Preswed14 Interesant

WORHSHOP EXPLORATORIU
Electronica, Telecomunicaiile si Tehnologia

Informaiei n Lume si n ar
22-23 Septembrie 2010
la
Facultatea de Electronic, Telecomunicaii i Tehnologia Informaiei
Universitatea Politehnica din Bucureti
Chairs:
Prof.dr.ing. Corneliu Burileanu, Prorector UPB
Prof.dr.ing. Teodor Petrescu, Decan ETTI
.l.dr.ing. Bogdan Ionescu, LAPI - UPB, LISTIC - UDS Frana
WORHSHOP EXPLORATORIU
Electronica, Telecomunicaiile i Tehnologia Informaiei

n Lume i n ar
22-23 septembrie 2010
Facultatea de Electronic, Telecomunicaii i Tehnologia Informaiei,
Corp A parter, Sala Orange, B-dul Iuliu Maniu 1-3,
Sector 6, Bucureti 061071, Romania
Miercuri, 22 Septembrie 2010

8:30- 11:00
Sesiunea 1
nregistrarea participanilor
Prof.dr.ing. Corneliu BURILEANU
Prorector - UPB
Deschiderea workshop-ului i prezentare program
Prof.dr.ing. Teodor PETRESCU
Decan - ETTI
Prezentarea direciilor de cercetare ale Facultii de Electronic, Telecomunicaii
i Tehnologia Informaiei
11:00-11:30
Pauz cafea
11:30-13:30
Sesiunea 2
Cerc.dr. Irina Diana COMAN.......................................................................pag.4

Free University of Bozen-Bolzano (Italia)
Sisteme Automate de Msurare i Analiz a Proceselor de Dezvoltare
Software
Drd.ing. Iuniana OPRESCU.......................................................................pag.35
LAAS - CNRS & Orange Labs (Frana)
Virtualizarea Planului de Control n Cadrul Protocolului BGP
.l.dr.ing. erban OBREJA.......................................................................pag.63
Catedra de Telecomunicaii - UPB
Internetul Viitorului
13:30-14:30
Pauz prnz
14:30-16:30
Sesiunea 3
.l.dr.ing. Cosmin Radu POPA.................................................................pag.88

Catedra de Dispozitive, Circuite i Aparate Electronice - UPB
Prelucrarea Inteligent a Informaiei
.l.dr.ing. Radu Mihnea UDREA.............................................................pag.118
Tehnologia Limbajului i Dialogul Om-Calculator
As.dr.ing. erban OPRIESCU...............................................................pag.142
LAPI, Catedra de Electronic Aplicat - UPB
Recunoaterea i Urmrirea Micrii folosind Camere ToF
16:30-17:00
Pauz cafea
17:00-18:00
Sesiunea 4
Conf.dr.ing. Constantin PALEOLOGU...................................................pag.155

Retele de Telecomunicaii Performante
.l.dr.ing. Nicolae Gheorghe MILITARU................................................pag.170

Dispozitive Electronice i Optoelectronice de nalt Performan
Drd.as.ing. Iulia Andreea MOCANU.......................................................pag.178
Aplicaii ale Metamaterialelor n Domeniul Microundelor
20:00
Sear Gala Cin restaurant Crama Domneasc

Str. elari nr. 13-15, Sector 3, Bucureti
(Centrul Istoric al Bucuretiului, n zona Lipscani, lng Muzeul Curtea Veche,
vis-a-vis de biserica Sf. Anton) http://www.cramadomneasca.com/.
Joi, 23 Septembrie 2010

8:30- 11:00
Sesiunea 5
Prof.dr.ing. Mihaela ULIERU..................................................................pag.192

Government of Canada's Science, Technology and Innovation Council (Canada)
How ICT is Reshaping the Economy
.l.dr.ing. Bogdan IONESCU..................................................................pag.205
LAPI, Catedra de Electronic Aplicat - UPB,
LISTIC, Polytech Annecy-Chambery (Frana)
Indexarea Automat dup Coninut a Secvenelor de Imagini
11:00-11:30
Pauza cafea
11:30-13:30
Sesiunea 6
Ing. Lucia STEFAN..................................................................................pag.231

Director Archiva Ltd (Anglia)
E-Archiving: Arhivistica Informaiei / Documentelor Electronice
As.dr.ing. Laura FLOREA.......................................................................pag.241
LAPI, Catedra de Electronic Aplicat - UPB
Prelucrarea i Analiza Imaginilor Medicale
13:30-14:30
Pauz prnz
14:30-16:30
Sesiunea 7
Prof. dr. ing. Grigore BOZGA.................................................................pag.263

UPB, coordonator Panel n cadrul proiectului DSE
Dezbatere Criteriile de Evaluare a Cercetrii pe Domenii Specifice n
Universiti- Proiect FSE Doctoratul n coli de Excelen
16:30-17:00
Pauz cafea
17:00-18:00
Sesiune special
Prof. dr. ing. Grigore BOZGA

UPB, coordonator Panel n cadrul proiectului DSE
Dezbatere Criteriile de Evaluare a Cercetrii pe Domenii Specifice n
Universiti- Proiect FSE Doctoratul n coli de Excelen
Sisteme automate de masurare si analiza a

proceselor de dezvoltare software
Irina Diana Coman

IrinaDiana.Coman@unibz.it
Septembrie 2010
Center for Applied Software Engineering

Free University of Bozen - Bolzano
Cuprins

Despre mine
Masurarea si analiza proceselor de dezvoltare
software
Domenii de interes
Studii
Studii
Studii
Studii
Studii
Studii
Studii
Studii
Publicatii
Carte

Irina Diana Coman, Adoption and Usage of AISEMA Systems: Automated InProcess Software Engineering Measurement and Analysis, Lambert Academic
Publishing, August 2010, ISBN 978-3-8383-7500-7
Publicatii
Carte

Irina Diana Coman, Adoption and Usage of AISEMA Systems: Automated InProcess Software Engineering Measurement and Analysis, Lambert Academic
Publishing, August 2010, ISBN 978-3-8383-7500-7
Articole in jurnal

Irina Diana Coman, Alberto Sillitti, Automated Segmentation of Development
Sessions into Task-related Subsections, Intl. Journal of Computers and
Applications, Vol. 31 (3), 2009.
Articole in conferinte (peer-reviewed, selectie)

Irina Diana Coman, Alberto Sillitti, Giancarlo Succi, A Case-study on Using an
Automated In-process Software Engineering Measurement and Analysis System
in an Industrial Environment, Proc. of 31st Intl. Conference on Software
Engineering (ICSE), May 2009.

Irina Diana Coman, Giancarlo Succi, An Exploratory Study of Developers' Toolbox
in an Agile Team, Proc. of 10th Intl. Conference on Agile Processes and eXtreme
Programming in Software Engineering (XP2009), May 2009.

Irina Diana Coman, Alberto Sillitti, Automated Identification of Tasks in
Development Sessions, Proc. of 16th Intl. Conference on Program
Comprehension (ICPC), June 2008.
Alte activitati
In 2009 si 2010, organizator al scolii de vara CASE

Summer School on Applied Software Engineering
(iulie, Bolzano).
Predare (FUB):

Cursul de Ingineria Programarii

Asistent pentru cursurile Managementul procesului de
dezvoltare software, Ingineria Programarii proiect
Membru al comitetului de program al European

Agent Systems Summer School (2009, 2010).
Reviewer pentru Intl. Conf. on eXtreme
Programming (2007).
Masurarea si analiza proceselor de

dezvoltare software

Suport decizional
Planificare
Evaluare
Imbunatatire
Comunicare
Personal Software Process(PSP), Team Software
Process (TSP)
Dar de ce sisteme automate?

PSP e foarte bun

PSP e foarte bun
DAR.

PSP e foarte bun
DAR.

PSP e foarte bun
DAR.

PSP e foarte bun
DAR.
Solutia: sisteme automate pentru colectarea si analiza datelor
Sisteme AISEMA
Automate
In proces
Datele disponibile imediate
Masurare in ingineria programarii

Neinvazive, ieftine, date precise si corecte
Metrici pentru proces si pentru produse
Analiza

(industrie) referinta, trenduri, situatie actuala, rapoarte

(cercetare) studiul si evaluarea metodologiilor si practicilor de
lucru, a aplicatiilor software si a imbunatatirilor propuse
Frequency
AISEMA pe scurt
De ce sa studiem AISEMA?

Potential mare, dar inca nefolosit.

Multe necunoscute:

Probleme:

Cerinte?
Probleme?
Beneficii concrete?
Fratele cel mare
Control
Date multe si ieftine

Utile si pentru cercetare?
Studiu de caz: adoptia si utilizarea

PROM
Mediu:

Departament IT in firma non-IT

10-20 programatori, team leader, manager
Durata: 2 ani
Scopuri:

Imbunatatirea procesului
Modalitate obiectiva si cuantificabila de a face
vizibila munca si contributia departamentului IT.
Rezultate

Procesul de adoptie detaliat in 5 pasi

Probleme la adoptie si modalitati concrete de
rezolvare (5 lectii si exemple concrete de beneficii)
Propuneri de imbunatatire a sistemelor AISEMA cu:

Auto-monitorare, diagnosticare si reparare

Inferenta automata a unor informatii de nivel inalt din
datele colectate
Analiza datelor pe 3 coordinate majore:

Profilul utilizatorului individual

Pair-Programming
Eficienta echipei si software-ul utilizat
Lectia 4 Vizualizarea datelor

Prezentarea datelor AISEMA:

Simplu;
Clar;
Sumar.
Agregarea datelor o directie de explorat.
Domenii de interes

Data mining
Masurari automate in ingineria programarii
Colectarea si analiza datelor
Asigurarea integritatii datelor
Vizualizarea datelor
Metrici si agregare
Practici si procese de dezvoltare software
Metode Agile, Pair-Programming
Sisteme autonome
Open Source
Mulumesc pentru atenie!
Lesson 1
How to gain support from developers:

Ensure data privacy.

Give developers full control over their own
data.
Give developers access to all information they
require about the system and its usage.
Take into account developers suggestions.
Offer developers a free choice on whether to
use the system.
Lesson 2
Be ready for a long start-up:

Planning can take very long (9 months):

Training takes time:

Busy calendars and different roles and backgrounds;

Possible extensions of the system needed;
People need time to get used with the idea of measurement
and being measured;
People need not to feel pressured to take a decision;
It takes time to identify and address peoples concerns;
Deployment can take very long (9 months):

Many issues surface only during intensive usage;

As people get used to the system, they come up with new
ideas for improvement and usage.
Lesson 3
Use any other data that are already being

collected:

Richer analyses;
Smooth adoption;
Requires flexibility of the AISEMA system.
Lesson 4
Data presentation is as important as data

accuracy:

Simplicity;
Clarity;
Brevity.
Developers, managers and team leaders will

NOT use the data unless the presentation is
adequate.
Lesson 5
Provide different aggregated views instead of

hierarchical breakdown:
Manager:

Leader:

Quality of software;
Measurement of performance of IT department.
Process evaluation;
Project status.
Developers:

Automation of time-consuming tasks;

Self-evaluation.
Lesson 6
Continuously check data accuracy;

Most time-consuming:

Software failures;
Changes to software systems;
Hardware failures.
Finding out that there IS a problem;
Identifying the source of the problem.
Solution:
Self-Healing.
Enhancements of AISEMA Systems

Ensuring continuous data accuracy

Self-healing capabilities
Automated inference of higher-level

information from low-level data
Automatically splitting development sessions into

task-related sub-sessions.
Self-healing Capabilities

Ensure continuous data accuracy

The system is more than its components:
Self-monitoring of each component is not enough!
The data are the main result:

Fast notification of problems (DataInspector):

Continuous data inspection

Data anomalies - signs of malfunctioning
Data invariants: static, dynamic , hiding complexity from user
Recovery (DataInspector and PROMConsole):

Identification of cause: check components and connectivity

Propagation of actions to client components
Inferring Task-related Sub-sections

How could we split a development session

into task related subsections?
Task = implementation of a specific, small
piece of functionality.
Task A
Task B
Identifying Number of Sub-sections
Steps:
1. Compute intervals of intensive access;
2. Find agglomerations of methods.
Inferring Sub-sections
Main Contributions of Thesis

1.
2.
3.
4.
5.
6.
7.
Description of adoption process of an AISEMA

system in industry.
Practical solutions to address issues in adoption
of AISEMA systems.
A model and implementation for enhancing
AISEMA systems with self-healing capabilities.
A method for automated detection of higherlevel information from AISEMA data.
Insights from AISEMA data analyses (user
working profiles, PP, tool usage and team
effectiveness).
Publications
Irina Diana Coman, Alberto Sillitti, Giancarlo Succi, A Case-study on Using an Automated
In-process Software Engineering Measurement and Analysis System in an Industrial
Environment, to appear in Proc. Intl. Conf. On Software Engineering (ICSE), May
2009.
Irina Diana Coman, Giancarlo Succi, An Exploratory Study of Developers Toolbox in an
Agile Team, to appear in Proc. XP2009, May 2009.
Irina Diana Coman, Alberto Sillitti, Automated Segmentation of Development Sessions
into Task-related Subsections, to appear in Intl. Journal of Computers and
Applications.
Irina Diana Coman, Alberto Sillitti, Automated Identification of Tasks in Development
Sessions, Proc. Intl. Conf. on Program Comprehension (ICPC), June 2008.
Irina Diana Coman, Alberto Sillitti, Giancarlo Succi, Investigating the Usefulness of PairProgramming in a Mature Agile Team, Proc. Intl. Conf. on Agile Processes and
eXtreme Programming in Software Engineering (XP), June 2008.
Irina Diana Coman, An Analysis of Developers Tasks using Low-Level, Automatically
Collected Data, Proc. of Doctoral Symposium at Joint Meeting European Software
Engineering Conf. and ACM SIGSOFT Symposium on Foundations of Software
Engineering, (ESEC/FSE), September 2007.
Irina Diana Coman, Alberto Sillitti, An Empirical Study on Inferring Developers Activities
from Low-Level Data, Proc. Intl. Conf. on Software Engineering and Knowledge
Engineering (SEKE), June 2007.
Thank you!
Research Questions

Requirements for adoption?

Challenges?
Benefits?
Full support of developers?
Automated inference of higher level information?
Usage of AISEMA data for studies of work
practices and tool usage?
Usage of AISEMA data for studies of PP in
industrial settings?
Answer: Requirements of Adoption

Flexibility to use all existing data

Different views for different roles
Complete data privacy
Full control of developers over their own data
Clear, synthetic data presentations in addition to
detailed ones
Team willing to accept long initial setup
Extensive training and time to accept the idea
Free and individual choice on the adoption
Answer: Challenges and Benefits

Challenges:

Gaining support of developers

Maintaining continuous data accuracy
Collecting data on remote and non-computer related
activities
Defining thresholds for software metrics
Inferring higher-level information (for example current
task)
Benefits:

Extensive data at low cost and no burden

High personal involvement
Initial baseline and automation of analyses
Answer: Full Support of Developers

Ensure data privacy

Full access to all information
Free and individual choice
Full control over ones own data
Consider all suggestions regarding features and
usages
Answer: Automated Inference

The activity-based method for automatically

splitting development sessions into task-related
subsections.
Medium level activities of developers (such as
exploring code or implementing a feature) are
often very much interleaved and thus hard to
detect automatically
Answer: AISEMA Data for Studies

Tool usage and work practices

Very useful for long time frame studies

Best when supplemented with qualitative data
PP in industrial context

Very useful for extensive, long-term studies

Allow trend studies as well as more focused studies
For some type of analyses additional qualitative data
are also needed
PROM
Open Issues
Aggregation of data:

Between 5 8 final numbers;

Ability to deal with data on different scales;
Adequate level of sensitivity;
Robustness to addition of new metrics.
Ensuring continuous data accuracy:

Self-healing capabilities.
Higher-level information:

1.
2.
Tasks and activities1, 2.

Coman, I. D. and Sillitti, A., An Empirical, Exploratory Study on Inferring Developers Activities from Low-Level Data. In Proc. of
19th Intl. Conference on Software Engineering and Knowledge Engineering, June 2007.
Coman, I. D. and Sillitti, A., Automated Identification of Tasks in Development Sessions, Proc. of 16th Intl. Conference on Program
Comprehension, June, 2008.
Goals Revisited
Manager:

Team leader:

Evaluate work products;

Make IT effort visible to non-IT manager.
Evaluate the status of projects;
Drive improvement of team based on objective
measurements.
Developers:

Automate time-consuming tasks;

Reliable, objective view of self-performance.
Data Invariant - Structure
Allows definition of complex data invariants on top of simpler existing

ones hiding complexity from the user.
Currently the definition of invariants is manually done by users.
Data Invariant - Examples
Self-healing: Open Issues

Automatically infer data invariants
Automatically infer possible causes of data

anomalies
Automatically design healing actions
Data
Data: stream of events of uninterrupted

working time with a single application and
file/class/method/location.
Time granularity: 1 sec.
Approach
Two hypotheses:
H1: The crucial locations for a task (core

locations) are accessed more than other locations
during the task solving;
H2: The core locations are intensively accessed
together during the solving of a task.
Two steps:

Identify number of subsections;

Identify boundaries of subsections.
Identifying Sub-sections
Intervals of intensive access: a method is

accessed for a minimum % of the time:

Degree of access (DOA) > threshold

Threshold = 1/3 *median(events)
The number of intensively accessed methods

at any given time:
1 subsection for each agglomeration of methods

simultaneously accessed intensively.
Data Analysis Pair-Programming

When do developers in mature Agile teams

find PP useful?
Is indeed PP considered useful for
knowledge transfer?
Study:
3 months, 14 veteran developers, 2 newcomers
Coman, I. D., Sillitti, A. and Succi, G. Investigating the Usefulness of Pair-Programming in a Mature Agile Team, Proc. of XP 2008.
Trends in PP
Does PP vary between iterations?
Is PP Used for Knowledge Transfer?

Group
Newcomers
Avg. PP of a
developer (1st
month)
74.33%
Avg. PP of a
developer (2nd
month)
29.31%
Veterans
52.68%
43.55%
Welch t-test for the difference between newcomers and veterans:

- statistical significant in the 1st month (p = 0.01)
- not statistical significant after the 2nd month (p = 0.1359)
Is PP Used for Knowledge Transfer?
Synthetic Data Presentation (example)
Analiza procesului de dezvoltare software

a
1
5

d
Identificarea cailor critice ale

proiectului, a activitatilor care
provoaca prelungirea proiectului.
Estimarea timpului si costului pentru
proiecte viitoare (Activity Based
Costing).
Documentarea activitatilor
efectuate.
Managementul calitatii (de ex. PSP,
TSP)
Un exemplu
Utente:
Mario Rossi
Data/Ora:
1 Luglio 2004, 16:10
Durata:
30 Minuti
# Caratteri aggiunti
5330
# Caratteri cancellati
2234

!
#
%&&

(
)
%(

+(

,(

%(

+(

,(

%

%

%

+

+

+

+

,.%/

2
1

#
3

#

28

!
%+(,7(%%+(
1

0

1

9
0
4%(56(%%+(%%%
5.%/

!
2
#

!

!

!
%+(,7(%%+(
1

0

1

0
4%(56(%%+(%%%
6.%/
0
%67
%+(,7(%%+(
0
55

4%(56(&(&.6
8
6,5
+%5
4,(%+,(%+(&.%,
8
6,5
65(+7(%,(4%.%/
8
%67
0
6,5
4%(56(%%+(%%%
0
+%5
7.%/
#

!

/.%/
(
)

+(

,(

#

3
)

:
;
3
##

<
! <

3 ##
1#

3
3#

=
>
#

1!
?#
!#

!
#=
#
0
3

>

?
1

#3
&.%/

9
9
1
1
1

!#
;
9 #
:
!
)!

#
!!
#

>

3

0

#

##

1

.%/

!
3
>

3
>

#

>

#
%.%/

3
!

#
#

3
>

3
)

1

>
+.%/
9

> !
> !
%
> !
+
> !
,
> !
> !
%
> !
+
> !
,
,.%/
9

3

#

3

> 1
3
!

#

# #
#

#
> !
%
> !
,
> !
+
> !
5.%/

> !)
1#

> !)
)
#
# # .

> !)

1

#

!
)

? #

##=
##
> !)
:
#

6.%/
(

!)
5(,(+(&.%,
> !
%
> !
,
> !
+
> !
7.%/
%(
#

>

#

!)
5(,(+(&.%,
/.%/
+(
> 3

> !)

#

> !)
!
5(,(+(&.%,
> !
2
4.%/
,(
#

>

1

?> !)
> !
%&.%/
5(
)
:

)
>

%.%/
6(
1
#

>

!

%%.%/
7(

9

9

9

9

9

9

9

9

9
%+.%/
(
)
%(

!
,(

%%.%/

!#
#

> !)
# #

)

>

>
30
%5.%/
(
)
%(

+(

#%&

%&&

#

#

<
#

#
)#>

;
!1
#<
3
3#
1

0 1

><
!
#

%&&<

>
%<
<

#

!

0;

<
>
#
;>
# #
!
;
#
<
!!<

?

1

> <
#1 <
#>
;

0<
;
? #
;
0

!1
#<
#

#<

!
;>
1
!

;

<
90;
%/.%/
Workshop: Electronica, Telecomunicaiile i

Tehnologia Informaiei n lume i n ar
Future Internet
Author:
Sl. dr. ing. erban Georgic Obreja
University Politehnica Bucharest

Bucureti,
UPB, ETTI
9/24/2010
Content
Future Internet (FI)

Internet evolution
Current Internet operation/limitations
KEY issues on Future Internet concepts and design
Future Content-Centric Internet Architecture
UPB Rersearch Projects related to FI

ENTHRONE FP6 IST-507637

End to End QoS through Integrated Management of Content

Networks and terminals
ALICANTE - FP7
MediA Ecosystem Deployment through Ubiquitous ContentAware Network Environments

9/24/2010
Bucureti,
UPB, ETTI
Internet evolution
Sources:
Nowadays is a global infrastructure supporting the economy

as well as the provision of societal services
Pervasive /ubiquitous;
Report from the National ICT Research Directors Working Group on Future
Internet (FI) of the EC on Information Society and Media- Nov. 2008
A.Peltomki, Stimulating an Innovation Ecosystem for Future Internet
Technologies, EC Information Society and Media Directorate-General, Lulea,
Sweden June 2009,
~ 25% of the world population - access to Internet
Mobility and nomadic usages are becoming the norm

By 2012, at least 1 Bn of Internet users will use mobile as their only

access means, adding to the today 1.5 Bn of fixed users
9/24/2010
Bucureti,
UPB, ETTI
Internet evolution
It has enabled user and consumer empowerment,

Tool for modernization of many domains:

through the emergence of eCommerce and social networks
eEducation, eGovernment, eHealth, etc.
Expected to contribute significantly to solve emerging

challenges such as climate change and energy efficiency.
It has favored innovation and the emergence of new
disruptive business models:
in 2008, ~300 million use VoIP Skype
9/24/2010
Bucureti,
UPB, ETTI
Internet evolution
Software: novel A/V consumption models

Web 2.0 and social networks are growing.

Popular social sites attract more than 120 million regular users.
Support for entrepreneurs creativity:

e.g. YouTube (global users downloading ~10 hours of video/min from

that site
its native openness, made possible thousands of innovation world wide,

to develop a huge range of applications
There is a recognized clear correlation factor between growth

of broadband Internet access and employment growth
9/24/2010
Bucureti,
UPB, ETTI
Current Internet
Today Internet usage

Content and services discovery & retrieval

Content delivery and streaming
Web services access
Users

Care about the content or services

Dont care about the service location
9/24/2010
Bucureti,
UPB, ETTI
Current Internet
Current Internet Architecture

Content Servers or Caches

Supporting Servers

Search Engines, DNS, AAA, DRM
Core and edge routers, Residential Gw

Fixed or mobile terminals
9/24/2010
Bucureti,
UPB, ETTI
Current Internet
9/24/2010
Bucureti,
UPB, ETTI
Current Internet
Content discovery by Search Engines

Content discovery by the User

find, classify and index content or services

Queries a Search Engine
Gets a number of URLs, where the content is
stores
Content Delivery/ Streaming

The user selects a URL

The content is delivered
9/24/2010
Bucureti,
UPB, ETTI
Current Internet
The actual architecture works

for current applications and usage

as long as sufficient resources available
Bandwidth, delay, etc
Main limitations factors

Growing number of users

Users demand high quality multimedia services
Increasing number of real time audio and
video communications
VoIP, videoconferencing
9/24/2010
Bucureti,
UPB, ETTI
Current Internet
Improve the Internet architecture

For more efficient use of the available resources
Provide an attractive business environment

Improvements
The content could be stored/cached closer to the end users
The routers could identify/analyse the content flowing through them
The network could identify what is the best path to the user
Less congestion, more bandwidth, lower delay
The content could be interactively adapted

Staticaly based on network and terminal capabilities

Dinamicaly the user can change the point of view (zoom) for a
streaming session
The content could be selected and adapted to the context

9/24/2010
Bucureti,
UPB, ETTI
FI design concepts
What we expect from FI?

New ways of multimedia content creation and

consumption
Future Internet will bring new experiences and

communications features
FI should

Accommodate services, methods, procedures and

techniques for creation and consumption of new
communications
Seamless, personalized multimedia communications
Access a grid of networks PAN, BAN, fixed, mobile
9/24/2010
Bucureti,
UPB, ETTI
FI design concepts
FI means New Experience

New Media

New
New
New
New
services
communications means
applications
streaming capabilities
9/24/2010
Bucureti,
UPB, ETTI
FI design concepts
FI will be content-centric
Content is clearly one of the main drivers of Internet

changes
Every year, the Internet traffic grows by 60%. This is
mainly due to video, and will be further amplified with
the advent of on-line 3D content.
FI content-centric
component interrelation
9/24/2010
Bucureti,
UPB, ETTI
FI design principles
FI may be realized in a vast number of

alternative ways
The need of virtualization
Need of design principles

Support multiple and new Business Model

Multiple stakeholders
Open environment
FI could host various embedded architectures
9/24/2010
Bucureti,
UPB, ETTI
Simplicity
In Internet intelligence at the edges to keep IP as

simple as possible
FI complex problems could require complex solutions
Sustainability, Scalability and Robustness

Special care designing the solutions could be less reliable

and flexible
Flexible enough to continuously evolve

To serve large number of entities
Easily recover if faults occur
Loose Coupling
9/24/2010
Bucureti,
UPB, ETTI
Excerpt of 7): Future Internet Towards Research Challenges 07

APR 2009, http://www.future-internet.eu/
9/24/2010
Bucureti,
UPB, ETTI
Future Content-Centric
Internet Architecture
Two approaches
First approach

Logical content-centric
Evolutionary approach
Second approach

Content-aware functionality
Forward-looking object approach
Clean-state approach
9/24/2010
Bucureti,
UPB, ETTI
Logical content centric approach

Intelligence will be moved in the network

Assumes a progressive rather than aggressive
evolution towards FI
It may consist of different virtual hierarchies of
nodes

Overlay, clouds or virtual groups of nodes

Nodes may belong to different hierachies
9/24/2010
Bucureti,
UPB, ETTI
9/24/2010
Bucureti,
UPB, ETTI

Network Provider Infrastructure

Network nodes routers, switches, access points, etc.

Limited functionality and intelligence
Users are connected as producers and consumers
Should hide all unnecessary complexity
Network topology, handover
Provide support for more intelligent nodes to take

necessary decisions in order to support the required
functionality
9/24/2010
Bucureti,
UPB, ETTI

Distributed Content/Services Aware Overlay

Consists of Content Aware Network Nodes

These nodes will have the intelligence
to filter the content and Web Services that flow through them
(via Deep Packet Inspection)
To identify streaming sessions and traffic (via signaling analysis)
Part of this information may be stored locally and/or

reported to the higher layer.
May be split in multiple overlays/clouds
Content caching, network monitoring, content adaptation,

content indexing, routing, etc.
9/24/2010
Bucureti,
UPB, ETTI

Information Overlay
Offers mechanisms for

Control the content and service construction, representation and

deployment
The content/service location/caching/deployment and network
resources information
Accounting and billing of content
Content publishing/subscription
Application overlay
Support for the applications

9/24/2010
Bucureti,
UPB, ETTI
Real Content Centric Architecture

Based on Content Objects

Content Object polymorfic container which may
consist of media, rules, behaviour, relations

Media anything that a human can perceived

Rules - the way the object is treated or manipulated
Behaviour - the way the object affects other objects
Relations between an object with other objects
Refers to time, space, synchronisation
Characteristics describe the objectand the relation

with other objects
Bucureti,
9/24/2010
UPB, ETTI
Real Content Centric Architecture example

Source: Van Jacobson, Diana K. Smetters, James D. Thornton, Michael F. Plass,

Nicholas H. Briggs, Rebecca L. Braynard, Networking Named Content, Palo Alto
Research Center, Palo Alto, CA, October 2009
Traditional
TCP/IP stack
CCN
9/24/2010
Bucureti,
UPB, ETTI
Real Content Centric Architecture - example

CCN packet types
9/24/2010
Bucureti,
UPB, ETTI

CCN forwarding engine model
9/24/2010
Bucureti,
UPB, ETTI

FIB
Forward Interest packets towards potential sources of

matching Data
Allows for a list of outgoing faces rather than a single
one (as in IP model)

CCN is not restricted to forwarding on a spanning tree

Allows multiple sources for data and can query them in parallel
9/24/2010
Bucureti,
UPB, ETTI

Content Store

Similar with the buffer memory for an IP router

Differences
IP packet belongs to a point-to-point conversation

No further value after being forwarded it is discarded
(IP packet it is forgotten)
CCN packets are potentially useful to many consumers
CCN remembers arriving Data packets as long as
possible (LRU or LFU replacement)
9/24/2010
Bucureti,
UPB, ETTI

PIT Pending Interest Table

Keeps track of Interests forwarded upstream towards

content source(s) so that the returned Data can be
sent downstream to its requester(s)
IN CCN only Interest packets are routed
As they propagate upstream towards potential Data sources,

they leave a trail of bread crumbs for a matching Data
packet
Every PIT entry is a bread crumb it is erased as soon as
they have been used to forward a matching Data Packet
9/24/2010
Bucureti,
UPB, ETTI
CAN/CCN : pros and cons

Pros

Adapted to the current characteristics of the FI from the

services point of view ( content and service orientation)
More flexible (policies applicable for different functions)
Decoupling identity location
Better security
Implementable : seamlessly or revolutionary-style
based on network virtualisations and overlays
(achievable in parallel Internet planes)
9/24/2010
Bucureti,
UPB, ETTI
CAN/CCN : pros and cons

Cons

No more network neutrality

Destroy the traditional TCP/IP stack layering concepts
(partially)
Significantly increase of the network nodes complexity (
speed problems in routers)
No more traditional concept : intelligence at the edge
and network: simple, stupid, but flexible
9/24/2010
Bucureti,
UPB, ETTI
UPB Rersearch Projects

related to FI
9/24/2010
Bucureti,
UPB, ETTI

related to FI

Integrated system for End to End QoS

management in Heterogeneous Networks
Several business actors

End to End QoS through Integrated Management of

Content Networks and terminals
Content Providers, Content Consumers, Service

Providers, Network Providers, Adaptation,
License Authority
Based on MPEG 21 metadata management

Heterogeneous networks core and access
SLA/SLS management
Bucureti,
9/24/2010
UPB, ETTI
ENTHRONE content delivery

chain
Bucureti,
UPB, ETTI
9/24/2010
Enthrone Integrated
Management System
EIMS - ENTHRONE Integrated Management Supervision
QoS i Adaptare
Administrare Metadate /
Cutare
Funcii avansate
Funcii auxiliare
EIMS Manager End to

End QoS
EIMS Manager
Metadate
EIMS Manager
Multicast
EIMS Manager Sesiune
EIMS Manager Servicii
EIMS Manager Cutare
EIMS Manager CDN
EIMS Manager Interfat

utilizator
EIMS Manager
Adaptare
Interfee
9/24/2010
Bucureti,
UPB, ETTI
Network Service Management

SP/NP
Administratori domenii de reea
EIMS@SP
NSM@SP
Furnizor
Servicii
SP
TM
EIMS@CP
EIMS@NP
EIMS@NP
NSM@NP
NSM@NP
Furnizor de
Coninut
CP
Reea core
1
Reea core
3
RM@ANP
omer
Reea de
acces
AN
Utilizatori
(sTVM)
Content
server
Conduct pSLS-link = conduct agregat,

multidomeniu, cu retricii de QoS
CPE
Conducte cSLS n AN
9/24/2010
Bucureti,
UPB, ETTI
Traffic trunks

NSM@NP has knowledge about network resources in terms of

traffic trunks
Intradomain/Interdomain traffic trunks
Between two domain egress points there are traffic trunks for
each QoS class
Bucureti,
9/24/2010
UPB, ETTI
Overlay topology path finding solution

EIMS@NP2
EIMS@NP3
EIMS@NP1
EIMS@NP6
EIMS@NP7
EIMS@NP4
EIMS@NP5
Overlay topology

Each network domain is abstracted to a node in the topology

The node is represented by the EIMS manager associated to the domain
Each node has a cost associated
The cost is calculated based on min/max delay&jitter, transit cost,
bandwidth capacity
The topology at a node is built by recursively interrogating each other
nodes about their directly connected neoghbors.
9/24/2010
Bucureti,
UPB, ETTI
Building the E2E QoS enabled pipe

First step

Builds the overlay topology

Computes several routes to each destination in the topology is
calculated based on shortest path algorithms
Second step

Performs QOS negotiation along the selected paths.

The chosen path is the path for which the negotiation is successful
The process of QoS negotiation is stopped when the first QoS
enabled path (pSLS pipe) is found
9/24/2010
Bucureti,
UPB, ETTI
Test platform- logical structure

Furnizor Servicii - SP
SP
Front-End
Terminal
Browser
E2EQoSMngr
SearchMngr
MetadataMngr
DIBrowser
CustSrvMngr@SP
TDM server
TDM client
PLayer
ResMngr@ANP
ServMon@SP
NetSrvMngr@SP
McastMngr@SP
E-Cast Node
/ X Layer
NetSrvMngr@NP
IP Multicast
IntraNetResMngr@NP
CustSrvMngr@TVM
ServMon@NP
E-Cast Node
AdaptMngr@CP
NetDimensioning
aTVM
aTVM
Reea Acces - AN
sTVM
Furnizor Coninut - CP
Furnizor Reea - NP
Bucureti,
UPB, ETTI
9/24/2010
Experimental test bed

BSOFT sTVM
To GEANT
Border_elcom
Border_an
aTVM
EIMS-NP1
E1
CP
subscribed
pSLS link
NetSrvMngr@SP
10.242..11.2
NetSrvMngr@NP
Invoked
pSLS link
Intra-NetResMngr@NP
E4
EIMS@ SP
CustSrvMngr
SrvMonSP
SearchMngr
E2EQoSMngr
DIDBrowser
E1
NP1
A1
TVM Media
stream
CP
2
6
Border_as1as2
E3
E6
3
7
Border_as1as3
Border_as2as1
11
10
12
A326 -
BSOFT player
10.242.103.100
End Users
Border_as3as1
EIMS-NP
EIMS-NP2
E1
NetSrvMngr@NP
Intra-NetResMngr@NP
10.
NetSrvMngr@NP
subscribed
cSLS link
Intra-NetResMngr@NP
Invoked
cSLS link
AN
NP2
11
W1
WiMAX mngr
Border_an103
9/24/2010
NP3
BSOFT player
10.242.105.100
End Users
Bucureti,
UPB, ETTI
Tests
9/24/2010
Bucureti,
UPB, ETTI

related to FI
ALICANTE FP6 IST-507637

ALICANTE, 2010-2013, FP& Integrated Project

(IP)
MediA Ecosystem Deployment Through
Ubiquitous Content-Aware Network
Environment- FI oriented project, 17 partners
9/24/2010
Bucureti,
UPB, ETTI
ALICANTE features
Applying new concepts ( Future Internet oriented) of

Content Aware Networking (CAN)

Network Aware Application (NAA)
novel virtual Content-Aware Network (CAN) layer

as a part of a full layered architecture

focused, but not limited to, on multimedia distribution with Quality of
Services (QoS) assurance
Flexible cooperation between
Providers/operators and end-users,

users access services/content, being consumers/providers

Business model: EU (CC, CP), CP, SP, NP, VCANP
9/24/2010
Bucureti,
UPB, ETTI
ALICANTE high level

architecture
9/24/2010
Bucureti,
UPB, ETTI
ALICANTE high level

architecture
Overall Architecture

User Env
Service Env
HB-layer
CAN layer
Infrastructure layer
9/24/2010
Bucureti,
UPB, ETTI
References

Report from the National ICT Research Directors Working Group on Future Internet (FI) of
the EC on Information Society and Media- Nov. 2008)
A.Peltomki, Stimulating an Innovation Ecosystem for Future Internet Technologies, EC
Information Society and Media Directorate-General, Lulea, Sweden June 2009
Schnwlder ,j. et al. , Future Internet = Content + Services + Management, IEEE
Communications Magazine, July 2009
FUTURE INTERNET ASSEMBLY, Madrid, Spain, 9th 10th December 2008, MEETING
REPORT, December 2008
DG Information Society and Media Directorate, for Converged Networks and Service Future
Internet 2020, VISIONS OF AN INDUSTRY EXPERT GROUP, May 2009
G. Tselentis et al. (Eds.), Towards the Future Internet, IOS Press, 2009
Future Internet Towards Research Challenges 07 APR 2009, http://www.futureinternet.eu/fileadmin/documents/prague_documents/FI_-_From_Functionalities2Challenges09_04_08.pdf
Future Internet Initiatives, http://www.nessi-europe.com/Nessi/ (Networked European
Software and sevices initiative)
Pavlou G., Towards a Service-aware Future Internet Architecture, Future Internet Assembly
Madrid, Dec 2008
Bucureti,
9/24/2010
UPB, ETTI
V Multumesc!
ntrebri?
9/24/2010
Bucureti,
UPB, ETTI
INTELLIGENT INFORMATION PROCESSING
Cosmin Radu Popa

Faculty of Electronics, Telecommunications and Information Technology
Bucharest, Romania
September 2010
0. OUTLINE
1. INTRODUCTION
2. LINEARIZATION OF DIFFERENTIAL AMPLIFIERS
3. LINEARIZATION OF ACTIVE RESISTIVE STRUCTURES
4. INCREASING THE PRECISION OF COMPUTATIONAL CIRCUITS
5. OPTIMIZATION OF MULTIFUNCTIONAL STRUCTURES
6. CONCLUSIONS AND FUTURE WORKS
0. OUTLINE
1. INTRODUCTION
1. INTRODUCTION
ADVANTAGES OF ANALOG INFORMATION PROCESSING
1. Low-power operation
2. Reduced area
3. High speed
1. INTRODUCTION
APPLICATIONS OF ANALOG INFORMATION PROCESSING
1. Smart sensors
2. Mobile computers
3. Mobile communication devices
4. Medical devices
1. INTRODUCTION
OBJECTIVES OF ANALOG INFORMATION PROCESSING
1. Increasing the precision of circuit functions
Implementing of linearization techniques for CMOS differential amplifiers
Reduction of linearity errors for active resistive structures
Minimizing the errors of computational circuits
Improving the performances of multifunctional structures
1. INTRODUCTION
2. Low-power operation
Weak inversion operation of MOS devices
Utilization of specific design techniques
Utilization of multifunctional structures
1. INTRODUCTION
3. Low-voltage operation
Current-mode operation
Utilization of DTMOS transistor (Dynamic Threshold MOS)

Increasing the linearity range
- differential amplifiers
- active resistive structures
- multipliers circuits
1. INTRODUCTION
4. Reduction of the silicon area
Utilization of active resistive structures
Bulk-drive of MOS devices
Utilization of FGMOS transistors (Floating Gate MOS)
Designing multifunctional structures
1. INTRODUCTION

LINEARIZATION OF DIFFERENTIAL AMPLIFIERS

Bulk-drive of MOS transistors
Constant sum of
gate-source voltages
Symmetrical structures
Utilization of
square-root circuits
Biasing of DA at a current
depending on vID
Biasing of DA at a voltage
depending on vID
Anti-parallel structure
2.1. Linear differential amplifier having a

proper biasing voltage

Block diagram
v1
IO
v2
IO
T
V
V2T
v1T
DA
iO
iO = Gm (v1 v2 )

iO = 2 KIO (v1 v2 ) =
Differential amplifier
V
V2T
v1T
iD1
iD2
iO
iO =
K
(v1T v2T )(2V v1T v2T 2VT
2

iO = 2 KIO (v1
Translation block
IO
IO
v1
v2
IO
IO
v1T
v2T
VC
v1T = v1 VC VT
v2T = v2 VC VT

iO = 2 KIO (
Arithmetical mean block
IO/2
IO/2
v1
v2
IO
IO
V
v + v2
V = 1
2

Linear differential amplifier
VDD
T11
T12
T10
T13
T13
IO/2
T
T14
T9
v1
T11
T12
T10
IO/2
T15
T15
T14
T9
v2
0
2i1
2i2
IO
IO
V
i1
i2
v1T
T5
IO
T1
T2
i1
i2
v2T
iO
T5
IO
IO
T16
T6
T4
T3
T3
T8
T2 T8
T2
T4
T6
DA
iO = 2 KIO (v1 v2 ) = Gm (v1 v2 )
2.2. Linear differential amplifier using

square-rooting circuits

Square-rooting circuit
VDD
VDD
i1
IO
IO
i2
v1
i1
i2
T2
T2
i1,2
v2
iO
IO
IO
IO
i1,2
T
2
T1,2
TO
1
-VDD
iO = Gm (v2 v1 )
G m = 2 KIO
i1 ,2 ' = 2 i1 ,2 IO
1. INTRODUCTION

10

OBJECTIVES IN THE DESIGN OF
ACTIVE RESISTIVE STRUCTURES
Reduction of silicon area
Linearity increasing
Increasing the maximal range
of the input voltage
The control of the equivalent resistance
using a reference voltage/current
Obtaining of a positive/negative
equivalent resistance
Improving of the power
supply rejection

CONCRETE METHODS FOR IMPLEMENTING
ACTIVE RESISTIVE STRUCTURES
Symmetrical structures
Complementary circuits
Utilization of linear differential amplifiers
Biasing of DA at a current dependent
on the differential input voltage
Biasing of DA at a voltage dependent
on the differential input voltage
Simulation of a linear I(V)
characteristic by mirroring
11
3.1. Symmetrical active resistive structure

Active resistive structure schematic
I2
I1
IXY
IXY
VX
I1
I0 I0
VY
I2
I
2I1
I0
V0
I01
I02
V0
I0
2I2
VC
12

Analysis of the active resistive structure operation
I1 =
I2 =
K VX VY VO
VT
+
2
2
2
K
2
VX VY VO
+
VT
2
2
I XY = I1 I 2 =
RECH. =
K
(VX VY )(VO 2VT )
2
VX VY
2
2
=
=
I XY
K (VO 2VT )
KVC

Second-order effects nonlinearity
Mobility degradation
K=
K0
1 + G (VGS VT )
I XY a1 (V X VY ) + a3 (V X VY )3
THD3 =
1 G
(VX VY )2
4 VC
Linearization technique
Principle: anti-parallel structure:
- changed inputs (VX V
Y)
- different biased (VC1 VC2 )
13
a11, 2
K VC 1, 2
2
I XY1, 2 a11, 2 (V X VY ) + a 31, 2 (V X VY )3

2
a31, 2
K G VC 1, 2
K G
+
8
8
I XY (a11 a12 )(V X VY ) + (a31 a32 )(V X VY )3
THD3 '
1 2
G (VX VY )2
2
THD3
1
THD3 ' 2 G VC
3.2. Active resistive structure using

complementary computational structures
14

X2
IOUT
IO
IXY
VX
IOUT =
IXY
IXY
K
(VX VY )2
4
I XY = 2 IOUT IO
VY
V VY
1
=
RECH . = X
I XY
KIO
1. INTRODUCTION
15
COMPUTATIONAL CIRCUITS
Multipliers
Quadratic core circuits
Logarithmic core circuits
Geometrical mean and
Euclidean distance circuits
Programmable function generators
4.1. Exponential circuits with third-order approximation
16

Exponential circuit with third-order approximation
X2 block
VDD
I
I 0 exp in
I0
I 0
I
I
1 + in + 1 in
I 0 2 I 0
1 I
+ in
6 I0
I0
Iin
Iout 1
I
I2
Iout 1 = I0 + in + in
2 16 I0
Iin
2
Iin
= 16 Iout1 16 I0 8 Iin
I0
I0
Iin
I0
Iout
Iout 1

X3 block
I out2 = I in +
2
I in
2 I0
3
I in
16 I 02
3
Iin
8I2
= 16 Iout 16 Iin in
2
I0
I02
VDD
Iin
Iin2 /I0
Iout 2
Iin
Iin2 /I0
I
I 0 exp in
I0
I 0 + 40 I 0 I out + 5 I in+ 8 I out

1
2
3
3
Iin
I0
Iout
Iout 2
17

VDD
I0
Iin
Iout 2
Iin
I0
Iout 1
Iin
Iin2 /I0
I0 exp (Iin/I0 )
1. INTRODUCTION

18
5.
OPTIMIZATION OF MULTIFUNCTIONAL STRUCTURES
Advantages:
The possibility of implementing multiple functions using the same core
Increasing the circuit modularity
Reduction of the silicon area and of the design costs per implemented function
Possibility of reconfiguring the circuit
Extension of the optimization techniques from the core to the entire
multifunctional structure
5.
Implemented functions:
Linear multifunctional structures
Signal gain with small distortions
Signal multiplying with low errors
Simulating a linear I(V) characteristic (RECH > 0, RECH < 0)
General multifunctional structures

Signal squaring
Signal square-rooting
Implementation of mathematical functions
19
5.
Objectives:
Linearity increasing
Increasing the maximal range of the input voltage
Precision increasing
Low-power low-voltage operation
5.
5.1. Linear multifunctional structure with current biasing
20
5.
Block diagram
CM
CM
iO
i1
CM
i2
v1
i2
v2
DA
i1
i1
i1 i2
v1
IO
i2
v2
DA
i1-i2
iO = Gm (v1 v2 )
i1-i2
IO
RECH = 1 / Gm
CM
iO
i1
v1
i2
CM
v2
DA
CM
i1
i2
IO
I
v1
X2
iO =
5.
i1
IOX
X2
v2 v3
v1
v4
i1 i2
i2
v2
DA
i2-i1
i2-i1
IO
K
(v1 v2 )(v3 v4 )
2
RECH = 1 / Gm

VDD
CM
T8
T7
iO
i1
i1
v1
i2
DA
iO
i2
v1
v2
v2
IO
i1
IO
X2
IO
21
5.

Drain currents:
I ' I '
i1 = O + O
2
2
K ( v1 v2 )2 K 2 ( v1 v2 )4
IO '
4 IO ' 2
I ' I '
i2 = O O
2
2
K ( v1 v2 )2 K 2 ( v1 v2 )4
IO '
4 IO ' 2
Differential output current:

iO =
5.
v1 v2
2
4 KIO ' K 2 ( v1 v2 )2

Biasing current:
IO ' = IO + I = IO +
K
( v1 v2 )2
4
resulting:
iO = KIO ( v1 v2 )
iO = Gm ( v1 v2 )
where:
Gm = KIO
22
5.

Squaring circuit
VDD
T10
4K
T11
4K
vS
vM
T1
v1
T2
vN
T3
T4
T5
4K
T7
4K
v2
T9
8K
I
-VDD
I =
5.
K
(v 1 v 2 )2
4
Squaring circuit
Average potential:
v +vN
vS = M
2
where:
v + V DD
vM = 1
2
output current:
I D5 =
v + V DD
vN = 2
2
I = I D 5 + I D7 I D 9
4K
(v M v 1 VT )2
2
resulting:
I D7 =
4K
(v N v 2 VT )2
2
8K
(V DD v S VT )2
I D9 =
2
I =
K
(v 1 v 2 )2
4
23
5.
Multiplication circuit
VDD
CM
iO
i1
T8
T7
i2
v2
DA
iO
i2
i1
v1
v1
v2
IO
IO
i1
I
v1
X2
v2 v3
IOX
IOX
X2
X2
v4
X2
v3
v4
iO = KIOX (v1 v2 )
IOX
5.
iO =
K
= (v3 v4 )2
4
K
(v1 v2 )(v3 v4 )
2
Positive active resistive structure

VDD
T8
T7
CM
T5
CM
i2
i1
v1
i1
v1
i1-i2
i1 i2
DA
T6
i1
i2
i2
i12
i1
T1
i1
i2
T3
T4
i2
T2
i12
v2
v2
IO
i1-i2
i1
IO
IO
X2
24
5.

Output differential current:
i12 = i1 i2 =
v1 v2
2
2 KIO ' K 2 ( v1 v2 )2
Biasing technique:
IO ' = IO + I = IO +
K
( v1 v2 )2
2
resulting:
i12 =
KIO
(v1 v2 )
2
RECH =
v1 v2
2
=
i12
KIO
Positive equivalent resistance:
5.
Negative active resistive structure

VDD
CM
i1
i2
i1
i1 i2
v1
T5
i2
i2
DA
i2-i1
T8
T7
CM
v2
i2-i1
v1
i12
T6
i1
T1
i1
i2
T3
T4
i1
i2
T2
i12
v2
IO
IO
i1
i12 = i2 i1
RECH =
IO
X2
v1 v2
2
=
i12
KIO
25
5.
5.2. General multifunctional structure with voltage biasing
5.
Block diagrams
v1
v2
v1
IO
IO
DA
DA
iO
iO = Gm (v1 v2 )
v2
v1
T
V
IO
IO
V2T
v2T
v1T
2IO
iO =
X2
DA
DA
iO
v3
v2
IO
T
v1T
iO
RECH = 1 / Gm
M
IO
v2T
v1T
iO
v1
V
V2T
v1T
v2
IO
IO
v4
K
(v v )(v v )
2 1 2 3 4
i1
i2
iO
iO =
K
(v1 v2 )2
4
26
5.

VDD
T11
v1
IO
T14
T9
v1
T13
T13
IO/2
IO
T10
IO/2
T15
T15
T14
T9
v2
V2T
v1T
T11
T12
T10
v2
T12
2i1
DA
2i2
IO
IO
V
i1
i2
v1T
iO
T5
IO
T1
T2
i1
i2
v2T
T5
iO
IO
IO
T4
T6
T16
T3
T4
T3
T8
T6
T2 T8
T2
DA
iO = 2 KIO (v1 v2 ) = Gm (v1 v2 )
5.
Squaring circuit
VDD
T11
v1
IO
IO
T14
T9
v1
T13
T13
IO/2
T10
IO/2
T15
T15
T14
2i1
V
i1
2IO
i2
v1T
i2
T5
IO
iO
v2
2i2
IO
IO
DA
i1
T9
v2T
v1T
T11
T12
T10
v2
T12
T1
T2
i1
i2
v2T
T5
IO
IO
T16
T6
T4
T3
T3
T2 T8
T8
2IO
T4
T6
iO
DA
iO =
K
(v1 v2 )2
4
27
5.
Multiplication circuit
VDD
T11
v1
v2
M
T
T14
T9
v1
T13
T13
T10
IO/2
IO/2
T15
T15
T14
T9
v2
V2T
v1T
T11
T12
T10
IO
IO
T12
2i1
DA
2i2
IO
IO
V
i1
iO
i2
v1T
T5
IO
T1
T2
i1
i2
v2T
T5
iO
IO
IO
v3
v4
T4
T6
T16
T3
T3
T8
T4
T6
T2 T8
T2
DA
iO = 2 KIO (v1 v2 )
iO =
K
IO = (v3 v4 )2
4
5.
K
(v1 v2 )(v3 v4 )
2

VDD
T11
T12
v1
IO
IO
T14
T9
T
T9
T14
v2
0
3i1
3i2
IO
IO
v2T
DA
i1
T1b
v1T
iO
T10
IO/2
T15
T15
V
v1T
T
v1
T13
T13
IO/2
v2
T11
T12
T10
iO
T5
IO
i2
T1b
i1
T1a
T1a
i1
i2
i2
v2T
T5
IO
IO
T4
T16
T3
T3
T8
T6
T4
T2a T8
T2a
T6
DA
iO
RECH =
T2b
T2b
iO
1
1
=
Gm
2 KIO
28
5.
Negative active resistive structure

VDD
T11
T12
T12
T10
v2
M
T
IO
T14
T9
v1
IO
3i1
iO
T15
T15
T9
T14
v2
3i 2
IO
IO
V
iO
v2T
i1
DA
iO
T10
IO /2
V
v1T
T13
T13
IO/2
v1
T11
T1b
v1T
i2
T1b
i1
T5
IO
T1a
T1a
i1
i2
i2
v2T
T5
IO
IO
T4
T16
T3
T3
T8
T6
T2a T8
T2a
T4
T6
DA
T2b
T2b
iO
RECH =
1
1
=
Gm
2 KIO
1. INTRODUCTION
29

Optimized structures
Differential amplifiers
Active resistive structures
Computational circuits
Multifunctional structures
Future works
Extending the capability of implementing mathematical functions
Finding new applications
Reduction of circuits complexity
Minimizing of errors introduced by the second-order effects
30
UNIVERSITATEA POLITEHNICA DIN BUCURETI

FACULTATEA DE ELECTRONIC, TELECOMUNICAII I
TEHNOLOGIA INFORMAIEI
Tehnologia Limbajului si Dialogul Om-Calculator
.l. Dr. ing. Radu Mihnea UDREA

Catedra de Telecomunicaii UPB
mihnea@comm.pub.ro
Introducere
Tehnologia limbajului vorbit

sisteme de tele/videoconferin,
telefonia mobil,
sisteme controlate vocal,
aplicaii biomedicale (proteze auditive)
achiziia vocii ntr-un mediu acustic (posibil)

afectat de perturbaii
Introducere
Sunt achiziionate i

ot
m
o
g
calea direct
zg
zgomotele de fond,
reflexiile semnalului
vocal din pereii sau
obiectele prezente n
sal (reverberaii),
semnalul vocal
retransmis (ecou)
om
o
re
fle
xii
ecou
Introducere
Algoritmii de reducere a zgomotului sunt necesari

pentru a mri calitatea perceput a vorbirii

pentru a mbunti anumii parametri n sistemele digitale
de prelucrare a semnalului vocal (recunoatere,
compresie).
Natura zgomotului este un factor important n

decizia asupra folosirii unei metode de mbuntire
a vorbirii.
Introducere
Clasificarea tipurilor de zgomot:

Zgomote de fundal zgomote acustice, aditive, generate

de micare, vibraii sau coliziuni (vehicule n micare,
aparate de aer condiionat, ventilatoare), persoane care
vorbesc n fundal, fenomene meteo etc.
Zgomote de impulsuri descrcri electrostatice,
zgomotul prezent n nregistrrile de arhiv.
Zgomotele de procesare zgomote rezultate n urma
procesrii analogice/digitale a semnalelor.
Zgomote neaditive zgomote datorate neliniaritii
microfoanelor, difuzoarelor i alte distorsiuni.
Metode de mbuntire
Tehnici de mbuntire folosind o singur surs

de intrare (un singur microfon)

se folosesc de diferite statistici ale semnalului vocal i ale

zgomotului.
performana acestor metode este n general limitat n
cazul nestaionaritii zgomotului.
performana scade drastic la RSZ mic.
reprezint situaii dificile deoarece nu este disponibil nici
un semnal de referin pentru zgomot.
Metode de mbuntire
Tehnici de mbuntire folosind mai multe

microfoane
au avantajul mai multor semnale de intrare i permit

folosirea unui semnal referin de zgomot ntr-un
dispozitiv de anulare adaptiv a zgomotului.
dac exist mai multe microfoane pot fi luate n
considerare proprietile spaiale ale semnalului i ale
sursei de zgomot.
sunt mai complexe.
Sisteme cu un singur microfon

Sistemele de mbuntire a vorbirii existente pot fi

mprite n patru categorii:

Reducerea zgomotului utiliznd periodicitatea vorbirii sau

a zgomotului;
mbuntirea bazat pe modelul vorbirii;
mbuntirea vorbirii bazat pe estimarea spectrului de
amplitudini pe termen scurt;
mbuntire bazat pe criteriul perceptual.
Metoda scderii spectrale

Se realizeaz scderea unui estimat al mediei

spectrului zgomotului din spectrul semnalului
afectat de zgomot.
Estimatorul scderii spectrale n amplitudine:
j ( k )
S (k ) = | Y (k ) | | D(k ) | e y
Estimatorul scderii spectrale de putere:

2
2
2
S ( k ) = Y ( k ) D ( )
Limitri ale scderii spectrale

Distorsiuni cauzate de variaii ale spectrului de

putere instantaneu al zgomotului fa de spectrul
mediu estimat (zgomot muzical).
Suprascderea spectral
Estimatul spectrului de putere pe termen scurt

[Berouti]:
S ( k )
Y k 2 D k 2
( )
( )
=
2
D ( k )
daca Y ( k ) > D ( k )
2
in rest
unde este factorul de suprascdere folosit pentru

supraestimarea spectrului de zgomot
iar este limita spectral inferioar.
Valoarea lui este funcie de estimatul aposteriori al

raportului semnal-zgomot:
4,75
3
= 0 RSZ
20
RSZ 5dB
5dB RSZ 20dB
RSZ 20dB
4
2
N 1
Y (k )
RSZ [dB ] = 10log10 Nk =01
2
D (k )
k =0
0 este valoarea dorit a lui la 0 dB RSZ.
1
20 10
10
20
30
RSZ
[dB]
Sistemul pentru scderea spectral

y (n) = s (n) + d (n)
arg{Y ( )}
Analiza
Y ( )
Segmentare
+ FFT
Sinteza
Limitare cu S ( )
IFFT +
factorul
suprapunere
sumare
D ( )
Scderea
spectral
Calculul
factorului a
s(n)
D ( )
Detecia activitii
vocale i estimarea
spectrului zgomotului
Spectrograma semnalului afectat de zgomot gausian

filtrat trece jos
Amplitudinea medie a zgomotului
3
2.5
2
1.5
1
0.5
0
0
500
1000
1500
2000
2500
Frecventa (Hz)
3000
3500
4000
Spectrograma semnalului dup scderea spectral a

mediei spectrului de putere a zgomotului
Amplitudinea medie a zgomotului
3
2.5
2
1.5
1
0.5
0
0
500
1000
1500
2000
2500
Frecventa (Hz)
3000
3500
4000
Spectrograma semnalului dup suprascderea

spectral cu factorul
Scderea spectral multiband

Zgomotul colorat afecteaz neuniform spectrul

semnalului vocal n diferite benzi de frecven.
Metod multi-band a scderii spectrale [1]
mprirea n benzi critice (scara Bark).

0
Benzi critice [kHz]

6
9
12
15
10 12 14 16 18 20 22 24
Uniti Bark
[1] R. M. Udrea, D. N. Vizireanu, S. Ciochina, S. Halunga Nonlinear spectral subtraction

method for colored noise reduction using multi-band Bark scale, Signal Processing,
Volume 88 Issue 5, Elsevier North-Holland, Inc. , May 2008, pp. 1299-1303.

Spectrograma semnalului afectat de zgomot colorat

Spectrograma semnalului dup suprascderea

multiband pentru 4 subbenzi
Metode bazate pe modelare statistic

Metode neliniare de estimare a spectrului de

amplitudine folosind diferite modele statistice i
criterii de optimizare

Estimatori de plauzibilitate maxim

Estimatori optimali n sens MMSE
Estimatori maximum a posteriori
Se bazeaz pe funcia densitate de probabilitate a

zgomotului i a coeficienilor TFD a semnalului
vocal i folosesc n anumite cazuri distribuii de tip
gausian
10
Metode bazate pe criteriul perceptual

Introducerea criteriilor perceptuale n procesul de

reducere a zgomotului [2]
in seama de efectele de mascare acustic a zgomotului

rezidual.
Pragul de mascare datorat semnalului vocal fr zgomot
nu este cunoscut.
Poate fi calculat din estimatul semnalului fr zgomot
dup aplicarea unei metode de mbuntire.
[2] R. M. Udrea, D. N. Vizireanu, S. Ciochina An improved spectral subtraction method for

speech enhancement using a perceptual weighting filter, Digital Signal Processing,
Volume 18, Issue 4, Academic Press, Inc. , July 2008 , pp. 581-587.
Pragul de mascare auditiv
11

Calcularea
pragului de
mascare
y ( n ) = s ( n) + d ( n )
Analiza
Segmentare
+ FFT
T ( )
Y ( )
Scderea
Spectral
Modificat
D ( )
Detecia activitii
vocale i estimarea
spectrului zgomotului
Ponderare
Perceptual
S ( )
Sinteza
IFFT +
Suprapunere
Sumare
s ( n )
G ( )
Faza ( )
12
Estimarea spectrului zgomotului

Pe durata pauzelor vorbirii

impune un detector de activitate vocal (Voice Activity

Detector - VAD) perfomant n medii zgomotoase
Estimarea spectrului zgomotului

Algoritmi de estimare continu a spectrului

zgomotului

Algoritmi de urmrire a minimului

Algoritmi de mediere temporal recursiv pentru
estimarea zgomotului
Algoritmi de mediere recursiv bazai pe incertitudinea
prezenei semnalului vocal
13
Algoritmi de urmrire a minimului

Nivelul puterii semnalului afectat de zgomot scade

aproape de nivelul puterii zgomotului n anumite
benzi de frecven, chiar n prezena semnalului
vocal.
Urmrind minimul spectrului de putere al
semnalului zgomotos n fiecare band de frecven,
se poate obine un estimat al nivelului zgomotului n
acea band
Algoritmi de mediere temporal

recursiv pentru estimarea zgomotului
Putem estima spectrul zgomotului pe benzi

individuale de frecven atunci cnd RSZ efectiv
pentru banda respectiv de frecven este foarte mic.
Spectrul zgomotului este estimat ca o medie
ponderat a estimailor anteriori ai zgomotului i
spectrului curent al zgomotului
d2 ( , k ) = a ( , k ) d2 ( 1, k ) + (1 a ( , k ) ) Y ( , k )
Ponderile a(,k) se calculeaz n funcie de

RSZ efectiv pentru fiecare component spectral.

probabilitatea prezenei semnalului vocal.
14
Algoritmi de mediere temporal

recursiv pentru estimarea zgomotului
Rezultate experimentale
Zgomote de diverse tipuri (multivorbitor, motor de

automobil, gar, tren) au fost suprapuse peste
semnalele vocale cu rapoarte semnal-zgomot de
intrare de 0dB, 5dB, 10dB, 15dB.
Semnalul vocal i zgomotul au fost eantionate cu
frecvena de eantionare de 8kHz.
15
Semnalul afectat de zgomot

The sky that morning was clear and bright blue
Zgomot nregistrat n tren, RSZ 10dB
Metoda scderii spectrale n amplitudine
16
Metoda supra-scderii spectrale
Metoda scderii spectrale multiband

descompunere uniform n frecven
17
Metoda scderii spectrale multiband

descompunere pe scara Bark
Metoda estimrii MMSE
18
Metoda estimrii log-MMSE
Metoda estimrii log-MMSE

(cu estimarea prezenei semnalului vocal)
19
Evaluarea calitii semnalului mbuntit

Metodele obiective dau o msur comparativ de

apreciere a mbuntirii semnalului.
Metodele folosite n evaluarea algoritmilor sunt:

Metoda anvelopei spectrale ponderate (WSS - Weighted

Spectral Slope);
Metoda raportului de plauzibilitate logaritmic (LLR Log-likelihood ratio);
Metoda distanei Itakura-Saito (IS);
Metoda de evaluare perceptual a calitii semnalului
vocal (PESQ - Perceptual Evaluation of Speech Quality).
Evaluarea calitii semnalului mbuntit
20
Scderea spectral cu estimarea

continu a zgomotului
Simulrile au fost validate prin experimente n timp

real efectuate pe un procesor de semnal n virgul
fix din familia Freescale MSC8101.
A fost studiat implementarea pe procesor folosind

limbajul C orientat pe arhitectura DSP-ului combinat cu
funcii implementate direct n asamblare.
Au fost propuse tehnici eficiente de optimizare a
implementrii n sensul reducerii memoriei ocupate de
date dar i a timpului de execuie prin reducerea
complexitii operaiilor.
21
Continuarea cercetrii
Studiul tehnicilor de
reducere a zgomotului i a
reverberaiilor, folosind mai
multe microfoane,
o
zg
ot
m
pot exploata att

caracteristicile spectrale ct
i spaiale ale semnalelor
achiziionate de microfoane.
algoritmi de tip adaptiv,
rspunznd astfel la
manifestarea diferitelor surse
de zgomot sau la schimbrile
din mediul acustic ambiant
calea direct
zg
om
o
re
fle
xii
ecou
ir de microfoane
Concluzii

Proiecte de cercetare-dezvoltare pe baz de

contract/grant (CNCSIS-PNII-IDEI, UE-FP6).
Bursa postdoctorala (FSE-POS DRU)
Rezultatele cercetrii au fost publicate n 14 articole
(ca prim autor)

2 articole n reviste naionale

4 articole n conferine naionale
6 articole prezentate la conferine internaionale indexate
ISI
2 articole n reviste indexate ISI
22
Referine
R. M. Udrea Teza de doctorat Metode de reducere a perturbailor n

sistemele de prelucrare a semnalului vocal, UPB, 2009
R. M. Udrea, D. N. Vizireanu, S. Ciochina, S. Halunga Nonlinear spectral
subtraction method for colored noise reduction using multi-band Bark scale,
Signal Processing, Volume 88 Issue 5, Elsevier North-Holland, Inc. , May
2008, pp. 1299-1303.
R. M. Udrea, D. N. Vizireanu, S. Ciochina An improved spectral subtraction
method for speech enhancement using a perceptual weighting filter, Digital
Signal Processing, Volume 18, Issue 4, Academic Press, Inc. , July 2008 , pp.
581-587.
R. M. UDREA, D.N. Vizireanu, I. Pirnog , A Perceptual Approach for Noise
Reduction using Nonlinear Spectral Subtraction, 8th International Conference
on telecommunications in Modern Satellite, Cable and Broadcasting Services
TELSIKS2007, Nis, Serbia si Muntenegru, 26-28 September, 2007, p. 649653.
Referine
R. M. UDREA, S. Ciochin, D. N. Vizireanu, Multi-band Bark Scale Spectral

Over-Subtraction for Colored Noise Reduction, ISSCS 2005, IEEE
International Symposium on Signals Circuits and Systems, 14-15 iulie, 2005,
Iai, Romnia, Vol I, pp. 311-314.
R. M. UDREA, S. Ciochin, D. N. Vizireanu, Reduction of Backgroung Noise
from Affected Speech using a Spectral Subtraction Algorithm based on Masking
Properties of the Human Ear, 7th International Conference on
telecommunications in Modern Satellite, Cable and Broadcasting Services
TELSIKS2005, Nis, Serbia si Muntenegru, 28-30 September, 2005, pp. 135138.
R. M. UDREA, D. N. Vizireanu, M. Rducanu, Noise Reduction using
Spectral Over-Subtraction Algorithm implemented on Motorola
MSC8101DSP, Proceedings of the International Conference Communications
2004, Bucureti, Vol. 1, pp. 193-199, 3-5 Iunie, 2004.
R. M. UDREA, D. N. Vizireanu, Speech Enhancement Using a Modified
Spectral Over-Subtraction Algorithm implemented on DSP, 27th International
Conference on Fundamentals of Electronics and Circuit Theory, IC-SPETO
2004, Gliwice - Niedzica, Polonia, Vol II, pp. 395-398, 26-29 Mai, 2004.
23
Referine
R. M. UDREA, S. Ciochin, Speech Enhancement Using Spectral

Oversubtraction and Residual Noise Reduction, IEEE International
Symposium on Signals Circuits and Systems SCS 2003, Iai, Romnia, Vol I,
pp. 165-169, 10-11 Iulie, 2003.
R. M. UDREA, S. Ciochin, D. N. Vizireanu, An Improved Speech
Enhancement Method using Spectral Oversubtraction and Perceptual
Weighting, Proceedings of the Symposium Etc. 2004, Timioara, Romania,
pp. 315-319, 22-23 Octombrie, 2004,
R. M. UDREA, S. Ciochin, D. N. Vizireanu, Modified Noise Reduction
Algorithm using Spectral Over-Subtraction, The 30th Internationally Attended
Scientific Conference of the Military Technical Academy, Bucureti, Romania,
pp. 247-253, 6-7 Noiembrie 2003.
R. M. UDREA, S. Ciochin, Implementation of an Optimized Algorithm for
Acoustic Noise Reduction Based on Hartley Transform, Buletinul tiinific al
Universitii "Politehnica" din Timioara, Timioara, Romania, pp. 223-227,
Septembrie 2000.
Va mulumesc pentru atenie!
24
Recunoasterea si Urmrirea Miscrii

folosind Camere ToF
As.dr.ing. Serban OPRISESCU,
Laboratorul de Analiza si Prelucrarea Imaginilor (LAPI)
http://imag.pub.ro
Facultatea de Electronica, Telecomunicatii si Tehnologia Informatiei,
Universitatea Politehnica din Bucuresti
Email: soprisescu [la] imag.pub.ro
Introducere
Camerele Time-of-Flight (ToF):
sisteme active
inregistreaza in fiecare pixel intensitatea si
distanta obiect-camera;
Pot fi utilizate cu succes pentru
detectie si recunoastere de actiuni
supraveghere video
monitoirzarea respiratiei etc.
www.artts.eu
IST-34107 (2006-2009)
Principiul camerei ToF
Cateva ecuatii
-Esantionam semnalul receptionat de 4 ori/perioada: A0,A1,A2,A3:
-Defazajul emisie-receptie:
-Distanta in fiecare pixel:
d=
c T
4
-Intensitatea in fiecare pixel:
Exemplu: f=20MHz => T=50ns

obtinem distanta maxima: d=7,5m
Probleme ale camerei ToF

Imaginea de distanta si cea de amplitudine
au cateva inconveniente:
Intensitatea scade cu patratul distantei:
obiectele cu reflectivitate similara se vor
vedea mai intunecate la distanta.
Erori sistematice in masurarea distantei: erori
mai mari pentru obiecte foarte intunecate.
Exemplu de scadere a intensitatii

cu distanta
Imagine de intensitate ToF
Poza cu webcam
Exemplu de erori de distanta
Distanta ToF
Intensitatea ToF
Corectia intensitatii functie de distanta:

2
A' (i, j ) = A(i, j ) D ANF
(i, j )
Imagine originala
Imagine corectata
Utilizand o imagine de distanta filtrata cu ANF:
Corectia erorilor de distanta:
Inainte de corectie
Dupa corectie
Utilizarea camerei ToF pentru

supraveghere video
Detectia prezentei de persoane pe baza histogramei
- Tehnica comparrii histogramei
- Numrarea modurilor histogramei
Tehnici de detecie si estimare a miscarii
Dezvoltarea unui program de urmrire a miscarii mai
multor persoane simultan
Detectia prezentei bazata pe histograma
Scena vida
(imaginea de distanta ToF)
Histogramele corespunzatoare
O persoana intrnd in scena
Detectia bazata pe histograma

1) Tehnica comparrii histogramei
SDi =
1
GM N
H ( j) H
i
i +1
( j)
j =1
unde Hi(j) = valoarea histogramei pentru frame i si nivel de gri j [0,G]
SDi
T = +
Nr frame (i [1,400])
Detectia bazata pe miscare

Algoritmul Block-matching
min (d m ),
d mP
(d m ) =
( n) I k 1 ( n d m ) , m,
nBm
Gsirea celui mai bun vector de deplasare dm pentru fiecare bloc Bm dintr-o
zona de cutare P, P = {n = (n1 , n 2 ) : P n1 P, P n 2 P},
folosind o metoda de cutare
A fost implementat un algoritm foarte rapid:
Cautarea blocului (estimarea) se face doar daca detectam miscare
(zero motion detection)
Rezultate
a) Frame-uri cu detectie
pozitiva (adevr teren)
b) Diferentele de
histograma
c) Decizia bazata pe b)
d) Detectia bazata pe
miscare
e) Numararea modurilor
histogramei
Supraveghere video Aplicatie
Urmarirea miscarii Aplicatie
Folosirea camerei ToF pentru

recunoasterea de actiuni
Avantaje:
- Segmentare directa din imaginea de distanta
- vedere pe timp de noapte
- traiectorii 3D
Metoda:
Caracteristici extrase din traiectoriile punctelor cheie
Extragerea siluetei -> Skeleton -> Puncte cheie
Imagine de intensitate
Inagine de distanta
centroid
Imagine segmentata
Skeleton cu puncte cheie
Sarcina: decizia intre 6 actiuni
WALK
CARRY
RUN
BENT
JUMP
BOX
Traiectoriile punctelor cheie
Traiectorii centroid pe x/y
Traiectorii puncte cheie pe x
10
Traiectorii -> Caracteristici

1) Variatia unei functii:
f= fMax-fMin
2) Variatia totala v a unei functii:
N
V = f (k ) f (k 1)
k =2
3) Viteza medie reala, calculata ca medie a vitezelor instantanee din fiecare

frame:
1 N
S r = ( f ( k ) f (k 1) ) / t
N k =2
4) Viteza medie absoluta, calculata ca media valorii absolute a fiecrei viteze
instantanee:
Sa =
1
N
k =2
( f (k ) f (k 1))
t
Arbore de decizie
Sax , Say = viteze absolute pe
x si y
Srx = viteza medie centroid
pe x;
SaH = viteza medie absoluta
a mainii;
TvsHx = variatia totala a
mainii pe x;
Vy = variatia totala a
centroidului pe y.
T1,, T14 = praguri
calculate intr-o etapa de
invatare (ex. pe walk).
? = actiune necunoscuta.
11
Matricea de confuzie de mai sus a fost calculata testand algoritmul de

recunoatere pe 60 de secvene, cate 10 secvene per aciune.
Exemple
12
Cercetarea continua in cadrul bursei post-doctorale tip EXCEL*

Cu tema Dezvoltarea unor sisteme cu camere ToF (Time of Flight) i
fuziune de senzori 2D/3D pentru aplicaii de monitorizare
Va mulumesc pentru atenie

* Proiect cofinanat din Fondul Social European prin Programul Operaional Sectorial pentru
Dezvoltarea Resurselor Umane 2007 2013
Axa prioritar: Educaia i formarea profesional n sprijinul creterii economice i dezvoltrii societii
bazate pe cunoatere
Domeniul major de intervenie: Programe doctorale i postdoctorale n sprijinul cercetrii
Titlul proiectului: Excelen n cercetare prin programe post-doctorale n domenii prioritare ale societii
bazate pe cunoatere (EXCEL)
Cod Contract: POSDRU/89/1.5/S/62557
Beneficiar: Universitatea POLITEHNICA din Bucureti
Excelen n cercetare prin programe postdoctorale
n domenii prioritare ale societii bazate pe cunoatere
EXCEL
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
REFERINTE
W. Hu, T. Tan, L. Wang and S. Maybank, A survey on visual surveillance of object motion and behaviors, IEEE Trans.
on Systems, Man, and Cybernetics Part C, 34(3), pp. 334-352, 2004.
A. Ali and J.K. Aggarwal, Segmentation and recognition of continuous human activity, Workshop on Detection and
Recognition of Events in Video, pp. 2835, 2001.
S. Oprisescu, D. Falie, M. Ciuc and V. Buzuloiu, Measurements with ToF cameras and their necessary corrections,
Proceedings of IEEE ISSCS Conference, vol 1, pp 221-224, Iasi, 2007.
D. Falie and V. Buzuloiu, Distance errors correction for the Time of Flight (ToF) cameras, IEEE Workshop on Imaging
Systems and Techniques, 2008. IST 2008, Crete, pp. 123-126, 2008.
S. Oprisescu, C. Burlacu, V. Buzuloiu and M. Ivanovici - Action recognition for simple and complex actions using time
of flight cameras, IPCV09 - Int. Conf. on Image Processing, Comp. Vision and Patt. Recognition, Las Vegas, USA, July
13-16, 2009.
J.K. Aggarwal and Q. Cai, Human motion analysis: a review, CVIU Journal, vol. 73, no.3, pp. 428440, 1999.
J.J. Wang and S. Singh, Video analysis of human dynamics - a survey, Real-Time Imaging, vol. 9, no. 5, pp. 321346,
2003.
E. Yu and J.K. Aggarwal, Detection of fence climbing from monocular video, 18th International Conference on Pattern
Recognition (ICPR'06), Hong Kong, vol. 1, 2006, pp. 375-378.
M. Blank, L. Gorelick, E. Shechtman, M. Irani, R. Basri, Actions as space-time shapes, 10th IEEE Int. Conf. on
Computer Vision (ICCV'05) vol. 2, 2005, pp.1395-1402.
L. Gorelick, M. Galun, E. Sharon, A. Brandt and R. Basri, Shape representation and classification using the Poisson
equation, IEEE Trans. Patt. An. & Mac. Intell., vol. 28, no. 12, pp. 1991-2005, 2006.
Nazl kizler and D. A. Forsyth, Searching for complex human activities with no visual examples, International Journal
of Computer Vision, vol. 80, no 3, pp. 337-357, Dec. 2008.
C. Burlacu, V. Buzuloiu, and C. Vertan, Color image segmentation by featured key-points, Proc. Int Conf.
Communications 2004, vol. 1, pp. 243-244, Bucuresti, Romania, June 3-5, 2004.
M. van Eede, D. Macrini, A. Telea, C. Sminchisescu and S. Dickinson, Canonical skeletons for shape matching, 18th
International Conference on Pattern Recognition (ICPR'06) vol. 2, 2006, pp. 64-69.
A. Yilmaz and M. Shah, Recognizing human actions in videos acquired by uncalibrated moving cameras, 10th IEEE Int.
Conf. on Computer Vision (ICCV'05) vol. 1, 2005, pp. 150-157.
13
Speech Enhancement in Hands-Free Communication Devices
Adaptive Algorithms for Acoustic

Echo Cancellation
Constantin Paleologu
Department of Telecommunications
pale@comm.pub.ro
Introduction
acoustic echo cancellation (AEC)
required in hands-free communication devices, (e.g., for mobile
telephony or teleconferencing systems)
! acoustic coupling between the loudspeaker and microphone
an adaptive filter identifies the acoustic echo path between the
terminals loudspeaker and microphone
specific problems in AEC
the echo path can be extremely long
it may rapidly change at any time during the connection
the background noise can be strong and non-stationary
important issue in echo cancellation
the behaviour during double-talk
the presence of Double-Talk Detector (DTD)
Introduction
(cont.)
d(n)
x(n)
y(n)
Digital
filter
e(n)
Adaptive
algorithm
Adaptive filter
Cost function J[e(n)] minimized
e( n ) = d ( n ) y ( n )
Introduction
(cont.)
AEC configuration
x(n)
Far-end
(n)
Adaptive filter
to the Far-end
v(n)
DTD
h
Acoustic echo path
synthetic
echo
acoustic
echo
e(n)
d(n)
microphone signal
Background
noise
Near-end
v(n)
Performance criteria: - convergence rate vs. misadjustment

- tracking vs. robustness
(a)
0.01
0.005
0
-0.005
-0.01
0
100
200
300
400
500
600
esantioane
Samples
(b)
700
800
900
1000
-20
dB
-30
-40
-50
-60
0
0.5
1.5
2.5
3.5
frecventa [kHz]
Frequency
[kHz]
Fig. 1. Acoustic echo path: (a) impulse response; (b) frequency response.
(a)
0.1
0.05
-0.05
0
100
200
300
400
500
600
esantioane
Samples
(b)
700
800
900
1000
dB
-20
-40
0
0.5
1.5
2.5
3.5
frecventa [kHz]
Frequency
[kHz]
(a)
0.1
0.05
0
-0.05
-0.1
0
500
1000
1500
esantioane
Samples
(b)
2000
2500
3000
10
dB
0
-10
-20
-30
0
0.5
1.5
2.5
3.5
frecventa [kHz]
Frequency
[kHz]
Adaptive algorithms for AEC

requirements
fast convergence rate and tracking
low misadjustment
double-talk robustness
most common choices
normalized least-mean-square (NLMS) algorithm
step-size parameter
affine projection algorithm (APA)
(n) = (n 1) +
0<1
update-term
step-size parameter (controls the performance of these algorithms)

large values fast convergence rate and tracking
small values low misadjustment and double-talk robustness
conflicting requirements variable step-size (VSS) algorithms
x(n)
(far-end)
(n)
e(n)
(to the far-end)
e(n) = 0
v(n)
(near-end)
e(n) = v(n)
( n) = 1
v ( n )
e ( n )
?
e2 ( n ) = e2 ( n 1) + (1 ) e 2 ( n )
[J. Benesty et al, A nonparametric VSS NLMS algorithm, IEEE Signal Process. Lett., 2006]
1) near-end signal = background noise (single-talk scenario)

v(n) = w(n)
background noise power estimate
w
( n) = 1
e ( n )
[J. Benesty et al, A nonparametric VSS NLMS algorithm, IEEE Signal Process. Lett., 2006]
NPVSS-NLMS algorithm
Problem: background noise can be time-variant
2) near-end signal = background noise + near-end speech
v(n) = w(n) + u(n)
(double-talk scenario)
v2 ( n ) = w2 ( n ) + u2 ( n )
near-end speech power estimate
???
Problem: non-stationary character of the speech signal
Solutions for evaluating the near-end signal power estimate
v2 ( n ) = ?
1. using the error signal e(n), with a larger value of the weighting factor:
e2 ( n ) = e2 ( n 1) + (1 ) e2 ( n )
= 1 1/(KL), with K > 1
v2 ( n ) = v2 ( n 1) + (1 ) e2 ( n )
= 1 1/(QL), with Q > K
simple VSS-NLMS (SVSS-NLMS) algorithm
SVSS ( n ) = 1
v ( n )
e ( n )
! The value of influences the overall behaviour of the algorithm.
2. using a normalized cross-correlation based echo path change detector:

v2 ( n ) = e2 ( n )
x2 ( n )
reTx ( n ) rex ( n )
x2 ( n ) = x2 ( n 1) + (1 ) x 2 ( n )
rex ( n ) = rex ( n 1) + (1 ) x ( n ) e ( n )
NEW-NPVSS-NLMS algorithm
v ( n )
if ( n ) <
1
NEW NPVSS ( n ) = e ( n )
1
otherwise
[M. A. Iqbal et al, Novel Variable Step Size NLMS Algorithms for Echo Cancellation ,
Proc. IEEE ICASSP, 2008]
red n e2 n
d2 ( n ) = d2 ( n 1) + (1 ) d 2 ( n )
(n) =
d2
( )
( )
( n ) red ( n )
red ( n ) = red ( n 1) + (1 ) e ( n ) d ( n )
convergence statistic
! The value of influences the overall behaviour of the algorithm.
3.
(n) = 1
d2 ( n ) 2y ( n )
e ( n )
Practical VSS-NLMS (PVSS-NLMS) algorithm

[C. Paleologu, S. Ciochina, and J. Benesty,Variable Step-Size NLMS Algorithm for
Under-Modeling Acoustic Echo Cancellation , IEEE Signal Process. Lett., 2008]
VSS affine projection algorithm (VSS-APA)

[C. Paleologu, J. Benesty, and S. Ciochina,Variable Step-Size Affine Projection Algorithm
Designed for Acoustic Echo Cancellation , IEEE Trans. Audio, Speech, Language Process.,
Nov. 2008]
main advantages
non-parametric algorithms
robustness to background noise variations and double-talk
they assume that the adaptive filter has converged to a certain degree.
Table I. Computational complexities of the different variable-step sizes.

Algorithms
Additions
Multiplications
Divisions
Square-roots
NPVSS-NLMS
SVSS-NLMS
NEW-NPVSSNLMS
2L + 8
3L + 12
PVSS-NLMS
L = adaptive filter length
Simulation results
conditions
Acoustic echo cancellation (AEC) context, L = 1000.
input signal x(n) AR(1) signal or speech sequence.
background noise w(n) independent white Gaussian noise
signal (variable SNR)
measure of performance normalized misalignment (dB)
20log (|| h h (n) || / || h ||)
10
algorithms for comparisons

NLMS
NPVSS-NLMS
SVSS-NLMS
NEW-NPVSS-NLMS
PVSS-NLMS
0
NLMS w ith step-size (a)
NLMS w ith step-size (b)
-5
NPVSS-NLMS
Misalignment (dB)
-10
-15
-20
-25
-30
-35
-40
10
15
20
25
30
Time (seconds)
Fig. 4. Misalignment of the NLMS algorithm with two different step sizes
(a) = 1 and (b) = 0.05 , and misalignment of the NPVSS-NLMS algorithm.
The input signal is an AR(1) process, L = 1000, = 1 1/(6L), and SNR = 20 dB.
5
NLMS w ith step-size (a)
NLMS w ith step-size (b)
NPVSS-NLMS
Misalignment (dB)
-5
-10
-15
-20
-25
-30
-35
-40
10
15
20
25
30
Time (seconds)
Fig. 5. Misalignment of the NLMS algorithm with two different step sizes
(a) = 1 and (b) = 0.05 , and misalignment of the NPVSS-NLMS algorithm.
The input signal is an AR(1) process, L = 1000, = 1 1/(6L), and SNR = 20 dB.
Echo path changes at time 10.
0
NPVSS-NLMS
SVSS-NLMS
NEW-NPVSS-NLMS
-2
PVSS-NLMS
Misalignment (dB)
-4
-6
-8
-10
-12
-14
-16
-18
10
20
30
40
50
60
Time (seconds)
Fig. 6. Misalignment of the NPVSS-NLMS, SVSS-NLMS [with = 1 1/(18L)],

NEW-NPVSS-NLMS (with = 0.1), and PVSS-NLMS algorithms. The input signal
is speech, L = 1000, = 1 1/(6L), and SNR = 20 dB.
5
NPVSS-NLMS
SVSS-NLMS
NEW-NPVSS-NLMS
PVSS-NLMS
Misalignment (dB)
-5
-10
-15
10
20
30
40
50
60
Time (seconds)
Fig. 7. Misalignment during impulse response change. The impulse response changes
at time 20. Algorithms: NPVSS-NLMS, SVSS-NLMS [with = 1 1/(18L)], NEWNPVSS-NLMS (with = 0.1), and PVSS-NLMS. The input signal is speech, L = 1000,
= 1 1/(6L), and SNR = 20 dB.
20
NPVSS-NLMS
SVSS-NLMS
NEW-NPVSS-NLMS
15
PVSS-NLMS
Misalignment (dB)
10
-5
-10
-15
10
20
30
40
50
60
Time (seconds)
Fig. 8. Misalignment during background noise variations. The SNR decreases from 20 dB to
10 dB between time 20 and 30, and to 0 dB between time 40 and 50. Algorithms: NPVSSNLMS, SVSS-NLMS [with = 1 1/(18L)], NEW-NPVSS-NLMS (with = 0.1), and
PVSS-NLMS. The input signal is speech, L = 1000, = 1 1/(6L), and SNR = 20 dB.
10
20
SVSS-NLMS
NEW-NPVSS-NLMS
PVSS-NLMS
15
Misalignment (dB)
10
-5
-10
-15
10
20
30
40
50
60
Time (seconds)
Fig. 9. Misalignment during double-talk, without DTD. Near-end speech appears between
time 15 and 25 (with FNR = 5 dB), and between time 35 and 45 (with FNR = 3 dB).
Algorithms: SVSS-NLMS [with = 1 1/(18L)], NEW-NPVSS-NLMS (with = 0.1), and
PVSS-NLMS. The input signal is speech, L = 1000, = 1 1/(6L), and SNR = 20 dB.
0
PVSS-NLMS (VSS-APA w ith p = 1)
-2
VSS-APA w ith p = 2
VSS-APA w ith p = 4
Misalignment (dB)
-4
-6
-8
-10
-12
-14
-16
-18
-20
10
20
30
40
50
60
Time (seconds)
Fig. 10. Misalignment of the VSS-APA with different projection orders, i.e., p = 1
(PVSS-NLMS algorithm), p = 2, and p = 4. Other conditions are the same as in Fig. 12.
11
5
APA w ith p = 2
VSS-APA w ith p = 2
Misalignment (dB)
-5
-10
-15
-20
10
20
30
40
50
60
Time (seconds)
Fig. 11. Misalignment during impulse response change. The impulse response changes
at time 20. Algorithms: PVSS-NLMS algorithm, APA (with = 0.25), and VSS-APA.
Other conditions are the same as in Fig. 12.
2
APA w ith p = 2
VSS-APA w ith p = 2
Misalignment (dB)
-2
-4
-6
-8
-10
-12
-14
-16
-18
10
20
30
40
50
60
Time (seconds)
Fig. 12. Misalignment during background noise variations. The SNR decreases from
20 dB to 10 dB between time 20 and 40. Other conditions are the same as in Fig. 12.
12
10
APA w ith p = 2
VSS-APA w ith p = 2
Misalignment (dB)
-5
-10
-15
10
20
30
40
50
60
Time (seconds)
Fig. 13. Misalignment during double-talk, without a DTD. Near-end speech appears
between time 20 and 30 (with FNR = 4 dB). Other conditions are the same as in Fig. 12.
Comparisons with other VSS-type APAs

algorithms for comparisons
classical APA, = 0.2
variable regularized APA (VR-APA)
[H. Rey, L. Rey Vega, S. Tressens, and J. Benesty, IEEE Trans. Signal Process.,
May 2007]
robust proportionate APA (R-PAPA)

[T. Gnsler, S. L. Gay, M. M. Sondhi, and J. Benesty, IEEE Trans. Speech Audio
Process., Nov. 2000]
ideal VSS-APA (VSS-APA-id) - assuming that v(n) is available
13
0
APA
VR-APA
VSS-APA
Misalignment (dB)
-5
VSS-APA-id
-10
-15
-20
-25
-30
0
10
15
20
25
Time (seconds)
30
35
40
Fig. 14. Misalignments of APA with = 0.2, VR-APA, VSS-APA, and VSS-APA-id.
Single-talk case, L = 512, p = 2 for all the algorithms, SNR = 20dB.
0
APA
VR-APA
R-PAPA
VSS-APA
-5
Misalignment (dB)
VSS-APA-id
-10
-15
-20
-25
-30
0
10
15
20
25
Time (seconds)
30
35
40
Fig. 15. Misalignments of APA, VR-APA, R-PAPA, VSS-APA, and VSS-APA-id.

Background noise variation at time 14, for a period of 14 seconds (SNR decreases
from 20dB to 10 dB). Other conditions are the same as in Fig. 14.
14
without DTD
with Geigel DTD

APA
30
VSS-APA-id
10
-10
-20
-20
10
15
20
25
Time (seconds)
30
35
40
VSS-APA-id
10
-10
-30
0
R-PAPA
VSS-APA
20
Misalignment (dB)
Misalignment (dB)
20
APA
30
VR-APA
VSS-APA
-30
0
10
15
20
25
Time (seconds)
30
35
40
Fig. 16. Misalignment of the algorithm during double-talk. Other conditions

are the same as in Fig. 14.
Conclusions and Perspectives

a family of VSS-type algorithms was developed in the context
of AEC.
the VSS formulas do not require any additional parameters
from the acoustic environment (i.e., non-parametric).
they are robust to near-end signal variations like the increase of
the background noise or double-talk.
the experimental results indicate that these algorithms are
reliable candidates for real-world applications.
[C. Anghel, C. Paleologu, et al,FPGA implementation of a variable step-size affine
projection algorithm for acoustic echo cancellation, in Proc. EUSIPCO 2010]
Future work towards proportionate adaptive algorithms.

[C. Paleologu, J. Benesty, and S. Ciochin, Sparse Adaptive Filters for Echo
Cancellation, Morgan & Claypool Publishers, ISBN 978-1-598-29306-7, 2010]
[C. Paleologu, S. Ciochin, and J. Benesty, An Efficient Proportionate Affine
Projection Algorithm for Echo Cancellation, IEEE Signal Processing Letters, 2010]
[J. Benesty, C. Paleologu, and S. Ciochin, Proportionate Adaptive Filters from a
Basis Pursuit Perspective, IEEE Signal Processing Letters, to appear]
15
Dispozitive electronice i optoelectronice

de nalt performan
Structuri filtrante avansate de microunde
Nicolae MILITARU, E.T.T.I.-U.P.B.
Prezentare general
Domeniul de cercetare
Direcia de cercetare
Inginerie electronic i telecomunicaii
Dispozitive electronice i optoelectronice de nalt

performan
Obiectiv
Obinerea unor dispozitive compacte cu performane

mbuntite pentru domenii de vrf din tehnica
telecomunicaiilor
Structuri filtrante de microunde de tipul

trece-band n tehnologii planare
Rol
Selecia n domeniul frecven a semnalelor
Cerine

Selectivitate ridicat
Atenuare de inserie redus
Timp de ntrziere de grup contant n banda de
trecere
Band de oprire ct mai larg
Atenuare n banda de oprire ct mai ridicat
Dimensiuni mici
Tehnologie ieftin (pre de cost redus)

Configuraii
n linie
Qext1
1
K 1 2
K 23
C12
C23
K 3 4
C34
Qext4
N :1
1: N
C
L
C C12
C C12 C23
C C 23 C34
C C34
Filtru trece-band de ordin patru, cu constante concentrate.

d io1
P1
d 23
d12
K12 K e
d34
K 2 3 K m
d io2
P2
K 3 4 K e
Filtru trece-band de ordin patru, n tehnologie microstrip.

Configuraii
n linie
0
|S11|, |S21|, [dB]
-10
-20
-30
-40
|S |, model circuit
21
|S |, model circuit
-50
11
|S21|, model planar

-60
0.9
|S11|, model planar

0.95
1
F [GHz]
1.05
1.1
Prezentare comparativ performane

FTB cu constante concentrate FTB microstrip

Configuraii
Cu rezonatoare multiplu cuplate
a) filtru ndoit
c) filtru cu cuadruplei n cascad
b) filtru cu triplei n cascad
d) filtru cu linii de acces multiplu cuplate
Fig. 2. Reprezentare simbolic a unor posibiliti practice de realizare a filtrelor cu cuplaje multiple.

Configuraii
d1 4
d 01
d 4 5
Configuraia de cuplaj.
w
Rez 1
Rez 4
d1 3
-C14
-C14
C14
d 2 4
L23
L
Rez 3
Port2
L24
-L24
-L24
2L
-L13
-L13
L13
-L23
-L23
Port1
Rez 2
d 2 3
Modelul n tehnologie microstrip.
Modelul cu constante concentrate.

Configuraii

0
-10
-20
|S11|, |S21| [dB]
-30
-40
-50
|S | sinteza
21
|S | sinteza
11
|S | sim. elmg.
11
|S | sim. elmg.
21
|S | sim. elmg. p.
21
|S | sim. elmg. p
-60
-70
11
-80
2200
2250
2300
2350
2400
2450
2500
2550
2600
F [MHz]
Comparaie ntre performanele teoretice ale FTB i cele anticipate prin simulare.

Tehnologii utilizate
Microstrip
Coplanar
Multistrat
Microstrip cu plan de mas decupat

Proiectarea unor filtre n tehnologie

microstrip
FTB clasice
FTJ prototip
Transformri de frecven
Transformarea FTB cu constante concentrare (bobine,
condensatoare) ntr-un filtru de microunde n tehnologie
planar
FTB cu rezonatoare de tip distribuit
Stabilirea configuraiei de cuplaj

Generarea matricei cuplajelor
Determinarea valorilor cuplajelor necesare
Obinerea geometriei (layout-ul) filtrului planar
Optimizarea geometriei


microstrip
FTB cu rezonatoare microstrip cuplate multiplu
Stabilirea configuraiei de cuplaj

Generarea matricei cuplajelor
0
5
1
a)
b)
0.78529
0
0
0
0
0
0.78529 0.03377
0
0.71770 0.05591
0
0
0
0.52370 0.48362 0.62656
0
M1 =
0.71770 0.49362 0.01663 0.35002

0
0
0
0.05591 0.62656 0.35002 0.03377 0.78529
0
0
0
0.78529
0
0
Configuraia de cuplaj a unui FTB de ordin 4 cu rspuns

asimetric (a) i geometria sa n tehnologie microstrip (b).


microstrip
Determinarea cuplajelor necesare
Coeficienii de cuplaj, k (cuplaj rezonator rezonator)

Factorii de calitate externi, Qext (cuplaj rezonator linie de
acces 50
Obinerea geometriei (layout-ul) filtrului planar

0
5
1


microstrip
Optimizare
Corecia cuplajelor rezonator linie de acces i corecia

frecvenei de rezonan a rezonatoarelor terminale
Ajustarea cuplajelor ntre perechile de rezonatoare
Ajustarea frecvenelor de rezonan ale rezonatoarelor
Metod avansat de optimizare suplimentar
Combin acurateea simulrii electromagnetice cu simplitatea

i viteza optimizrii folosind un simulator de circuit

microstrip
Optimizarea geometriei
0
Lumped S11
-10
Lumped S21
-20
Magnitude (dB)
Lossless sim S11
-30
Lossless sim S21
-40
-50
-60
-70
-80
-90
-100
3.2
3.4
3.6
3.8
Frequency (GHz)
4.2
4.4

Modele experimentale
Cercetri viitoare
Filtre de microunde n tehnologie multistrat cu

rezonatoare multiplu cuplate
Filtre trece-band alctuite din rezonatoare microstrip

extrem de compacte, cu moduri duale de oscilaie
Filtre de microunde cu rezonatoare mpachetate care

folosesc cuplaje prin fant
APLICAII ALE
METAMATERIALELOR N
DOMENIUL MICROUNDELOR
Doctorand as. univ. ing.

Iulia Andreea Mocanu
Cuprins
1. Introducere
2. Aplicaii ale metamaterialelor realizate cu
structuri 3D
structuri 2D
structuri 1D
5. Perspective
1. Introducere
1.1. Definiii:
1999: Roger M. Walser:

metamaterialele
sunt
structuri
artificiale,
tridimensionale, periodice, proiectate pentru a produce o
combinaie optim de dou sau mai multe rspunsuri la o
anumit excitaie.
2006: Caloz-Itoh:
metamaterialele sunt structuri artificiale, omogene
electromagnetic, cu proprieti neobinuite i inaccesibile
direct n natur.
1. Introducere
1.2. Clasificarea materialelor

p - dimensiunea medie a celulei structurale

g - lungimea de und n ghid
p < g 4
omogenitate
1. Introducere
1.2. Clasificarea materialelor
Diagrama
permitivitii electrice
-permeabilitii magnetice
i indicele de refracie
1. Introducere
1.3. Proprieti specifice metamaterialelor
Refracie negativ a unui fascicul monocromatic

la suprafaa de separaie dintre aer i o plac de metamaterial
1. Introducere
1.3. Proprieti specifice metamaterialelor
Triada intensitate cmp electric, intensitate cmp

magnetic, vector de und i vectorul Poynting pentru:
a) mediu de tip RH; b) mediu de tip LH.
1. Introducere
1.4. Dinamica domeniului
Numrul de lucrri publicate

pn n anul 2002
Scalarea dimensiunii
metamaterialelor
Cuprins
1. Introducere
structuri 3D
structuri 2D
structuri 1D
5. Perspective
2. Aplicaii ale metamaterialelor

realizate cu structuri 3D
2.1. Invizibilitate total
Cuprins
1. Introducere
structuri 3D
structuri 2D
structuri 1D
5. Perspective

3.1. Focalizarea cu ajutorul lentilelor LH plate
Efect dublu de focalizare ntr-o lentil LH format

dintr-o plac de material de tip LH , ncadrat de dou medii de tip RH

Lentil Veselago-Pendry pentru

microunde,
cu indice de refracie negativ
(C. Soukoulis, OPN, June 2006,
pp. 16-21)

Imaginea unui
obiect bidimensional
NANO.
a) Imaginea FIB
b) Imagine superlentil
c) Profilul seciunii transversale;
d) Rezultatul controlului
imaginii al aceluiai obiect;
e) Media seciunii transversale.
Cuprins
1. Introducere
structuri 3D
structuri 2D
structuri 1D
5. Perspective

4.1. Metoda liniei de transmisiune pentru
structuri LH
a) Circuitul cu constante concentrate pentru o unitate de lungime

infinitezimal dintr-o linie PRH;
b) Circuitul cu constante concentrate pentru o unitate de lungime
infinitezimal dintr-o linie PLH.

structuri CRLH
Implementarea n tehnologie microstrip

a unei celule de linie de transmisune LH
Circuitul cu constante concentrate

pentru o unitate de
lungime infinitezimal
dintr-o linie CRLH

structuri CRLH cu pierderi
Implementare in tehnologie
microstrip a unei linii de transmisiune CRLH cu 24 de celule (Itoh)

structuri CRLH cu pierderi
Parametrii S pentru
o linie de transmisiune CRLH
echilibrat, cu pierderi.
a) |S21| pentru Q =100,
b) |S11| pentru Q =100,
c) |S21| pentru Q =500,
d) |S11| pentru Q =500.

4.4. Cuplor simetric CRLH realizat cu 7 celule
microstrip a unui cuplor simetric CRLH cu 7 celule

Parametrii de mprtiere
pentru cuplorul simetric
CRLH cu 7 celule:
a) |S11|, b) |S21|,
c) |S31|, d) |S41|.

pentru cuplorul simetric
CRLH cu 3 celule:
a) |S11|, b) |S21|,
c) |S31|, d) |S41|.

4.6. Cuplor asimetric CRLH realizat cu 7 celule
microstrip a unui cuplor asimetric CRLH cu 7 celule

pentru cuplorul asimetric
CRLH cu 7 celule:
a) |S11|, b) |S21|,
c) |S31|, d) |S41|.

Directivitatea pentru
cuplorul asimetric CRLH cu
7 celule:
a) |S11|, b) |S21|,
c) |S31|, d) |S41|.
Cuprins
1. Introducere
structuri 3D
structuri 2D
structuri 1D
5. Perspective
5. Perspective
msurtori pentru cuploarele propuse,
optimizarea performanelor i a dimensiunilor ,
realizarea altor dispozitive de microunde,
studiul altor tipuri de linii de transmisiune right-left

handed: D-CRLH, E-CRLH.
How Autonomic and Cognitive

Networking Technologies are
Reshaping the Economy
The (unaccounted for)
Wealth of Networks
MIHAELA ULIERU
From the Industrial to the

Ubiquitous Civilization
Such networks are coordinated to create a ubiquitous environment
(source: James Law Cybertecture International ltd., 2006)
ACNT are the Controllers of

Cyber-Physical Systems
Integration and networking of systems across all scales to create

intelligent environments of self-organizing artifacts
Adapt to and anticipate users needs and wants
How do we organize billions of mobile

smart objects that are highly dynamic,
short living,?
ORDER FOR FREE

EFFORTLESS COORDINATION
INCREASED PRODUCTIVITY AND

EFFICIENCY
(source: Makoto Miwa, Panasonic, 2006)
AUTOMATE SPECIALIZED
WORK (increase time available
to do creative work)
BOOST ORGANIZATIONAL
CAPITAL (self-organized
production)
Supply chain management
Asset management
INCREASE THE EFFECTIVENESS OF

OPERATIONS AND PRODUCTIVITY
Preventive maintenance
MEASURABLE,
TANGIBLE (aka
ACCOUNTED FOR!)
Real-time inventory
ENABLE SELF-ORGANIZATION
TO MAXIMIZE PERFORMANCE
IMPROVE SOCIAL WELFARE
Reduce crime rate (TANGIBLE)
Peace of Mind (INTANGIBLE)
Common Situational Awareness

Introducing eParade, a first-in-class public
safety service.
First responders are empowered by
accessibility to relevant and locationspecific information at their finger tips.
eParade is an innovative service the
Ottawa Police and Wesley Clover have
pioneered to bridge traditional barriers
in effectively collaborating relevant
information between responders.
Consumer surplus (Flickr) cost is removed, yet it

counts as value for the welfare of the operation
COLLAPSE OF COORDINATION COSTS
Org
Org 44
Org
Org 33
Agent
Agent 55
Organisation
Organisation 22
Organisation
Organisation 66
Organisation
Organisation 11
Unit 1
Unit 3
Unit 2
Unit 1
Unit 3
Unit 2
ACNT effects
Increase performance (tangible)

Point to ineffective policies hindering intraand inter-organizational collaboration
(intangible boost organizational capital by
collapsing coordination costs);
Empower the individual contribution and peer
production (intangible - boost human capital);
Reveal the impact of professional, social and
cultural aspects (intangible) on operational
effectiveness / productivity
The world is changing very fast. Big will

not beat small anymore. It will be the fast
beating the slow.
Rupert Murdoch
The iPhone Prosumer Community

GreatApps Company
Product: iSteam
Profit: $100,000 in 1 month
Ethan Nicholas
Product: iShoot
Profit: $600,000 in 1 month
http://www.news.com.au/technology/story/0,2564
2,24916555-5014239,00.html
http://www.wired.com/gadgetlab/2009/02/shootis-iphone/
Inspiration
Consumption
Consumption
Production
Production
Distribution
NETWORKED
INDIVIDUALISM
(Barry Wellman U Toronto)
2
Challenges the Institution
3
Access to knowledge
4
Self-catalyzing
1
Innovative marketing model
PEER PRODUCTION
Sales
How is this changing the world now?
Head: Corner store

80% of the sales (area)
for 20% of the books
Transaction costs
(per book)
Traditional cost-value threshold.

Additional value not worth cost
of coordination.
Tail: Amazon.com
20% of the sales (area)
for 80% of the books
Number of books published
(Unaccounted) value
Techno-economic
Networks, Digitalization, Innovation
Head: institutionalized production
20% of products (area)
individualized (customized)
production
20% producers
Tail: open source crowdsourcing
80% of products (area)
80% of the producers
(productivity unavailable to
hierarchy)
Number of producers
(Unaccounted) value
How to leverage on this

unused human capital?
Head: institutionalized production
20% of jobs uses 80% available human
capital
Tail: Available but UNUSED
human capital
80% of jobs (area) use
20% human capital
(productivity unavailable to
hierarchy)
Number of producers
NEED TO REVOLUTIONIZE OUR

INSTITUTIONAL STRUCTURES !
300,000s
Government.
Contributors
34 Million
Citizen.
Contributors
(in Canada)
100s Service 1000s

1
5
Dept.
Contributors
Team
Team
Contributors
Lead Members
Unleashing the Ecosystem
POLICIES AND TECHNOLOGIES THAT ENABLE Response-Able Communities
Session on ICT for

eGovernance
SOS Networks
Emerging Leaders
Social Entrepreneurs
Agents of Change
10
WITH Citizens FOR Citizens
SOCIAL INNOVATION GENERATION
Cyber
Physical smart application
Miorandi et al IEEE-SMC 2010
Social Networking Infrastructures

ACNT Fueling the Power of Crowdsourcing
Internet
The Mind Electric
To make human
connections
readable by
machines and
vice versa so
that the web of
machines and
people can work
in concert.
See Flagship on Living

Technologies
Everyware
Web 2.0
11
Living Technologies
TOWARD WEB 3.0

The Holodeck as Future Workground
Exodus to the Virtual World: How Online Fun is Changing Reality [Edward Castronova]
Also - Jane McGonigal's TED talk
12
13
Indexarea automat a secvenelor de imagini

Clasificarea automat dup gen
.l.dr.ing. Bogdan-Emanuel IONESCU*
bionescu@alpha.imag.pub.ro
*activitatea de cercetare a fost finanat parial din proiectul FSE - European

Structural Funds EXCEL POSDRU/89/1.5/S/62557 (2010-2013)
LAPI Laboratorul de
Analiza i Prelucrarea
Imaginilor
Universitatea
Politehnica din
Bucureti
LAPI Laboratorul de Analiza i Prelucrarea Imaginilor

Prof. Vasile BUZULOIU
directorul laboratorului
> personal didactic i de cercetare:
- 2 profesori
- 1 confereniar
- 4 efi de lucrri
- 3 asisteni
- 3 cercettori principali
- >8 doctoranzi
>2010 devenit Centru de Cercetare al

Universitii Politehnica din Bucureti
> http://alpha.imag.pub.ro
Prezentare laborator LAPI
WED14, Bucureti, 22-23 Septembrie 20102

> proiecte de cercetare naionale i internaionale:
- 2 proiecte FP6 - European 6th Framework Programme,
- 3 CERN European Organization for Nuclear Research,
- >4 FSE Fonduri Structurale Europene (POS-DRU, POS-CCE),
- >6 proiecte naionale finalizate i n curs de desfurare,
- proiecte industriale (Tessera Romnia - Fotonation).

> colaborri universitare externe (active >15 universiti):
- Polytech Annecy-Chambery, Frana,
- Universit de Poitiers, Frana,
- INSA Lyon, CPE Lyon, Frana,
- INPG Grenoble, Frana,
- ENST, ENSTA Paris-Frana,
- University of Malaga, Spania,
- National University of Ireland, Galway-Irlanda, etc.
i din mediul privat:
- CERN, Geneva-Elveia,
- EbooSolutions, Annecy-Frana,
- Tessera - Fotonation, SUA,
- SensoMotoric Instruments, Berlin-Germania, etc.

> tematici de cercetare abordate:
urmrirea privirii i comunicarea cu ajutorul acesteia
(Eye Gaze Communication)
- segmentare iris,
- eye tracking,
- detecia gradului de
oboseal a oferului,

> tematici de cercetare abordate (continuare):
detecia evenimentelor
folosind Time-of-Flight cameras
- calibrare i corecie,
- detecia de evenimente,
- segmentare 3D,
> (prezentare erban OPRIESCU)


indexarea dup coninut a

documentelor multimedia
- analiz coninut baze de imagini i video,
- sisteme de navigare 3D,
- rezumare automat, etc.

imagistic medical
-calibrare, imbuntire,
-diagnosticarea automat a
osteoporozei,
-analiza imaginilor radiologice pentru
urmrirea protezelor ortopedice, etc.
> (prezentare
Laura FLOREA)

algoritmi de analiz i
prelucrare a fotografiilor
digitale
- red-eye removal,
- auto-sharpening,

recunoaterea automat a gesturilor minii
- recunoatere ipostaze statice,
- sistem de recunoatere gesturi
dinamice (15 img./s)

10
Planul prezentrii
Problematica indexrii dup coninut
Problematica clasificrii automate a genului video
Abordarea propus
Descrierea coninutului video (aciune, culoare, structur)
Concluzii i perspective
Planul prezentrii
WED14, Bucureti, 22-23 Septembrie 2010, 1/41

11
Problematica indexrii
dup coninut

12
societatea actual este una bazat pe informaie oamenii
schimb/acceseaz un volum de date multimedia important:
text, ex. eBooks,
eDocuments, etc.
audio, ex. muzic,

voce, etc.
imagini, ex. poze,

grafic, etc.
secvene de
imagini ex. filme,
flux video, etc.
> tehnologie actual avansat (capacitate de stocare ridicat,

protocoale de transmisie a datelor rapide, dispozitive mobile
multimedia, etc.)

13
> problem: cum accesm informaia relevant din aceste colecii
imense de date ?
ex. video = cantitate mare de date
- 1 minut > 1.500 imagini (la 25 cadre/s)
- 1 baz video > mii de filme (>135.000.000 imagini)
vrem s cutam o anumit informaie,
vrem s rsfoim baza de date,
vrem s vizualizm eficient conunutul (video),
> soluia existent: sistemele de indexare automat dup coninut sau

Content-based Indexing Systems (sunet: CBAR, imagini: CBIR, video:
CBVR, etc.)

14
> indexare = adugarea de informaii suplimentare relative la
coninutul datelor ce se doresc a fi indexate = adnotare
> exemplu simplu:
0001010010
1100011010
1111110001
0011110110
0001111111
informaie brut
program: Windows 7 Explorer

15
Procesul de indexare
> arhitectura de baz a unui sistem de indexare video:
baza video
datele
propriu-zise
interfaa cu utilizatorul
[rezumate]
sistemul de navigare
atribute/indeci
(numerici, textuali, etc.)
sistemul de cutare
adnotare coninut
Procesul de indexare

16
Sintactic vs. semantic

> global, distingem dou tendine:
adnotare sintactic (low-level)
culori
text
schimbare
textur
obiect de interes
traiectorie
sunet
adnotare semantic (high-level)
n meciul de fotbal al lui Real Madrid, Ronaldo numrul 9 a marcat golul.

Sintactic vs. semnatic

17
Sursele de informaie
> video = informaie spaio-temporal:
culoare: coninut vizual;
textur: materiale din scen;
form: obiecte, personaje;
motenite de la
sistemele de indexare
de imagini
structura temporal: ritm, aciune;

micare local sau global;
specifice video
sunet muzic, vorbire;

alte surse ex. culoare piele, text ncrustat, etc.
Sursele de informaie

18
Problematica clasificrii
automate a genului video

19
Clasificarea dup gen

> o subproblem de actualitate a indexrii =
separarea documentelor video n funcie de gen:
talk show
muzic sport
sub-gen:
coninut:
technic:
mixt
film
animaie
soap
desene anim.
dram
documentar
reclame
artistice
aciune
3D
tiri
ppui
comedie
creion
abstracte
pictur
obiectivul cercetrii: determinarea de descriptori de coninut suficient

de relevani pentru a face distincie ntre genurile de baz.
Clasificarea dup gen

20
State-of-the-art
> mai multe abordri existente (state-of-the-art n [D. Brezeale, D.J. Cook08] ):
[M.J. Roach, J.S.D. Mason01]
uni-modal
- adnotare: micare;
- classificator: Gaussian Mixture Model;
- genuri: sport, desene animate & tiri
[X. Yuan, W. Lai, T. Mei, X.S. Hua, X.Q. Wu, S. Li06]
erori de detecie < 6%.

spaio-temporal
- adnotare: temporal & spaial;

- classificator: Decision Trees & Support Vector Machine;
- genuri: - film, reclame, tiri, muzic & sport
precizie 88.6%;
- filme n: aciune, comedie, horror & desene precizie 81.3%;
- sport n: basebal, footbal, volei,
precizie 97%;
tenis, basket & soccer
State-of-the-art

21
State-of-the-art
> mai multe abordri existente (state-of-the-art n [D. Brezeale, D.J. Cook08] ):
[M. Montagnuolo, A. Messina09]
multi-modal
- adnotare: vizual-perceptual, temporal, cognitiv, text, sunet;

- classificator: Parallel Neural Networks;
- genuri: fotbal, desene, muzic, buletin vreme, tiri,
acuratee 95%
talkshow & reclame.
Obs. - testarea este realizat de regul pe baze de test limitate sau pe
cazuri particulare;
- msurile de evaluare ale performanei sunt non standard sau
insuficiente (ex. doar eroare de detecie, doar precizie, etc.)
State-of-the-art

22
Abordarea propus
filme de animaie
documentare
muzic
tiri
- filme de animaie: palet de culoare specific;

- documentare: linia orizontului este frecvent, ritm lent;
- muzic: culori ntunecate, ritm alert;
- tiri: siluete persoane/fee;
etc.
> abordarea propus: adnotare folosind informaia temporal, de
culoare i de contur;
Abordarea propus

23
Descrierea coninutului video

structura temporal

24
Segmentarea temporal
[filme de animaie CITIA-Annecy]
= descompunerea n plane (uniti structurale de baz).

timp
imagine1
...
imaginei
plan1
...
imaginei+1
...
plan2
...
imaginej
plani
...
imagineN
planM
detecia tranziiilor video:

cut
cuts
fade-ins
dissolves
fade-outs
Segmentarea temporal

25
Segmentarea temporal: cut

> metoda propus: bazat pe histogram (cut = tranziie abrupt):
calcul distane
film
imagini reinute
k
eantionare
spaial+temp.
calcul
histograme
reducere
de culoare
d1
d2
d4
d1, d2, d3, d4
k+1
...
...
...
distane
d3
...
thresholding
cuts
evoluia temporal a valorii

medii de distan
Detecia de cut
derivata de ordin 2 a evoluiei

temporale

26
Segmentarea temporal: fade

> metoda propus: bazat pe intensitate (fade = variaie gradual):
Y
...
...
conversie
YCbCr
Cb
timp
Cr
evoluia temporal a Var{Y}, E{Y}, |E{Cb}-E{Cr}|
fade-in
fade-out
timp
Detecia de fade
decizie (fade-in):
- Var{Y} 0 (start)
- E{Y} sau
|E{Cb}-E{Cr}|
- durat[tmin;tmax]
(fade-out sens invers)

27
Segmentarea temporal: dissolve

> metoda propus: bazat pe intensitate (dissolve = pixeli care
apar/dispar ):
film
k
FPk
eant. spaial
niveluri de gri
k+1
calcul pixeli
ce apar/dispar
FPi =
FadeOuti + FadeIni
X Y
FPk+1
FPN
dissolve
dissolve
CT
TT
Detecia de dissolve
prag dublu
FPi >CT & max

FPi >TT & &
dissolve

28
Segmentarea temporal: plane

> determinarea planelor video:
cuts
fade-in
fade-outs
dissolves
ex.blitz
timp
plan 1
plan 2
plan 3
Planele video
plan 4
plan 5
timp

29
Descriptorii de aciune
> obiectiv: informaie relativ la ritm vizual i aciune;
%tranziii graduale:
GT =
Tdissolve + T fade in + T fade out

Ttotal
ritm: ~ tempo
T = 5 s (i ) : numrul relativ de schimbri de plan ce au loc ntr-o

fereastr de durat T pornind din momentul i;
v T = 5 s = E { T = 5 s (i )} : valoarea medie pe secven;

actiune: n general este corelat cu frecvena schimbrilor de plan:
hot action
Coninutul de aciune
low action
30
Descriptorii de aciune
actiune: (continuare);
plane de aciune
adnotare vizual
cuts
plan video
T
HA = HA
Ttotal
1 if T = 5 s (i ) > 2 .8
HA shot (i ) =
0 altfel
TLA
Ttotal
1 if T = 5 s (i ) < 0 .7
LA shot (i ) =
0 altfel
LA =
Coninutul de aciune
timp

31

culoare

32
Descriptorii de culoare
obiectiv: informaie relativ la distribuia de culoare i percepia
culorilor;
palet
p%
plan 1
film
plan 2
detecie
plane
...
p%
rezumare
plane
...
plan M
p%
reducere culori
& calcul hist.
medie
hist.
global
ponderat
...
hist. medie
ponderi plane
histograma global ponderat:
pondere
1 Ni j
T
hGW (c) = hshoti (c) i , i = shot
Ttotal
i =0 N i j =0
culoare
M
Coninutul de culoare

33
> cum trecem de la descrieri numerice (ex. histograme) la descrieri
perceptuale ? numele culorilor;
B
intensitate
nuan
A
Nume: Dark Hard Blue
couleurs adjacentes
roides co
urs f
ule
ule
ur
co
saturaie
d es
au
ch
couleurs complmentaires
Webmaster 216 culori
Roata de culoare a lui Itten
histograma de culori elementare:
hE (ce ) =
215
h
c=0
GW
(c )
Name ( c e ) Name ( c )
34
proprieti ale culorilor:
215
Plight = hGW ( c )
Wlight Name ( c )
c =0
: procentul de culori luminoase din secven,

Wlight{light, pale, white};
Pdark : procentul de culori ntunecate, Wdark{dark, obscure, black};

Phard : procentul de culori saturate, Whard{hard, faded} elem.;
Pweak : procentul de culori slab saturate, Wweak{weak, dull};
Pwarm : procentul de culori calde, Wwarm{Yellow, Orange, Red, YellowOrange, Red-Orange, Red-Violet, Magenta, Pink, Spring};
Pcold : procentul de culori reci, Wcold{Green, Blue, Violet, Yellow-Green,

Blue-Green, Blue-Violet, Teal, Cyan, Azure};

35
proprieti ale culorilor (continuare):
es
aud
ch
couleurs adjacentes
roides co
ule
ur
urs f
ule
co
Card { c / hGW ( c ) > 0 .01 } : variaie de culoare = numrul de culori

=
216
folosite n secven;
couleurs complmentaires
Pvar
Pdiv =
Card { ce / hE ( ce ) > 0 .04 } : diversitate de culoare = numrul de

13
nuane folosite n secven;
relaia dintre culori:
Padj =
Card {ce / Adj ( ce , c 'e ) = True }

: adiacen = culori apropiate (roat
2 N ce
de culoare);
Pcompl : complementaritate = culori opuse (roat de culoare);


36

contururi

37
[IJCV, Cristoph Rasche10]
Descriptorii de contur
obiectiv: informaie relativ la contururile obiectelor i a relaiilor dintre

acestea;
metod de descriere a curburii segmentelor;
iniial: edge detection pentru diferite niveluri de
detaliu (Canny);
=1
siluete de animale
(ondulate, inflexiuni,
iregulariti reduse)
=3
=5
scene naturale
(erpuite, iregulate)
Descriptori de contur

38
semntura contururilor: conversie contururi 2D funcii 1D;

arc neted
v
v
inflexiune
L feature
> dac fereastra se gsete pe o parte a liniei drepte definete

funcia bowness (v);
> dac acesta se gsete pe ambele pri definete funcia de
inflexiune (v);
(v )
(v )
> pentru diferite valori ale ferestrei v definim spaiul local/global LG;

39
propriti de contur:
b : gradul de curbur;
drept vs. arc;
cerc vs. cerc complet;
: grad de circularitate;
e : iregularitate
y : simetrie
zig-zag vs. sinusoidal;

iregular vs. simetric;
+ parametri de aparen:
c m , c s : media, dispersia intensitii pixelilor de-a lungul

conturului;
iregularitate simetrie
f m , f s: gradualitate (fuzzy, filtrare I * DOG).

40
relaia dintre contururi: contururile vecine sunt grupate n perechi;

douter
dcorner
dcenter
L feature
T feature
Paralel
> selecie bazat pe proximitatea punctelor terminale, similaritatea

structurii, simetria alinierii etc.
parametri: direcia angular a perechii, distana dintre puncte
terminale, lungimea medie, etc.

41
> descriptori puternici pentru indexarea imaginilor statice:
selecie
[60000 imagini]
scade
similaritatea
42

43
Rezultate clasificare dup gen

> testare clase de descriptori n cazul a 7 genuri uzuale:
desene
pub.
document.
filme
muzic
tiri
sport
> baza de test: 210 secvene, 30 per gen, > 91 ore;

- animaie
: 20h30min (desene animate, scurt metraje, serii, filme);
- publicitate
: 15min;
- documentare
: 22h (natur, oceane, civilizaie i istorie);
- filme
: 21h57min (lung metraj, episoade, sitcom);
- muzic
: 2h30min (pop, rock, dance);
- tiri
: 22h (TVR telejurnal).
- sport
: 1h55min (n principal fotbal).
> problema dreptului de autor!


44

clasificarea datelor: modul de abordare,
binar: un gen vs. toate celelalte;
metode:
- K-Nearest Neighbors (k=1, distan cosinus);
- Support Vector Machine (nucleu liniar);
- Linear Discriminant Analysis (decorelare cu PCA);
antrenare? poate distorsiona rezultatele:
set de antrenare ales aleator + clasificare (repetat x1000);
seturi de antrenare de dimensiuni diferite.

45

clasificarea datelor: evaluare performan,
P=
TP
TP + FP
: precizie (precision) msur a falselor detecii;
R=
TP
TP + FN
: amintire (recall) msur a non deteciilor;
Fscore = 2
PR
P+R
CD =
N GD
N total
: Fscore evaluare global a falselor detecii i a

non deteciilor;
: rat de detecie corect ia n calcul ambele

clase;

46

clasificarea datelor: discuie asupra preciziei i reamintirii;
> pentru fiecare gen rate ~> 80%;
SVM
SVM
SVM
KNN
KNN
KNN
LDA
LDA
> cele mai bune rezultate:

- documentare: SVM, KNN contur-culoare-aciune;
- filme: LDA, SVM culoare-aciune & contur-culoare-aciune;
- tiri: KNN, SVM contur-culoare-aciune & culoare-aciune;
- sport: LDA, KNN culoare-aciune & contur-culoare-aciune;

47

clasificarea datelor: discuie asupra performanelor globale;
(toate genurile)
SVM
SVM
antrenare 10%-20% !
> cele mai bune rezultate:

- 1st SVM contur-culoare-aciune;
- 2nd KNN contur-culoare-aciune;

48

49
> am discutat problema clasificrii automate dup gen a secvenelor
video folosind descriptori de culoare, aciune i contur,
> teste pentru 91 ore video i 7 genuri clasice:
pentru fiecare gen precizie i reamintire > 80% (best 100%);
global, combinaia contur-culoare-aciune furnizeaz cele mai bune
performane, Fscore=80%, CD=94%;
> rezultate preliminarii!
abordarea deteciei de concepte, baz de test mai vast,
sunet (cooperare n curs Johannes Kepler University, Linz).

50
mulumesc !
demonstraie?
ntrebri?
Email: bionescu@alpha.imag.pub.ro
Web: http://imag.pub.ro/~bionescu

51
Lucia Stefan
Archiva Ltd.
London, UK
Definitia populara:
depozitare intr-un spatiu exterior
Scanare si depozitare filelor in afara
sistemului curent
Arhive de tip zip
..... Etc
Se face confuzie intre custodie si

arhivare
Arhivarea presupune analiza, selectia si
organizarea documentelor electronica de

arhivat, adica definirea:
Materialul electronic de arhivat
Procedeelor de arhivare si cautare
Accesului la materialul arhivat
Rolurilor participantilor la arhivare
Definirea standardelor applicate
Prin arhivare, se urmareste ca informatia
sa pastreze urmatoarele caracteristici

(conform ISO 15489, partea 1):
Autenticitate
Credibilitate
Integritate
Utilizabilitate
Ref:
http://www.nationalarchives.gov.uk/documents/generic_
reqs1.pdf
Deteriorarea suportului:
Suportul electronic gen CD, DVD se
mentine maximum 10 ani spre deosebire

de piatra (milenii), pergament (1000 ani),
hartie (200 ani) microfisa (100 ani)
Deteriorarea Hardware
Invechirea sistemului de operare si a
softurilor
Evolutia tehnologica
Invechirea formatelor electronice
Pierderea contextului
S-au propus trei solutii:

1.Muzeu al vechilor sisteme: hardware, OS
si software
2.Simularea (emulation)
3.Migrarea documentelor la intervale
regulate de timp
1 nu este solutie realista
2 si 3 optiuni viabile
Arhivistica traditionala: proces pasiv

Arhivistica electronica: proces activ
Selectia materialui
Prezentarea
materialui
Transfer
Regasirea materialui
Arhivare pasiva
Arhivare activa
Ingest migrarea catre alte sisteme
- reconcilierea
Administrarea suportului electronic
(media)
- administrarea copiilor si back-upuriloe
- evitarea degradarii suportului si format
obsolescence
Administrarea formatului filei
- Conversia la alte formate
- Gestionarea copiilor in diverse formate ale
aceluiasi document
Mecanismele de autentificare
Securitatea
Audit si certificare: pentru arhive de
incredere
Identificarea materialului de arhiva

Controlul extern (legislatie, regulamete,
control managerial)
Definirea produsului final (in ce format,
media, etc) materialul va fi arhivat

Resurse necesare (umane, infrastructura,
tehnologie)
Reconcilierea: verificarea rezultatului
evitarea degradarii suportului si format
obsolescence
Criterii de alegere a formatului media:
-longevitate: optim 10 ani
-capacitate: adaptata informatiei de arhivat
-viabilitate: evitarea si detectarea erorilor
-adoptie: cat este acceptat de industrie
-cost
-rezistenta la distrugere fizica
Format nativ sau redare in alt format? (ex
format nativ word sau redare in pdf/a)

Criterii de alegere:
- adoptie
- platform independent
- open stadard
- transparenta
- Metadata support
In practica: formatele MS Office, XML,
PDF/A, JPEG
Unitatea
de schimb dintre OAIS si mediul

exterior sunt Pachetul de Informatii
(Information Package).
Tipuri de pachete
Submission Information Package (SIP)
Archival Information Package (AIP)
Dissemination Information Package (DIP)
Un
pachet de informatii este un container

conceptual de doua tipuri de informatii:
Continutul (Content Information)
Informatia descriptiva de arhivare (Preservation
Description Information (PDI).
Long-term archive and catalogue provided by EUMETSAT Unified
Archive and Retrieval Facility (U-MARF) including on-line catalogue

access and user services.
Metadata for
Cataloguing
User
Search
Browse
Order
EPS Ground Segment
MSG Ground Segment

MTP Ground Segment
Products
Formatting &
media delivery
workshop PFD
Near line archive
on tape
Technologies:
Server: SUN Cluster environment and Solaris 10.
GS communication via redundant Ethernet networks with
automatic failover.
Internal Gigabit Ethernet network.
RAID array: EMC2 (CX600 - 35 TBytes).
SAN using Fibre Channel.
Archive library and tapes:
Sun StorageTek SL8500 (1448 slots enabled 724 TByte
native, 1.5 PByte compressed).
Primary media: Sun StorageTeek T10000 (12 drives).
Secondary media: LTO-4 (4 drives) (LTO-2 4 drives).
Archive HSM: SAM-FS (previously AMASS).
Distribution media: DVD, DAT(DDS-4), DLT (DLT8000), LTO-2, -4.
Compression: BZIP2,GZIP,PKZIP,UNIX Compress.
Catalogue DBMS: Oracle 11.
Internal communication between subsystems: Corba (TAO).
Bespoke SW: C++, Java 5.
Web Services, OGC standards
Data access and interoperability:

Data access through catalogue of metadata and catalogue
web portal user access layer:

Catalogue search and browse information;
Ordering and follow-up;
User management functions.
Implementation of international inter-operability access
planned until end of 2009: ISO metadata standards, ESA

HMA / OGC standards, WMO Information System standards.
Reprocessing:
Specific API and internal communication network with
Mission G/S allow optimal retrieval of products for

reprocessing.
Auxiliary data stored in U-MARF and retrieved for
reprocessing.
U-MARF catalogue products versioning scheme.
Archives data exploitation aspects (based on operational
status from 2007):

Performance figures & requirements:
Ingestion throughputs and timelines from GS (re-)processing
facilities:
Average daily ingestion rate: 186 Gbytes (EPS), 108 GBytes
(Meteosat).
Retrieval performances (under full ingestion rate):
over 1800 registered users for the on-line ordering (about 50
new users per month);
Archive and catalogue capacity sizing:
archive holding: about 375 TBytes, > 6,900,000 files;
incremental growth: to days configuration 3.2 PBytes.
Other salient operational aspects:
Availability constraints:
max 10 minutes downtime (fulfilled by concept of back-up
ingestion chain).
providing 2 days autonomy of operations on bulk orders).
Distinct OPEration and VALidation configurations.
Intrebari?
lucia.stefan@uclmail.net
Multumesc
Prelucrarea i analiza
imaginilor medicale
As. Dr. ing. Laura Maria Florea
Laboratorul de Analiza i Prelucrarea Imaginilor - LAPI
Cuprins

Ce sunt imaginile medicale?

De ce? Cum?
Aplicaii
Imagini medicale - raze X
Ecografii
Imagini de rezonan magnetic
Termografii
Imagini microscopice
De ce? Cum?
Scop: asistarea personalului medical n punerea

diagnosticului
Metode:
Digitizare (acolo unde este cazul)
cazul)
Prelucrare pentru vizualizare mai bun
bun
Analiz
Analiz automat
automat
Creare i indexare de baze de date
Aplicaii pentru imagini de raze X

Aplicaii n ortopedie
Fixarea protezei de old n canalul femural n cazul
artroplastiei de old
Detec
Detecia osteoporozei n imagini de calcaneu
provenite din investiga
investigaii ocazionale
Analiza automat a mamografiilor
Constantin
VERTAN
Laura
FLOREA
Alina
SULTANA
Proiecte
Sistem de
analiz digital a
imaginilor radiologice pentru
urmrirea protezelor ortopedice
SASOR/PRETOPIC
Osteoporosis Detection
Contribuii la mbuntirea
diagnosticului medical cu
ajutorul analizei i prelucrrii
imaginii digitale
Corneliu
FLOREA
Artroplastia de old
Operaie de nlocuire a ncheieturii oldului cu o

protez
Pacienii trebuie s se prezinte regulat la control

la medicul specialist
Dorim ajutarea medicului prin determinarea

automat a unor parametri clinici ai fixrii
protezei n canalul femural
Artroplastia de old
Digitizarea imaginilor radiologice

Scanner de film
Calitate bun
bun
Cost ridicat
Aparat de fotografiat
Ieftin, la ndemn
Pierdere de informa
informaie
gam
dinamic redus
Metoda multi-expunerii
( (Ev , f ) 2 ) f
N
Evi
f HDR =
i =1
(Ev , f )
i
i =1
Metoda ordonrii totale

Folosim o singur imagine

Diferenierea pixelilor prin informaia din
vecintate:

g(i,j)=[g
g(i,j)=[g1(i,j), g2(i,j), g3(i,j)
(i,j)]
ordonare lexicografic
lexicografic strict
formarea histogramei dorite
Rezultate diferite n funcie de alegerea

vectorilor de parametri, g(i,j)
Segmentarea imaginilor

Segmentare pe histogram: prguire simpl,

Otsu, Fisher, etc.
Segmentare n spaiul caracteristicilor: FCM
Mean Shift
Expectation-Maximization (EM)
Segmentarea orientat pe contururi: gradient,
compas, Harris, Canny
Segmentarea imaginilor: EM, Canny
Segmentare robust
Nici una dintre metodele amintite nu a putut

separa corect canalul femural
Folosim cunotine apriori despre imaginile

radiografice de proteze de old
Construim o metod adaptiv care s foloseasc

informaie de intensitate i informaie spaial
Segmentarea robust: rezultate
Parametri pentru diagnosticul clinic

Unghiul protez-os:
unghi mare
semnaleaz o presiune mare exercitat de

protez asupra femurului
Profilul de distan protez-os

distan
distane
mici pe toat suprafa

suprafaa pentru o fixare bun
Zona de adiacen protez-os

minim 3 zone
de adiacen
adiacen pentru o protez fix
10
Detecia osteoporozei
Boal caracterizat de pierderea de mas osoas

i de modificri micro-arhitecturale ale texturii
osoase
Poate fi detectat prin examinri cu echipamente

specializate sau subiectiv de ctre medicul
specialist
Analiza texturilor
os sntos
os osteoporotic
Scop: detecia
osteoporozei n imagini
nespecifice
Radiografii de clci
3 zone de interes
definite de medic
specialist
Analiza texturilor
Apariia osteoporozei implic modificarea

texturii osoase
Descriptori de textur pe baz de:

Fractali
Modelarea parametric
parametric a histogramei
Bancuri de filtre trece jos i Gabor
Descriptori de structur
structur color (MPEG(MPEG-7)
Baza de date cu
imagini marcate manual
ROI 0
Caracteristici
de textur
ROI 1
Decizie
Clasificare
supervizat
3.8% MD
7.6% FA
ROI 2
Analiza mamografiilor
Tip special de radiografie utilizat pentru

examinarea cancerului de sn
Cele dou anomalii:

Masele
Calcifierile
mbuntirea mamografiilor
Constrngeri de achiziie:
Timp de expunere redus
Concentra
Concentraie de radia
radiaie redus
Contrast sczut i detalii insufucient eviden
evideniate
Creterea contrastului i eliminarea zgomotului:

mbunt
mbuntirea contrastului folosind TopTop-Hat
mbunt
mbuntirea folosind prelucrarea logaritmic a
imaginilor (LIP)
mbuntire folosind Top-Hat

Tehnic morfologic de mbuntire a

contrastului
Tw ( f ) = f f o g
Tb ( f ) = f f g
mbuntire folosind LIP

Mamografia
original
Normalizare
clasic
mbuntire LIP
Ptracu
mbuntire
Jourlin
mbuntire LIP
Vertan
Detecia microcalcificaiilor
Microcalcificaiile: depozite mici, punctiforme,

de calciu; folosite ca markeri pentru detecia
timpurie a cancerelor asimptomatice
Detecie pe baza rezoluiei spaiale i a

contrastului
Algoritm adaptiv de cretere a regiunilor:

Extragerea m
mtii snului i eliminarea mu
muchiului
pectoral
mbunt
mbuntire cu TopTop-Hat alb
Prguire adaptiv
Etichetare
RegionRegion-merging
Filtrare fals pozitive
Detecia maselor

Masele: regiuni opace, dense

Probleme de detecie:
Contrast slab fa
fa de esutul din vecintate
Diversitate mare a formelor i dimensiunilor
Detecia maselor:
Pe baza unei proiec
proiecii
Pe baza a dou proiec
proiecii mediolaterale
Extragerea potenialelor zone

tumorale
difereniere ntre zonele suspecte i zonele de

esut fibroglandular sntos folosind informaii
de contrast, circularitate, arie
Detecia asimetriilor
Radiologii compar cele dou vederi mediolaterale sau cranio-caudale:
asimetriile indic existen

existena unei posibile leziuni
informa
informaie relevant n cazurile pacientelor
asimptomatice
Detecia asimetriilor
Pai

prere-procesarea imaginilor ini

iniiale i alinierea lor
extragerea conturului pentru fiecare sn
selec
selecia punctelor de coresponden
coresponden dintre cei doi
sni (curbura
(curbura fiecrui sn)
sn)
aplicarea unei transform
transformri spa
spaiale de potrivire
algoritmi de compara
comparaie pentru cele dou
dou
mamografii pe baza mediilor regiunilor
corespondente
Rezultate
Concluzii

Digitizare
Prelucrare:
Cre
Creterea gamei dinamice prin:
Multiexpunere
Metoda
ordonrii totale a pixelilor
mbunt
mbuntirea contrastului prin:
Transformata
TopTop-Hat
Metode de prelucrare logaritmic a imaginilor
Concluzii
Segmentare:
ExpectationExpectation-Maximization
Canny
MeanMean-Shift
Analiz:
Textural
Pe baz de contrast,
contrast, form
form,, arie, circularitate
etc
Bibliografie selectiv
Artroplastia de old
C. Vertan,
Vertan, L. Florea,
Florea, C. Florea,
Florea, A. Oprea,
Oprea, M. Popescu,
Popescu, S. Cristea:
Cristea: ClinicalClinicalradiological correlation of the fit of the femoral component of a hip
prosthesis: THA followfollow-up by digital XX-ray image processing
processing, EFORT
Congress, Nisa,
Nisa, Fran
Frana, 2008.
C. Florea, C. Vertan, L. Florea, A. Oprea: On the Use of Logarithmic
Image Processing Models in Total Hip Prostheses XX-ray Visualization
and Analysis
Analysis, Medical Image Understanding and Analysis MIUA, pp. 127127131, Dundee, Marea Britanie, 2008.
L. Florea, C. Vertan, C. Florea, A. Oprea: Dynamic Range Enhancement
of Consumer Digital Camera Acquired Hip Prosthesis Xray Images
Images,
EUSIPCO, pp. 11031103-1106, Poznan, Polonia, 2007.
C. Florea,
Florea, C. Vertan,
Vertan, L. Florea:
Florea: Logarithmic ModelModel-based Dynamic
Range Enhancement of Hip Xray Images
Images, ACIVS, vol. LNCS 4678,
Eds. J. BlancBlanc-Talon et al, pp. 587587-596, Delft, Olanda,
Olanda, 2007.
10
C. Vertan,
Vertan, L. Florea,
Florea, I. tefan,
tefan, M. Popescu,
Popescu, . Cristea:
Cristea: ClinicalClinical-radiological
correlation of the bone structure analysis by multimedia MPEGMPEG-7
descriptors: osteoporosis screening by calcaneal digital XX-ray image
processing
processing, EFORT Congress, p. 10.134, Floren
Florena, Italia, 2007.
C. Vertan, I. tefan, L. Florea: Detection of postmenopausal alteration of
bone structure in digitized Xrays
Xrays, CAIP, vol. LNCS 4673, Eds. W. G.
Kropatsch, M. Kampel, A. Hanbury, pp. 278278-284, Viena, Austria, 2007.
L. Srbu,
rbu, C. Vertan,
Vertan, I. tefan:
tefan: Gabor Textural Descriptors for Early
Osteoporosis Detection in Calcaneal X-Rays
Rays, n Proceedings
Communications, pp. 201201-204, Bucure
Bucureti,
ti, Romnia,
Romnia, 2006.
C. Vertan,
Vertan, L. Srbu,
rbu, I. tefan:
tefan: MPEGMPEG-7-like Description Scheme for
Osteoporosis Diagnosis Support
Support, n Proceedings AQTR, vol. 2, pp. 397397400 (ISBN 22-42444244-03600360-X), ClujCluj-Napoca, Romnia,
Romnia, 2006. (ISI Proceedings/
INSPEC)
Analiza mamografiilor

Alina Sultana, Mihai Ciuc,

Ciuc, Rodica Strungaru,
Strungaru, Laura Florea:
Florea: A New
Approach in Breast Image Registration
Registration, ICCP, Cluj,
Cluj, Romnia,
Romnia, 2010.
Alina Sultana, Mihai Ciuc, Rodica Strungaru, erban Oprisescu, Laura Florea:
Automated Bilateral Asymmetry Detection in Mammograms, ECIT,
Ia
Iai, Romnia,
Romnia, 2010.
A. Sultana, M. Ciuc,
Ciuc, L. Florea,
Florea, C. Florea:
Florea: Detection of Mammographic
Microcalcifications Using a Refined RegionRegion-Growing Approach
Approach,
ISSCS, Ia
Iai, Romnia,
Romnia, 2009.
Oprea,
Oprea, A.E.; Strungaru,
Strungaru, R.; Forsea A.; Ungureanu G.M.: "A
"A Novel
Approach in Melanoma Identification",
Identification", 4th European Conference of the
International Federation for Medical and Biological Engineering, Antwerp,
Belgium, 2008.
Oprea, A.E.; Strungaru, R.; Ungureanu G.M.: "A
"A Self Organizing Map
Approach to Breast Cancer Detection",
Detection", 30th Annual International IEEE
EMBS Conference Vancouver, British Columbia, Canada, 2008 .
11
V mulumesc
pentru atenie!
Activitatea de cercetare a fost par

parial finan
finanat
at din
POSDRU prin contractul de finan
finanare
POSDRU/89/1.5/S/62557.
12
PhD in Schools of Excellence
Doctoratul
nEconomic
coli de Sciences
Excelen
Social and
Evaluarea calitii cercetrii n universiti i

creterea vizibilitii prin publicare tiinific
Lorem ipsum dolor sitProfessor,
amet, consectetur
adipiscing
Fusce gravida sem laoreet
Ph.D.,
Danielelit.DAVID
dolor.
Cum sociis
penatibus
et
magnis
dis
parturient
montes, nascetur
Prof.
Dr.natoque
Grigore
Bozga
Babe-Bolyai University, Cluj-Napoca, Romania
ridiculus mus. Quisque
velSinai
nisl.School
In hac habitasse
platea
dictumst.
In id libero ut erat
Mount
of Medicine,
New-York,
USA
accumsan varius.
OBIECTIVUL GENERAL al Proiectului
Elaborarea, testarea i implementarea unei metodologii care s

asigure evaluarea la standarde internaionale a calitii cercetrii n
universiti.
REZULTATE ateptate:
Ierarhizarea pe domenii tiinifice a universitilor romneti
Identificarea grupurilor de cercetare ale cror lucrri se ridic la nivel

internaional
Exerciiul Naional al Evalurii Cercetrii (ENEC) va furniza factorilor de

decizie un instrument de msurare a calitii cercetrii din universitile
romneti
Creterea eficienei alocrii resurselor financiare pentru cercetare prin

formularea unor propuneri legislative care s coreleze finanarea cu poziia n
ierarhia pe domenii
Evaluarea cercetrii n tiine Inginereti

Domenii:
Inginerie Civil i Instalaii
Inginerie Mecanic i Mecatronic
Inginerie Aerospaial
Transporturi
Inginerie Chimic
tiina i Ingineria Materialelor
Mine, Petrol i Gaze
Inginerie Industrial
Inginerie Electric
Energetic
Electronic i Telecomunicaii
Ingineria Sistemelor
Calculatoare i Tehnologia Informaiei
Biotehnologii, Ingineria i Sigurana Alimentelor
tiina i Ingineria Mediului
Pentru fiecare domeniu se constituie un panel de experi strini i romni
Lista domenilor de cercetare ce vor fi evaluate
CRITERII PENTRU EVALUARE

Criteriul I - Rezultate obinute n activitatea de cercetare tiinific/creaie artistic
(50-70%)
Criteriul II - Mediul de cercetare tiinific/creaie artistic
(10-30%)
Criteriul III - Recunoaterea n comunitatea academic
(5-15%)
Criteriul IV - Resurse financiare atrase pentru cercetare
(5-15%)
Fiecare criteriu este cuantificat prin mai muli indicatori

Cuantificarea are 2 componente: (i). Cantitativ i (ii). Calitativ
Ponderile i indicatorii specifici domeniului
Evaluarea se face de ctre experi strini i romni (2/1)
Finalizarea procesului: septembrie 2011
Obs: Criterii similare utilizate n evalurile Comisiei Europene, RAE, etc.
DESCRIPTORI INDICATORI
Criteriul I: Rezultate obinute n activitatea de cercetare tiinific/creaie artistic
Pondere: 65%
Descriptori:
Articole
Cri tiinifice de autor
Brevete
Alte realizri supuse legii dreptului de autor i a drepturilor conexe care implic
creaia ca proces al cercetrii i inovrii din domeniile arhitectur i arte
Produse i/sau servicii inovative cu impact economic demonstrabil
Indicatori:
Maxim 3 indicatori
Criteriul II: Mediul de cercetare tiinific/creaie artistic
Pondere: 15%
Descriptori:
Realizarea de traduceri, antologii i editarea de volume
Conductori de doctorat
Organizarea de manifestri tiinifice i de creaie artistic de nivel internaional
Existena unor mecanisme proprii pentru atragerea tinerilor cercettori
Susinerea financiar de ctre universitate a cercetrii n domeniile asumate prin
strategii naionale
Program de investiii pentru laboratoare/ateliere dotate cu infrastructur specific
cercetrii tiinifice/creaiei artistice
Capacitatea universitii de a asigura accesul la literatura de specialitate
Capacitatea universitii de a susine programe postdoctorale
Indicatori:
Maxim 4 indicatori
Criteriul III: Recunoaterea n comunitatea academic
Pondere: 10%
Descriptori:
Lucrri invitate la conferine internaionale de prestigiu
Profesor invitat la universiti de prestigiu
Poziii de conducere n organizaii profesionale internaionale
Traduceri ale contribuiilor tiinifice proprii publicate de edituri din strintate
Citri i recenzii ale creaiei de autor
Membru al Academiei Romne, al academiilor de ramur i al academiilor din
strintate
Recenzor la edituri de prestigiu i reviste cotate Web of Science
Membru n Board-uri ale unor reviste cotate Web of Science
Indicatori:
Maxim 3 indicatori
Criteriul IV: Resurse financiare atrase pentru cercetare
Pondere: 10%
Descriptori:
Fonduri atrase pentru cercetare
- prin competiii naionale
- prin competiii internaionale
- prin contracte directe cu teri
Fonduri atrase din servicii/produse inovative/creative
Indicatori:
1 indicator
Criteriul I: Rezultate obinute n activitatea de cercetare tiinific/creaie artistic
Pondere: 65%
Descriptori n tiine Inginereti:
Articole tiinifice
- Articole n reviste din fluxul principal, cotate ISI;
- Articole n publicaii indexate n Baze de Date Internaionale (BDI).
Bazele de Date Internaionale luate n calcul sunt:
Engineering Village, www.engineeringvillage.com (include 15 baze de date)
SCOPUS, www.scopus.com
Factorul de calitate se va situa ntre valorile 0.7 1.2; fiecare dintre articolele supuse evalurii, va fi
ncadrat pe unul dintre urmtoarele 4 paliere:

de nivel internaional de vrf;

de nivel internaional;
de nivel naional;
de nivel local.
Cri tiinifice, monografii
Crile tiinifice considerate pentru evaluarea cercetrii n domeniul tiinelor Inginereti sunt clasificate n
vederea evidenierii nivelului de vizibilitate dup cum urmeaz:
Monografii n edituri de prestigiu din strintate. Se pune accentul pe monografii publicate n

edituri dintr-o list stabilit de coordonatorii panelurilor de tiine Inginereti:
Elsevier
Springer
John Wiley & Sons
McGraw-Hill
CRC Press
Francis & Taylor
Oxford University Press
Cambrige University Press
Academic Press
Kluwer Academic Publishers
Capitole de carte publicate n edituri de prestigiu din strintate (conform listei precedente)
Cri tiinifice publicate n ar, incluse n depozitul Bibliotecii Naionale
Factorul de calitate se va situa ntre valorile 0.7 1.2; fiecare carte supus evalurii calitative va fi ncadrat
pe unul dintre urmtoarele 4 paliere:


de nivel naional;
de nivel local.
Brevete de invenie
Acest descriptor cuantific rezultatele obinute n transferul cunotinelor acumulate prin cercetri
fundamentale i aplicative spre aplicaii practice inginereti, cu impact economic i social.
Brevete de invenie internaionale;

Brevete de invenie naionale.
Factorul de calitate se va situa ntre valorile 0.8 1.2; fiecare dintre brevetele supuse evalurii va fi ncadrat
pe unul dintre urmtoarele 3 paliere de calitate:
de nivel internaional de vrf,;
de nivel naional.
Produse i servicii inovative

Produsele i serviciile inovative sunt evaluate funcie de efectele economice nregistrate. Pentru
cuantificarea lor trebuie evitat suprapunerea cu fondurile atrase prin contracte cu mediul economic
(Criteriul IV) de ctre universitate.
Factorul de calitate asociat acestui indicator va fi determinat pentru un procent stabilit din produse sau
servicii considerate de ctre cercettorii din domeniul evaluat ca fiind reprezentative. Factorul de
calitate se va situa ntre valorile 0.7 1.2; fiecare dintre produsele/serviciile supuse evalurii va fi
ncadrat pe unul dintre urmtoarele 4 paliere de calitate:


de nivel naional;
de nivel local.
Pentru domeniul tiine Inginereti raportarea se va face pe toi cei patru indicatori, urmnd
sa fie alei doar trei dintre ei, selecia fcndu-se ntre Cri tiinifice de autor i
Produse i servicii inovative, n funcie de domeniu.
Criteriul II: Mediul de cercetare tiinific/creaie artistic
Pondere: 15%
Conductori de doctorat
Factor de calitate se stabilete funcie de:
(1) numrul i calitatea publicaiilor rezultate n medie pe teza de doctorat finalizat.
Se urmeaz aceiai factori de calitate ca la Articolele tiinifice de la Criteriul I.
(2) numrul tinerilor care dup obinerea titlului de doctor au fost acceptai la universiti prestigioase
(clasament Shanghai i /sau Times) pentru efectuarea de stagii post-doctorale.
Nivel internaional de vrf:
burse Marie Curie, Humboldt, Fulbright, JSPS, sau Universiti din Top 100;
Nivel internaional:
burse post-doctorale n Universiti din Top 500;
Nivel naional:
Universiti romneti sau Institute ale Academiei Romne.
Nu exist nivel local.
Organizarea de manifestri tiinifice internaionale
Se ia n considerare numrul manifestrilor tiinifice internaionale desfurate n perioada evaluat, conferine
ale cror volume (proceedings) ndeplinesc urmtoarele condiii pentru vizibilitate internaional:
denumirea, prezentarea lucrrilor i limba n care sunt publicate articolele s fie de circulaie
internaional (englez, francez, spaniol, german, rus);
conferina trebuie s aib comitet internaional de program;
lucrrile s fie evaluate n sistem peer-review, rata de respingere s fie peste 25%;
tirajul volumului de lucrri s depeasc numrul autorilor i s aib cod ISBN;
conferina s poat fi accesat pe web (organizator, comitet de program, cuprins volum de lucrri,
acces la rezumate);
minimum 50% din articole s fie din strintate (fr coautori din Romnia);
minimum 50 de articole publicate n volum.
Atragerea tinerilor cercettori
Se va lua n considerare numrul tinerilor cercettori (35 ani mplinii la momentul evalurii), fiecare
cercettor fiind ponderat cu un factor calculat astfel: nr. luni activitate full time / 60.
Factorul de calitate se va situa ntre valorile 0.7 1.2; atribuirea acestuia se face pe baza
urmtoarelor 4 paliere generale:
de nivel naional;
de nivel local.
Program de investiii n laboratoare, concretizat n infrastructura pentru cercetare
tiinific
Acest descriptor se evalueaz cantitativ prin valoarea investiiilor universitaii n laboratoarele de
cercetare din domeniul evaluat (n sute de mii EURO) pentru perioda evaluat (5 ani).
Sunt luate n considerare investiiile n laboratoare de cercetare, NU n laboratoare de uz didactic.
Pentru acest descriptor se face evaluarea n teren.
Factorul de calitate se va situa ntre valorile 0.7 1.2:

de nivel naional;
de nivel local.
Criteriul III: Recunoaterea n comunitatea academic
Pondere: 10%
Prelegeri invitate la conferine internaionale, profesor invitat la universiti de

prestigiu
Se consider numrul prelegerilor invitate pe toat perioada evaluat.
Citri ale lucrrilor tiinifice
Se consider numrul citrilor acumulate pe toata perioada evaluat.
Membri ai Academiei Romne, Academiei de tiine Tehnice, Academiilor

naionale din strintate
Se consider numrul cercettorilor care sunt membri ai academiilor de mai sus.
Factorul de calitate se va situa ntre valorile 0.7 1.2;
.
Recunoaterea internaional de vrf, reflectat prin:

200 de citri pentru perioada evaluat i indice Hirsch > 8 sau
Membru al unei Academii naionale din strintate.
Recunoatere internaional, reflectat prin:

100 de citri pentru perioada evaluat i Indice Hirsch > 5 sau
Membru al Academiei Romne.
Recunoatere naional, reflectat prin:

50 de citri pentru perioada evaluat sau
Indice Hirsch >3 sau
Membru al unei academii de ramur sau
5 prelegeri la universiti de prestigiu.
Recunoatere local:
Pe acest palier vor fi ncadrai toi cercettorii care nu au fost ncadrai n celelalte paliere.
Criteriul IV: Resurse financiare atrase pentru cercetare
Pondere: 10%
Evaluarea aferent Criteriului IV este integral cantitativ, fiind realizat n mod automat
de ctre platforma informatic de suport al evalurii, fr implicarea experilor
evaluatori.
Aceast evaluare se face pe baza contractelor de cercetare raportate de o universitate
pe un anumit domeniu, pe perioada evaluat.

Preswed14 Interesant

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Preswed14 Interesant

Încărcat de

Drepturi de autor:

Formate disponibile

WORHSHOP EXPLORATORIU

Electronica, Telecomunicaiile si Tehnologia

Electronica, Telecomunicaiile i Tehnologia Informaiei

Miercuri, 22 Septembrie 2010

Cerc.dr. Irina Diana COMAN.......................................................................pag.4

.l.dr.ing. Cosmin Radu POPA.................................................................pag.88

Conf.dr.ing. Constantin PALEOLOGU...................................................pag.155

.l.dr.ing. Nicolae Gheorghe MILITARU................................................pag.170

Sear Gala Cin restaurant Crama Domneasc

Joi, 23 Septembrie 2010

Prof.dr.ing. Mihaela ULIERU..................................................................pag.192

Ing. Lucia STEFAN..................................................................................pag.231

Prof. dr. ing. Grigore BOZGA.................................................................pag.263

Prof. dr. ing. Grigore BOZGA

Sisteme automate de masurare si analiza a

Irina Diana Coman

Center for Applied Software Engineering

In 2009 si 2010, organizator al scolii de vara CASE

Cursul de Ingineria Programarii

Membru al comitetului de program al European

Masurarea si analiza proceselor de

Dar de ce sisteme automate?

Dar de ce sisteme automate?

Dar de ce sisteme automate?

Dar de ce sisteme automate?

Dar de ce sisteme automate?

Solutia: sisteme automate pentru colectarea si analiza datelor

Datele disponibile imediate

Masurare in ingineria programarii

Neinvazive, ieftine, date precise si corecte

Metrici pentru proces si pentru produse

(industrie) referinta, trenduri, situatie actuala, rapoarte

Potential mare, dar inca nefolosit.

Date multe si ieftine

Utile si pentru cercetare?

Studiu de caz: adoptia si utilizarea

Departament IT in firma non-IT

Procesul de adoptie detaliat in 5 pasi

Auto-monitorare, diagnosticare si reparare

Analiza datelor pe 3 coordinate majore:

Profilul utilizatorului individual

Lectia 4 Vizualizarea datelor

Prezentarea datelor AISEMA:

Agregarea datelor o directie de explorat.

Mulumesc pentru atenie!

How to gain support from developers:

Ensure data privacy.

Be ready for a long start-up:

Planning can take very long (9 months):

Training takes time:

Busy calendars and different roles and backgrounds;

Deployment can take very long (9 months):

Many issues surface only during intensive usage;

Use any other data that are already being

Data presentation is as important as data

Developers, managers and team leaders will

Provide different aggregated views instead of

Automation of time-consuming tasks;

Continuously check data accuracy;

Enhancements of AISEMA Systems

Ensuring continuous data accuracy

Automated inference of higher-level

Automatically splitting development sessions into

Ensure continuous data accuracy