**3. Tunable photonic microwave filters**

Photonics-based microwave and millimeter-wave filters offer advantages such as ultra-wide bandwidth, immunity to electromagnetic interference, and lightweight (Seeds&Williams, 2006; Capmany et al., 2006; Minasian, 2006). These advantages open new opportunities in a wide range of potential applications especially when high selectivity, resolution, wide tunability, and fast reconfigurability characteristics are required.

 In recent years, numerous reconfigurable coherent-free photonic microwave transversal filter structures have been proposed and demonstrated, where multi-wavelength source is employed to suppress the optical interference in conjunction with modifying the optical tap weights or the time-delay increment between taps (Capmany et al., 1999; Polo et al., 2003; Ortigosa-Blanch et al., 2006; Hunter&Nguyen, 2006; Ning et al., 2007; Blals&Yao, 2008). Spectral slicing of an RF-modulated broadband optical source has been employed to generate different wavebands. However, the use of Bragg gratings or arrayed waveguide gratings for realizing spectral slicing results in fixed time-delay increments, which limit the

Photonic Microwave Signal Processing Based on Opto-VLSI Technology 365

attenuation for that spectral component. Therefore, by manipulating the phase hologram of individual pixel block, the power of each waveband component can be independently adjusted according to the required tap weights; and by partitioning the pixels into appropriate blocks, the optical taps with specific wavelength spacing can be picked out from the wideband light source while all the other wavebands are steered off-track and attenuated dramatically (Xiao *et al.*, 2008). For example, by applying an appropriate phase holograph consisting of 5 different blazed gratings, 5 different wavebands could be steered back and coupled to the collimator, as shown in Fig. 5. In other words, by configuring the phase hologram employed to the Opto-VLSI processor, tap weights and tap separations can

**λ<sup>1</sup> λ<sup>2</sup> λ<sup>3</sup> λ<sup>4</sup> λ<sup>5</sup>**

The optical taps that are coupled back into the fiber collimator are fed, through the circulator, to a high dispersion optical fiber, where the different RF-modulated wavebands experience different delay times. The delayed RF-modulated wavebands are finally photodetected by a photo-receiver built into a Network Analyzer, which displays the microwave filter response. An optical spectrum analyzer (OSA) is also used as a monitor driven by a

In the experiments, the RF signal generated by the Network Analyzer was used to intensity modulate the broadband ASE source using a JDS Uniphase electro-optical modulator of 4- GHz bandwidth. A 256-phase-level one-dimensional Opto-VLSI processor was used, which has 1×4096 pixels, with 1 μm pixel size and 0.8 μm dead spacing between adjacent pixels. A Labview software was specifically developed to appropriately partition the pixel blocks so

small fraction of the light detected by the photodiode of the Network Analyzer.

**λ**

**Diffraction grating**

pixel

be adjusted simultaneously, independently and continuously.

**Opto-VLSI**

Selected Waveband

Φ(2π)

Fig. 5. The principle of optical waveband selection.

Phase Hologram

**Collimator**

μW

**λ<sup>1</sup> λ<sup>5</sup>**

tunability of the photonic microwave transversal filter. Another approach to generating optical taps is the use of a tunable laser array, where each tunable laser element is dedicated to control the weight of a single optical tap. However, the main disadvantage of this approach is the high cost and reliability of the filter structure, especially when the number of taps increases.

As a powerful reconfigurable holography technology, a novel tunable photonic microwave filter has been proposed and experimentally demonstrated based on the use of an Opto-VLSI processor (Xiao et al., 2009). Through computer-generated phase holograms uploaded onto the Opto-VLSI processor, arbitrary spectral slicing with adaptive wavelength separations as well as independent tap weight control can be achieved. This demonstration has a significant advantage that the time-delay increment, tap numbers and tap weights can be adjusted independently and simultaneously, simply by electronics. This structure has the highest flexibility compared to previously reported microwave filter structures (Capmany et al., 2006; Zheng et al., 2006). The proposed tunable filter structure is a practical solution to realizing flexible and tunable microwave filters.

In Fig. 4, the proposed tunable photonic microwave filter is illustrated through an experimental setup. A broadband light source of amplified spontaneous emission (ASE) is externally modulated by an RF signal through an electro-optic modulator (EOM). The modulated light is amplified by an erbium doped fiber amplifier (EDFA) and routed via a circulator into a collimator which collimates the light into a 1-mm diameter beam. A 1200 line/mm grating plate disperses the incident beam into spectral components along different directions and linearly maps them onto the active window of an Opto-VLSI processor.

Fig. 4. Experimental setup for the tunable photonic microwave filter structure.

The wavelength of the optical field incident onto the Opto-VLSI processor varies along the pixels, which can be logically partitioned into pixel blocks by programming the Opto-VLSI processor. The spectral component, falling within the specific pixel block of the Opto-VLSI processor, can be either steered back along the incidence path thus coupling it back into the fiber collimator with minimum attenuation, or deliberately steered "off-track" so that its power is partially coupled back into the fiber collimator leading to an appropriate optical

tunability of the photonic microwave transversal filter. Another approach to generating optical taps is the use of a tunable laser array, where each tunable laser element is dedicated to control the weight of a single optical tap. However, the main disadvantage of this approach is the high cost and reliability of the filter structure, especially when the number

As a powerful reconfigurable holography technology, a novel tunable photonic microwave filter has been proposed and experimentally demonstrated based on the use of an Opto-VLSI processor (Xiao et al., 2009). Through computer-generated phase holograms uploaded onto the Opto-VLSI processor, arbitrary spectral slicing with adaptive wavelength separations as well as independent tap weight control can be achieved. This demonstration has a significant advantage that the time-delay increment, tap numbers and tap weights can be adjusted independently and simultaneously, simply by electronics. This structure has the highest flexibility compared to previously reported microwave filter structures (Capmany et al., 2006; Zheng et al., 2006). The proposed tunable filter structure is a practical solution to

In Fig. 4, the proposed tunable photonic microwave filter is illustrated through an experimental setup. A broadband light source of amplified spontaneous emission (ASE) is externally modulated by an RF signal through an electro-optic modulator (EOM). The modulated light is amplified by an erbium doped fiber amplifier (EDFA) and routed via a circulator into a collimator which collimates the light into a 1-mm diameter beam. A 1200 line/mm grating plate disperses the incident beam into spectral components along different directions and linearly maps them onto the active window of an Opto-VLSI processor.

Fig. 4. Experimental setup for the tunable photonic microwave filter structure.

The wavelength of the optical field incident onto the Opto-VLSI processor varies along the pixels, which can be logically partitioned into pixel blocks by programming the Opto-VLSI processor. The spectral component, falling within the specific pixel block of the Opto-VLSI processor, can be either steered back along the incidence path thus coupling it back into the fiber collimator with minimum attenuation, or deliberately steered "off-track" so that its power is partially coupled back into the fiber collimator leading to an appropriate optical

of taps increases.

realizing flexible and tunable microwave filters.

attenuation for that spectral component. Therefore, by manipulating the phase hologram of individual pixel block, the power of each waveband component can be independently adjusted according to the required tap weights; and by partitioning the pixels into appropriate blocks, the optical taps with specific wavelength spacing can be picked out from the wideband light source while all the other wavebands are steered off-track and attenuated dramatically (Xiao *et al.*, 2008). For example, by applying an appropriate phase holograph consisting of 5 different blazed gratings, 5 different wavebands could be steered back and coupled to the collimator, as shown in Fig. 5. In other words, by configuring the phase hologram employed to the Opto-VLSI processor, tap weights and tap separations can be adjusted simultaneously, independently and continuously.

Fig. 5. The principle of optical waveband selection.

The optical taps that are coupled back into the fiber collimator are fed, through the circulator, to a high dispersion optical fiber, where the different RF-modulated wavebands experience different delay times. The delayed RF-modulated wavebands are finally photodetected by a photo-receiver built into a Network Analyzer, which displays the microwave filter response. An optical spectrum analyzer (OSA) is also used as a monitor driven by a small fraction of the light detected by the photodiode of the Network Analyzer.

In the experiments, the RF signal generated by the Network Analyzer was used to intensity modulate the broadband ASE source using a JDS Uniphase electro-optical modulator of 4- GHz bandwidth. A 256-phase-level one-dimensional Opto-VLSI processor was used, which has 1×4096 pixels, with 1 μm pixel size and 0.8 μm dead spacing between adjacent pixels. A Labview software was specifically developed to appropriately partition the pixel blocks so

Photonic Microwave Signal Processing Based on Opto-VLSI Technology 367

Fig. 6. Microwave filter tuning through tap weight control. (a), (d), and (g) phase hologram applied to the Opto-VLSI processor; (b), (e), and (h) selected RF-modulated wavebands; (c), (f), and (i) measured (solid line) and simulated (dashed line) filter responses for each case.

Fig. 7. Microwave filter tuning through time-delay increment control. (a,d,g) phase hologram applied to the Opto-VLSI processor; (b,e,h) selected RF-modulated wavebands; (c,f,i) measured (solid line) and simulated (dashed line) filter responses for each case.

that the optimum wavebands are selected. The pixel blocks are then driven by optimized phase holograms (blazed gratings) that steer the selected wavebands so that they are coupled back into the collimator. By optimizing the size and the phase profile of each pixel block, any desirable weight and time-delay increment is synthesized after the wavebands are launched into the high dispersion fiber (HDF), which has dispersion coefficient 382.5ps/nm and insertion loss 4.6 dB.

The response of the filter structure shown in Fig. 4 can be expressed as:

$$H(f) = \sum\_{r=0}^{M} a\_r \exp[-j2\pi rf\tau] \tag{3}$$

where f is the RF frequency, M is the number of the detected RF-modulated wavebands, *<sup>r</sup> a* is the rth tap weight, which is proportional to the optical power of the rth waveband, and τ is the time delay between adjacent wavebands introduced by the high dispersion fiber. The time delay, τ, can also be expressed in terms of the dispersion of the HDF as

$$
\tau = \alpha \cdot \Delta \mathcal{X} \tag{4}
$$

where α denotes the dispersion coefficient of the HDF, and Δλ is the adjacent waveband separation.

The principle of the photonic microwave filter was demonstrated using 5 optical taps. Figures 6(a), 6(d), and 6(g) show the phase holograms applied to the Opto-VLSI processor to generate a constant time-delay increment with variable filter weights. The Opto-VLSI processor was partitioned into 5 pixel blocks corresponding to 5 wavebands, and the size of each pixel block was 400 pixels, resulting in a center-to-center waveband separation of 3.60 nm. For each pixel block, an optimum phase hologram was appropriately uploaded, so that the power level of the specific waveband was attenuated to an appropriate intensity. As shown in Figure 6(a), an appropriate phase hologram for each pixel block was employed to generate a normalized weight profile of [1, 1, 1, 1, 1]. Fig. 6(b) shows the selected wavebands measured by the OSA, and Fig. 6(c) shows the corresponding filter response. Note that the waveband coupled back into the collimator depends on the steering angle associated to this waveband and the numerical aperture of the collimator, and that the separation between adjacent wavebands depends on their pixel block sizes. The measured linewidth for each waveband was about 0.5 nm. Note that the spectral range illuminating each pixel block was around 3.6 nm. However, because this spectral range was diverging, only 0.5nm of this range was actually steered back and coupled into the collimator. By reconfiguring the phase holograms uploaded onto the various pixel blocks, the filter weights were varied to [0.4, 0.8, 1, 0.8, 0.4], and then to [0.2, 0.8, 1, 0.8, 0.2] as shown in Figs. 6(e) and 6(h). The corresponding filter responses are shown in Figures 6(f), and 6(i), respectively, where changes in rejection band and bandwidth are demonstrated, as a result of waveband attenuation.

Note that in Figures 6(c), 6(f) and 6(i) the solid lines denote the experimental results, which agree well with the simulation results calculated from Eq. (3,4), shown in dashed lines. All the filter responses shown in Fig. 6 exhibit a free-spectral range (FSR) of about 722 MHz and this is in good agreement with the specified dispersion coefficient of the HDF used in the experiments. The time delay increment, which is the product of the waveband separation and dispersion coefficient of the HDF, was 1.38 ns.

that the optimum wavebands are selected. The pixel blocks are then driven by optimized phase holograms (blazed gratings) that steer the selected wavebands so that they are coupled back into the collimator. By optimizing the size and the phase profile of each pixel block, any desirable weight and time-delay increment is synthesized after the wavebands are launched into the high dispersion fiber (HDF), which has dispersion coefficient

0

=

τ α= ⋅Δλ

time delay, τ, can also be expressed in terms of the dispersion of the HDF as

denotes the dispersion coefficient of the HDF, and Δ

demonstrated, as a result of waveband attenuation.

and dispersion coefficient of the HDF, was 1.38 ns.

*M r r*

( ) exp[ 2 ]

π τ

= − (3)

λ

(4)

is the adjacent waveband

*H f a j rf*

where f is the RF frequency, M is the number of the detected RF-modulated wavebands, *<sup>r</sup> a* is the rth tap weight, which is proportional to the optical power of the rth waveband, and τ is the time delay between adjacent wavebands introduced by the high dispersion fiber. The

The principle of the photonic microwave filter was demonstrated using 5 optical taps. Figures 6(a), 6(d), and 6(g) show the phase holograms applied to the Opto-VLSI processor to generate a constant time-delay increment with variable filter weights. The Opto-VLSI processor was partitioned into 5 pixel blocks corresponding to 5 wavebands, and the size of each pixel block was 400 pixels, resulting in a center-to-center waveband separation of 3.60 nm. For each pixel block, an optimum phase hologram was appropriately uploaded, so that the power level of the specific waveband was attenuated to an appropriate intensity. As shown in Figure 6(a), an appropriate phase hologram for each pixel block was employed to generate a normalized weight profile of [1, 1, 1, 1, 1]. Fig. 6(b) shows the selected wavebands measured by the OSA, and Fig. 6(c) shows the corresponding filter response. Note that the waveband coupled back into the collimator depends on the steering angle associated to this waveband and the numerical aperture of the collimator, and that the separation between adjacent wavebands depends on their pixel block sizes. The measured linewidth for each waveband was about 0.5 nm. Note that the spectral range illuminating each pixel block was around 3.6 nm. However, because this spectral range was diverging, only 0.5nm of this range was actually steered back and coupled into the collimator. By reconfiguring the phase holograms uploaded onto the various pixel blocks, the filter weights were varied to [0.4, 0.8, 1, 0.8, 0.4], and then to [0.2, 0.8, 1, 0.8, 0.2] as shown in Figs. 6(e) and 6(h). The corresponding filter responses are shown in Figures 6(f), and 6(i), respectively, where changes in rejection band and bandwidth are

Note that in Figures 6(c), 6(f) and 6(i) the solid lines denote the experimental results, which agree well with the simulation results calculated from Eq. (3,4), shown in dashed lines. All the filter responses shown in Fig. 6 exhibit a free-spectral range (FSR) of about 722 MHz and this is in good agreement with the specified dispersion coefficient of the HDF used in the experiments. The time delay increment, which is the product of the waveband separation

The response of the filter structure shown in Fig. 4 can be expressed as:

382.5ps/nm and insertion loss 4.6 dB.

where

separation.

α

Fig. 6. Microwave filter tuning through tap weight control. (a), (d), and (g) phase hologram applied to the Opto-VLSI processor; (b), (e), and (h) selected RF-modulated wavebands; (c), (f), and (i) measured (solid line) and simulated (dashed line) filter responses for each case.

Fig. 7. Microwave filter tuning through time-delay increment control. (a,d,g) phase hologram applied to the Opto-VLSI processor; (b,e,h) selected RF-modulated wavebands; (c,f,i) measured (solid line) and simulated (dashed line) filter responses for each case.

Photonic Microwave Signal Processing Based on Opto-VLSI Technology 369

resolution and the narrow bandwidth of analogue-to-digital converter, or by high power

The processing of radio frequency (RF) and microwave signals in the optical domain is an attractive approach to overcome the bottlenecks encountered in conventional electronic signal processing systems (Capmany *et al.*, 2005). A wide range of emerging RF signal processing applications require specifically high resolution, wide-range tunability, and fast reconfigurability. These requirements are difficult to achieve using conventional all-

The use of photonics-based true time delay units has extensively been investigated in the last decade for applications ranging from modern microwave radar to wireless communication systems. In particular, broadband microwave phased-array antennas require the generation of variable true-time delays at each antenna element to realize beam or null steering, and optical fibers have been the best candidates for true-time delay synthesis. Compared with all-electrical techniques, optical true-time delay generation offers the advantages of broader bandwidth, lower insertion loss, higher phase stability, smaller size, lighter weight, and excellent immunity to both electromagnetic interference and crosstalk (Frigyes&Seeds, 1995; Italia *et al.*, 2005b; Y. Chen&Chen, 2002; Rideout *et al.*, 2007). Several approaches have been adopted to realise tunable true-time delay units, including the use of in-fiber chirped Bragg gratings (FBGs) (Italia *et al.*, 2005b), white cells or fiber delay lines in conjunction with MEMS (Mital *et al.*, 2006b; Anderson *et al.*, 2006; Vidal *et al.*, 2006), integrated optical waveguides (C. M. Chen *et al.*, 2010), optically-switched fiber delay structures (Tong&Wu, 1998), dispersion-enhanced photonic-crystal fibers (Jiang *et al.*, 2005), and higher-order mode dispersive multi-mode fibers (Raz *et al.*, 2004). However, most of these reported true-time delay architectures have mainly been used for realising beam steering in phased array antennas, and therefore, they do not have the flexibility to simultaneously generate multiple arbitrary true-time delays. In addition, such architectures can only generate discrete true-time delays, making them impractical for broadband null steering (Zmuda *et al.*, 2000). Zmuda et al. have reported a few adaptive true-time delay architectures based on the use of multiple tunable lasers in conjunction with high-dispersion fibres for the implementation of broadband nulling in microwave phased arrays (Zmuda *et al.*, 2000; Zmuda *et al.*, 1998). However, the use of multiple tunable lasers requiring continuous calibration makes the system implementation very expensive and impractical. Recently, a novel true-time delay unit has been demonstrated through uploading appropriate holograms onto an Opto-VLSI processor to synthesize multiple arbitrary time delays (Juswardy et al., 2009). This true-time-delay unit, which consists of a broadband optical source using Amplified Spontaneous Emission (ASE) and high dispersion fibers, has the capability to generate multiple true-time delays for several antenna elements simultaneously, making it attractive for broadband null-steering in phased array antennas. The principle of the phased-array antenna architecture shown in Fig. 8 is demonstrated using the experimental setup illustrated in Fig. 4. In this setup, a broadband ASE source was modulated, via a JDS Uniphase electro-optical modulator (EOM) with a half-wave voltage of 6 V, by an RF signal which was generated using a 20 GHz network analyzer. The RFmodulated optical signal was amplified by an EDFA and collimated at 1-mm diameter and then launched onto a diffractive grating plate. The latter demultiplexed the collimated ASE beam into multiple RF-modulated wavebands, which were then mapped onto the active window of a 256-phase-level 1×4096-pixel Opto-VLSI processor of 1-µm pixel size and 0.8-

consumption of broadband analogue-to-digital converter.

µm dead spacing between adjacent pixels.

electronic processing, but feasible with photonics-based signal processing.

The unique capability of the proposed tunable filter structure is its ability to continuously tune the time delay increment. This is accomplished by changing the spacing between adjacent pixel blocks (i.e., the pixel block size and center). Figures 7(a), 7(d), and 7(g) show the phase holograms applied to the Opto-VLSI processor to synthesize a normalized weight profile of [0.6, 0.8, 1.0, 0.8, 0.6] but different time delay increments. The corresponding wavebands are shown in Figures 7(b), 7(e), and 7(h). For these three cases, the sizes of the pixel blocks were 200, 300, and 400 pixels, respectively, corresponding to waveband separations of 1.72 nm, 2.60 nm, and 3.60 nm, respectively. Note that by optimizing the phase hologram for each pixel block, a normalized weight profile [0.6, 0.8, 1, 0.8, 0.6] was maintained in all cases. Figures 7(c), 7(f), and 7(i) show the corresponding measured (solid) and simulated (dashed) filter responses. It is evident from Fig. 7 that the tuning of the waveband separation (ie. the time delay increment) controls the free-spectral-range (FSR) as well as the bandwidth of the filter. For example, in Fig. 7, the FSR was reduced from 1.52 GHz, to 1.01 GHz, and then to 722MHz, and the filter bandwidth dropped from 216 MHz to 142 MHz and then to 86 MHz, when the time-delay increment was increased from 0.66 ns, to 0.99 ns, and then to 1.38 ns, respectively. Note that a good agreement between the simulated and measured filter responses is displayed in Fig. 7.

By investigating the filter responses shown in Figures 6(c, f, i) and Figures 7(c, f, i), one can see that the spectral response fades remarkably as the RF frequency increases. This is due to the interaction between the dispersion of the HDF and the nonzero optical bandwidth of each waveband (Taylor *et al.*, 2007; Pastor *et al.*, 2003a). This limitation is inherent to all filters that use spectral slicing. The bandwidth of the sliced wavebands in Fig. 6 and Fig. 7 was about 0.5 nm. Preliminary experimental results have shown that a narrower waveband bandwidth can be achieved using a higher dispersion grating plate and/or a larger distance between the grating and the Opto-VLSI processor.

Even though the optical tap generation involves fiber to free-space and free-space to fiber coupling, standard passive micro-assembly can be used to realise low-cost, robust and stable alignment without the need of automated high- precision stages (Baxter, 2006). Note that the maximum number of optical taps that can be generated depends on the bandwidth of the ASE source, the size of the active window of the Opto-VLSI processor, and the waveband separation. For our current experimental system, which based on the 1-D Opto-VLSI processor, up to 12 optical taps with the waveband separation of 2.60 nm can be generated. Note that, by using a 2-D Opto-VLSI processor, one can increase the number of optical taps significantly.
