switching behavior was observed experimentally, which we believe is caused by our short ring length. However, we proposed a novel timing scheme to overcome this limitation. It was demonstrated experimentally that our novel timing scheme effectively removed the unwanted Q-switching behavior. Based on our numerical model, we expect to be able to increase the scanning range significantly using a tunable filter in the system. Our future investigation aims at the successful implementation of such a filter in order to achieve a longer scanning range.