Coherence-Free Mv Photonic Bandpass Filter Using a Frequency-Shifting Recirculating Delay Line

262JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 3, FEBRUARY 1, 2010Coherence-Free Microwave Photonic Bandpass Filter Using a Frequency-Shifting Recirculating Delay LineCibby Pulikkaseril, Student Member, IEEE, Erwin H. W. Chan, and Robert A. Minasian, Fellow, IEEE, Fellow, OSAAbstract—A new microwave photonic filter that realizes a high-resolution bandpass filter response with coherence-free operation and without the limitations of phase-induced intensity noise is presented in this paper. It is based on a frequency-shifting recirculating delay line, which can simultaneously achieve a multitap, high-resolution filter and a very large suppression of noise, while also enabling the use of a conventional narrow-linewidth laser source without suffering from coherent interference effects. value of 200 and an Experimental results demonstrate a filter SNR improvement of 40.3 dB compared to conventional recirculating delay line structures. Index Terms—Bandpass filter, microwave photonics, optical filters, optical fiber devices, photonic signal processing.I. INTRODUCTIONPHOTONIC signal processing is a powerful technique for processing high-bandwidth signals. The unique, high time-bandwidth product capabilities of photonic signal processing can overcome the inherent bottlenecks caused by limited sampling speeds in conventional electrical signal processors. Moreover, this approach allows direct processing of high-frequency signals that are already in the optical domain [1]. Optical fiber, with low loss (independent of RF), is an excellent medium for processing high bandwidth signals [2], and also provides advantages of electromagnetic interference (EMI) immunity. Many applications require high-frequency selectivity and high-resolution bandpass filtering. This generically requires a large number of taps in the impulse response of the discrete-time signal processor. Filters based on a recirculating delay line (RDL) can generate a large number of optical taps using simple structures. However, the RDL [3] and other previously reported high-resolution photonic processors [4] are principally limited by the excessive phase-induced intensity noise (PIIN), which is generated by the optical interference in summing the multiple delayed optical signals [5]–[7]. To obtain a robust transfer characteristic that is insensitive to environmental perturbations, conventional approaches have required the use of an incoherent approach, in which the coherence time of the light source is made smaller than the minimum delay time of the processor. However, when the light from the same optical source is delayed in different optical paths and then recombined together in the optical domain, the phase noiseManuscript received July 15, 2009; revised November 17, 2009. First published December 15, 2009; current version published January 15, 2010. This work was supported by the Australian Research Council. The authors are with the School of Electrical and Information Engineering, University of Sydney, Sydney, N.S.W. 2006, Australia (e-mail: cibby@yd. edu.au). Digital Object Identifier 10.1109/JLT.2009.2038349of the light is converted to intensity noise [6], [7], and this problem compounds for RDLs that have a large number of taps. This is important because high-resolution processors require a large number of taps; thus, efforts to increase the resolution of the processor are accompanied by an increase in the dominant PIIN noise. Hence, this limits the filter performance as the PIIN significantly degrades the SNR. Several methods of removing PIIN have been proposed, including the use of different optical wavelength sources [8], double-pass modulation [9], and a Sagnac loop [10]; however, these approaches are restricted to only a limited number of taps. We have also previously reported an approach based on performing multiple successive intensity modulations to realize a coherence-free bandpass filter with an increased number of taps [11]; however, since this requires a separate modulator for each pair of taps generated, this approach cannot generate the many hundreds of taps required in high-resolution filters. The object of this paper is to present a new concept to solve the problem of generating a large number of taps without the limitations of PIIN noise. It is based on a frequency-shifting RDL (FS-RDL). This can simultaneously achieve a high-resolution filter and a very large suppression of PIIN noise. Moreover, it enables the use of a conventional narrow-linewidth laser as the optical source without suffering from coherent interference effects. Experimental results are presented that demonstrate a filter value of 200 and a significant PIIN noise reduction of 41 dB. This paper is structured as follows. The principle of operation and structure of the FS-RDL is detailed in Section II. Section III presents an analysis of the dominant sources of noise in the FS-RDL. Section IV presents the simulation results, and Section V describes the experimental results. Finally, conclusions are presented in Section VI. II. FS-RDL TOPOLOGY The new FS-RDL structure for high-resolution, low-PIIN signal processing is shown in Fig. 1. Modulated light from a laser is injected into a frequency-shifting recirculating loop. We first proposed this structure for microwave photonic filters in [1], and here, we present the full noise analysis, design, experimental validation, and results. The optical frequency shifter in the loop shifts each recirculating tap by , e.g., a tap that has circulated times will experience a shift of . The optical amplifier in the loop compensates for the loop losses and allows control over the loop gain. This processor structure can create many taps because numerous recirculations can occur. Since the FS-RDL creates filter taps with different wavelengths arriving at the photodetector, the frequency separation of these taps causes the PIIN noise to be shifted away from0733-8724/$26.00 © 2010 IEEEPULIKKASERIL et al.: COHERENCE-FREE MICROWAVE PHOTONIC BANDPASS FILTER263The noise power spectrum can be obtained from the autocorrelation of the output optical intensity, which is given byFig. 1. General structure of the FS-RDL.baseband to the beat frequency, which is equal to the frequency shift . By designing the frequency shift to be larger than the bandwidth of the photodetector, the majority of the PIIN is automatically filtered out. Moreover, since every tap generated by the structure has a different wavelength, the filter is essentially free of coherent interference, thus enabling the use of a narrow-linewidth telecommunications-type laser as the optical source for the system. III. NOISE ANALYSIS A. Phase-Induced Intensity Noise We present the noise analysis for the PIIN when there is a frequency shifter in the recirculating loop, assuming a single polarization state. The PIIN spectrum is obtained from the electric field at the filter output, and from Fig. 1, we observed that the output electric field is given by(5) Following the procedure in [6] and [7], the autocorrelation can be expressed as(6) (1) is the optical carrier angular frequency, is where the characteristic delay time of the loop, is the angular frequency shift, is the random time-varying phase, and (2) with being the amplitude of the input electric field and where is the average optical intensity in the loop and(7) and represents the coherence time of the optical source. We use the definition of coherence time proposed by Moslehi [7], which states that coherence time is equal to half width of the self-coherence function of the laser light field. According to the Wiener--Khinchine theorem, the power spectrum of a random process is given by the Fourier transformation of its autocorrelation function [12]. From (6), we obtained that there are only two terms that need to be transformed: , which contains the effect of the frequency shifter, and , which is related to the linewidth of the optical source. The one-sided transform of is well known and given by (8) , can be obtained by using The transform of the sifting property of the Dirac delta function, and is given by (9)(3) is the coupling coefficient of the optical coupler and where is the gain of the optical amplifier. Since the output optical intensity is the modulus of the output electric field squared, it can be expressed as(4) where * denotes complex conjugation.where page).is given by (10) (given at the bottom of the next264JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 3, FEBRUARY 1, 2010Using (6), (8), and (9), the FS-RDL PIIN spectrum is given byThe output ASE field can be written as(16) where is the center frequency of the ASE field, is the random time-varying phase of the ASE field, and (11) where (12) Equation (11), which gives the PIIN noise spectrum when there is a frequency shifter in the recirculating loop, reduces to is set to zero, i.e., for the case the results of [6] and [7] when when there is no frequency shifter in the recirculating loop. B. Signal-Spontaneous Beat Noise The FS-RDL also produces signal-spontaneous (S-SP) beat noise in addition to PIIN, where the input optical carrier beats with amplified spontaneous emission (ASE) produced by the optical amplifier in the loop. S-SP beat noise occurs in the conventional RDL, and has been analyzed by Kringlebotn et al. [13], though the authors were unable to measure the S-SP beat noise, as it was dominated by PIIN. The FS-RDL, on the other hand, produces a much lower level of PIIN. At the operating frequencies of interest of the FS-RDL, S-SP beat noise is expected to be the dominant noise source. The S-SP beat noise analysis is obtained from the correlation of the signal and ASE fields at the photodetector, which is given by (17) with being the amplitude of the ASE field at the amplifier output and (18) This shows that each recirculation of the ASE field will experience the same overall gain as the signal field, but will not necessarily exit the FS-RDL with the same amplitude and phase. This is due to the fact that the ASE field originates from inside the loop. Note that the center frequencies of the ASE and signal fields are equal. This is because the ASE will experience self-filtering effects inside the loop [15]. Additionally, the FS-RDL permits the use of a narrow-linewidth laser source; if such a source is used, the coherence time of the ASE field will be much smaller than the coherence time of the laser source. As a result, the linewidth of the ASE field dominates. By using these simplifications, and following a similar procedure to the PIIN analysis, the cross correlation of the S-SP beat noise can be expressed as(13) represents the signal field at the output of the where FS-RDL, given by (1), and is the ASE field at the output of the FS-RDL. As these two sources are uncorrelated, the correlation can be expressed as the product of their separate self-coherence functions, as defined by Nazarathy et al. [14], as where(19)(20) with being the coherence time of the ASE generated by the optical amplifier inside the loop. Similarly to the derivation of the FS-RDL PIIN, the correlation of the S-SP beat noise can be obtained by expanding (19) and collecting similar terms. Unfortunately, the inherent asymmetry of the S-SP beat noise complicates the analysis, thereby creating a correlation function with both real and imaginary(14) where the self-coherence function of an optical source is (15)(10)PULIKKASERIL et al.: COHERENCE-FREE MICROWAVE PHOTONIC BANDPASS FILTER265parts; from (14), however, only the real part is of interest. Hence, the correlation of the S-SP beat noise can be expressed asC. S-SP Beat Noise In addition to the PIIN noise and S-SP noise, the FS-RDL also produces spontaneous-spontaneous (SP-SP) beat noise, where the ASE field beats with itself. Since the ASE noise power is generally expected to be less than the signal power, the SP-SP beat noise should occur at a level below S-SP beat noise. The ASE field can be considered as an unpolarized thermal noise source, which allows the use of circularly complex Gaussian statistics [17], which, for our spectrum of interest, allows the autocorrelation of the output ASE intensity to be written as the product of two second-order moments as(25) (21) The FS-RDL SP-SP beat noise spectrum can be obtained by following a similar procedure to the derivation of the FS-RDL S-SP beat noise spectrum, and is given by where is the ASE intensity, which is propor[16], where is the amplifier tional to noise factor, is the photon energy, and is the optical bandwidth of the ASE field, and (22) The noise power spectral density of the S-SP beat noise is obtained by taking the Fourier transform of (21). The correlation of the S-SP beat noise shares similar components with (6), which suggest that the same method can be used to obtain the noise spectrum. The FS-RDL S-SP beat noise spectrum is given by where (27) The SP-SP beat noise expression is much less complex than the PIIN and S-SP expressions; this is because the ASE noise originates from the loop and all recirculations see the same phase change when exiting the optical coupler. IV. SIMULATIONS AND DISCUSSION Simulations were carried out to evaluate the amount of noise reduction that the FS-RDL microwave photonic filter can achieve relative to the conventional RDL microwave photonic filter operating under the same frequency response characteristic conditions. The reference filter used for the purpose of these comparisons was a microwave photonic filter with a value of 177 and an FSR of 100 MHz. To attain this value with a coupler coefficient of , an optical amplifier gain of was used. For the FS-RDL, a frequency shifter with 3 GHz shift was used in the loop and a laser source linewidth of MHz was selected. For the conventional RDL, an optical source linewidth of GHz was used to prevent coherence interference so that the filter response can be insensitive to environmental perturbations. For both cases, the ASE was characterized by a linewidth of GHz, and normalized optical intensities of and . It should be noted that to prevent aliasing in the FS-RDL, which potentially could occur due to the frequency-shifted modulated carriers beating at the photodetector, the frequency shift must be more than three times the RF modulation frequency [1]. Thus, the frequency of operation is limited by the frequency shift available in the loop. However, optical frequency shifters with 25 GHz frequency shift have been reported using lithium--(26)(23) where (24) Equation (23) is similar to the PIIN spectrum given in (11), with the replacing and being proportional to , instead of only . Since the linewidth of the ASE field is very broad, the S-SP beat noise spectrum is expected to be flat, with resonance peaks at integer multiples of the filter-free spectral range (FSR). Furthermore, (23) is proportional to the intensity of the ASE created by the optical amplifier, which indicates that the level of the S-SP beat noise spectrum is dependent on the operating conditions of the optical amplifier.266JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 3, FEBRUARY 1, 2010Fig. 2. Simulated comparison between noise sources of the conventional RDL GHz) and the FS-RDL (  : MHz). Parameters used are I (  : ; v ;K : ; ;I GHz, f GHz, N : . and G1 =2 = 0 01 1 1 = 1 965= 3751 =15 1 =3= 250= =05Fig. 3. Noise reduction of the FS-RDL PIIN compared to the conventional RDL PIIN as a function of (a) source linewidth and (b) frequency shift.niobate technology [18]. Hence, the FS-RDL can be operated at microwave frequencies. A comparison between the noise performance of the FS-RDL obtained from (11), (23), and (26), and the expression for the PIIN spectrum of an RDL [7] is shown in Fig. 2. For the FS-RDL, as expected, the main lobe of the PIIN spectrum has been shifted to 3 GHz, and consequently, it can achieve a very large noise reduction at baseband, whereas for the conventional RDL, the PIIN noise is very high at baseband, where the filter is designed to operate at. As an example, for a nominal filter passband frequency of 1 GHz, the FS-RDL actually reduces the PIIN noise to be below the S-SP beat noise. Taking into account all the noise sources, the FS-RDL demonstrates a significant noise reduction of 42.2 dB compared to the conventional RDL. To investigate the factors that influence the amount of PIIN noise reduction that can be obtained at the filter passband center, the effect of the linewidth of the laser source and the frequency shift in the loop were investigated. Fig. 3 shows the amount of noise reduction at the passband center that is achievable by varying these two parameters; in Fig. 3(a), the linewidth of the source is varied for a loop with a 3 GHz frequency shifter, and in Fig. 3(b), the frequency shift is varied for a laser linewidth of 1.5 MHz. It can be seen from Fig. 3(a) that a large PIIN noise reduction can be obtained by using a narrow-linewidth laser. Fig. 3(b) reveals an interesting result, namely, that to achieve maximum noise reduction of PIIN noise in the FS-RDL, the frequency shift should be an integer multiple of the filter FSR. However, even if this condition is not fulfilled, there are still very significant reductions in the PIIN noise. This effect is described by (11), with the function given by , which is the only term that depends on . When the FS-RDL frequency shift is an integer multiple of the filter FSR, the notches that appear in will align to the resonant FSR peaks, thereby reducing the noise power spectral densities at these frequencies. When other frequency shifts are used, the PIIN can increase above this value. This phenomenon is illustrated in Fig. 4, where, using the identical parameters as indicated in Fig. 2, appropriate selection ofFig. 4. FS-RDL PIIN spectrums for different frequency shifts.the FS-RDL frequency shift can cause a 20.1 dB difference in the PIIN at the filter passband center frequency. Selecting the FS-RDL frequency shift to be an integer multiple of the filter FSR also causes upshifted modulation sidebands to align to passbands around the frequency shift frequency; however, this is not an issue because the photodetector bandwidth needs to be smaller than the frequency shift in order to filter out the upshifted PIIN noise and also to avoid aliasing, and hence, these upshifted signals are automatically filtered out. Theoretically, the FS-RDL can reduce the PIIN noise at 1 GHz by 54.9 dB, using a frequency shift of 3 GHz and an optical source with a linewidth of 1.5 MHz, which can be achieved by commercial DFB lasers. V. EXPERIMENTAL RESULTS Experiments were conducted to verify the concept of the FS-RDL microwave photonic filter. The experimental setup is shown in Fig. 5. The optical source was a tunable laser (Santec) having a wavelength of 1559.9 nm and a linewidth of less than 500 kHz. The output of the tunable laser was modulatedPULIKKASERIL et al.: COHERENCE-FREE MICROWAVE PHOTONIC BANDPASS FILTER267Fig. 5. Experimental setup of the FS-RDL microwave photonic filter.by an electro-optic modulator (EOM) (Sumitomo) biased at quadrature. The modulated output was then fed into the loop structure, with a polarization controller at the input of the loop as well as inside the loop itself. With the use of these two polarization controllers, the effects of polarization-dependent loss (PDL) in the loop [19], most notably from the frequency shifter, could be successfully mitigated. The loop comprised an optical coupler with a coupling ratio of 50%, and the frequency shifter was an acousto-optic frequency shifter (Brimrose), since this was the only one available, which delivered a downshift of 750 MHz. The frequency shifter had an insertion loss of 6.3 dB with a PDL of 4.1 dB. Note that an electro-optic frequency shifter would have been desirable to achieve larger frequency shifts in the loop so as to enable operation at higher frequencies. This also necessitates the loop to have a broad gain bandwidth in order to achieve a sharp filter response. Note that frequency shifters have a finite amount of carrier suppression ratio. The unwanted original optical carrier with the same optical frequency as the desired optical signal will beat at the photodetector. This can create additional PIIN at baseband. This effect depends on how high the extinction ratio is, and therefore, it is important for the frequency shifter to have a large suppression of the original optical carrier. Acousto-optic frequency shifters use separate spatial locations to discriminate between different frequency components, so the carrier suppression is extremely high relative to the first-order component, and electro-optic frequency shifters can achieve large original carrier suppression [20], with the result that there are negligible PIIN and coherent interference problems. The optical amplifier in the loop comprised a 0.9 m length of high-concentration Er-doped fiber, pumped forward and backward with 980 nm pump lasers through wavelength division multiplexing (WDM) couplers on either side of the doped fiber. Excess ASE from the amplifier was removed by use of the self-filtering effect, which eliminated the need for an optical filter in the loop [15]. The FS-RDL output was detected by a photodetector. Since the FS-RDL FSR is determined by the loop length, tunability of the passband center could potentially be achieved by insertion of a tunable delay line in the loop. The performance of the FS-RDL was compared to a conventional RDL as a reference, which had the same structure, except that the frequency shifter was replaced with a similar length of fiber. Since the conventional RDL suffers from coherent interference, linewidth broadening must be used for the outputFig. 6. Frequency response of the FS-RDL compared to the conventional RDL.to be stable, and this was obtained by using a phase-modulation broadening scheme [21], which allowed the tunable laser linewidth to be broadened to 400 MHz. Additionally, to ensure that the RF signal power at the filter passband was equal in both cases, a variable optical attenuator was placed before the photodetector and adjusted to achieve equal power in both cases. Both the FS-RDL and the conventional RDL were operated with the same frequency response characteristics, namely, a value of 200 and an FSR of around 45 MHz. Fig. 6 displays a comparison between the frequency response of the FS-RDL and the conventional RDL, which shows that the frequency response shapes are basically similar, the slight offset in the center frequencies being due to the fact that the FS-RDL and the RDL had slightly different loop lengths. This demonstrates that the frequency shifter has no effect on the filter frequency response. It is important to note that the FS-RDL had a clean response, whereas the conventional RDL response has been averaged with an averaging factor of 16, in order to reduce the noise in the response, especially in the stopband. The predicted bandpass filter response is also shown in Fig. 6. Excellent agreement between the predicted and measured results can be seen. Fig. 7 shows a comparison of the measured noise performance for the two cases measured on a spectrum analyzer with a 10 kHz resolution bandwidth. This measurement was taken by recording the spectrum of the photodetector current when no input signal was applied to the EOM. The FS-RDL shows a large noise reduction of 41.1 dB compared to the conventional RDL at a passband frequency of 235.8 MHz. The SNR of the FS-RDL was measured by applying a 5 dBm RF tone at the filter passband of 235.8 MHz, to the EOM, and taking the ratio of the measured detected signal level to the measured noise level without the 5 dBm RF tone. The measured FS-RDL SNR was 60.4 dB in a 10 kHz resolution bandwidth. For comparison, the SNR of the conventional RDL was measured under the same condition. In this case, the SNR was 20.1 dB. This shows that the overall SNR of the FS-RDL is 40.3 dB higher than the conventional RDL, which is a very significant improvement. Fig. 8 shows a comparison of the FS-RDL noise spectrum between theory using (11) and (23), and experiment that shows good agreement. The relative values of and were selected to match experimental results. The fairly flat profile with frequency of the noise peaks268JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 3, FEBRUARY 1, 2010Fig. 7. Noise spectrum comparison between the FS-RDL and the conventional RDL.Fig. 9. Comparison between the theoretical PIIN noise spectrum predicted from the analysis of the FS-RDL and the measured noise spectrum.spectrum predicted from the analysis of the FS-RDL and the measured noise spectrum, indicating that our theory matches well to experimental results. VI. CONCLUSION A new technique to realize a high-resolution microwave photonic filter without the limitations of PIIN noise has been presented. It is based on an FS-RDL. This can simultaneously achieve a multitap, high-resolution filter and a very large suppression of PIIN noise. Moreover, it enables the use of a conventional narrow-linewidth laser as the optical source witshout suffering from coherent interference effects. Experimental results have demonstrated a filter value of 200 and a very significant noise reduction of 41 dB compared to a conventional RDL operating under the same condition. This offers high-resolution photonic signal processing with low noise performance. REFERENCES[1] R. A. Minasian, “Photonic signal processing of microwave signals,” IEEE Trans. Microw. Theory Tech., vol. 54, no. 1, pp. 832–846, Feb. 2006. [2] K. P. Jackson, S. A. Newton, B. Moslehi, M. Tur, C. C. Cutler, J. W. Goodman, and H. J. Shaw, “Optical fiber delay-line signal processing,” IEEE Trans. Microw. Theory Tech., vol. MTT-33, no. 3, pp. 193–210, Mar. 1985. [3] B. Moslehi and J. W. Goodman, “Novel amplified fiber-optic recirculating delay line processor,” J. Lightw. Technol., vol. 10, no. 8, pp. 1142–1147, Aug. 1992. [4] E. H. W. Chan, K. E. Alameh, and R. A. Minasian, “Photonic bandpass filters with high skirt selectivity and stopband attenuation,” J. Lightw. Technol., vol. 20, no. 11, pp. 1962–1967, Nov. 2002. [5] J. Capmany, “Investigation of phase-induced intensity noise in amplified fibre-optic recirculating delay line,” Electron. Lett., vol. 29, no. 4, pp. 346–348, Feb. 1993. [6] M. Tur, B. Moslehi, and J. W. Goodman, “Theory of laser phase noise in recirculating fiber-optic delay lines,” J. Lightw. Technol., vol. LT-3, no. 1, pp. 20–31, Feb. 1985. [7] B. Moslehi, “Analysis of optical phase noise in fiber-optic systems employing a laser source with arbitrary coherence time,” J. Lightw. Technol., vol. LT-4, no. 9, pp. 1334–1351, Sep. 1986. [8] J. Capmany, D. Pastor, and B. Ortega, “New and flexible fiber-optic delay-line filters using linearly chirped Bragg gratings and laser arrays,” IEEE Trans. Microwave Theory Tech., vol. 47, no. 7, pp. 1321–1326, Jul. 1999.Fig. 8. FS-RDL noise spectrum comparison with theory.shown in Fig. 8 indicates that S-SP beat noise is dominant, since the PIIN noise has been suppressed. Finally, to test the PIIN analysis for the FS-RDL, we modified the experimental parameters so that the PIIN noise could be visible. From the simulation shown in Fig. 2, it can be seen that the PIIN noise profile has a characteristic curvature with frequency, and when the S-SP noise is dominant, the PIIN noise only rises above the S-SP noise and becomes visible at frequencies that are close to the frequency shift value. In order to access this region and produce several noise peaks at frequencies close to the frequency shift value of 750 MHz, we reduced the FSR of the FS-RDL. This was achieved by adding a length of optical fiber to the loop to reduce the FSR to 8.5 MHz. The operating value was 100. Fig. 9 shows the measured FS-RDL noise spectrum at frequencies close to the frequency shift value of 750 MHz. The PIIN noise peak profile curvature with frequency can be seen. Also shown in Fig. 9 is the predicted spectrum from the PIIN noise analysis. The parameters used in this simulation were kHz, GHz, and MHz. Fig. 9 shows a comparison between the theoretical PIIN noise。