You are on page 1of 6

EVALUATION OF ACHIEVED VIDEO QUALITY IN WIRELESS

MULTIMEDIA TRANSMISSION USING UDP-LITE

Janne Vehkaperä, Johannes Peltola, Jyrki Huusko, Mikko Myllyniemi, and Mikko Majanen
VTT Technical Research Centre of Finland
Kaitoväylä 1, Oulu, P.O. Box 1100, FIN-90571
Finland
firstname.lastname@vtt.fi

ABSTRACT allowing corrupted data to be passed to higher layers and


utilizing the available bandwidth more efficiently.
The purpose of this paper is to study the performance of User Datagram Protocol (UDP) [4] is a best-effort
the partial checksum mechanism of UDP-Lite transport transport protocol and it has become very popular in
protocol compared to the standard UDP protocol in delay-sensitive multimedia applications due to its
wireless environment in terms of achieved video quality. simplicity. However, UDP does not pass corrupted
Video quality is evaluated in PSNR using simulation packets; it simply discards them. UDP-Lite [2] is a
results achieved with a simulation platform developed in modified version of the UDP transport protocol
the framework of IST PHOENIX project. The platform introduced by IETF and it has been developed to allow
offers the necessary components for modeling all relevant delivery of corrupted packets to higher layers by
ISO/OSI layers for wireless IP video streaming, including introducing more lightweight checksum calculation.
partial checksum functionality on transport and data link Suitability of UDP-Lite for multimedia transmission has
layers. The achieved gain in video quality in terms of been studied in several papers [1], [3], [6] and [10].
PSNR for IEEE 802.11a simulation model validates that However, in-depth analysis of the effect of UDP-Lite to
the video streaming benefits of using UDP-Lite transport the achieved video quality compared to UDP has not been
protocol instead of using regular UDP in wireless made. Studies have been concentrated on better
channels. The enhancement was 2 – 3 dB in PSNR for throughput provided by UDP-Lite and the achieved video
AWGN channel and for fading channels the difference quality is not deeply analyzed using common video
was smaller, being 1dB or less. The major conclusion of quality metrics.
our study is that the best improvement in video quality In this paper, we study the performance of the partial
using UDP-Lite is achieved in conditions where bit errors checksum mechanism proposed by UDP-Lite compared to
appear in very short bursts. Also a minor enhancement the standard UDP protocol in terms of achieved video
using UDP-Lite is achieved in transmission conditions, quality. To compare the performance of these protocols,
where signal-to-noise ratio in the channel is very low. we have made several simulations utilizing a software
platform developed in the framework of the IST
PHOENIX project (www.ist-phoenix.org) [5]. We
KEY WORDS evaluate the performance of UDP-Lite using different
Wireless multimedia, transport layer, UDP-Lite, video wireless channel models namely additive white Gaussian
quality, and MPEG-4. noise (AWGN), fast fading, and frequency non-selective
block fading channel models.
This paper is organized as follows. Section 2 introduces
1. Introduction the system architecture for the optimized wireless
multimedia transmission and the proposed solutions for
The new generation of wireless networks enable more different layers. Section 3 introduces the developed
efficient transmission of multimedia content to mobile simulation model and Section 4 presents simulation
terminals and video communication is becoming more results achieved utilizing the simulation model. Section 5
and more popular. However, the transmission over summarizes some key conclusions of this paper.
wireless links stays challenging due to varying
transmission conditions and high transmission error rates.
Novel multimedia applications can handle corrupted 2. System Architecture and Protocol Stack
packets and they often even prefer corrupted packets
instead of lost packets. However, erroneous data is The Figure 1 represents the protocol architecture for user
usually not delivered to the application because it is datagram protocol (UDP) and UDP-Lite performance
discarded by checksum process performed at lower layers analysis. The system consist of transmitter (source coder)
including data link and transport layer checksum and receiver (source decoder) units. Data streams are
mechanisms. Sending high bit rate and delay-sensitive packetized with Real-Time Transport Protocol (RTP) and
multimedia stream through wireless networks requires the data stream is delivered through IPv6 network. The
support from the application, transport and link layers for model consist the wired IPv6 network. The last hop in the
system is considered wireless. The Section 3 of this paper but on the other hand, it allows errors in less important
describes the wireless channel and modelled protocols in parts of the packet. It has been shown that the number of
more detail. discarded packets is significantly reduced when using
UDP-Lite instead of UDP [3]. This is something that error
resilient video decoder may utilise for better video quality
after error concealment tools.
The partial checksum is achieved by changing the UDP’s
Length field to Coverage field – it specifies the coverage
of the checksum and it is defined by the sending
application on a per-packet basis. This change is possible
because some redundancy is included in headers (UDP
length field = IP length field – size of IP header). By
setting Coverage value equal to packet length turns UDP-
Lite into traditional UDP. The UDP-Lite header together
with IPv6 pseudo header is depicted in the Figure 2.

Figure 1. Protocol architecture for multimedia


transmission

2.1 Transport layer

The function of the transport layer protocols is to provide


end-to-end communication between two (or more) hosts.
For multimedia streaming in IP networks, current de-facto
standard is to use RTP together with UDP.
RTP [7] provides (among other things) timestamps and
sequence numbers to packets. Sequence numbers can be
used for detecting lost packets or in sequencing of the out- Figure 2. IPv6 pseudo header and UDP-Lite header
of-order packets. Timestamps are used to maintain
synchronisation with the media time and the play back 2.2 Data Link Layer
time. RTP also provides payload type identification. The
companion control protocol, RTCP, monitors the quality In order to utilize UDP-Lite effectively in multimedia
of service (among other things). It may, for example, delivery, the data link layer should also support the
inform the sender to slow down the transmission rate due transmission of corrupted data. Two different solutions
to the detected increase in packet loss rate. can be applied for transmitting the corrupted data to upper
UDP [4] provides connectionless, best-effort, maybe- layers at medium access control (MAC) as described in
duplicates, unordered transport service. It is quite a simple [1]. The solutions are; to ignore the checksum at MAC
protocol and has a low overhead. It adds demultiplexing level, or to provide the partial checksum for MAC
and (optional) data integrity service (checksum) to IP. transmission frames.
Using UDP checksum is optional with IPv4 but Ignoring totally the checksum mechanism for
mandatory with IPv6 because IPv6 does not have any transmission frame is inadvisable since it can lead to the
checksum mechanism of its own. Also, disabling UDP situation, where the system crashes due to the erroneous
checksum in IPv4 is not reasonable since damaged packet header information and faulty protocol
transport and higher layer headers might pass unverified interpretation. Thus, partial checksum mechanism is
and the packets might be delivered incorrectly. UDP has preferred also for MAC level.
neither error reporting nor error recovery mechanism: We have defined a novel partial checksum solution for the
damaged packets are simply discarded. carrier sense medium access with collision avoidance
UDP-Lite [2] introduces the partial checksum to UDP. If (CSMA/CA) type of medium access, which is used in our
errors are detected in the sensitive part (e.g. in the header) simulation model. Our mechanism differs slightly from
of the packet, it will be discarded. If errors are in the some previously published partial checksum mechanism
insensitive part of the packet (e.g. in the payload), the such as one described in [8].
packet is not discarded. The sensitive part of the packet The MAC frame format for multimedia transmission is
always includes at least the IP pseudo header and the illustrated in Figure 3. The proposed solution enhances
UDP header. Additionally, the sensitive part can be the IEEE 802.11 standard in multimedia delivery. In the
extended to cover also parts of the payload, e.g. RTP IEEE 802.11a/b/g based systems, the multimedia data
header. Thus, UDP-Lite allows the data integrity checking frame is identified with frame control field, and the
in vital parts of the packets, thus prohibiting all kind of mechanism is backward compatible with original
misbehaviours due to the corrupted header information,
standard. Our mechanism is based on fast byte-by-byte • IPv6 network protocol,
cyclic redundancy check (CRC) operations instead of bit- • IPv6 network and mobility modeling,
by-bit hardware operations, which are utilized for • Robust Header Compression (RoHC),
example in majority of WLAN hardware solutions. The • Enhanced MAC protocol,
benefits of fast byte-by-byte software solution are • Physical layer controller,
represented e.g. in [9] and with state-of-the-art hardware • Channel en/decoder (convolution, RCPC, and
solutions the proposed partial checksum mechanism is low density parity check (LDPC) codes),
feasible. Including the easy CRC calculation and per • Interleaver and modulator (OFDM, TCM,
frame checking routines, our MAC frame structure TTCM, STTC),
benefits from the dynamic CRC coverage area for
• Radio channel model (AWGN, fast and slow
different data frames and streams.
fading channels).
Data lines:

Payload
Source Cross layer communication Source
decoding Block specific control decoding

Streaming Streaming
protocol protocol
e.g. RTP & End to end QoS e.g. RTP &
RTCP RTCP

Figure 3. MAC frame format for multimedia data Transport Transport


with partial checksum support protocol
e.g. UDPLite
protocol
e.g. UDPLite

IP
IP IP IP
network

3. Simulation Model Data link control Data link control


and Medium and Medium
Access Control Access Control

The simulation framework in this paper utilizes a


Channel
simulation platform developed in EU-IST funded decoding and
demodulation
Channel
decoding and
demodulation
PHOENIX project. The primary purpose of this
simulation tool is to allow development of different cross-
layer optimization techniques for wireless video
Sending IP
streaming communication systems. The simulation terminal Radio link transmitter Receiving wireless terminal

platform contains different models for all system layers Figure 4. System model of the simulation environment
from the physical layer up to the application layer present
in IP based video communication networks. Accurate The most important blocks affecting into the simulations
modeling of the physical layer functionality and channel and parameters used within each block in this study are
as well as seamless integration of the application layer introduced in more detail in the following chapters.
tools such as controllers and video codecs distinct this
simulation platform from traditional network simulation 3.3 Source codec
tools such as Opnet or ns-2. Modeling of the physical
layer allows obtaining realistic simulation results for We have used MPEG-4 simple profile video coding
techniques, such as UDP-Lite where generated error standard for video compression. MPEG-4 standard
patterns in the channel have strong effect on the includes several techniques, such as resynchronization
transmission performance. markers, data partitioning, reversible variable length
As stated earlier, all of the system layers have been codes (RVLC), and header extension codes (HEC) which
implemented into the simulation platform in order to improve both packet and bit error resilience of the
provide a realistic evaluation of the system performance compressed video. The decoder includes additional error
[5]. System model for the simulation environment is concealment methods such as spatial error concealment,
illustrated in Figure 4. Following system layers are where correctly received image areas are used to predict
included in the platform and can be used in the lost image areas to further improve perceived video
simulations: quality.

• Application layer controller, 3.4 Transport layer


• Source codec (MPEG-4,H.264, and H.264/SVC),
• Ciphering, The bit stream produced by the source encoder is further
• Unequal error protection (UEP) at application packetized using RTP protocol, which adds timestamps
layer using rate compatible punctured and sequence numbers into the packet in order to improve
convolution (RCPC) coding, the transmission above best-effort transport protocols.
• RTP packetization and RTCP signaling, After RTP packetization, UDP or UDP-Lite protocol is
• Transport layer (UDP/UDP-Lite/Datagram used to achieve low-delay transmission of video content.
congestion control protocol (DCCP), In the UDP-Lite protocol implementation, a checksum
calculation can be defined to cover IP pseudo header and
UDP-Lite header or to cover also a defined portion of the 4. Simulation Results
payload in addition to these headers. In UDP-Lite
simulations, the checksum is calculated over pseudo IP, In order to evaluate the possible enhancement in video
UDP-Lite and RTP headers. In UDP simulations, the quality achieved using UDP-Lite instead of UDP, we
checksum is calculated using the whole packet (headers have simulated the transmission of MPEG-4 video over
and payload). wireless channel using introduced simulation platform.
The scenario considered is a video transmission using
3.5 IPv6 network IEEE 802.11a WLAN radio. The maximum bit rate is 12
Mb/s from which 1 Mb/s is considered for simulated
The simulation platform includes a simulated wired IPv6 transmission rate since the capacity of the channel is
network where topology of the network and delay shared among several users. The transmitter and the
introduced by the network can be defined. In our receiver are considered to be inside a building in an office
simulations, the effect of wired network is minimized in environment and the transmitter is considered to stay in
order to test the performance of the UDP-Lite protocol in one position and the receiver is assumed to move slowly
the last wireless hop. This means that the delays in a floor area. We have considered three different types
introduced by the wired IPv6 network are short and no of channel models for our simulations: AWGN, fast
packet losses occur in wired network. fading, and frequency non-selective block fading
channels. Fading channel models with average SNR of 6
3.6 Data link layer dB are illustrated in Figure 5. QPSK modulation and
RCPC channel coding are used in every simulation. For
An enhanced IEEE 802.11 MAC layer is assumed, since each channel type we have calculated the packet loss ratio
no fragmentation or packing is considered for (PLR) and PSNR video quality metric for different
transmission frames. Partial CRC is performed for each channel SNR conditions.
data frame using the mechanism defined in Section 2.
Automatic retransmissions are not considered since only a Fast fading channel model (mean SNR=6 dB)
small delay introduced by retransmission cycle for MAC 15

data frames is assumed. Simulations represent the worst


SNR [dB]

10
case scenario, where data frames are discarded after the
5
predefined retransmission cycle.
0
0 10 20 30 40 50 60
3.7 Radio channel Time [s]
Frequence non-selective block fading
channel model (mean SNR=6.3 dB)
The platform includes a full physical layer 20
implementation for generic radio interface including
SNR [dB]

10
channel models for AWGN, fast and slow fading
0
channels. Radio model includes different interleaving,
channel coding and modulation schemes. For this study -10
0 10 20 30 40 50 60
we have used simulation model for IEEE 802.11a using Time [s]
QPSK modulation. The actual modulation scheme should
Figure 5. Fast fading and frequency non-selective
not have great impact on PSNR video quality results, thus
channel (below) models
results should be also applicable for IEEE 802.11b
systems.
As a test video sequence, we have used CIF-resolution
Foreman sequence compressed using MPEG-4 at average
3.8 Cross layer signalling
bit rate 448 kbit/s. The frame rate of the video is 30 fps
and intra-refresh is made after every 15 frames. Error
The simulation platform has been designed to support
resilience tools are used and the video packet size was
cross-layer optimization research for wireless video
2500 bits which means in our simulations that one
streaming. This support is enabled by offering adaptive
predicted frame (P-frame) is typically divided into two
system components, cross-layer controllers and dedicated
separate packets. In the case of AWGN channel,
cross-layer communication control signals across
simulations have also been done with a video packet size
different layers. Adaptive system components indicate
of 1500 bits in order to study the effect of the packet size
that all components should be able to change their
to the video quality. In the following, we will present the
parameters or operation mode on demand. Cross layer
simulation results for each channels type.
controllers collect information about the status of system
layers and based on this information they adjust the
operation of different components. For this work the cross 4.1 Achieved quality in AWGN channel
layer functionality has been used to signal the coverage
area for UDP and data link layers checksum calculation. PSNR and PLR as a function of channel SNR are shown
for AWGN channel in the Figures 6 and 7. Distinct
difference between UDP-Lite and UDP can be noticed 4.2 Achieved quality in fast fading channel
when the packet size is 2500 bits. Effective PSNR
improvement for UDP-Lite is located at channel SNR Figure 8 presents simulation results for fast fading
range from 1dB to 3dB. In this region 2 - 3dB channel, where the sampling period for fast fading gain
improvement in PSNR was measured. If the channel was was 1 ms and Doppler frequency for time correlated
very bad (SNR smaller than 0dB) or very good (SNR Rayleigh fading was 5 Hz. These fading parameters
greater than 4), the difference between UDP and UDP- correspond roughly to walking inside the office area.
Lite saturates to zero. Same behaviour can be seen from
PLR results. In the effective range, PLR differ about 30 Peak signal-to-noise ratio in fast fading channel
30
percentage units, but outside of this region there is no
UDP-Lite
difference between UDP and UDP-Lite results. 25

PSNR [dB]
UDP
20
Peak signal-to-noise ratio in awgn channel 15

30 10
-4 -2 0 2 4 6 8
PSNR

channel snr [dB]


UDP-Lite
20 Packet loss rate in fast fading channel
UDP 1

UDP-Lite
10
0 2 4 6 UDP

PLR
channel snr [dB] 0.5

Packet loss rate in awgn channel


0.9
0
UDP-Lite -4 -2 0 2 4 6 8
0.6 channel snr [dB]
PLR

UDP

0.3 Figure 8. PSNR and PLR curves in fast fading channel


0
0 2 4 6 Achieved video quality in fast fading channel simulation
channel snr [dB] was moderate. The average PSNR was below 25 dB for
Figure 6. PSNR and PLR curves in AWGN channel both UDP and UDP-Lite for all measured channel SNR
for packet size of 2500 bits values. UDP-Lite gave better PSNR values, but the
difference between UDP and UDP-Lite was only about
Peak signal-to-noise ratio in awgn channel 1dB. The small difference in PSNR is partly explained by
with packet size 1500 bits the small difference also in PLR statistics. The results can
30 be explained by the fact that fading causes a long burst or
PSNR [dB]

20
errors which reach the headers as well as payload, thus
UDP-Lite
UDP
UDP-Lite discards most packets affected by error bursts.
10

0 1 2 3 4 4.3 Achieved quality in frequency non-selective block


channel snr [dB]
fading channel
Packet loss ratio in awgn channel
with packet size 1500 bits
1 Figure 9 illustrates simulation results for frequency non-
selective block fading channel with uncorrelated log-
PLR

0.5 UDP-Lite normal distributed slow fading, where is 16 dB and the


UDP
coherence time for slow fading gain is 2000 ms. The
0
0 1 2 3 4
results for this channel model are similar to results for the
channel snr [dB] fast fading channel in terms of differences in PSNR and
Figure 7. PSNR and PLR curves in AWGN channel PLR values between UDP and UDP-Lite transmission.
for packet size of 1500 bits The differences in PSNR are even slightly smaller than in
the case of fast fading channel being mostly under 1dB.
Figure 7 presents AWGN results when the packet size is
1500 bits and thus smaller than in Figure 6. It can be seen
that when a larger packet size is used, UDP-Lite reaps
more benefit compared to UDP. When using a smaller
packet size there are relatively more errors in header parts
than in the payload part and this causes smaller PLR
difference for smaller packets. In addition, errors in the
payload may cause loss for VLC synchronization which
effectively may have similar effect for small payload than
losing short packet altogether.
Peak signal-to-noise ratio in frequency if the amount of retransmitted data is not high and it does
non-selective block fading channel not affect on the real-time requirements.
35
PSNR [dB]

30 UDP-Lite
UDP
25

20
Acknowledgements
15
-2 0 2 4 6 8 10
This work was carried out in the PHOENIX project (FP6-
channel snr [dB]
Packet loss rate in frequency 2002-IST-1-001812), which was partially funded by the
non-selective block fading channel European Commission within the European Union Sixth
0.6
Framework Programme and Information Society
0.4 Technologies. The authors would like to thank the whole
PLR

0.2
UDP-Lite project consortium and the colleagues who have
UDP
participated in the development of the PHOENIX system.
0
-2 0 2 4 6 8 10
channel snr [dB]

Figure 9. PSNR and PLR curves in frequency non- References


selective block fading channel
[1] L-Å. Larzon, M. Degermark, and S. Pink, UDP Lite
for Real Time Multimedia Applications. Proc of the IEEE
5. Conclusions International Conference of Communications (ICC),
Vancouver, British Columbia, Canada, June 1999.
In this paper, we have evaluated the achieved [2] L-Å. Larzon, M. Degermark, S. Pink, L-E. Jonsson
enhancement in video quality in wireless transmission, (ed.), and G. Fairhurst (ed.), The Lightweight User
where corrupted packets are allowed to reach the video Datagram Protocol (UDP-Lite). RFC 3828, July 2004.
decoder instead of dropping packets in the network layers. [3] R. Reine and G. Fairhurst, MPEG-4 and UDP-lite for
We have used both UDP-Lite and modified MAC Multimedia Transmission. 4th Annual Post Graduate
protocols which introduce the partial checksum Networking Conference, PGNET 2003, Liverpool, UK,
calculation for packet payload in order to allow passing June 2003.
corrupted packets to error-resilient video decoder. To [4] J. Postel, User Datagram Protocol. RFC 768, August,
evaluate the achieved video quality in wireless 1980.
transmission using UDP-Lite, we introduced the [5] M.G. Martini, M. Mazzotti, C. Lamy-Bergot, P.
simulation platform which has been developed in IST- Amon, G. Panza, J. Huusko, J. Peltola, G. Jeney, G.
PHOENIX project. The simulation platform can be used Feher, and S.X. Ng, A Demonstration Platform for
for evaluating different models and algorithms in all Network Aware Joint Optimization of Wireless Video
system layers from the physical layer up to the application Transmission. Proceedings of 15th IST Mobile & Wireless
layer present in IP based video communication networks. Communication Summit, Myconos, Greece, June 4-8,
We have used this simulation environment to evaluate the 2006.
performance of the UDP-Lite protocol using three [6] S.A. Khayam, S. Karande, H. Radha, and D.
different wireless channel models. Loguinov, Performance analysis and modeling of errors
Simulations for the considered wireless channel models and losses over 802.11b LANs for high-bit-rate real-time
conclude that allowing corrupted packets to reach the multimedia, Signal Processing: Image Communication,
error-resilient video decoder will improve the achieved 18(7), 2003, 575-595.
video quality. The greatest improvement in video quality [7] H. Schulzrinne, S. Casner, R. Frederick, and V.
was achieved in AWGN channel where bit errors have Jacobson, RTP: A Transport Protocol for Real-Time
more random pattern. In fast fading channel and Applications. RFC 3550, July 2003.
especially in frequency non-selective block fading [8] E. Masala, M. Bottero, J.C. De Martin, MAC-Level
channel, the improvement in video quality was smaller. In Partial Checksum for H.264 Video Transmission over
frequency non-selective block fading channel the bit 802.11 Ad Hoc Wireless Networks, in Proc. of IEEE 61st
errors occur usually in bursts and this will reduce the Semiannual Vehicular Technology Conference (VTC),
quality of the channel to intolerable and significant Stockholm, Sweden, May-June 2005.
improvement in video quality using UDP-Lite instead of [9]. T. Ritter, The Great CRC Mystery, Dr. Dobb's
UDP is not achieved. According to our simulations, the Journal of Software Tools, vol 1, number 2, February,
key finding of our study is that the best improvement in 1986, pp. 26-34, 76-83.
video quality using UDP-Lite wireless transmission is [10] A. Singh, A. Konrad, and A.D. Joseph, Performance
achieved in transmission conditions where only short evaluation of UDP lite for cellular video. Proc. of the 11th
bursts of bit errors occur. In poor transmission conditions NOSSDAV, Port Jefferson, New York, USA, June 25-27,
only a minor enhancement in video quality using UDP- 2001.
Lite instead of UDP is achieved. A retransmission could
be more effective solution in poor transmission conditions

You might also like