Filter Bandwidth Definition of the WaveShaper S-series Programmable Optical Processor Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors Section 1: Introduction The WaveShaper family of Programmable Optical Processors allow creation of user-customized filter profiles over the C- or L- band, providing a flexible tool for industrial or research laboratories. Bandpass filter profiles can be created with a bandwidth ranging from 10 GHz to 5 THz, in 1 GHz steps. However, upon measuring the spectral response of the filter, users can be confused by the bandwidth definition, which conforms to neither a full-width half-maximum (FWHM) nor a 1/e measurement. Furthermore, an inspection of the shape of a nominally flat-top filter shows that there is a finite slope to the roll-off edge of the filter. This can limit the adjacent channel extinction, particularly for very narrow filter bandwidths. What does the specified bandwidth refer to - and is it constant for all devices and channels? 1 Introduction 2 Theory 3 Experimental Results 4 Conclusion This white paper aims to provide the reader with an understanding of how the Finisar WaveShaper Programmable Optical Processor can generate spectral filters that are defined as having a bandwidth, B. First, a theoretical treatment of the filter shape is presented, with a prediction of how the filter bandwidth should be specified. Second, experimental results are shown which confirm the theoretical analysis. Finally, these findings are summarized and clearly stated for the reader. Page 2 This document aims to provide the reader with an understanding of how the Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors Finisar WaveShaper generates spectral filters that are defined as having a bandiwidth, B. First, a theoretical treatment of the channel shape is presented, with a prediction of how the channel bandwidth should be specified. Second, experimental results are shown which confirm our theoretical analysis. Finally, these findings are summarized and clearly stated for the reader. Theory Section 2: Theory In this section, an expression for the channel shape produced by the Finisar WaveIn this section,along an expression the default, filter shapedefining produced by Shaper tunable filter is presented, with anforanalysis offlat-top appropriately the WaveShaper family of Programmable Optical Filters is presented, along with an the filter bandwidth specification. It should be noted that this analysis is de- analysis to appropriately define the filter bandwidth specification. It should be noted rived for a simple that flat-top filter isresponse by the WaveShaper, with no this analysis derived for produced a simple flat-top filter response with no additional additional amplitude or phase shaping. amplitude or phase shaping. 2.1 Prediction of filter shape Prediction of channel shape An ideal optical filter of bandwidth, B, should look like a rectangular function, An ideal optical filter of bandwidth, B,ideal should look like from -B/2 to B/2. That is, an channel shape canaberectangular expressed as function, from −B/2 to B/2. That is, an ideal channel shape can be expressed as R(f ) = 1, −B/2 ≤ f ≥ B/2 0, otherwise. (1) This rectangular function is generated on the liquid crystal-on-silicon (LCOS) chip, which sits in the Fourier plane of a 4–f imaging system. Propagation through THEORY This rectangular function is generated on the liquid crystal-on-silicon (LCOS) chip, which sits in the Fourier plane of a 4-f imaging system. Propagation through the THEORY optical system can be reduced to an optical transfer function that is composed of a system relatively can be reduced to anspectrum, optical transfer narrow Gaussian given by function that is composed 1 the optical the system can be reducedspectrum, to an optical transfer of optical a relatively narrow Gaussian given by function that is composed of a relatively narrow Gaussian spectrum, given by 2 2 L(f ) = Ae−(f −fo ) /2σ , (2) −(f −fo )2 /2σ 2 , (2) L(f ) = Ae where A is the spectrum amplitude, fo is the filter center frequency and σ 2 is 2 where is the spectrum amplitude, the filter center frequency is 2the center frequency and σ is specwhere A isvariance the spectrum amplitude, fo is thefo isfilter the spectral of Athe Gaussian spectrum. The spectral varianceand is σrelated, tral variance of the Gaussian spectrum. The spectral variance is related, through the the spectral of deviation, the Gaussian spectrum. The spectral is spectrum related, through thevariance standard to the 3 dB bandwidth of avariance Gaussian standard deviation, to the 3 dB bandwidth of a Gaussian spectrum by the following through the standard deviation, to the 3 dB bandwidth of a Gaussian spectrum by the following relation: relation: by the following relation: BW3dB . (3) σ= √ BW 2 3dB 2 ln 2 σ= √ . (3) 2 2 lnfunction 2 The bandwidth of the optical transfer Gaussian is roughly 10 GHz, The suggests bandwidth of the the standard optical transfer function roughly 10 GHz, which that deviation is equalGaussian to 4.2466is GHz. The bandwidth of the optical transfer function Gaussian is roughly 10 GHz, whichThe suggests that the standard is equalistothen 4.2466 GHz. filter shape generated by deviation the WaveShaper given by the convolution which suggests that the standard deviation is equal to 4.2466 GHz. filter(2), shape of The (1) and or generated by the WaveShaper is then given by the convolution of (1) and (2), or S(f ) = R(f ) ∗ L(f ) S(f ) = R(f ) ∗ L(f ) +∞ = +∞ R(f ) L(f − f ) df . −∞ R(f ) L(f − f ) df . = −∞ (4) (4) Page 3 Normalized Normalized powerpower (dB) (dB) . (3) σ= √ The bandwidth of the optical transfer Gaussian is roughly 10 GHz, 2 2 function ln 2 The bandwidth of the optical transfer function Gaussian is roughly 10 GHz, BW3dBof Programmable Optical Processors Filter Bandwidth Definition of the WaveShaper S-series . (3) σdeviation = √ function which suggests that of thethe standard is equal to 4.2466 is GHz. The bandwidth optical transfer Gaussian roughly 10 GHz, which suggests that the standard deviation 2 2 lnis2equal to 4.2466 GHz. Thesuggests filter shape by the WaveShaper is then givenGHz. by the convolution which thatgenerated the standard deviation is equal to 4.2466 The shape generated by the WaveShaper is then givenisbyroughly the convolution The filter bandwidth of the optical transfer function Gaussian 10 GHz, of (1) and (2), or The filter shape generated by the WaveShaper is then given by the convolution The filter shape generated is then given by the convolution of (1) and (2), or of (1) and (2), or which suggests that the standard deviation is equal to 4.2466 GHz. of (1) and (2), or The filter shape generated by the WaveShaper is then given by the convolution of (1) and (2), or S(f ) = R(f ) ∗ L(f ) S(f ) = R(f ) ∗ L(f ) S(f ) = R(f ) ∗ L(f ) Prediction of channel shape (4) +∞ (4) +∞ S(f ) = R(f ) ∗ L(f ) (4) Prediction of channel shape R(f ) L(f − f ) df . = = −∞ +∞ R(f ) L(f − f ) df . = −∞ R(f ) L(f − f ) df . (4) +∞ = 0, A = 1, the integral Substituting (1) and (2) into−∞ (4), and setting f 20 GHz o into and Substituting (1) and Substituting (2) = into (4), and setting = =1,1, the theintegral integral (1) R(f and (2) (4), = 0, becomes ) L(f − ff)osetting df= . 0,fo A 40 AGHz 0 becomes −∞ 60 GHz = 0, A = 1, the integral Substituting (1) and (2) into (4), and setting f 20 o becomes 80 40 GHz 0 becomes 100 GHz 60 GHz B/2 Substituting (1) and (2) into (4), f = 0, A80=GHz1, the integral B/2 and setting -10 −(f −f )22 /2σ 22 o S(f ) = B e−(f −f ) /2σ df . (5) 100 GHz df . (5) S(f ) = − B/2/2 e becomes -10 2 2 − B/2 −(f −f ) /2σ e df . (5) S(f ) = −aBB/variable In order-20to evaluate this integral, substitution is required. Redefining /2 2 In order to evaluate this integral, a variable substitution is required. Redefining 2 −(f integral, −f )2 /2σa InS(f order to evaluateethis variable substitution is required. Redefining df . (5) ) = the In variable u to be equal to -20 order to thisto integral, a variable substitution is required. Redefining the variable u evaluate to bethe equal − B/2 to variable u to be equal the variable-30 u to be equal to In order to evaluate this integral, a variable f − f substitution is required. Redefining −f, (6) u = f√ -30 (6) u= √ 2σ , the variable u to be equal to f − f 2σ -40 u= √ , (6) allows (5) to be rewritten as 2σ allows (5) to be rewritten as f −f -40 (6) u = √ , allows (5) to be rewritten as √ 2σuu22 −u2 -50 allows (5) S(u) to be rewritten = 2√2σas e−u2 du, (7) S(u) = 2√2σ u1u2 e du, (7) allows (5) -50 to be rewritten as 2 -100 -80 -60 -40 -20 0 60 80 100 −u 40 u1 20 S(u) = 2 2σ du, (7) e Frequency offset (GHz) where -100 u u 1 2 -80 -60 -40 -20 √ 0 20 40 60 80 100 where −u2 Frequency offset (GHz) S(u) = 2 2σ e du, (7) Figure 1 – The filter shape, S(f ), calculated for different where where u1 filter bandwidths, B. Figure 1 – The filter shape, S(f ), calculated for different filter bandwidths, B. where 2 2 2 2 −f √ − B/22σ −f (8) u1 = B/2√− f u2 = √ 2σ . (8) B/2 2σ −f . u2 = √ 2σ Evaluation of (7) requires use of the error function, which is commonly defined Evaluation of (7) requires use of the error function, which is commonly defined u1 = − B/2 as Evaluation of (7)asrequires use of the error function, which is commonly defined √ a as π −t2 erf(a) = e dt. (9) √2 0 a π 2 erf(a) = e−t dt. Using (9), the integral in (7) resulting in the final form(9) of 2 can be evaluated, 0 the Using channel shape generated WaveShaper, given (9), the integral in by (7)the canFinisar be evaluated, resulting inby the final form of WaveShaper, − B givenby the channel shape generated √ by theBFinisar /2 − f /2 − f − erf √ (10) S(f ) = σ 2π erf √ . B/2 2σ − B/22σ √ −f −f √ √ − erf . (10) S(f ) = σ 2π erf Examples of this filter shape are shown 2σ in Figure 1, S(f 2σ) calculated for several different values B filter and an optical transfer of 10 GHz. The Examples of of this shape are shown in function Figure 1,bandwidth S(f ) calculated for several net effectvalues of convolution a Gaussian is function to smooth out the sharp of different of B and with an optical transfer bandwidth of 10 features GHz. The the rectangular function,with which makes theis output spectrum than the net effect of convolution a Gaussian to smooth out thenarrower sharp features of original rectangular profilewhich at themakes top and at the bottom,narrower necessitating the Page 4 the rectangular function, thewider output spectrum than the Normali -30 Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors -40 -50 -100 -80 -60 -40 -20 0 20 40 Frequency offset (GHz) 60 80 100 20 GHz 40 GHz 60 GHz 80 GHz 100 GHz 0 Figure 1 – The filter shape, S(f ), calculated for different filter bandwidths, B. Normalized power (dB) -10 -20 −f 2σ B/2 − f . u2 = √ 2σ -30 u1 = -40 -50 -100 -80 -60 -40 − B/2 √ -20 0 (8) 20 40 60 80 100 Frequencyfunction, offset (GHz) Evaluation of (7) requires use of the error which is commonly defined as Figure 1 The filter shape, S(f), calculated for different filter bandwidths, B. √ a π 2 erf(a) = e−t dt. (9) 2 Using (9), the integral in (7) can be0 evaluated, resulting in the final form of the channel generated the WaveShaper Programmable Optical Processor, Using (9), the shape integral in (7)by can be evaluated, resulting in the final given form of by the channel shape generated by the Finisar WaveShaper, given by B −B /2 − f /2 − f √ √ S(f ) = σ 2π erf − erf . 2σ 2σ √ (10) Examples of this filter shape are shown in Figure 1, S(f ) calculated for several different values ofExamples B andofan transfer function of for 10several GHz.difThe thisoptical filter shape are shown in Figurebandwidth 1, S(f) calculated values of B and transferis function bandwidth 10 GHz. The features net effect of net effect offerent convolution withanaoptical Gaussian to smooth outofthe sharp of convolution with a Gaussian is to smooth out the sharp features of the rectangular the rectangular function, which makes the output spectrum narrower than the function, which makes the output spectrum narrower than the original rectangular original rectangular profile at the top and wider at the bottom, necessitating the profile at the top and wider at the bottom, necessitating the existence of a point in the existence ofspectrum a pointthat in the spectrum that crosses the rectangular profile crosses the rectangular profile boundary. This “crossover point”boundary. marks the relative power level at which the actual filter shape has a bandwidth, B. 3 Page 5 THEORY Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors THEORY Figure 2 – Identifying the location of the crossover point, where the WaveShaper filter shape (red) crosses the ideal rectFigure Identifying the location of the crossover where thepoint, filter shape (red) crosses the Figure 2 –2 Identifying the location of thepoint, crossover angular window (blue). ideal rectangular window (blue. where the WaveShaper filter shape (red) crosses the ideal rectangular window (blue). Defining the channel bandwidth specification 2.2 Defining the channel bandwidth specification In the above the analysis, the desired filter profile specification is a rectangular function with bandDefining channel bandwidth width, B. The actual filter shape generated by the WaveShaper, however, is narIn the above analysis, the desired filter profile is a rectangular function with bandIn the analysis, the spills desiredout filterof profile a rectangular function rower at the top part of theabove channel, and the isrectangular windowwith at width, B. The actual filterB.shape generated by the WaveShaper, however, istop narbandwidth, The actual filter shape generated, however, is narrower at the part of the bottom, even though the channel is labeled as a B bandwidth filter. rower at the top the part of the channel, and spills out of the rectangular window at channel, and spills out of the rectangular window at the bottom, even though the To clearly define the bandwidth of the filter WaveShaper filterB. shape, it is instrucchannel as a bandpass the bottom, even thoughis labeled the channel is labeledwith as bandwidth, a B bandwidth filter. tional to look at the power level at which S(f ) crosses the rectangular function To clearly define the bandwidth of bandwidth the WaveShaper filter shape,it isitinstructional is instruc-to To clearly define the of the actual filter shape, boundary. This is illustrated in Figure 2, where the crossover point, Pcx , is identhe power level which S(f) crosses the rectangular function boundary. This is tional to look at look the atpower level atatwhich S(f ) crosses the rectangular function tified as the pointillustrated where in the generated filter shape from the WaveShaper crosses Figure 2, where the crossover point, is identified as the point where the boundary. This is illustrated in Figure 2, where the crossover point, Pcx , is idengenerated filter shape crosses the ideal rectangular window. the ideal rectangular window. tified as the point where the generated filter shape from the WaveShaper crosses constant powerpower level,level, thisthispower If the crossover point, cx , occurs If the P crossover point,atPcxa , occurs at a constant power level level can the ideal rectangular window. be useddefine to strictly define the bandwidth of the filter response.filter response. can be used to strictly the bandwidth of the WaveShaper If the crossover point, Pcx , occurs at a constant power level, this power level a given filter B,B,isis given givenby The power ratio atThe which Pratio cx occurs, power at whichfor Pcx occurs, a givenbandwidth, filter bandwidth, can be used to strictly define the bandwidth of theforWaveShaper filter response. by The power ratio at which Pcx occurs, for a given filter bandwidth, B, is given by S(−B/2) , (11) Pcx = S(0) S(−B/2) , Using (10), this can be evalu(11) = P cx where S(f ) was defined in the previous section. S(0) where S(f) was defined in the previous section. Using (10), this can be evaluated and ated and simplified to simplified to in the previous section. Using (10), this can be evaluwhere S(f ) was defined ated and simplified to erf (γ) , (12) Pcx = 2 erf (γ/2) erf (γ) , (12) Pcx = where 2 erf (γ/2) where 4 B γ=√ . 2σ B γ=√ . 2σ (13) (13) 4 Page 6 In the aboveofanalysis, the desired filter profile a rectangular function with bandFilter Bandwidth Definition the WaveShaper S-series ofisProgrammable Optical Processors width, B. The actual filter shape generated by the WaveShaper, however, is narrower at the top part of the channel, and spills out of the rectangular window at the bottom, even though the channel is labeled as a B bandwidth filter. To clearly define the bandwidth of the WaveShaper filter shape, it is instrucDefining the channel bandwidth specification tional to look at the power level at which S(f ) crosses the rectangular function boundary. This is illustrated in Figure 2, where the crossover point, Pcx , is iden1.5 tified as the point where the generated filter shape from the WaveShaper crosses 1.5 the ideal rectangular window. 1.4 If the crossover point, Pcx , occurs at a constant power level, this power level 1.4 1.3 1.2 1.1 erf( )/erf( /2) erf( )/erf( /2) 1.3 to strictly define the bandwidth of the WaveShaper filter response. can be used The power ratio at which Pcx occurs, for a given filter bandwidth, B, is given 1.2 by 1.1 1.0 0.9 0.8 0.7 1.0 0.6 0.9 Pcx = 0.5 1.0 S(−B/2) , S(0) 1.5 2.0 2.5 (11) 3.0 3.5 4.0 4.5 5.0 where S(f ) was defined in the previous section.B/BUsing (10), this can be evaluG 0.8 ated and simplifiedFigure to 3 0.7 Error function ratio trend as the ratio of filter bandwidth to optical transfer function bandwidth increases. 0.6 where Pcx = 0.5 1.0 where 1.5 2.0 2.5 erf (γ) , 2 erf (γ/2) 3.0 B/BG 3.5 (12) 4.0 4.5 5.0 Figure 3 – Error function ratioB trend as the ratio of Wave√ . transfer function bandShaper filter bandwidth γto=optical 2σ width increases. 4 (13) Previously, a relation between the 3 dB bandwidth of the Gaussian filter, BG, and Previously, a relation between theis3used dBhere bandwidth σ was stated in (3), which to express of γ asthe Gaussian filter, BG , and σ was stated in (3), which is used here to express γ as √ B γ = 2 ln2 . BG (14) Hence, an expression for the crossover point, Pcx , is found in terms of the ratio Hence, an expression for the crossover point, Pcx, is found in terms of the ratio between the WaveShaper filter bandwidth, B, and the innate optical transfer between the filter bandwidth, B, and the innate optical transfer function bandwidth, bethis noted thatforthis only holds for a Gaussian function bandwidth, BG . It BG. It should beshould noted that only holds a Gaussian spectrum. spectrum. Further conclusions can be drawn by considering the nature of the error funcFurther conclusions cantends be drawn considering the nature of the error tion, which to ½ as γ by increases. This is illustrated in Figure 3 where the funcratio of 1 /2 as γ isincreases. illustrated in Figure 3, where the tion, which tendserror to functions calculated asThis B/BGisis increased. γ ratio of erf(γ)/erf( 2 ) is calculated as B/BG is increased. Clearly, for large ratios of B/BG , the ratio of error functions tend to one; conservatively speaking, it appears safe to conclude that when the WaveShaper filter response is at least 3 times larger than the optical transfer function bandwidth, the crossover point occurs at for B BG , Pcx ≈ 12 . (15) This translates to, on a decibel scale, a level that is -6.02 dB down from the channel peak. This is illustrated in Figure 4, which plots the expression given in Page 7 0.6 Filter Bandwidth Definition0.5of the WaveShaper S-series of Programmable Optical Processors 1.0 1.5 2.0 2.5 3.0 3.5 B/BG 4.0 4.5 5.0 Figure 3 – Error function ratio trend as the ratio of WaveShaper filter bandwidth to optical transfer function bandwidth increases. -3.5 Previously, a relation between the 3 dB bandwidth of the Gaussian filter, BG , and σ was stated in (3), which is used here to express γ as Crossover point level (dB) -4.0 √ B . γ = 2 ln2 BG -4.5 (14) -5.0 Hence, an expression for the crossover point, Pcx , is found in terms of the ratio between the WaveShaper filter bandwidth, B, and the innate optical transfer function bandwidth, BG . It should be noted that this only holds for a Gaussian -5.5 -6.0 spectrum. Further conclusions can be drawn by considering the nature of the error funcBandpass filter bandwidth (GHz) tion, which tends to 1/2 as γ increases. This is illustrated in Figure 3, where the Figure 4 Location crossover point with respect to actual filter bandwidth. as B/BGof is increased. ratio of erf(γ)/erf( γ2 ) is calculated Clearly, for large ratios of B/BG , the ratio of error functions tend to one; confor large of B/BG, thethat ratio of error functions tend to one;filter conservaservatively speaking, itClearly, appears saferatios to conclude when the WaveShaper tively speaking, it appears safe to conclude that when the desired filter response is at response is at least 3 times larger than the optical transfer function bandwidth, -6.5 0 10 20 30 40 50 60 70 80 90 100 least 3 times larger than the optical transfer function bandwidth, the crossover point the crossover point occurs occurs at at for B BG , Pcx ≈ 12 . (15) This translates to, on a decibel scale, a level that is -6.02 dB down from the This translates to, on a decibel scale, a level that is -6.02 dB down from the filter channel peak. This is illustrated in Figure 4, which plots the expression given in peak. This is illustrated in Figure 4, which plots the expression given in(12) as a function of the actual filter bandwidth. For values of B that are at least three times larger than BG, it is evident that the crossover point is located at, roughly, -6 dB from the filter 5 peak. To reiterate, when a WaveShaper Programmable Optical Processor is set to produce a rectangular function of bandwidth B, which is at least three times larger than the optical transfer function bandwidth, the output filter response will have, roughly, a 6 dB bandwidth of B. If the bandwidth B is less than three times the optical transfer function bandwidth, the bandwidth is defined at the crossover point indicated in Figure 4. Page 8 Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors Section 3: Experimental Results predicted measured 0 Normalized power (dB) -10 -20 -30 -40 -50 -60 -40 -20 0 Frequency offset (GHz) 20 40 60 Figure 5 Comparison between predicted and measured channel shape for a 50 GHz filter response. In order to compare the presented theory to measured results, production data from a range of WaveShaper Programmable Optical Processors were extracted and analyzed. One 50 GHz channel, from a WaveShaper 1000S, was taken to compare to the predicted filter shape shown in (10). Measured results are compared to this predicted channel shape in Figure 5. The spectra match reasonably well, with the exception of a small broadening on one side of the experimental data, which is attributed to the asymmetric beam shape through the device. To compare the output of a multiport device with filter profiles sent to unique ports, a WaveShaper 4000S was set to output adjacent 10 GHz bandpass filters to each port. The measured data is shown in Figure 6 compared to predicted channel shapes, as derived in the previous section. It should be noted that an optical transfer function bandwidth of 12 GHz was used in these calculations, which takes into account the resolution bandwidth of the optical spectrum analyzer. Page 9 Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors 5 port 1 port 2 port 3 port 4 Normalized power (dB) 0 -5 -10 -15 -20 193.95 193.96 193.97 193.98 193.99 Frequency (THz) 194.0 194.01 194.02 Figure 6 Four port operation with 10 GHz channel spacing comparing (solid) measured and (dashed) predicted data. The predicted model produces an excellent fit compared to the nominal 10 GHz bandpass filter responses, with a slight broadening of the measured data from port 3. From (12), the predicted crossover point of a 10 GHz bandpass filter, using an optical transfer function bandwidth of 12 GHz, is -3.03 dB. From the measured spectra in Figure 6, the average crossover point occurred at -3.38 dB. Similarly, the behaviour of 50 GHz channels can be predicted with the presented models. A WaveShaper 4000S was therefore set to output adjacent 50 GHz channels to unique ports, and the measured data was compared with predicted spectra, with the results shown in Figure 7. 5 port 1 port 2 port 3 port 4 Normalized power (dB) 0 -5 -10 -15 -20 193.9 193.95 194.0 194.05 Frequency (THz) 194.1 194.15 Figure 7 Four port operation with 50 GHz channel spacing, comparing (solid) measured and (dashed) predicted data. Page 10 Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors 90 80 70 528 samples Avg: -6.2 dB Std.Dev: 0.41 dB Occurances 60 50 40 30 20 10 0 -8.0 -7.5 -7.0 -6.5 -6.0 Channel crossover point (dB) -5.5 -5.0 Figure 8 Survey of crossover points for six interleaved devices, resulting in 528 samples. As the generated filter bandwidth of 50 GHz was much larger than the optical transfer function, the experimental data shows an even better fit to theory. The crossover point of the measured spectra occurred at an average of -5.98 dB; in comparison, the predicted spectra have crossover points at -6.02 dB. To draw a conclusion about the location of the crossover point of a WaveShaper Programmable Optical Processor filter response, a relatively large sample size was required. Six WaveShaper processors were measured; each device was set to produce 50 GHz interleaver patterns, odd and even, sliced into distinct channels, and the crossover point was measured for each slice. This resulted in a total of 528 measurements, giving an appropriate sample size to determine statistical data about the location of the crossover point of a WaveShaper bandpass filter. The crossover point analysis is shown in Figure 8, tabulated in histogram format. For all samples, the mean crossover point location occurs at -6.2 dB, with a standard deviation of 0.41 dB; the standard deviation is acceptable considering the resolution of the measurement data. The simple mathematical model presented in this white paper gives accurate agreement with measured spectral data, allowing accurate definition of the crossover point, confirmed by statistical data. Page 11 Filter Bandwidth Definition of the WaveShaper S-series of Programmable Optical Processors Section 4: Conclusion Conclusion CONCLUSION CONCLUSION Conclusion This paper has presented a theoretical analysis of the WaveShaper output bandThis paper has presented a theoretical analysis of the bandpass filter shape pass filter shape,generated given asby the convolution between a rectangular profile given and as a the WaveShaper family of Programmable Optical Processors, This paper has presented a theoretical analysis of the WaveShaper output bandconvolution between a matching rectangular profile and atheory narrow and Gaussian spectrum. Good narrow Gaussian the spectrum. Adequate between experimental pass filter shape,matching given as the convolution between a rectangular profile a between theory and experimental data was found, confirming theand accuracy data was found. this approach. narrow Gaussian of spectrum. Adequate matching between theory and experimental Additionally, it was predicted that, for a reasonably small Gaussian spectrum, data was found. Additionally, it was predicted that the crossover point between the output filter the crossover point between the output filter response and the initial rectangular response and the initial rectangular profile of bandwidth would occur at roughly -6 Additionally, it was predicted that, for a reasonably small BGaussian spectrum, profile of bandwidth B would occur at roughly -6 dB, if the width of the rectangular dB, if the width of the rectangular profile was greater than 30 GHz. For filter profiles the crossover point between the output filter response and the initial rectangular profile was greater than than 30 GHz. Forcrossover filter profiles 30 GHz, the narrower 30 GHz, the point wasnarrower found to be than given by profile of bandwidth B would occur at roughly -6 dB, if the width of the rectangular crossover point was found to be given by profile was greater than 30 GHz. For filter profiles narrower than 30 GHz, the crossover point was found to be given byerf (γ) , (16) Pcx = 2 erf (γ/2) erf (γ) where , (16) Pcx = 2 erf (γ/2) where √ B where γ = 2 ln2 . (17) BG √ B This was confirmed by experimental data; . over 528 measurements for (17) proγ = 2 ln2 BG duction devices were used to build a statistical evaluation of the location of the This was confirmed by experimental data; over 528 measurements for procrossover point of 50 GHz bandpass filter responses. The mean crossover point duction devices were used to build a statistical evaluation of the location of the the sample set -6.2 dB,This with aconfirmed standard deviation of 0.41 dB. by experimental data; over 528 measurements for produccrossover point of 50 GHzwas bandpass filter responses. The mean crossover point tion devices were used to build a statistical evaluation of the location of the crossover the sample set -6.2 dB, with a standard deviation of 0.41 dB. point of 50 GHz bandpass filter responses. The mean crossover point of the sample set was found to be -6.2 dB, with a standard deviation of 0.41 dB. Acknowledgements: This white paper was written by Cibby Pulikkaseril and was made possibleThis with the assistance of Michaël Roelens, by Jeremy Bolger, Simon Poole and Acknowledgements: white paper was written Cibby Pulikkaseril Steve Frisken. and was made possible with the assistance of Michaël Roelens, Jeremy Bolger, Acknowledgements: This white paper was written by Cibby Pulikkaseril Simon Poole and Steve Frisken. and was made possible with the assistance of Michaël Roelens, Jeremy Bolger, Simon Poole and Steve Frisken. Page 12