Academia.eduAcademia.edu

Figure 1 – uploaded by Bazil Raj

See full PDFdownloadDownload figure
Table 1. Analysis of variables (ANOVA) for control channel x Table 2. Analysis of variables (ANOVA) for control channel y

Table 1 Analysis of variables (ANOVA) for control channel x Table 2. Analysis of variables (ANOVA) for control channel y

Related Figures (24)

Fig. 1. Optical propagation path: The receiver system is located on the IT block building at the left (marked by red bal- loon) and the transmitter system is located on a tower at the right (marked by blue balloon).
Fig. 1. Optical propagation path: The receiver system is located on the IT block building at the left (marked by red bal- loon) and the transmitter system is located on a tower at the right (marked by blue balloon).
Fig. 2. Schematic of the free space optical communication experimental setup for 0.5 km data link: seen left -receiver with beam steering opto-electronic assembly and seen right-transmitter (red arrow line: optical path and black arrow line: sig- nal path).
Fig. 2. Schematic of the free space optical communication experimental setup for 0.5 km data link: seen left -receiver with beam steering opto-electronic assembly and seen right-transmitter (red arrow line: optical path and black arrow line: sig- nal path).
Fig. 3. Hardware design flow graph of data normalization algorithm in PVs preprocessor unit. Error signal Vp, and Vz, variations —10 V to +10 V are normalized with respect to Vp.r. Error signal of 0 V is equal to 800H = (2048) 9.
Fig. 3. Hardware design flow graph of data normalization algorithm in PVs preprocessor unit. Error signal Vp, and Vz, variations —10 V to +10 V are normalized with respect to Vp.r. Error signal of 0 V is equal to 800H = (2048) 9.
ing operation in beam stabilization, x and y-displacement errors contribute the greatest effect on the suc- cess of its operation. Therefore, only these parameters are considered in this work as the practical con- straints along with the desired response surface.
ing operation in beam stabilization, x and y-displacement errors contribute the greatest effect on the suc- cess of its operation. Therefore, only these parameters are considered in this work as the practical con- straints along with the desired response surface.
Fig. 4. Direct controller design based on the response surface model with pipeline registers.
Fig. 4. Direct controller design based on the response surface model with pipeline registers.
Fig. 5. (a) Typical back propagation neural network trained pattern coded with weights and bias values Standalone Linear Time Invariant (LTI) system with neural-controller in (b) open-loop: incident light falls on the OPD and FSM. FSM driven signals are approximately proportional to the values of process variables and (c) closed-loop: incident light falls on FSM. The reflected light falls on the OPD and control signal is applied in feed back control configuration.
Fig. 5. (a) Typical back propagation neural network trained pattern coded with weights and bias values Standalone Linear Time Invariant (LTI) system with neural-controller in (b) open-loop: incident light falls on the OPD and FSM. FSM driven signals are approximately proportional to the values of process variables and (c) closed-loop: incident light falls on FSM. The reflected light falls on the OPD and control signal is applied in feed back control configuration.
Fig. 6. Coordinate Rotation Digital Computer: clock manager, data acquisition, neural-controller and UART implemen- tation in the FPGA device. A, B, C, D, E, Fand G are Main sub-circuits; a, b, c, d, e, fand g are Pipeline stages.
Fig. 6. Coordinate Rotation Digital Computer: clock manager, data acquisition, neural-controller and UART implemen- tation in the FPGA device. A, B, C, D, E, Fand G are Main sub-circuits; a, b, c, d, e, fand g are Pipeline stages.
Fig. 7. Synaptic weight & bias value memory organization in single precision (32 bit) floating point data format in RAM for (a)hidden layer!1(HL1) (b) output layer (OL) and (c) hidden layer2 (HL2).
Fig. 7. Synaptic weight & bias value memory organization in single precision (32 bit) floating point data format in RAM for (a)hidden layer!1(HL1) (b) output layer (OL) and (c) hidden layer2 (HL2).
Fig. 8. (a) Hardware design flow-graph of hyperbolic tangent sigmoidal (bipolar) activation function model as the second order non-linear function and (b) plot of real & modeled neuron behavior with o = 0.032, 0 = 1, B = 255, ne {—255,255} and L = +f/2.
Fig. 8. (a) Hardware design flow-graph of hyperbolic tangent sigmoidal (bipolar) activation function model as the second order non-linear function and (b) plot of real & modeled neuron behavior with o = 0.032, 0 = 1, B = 255, ne {—255,255} and L = +f/2.
Fig. 9. Neuron output data routing ring circuits and serial in and serial out circular shift register array to accomplish the multiply-accumulation computation for (a) Hidden Layer 2 (HL2) and (b) Output Layer (OL). implemented in HLI, a1 x 12 demultiplexer is used in first DRRC to sort the input value into the corre- sponding second input of 2 x 1 multiplexer with respect-to command ‘sort_ctrl’ signal. The first nine reg- isters output are given to the HL2 multiply-accumulation unit. The outputs of the HL2 neurons are directly given to the second input of the | x 2 multiplexer of OL DRRC. The first two registers output are given in parallel to the OL multiply-accumulation computations.
Fig. 9. Neuron output data routing ring circuits and serial in and serial out circular shift register array to accomplish the multiply-accumulation computation for (a) Hidden Layer 2 (HL2) and (b) Output Layer (OL). implemented in HLI, a1 x 12 demultiplexer is used in first DRRC to sort the input value into the corre- sponding second input of 2 x 1 multiplexer with respect-to command ‘sort_ctrl’ signal. The first nine reg- isters output are given to the HL2 multiply-accumulation unit. The outputs of the HL2 neurons are directly given to the second input of the | x 2 multiplexer of OL DRRC. The first two registers output are given in parallel to the OL multiply-accumulation computations.
Fig. 10. (a) UART — RS 232 communication protocol architecture and (b) communication frame format.
Fig. 10. (a) UART — RS 232 communication protocol architecture and (b) communication frame format.
Fig. 11. Time series plot of performance of open loop experiment. Upper plot: reference x & y positions estimated using reference control data C,(1) & CO, middle plot: actual x & y positions estimated using direct controller data C,,(2) & Cy yf) and bottom plot: actual x & y positions estimated using neural-controller data C,(1) & Cy (0.
Fig. 11. Time series plot of performance of open loop experiment. Upper plot: reference x & y positions estimated using reference control data C,(1) & CO, middle plot: actual x & y positions estimated using direct controller data C,,(2) & Cy yf) and bottom plot: actual x & y positions estimated using neural-controller data C,(1) & Cy (0.
Table 3. Validation test results of developed direct and neural controller for Cx and Cy
Table 3. Validation test results of developed direct and neural controller for Cx and Cy
Fig. 12. Histogram of time series plot of control data deviations (errors). Upper plot: deviation errors e,,() and e, y(t) esti- mated from direct controller. Bottom plot: deviation errors e7,(f) and e>,(7) estimated from neural-controller.
Fig. 12. Histogram of time series plot of control data deviations (errors). Upper plot: deviation errors e,,() and e, y(t) esti- mated from direct controller. Bottom plot: deviation errors e7,(f) and e>,(7) estimated from neural-controller.
Table 4. Empirical standard deviations and position magnitude average of experiment using the population length of 10000 data points in open & closed loop control configuration
Table 4. Empirical standard deviations and position magnitude average of experiment using the population length of 10000 data points in open & closed loop control configuration
Fig. 13. Maximum error percentage of control signal predicted by the direct and neuro controller in open loop control configuration for x and y channel.
Fig. 13. Maximum error percentage of control signal predicted by the direct and neuro controller in open loop control configuration for x and y channel.
Fig. 14. Time series plot of beam displacements on the OPD in open and closed-loop control configuration for (a) x and (b) y-position.
Fig. 14. Time series plot of beam displacements on the OPD in open and closed-loop control configuration for (a) x and (b) y-position.
Fig. 15. Comparison of Power Spectral Density (PSD) of time series of position data in open and closed-loop control con- figuration for (a) x and (b) y-channels.
Fig. 15. Comparison of Power Spectral Density (PSD) of time series of position data in open and closed-loop control con- figuration for (a) x and (b) y-channels.
Fig. 16. Alternative (open (red) and closed (blue) loop control ) 10 second average values of absolute values of position displacement on (a) x-channel and (b) y-channel for 900sec in 1 minute interval.
Fig. 16. Alternative (open (red) and closed (blue) loop control ) 10 second average values of absolute values of position displacement on (a) x-channel and (b) y-channel for 900sec in 1 minute interval.
Fig. 17. A portion of beam angle of arrival measurement carried out on Jan. 11", 2014 when Zoffset 8 constant.
Fig. 17. A portion of beam angle of arrival measurement carried out on Jan. 11", 2014 when Zoffset 8 constant.
Fig. 18. Laser beam spot centroid motion with and without beam steering control: (a) beam wandering on 2D-4Q plane with horizontal and vertical scales of real space. (b) Histogram of plane center -beam centroid hypotenuse distance.
Fig. 18. Laser beam spot centroid motion with and without beam steering control: (a) beam wandering on 2D-4Q plane with horizontal and vertical scales of real space. (b) Histogram of plane center -beam centroid hypotenuse distance.
Fig. 19. Normalized impulse response of samples average verses impulse time with and without beam steering.
Fig. 19. Normalized impulse response of samples average verses impulse time with and without beam steering.
Fig. 20. Time series plot of scintillation index (6; ) measured at the receiver station (after propagating 0.5 Km) using ensample average of irradiance of optical signal with the beam steering turned on (red) and off (blue).
Fig. 20. Time series plot of scintillation index (6; ) measured at the receiver station (after propagating 0.5 Km) using ensample average of irradiance of optical signal with the beam steering turned on (red) and off (blue).
Fig. 21. Figures in the Left show the time series of photodiode output without (top) and with (bottom) beam steering; in the right show the corresponding histogram.
Fig. 21. Figures in the Left show the time series of photodiode output without (top) and with (bottom) beam steering; in the right show the corresponding histogram.
Related topics:
Computer ScienceDigital System DesignNeural Network Based ControllersScintillation Physics
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved