Fig. 1. Experimental setup of three globally coupled SLs. DFB1, DFB2, DFB3, three distributed feedback lasers; LDC; laser diode controller; FC, fiber coupler; OC, optical circulator; VOA, variable optical attenuator; τ11,τ22,τ33, feedback delay time; PD, photodiode; OSC, oscilloscope; OSA, optical spectrum analyzer.
Fig. 2. (a1)–(a3) The chaotic time series from the three DFB lasers; (b1)–(b3) the ACFs; (c1)–(c3) the power spectra. The attenuation is 9 dB, I1,I2,I3=28.34,24.5,26.6 mA, T1,T2,T3=27.75,15.5,18°C.
Fig. 3. (a) ρm as a function of attenuation; (b) ρm as a function of I2.
Fig. 4. (a1)–(a3) Time series of signals at states I, II, and III, respectively; (b1)–(b3) the corresponding power spectrum.
Fig. 5. (a1)–(a3) The chaotic time series from the three SLs; (b1)–(b3) the ACFs, (c1)–(c3) the power spectra. The parameters are: Im=20,22.5,20 mA; krm=11.7,16.7,11.7 ns−1; τmm=2,2.02,2.04 ns; m=1,2,3.
Fig. 6. (a1)–(a3) The two-dimensional map of ρm as functions of the coupling strength kr2 and bias current I2 of DFB1, DFB2, and DFB3, respectively; (b1)–(b3) the PE of DFB1, DFB2, and DFB3, respectively. I1=I3,I2=I1+2.5 mA; kr1=kr3,kr2=kr1+5 ns−1; τmm=2,2.02,2.04 ns.
Fig. 7. TDS concealment with different time delays. I1=I3. (a) I2=I1−1 mA, krm=12,10,12 ns−1, τmm=3,3.02,3.06 ns,m=1,2,3; (b) I2=I1−1 mA, krm=12,11,12 ns−1, τmm=3,3.1,3.2 ns, (c) I2=I1−2 mA,krm=13.3,12.3,13.3 ns−1; τmm=4,4.07,4.13 ns, (d) ρm as a function of τ11. τ22=τ11+0.3 ns, τ33=τ11+0.7 ns, Im=21,19,21 mA, krm=12,10,12 ns−1, m=1,2,3.
Fig. 8. Architecture for the eight-armed bandit problem processed in parallel based on triple-channel chaos.
Fig. 9. Evolution of CDR for the triple-channel signals with different correlations and for the one-channel scheme. The vertical bars indicate the standard deviation around the mean value for three sets of simulated signals. P=[0.2,0.2,0.8,0.2,0.2,0.2,0.2,0.2].
Fig. 10. CC with different sampling intervals for the one-channel and triple-channel schemes, respectively. The vertical bars indicate the standard deviation around the mean value for eight sets of simulated signals. P=[0.8,0.2,0.2,0.2,0.2,0.2,0.2,0.2].
Fig. 11. (a), (b) CC and ρm as functions of attenuation; (c) CDR as a function of learning cycles. The vertical bars indicate the standard deviation around the mean value for 11 sets of signals with ρm<0.2 and ρm>0.3, respectively. The sampling interval is 10 ps. P=[0.3,0.2,0.8,0.1,0.2,0.3,0.5,0.4].
Fig. 12. CC as a function of bias current, for a comparison of the triple-channel scheme (red solid line), the previously investigated parallel scheme (blue dotted line), and the one-channel scheme (black solid line). The vertical bars indicate the standard deviation around the mean value for three runs. P=[0.3,0.2,0.8,0.1,0.2,0.3,0.5,0.4].
Fig. 13. (a) Evolution of the CDR for different distributions of reward probability in a changing environment. P1=[0.8,0.2,0.2,0.2,0.2,0.2,0.2,0.2], P2=[0.7,0.2,0.3,0.2,0.2,0.2,0.2,0.2]. (b) Threshold value adaption for P2.
Fig. 14. Evolution of the averaged CDR with randomly selected signals for eight-armed and 16-armed bandit problems. P=[0.3,0.2,0.8,0.1,0.2,0.3,0.5,0.4], and P=[0.3,0.2,0.8,0.1,0.2,0.5,0.2,0.2,0.2,0.3,0.3,0.4,0.5,0.1,0.1,0.2], respectively.