Upload
h
View
215
Download
1
Embed Size (px)
Citation preview
1-4244-0216-6/06/$20.00 ©2006 IEEE
Mobile Multimedia Group Conferencing - Enriching H.264-based Video by Mobile Source
Specific Multicast Communication Hans L. Cycon, Thomas C. Schmidt, Matthias Wählisch, Mark Palkow and Henrik Regensburg
Abstract — In this paper we report on a multimedia communication software with a distributed architecture and its applications. It is a simple, ready-to-use scheme for distributed presenting, recording and streaming multimedia content over unicast or multicast networks. Dedicated end-to-end bandwidth management optimizes its network resource consumption. Additionally we report on group communications schemes for future applications within mobile networks. We present straightforward extensions to session signalling and source specific multicast routing for transforming (morphing) previous delivery trees into optimal trees rooted at a relocated source. This extension scheme only requires basic signalling mechanisms, explicit joins and prunes, which are present in current multicast routing protocols such as PIM-SM. First evaluations grounded on real-world Internet topologies indicate network performance superior to traditional distribution schemes*.
Index Terms — Video and Multimedia Group Conferencing,
E-learning, Mobile IPv6, Source Specific Multicast Mobility
I. INTRODUCTION
Mobile internet usage becomes more and more a day-to-
day application. Additionally visual devices performing
synchronous communication such as voice or video
conferencing over IP (VoIP/VCoIP) are now almost
ubiquitous. This raises new challenges for the Internet
infrastructure, such as mobile conference users. The
availability of new, truly mobile IP enabled sub network
layers not only offers connectivity to nomadic users at
roaming devices, preserving communication sessions
beyond IP subnet changes, but re-raises questions
concerning the quality of IP service: The constant bit rate
scenarios of voice and video conferencing will appear
significantly disturbed by packet loss intervals, delays or
* Hans L. Cycon is with the FB 1, FHTW Berlin, Treskowalle 8, 10318 Berlin, Germany (e-mail: [email protected]).
Thomas C. Schmidt is with the Department of Electrical Engineering and Computer Science, HAW Hamburg, Berliner Tor 7, 20099 Hamburg. He is also with the Computer Center, FHTW Berlin, Treskowallee 8, 10318
Berlin, Germany (e-mail: [email protected]). Matthias Wählisch is with the Computer Center, FHTW Berlin,
Treskowallee 8, 10318 Berlin, Germany (e-mail: [email protected]).
Mark Palkow is with the daViKo GmbH Berlin, Hoenower Strasse 35/PF 16, 10318 Berlin, Germany (e-mail: [email protected])
Henrik Regensburg is with the FB 1, FHTW Berlin, Treskowalle 8,
10318 Berlin, Germany (e-mail: [email protected]).
jitter exceeding 100 ms. Thus, when heading towards
VCoIP as a standard Internet service, important steps for
global usability have to be taken, focusing on ease and
quality.
In VCoIP conferencing scenarios each member
commonly simultaneously acts as a group listener and a
source. Therefore group communication and especially IP
multicasting will be of particular importance to mobile
environments, where users commonly share frequency
bands of limited capacities. While mobility of a listener can
relatively easily be handled, source mobility presents a
severe problem for multicast packet distribution.
There are two Internet approaches dealing with multicast
group communication: Source Specific Multicast (SSM)
[11] and Any Source Multicast (ASM) [12]. SSM, still in its
design process, is considered a promising improvement of
group distribution techniques. In contrast to ASM, optimal
multicast source trees are constructed immediately from
(S,G), i.e. Source address - Group address router states
subscriptions at the client side, without utilizing network
flooding or Rendezvous Points. Source addresses are to be
acquired by out of band channels, i.e. a SIP [9] session
initiation in conferencing scenarios. As a consequence,
routing simplifies significantly. In VCoIP conferencing
scenarios each member commonly simultaneously acts as a
group listener and a sender. In a mobile scenario, routing
thus invalidates with changing source addresses. Up until
now SSM source mobility remains as an unsolved problem
[17, 18].
In this paper we present a multimedia communication
system including a VCoIP (Video Conferencing over IP)
software with a distributed architecture and some
applications. We further on discuss session mobility with the
special focus on real-time multicast.-
The paper is organized as follows. Section II presents the
basic video conference software and some of its applications
and test results. In Section III we review the basic problems
of multicast source mobility and related work, sketch our
approach to SIP-based source specific group initiation and
introduce tree morphing, our new approach to source
specific multicast sender mobility. Finally, section IV is
dedicated to conclusions and outlook.
II. A DISTRIBUTED COMMUNICATION SYSTEM
A. The Basic Software
The digital audio-visual conferencing system we use is a
server-less multipoint video conferencing software without
MCU developed by the authors [2]. It has been designed in a
peer-to-peer model as a lightweight Internet conferencing
tool aimed at email-like friendliness of use. Guided by the
latter principle, it refrained from implementing H.323 client
requirements [1].
The system is built upon a fast H.264/MPEG-4 AVC
standard conformal video codec implementation [3]. It is a
Baseline profile implementation, optimized for real-time
decoding and encoding by several accelerating measures
like diamond shape motion search, MMX enhanced SAD
motion estimation, fast mode selection and a fast subpel
search strategy.
There is also an application-tailored fast wavelet-based
video codec [4] used for higher available data rate.
By controlling the coding parameters appropriately, the
software permits scaling in bit rate from 24 to 1440 kbit/s on
the fly. Audio data is compressed using a 16 kHz-speech-
optimized variable bit rate codec [5] with extremely short
latencies of 40 ms (plus network packet delay). All streams
can be transmitted by unicast as well as by multicast
protocol. Audio streams are prioritized above video since
audio communication is more sensitive to distortions in
erroneous networks.
An application-sharing facility is included for
collaboration and teleteaching. It enables participants to
share or broadcast not only static documents, but also any
selected dynamic PC action like animations including mouse
pointer movements. All audio/video (A/V) - streams
including dynamic application sharing actions can be
recorded on any site. These data can be displayed locally or
automatically converted into a web streaming format, which
is internet wide available.
This system is equally well suited to intranet and wireless
video conferencing on a best effort basis, since the
audio/video quality can be controlled to adapt the data
stream to the available bandwidth. In faulty network
conditions like poor WLAN links we use unicast TCP
transmissions to avoid distortions by packet loss. For point-
to-multipoint situations like virtual classrooms there is also
a possibility to switch to multicast network transmission via
UDP to minimize computation and transmission load for
networks and senders. To avoid problems with non-routable
private network segments we use Simple Traversal of UDP
over NATs (STUN). This protocol is mostly used for
assisting devices behind a NAT, a firewall or router with
port blocking, see [16].
The joined use of high bandwidth UDP traffic with TCP
updates bound to real-time demands is known to suffer from
distortions due to TCP traffic suppression. Application
sharing in conferencing applications thus is endangered to
encounter disruptions in the event of network congestion.
For a service enhanced synchronous use of UDP media
sessions and application sharing with reliable data transport
requirements, we implemented end-to-end load balancing
employing proprietary extensions to UDP, reliable (RUDP).
We work on its packet identifiers to control application
sharing data flows. On the occasion of a significant amount
(e.g., 5) of unacknowledged packets from shared
application, we slow down video packet transmission to
reserve required resources for real-time application updates,
see [8]. Audio communication remains undisturbed of load-
balancing actions.
Global connection between conference participants in the
internet will be established by a dynamic user session
recording. We denote this by User Session Locator (USL)
and store appropriate session information within a
Lightweight Directory Access Protocol (LDAP) directory
server (see [6] for further details).
B. Application Scenarios
Based on the system’s capabilities, new scenarios for
synchronous and asynchronous distributed learning evolve.
Actually the system is designed for and used in various
learning scenarios.
(i) Synchronous distributed learning scenario
Teacher and students are connected by LAN or WLAN.
All participants establish mutual connections via web server.
The teacher can than send his PC presentations and
applications to the students PCs. Students (outside or inside
the lecture room) can participate active and/or passive by
real-time audio/video with latencies (in LANs) well below
50 ms. Since all participants can send their presentations or
applications via WLAN to a beamer in a conference room,
this can be used as a “wireless” connected beamer which
can present full video formats.
All participants can also initiate co-operations in small
groups via full video conferencing. Within the peer to peer
network each student can send, receive and work on any PC-
applications for collaboration.
Additionally there is also a live streaming option. Any
participant is able to send one audio/video stream to a
preconfigured windows media streaming server via push
distribution. This client works like an A/V gateway. All
H.264 video conferencing video streams will be mixed to
one video stream and then live-transcoded into windows
media format. So additionally also passive participants can
join the videoconference, without overloading the
bandwidth between the peer to peer videoconferencing
groups. The streaming server adapts the available
connection bandwidth of the passive viewers by using multi-
bit rate transcoding profile.
Fig. 1. Video Conferencing Live Streaming
Thus all viewers get the best optimization for there
appropriate connection bandwidth.
(ii) Asynchronous distributed learning scenario
Each station can record all sessions. The recordings can
be stored locally or made net-wide accessible a by
converting it e.g. into a MS streaming format and uploaded
to an e-learning platform. Lecture room presentation or
distributed group work are thus ready to be played back
anywhere at any time by streaming video.
III. MOBILITY AWARE SOURCE SPECIFIC MULTICAST
COMMUNICATION
A. The Mobile Multicast Source Problem
Mobility today must be seen as one of the major driving
forces for multimedia data transmission. Cellular phones
and portable paddles we expect to carry individual Internet
addresses soon, as available from IPv6 address space, and to
operate mobility supporting Internet protocols as the
recently released MIPv6 [7]. Multimedia applications s. a.
our video conferencing system will request seamless support
for mobile group conferencing, thereby occurring as
simultaneous multicast sender and receiver to the Internet
infrastructure.
Source mobility presents a severe problem for multicast
packet distribution. Even though multicast routing itself
supports dynamic reconfiguration, as members may join and
leave ongoing group communication over time, multicast
group membership management and routing procedures are
intricate and too slow to function smoothly for mobile users.
In addition multicast imposes a special focus on source
addresses. Applications commonly identify contributing
streams through source addresses, which must not change
during sessions, and routing paths in most protocols are
chosen from destination to source.
Routing overheads in Any Source Multicast (ASM) [12]
up until now have hindered the widespread availability of
multicast services. It is currently expected that the
simplified, interdomain-transparent group communication
scheme of Source Specific Multicast (SSM) [11] will offer
the basis for widely available group communication support
at the network layer.
Source addresses in source specific multicast are
requested to be known prior to group subscription. They
need to be shared by the entire group of conference
members and thus are to be provided by the signaling
scheme for group initiation. The latter can be achieved by
appropriate SIP [9] negotiations. Additionally, source
addresses carry the dual meaning of client–subscribed
source–group identifiers at the one hand, and routing
location information at the other hand. Both semantics need
to be followed by receivers and intermediate routers.
The ‘lightweight’ SSM approach to group communication
can thus be considered as a highly appropriate Internet
solution for multimedia conferencing, but enforces
extensions to session initiation and mobility management.
B. Session Initiation for Source Specific Group Conferencing
Session initiation for conferencing applications
commonly is negotiated by the session initiation protocol
SIP [9]. Within its invite message SIP reports on
contributing node parameters and media session specific
characteristics via an SDP data set. SIP accounts for
multicast group conferencing by allowing a multicast
distribution of its signaling via an maddr header field.
However, SIP core procedures are bound to Any Source
Multicast communication, which significantly simplifies the
coordination of primarily unknown members of the
distributed conferencing system.
To enable group communication by Source Specific
Multicast, SIP dialogs need alteration in the following way:
A new conferencing member willing to join a previously
established group conference invites any party and receives
acknowledgement including multicast session descriptions
via unicast. The invited party then has to repeat the
acknowledgement to a previously established SSM signaling
domain, in order to trigger an active source subscription of
all previous group members to the newly established caller.
All additional group members subsequently will advertise
their session affiliation, while the initially called party will
signal a turnover of its newly established SIP signaling
channel to SSM multicast. As soon as the new conferencing
member has completed subscription to SIP signaling and
media session groups for all conference party's addresses, a
Source Specific Multicast group conference is fully
established among peer-2-peer members in the absence of
any coordinating instance.
C. Tree Morphing – Introducing Multicast Routing Trees Adaptive to Mobility
In the present section we will briefly introduce our new
concept of multicast routing, adaptive to source mobility. A
mobile multicast source (MS) away from home will transmit
unencapsulated data to a group using its Home Address
(HoA) on the application layer and its current Care-of-
Address (CoA) on the Internet layer, just as unicast packets
are transmitted by MIPv6. Likewise data packets will carry
a mobility destination option header to pass HoA as source
identifier to the application layer at the receiver side. In
extension to unicast routing, though, the entire Internet
layer, i.e. routers included, will be
R
R R
D2D1
R
R
R RpDR nDR
MS MS
Fig. 2. Elongation of the Root of the Delivery Tree
aware of the permanent HoA. Maintaining address pairs in
router states like in binding caches will enable all nodes to
simultaneously identify (HoA,G) – based group membership
and (CoA,G) – based tree topology.
When moving to a new point of attachment, the MS will
alter its address from previous CoA (pCoA) to new CoA
(nCoA) and eventually change from its previous Designated
multicast Router (pDR) to a next Designated Router (nDR).
Subsequent to handover it will immediately continue to
deliver data along an extension of its previous source tree.
Delivery is done by elongating the root of the previous tree
from pDR to nDR (s.Fig. 2). This extension is achieved
through a state update message, carried in a Hop-by-Hop
option header and sent to the multicast destination address
using source routing through pDR. All routers along the
path, located at root elongation or previous delivery tree,
thereby will learn MS’s new CoA and implement
appropriate forwarding states.
Routers on this extended tree will use RPF checks to
discover potential short cuts. Registering nCoA as source
address, those routers, which receive the state update via the
topologically incorrect interface, will submit a join in the
direction of a new shortest path tree and prune the old tree
membership, as soon as data arrives. All other routers will
simply overwrite their (pCoA,G) state with
(nCoA,G).Thereby all parts of the previous delivery tree,
which coincide with the new shortest path tree, are re-used.
Only branches of the new shortest path tree, which have not
previously been established, need to be constructed. In this
way the previous shortest path tree will be morphed into a
next shortest path tree as shown in figure 3 and 4.
R
R R
D2D1
R
R
R
R RpDR nDR
MS MS
Fig. 3. Intermediate Morphing State
R
R R
D2D1
R
R
R RpDR nDR
MS MS
Fig. 4. Final Morphing State
Note that this algorithm does not require data
encapsulation at any stage. It is not built upon a specific
multicast routing protocol, but will require the following
functional mechanisms compliant with current protocols
such as PIM-SM [13]:
• Outgoing router interfaces need to maintain (S,G) states to denote their partition in the distribution tree. These states will be extended to include the Home Address identifier (S, G, HoA).
• Routers need the ability to explicitly join an (S,G) state.
• Routers need the ability to explicitly prune an (S,G) state. Alternatively, but with lower efficiency, routing states may time out.
• Finally, the computation of standard Reverse Path Forwarding (RPF) check is used.
For the details of signaling and routing protocol
extensions under SSM mobility we refer the reader to [14].
D. Performance Evaluation
Mobility initiated handovers may in general lead to
packet loss and delay. The tree morphing multicast routing
scheme will not produce any packet loss in addition to
mobile IP handovers, as can be easily concluded from
primary packet forwarding relying on unicast source routes.
For a first evaluation measure we will subsequently
concentrate on handover initiated packet delay as a result
from initially suboptimal delivery trees. Based on real-world
Internet topologies we simulate the packet distribution and
compare our results to the bi-directional tunneling approach
[7], which currently is the only stable mobility solution for
SSM source mobility.
To judge on performance quality of the tree morphing
(TM) scheme, we now analyze its delay effects within
realistic Internet topologies. We per-formed a stochastic
discrete event simulation based on the network simulator
platform OMNeT++ 3.1 [10] and several real–world
topologies of different dimensions. The selection of network
data in our simulation must be considered critical, as key
characteristics of multicast routing only make an impact in
large networks, and as topological setup fixes a dominant
part of the degrees of freedom in routing simulations.
We chose the ATT core network [15] as a large (154
nodes), densely meshed single provider example. For inter–
provider data we extracted sub-samples of varying sizes
from the ”SCAN + Lucent” map [19, 20], the result of two
extensive Internet mapping projects containing 284.805
network nodes connected by 449.246 links. Sample sizes,
154 and 15.400 nodes, vary by two orders of magnitude.
The delay excess relative to optimal routes has been
calculated as characteristic performance measure under the
assumption of homogeneous link delays. Extreme values,
i.e. maximal delays at initial elongation phase and minimal
after convergence, were evaluated for tree morphing (TM)
as functions of the distance from pDR to nDR. In detail,
designated routers within a given topology were randomly
chosen according to their predefined distances. For each pair
of edge routers at the mobile source a uniformly distributed
set of 20 receivers was established and delay values were
taken from average reception time. Sampling of source
positions was repeated 20 times for each parameter set in
order to better explore the large phase space. Comparisons
are drawn with bi-directional tunneling (BT), which does
not depend on designated router distances, but on home
agent (HA) position. The delay excess in BT as function of
HA position does not converge to a characteristic value, but
rather admits a broad distribution. The latter has been
derived from scattering HA positions uniformly within the
sample networks.
The results of our simulations are displayed in figure 5
and 6. pDR to nDR distances were chosen between 2 and
10, except for the ATT network, which exhibits a maximal
edge router separation of 5.
Fig. 5. Internet 15.400 N odes
Fig. 6. ATT Core Network
Error bars indicate the standard deviation of initial TM
delay excess, as calculated from events differing in location
of the mobile source. Plotted lines indicate the linear
regression curves derived from this result set. Delay excess
distributions for scattered HAs in BT are laid underneath
TM curves in grey dots.
It can be observed that initially maximal delays of the tree
morphing scheme tend to remain below the average of
permanent BT packet retardation. Convergence of the TM
then will lead to (relatively) undelayed packet delivery,
which is never met in BT. Little dependence on network size
becomes visible for TM — relative delays more strongly
change with topologic characteristics. In a densely meshed
provider network such as the ATT core, packet transitions
are rapid and therefore initial delays from tree elongation
account more dominantly for our relative measure. In the
contrary it is interesting to note that delays from BT admit a
systematic dependence on network size: BT average delay
excess increases from 45 % in the small ATT network to
about 120 % at sample size 15.400. From these observations
it can be concluded that bi-directional tunneling attains
appropriate performance for small communities within a
densely meshed core network, but becomes infeasible in
large inter-provider domains. The tree morphing even in its
initially weakest phase exhibits fairly uniform performance,
no matter how large the underlying network is.
IV. CONCLUSION AND OUTLOOK
In this paper we presented a distributed communication
conferencing software and some of its applications. The
applications include an easy-to-use scheme for distributed
presenting, recording and streaming of multimedia content.
The video conferencing module is based on a H.264/MPEG-
4AVC software implementation. In addition to this we use
also some costumer tailored wavelet-based codecs.
Concordantly we presented an approach to solve the mobile
source problem in SSM routing. This novel scheme of
morphing a previous distribution tree into a new shortest
path tree operates based on common multicast routing
protocols with simple algorithmic extensions. After a
handover it allows for immediate data transmission and
strictly avoids tunneling. All procedures are robust and of
rapid convergence. First performance simulations indicate
an overall low initial delay of the tree morphing scheme,
outperforming the conventional bi-directional tunneling
approach.
In future work we will optimize the communication
system for more inhomogeneous networks and quantify
further characteristic measures of the scheme by
simulations.
REFERENCES
[1] ITU-T Recommendation H.323: “Infrastructure of audio-visual services – Systems and terminal equipment for audio-visual services:
Packet-based multimedia communications systems”, Draft Version 4, 2000.
[2] The daViKo homepage, http://www.daViKo.com, 2005.
[3] J. Ostermann, J. Bormans, P. List, D. Marpe, M. Narroschke, F. Pereira, T. Stockhammer, and T. Wedi : Video Coding with H.264 / AVC: Tools, Performance, and Complexity , IEEE Circuits and
Systems Magazine, vol. 4, no. 1, pp. 7-28, April 2004. [4] H. L. Cycon, M. Palkow, T. C. Schmidt, M. Wählisch, and D. Marpe:
“A Fast Wavelet-Based Video Codec and Its Application in an IP
Version 6-Ready Server less Videoconferencing System.”, International Journal of Wavelets, Multiresolution and Information Processing (IJWMP), vol. 2, no. 2, pp. 165-171, June 2004
[5] The Speex Project homepage, http://www.speex.org, 2005 [6] T. C. Schmidt, M. Wählisch, H. L. Cycon, and M. Palkow, “Global
serverless videoconferencing over IP”, Future Generation Computer
Systems, vol. 19, no. 2, pp. 219–227, February 2003. [7] D.B. Johnson, C. Perkins and J. Arkko: “Mobility Support in IPv6”,
IETF RFC 3775, June 2004.
[8] http://www.javvin.com/protocolRUDP.html [9] J. Rosenberg, H. Schulzrinne, et al, "SIP: Session Initiation Protocol",
IETF, RFC 3261, Juni 2002.
[10] A. Varga et al., “The OMNeT++ discrete event simulation system,” http://www.omnetpp.org, 2005.
[11] S. Bhattacharyya, “An Overview of Source-Specific Multicast
(SSM),” IETF, RFC 3569, July 2003. [12] S. E. Deering, “Host Extensions for IP Multicasting,” IETF, RFC
1112, Aug. 1989.
[13] B. Fenner, M. Handley, H. Holbrook, and I. Kouvelas, “Protocol Independent Multicast - Sparse Mode (PIM-SM): Protocol Specification (Revised),” IETF, Internet Draft – work in progress 11, October 2004.
[14] T. C. Schmidt and M. Wählisch, “A First Performance Analysis of the Tree Morphing Approach to Source Mobility in Source Specific
Multicast Routing”, in Proceedings of the IEEE ICN'06, IEEE Press,
2006. [15] O. Heckmann, M. Piringer, J. Schmitt, and R. Steinmetz,“On Realistic
Network Topologies for Simulation,” in MoMeTools ’03:
Proceedings of the ACM SIGCOMM workshop on Models, methods and tools for reproducible network research. New York, NY, USA: ACM Press, August 2003, pp. 28–32.
[16] J. Rosenberg, J. Weinberger, C. Huitema, R. Mahy: “STUN - Simple Traversal of User Datagram Protocol (UDP) Through Network Address Translators (NATs)”, IETF, RFC 3489, March 2003.
[17] T. C. Schmidt and M. Wählisch, “Multicast Mobility in MIPv6: Problem Statement”, Internet Draft – work in progress, October 2005.
[18] I. Romdhani, M. Kellil, H.-Y. Lach, A. Bouabdallah, and H.
Bettahar, “IP Mobile Multicast: Challenges and Solutions,” IEEE Comm. Surveys & Tutorials, vol. 6, no. 1, pp. 18–41, 2004.
[19] “SCAN project: Internet Maps. SCAN+Lucent
map,”http://www.isi.edu/scan/mercator/maps.html, 2005. [20] R. Govindan and H. Tangmunarunkit, “Heuristics for internet map
discovery,” in Proceedings IEEE INFOCOM 2000, vol. 3. Tel Aviv,
Israel: IEEE Computer Society, March 2000, pp. 1371–1380.
Hans L. Cycon is currently teaching mathematics and signal processing at
FHTW Berlin, University of Applied Sciences. He received his diploma in physics in 1975 and his PhD in mathematics in 1979 and his habilitation in 1984 from the Technical University Berlin, Germany. His publications
fields are mathematical physics and signal processing i.e. image coding. Hans L. Cycon is leading several projects in developing wavelet based still image and video compression codecs. He is member of the German
delegation of the ITU/ISO standardization committee for JPEG 2000 still image standard.
Thomas C. Schmidt is teacher of Information Engineering at the HAW Hamburg and project manager at FHTW Berlin, where he was head of the computer centre for many years.
He studied mathematics and physics at Freie Universität Berlin and University of Maryland, USA. In 1993 he received his PhD in mathematical physics for a work in many particle theory of quantum
mechanics done at the theory group of the Hahn-Meitner-Institut Berlin. Since the late 1980s he has been involved in many computing projects, especially focusing on simulation and parallel programming, distributed
information systems and visualisation. His current fields of interest lie in the areas of mobile and multimedia networking and hypermedia information processing, where he has continuously conducted numerous
projects on national and international level.
Matthias Wählisch is a member of the networking group of the computer
centre of FHTW Berlin. He is studying mathematics and computer science at Freie Universität Berlin. His major fields of interest lie in networking protocols, where he looks back on seven years of professional experience in
project work and publication.
Mark Palkow presently is the Managing Director and Chief Developer at
the daViKo Gesellschaft für digitale audiovisuelle Kommunikation mbH that he founded in 2000. He received his diploma in communication engineering from the Fachhochschule Telekom Berlin in 1996. Since then
he has worked on several research projects at FHTW Berlin, the Old Dominion University Norfolk and the Heinrich Hertz Institut Berlin.
Henrik Regensburg is a member of the developer group of the “competence center media and networks” of FHTW Berlin. His major fields of interest lie in distributed video applications and coding,
networking and a/v-content authoring. He received his diploma in applied computer science from the University of Applied Sciences FHTW-Berlin in 2002. Since then he has worked on research projects at FHTW Berlin and
several freelance projects in commerce, all concerned with video conference technology and e-learning.