1. Introduction

In an optical network, optical performance monitoring (OPM) is the key enabling technology for reliable spectrum management. Up-to-date network telemetry is required for capacity scaling, network or component fault recovery, and network reconfiguration through performance prediction and planning. The monitoring includes information on bit-error-rate (BER), optical signal-to-noise ratio (OSNR), electrical signal-to-noise ratio (ESNR), loss, and power. The monitoring information is then provided to the network management agent for resource optimisation to maximise the reach versus rate. In practice, optical performance monitoring may be based on the measurement of one or several parameters. However, OPM is used herein to refer to “power” monitoring since power is one of the key indicators of performance in optical systems.

In transport optics, especially in WDM networks, spectral sensing is not straightforward. Traditionally, WDM channels are distributed over the 35 nm wide fibre-optic C-band (1530 nm to 1565 nm) with fixed centre frequencies arranged on the International Telecommunication Union (ITU) grid at intervals of 50 GHz or 100 GHz and an OPM card is used to measure the power. The working principle of the OPM card involves sweeping a tunable filter with 50 GHz resolution over the spectrum to make available ITU grid channel power readings. Due to their excessive power consumption, size, and cost, OPM cards are deployed only at a few points in the network; typically co-located with reconfigurable add-drop modules (ROADMs). However, current optical networks are elastic in nature, i.e., the channels are not located on a fixed regular grid, rather the channel centre wavelength can be placed at an arbitrary location within the spectrum. The flexible grid can support a variety of channel power profiles (i.e., bandwidth and power spectral density) with a channel centre frequency placement resolution as fine as 6.25 GHz. Flex-ready spectrum measurements are required to facilitate the deployment of the flex-grid system. As a result, flex-grid ready ROADM architectures are equipped with new modules that can measure power at a desired frequency location and resolution. However, due to cost issues, spectrum measurement is only performed at add-drop nodes and not at amplifier nodes. Moreover, a single OPM module is shared by the multi degree-ROADM, so the OPM measurement speed reduces as the number of lines it supports increases. Consequently, the performance of WDM channels in a section (ROADM-to-ROADM) is modeled based on an analytical or a semi-analytical analysis or a machine learning approach. The absence of OPM makes it difficult to have live and accurate network measurement; and hence hard to implement performance optimization. Complete knowledge of spectral content in a network is a prerequisite for the effective use of color-, direction-, contention-, grid-, filter-, gap- less ROADM; flexible modulation formats; flexible channels frequencies; and spectral assignment.

A variety of different approaches to the problem of spectral sensing with high resolution across a wide band have been disclosed in the literature. A recent review by Yang et al [1] classifies integrated spectrometers into four categories. The first category makes use of dispersive optics technology such as an arrayed waveguide grating (AWG) to spatially separate the different spectral components before sampling by a detector array [2]. The second category makes use of a bank of narrowband filters each with its own dedicated detector [3]. The third category makes use an interferometer based on Mach-Zehnder Interferometer (MZI) technology [4] or micro-electro-mechanical systems (MEMS) to produce spatial or temporal interferograms from which the spectrum is calculated via a Fourier transform. The fourth category makes use of computational techniques to reconstruct the spectrum from information encoded within a set of detectors [5] either via complex spectral to spatial mappings or detectors with distinct spectral response. References [2–13] are offered in this work only as examples of the vast prior art that can be mapped into one of these four categories. However, when scaled to combine acceptable resolution (∼ 1 GHz) with wideband operation (entire C-band) the practical implementation of the spectrometer architectures disclosed in the prior art is most often not viable due to excessive cost, loss, and footprint. Firstly, a resolution of ∼1 GHz requires the spectrometer to process a signal segment of at least 1 ns duration in which time the signal travels ∼ 20 cm. This not only requires a low loss platform with waveguide loss < 0.1 dB/cm but also, for compactness, sufficiently strong confinement to avoid bending losses of extensively folded or recirculating paths compromising the low waveguide loss. Secondly, to achieve 1 GHz resolution over a band extending over ∼4400 GHz requires more than 4400 samples. This necessarily requires time division multiplexing of a practical number of photodetectors among the samples. An integrated solution for a high resolution (sub-GHz) spectrometer to monitor the power in fixed- and flex- grid architectures across the entire C band 1530 nm to 1565 nm consequently remains challenging. Nevertheless, a three-stage architecture proposed in a previous publication [14] has been shown to be viable. The first stage is a tunable ring resonator (RR) that defines the resolution. The third stage is an AWG that isolates one RR resonance within each of its channels. The principal innovation is the ganged tuning of the RR and AWG to retain the RR resonance at the center of the AWG channel passband. This is achieved by a second stage that uses an MZI with delay imbalanced arms to form a coherent superposition of two interleaved AWG channel spectra corresponding to a pair of input ports. Further details can be found in [14]. This article describes a refinement of the architecture and method that enables the ganged tuning of the AWG to be virtualized. The coherent superposition of a pair of AWG input channels is replaced by the incoherent superposition of pairs of AWG outputs, which may be performed by processing the measured AWG channel powers. The MZI stage is eliminated, releasing the spectrometer from any requirement to control inter-stage optical path lengths, and thereby significantly easing manufacture.

In this paper, the circuit architecture design, modelling, and integration feasibility of an ultra high-resolution wideband spectrometer based on the refined architecture is presented. The purpose of the proposed circuit is to measure the spectral profile of WDM channels in flex- and fixed-grid architectures across 1530 nm to 1565 nm (C band). The architecture combines a RR and AWG subcircuit with a simple but novel information processing method that enables the spectrometer to scan the entire C-band with high resolution (∼1.4 GHz) using only one dynamic control. A three AWG configuration offers substantially zero adjacent channel leakage but for clarity of exposition this article focusses on the two AWG minimum configuration. Hardware economies may be made by replacing multiple AWG by a single time multiplexed multiple-input AWG. Details of the three AWG configuration may be found in Ref. [15,16].

The proposed spectrometer can be fabricated using any suitable photonic integrated circuit fabrication platform. However, the resolution bandwidth of the spectrometer primarily depends on the RR waveguide loss (dB/turn). Hence to meet the specification of the proposed spectrometer, the CMOS compatible Si₃N₄ photonic integration platform has been selected as it offers low loss, tight confinement, low dispersion, and a mature thermo-optic phase shifter technology. There are ample reports in the literature of the technological verification of all components required by the spectrometer fabricated using the Si₃N₄ integration platform [17–19], including a high port count AWG [20]. In this report, the detailed simulation and experimental verification of the proposed architecture is presented. Spectrometer operation with high resolution is demonstrated with the implementation of a ring resonator with ∼1.4 GHz bandwidth and two interlaced cyclic AWGs. The only dynamic control needed to scan the whole C-band is the tuning of the ring resonator over its FSR of 50 GHz. A virtual channel synthesis algorithm is applied to achieve a flat scanning response and to suppress crosstalk to some extent. Owing to the information processing method, the operation of the circuit is invariant to the optical path length between individual components substantially improving robustness to fabrication process variations.

2. Operating principle

The spectrometer architecture consists of a first stage fine filter followed by a second stage coarse filter. The fine filter is implemented using a ring resonator (RR) which generates a comb of regularly spaced narrow bandwidth resonances. The RR frequency response is periodic with period known as the free spectral range (FSR). The bandwidth of an individual RR resonance defines the resolution which should be to meet the target specification. An intra-ring phase shifter is used to tune the RR so that the comb of resonances is translated in frequency over one FSR. This ensures that spectral information is collected over a continuous frequency interval spanning the whole C-band.

The task of the coarse filter is to isolate an individual RR resonance among the multitude of resonances of the comb. This task requires the coarse filter to substantially extinguish all adjacent RR resonances while substantially transmitting the individual RR resonance irrespective of the frequency to which it is tuned. Since the comb of RR resonances is translated in frequency by the tuning mechanism over an interval $f \in {({ - \varDelta f} ,\varDelta f} ]$ spanning one FSR, to satisfy both these requirements, a fixed frequency ideal filter must have unit transmission over $ {({ - \varDelta f} ,\varDelta f} ]$ and zero transmission over the complement of this interval. A practical fixed frequency filter cannot meet this requirement and would not only introduce undesirable attenuation within the passband but also introduce unacceptably high adjacent channel leakage at the limits of the tuning range. It is necessary that the coarse filter be tuneable so that it can maintain the wanted resonance within a flat region of its passband and unwanted adjacent resonances within its stop band.

The analysis of virtual channel synthesis that follows is applicable to pre-detection (optical) processing, in which case all quantities are complex, and post-detection (electronic) processing, in which case all quantities are real and positive. The second stage of the architecture synthesizes a tuneable coarse filter channel by forming a weighted sum y with weights ${w_j}$ of the outputs ${x_j}$ of a collection of physical channels indexed by j with fixed centre frequency:

The physical channels are provided by one or more arrayed waveguide grating (AWG) devices which together offer a channel frequency spacing that is equal to the RR FSR divided by an integer $N \ge 2.\textrm{}$Collectively, the passbands of the physical channels cover the entire C-band so that each RR resonance falls within the passband of at least one physical channel. Moreover, if a physical channel has a RR resonance falling within its passband, all other RR resonances fall within the stop band of that channel. The values of the weights are chosen to meet the two requirements specified in the preceding and consequently are a function of the tuning frequency. Following similar reasoning to maximal ratio combining [21], the weights are set equal to the complex conjugate of the transmission function of their associated physical channel which results in a synthesised channel transmission function:

If the physical channel transmission falls rapidly outside its passband, only the pair of overlapping channels with centre frequencies adjacent to the RR resonance frequency contribute significantly to the weighted sum. The worst-case adjacent channel leakage occurs when the RR resonances frequency is midway between adjacent physical channel passband centre frequencies, i.e., the adjacent ring resonance is detuned from the passband centre of the contributing physical channels by a multiple $2N - 1$ of the passband half-width. For a Gaussian channel profile and a passband width specified by the FWHM then the adjacent channel leakage is suppressed by a factor of ${2^{{{({2N - 1} )}^2}}}$ or $27\textrm{}dB$ for $N = 2$ and $75\textrm{}dB$ for $N = 3$. The adjacent channel leakage consequently improves rapidly with N. However, the number of photodetectors required also scales with N. Consequently, the focus of this paper is on the $N = 2\textrm{}$ case. The adjacent channel leakage for $N = 2$ can be improved by slight reduction in the $- 3dB$ bandwidth of the physical channels.

In prior work [14], a method was proposed for forming the weighted superposition of the fields generated within an AWG by a pair of input access waveguides driven by a delay imbalanced Mach-Zehnder interferometers (MZI). Since the processing occurs in the optical domain the tuneable coarse filter may directly drive further optical stages which may be advantageous in some applications. The coherent summation however requires tight control of the phase of the field and hence close integration of the MZI, and AWG is necessary to ensure optical path lengths are matched. The requirement to control the phase also introduces the requirement to switch between two pairs of input waveguides each driven by its own MZI (see Ref. [14] for details).

In the case of the spectrometer application the weighted superposition may be performed after photodetection. All variables and functions in the weighted sum are then real-valued and positive. This eliminates any requirement to control path lengths exterior to the RR and AWG components. The complete architecture is thereby rendered robust to fabrication process variations. It is even possible to use an optical fibre to interconnect RR and AWG components implemented in different material platforms best suited to the component such as Si₃N₄ for the RR and doped-silica for the AWG, albeit a goal of this work is single-chip implementation.

There is some flexibility in the choice of the dependence of the weights as a function of tuning. The weight functions only need to approximate their associated channel profiles accurately over their passband and may rapidly and smoothly decay towards zero outside their passband. This expedient to some extent suppresses the sidelobe structure of practical AWG channel profiles. However, it cannot suppress crosstalk introduced by the principal contributing channel, i.e., a low level of spurious lobes within one RR FSR from the main lobe passband of the channel is necessary to avoid adjacent channel leakage.

One embodiment of the proposed $N = 2$ AWG architecture is shown schematically in Fig. 1(a) as an example. A RR first stage provides the fine filter while the second stage AWG pair provide the coarse filtering. The RR generates a train of narrow resonances spaced in frequency by its free spectral range (FSR), while the AWG pair isolates each RR resonance in one or other or both of their output channels. Hence, the output channel frequency spacing of each AWG must be equal to the RR FSR by design. The RR defines the spectrometer resolution bandwidth and is tunable in frequency over one FSR. A nominally identical AWG pair with $- 3\; dB$ channel passband-width close to half their output channel frequency spacing is required for this example. However, the AWG pair are not driven by the same input channel (port), rather they are driven by adjacent input ports as shown in Fig. 1(a). The input channel frequency spacing is equal to half of the output channel frequency spacing (i.e., $1/2$ FSR). The channel spectra of AWG₁ and AWG₂ thereby are interlaced and overlap as illustrated in Fig. 1(b) for a 50 GHz channel frequency spacing. For clarity, only three of the channel spectra of each AWG are shown with possible mappings of the comb of RR resonances. In practice, 50 GHz AWGs require 88 channels to cover the whole C-band.

 

Fig. 1. (a) Schematic of the proposed spectrometer; (b) interlaced optical spectrum of AWG₁ and AWG₂ with the resonance mapping over one FSR. AWG, arrayed waveguide grating; RR, ring resonator; DC, directional coupler; DA, detector array; DAQ, data acquisition Ch; channel; FSR, free spectral range. Green and purple stems are displaced slightly for better visualization. The green stem and purple stem denote the RR tuning of $\theta = 0$ to $\pi $ and $\theta = \pi \; $ to $2\pi $ respectively.

Download Full Size | PDF

The comb of RR resonances is tuned by an intra-ring phase shift $\theta $. The translation in frequency of the comb is proportional to the phase shift and ranges over one FSR as $\theta $ ranges over $2\pi $ radians. For simplicity of exposition, $\theta = 0$ is taken to correspond to the alignment of the RR resonances with AWG₁ channel passband centre frequencies. It follows that $\theta = \pi $ corresponds to the alignment of the RR resonances with AWG₂ channel passband centre frequencies and $\theta = \pi /2,3\pi /2$ corresponds to alignment respectively with the intersection between the upper (lower) AWG₁ -3 dB channel passband edge and the lower (upper) AWG₂ -3 dB channel passband edge.

The optical power of each output channel of both AWGs is measured by a photodetector array while the RR is scanned over an FSR under the control of the data acquisition system (DAQ). The principal novelty of the spectrometer is the construction of a virtual AWG that is tuned to retain the ring resonance within the passband of its synthesised channels. A virtual channel with index m is synthesised by the weighted sum of the optical power of the interlaced AWG pair. The weights depend on the frequency to which the RR is tuned or equivalently the tuning phase. For simplicity of exposition, it is convenient to use a Gaussian weighting function with the same FWHM as the AWG channel profile.

As the ring resonance is tuned over the first half of the FSR by a tuning phase shift from $\theta = 0$ to $\theta = \pi $ (green stem in Fig. 1(b)), AWG₁ channel m (Ch₁ m) and AWG₂ channel m (Ch₂ m) make the most significant contributions to the weighted sum. The contribution by Ch₁ m (Ch₂ m) is weighted essentially by unity (zero) at $\theta = 0$ and by essentially zero (unity) at $\theta = \pi $. At $\theta = \pi /2$ the contribution of Ch₁ m and Ch₂ m are equal, and each is weighted by one half. Similarly, as the ring resonance is tuned over the second half of the FSR by a tuning phase shift from $\theta = \pi $ to $\theta = 2\pi $ (purple stem in Fig. 1(b)), AWG₂ channel m (Ch₂ m) and AWG₁ channel $m + 1$ (Ch_1(m + 1)) make the most significant contributions to the weighted sum. The contribution by Ch₂ m (Ch_1(m + 1)) is weighted essentially by unity (zero) at $\theta = \pi $ and by essentially zero (unity) at $\theta = 2\pi $. At $\theta = 3\pi /2$ the contributions of Ch₂ m and Ch_1(m + 1) are equal, and each is weighted by one half.

The normalisation of the weights ensures that the synthesised channels have a perfectly flat passband to the extent that the weighting function closely approximates the channel profile over its passband. The error in the approximation creates some ripple. However, the ripple is small as the synthesised channel spectral profile is in close agreement with the ideal when the ring resonance is aligned at either passband centre or at the intersection of the -3 dB passband edges of the AWG channels summed. The data processing is performed in the electronic domain by the data acquisition system which also controls the RR tuning phase shifter. The construction of the weights requires knowledge of the RR resonance frequency or a stable calibration of the phase shifter characteristics. Precision tracking of the position of a resonance is not mandatory for successful virtual channel synthesis given the wide AWG passband relative to the RR bandwidth. Nevertheless, an integrated wavelength meter concept presented in [22] can be used to monitor the position of the RR resonance within a single channel of the AWG. Since the resonances are substantially periodic in frequency, albeit very slightly detuned by chromatic dispersion, the position of the resonance within each channel of the AWG can be determined easily.

The spectrometer requires only one control, which sets the intra-ring phase shift to tune the RR resonant frequency cyclically over one FSR. Device and circuit simulations; previously reported experimental demonstrations; and the process development kit; support the practicality of a 50 GHz FSR RR. Table 1 provides the specifications of the proposed circuit design. For a RR FSR of 50 GHz, the required number of AWG output ports is 88 to cover the entire C-band. The number of output ports can be reduced by increasing the FSR of the RR, since AWGs with up to 96 50-150 GHz ports are available commercially [23]. On the other hand, a RR having FSR of 220 GHz fabricated using double strip TriPleX waveguide technology is already reported in [8]. Hence, the number of output channels of the AWG required can be reduced by a factor of 3 (∼30 channels).

Table 1. Design specifications of the proposed circuit architecture shown in Fig. 1(a)

View Table | View all tables in this article

The VPIphotonics simulation tool is used to evaluate the performance of the combined circuit architecture and data processing method. The optical power at each output channel of both AWGs is monitored while scanning the RR over one FSR. The calculation of the power depends on the number of samples used in the simulation time window. Since the signals were widely spread, a moderate time window is used in the simulation to avoid an excessive memory requirement. A slight variation in the result is obtained for different time window settings. The simulated output is then processed according to the described algorithm. Figure 2(a) shows the variation in the measured total optical power (arithmetic sum of AWG₁ and AWG₂) as a function of the resonance position over one FSR for various AWG channel passband widths. Channel 1 & 2 of AWG₁ and channel 1 of AWG₂ are used for the calculation. The results show that the spectral measurement is almost flat with little ripple (∼0 dB) over the entire FSR for a passband width of 25 GHz. Figure 2(c[i-iv]) shows the optical spectrum of AWG₁ and AWG₂ at output channel one (Ch₁₁ & Ch₂₁) for ring resonance position of 0° and 90° respectively for a passband width of 25 GHz. It shows that a strong (∼-21→-22 dB) adjacent ring resonance component is present in the AWG₂ channel. Figure 2(c[v-viii]) shows the optical spectrum of AWG₁ and AWG₂ at the same resonance position for the pass band width of 20 GHz. The adjacent resonance cross talk is reduced to ${\sim}{-} 30$ dB while maintaining minimal ripple. Figure 2(b) shows the transmission spectra using weighted superposition method. As described earlier, this method can eliminate the ripple found in Fig. 2(a) along with improved crosstalk. Since the transmission characteristic of an AWG is substantially periodic, the simulated result will almost be same for any other channel measurement of the AWG. To substantially eliminate adjacent channel leakage, the architecture can further be upgraded to a $N = 3$ AWG configuration. The details can be found in Ref. [15,16].

 

Fig. 2. Simulated optical power transmission as a function of ring resonance for different passband width of the AWG channel, (a) without weight summation, (b) with weight summation for passband width of 25 GHz; (c) optical spectrum of AWG₁ and AWG₂ at channel one for ring resonance position of 0° and 90° for the passband width of 25 GHz [i-iv] and 20 GHz [v-viii] respectively.

Download Full Size | PDF

3. Integration feasibility & experimental setup

The proposed spectrometer can be fabricated in any mature low-loss photonic integration platform with sufficient index contrast to support ring resonators. If the excess loss of the ring per turn is negligible in comparison to the power coupled out per turn, the resolution bandwidth of the spectrometer is determined by the power cross-coupling ratio of the couplers. The ring excess loss per turn consequently limits the achievable resolution. An integration platform supporting the design of low-loss waveguides and low-loss waveguide bends is therefore paramount. Owing to its low loss, tight confinement, low dispersion waveguides and a mature thermo-optic phase shifter technology, the CMOS compatible Si₃N₄ photonic integration platform offered by LioniX International was selected to meet the specification of the proposed spectrometer circuit. The platform also offers good prospects for further loss-reduction [24] and to lower power consumption, temperature insensitive, alternatives to thermo-optic phase-shift elements [25,26].

For effective use of resources, the fabrication plan envisaged multi-project wafer runs (MPW) for test structures followed by a custom wafer fabrication run for prototypes for demonstration. The LioniX MPW runs supports designs using the asymmetric double strip (ADS) waveguide only and the low-cost photolithography used has a minimum feature size of 1 $\mu $m. Accordingly, the components and sub-circuits that constitute the proposed spectrometer are designed and simulated using ADS as the reference waveguide. The waveguide characteristic over full C-band is obtained by using the Photon Design software tool FIMMWAVE. The TE-like mode is used in all the simulation due to its tight confinement, hence it exhibits lower bend loss in comparison to the TM-like mode. The effective group index of the mode at the smaller wavelength edge (1530 nm), centre wavelength (1550 nm), and the longer wavelength edge (1565 nm) of the C-band is found to be 1.7725, 1.76841 and 1.7629 respectively.

Figure 3(a) shows the micrograph of a RR fabricated via a LioniX MPW run. The resolution of the spectrometer determined by the bandwidth of RR which is largely set by the cross-coupled power, and hence the spatial gap between the access and ring waveguides. Figure 3(a) shows that a straight access waveguide is used. To determine the gap, a custom method based on local bisymmetry and adiabatic mode evolution throughout the coupling region between access and ring waveguide has been adopted [27]. The method utilizes a simple mode solver to calculate the effective index difference of the two fundamental symmetric and antisymmetric local eigenmodes of the coupled region. A parameter set is extracted from these data to determine the relationship between the cross-coupled power and the minimum gap size. To design a ring resonator with <1 GHz bandwidth tunable over a 50 GHz FSR, the method proves to be simple, fast, and less computationally resource-hungry compared to other simulation platforms such as 3D- FDTD. For the MPW run, several ring resonators were designed and laid out on the mask having gaps from 1.2 µm to 1.8 µm with 0.2 µm increment with the objective that at least one RR works well. For the experimental results discussed in this report, a fabricated RR with 537 µm radius and 1.2 µm spatial gap between straight access and curved ring waveguide is used. The bend loss of an ADS waveguide of this radius of curvature is negligible over the C-band; the mode is fully bound and only absorption and scattering contribute loss. Detailed data on absorption and scattering loss is not available beyond a specification of a total straight waveguide loss <0.5 dB cm^-1 with the expectation of a typical value of 0.1-0.2 dB cm^-1. Using a parameter extraction method reported in [27], it can be conjectured that the RR used has a straight waveguide loss well below the upper limit.

 

Fig. 3. (a) Micrograph of the ring resonator equipped with a thermo-optic phase shifter covering almost the whole circumference of the ring waveguide; (b) Micrograph of arrayed waveguide grating (AWG) with 8 input channels and 32 output channels; (c) the experimental setup for recording the optical spectrum; (d) Schematic diagram of the experimental setup.

Download Full Size | PDF

Figure 3(b) shows the micrograph of the fabricated AWG. Bright Photonics BV provided the design which was fabricated via the LioniX MPW process. Two identical cyclic AWGs with 32 output channels have been employed in the experiment. The detailed design specifications are given in Table 2.

Table 2. Design specification of AWG₁ and AWG₂

View Table | View all tables in this article

Figure 3(c) depicts the experimental setup implemented for the spectrometer operation. A tunable laser (Agilent 81680A) capable of tuning over the whole C-band is employed as the input source. An optical component analyzer (Agilent N7788B) is connected to the ring resonator to maximize TE mode transmission. The input and output of the RR and AWGs are packaged with polarization maintaining (PM) fiber. The tuning of the RR is controlled by a precision source/measure unit (Agilent B2912A). The output from the RR is split by a 50:50 polarization maintaining splitter and fed into the two AWGs. The output is detected by an optical power sensor (Agilent 81632A) and analyzed by a lightwave measurement system (Agilent 8164A).

4. Experimental validation & discussion

Figure 4(a-c) shows the measured add-drop transmission spectra of the RR at various wavelengths across the C-band. The peak transmission of the RR varies from 1.25 dB at the longer wavelength edge to 1.75 dB at the shorter wavelength edge of the C- band. Figure 4(d) shows the zoomed-in view of the ring resonance at the design wavelength. The full width half maximum (FWHM) bandwidth of the RR is found to be ∼1.4 GHz. The FWHM bandwidth at the edges of the C-band is found to be within ${\sim}{\pm} $40 MHz from the resolution bandwidth mentioned at the center of the C band. Figure 4(e-f) shows the measured FSR at the center wavelength and longer wavelength edge. The FSR is ∼48.70 GHz at the short wavelength edge. The FSR variation is due to the variation of the group index across the band from the group index at the design wavelength, 1550 nm. The relative detuning between the RR and AWG is slight and may be easily compensated by digital signal processing to provide the quantitative spectrum provided the RR is scanned over a sufficient interval to ensure there are no spectral gaps in the measured data.

 

Fig. 4. Transmission spectrum of the RR obtained from fabricated chip using laboratory measurement; (a) smaller wavelength edge; (b) center wavelength; (c) longer wavelength edge; (d) zoom-in view of the ring resonance at the design wavelength. The measured full width half maximum (FWHM) is ${\sim} 1.4$ GHz; the FSR at the (e) center wavelength edge and (f) longer wavelength edge.

Download Full Size | PDF

The RR is tuned using an intra-ring phase shifter. Only thermo-optic phase shifters are offered by the LioniX MPW process. Figure 5 (a) shows the tuning of the ring resonances as a function of applied voltage to the thermo-optic phase shifter. Tuning over a complete FSR is achieved. Figure 5 (b) shows the I-V characteristic of the heater, while the frequency of the resonances as a function of heater power shown in Fig. 5(c). The phase shifter heater resistance is found to be 734 Ω by direct measurement, which is well aligned with the slope (∼730 Ω) of the I-V curve. The tuning wavelength (frequency) is found to be linearly proportional to the applied heater power as expected. However, electro-optic tuning offers smooth operation with better efficiencies [28], due to a linear voltage to index relationship and low drive voltage & power requirements. The use of piezo-electric actuators as an alternative means of providing an adjustable phase shift on the Si₃N₄ platform, augurs well for the future [26].

 

Fig. 5. (a) Experimental demonstration of the tuning of the ring resonance as a function of applied voltage to the thermo-optic phase shifter; (b) The I-V characteristic of the thermo-optic phase shifter and (c) Peak resonances in a particular FSR as a function of heater power.

Download Full Size | PDF

Two 32-channel cyclic AWGs designed by Bright Photonics BV to the specification given in Table 2 were fabricated. The AWG designs (AWG₁ & AWG₂) are identical. The constraint of a relative frequency shift between AWG₁ and AWG₂ channels being half (25 GHz in this case) of the AWG output channel spacing can be fulfilled by selecting proper input and output port combinations for the individual AWG at room temperature. The design suggests that selecting an adjacent input port to the nominal input port should achieve the 25 GHz relative frequency shift of the channel spectra. In case of slight detuning, one of the AWGs can be temperature tuned to obtain the required frequency shift. Once the input-output port combinations and, if necessary, the required relative temperature tuning are known and entered into DAQ processor, incoherent superposition of pairs of AWG outputs can be controlled indefinitely.

Figure 6(a) shows the interlaced AWG₁ and AWG₂ channel spectra when light is launched from input port 4 of AWG₁ and input port 5 of AWG₂. Three channels of the respective AWGs are considered for better presentation. The wavelength resolution is 1 picometer for these data acquisitions. It can be observed that the central frequency difference (∼47.5 GHz around 1550 nm) between two output channels of same AWG deviates from the design specification (50 GHz) and is subject to group velocity dispersion when the whole C-band is considered, which leads to the variation in the relative frequency shift between adjacent channels. As mentioned, the relative placement of the center of channels can be controlled by a coarse temperature tuning applied to one AWG, while maintaining the other AWG temperature fixed. Figure 6(b) shows that a 2°C increment of the AWG₁ temperature achieves more uniform channel spacing than in Fig. 6(a). According to the design specifications, interlaced channel spacing equal to half of the AWG output channel spacing and AWG output channel bandwidth less than AWG output channel spacing favour the achievement of low adjacent channel leakage. However, the long-arrayed waveguides of an AWG with 50 GHz channel spacing are vulnerable to fabrication process variations. The adjacent channel cross talk mainly depends on the phase errors in the arrayed waveguide sections. Distortion of the channel profile and significant sidelobe from nearest and next to nearest neighbouring channels are apparent. It is clear that the channel spectra of the fabricated AWGs are process limited.

 

Fig. 6. (a) Interlaced spectrum between AWG₁ and AWG₂ before coarse temperature tuning. Three adjacent output ports of each AWG are used to extract the spectra of six channels The experimental setup is similar to the one shown in Fig. 3(c) except for the presence of the chip containing the ring resonator. The outputs of the polarization maintaining 50/50 splitter are connected to the input port 4 of AWG₁ and input port 5 of AWG₂; (b) Interlaced spectrum between AWG₁ and AWG₂ after coarse temperature tuning. Four adjacent output ports of each AWG are used to extract spectra of eight channels. The temperature of AWG₁ is increased by 2°C by placing it on a hot plate.

Download Full Size | PDF

Figure 7(a) demonstrates the interlaced optical spectra detected at six channels consisting of Ch₂₂, Ch₂₃, and Ch₂₄ from AWG₁, and Ch₂₁, Ch₂₂ and Ch₂₃ from AWG₂. The circuit setup is now completed as Fig. 3(c). The ring resonator is introduced before the polarization maintaining splitter and DC voltage with discrete variation has been applied for tuning the ring resonator. Each resonant peak observed within each channel has been achieved by tuning the ring resonator. To resolve any ambiguity with a static presentation of the optical spectra in Fig. 7(a) each resonant peak is labelled by its corresponding tuning voltage applied to the ring resonator phase shifter. Tuning over one FSR of the ring resonator is examined here by changing the voltage discretely from 0 V to 19 V. With each discrete tuning condition, data has been recorded. The ring resonator has a periodic chain of resonances spaced by ∼50 GHz, and each AWG channel isolates one resonance within its output channel spacing (in the ideal case, 50 GHz). When the tuning voltage changes as shown in Fig. 7(a), the resonance also shifts in a dynamic fashion, and so each AWG channel isolates only that resonance within its output channel spacing for that tuning condition. Figure 7(a) also shows how the ring resonator resonance tracks the AWG spectral shape. It also shows the resonances tuned by any specific voltage are separated by ∼50 GHz, the same as the FSR of the RR. It can be conjectured from Fig. 6 that significant crosstalk at one channel from the sidelobes of another will be observed. Six channels are considered again with the selected tuning conditions shown in Fig. 7(b) to isolate the resonances located around the center and around the point where the transmission is 3 dB down from the maximum of each channel. Significant crosstalk can be observed. Although in the ideal case, the centres of these channels should have minimal intercept from adjacent channels, it can be observed that all channel suffers from crosstalk from nearest and next to nearest neighbouring channels of the same AWG. RR resonances located at the center of each channel due to the application of 6 V for AWG₁ and 15 V for AWG₂ show a crosstalk level of ∼8 dB – mostly due to the significant sidelobe resonances excited by the same tuning condition. The crosstalk from adjacent channels from the other AWG is also present, however they are more prominent in the resonances occurring at the edges of the channels as shown in Fig. 7(b) which is expected as overlapping between two adjacent channels occurs.

 

Fig. 7. (a) Experimental results of a full scanning cycle of the spectrometer around 1550 nm. The interlaced scanning spectra shown are taken from three output channels of each AWG constituting 6 channels (Ch₂₂, Ch₂₃, and Ch₂₄ from AWG₁ and Ch₂₁, Ch₂₂ and Ch₂₃ from AWG₂). Scanning is executed by discretely tuning the ring resonator over one FSR. Each resonant peak is numbered by its corresponding tuning voltage; (b) Zoom-in view of the spectra detected from 6 channels at selected tuning conditions applied to ring resonator which excites the resonances at around the center and 3 dB point of each channel.

Download Full Size | PDF

A custom fabrication run using a stable process is necessary to achieve experimentally the full potential of the spectrometer. However, since the AWG chips were fabricated, the LioniX MPW process has been updated from contact lithography to stepper lithography which should substantially reduce arrayed waveguide phase errors. These errors may be further reduced by increasing the channel frequency spacing from 50 GHz. A ring resonator having FSR of 220 GHz using Si₃N₄ has already been reported in Ref. [8]. For an FSR of 200 GHz, only 22 AWG channels are required to cover the whole C-band. This leads to a compact design offering improved crosstalk performance. For example, an adjacent channel crosstalk of ${\sim}{-} 22\; dB$ is obtained for a $16 \times 16$ cyclic AWG using high contrast silicon photonics having 189 GHz output channel spacing [29]. On the other hand, an adjacent crosstalk of $- 18\; dB$ is obtained back in 1992 [30] for $15 \times 15$ multiplexer having 87 GHz output channel spacing using InP technology. A crosstalk of $- 13\; dB$ is obtained for 200 GHz output channel spacing using Si₃N₄ technology [31]. Furthermore, AWG based on Si₃N₄ platform having adjacent channel crosstalk better than -39 dB with a channel spacing of 3.09 nm has been reported [32]. Crosstalk can further be improved using high resolution optical lithography [33].

The virtual channel synthesis procedure is performed offline after recording the optical spectra of 18 physical channels (Ch₂₀ - Ch₂₈ from AWG₁, and Ch₁₈ - Ch₂₆ from AWG₂) while scanning the RR over one FSR. Due to the process limited non-ideal impairments discussed above, many channels should be incorporated to map the instrument properly. It has been observed that 18 channels are adequate to achieve instrument response over a 150 GHz span around 1550 nm with sufficient accuracy, and thus to evaluate the performance of the synthesis method. As mentioned in section 2, after applying weights to the spectral profile of the associated physical channel, incoherent superposition of all these weighted channels achieves the synthesised channel transmission, which can be observed in Fig. 8 for selected tuning voltages applied to the ring resonator. A Gaussian weighting function with the same FWHM as the AWG channel is implemented for this purpose. The normalized synthesized channel transmission is plotted in linear scale to indicate the peak height fluctuation, which is correctable by a more precise calibration.

 

Fig. 8. Scanning results obtained by the spectrometer after the application of Gaussian weights to 18 channels and their incoherent superposition. The scan is performed by tuning the ring resonator with the applied voltage of (a) 0 V, (b) 7 V, (c) 10 V, (d) 12 V, (e) 14 V, (f) 16 V, (g) 18 V, and (h) 19 V at the thermo-optic phase shifter.

Download Full Size | PDF

Figure 8 is a static representation of the dynamic nature of a spectrometer scanning over 50 GHz around 1550 nm. Each figure shows the power spectral response of the synthesised virtual channel. With the discrete variation in the tuning voltage applied to ring resonator, the narrow resonant peak translates over the 50 GHz span. The solid line is used to indicate the present response of the tracking filter, while dotted responses are retained to illustrate the translation of the resonant response with the scanning of the ring resonator. In a power spectral analysis application, the power spectrum of the input multiplied by the response shown in each individual subfigure and integrated over frequency provides the measured output at that specific tuning frequency. The Gaussian weights are chosen such that when that tuning is equal to the peak of a physical AWG channel the associated photodetector has a high weight compared to adjacent channels. When tuned between a pair of physical AWG profiles, the weights of these two channels become comparable. If a spurious peak from a distant channel, which can be observed in Fig. 7, occurs within the channels that compose the virtual channel with large weights, then the spurious peak will be strongly suppressed because its photodetector has low weight at the virtual channel. The weights thus act to reduce the ripple in the instrument response and reduce the effect of crosstalk at the tuning frequency from distant channels. However, spurious peaks can be observed in Fig. 8 which are loosely located at 50 GHz apart from the virtual channel. The sidelobes observed in Fig. 6 cause their appearance. They can compromise the measurement at a synthesized virtual channel tuning frequency because of potential leakage into photodetector of that channel via these spurious peaks. To minimize these limitations, the AWG channel profile should approximate the ideal quadratic function in the passband, and zero outside its channel spacing.

For comparison, a rectangular weight function has also been applied for synthesising the virtual channel. A rectangular weight function is the ideal option for the synthesis process if the AWG channel profile follows the quadratic function. The weight is constant over the channel profile with a hand-over operation when tuned between a pair of AWG profiles. The constant weight for one AWG profile can be different from others to ensure a flat scanning response. Figure 9 shows a comparison between virtual channel synthesis method employing Gaussian and rectangular weight functions. The objective is to achieve a flat scanning response over the desired frequency span via signal processing so that the requirement of a flat passband for the AWG channel profile can be relaxed. It can be observed that for a rectangular weight function, the ripple in the scanning response can extend beyond ± 1.2 dB over the 150 GHz span shown in Fig. 9. The Gaussian weight function limits the ripple within ± 0.6 dB. From the figure, it can also be conjectured that the variance of the response from the unit transmission due to the ripple is less for the Gaussian weight function. Any hand-over operation is automated in the synthesis process utilizing Gaussian weights. The data processing is slowed compared to rectangular weight, however, a moving weighted average to limit the number of terms in the weighted sum can be used for speed. In section 2, for the simplicity of exposition, the peak of the ring resonance is aligned with the peak of the AWG channel spectra at zero intra-ring phase shift. This can also be relaxed by the weighted virtual channel synthesis method. Ideally the output channel frequency spacing of each AWG must be equal to the RR FSR. However, in practice there will be some offset (< 1 GHz) between the AWG channel centre frequencies and the reference (zero-detuning) grid of RR resonances especially at the edges of the C-band. The proposed algorithm is robust to this frequency offset impairment without any special measures provided the offset is small compared to the AWG channel bandwidth which experimentally is confirmed to be the case. Moreover, if the frequency of the resonance within each channel is known and the frequency scan is of sufficient extent that each AWG passband is covered, then even a substantial offset can be compensated in the data processing preserving the frequency precision of the spectral sample. One approach is to measure the position of the RR resonance in a redundant channel at each edge of the C-band using, for example, using a wavelength meter [22] and interpolate between these measurements to infer the position of the RR resonances in all channels.

 

Fig. 9. Comparison of the ripple characteristics of the scanning profile detected by photodetector when two different weights (Gaussian and rectangular) are used for the virtual channel synthesis.

Download Full Size | PDF

For a ring resonator having FSR of 50 GHz, an AWG of 86∼88 channels are required to cover the whole spectrum. The 32- channel AWG design presented here can easily be scaled up to 86 or 88 channels since commercial AWG with a higher number (96) of output channels are already available [23]. Since the designed 32-channel AWG is cyclic, the output spectrum is periodic with a period of 32. Hence, with the aid of a tunable optical bandpass filter the spectrum of the whole C-band can be measured to prove the concept.

The industry-specified target repetition rate is 10 Hz. Due to the inherent parallelism of the architecture, a scan of the complete C-band only requires the RR to be scanned over only one FSR. This implies tuning the ring resonance by 50 GHz in 0.1 s or 500 GHz/s. So, one would tune over the 1.4 GHz width of the resonance in 2.8 ms which is 2,000,000 times slower than the photon lifetime (∼1.43 ns). So, the envisaged tuning is quasi-static in respect of the photon dynamics. COMSOL thermal simulation models predict a settling time ${\sim} 0.2\textrm{}ms$. Experiments with the weighted superposition of decaying exponentials expected from solutions to the heat equation were robust to the long tail caused by the addition with equal weight of a term with ten times the time constant. Subject to up-down heater power ramps (which result in two scans of the whole C-band per cycle) the tuning response is useable at a repetition rate of 500 Hz and almost perfect at 100 Hz; a margin by a factor of 200 from the specification.

5. Conclusion

In summary, a simple circuit architecture for on-chip spectral monitor with high resolution is presented. The newly proposed data processing method and feasibility for photonic integration places the spectrometer in the forefront of the state of the art. The CMOS compatible Si₃N₄ platform is selected for fabrication due to its low loss and maturity. Experimental verification of a spectrometer operation incorporating a ring resonator and two interlaced AWGs has been demonstrated exploiting discrete components but fabricated using same material platform. High resolution (∼1.4 GHz) scanning has been achieved over 1600 GHz span centered at 1550 nm, which can easily be extended to whole C-band via exploiting the cyclic AWG response. Despite the limited performance of the AWG mostly due to the limitations of MPW fabrication, the adoption of the virtual channel synthesis data processing has resulted in correct operation of a complete spectrometer system. The incoherent summation technique employed in virtual channel synthesis requires a less compact chip but is more robust to fabrication process variations than the coherent summation architecture proposed previously. Based on the selection of proper weights, the spectral sensing can be performed with detection sensitivity with minimal ripple relative to a flat response. The capability of virtual channel synthesis data processing method accompanied by custom fabrication will offer compact single-chip realization of the spectrometer system with experimental results that closely reproduce the predictions of simulations.