Nonlinear stability analysis of a time-delay opto-electronic oscillator Dan Gauthier, Kristine Callan, Lucas Illing,* Zheng Gao, Eckehard Schöll† Duke University, Department of Physics, Fitzpatrick Institute for Photonics, Center for Nonlinear and Complex Systems * Reed College † Technische Universität Berlin University of Maryland, Applied Nonlinear Dynamics Seminar, Dec. 10, 2009 1 Opt-electronic oscillator basics Uses commercial telecommunication components τ Feedback loop generates t time ti delay d l Ikeda, et al., PRL 49, 1467 (1982) Goedgebuer, et al., IEEE J. Quantum Electron. 38, 1178 (2002) Meucci et al., PRE 66, 026216 (2002) Gastaud, et al., Electron. Lett. 40, 898 (2004) Pouut/Pin MZ: cosine squared transfer function VMZ [V] FAST DYNAMICS Bandwidth: 10-20 GHz 2 Experimental setup MZM AMP PD LD 3 Quick summary of observations • ultra-fast, large-amplitude pulsing • featureless f t l power spectrum t above b a critical iti l lloop gain i • “immediate” transition from V=0 state to broadband chaos (through complex transient) as loop gain increases 4 More details of what we did and what we observe 5 “Standard” MZM bias τ Pout/Pin Bias to center edge of fringe m=π/4 VMZ [V] ∝m “Most” people bias under these conditions (Meucci et al. is a counter example) IIncrease loop l gain i γ from f zero until til an iinstability t bilit iis observed b d ((proportional ti l tto laser power, amplifier gain, photodiode response, etc.) 6 Pout//Pin Bifurcation sequence: “Standard” bias VMZ [V] FP P QP? Q C? PD S Signal (mV V) Increase γ 15 0 -15 0 15 Time (ns) 7 “Non-standard” MZM bias τ Pout/Pin Bias to top of fringe VMZ [V] System should be linearly stable! IIncrease loop l gain i γ from f zero until til an iinstability t bilit iis observed b d ((proportional ti l tto laser power, amplifier gain, photodiode response, etc.) 8 Poutt/Pin Bifurcation Sequence: “Non-standard” bias VMZ [V] FP FP FP BBC Increase γ 9 Mathematical Model of the Opto-Electronic Oscillator 10 Time delays are important! τ Signals take a finite time to propagate through a high-speed photonic/electrical circuits In a typical optical waveguide or coaxial cable, v~0.6c v 0.6c Signal take about 1 ns to travel 1 ft. Dynamics described by time-delay differential equations Bandpass filtering in loop (typical of high-speed devices) ( 10 kH (~10 kHz – 20 GH GHz ffor our system)) Amplifier saturates for high input voltage 11 Dimensionless model high frequency cut-off MZM bias loop time delay x& ( s ) = − x( s ) − y ( s ) + γ cos 2 {m + d tanh[ h[ x( s − τ )]} − γ cos 2 m y& ( s ) = ε x( s ) normalized filter (bandwidth)-1 amplifier saturation ε = 2 ×10−6 d = 2.1 τ = 1,820 12 Linear Stability Analysis x& ( s ) = − x( s ) − y ( s ) + γ cos 2 {m + d tanh[ x( s − τ )]} − γ cos 2 m y& ( s ) = ε x( s ) fixed point ( x*, y*) = (0, 0) becomes unstable for increasing γ via a Hopf bifurcation γH = − b± d sin(2m) b± ≈ ±1 γH → ∞ m = 0 System linearly stable at “non-standard” bias (top of the fringe)! Not what we observe! 13 Observed Instability Boundary γH = − b± d sin(2m) b± ≈ ±1 typical system noise add broadband noise to system • system unstable when m=0 • depends d d on system t noise i • asymmetric function of m → fi it i perturbations, finite-size t b ti nonlinear li analysis l i needed d d 14 Observed Transient Behavior with m=0 Catch the system during slow increase in g and near instability boundary ps ((near measurement limit of oscilloscope) p ) Pulse width ~200 p τ Why does the system want to pulse with bandwidth limited pulses? 15 Asymptotic Analysis c[ x( s − τ )] x& ( s ) = − x( s ) − y ( s ) + γ cos 2 {m + d tanh[[ x( s − τ )]} − γ cos 2 m y& ( s ) = ε x( s ) ε = 2 ×10−6 T k Take ( x*, y*) = (0, 0) y=0 When y = 0 , a perturbation p in the form of a pulse p cycles y around the loop p to give another pulse at time τ, which cycles around the loop to given another pulse at time 2τ, etc. Discrete map xn +1 = c[ xn ] (pulse amplitude) 16 Fixed points of discrete map with m=0 xn +1 = c[ xn ] For m=0 and our value of d, there exist one or three fixed points of the map, depending on γ xs*1 = 0 always stable (identical to linear analysis) xs*2 exits for γ ≥ γ c and is stable * u exits for γ ≥ γ c and is unstable x ≈ −1/(γ d 2 ) 17 Visualize fixed points of discrete map xs*2 x * u x * s1 x γc γ 18 Finite-size perturbation destabilizes fixed point xs*2 x * u x perturbation amplitude lit d * s1 x γc γ th pulsing state ≈ −1/(γ d 2 ) γ 19 Inject pulses into the oscillator τ inject electrical pulses 20 Map out perturbation stability boundary full DDE's experiment map predictions 21 Instability boundaries for other m γ th determined for one particular size of the perturbation amplitude expect instability boundary to be given by min(γ H , γ th ) 22 Compare with observed instability boundary qualitatively q y similar, explains p asymmetry y y about m=0 23 Fixed points for m = 0.3 m=0.3 -x ~0.305 γH γc γ Expect to see Hopf bifurcation 24 Fixed points for m = -0.2 x m = -0.2 γc γH γ expect to see sub sub-critical critical saddle-node saddle node bifurcation …. except near m ~ -π/4 25 But why ultrafast pulsing for m=0 and featureless power spectrum???? 26 Continuous mapping Extend asymptotic analysis Fi d a mapping Find i off th the iinterval t l s-τ,s onto t the th interval i t l s,s+τ xs , s +τ = c[ xs −τ , s ] Start with an initial Gaussian pulse look at one iteration of continuous mapping 27 Strong pulse compression for m=0 -x X 1.0 0.8 initial pulse 0.6 after ft one round d trip ti 0.4 0.2 0.2 0.4 0.6 0.8 1.0 t s Conclusion: After one round trip, strong pulse compression. Pulse will continue to compress p until the p pulse bandwidth is comparable p to filter bandwidth (asymptotic analysis then breaks down). Such ultrafast pulses at irregular time intervals will have a flat power spectrum! 28 Consistent with observations … ultrafast pulses initiated by noise … featureless power spectrum 29 Continuous mapping for m = π/4 X-x 1.0 initial pulse 0.8 after one round trip 0.6 0.4 02 0.2 0.2 0.4 0.6 0.8 1.0 ts For "standard" MZM bias, pulse perturbations expand in time. Therefore, pulses smoothed out. out Likely to see periodic waveforms consistent with a Hopf bifurcation. 30 Summary • Opto-electronic oscillator shows surprising complex behavior that has been largely missed by other researchers • Finite-size perturbations (even from noise) forces the system to access states far from the trivial fixed point • A full f ll nonlinear li stability t bilit analysis l i iis needed d d tto understand d t db behavior h i • A continuous mapping demonstrates that the system wants to produce pulsing behavior at "non non-standard standard" MZM bias Question: Can feedback control stabilize ultrafast pulses??? • True Hopf bifurcations only expected near "standard" bias. This is a surprise to at least one theorist who has studied this system for many years! 31