connecting theory and practice – part a
DESCRIPTION
Technion Israel Institute of Technology. Connecting Theory and Practice – Part A. Spring 2013 Final Part A Presentation. Contents. Project Definition and Goals Work Plan Review Project Main Activities: Matlab Algorithm + Integration AWR Activities Simulations Gantt. Main Challenges. - PowerPoint PPT PresentationTRANSCRIPT
Connecting Theory and Practice – Part A
Spring 2013Final Part A Presentation
TechnionIsrael Institute of Technology
Professor: Yonina EldarSupervisor:
Debby Cohen
Consultants:
Eli Shoshan, Rolf Hilgendorf
Students: Etgar Israeli, Shahar Tsiper
Project Definition and Goals Work Plan Review Project Main Activities:
◦ Matlab Algorithm + Integration◦ AWR Activities
Simulations Gantt
Contents
Understanding, fixing and improving the Matlab code
Learning AWR tool and Modeling MWC Deeper understanding of the main issues the
system suffers from Integrating Co-Projects into one Matlab code
Main Challenges
◦ Matlab Reconstruction◦ AWR Activities – Part A◦ AWR Activities – Part B◦ A-Matrix Calibration◦ MWC Development Support Systems
Project Main Stages
Matlab reconstruction algorithm◦ The current algorithm
What is there, what is missing Tuning detection SBR4 (SBR2 ?) SBR Special How do we want to display results of reconstruction
Modulate back up to show the original signal Combine slices, find signal and demodulate to
baseband Enter first draft of MWC schematic
Work Plan Review – Matlab
Completed
Starting AWR activities◦ Understand current schematics of analog part of
new MWC◦ Get understanding of AWR tool◦ Define method for in- and output files
Matlab , CSV etc.◦ Enter first draft of MWC schematic
Work Plan Review – AWR activities
Completed
Matlab◦ Calculating recovery success % with correlation
technique◦ Complete rewrite of Sample and Expander
algorithms◦ Implementing support recovery with thresholds◦ Complete rewrite and debug of MWC System code◦ Full consistency check with article
Additional Tasks Completed
Integration from co-projects◦ Integrating & comparing different Mixing Series◦ Integrating Filter Banks Expander
Finally - Achieving great recovery results without noticeable redundant harmonics
Additional Tasks Completed
AWR◦ Precise modelling of most linear components
using S Parameters – Waiting for new card schematics to complete
◦ Mixing Series full integration into AWR◦ Almost complete modelling of old MWC card
More details in last meeting’s summary
More tasks completed
Main issues solved:◦ Redundant Harmonics have been (almost) eliminated◦ Reconstruction (-1) factor has been removed. Entire
Reconstruction method has been rebuilt◦ Parameters names are now consistent throughout the code -
mostly L, L0, m, M, etc. ◦ Fixing “minor” things – for example:
exp(jωt) vs. exp(-jωt) A.’ (transpose) vs. A’ (conj-transpose) A=SFD vs. A=conj(SFD)
Matlab Code Debug
Features Added◦ Constructing matlab libraries by subjects◦ Consistency Check◦ Error check – Original vs. Recovery signal◦ Improving signal generation (qpsk, sinc and it’s
powers)◦ Integrating different mixing series
Matlab Code Improvements
Full article-code consistency check has been implemented
Conditions must be met or user must authorize manual override
Consistency Check
Error check – Original vs. Recovery◦Comparison Method: Correlation between Original & Recovery
Signal:
◦ Where :◦ c = xcorr(x,y) returns the Cross-Correlation sequence in a length 2N-1 vector, where x and y
are length N vectors (if x and y are not the same length, the shorter vector is padded to the
length of the longer vector).
By default, xcorr computes raw correlation with no normalization.
◦ In the denominator, 2-norm of x and y – for normalization
◦ Matlab code (function handle):
CorrXY = @(x,y)max(abs(xcorr(x,y)))/(norm(x,2)*norm(y,2));
max | ( , ) | , 0 1|| || || ||xcorr x yx y
Error check – Original vs. Recovery
Parameter ValueSignal Type QPSK
Mix Series Type GoldN 4
Hardware Channels
4
q 5B 20MhzFp 24Mhz
Fnyq 6.144GhzM 263
Expander Type FIR
Frequency Domain Zoom-In
Time Domain Zoom-In
Console OutputParameter ValueSignal Type QPSK
Mix Series Type GoldN 4
Hardware Channels 4q 5B 20MhzFp 24Mhz
Fnyq 6.144GhzM 264
Expander Type FIR
MWC Analog Schematics
AWR – Top System ViewSig
Generator (AWR
or Matlab)
Pre Processi
ngSplitter + Mixer
Post Processi
ngSampling
+ A/D
Output to
Matlab Server
AWR – Hierarchy
Output to Matlab Server
Sampling + A/D
Post Processing
Splitter + Mixer
Pre Processing
Sig Generator (AWR or Matlab)
Signal Generator Analog generated input from AWR Digital input from Matlab is partially implemented
In this block all of the components are accurately modeled with S-Parameters
Pre Processing
AWR unable to model mixer in the way we intended. Fallback option – Mathematical Multiplier Mixing series are generated in Matlab
Mixer
LPF-105 was found inadequate for signal properties Missing properties for buffers and output driver
Post Processing
Quantization doesn’t work perfectly yet - WIP
Sampling + A/D
4 digital channels are multiplexed into 1 channel That channel is demuxed in Matlab environment Timing AWR and Matlab with triggers – WIP Full DSP of AWR output and A matrix calibration – Main open Task
Output to Matlab Server
Understanding AWR Signal Types Signals in AWR can be modeled with 4 different
types:◦ Real Signal – Signals are modeled using real numbers
as a very dense function of time (5*Fnyquist)◦ Complex – Signals are modeled using complex
numbers as a dense function of time◦ Complex Envelope – Signal data contains carrier wave
and modulated information separately◦ Digital Signal – I/O to/from Matlab
Complex Envelope Type AWR utilizes the CE representation of signals whenever possible
to gain the tremendous advantage in simulation:
This representation, that is utilized in all mixer components to shorten simulations by order-of-magnitude, annihilates the wide spectrum of the signals that are mixed with the series
S-Parameters - Background
S-Parameter files contain the behavior of linear components for different frequencies
They represent the following LTI system:
Usually our components will have b1 and a2 connected to GND Information for adding components using S-Parameters will be
documented in the project book
S-Parameters - Example LFCN-105_Plus25degC.S2P
Mix Series Modeling Frequency, M and tRise were all taken into
account:
Sine Signal as MWC Input Sine wave was used as MWC input Fp = Fnyq/M = 6.144GHz/261 ≈ 23.5Mhz Sine wave frequency - 400Mhz ≈ 17 Fp At the MWC output, we expect to see evenly spaced
deltas, by 23.5Mhz between them Different hardware channels should differ only by
amplitude Results comply with theory
Reminder: the mixer is implemented as a Mathematical Multiplier!
Sine Signal MWC Output – Zoom Out
dB
Sine Signal MWC OutputdB
Simulations – Results & Conclusions
◦Different kind of simulations had been made, with Constant/variable SNR. For example:
Demonstrate fp ≥ B condition Comparing Different mixing series Collapsed vs. non collapsed channels –
m = const { m ≡ q*HardwareChannels } Demonstrate m ≥ 2N Condition
Demonstrate M ≥ L condition
Comparison Method (Reminder)
◦ Comparison Method: Correlation between Original & Recovery Signal:
◦ Where :◦ c = xcorr(x,y) returns the Cross-Correlation sequence in a length 2N-1 vector, where x and
y are length N vectors (if x and y are not the same length, the shorter vector is padded to
the length of the longer vector).
By default, xcorr computes raw correlation with no normalization.
◦ In the denominator, 2-norm of x and y – for normalization
◦ Matlab code (function handle):
CorrXY = @(x,y)max(abs(xcorr(x,y)))/(norm(x,2)*norm(y,2));
max | ( , ) | , 0 1|| || || ||xcorr x yx y
Different fp comparison
For constant B, sampling with different fp rate
fp = [0.8,1,1.2,1.4]*B were taken
fp ≥ B is necessary for blind recovery
Different fp comparison
Simulation Parameters:Signal – 'sinc‘Mix Series Type - 'GoldSeries'N = 6Hardware Channels = 4q = 5B = 20MhzFp = 16,20,24,28 Mhz M = 499Fnyq = 6.144GhzNumber of simulations = 10
Comparison Between Different Mixing Series - WIP
Series Types:◦ Random Sequences (Using Matlab Function ‘rndsrc’)◦ One Random Sequence and it’s shift◦ Gold Sequence◦ One Gold Sequence and it’s shift◦ Lu M Sequences◦ Lu Lagendre Sequences
SNR = 30
Comparison Between Different Mixing Series – Ch=4, q=5
Simulation Parameters:Signal – 'sinc'N = 6HardwareChannels = 4q = 5B = 20MhzFp = 24MhzM = 263Fnyq = 6.144GhzSNR = 30dBNumber of simulations = 300
Comparison Between Different Mixing Series – Ch=20, q=1
Simulation Parameters:Signal – 'sinc'N = 6HardwareChannels = 20q = 1B = 20MhzFp = 24MhzM = 263Fnyq = 6.144GhzSNR = 30dBNumber of simulations = 120
Comparison Between Different Mixing Series
Interim conclusions: For HardwareChannels=4, q=5:
◦ ‘rndsrc’ – Most of the times gives good results ◦ Lu M and shifted Sequences - Mediocre results ◦ Gold Sequence – best results.◦ Lu Lagendre – bad results! (most of the time
Support recovery doesn’t succeed)
Comparison Between Different Mixing Series
Interim conclusions: For HardwareChannels=20, q=1:
◦ ‘rndsrc’ – Best results.◦ Gold Sequence – Mediocre results.◦ Lu Lagendre Sequence – Mediocre results
Different M comparison
For constant parameters, sampling with different series length
M = [155, 191, 263, 299] were taken fp = 24Mhz, q=5, fnyq =6.144*109
L0 = = 130 L = 2L0+1 = 261 M = = 257
M ≥ L and M ≥ Mmin is necessary for blind recovery
Different M comparison
Simulation Parameters:Signal – 'sinc‘Mix Series Type - 'GoldSeries'N = 6Hardware Channels = 4q = 5B = 20MhzFp = 24 Mhz M = [155, 191, 263, 299]Fnyq = 6.144GhzNumber of Simulations = 30
Collapsed vs. non collapsed – Sim1
For constant parameters, and constant SNR sampling with different number of q & hardware channels
HardwareChannels = [1,2,3,4,6,10,20] were taken q = [1,3,5,9,15,21] were taken m ≡ q*HardwareChannels
m≥2N is necessary for blind recovery
Collapsed vs. non collapsed – Sim1
Simulation Parameters:Signal – 'sinc‘Mix Series Type - 'GoldSeries'N = 6Hardware Channels = [1:4,6,10,20] q = [1,3,5,9,15,21] B = 20MhzFp = 16,20,24,28 Mhz M = 263Fnyq = 6.144GhzSNR = 30dBNumber of simulations = 30
Collapsed vs. non collapsed – Sim2
Let m ≡ q*HardwareChannels Setting m = constant, Comparing results for variable
SNR m = 105 was taken
◦ HardwareChannels = [105,35,21,15,1] and respectively q = [1,3,5,7,105] were taken
◦ N = 42 was taken
Collapsed vs. non collapsed – Sim2
Simulation Parameters:Signal – 'sinc‘Mix Series Type - 'Gold'N = 42Hardware Channels = [105,35,21,15,1] q = [1,3,5,7,105] B = 12MhzFp = 24 Mhz M = 399Fnyq = 6.144GhzNum of simulations = 30
Deeper theory understanding, improving and integrating new technologies into the Matlab code
Solving AWR issues Combining AWR and Matlab into one seamless
system Implementing the solutions on the actual system Writing comprehensive literature, covering main
methodologies used in AWR
Future Challenges
AWR◦ AWR Analog output synchronization with Matlab
Server – WIP◦ Track&Hold + A/D – WIP◦ Mixer and buffers aren’t modeled correctly – More
research necessary.
Tasks in progress - AWR
AWR◦ Inserting digital signal to AWR from Matlab◦ Experimenting with AWR Noise Figures until
reaching adequate levels◦ Solving Mixer’s modelling issues
More details in last meeting’s summary
Open Tasks
Matlab:◦ Used for full modeling of the MWC system –
Already given – need to be fixed
◦ Calibration Methods AWR:
◦ Implementing an analog model of the entire MWC system.
◦ Linking the analog AWR frontend and the digital Matlab backend
Labview:◦ Implementing calibration procedure
Systems Used In Project
Project Gantt – 2nd Stage Summer
2013week 1 20/10
week 2 27/10
week 3 3/11
week 4 10/11
week 5 17/11
week 6 24/11
week 7 1/12
Mid Presentation
week 8 8/12
week 9 15/12
week 10 22/12
week 11 29/12
week 12 05/01
week 13 12/01
week 14 19/01
Writing AWR literature
Refine MWC design
creation the input mixing series (AWR or matlab)
Integrating AWR output to Matlab
Implemetation of T&H + A/D - AWR
Get final models for all components
Integrating matlab signal to AWR
Ensure synchronization between patterns
Enter final schematic after solving AWR issues
Basic Verification of output data using matlab
AWR and Matlab real-time loopback
Anti-aliasing filter response
Creat logic design for LabView
Implementing the solutions on the actual system
Thank You!
Spring 2013Final Part A Presentation
Supervisors: Rolf Hilgendorf, Debby CohenStudents: Etgar Israeli, Shahar Tsiper
TechnionIsrael Institute of
Technology