seminaire gul 29oct2009 - SUPELEC · Modèles Théoriques de Graphes pour les Equipments Multi-standards de Radio Logicielle ... o Algorithmes Génétiques Cas d'Etudes d'Opérateurs

1

Seminar SCEE

Optimization of MultiOptimization of Multi--standards standards

Software Defined Radio Software Defined Radio

Equipments : Equipments :

A Common Operators’ ApproachA Common Operators’ Approach

SUPELEC - Campus de RennesSCEE – Signal, Communications et Electronique Embarq uée

IETR – UMR CNRS 6164Institut d'Electronique et Télécommunications de Re nnes

Sufi Tabassum GUL29th October, 2009

Sufi Tabassum GUL- SUPELEC – 29th

Oct. 2009 2IETR - INSTITUT D’ÉLECTRONIQUE ET DE TÉLÉCOMMUNICATI ONS DE RENNES

Presentation outline

� Multi-standard Design Issue

� Parametrisation Techniques for Software Radio

� Graph Modeling of SDR Equipment

� Optimization of Graph Models

� Case Studies of Common Operators

� Conclusions & Perspectives

� Publications

2



Plan de la présentation

� Problème de la Conception Multi-standard� Techniques de Paramétrisation pour la Radio Logicie lle

1. Technique par Fonctions Communes2. Technique par Opérateurs Communs

o Approche Pragmatiqueo Approche Théorique

� Modélisation Graphique des Equipments de Radio Logi cielle� Modèles Théoriques de Graphes pour les Equipments Multi-standards

de Radio Logicielle� Optimisation de Modèles de Graphes

� Paramètres de Coûto Types de Coûtso Fonction de Coût

� Optimisationo Recherche Exhaustiveo Recuit Simuléo Algorithmes Génétiques

� Cas d'Etudes d'Opérateurs Communs� DMFFT� LFSR� FRMFB

� Conclusions & Perspectives� Publications




� Multi-standard Design Issue� Parametrisation Techniques for Software Radio

1. Common Function Technique2. Common Operators’ Technique

o Pragmatic Approacho Theoretical Approach

� Graph Modeling of SDR Equipment� Graph Theoretical Models of Multi-standards SDR Equipments

� Optimization of Graph Models� Cost Parameters

o Types of Costso Cost Function

� Optimizationo Exhaustive Searcho Simulated Annealingo Genetic Algorithms

� Case Studies of Common Operators� DMFFT� LFSR� FRMFB


3















Multi-standard SDR design

GPS

Bluetooth

Figure 1 : Multi-standards SDR design

4



Multi-standard SDR Design

Velcro Solution

SDR Approach

Our Approach

Set of Reconfigurable

Common Operators

Approach x

Figure 2 : Multi-standards design trends



Multi-standard SDR Design

� To design multi-standard reconfigurable radio, we h ave choice between two extremes

� One extreme: go “Velcro” (Efficient but costly)

o One self-contained module per standard.

� Other extreme: go “Primitive” (Cheaper but deadlines!)

o Use only adders, multipliers, etc.

o Provide “higher” functionality by multiple calls of simpler modules.

� There may be other choices

� Intermediate granularity (e.g. in NoC, SoC etc.)

� Formalization at an intermediate granularity

� Our aim is to find the best trade-off between:

� “ Performance and Cost ”

5















Parametrisation

� Techniques of parameterization were introduced in order to

� Identify commonalities

� Exploit these commonalities

� Parameterization can be considered as a sub-part of the digi tal

radio design methodology.

� We propose

� “Parameterization technique should be considered as a method that searches

for and finds all the commonalities inside single/between several different

standards in order to optimize, the resources during equipment’s

implementation and/or execution phases.”

6



Parametrisation Techniques for SDR

� Parametrisation :

� A promising technique of designing radio systems entities

� Takes advantages of the programmable or at least reconfigurable

capabilities

�� An optimal sharing between hardware and software

resources.

� Parametrisation Techniques can be divided into :

1. Common Function Technique

2. Common Operators’ Technique



Common Function Technique

� CF technique : function sharing among different

standards [1].

� Non-duplication of redundant components &

� Possible save in complexity.

Figure 3 : Multi-standards transmitter data processing tasks

[1]. W. H. W. Tuttlebee, Software Defined Radio: Enabling Technologies. John Wiley & Sons Ltd. UK, 2002.Details about chapter 8 to be added

7



Common Function Technique

� Some conclusions about CF : As stated in [2], CF is not

sufficient because

� CF technique depends highly upon the selected standards.

� The evolution to new standards must require the addition of the distinct

components of each function in the associated CF.

� As a consequence, the CF should be redefined and redesigned to be

capable to meet the requirements of the advanced standards.

� Due to standards’ dependency, our feeling is that the CF

technique is not flexible enough for SDR applications.[2]. Arnd Rhemiere

Details to be added



Common Operator Technique

� Common operator technique : identification of common

elements based on structural aspects.

� CO Performs operations without knowing their application.

� CO is independent of implementation of a terminal.

� CO is to be considered as an Open Technique[3].

� CO can be called and re-called by distinct functions several

times all along the terminal.

[3]. L. Alaus, J. Palicot, C. Roland, Y. Louët, and D. Noguet, “Promising Technique of Parametrisation ForReconfigurable Radio, the Common Operators Technique: Fundamentals and Examples,” SpringerScience + Business Media LLC, 2009

8




Figure 4 : Generalized block diagram showing the breakdown of several standards

Very Fine Grain

Velcro

Incr

easi

ng G

ranu

larit

y




� The goal is to find the maximum number of common elements

and then share their functionalities between several

processing tasks.

� In order to attain the best cost-performance trade-off

� Identify some level of granularity

�� CO technique consists in increasing the granularity of the

basic primitive levels elements to address in an SDR.

9




� A trade-off could be proposed between CF and CO, with the

following rules:

1. A CF can call (one or several) functions,

2. A CF can use (one or several) operators,

3. A CO cannot call a function.

Figure 5 : Two technique of parametrisation (trade-offs)



Common Operator’ Technique

� Common Operators’ can be identified either by :

1. Pragmatic approach (PA) or

o The initial approach developed to identify or create possible COsand follows two stages:

– Existing Search &– Constructive Search

2. Theoretical approach (TA)

o An attempt to formalize SDR design.

o It consists in moving the classical breaking down of Figure.4 intoa graphical approach.

o Applies optimization process in order to find the most relevantCOs (or their combination) present in the graph derived.

o The optimisation process related to the TA is performed by aspecific cost function to be minimized.

10



The Theoretical Approach

� Our vision of the parameterization:

“The parameterization techniques are the means to find the

optimal path in the tree derived from Figure 2. Here we are

considering a global optimality”

� Three challenges arise:

1. The definition of the graph

2. The definition of cost parameters, finding costs and

development of the cost function

3. The optimization algorithms



Candidates for Common Operator

� A common operator is a tuneable operator which can carry out

some common functions of several standards.

� FFT/Butterfly[3], DMFFT[4] and LFSR[5] are strong candidates for

common operators as they can be used in the implementation of

numerous important tasks of a communication transceiver, e.g. filtering,

equalization, channelization, OFDM (de)modulation, convolutional

coding, scrambling, etc.

[3]. J. Palicot, C. Roland, “FFT: a Basic Function for a Reconfigurable Receiver,” ICT’03 Conference, Thaiti,France, 2003.

[4]. Ali Al Ghouwayel, Yves Lou¨et and Jacques Palicot, “A Reconfigurable Butterfly Architecture for Fourier andFermat Transforms,” IEEE WSR’2006, Karlsrhue, Germany, March 2006.

[5]. L. Alaus, D. Noguet and J. Palicot, “A Reconfigurable Linear Feedback Shift Register Operator for SoftwareDefined Radio Terminal,” ISWPC, Santorini, Greece, May 2008.

11















� We Model radio as graph :

� Progressively simpler processing elements (PE).

� When necessary, a PE is called multiple times (not

replicated).

� 2 critical parameters per PE:

o cost andtime(computational delay)

Graph Modeling

12



Overview of Our Methodology

� Our objective is to find commonalities and hence

Common Operators

� We want to optimize graphs of multi-standard

systems based on these Common Operators

� Our procedure consists of three steps:

1. Drawing graphs of a radio system

2. Assigning cost parameters to various blocks of radio system

3. Running optimization algorithms to find the optimum solution



Graph Modeling

Left: PE A needs:Either B OR C

Right: PE A needs:BothB AND C

� “ Hypergraph” in order to introduce two different types ofdependencies between the nodes:

� OR dependency

� AND dependency

Figure 7 : A hypergraph of tri-standard SDR systemFigure 6 : Two Possible Dependencies in Graph

13



Graph Modeling

NAND NOT XOR AND OR

WiFi #1 WiFi #2 WiFi #3 WiMAX

Scrambler/Randomiser

Convolutional Coder Interleaver

Constellation Mapper FFT-N

RS Encoder

Butterfly

Adder MultiplierLFSR

LUTa a’ b b’

Very Fine Grain

Velcro

Incr

easi

ng G

ranu

larit

y

Figure 8 : A simplified hypergraph of tri-standard SDR system













14



Cost Parameters

� Parameters associated with PE 1. Building Cost (BC) &

o The cost of the block/PE capable of computing a function, called building cost(BC) and it is paid once during the useful life of a radio.

2. Computational Cost (CC)o The computing time required to perform a particular function, called

computational cost (CC) and it is paid every time a component is brought intoplay.

� Parameters associated with arcs1. Number of Calls (NoC)

o TheNoC parameter specifies thenumber of times a lower level PE is called toperform the functionality of higher level PE.

� In order to solve the optimization problem that findsbalance between economy and computing efficiency weconsider these parameters in our cost/objectivefunction.



Types of Costs

� We have considered 3 types of costs for PEs

� Cost#1: In terms of equations

o Number of multiplications and/or additions, generic cost.

� Cost#2: FPGA/ASIC implementation

o In terms of gates/LUT/Slice/hardware synthesis (HW)

� Cost#3: DSP(SW)/HW execution time

o In terms of number of cycles/Time of execution

� These costs are based on:

� Design objectives and may be design phase (in terms of

equations : at early design phase for instance)

� Implementation figures availability

� Required precision

Sometimes BCSometimes CC

Obviously BC

Obviously CC

Sometimes BCSometimes CC

15



Cost Function

denotes the building cost of ith component in the system.

indicates if the ith node is present in the system or not.

is the total building cost of all the components that are present in the SDR system.

}{ Nn ,...,2,1=

indicates that there may be n standards present in an SDR where

i.e. if we choose N=3 then it means that there are three

321 ,,, SandSSstandards present namely in an SDR system.

denotes the computational cost of kth component in the system.

nS

}{ Nn ,...,2,1=

is the total computational cost of any of , where

.



Cost Function

nS }{ Nn ,...,2,1=is the total computational cost of all of , where

.

+= ∑∑∑ ∈∈ n k

Nnnkni

iiSbool

SDR SCCwNBCwCNnn

))((..min_

))((

Provide a balance between relative costs of BC and CCω

ωn

� Coarse grain oriented design� Towards Velcro

� Fine grain oriented design� Towards primitive level

If ω << ωn

If ω >> ωn

where

16















Optimization Techniques

� Techniques that give exact optimal solution� Exhaustive Search� Require a computing effort that increases exponentially with

number of nodes.

� Techniques that provide near-optimal solution in lesscomputing� Simulated Annealing� Genetic Algorithms

17



Exhaustive Search

� Exact methods require a high computing effort (increasesexponentially with number of nodes).

� Only for problems involving fewer nodes (say 10 nodes).

� Used to check if sub-optimal methods find a good solution (fo rsimple cases).

Figure 9 : A generic view of SDR equipment Figure 10 : Results of running ES on generic example

200200



Simulated Annealing

� Simulated Annealing : gives a near-optimal solution

� A random-search technique.

� Exploits an analogy of metal cooling into a minimum energy

crystalline structure .

Figure 11 : Results of running SA on generic example

Optimization tendency

18



Genetic Algorithms

� Genetic Algorithms (GA) are a family of computational model sinspired by evolution.

� An implementation of a GA begins with a population of (typica llyrandom) chromosomes. Over successive generations, thepopulation evolves toward an optimal solution.

Figure 12 : Results of running GA on generic example



Comparison on generic example

Figure 13 : Results of running GA on generic example

19



Sum-up

1. Use graph to list each possible design.

2. For each design, calculate its total cost: i.e. BC and CC

(time to perform each top PE).

3. Use Algorithms e.g. exhaustive search, simulated

annealing, etc. to optimize design (finds the lowest cost

solution).

4. Choose least expensive design that performs the

functionality of each top PE.

5. After exploring various optimization techniques we came

to the conclusion that SA is best suited to the problem at

hand.









o Types of costso Cost Function




20



Sub-design examples

� Goal: validate our approach

Means: integrating the results of other PhD student s of

research team

� Also permits to mature our approach

� Confront to reality: the theoretical approach

� DMFFT: In collaboration with Ali Al-GHOUWAYEL

� LFSR: In collaboration with Laurent ALAUS

� FRMFB: In collaboration with R. MAHESH

� A GUI was developed to facilitate these case studies



Figure 14 : Graphical user interface

GUI Tool

� In order to facilitate the drawing of graphs and associatingcosts for different costs parameters a GUI was developed.

Figure 15 : Parameters’ management window

21



Example #1: DMFFT

DMFFT : Dual mode FFT Context of an OFDM demodulation with RS

decoding

Figure 16 : An example of FFT sharing between OFDM demodulation and RS decoding over GF(Ft).



Example #1: DMFFT

Figure 17 : A simplified tri-standard SDR system

22



Example #1: DMFFT

Best solution �� implement FFT through DMFFT and RBPECO �� RBPE

BC/CC

=128

FFT-N

1



Example #2: LFSR

A linear feedback shift register (LFSR) is ashift register whose input bit is a linear functionof its previous state.

There are two implementation styles of LSFRs:Fibonacci RF-LFSR and Galois RG-LFSR.

The study of common operators has lead us tobuild the architectures called R-LFSR and ER-LFSR respectively.

23



Example #1: DMFFT



Example #2: LFSR

x-LFSR

22ER 8R14ER 2R 6RG

x-LFSR

WiFi

24



Example #2: LFSR

22ER 8R14ER 2R 6RG

x-LFSR

WiFi

Best solution ��

CO ��



Example #3: Channelizers

Best solution ��

FRMFBCO �� FRMFB

Figure 16 :

25




OFDM Modulation/

Demodulation

FFT

Figure 16 :




Best solution �� Implement RS decoding and Channelization through DMFFT and RBPE

CO �� RBPE

Figure 16 :

26



Conclusion on use cases

� Graphs

� Modeling air interface standards as graphs

� Cost parameters and costs

� Optimization techniques

� local optimal vs global optimal� Confirm results found in other studies without the

graphical approach (these are usually local optimum s)� Here can be mixed/merged with others to find global

optimums.









o Types of costso Cost Function




27



Conclusions

� Ever increasing demands of users, manufactures and service

providers : Focus has been shifted toward SR approach.

� Optimization aspects of flexible radio system design : redu cing

the cost, complexity and size of the radio terminals/base-

stations

� We presented various case studies to validate our approach.

� We also elaborated how our approach can be used to find the

global optimal solution.



Perspectives

1. Scheduling issue

2. Investigation regarding other optimization algorithms e. g. GA

etc.

� Two new PhDs are starting to address 1 & 2.

3. More and more complete graphs

� More and more standards

� PHY layer was addressed

� Aim is to draw graph from application to PHY layer including

intermediate layers to exploit cross layer optimization aspects

4. Exploration of other techniques to formulate cost function .

5. Detailed study of costs for mixed designs consisting of FPGA ,

DSP etc.

28



Perspectives

6. Find new COs

7. Provide an industrial tool for designers of future SDR and CRequipments’

� That allows to optimize multi-standards SDR design &

� That provides a choice to orientate the design towards high

granularity or low granularity

� It is certainly interesting to follow the research directio nsmentioned above to further enhance the proposed ideas.



Publications

� International Journal Papers1. S. T. Gul , Ali Al-Ghouwayel, C. Moy and Y. Louët, “A Novel Design of Reconfigurable Fourier Transform

Operator Over CC and GF(GF(FFtt)) for Future Multi-standards SDR Equipments,” International Journal ofCommunication Networks and Distributed Systems-IJCNDS, Accepted in June 2009 for publication by the end of2009.

2. S. T. Gul , L. Alaus, C. Moy, J. Palicot and D. Noguet, “Optimal set of LFSR Common Operators for Multi-Standards Cognitive Radio Terminals,” International Journal of Autonomous and Adaptive Communications-IJAACS, Special Issue on Cognitive Radio Systems, Accepted in July 2009 for publication in 2010.

3. S. T. Gul , R. Mahesh, C. Moy, J. Palicot and A. P. Vinod “Filter Bank Techniques for SDR Channelizers andtheir Optimization using Graph Theoretical Approach,” EURASIP Journal on Advances in Signal Processing,Special Issue on Filter Banks for Next Generation Multicarrier Wireless Communications, Submitted, June, 2009.

� International Conference Papers1. S. T. Gul , C. Moy and J. Palicot, “Graphical Modeling and Optimization of Air Interface Standards for

Software Defined Radios,” 12th IEEE International Multitopic Conference-INMIC2008, Karachi, Pakistan, pp.473-479, December 2008.

2. S. T. Gul , R. Mahesh, C. Moy, A. P. Vinod and J. Palicot, “A Graphical Approach for the Optimization of SDRChannelizers,” International Union of Radio Science (Union Radio Scientifique Internationale-URSI); XXIXGeneral Assembly, Chicago, Illinois, USA, August 2008.

3. S. T. Gul , C. Moy and J. Palicot, “Two scenarios of flexible multi-standard architecture designs using amulti-granularity exploration,” The 18th Annual IEEE International Symposium on Personal, Indoor and MobileRadio Communication- PIMRC’07, Athens, Greece, pp. 1-5, September 2007.

� International Conference Workshops1. S. T. Gul , L. Alaus, D. Noguet, C. Moy and J. Palicot “The Common Operator Technique: An Optimization

Process to Identify and Design a Set of Common Operators to Perform SDR Equipment (projectNEWCOM++),” Dynamic Spectrum Management in Cognitive Radio Networks, ICT-MobileSummit 2009,Santander, Spain, June 2009.

29



Thanks for your attention

Questions?



Relative Weights of BC & CC

Documents

seminaire gul 29oct2009 - SUPELEC · Modèles Théoriques de Graphes pour les Equipments Multi-standards de Radio Logicielle ... o Algorithmes Génétiques Cas d'Etudes d'Opérateurs