1. Introduction

Programmable optical fiber delay line (OFDL) has attractive applications, such as antenna beam forming in phase array radar system [1] and microwave signal processing[2,3], due to various advantages in comparison with the traditional methods. It can provide true delay of microwave signal, very high time-bandwidth product, immunization to electromagnetic interference (EMI) and the possibility of spatial and wavelength parallelism using WDM techniques. So far, two types of optical fiber delay lines have been proposed as buffers for optical communication networks: re-circulating structure and traveling-wave structures [4]. Re-circulating buffers are compact and require fewer components to implement. However, the fiber loop length is fixed and hence it is not suitable for applications in phase array radar system and microwave signal processing. Traveling-wave buffer has the potential for the above applications. It can be cascaded to provide desired delay using, e.g., 2×2 optical switches or broadcast-and-select configuration [4]. Of course, high-speed (≤1ns) and low-loss (≤1dB) optical switches are required [1,5].

Two typical high speed devices, i.e. LiNbO₃ switch [6] and semiconductor optical amplifier (SOA) ON–OFF gate [4], can be considered. For LiNbO₃ 2×2 switch, the large insertion loss, polarization dependence and crosstalk need to be addressed before they can be used in phase array radar system and microwave signal processing. SOA ON–OFF gate has nanosecond switching speed, large on-off extinction ratio and wide optical bandwidth. It seems more suitable for traveling-wave type of fiber delay lines[7].

In this paper, we proposed a novel 3n feed-forward OFDL using SOAs as switching components. Theoretical analysis and experimental verification are presented. Important issues, such as the attenuation [8], polarization [9,10], delay precision, and etc., are addressed in detail.

2. Design of kⁿ feed-forward optical fiber true delay line

Traditionally, the programmable traveling-wave fiber delay line is implemented by using 1×N optical switch, or equally by using 1×N coupler plus on-off gate (e.g., SOA) in each arm, as shown in Fig. 1(a) [11]. In this configuration, the maximum number of possible delays, y, is equal to the mount of SOAs, x. It is apparently not suitable to be scaled for large delays. The delay performance can be enhanced by configuring it in an alternative way with 2×2 switches(or 2×2 couplers plus ON-OFF SOA gates) and same number of SOAs as shown in Fig. 1(b). In this configuration, the total number of delays can be expressed as follows:

where the number of SOAs, x, is supposed to even. We define such structure as 2ⁿ structure, where, n=x/2, is the number of stages.

Similarly, we can use x SOAs to build k^J structure as shown in Fig. 1(c), where k is the dimension of each stage and J is the number of stages(J=x/k is an integer). It can be derived the total number of delays, y, is given by:

Equation (2) can be rewritten as:

Equation (3) shows that for the same x, there is an optimal k that gives maximum y. We use L(k) to express the right side of Equation(3):

The peak value of L(k) can be obtained mathematically by differential operation on L(k) as follows:

So, the k can be obtained by solving Equations (4) and (5):

 

Fig. 1. The architectures of traveling-wave fiber delay lines (y=x for (a), y=2^x/2 for (b), and y=k^x/k for (c), where y is the maximum number of possible delays, and x is the number of SOAs)

Download Full Size | PDF

However, as shown in Fig. 1, the physics meaning of k is the number of delay lines in each stage, so, k should be integer. Figure 2 shows the relationship between L(k) and k. It can be easily found that the optimal value k is 3. It can also be proved that 4ⁿ structure is exactly the same as 2ⁿ structure:

When the number of SOAs used is small, the difference between the maximum numbers of delays available for k=2 and k=3, respectively, are also small. However, as more SOAs are used, the difference becomes significant, as shown in Fig. 3. From the figure, we can see that fast tunable true delay line with large number of delays can be obtained by employing 3ⁿ structure and certain number of SOAs.

 

Fig. 2. L(k) as a function of k

Download Full Size | PDF

 

Fig. 3. Dependence of total delay number on the delay units per stage (x: the number of SOAs)

Download Full Size | PDF

3. Implementation and performance measurement

According to the above analysis, 3ⁿ feed-forward optical fiber true delay line based on SOAs can provides large amount of delays with fast tunability and easier scalability. For practical applications, problems such as insertion loss and polarization dependence should be solved. The former can be overcome by SOA itself for it provides gain in addition to ON-OFF gating. In fact, this is also one of advantages of employing SOAs. We use Farady rotation mirrors (FRMs) to alleviate the polarization dependence in optical fiber [12] and in SOAs. The SOAs we used are bi-directional SOA without inside optical isolators at input and output. Due to the low backward reflection of the SOA and the polarization rotation by FRM, no oscillation was observed. Figure 4 shows the structure for practical implementation and we built a two-stage (the first two stages) feed-forward optical fiber true delay line for experimental verification. For each stage, the light is fed from the input port of the optical circulator. It experiences delay whose amount is determined by turning on one of the SOAs and is then forwarded to the next stage via the output port.

The main insertion loss is caused by 1×3 coupler. The light beam passes the coupler twice and experience about 12 dB loss (excluding the coupler’s excess loss). By taking the losses resulted from the FRM and the circulator into consideration, the total loss of each stage is around 13 dB. It should be noted that the beam also passes SOA twice. Usually, an SOA can provide with 5–8 dB gain for small optical signals, e.g. -10dBm, under moderate drive current (~80mA). Therefore, by careful design of the drive current of each SOA, it is possible to obtain a delay stage with no insertion loss. This feature is particularly important for practical applications. It enables the proposed structure to be cascaded without limitation on the stage number.

 

Fig. 4. A practical structure of 3ⁿ feed-forward optical fiber true delay line

Download Full Size | PDF

The resolution is another key issue to be addressed. The delay introduced by a light path distance of 0.15mm, or fiber length of 0.103mm (group refractive index of 1.46 is supposed), is about 0.5 ps. It corresponds to a phase control precision of about 1.8° in X-band (λ~3cm). In order to reach the above precision level, we used hydrogen flame to precisely adjust the fiber length, with the help of the high precision reflectometer (HP8504)[12]. This equipment is used to precisely measure the lightpath difference between the measured arm and reference arm of Michelson interferometer. The measurement span (equivalent distance in air) is 1–400 mm with resolution of 20 µm. Figure 5 shows the result when we set all six SOAs ON. It can be clearly seen that there are 9 peaks, which represent the all possible delays to be obtained, and each delay can be programmablely switched by setting of the relative one SOA per stage. The measured relative light path distances in three arms of the first stage are 20.756mm, 20.914mm, 21.056mm, respectively. The correspondent fiber lengths are 14.216 mm, 14.325mm, 14.422mm, respectively. These data indicate that τ as shown in Fig. 3 is 0.5 ps and the precision is kept within 0.03ps. One may note that there are insertion losses in stages 2 and 3. This is because of the limited drive current of the circuit we used, resulting in small SOA gain incapable to compensate the inherent insertion loss mentioned above.

Figure 6 shows the measured result of tuning speed. The rise time is 18.6 ns and the fall time id 39.4 ns. Such speed can enhance the scanning performance of microwave significantly. The scanning frequency may be improved from tens of kHz up to 10 MHz. The drive circuit for the SOAs was implemented with conventional electronic devices at hand. There is large potential to improve the tuning speed, especially the switching-off performance.

One may note that there is extra delay in switching-off process, which is mainly resulted in from the large junction capacitance in the device.

We also measured the polarization dependence loss (PDL) using polarization controller. The measured result is 0.18 dB. In comparison with the PDL of SOA (~0.3dB), there is improvement that is mainly due to the employment of FRM.

The relatively low noise figure of the SOA may bring a certain limitation to the proposed structure for some applications, especially for those requiring high transmission SNR. This indicates that for expansion purpose, noise performance should be taken into consideration, which is a topic worthy of intensive study and, however, beyond the scope of the paper.

 

Fig. 5. Measured result of relative fiber lengths x-axis: light path difference(0.2 mm/div), y-axis: relative intensity of the output(dB)

Download Full Size | PDF

 

Fig. 6. Photograph of tuning speed measurement (lower trace: drive pulse, upper trace: optical output)

Download Full Size | PDF

4. Conclusions and discussion

In this paper, we proposed a new kind of 3ⁿ feed-forward optical fiber delay lines. Theoretical analysis shows that it can provide large delays and is easy to be expanded. Experimental demonstration to implement such delay lines using SOAs and Farady rotation mirrors is presented. Delay step as small as 0.5 ps with precision of about 0.03 ps was realized. Measured results verify the feasibility of the proposed method. Its excellent expandability, low insertion loss, low polarization dependence loss and high tuning speed are suitable for application in phase array radar system and microwave signal processing.