You are on page 1of 8

V. Sgardoni et al.: Robust Video Broadcasting over 802.

11a/g in Time-Correlated Fading Channels 69

Robust Video Broadcasting over 802.11a/g


in Time-Correlated Fading Channels
Victoria Sgardoni, Mantalena Sarafianou, Pierre Ferré, Andrew Nix and David Bull

Abstract — In order to deliver video streams efficiently over the number of lost MAC frames at a receiver can be high
WiFi to many thousands of consumer handheld devices, broadcast (typically greater than 20% within a video frame). This
protocols must be employed. In this mode of operation the results in serious deterioration of the received video quality.
received video quality can deteriorate rapidly as a result of high Cross-layer MAC-PHY optimization for video
application layer packet loss which occurs because MAC frame
retransmission cannot be used. In this paper we develop a robust
transmission over WiFi has been explored by a number of
video delivery solution for broadcast transmission over 802.11a/g. authors, e.g. [2, 3, 4]. The wireless broadcast of IP video
Using a cross-layer WiFi simulator in combination with an to multiple users in heterogeneous environments presents
accurate time-correlated fading channel, the received video a number of new challenges. Most importantly, MAC and
quality is evaluated for broadcast H.264 video sequences. IP layer feedback from a given receiver or video decoder
Application layer cross-packet forward error correction is then is no longer possible. Each receiver experiences a unique
used together with error concealment at the video client.
Furthermore, the application of an external packet interleaver is
radio channel and user specific source adaptation in no
considered. Combining a block size of two hundred packets longer possible. Only a handful of papers address the
(which introduces a 4.8 second delay) and an application layer specific needs of broadcast video over wireless networks
FEC code rate of 0.75 our results demonstrate that video can be [5, 6, 7].
successfully broadcast over WiFi to many thousands of handheld In this paper we estimate the received broadcast video
terminals at large-scale spectator events.1 quality by simulating the transmission of a video sequence
Index Terms — WLAN IEEE 802.11a/g, broadcast, over the 802.11a/g MAC and PHY layers. A MAC-PHY
forward error correction, interleaving simulator is used to model the transmission of a time series
of queued MAC frames over the wireless channel [8].
I. INTRODUCTION MAC layer FLR and frame delay are evaluated. As
IEEE 802.11 is increasingly being used for multimedia discussed in [8], most 802.11a/g MAC and PHY studies use
distribution to mobile terminals. This paper investigates static channel models, where the PHY layer packet error
broadcast video streaming to thousands of handheld rate (PER) is independent of time. However, it is well
devices, such as mobile phones and Personal Digital known that packet errors over a wireless medium are bursty
Assistants (PDAs). The study was performed as part of the in nature [9]. Therefore, an accurate time correlated fading
VISUALISE project [1]. This work enables spectators at channel is used here to simulate the PHY layer PER. We
sporting events to use mobile WiFi devices to radically investigate a number of methods to enhance the quality of
enhance their visual experience. Typical events include broadcast video encoded using the H.264/AVC standard.
motor sports, such as the World Rally Championship These include H.264 error concealment and the use of cross
(WRC), and stadium based activities such as cricket or packet application layer Forward Error Correction (AP-
baseball. Live and recorded video streams are pushed to FEC). In the latter case an erasure block code is applied to a
spectator terminals together with scoreboard, game block of NALUs (Network Abstraction Layer Units, see
statistics, live timing and location tracking information. section IV) to provide an additional layer of video
Spectators have the opportunity to watch their favorite protection. We also investigate the effect of interleaving
players and/or teams from a range of remote cameras. packets within a large block using an external interleaver.
The 802.11a/g standard combines a Coded Orthogonal This approach allows shorter, and hence more
Frequency Division Multiplexing (COFDM) Physical layer computationally efficient, erasure codes to be applied. The
(PHY) with the legacy 802.11 Medium Access Control impact of the time varying wireless channel (determined by
(MAC) layer. When data is sent as a broadcast (or the Doppler spread) and its influence on packet loss and
multicast) stream, MAC layer retransmission cannot be video quality are also studied. Our simulation results are
used since clients no longer have access to a feedback supported by a number of experimental measurements and
channel. While broadcasting potentially allows thousands field trials.
of terminals to receive the video stream, the received MAC
frame loss rate (FLR) is often too high for successful video
II. OVERVIEW OF IEEE 802.11 a/g
decoding. If the radio frequency signal level is poor then
The IEEE 802.11 a/g MAC and PHY layers are
1
described in several publications, e.g. [8, 10], including the
All authors are with the Centre for Communications Research (CCR),
University of Bristol, UK. This work was supported in part by the UK
standard [11]. As such we limit ourselves to the key points
Technology Strategy Board. (Contact: Andy.Nix@bristol.ac.uk) relevant to this paper.

Manuscript received January 15, 2009 0098 3063/09/$20.00 © 2009 IEEE

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.
70 IEEE Transactions on Consumer Electronics, Vol. 55, No. 1, FEBRUARY 2009

Once a MAC frame has been produced for transmission, format (which is well suited to mobile phone displays and
in unicast mode the MAC layer at the transmit station processing capabilities). This was achieved using a
expects to receive an acknowledgement (ACK) following hardware video encoder2. The following results were
successful reception by the intended user. This process is generated for the ‘ArtCar’ video sequence, which was taken
described in [10]. When multicast and broadcast protocols from inside a rally car at the WRC’07 event is South
are used to simultaneously serve video packets to many Wales, UK. The video sequence consists of 410 frames
thousands of terminals, MAC layer ACK is no longer encoded at 12.5fps, with I-frames sent every 25th frame,
provided. Instead each PHY layer packet is sent one after one reference frame and error concealment based on
the other without retransmission. As a result, in the previous frame copy.
multicast and broadcast scenario each WiFi terminal can The H.264/AVC standard specifies two layers, the Video
experience a high PHY layer PER. This translates to a high Coding Layer (VCL) and the Network Abstraction Layer
NALU loss rate at the H.264 video decoder. (NAL). The NAL is the interface between the VCL layer
MAC data frames are mapped to Protocol Data Units and the underlying network layers. This helps to achieve
(PDU) packets for transmission over the PHY layer. The compatibility with many heterogeneous networks. The
PHY layer simulator described in [12] supports correlated coded video stream is divided into NAL units (NALU). In
time-varying channel gains for each tap in the channel the case of transmission over IEEE 802.11 systems, a
impulse response, as described in [8]. This PHY layer NALU is delivered to the transport layer in a packet based
model is used to evaluate the outcome of each PDU packet format, according to the Real-time Transport Protocol
transmission. A MAC frame is lost if an error is (RTP). Each NALU represents a packet of specific length,
encountered during the MAC layer frame check sum (FCS) containing a header and payload data. After H.264
process. encoding, a number of fixed length NALUs are created. In
our study we assume a NALU size of 750 bytes.
III. THE CHANNEL MODEL Error Control: Cross-packet application layer FEC (AP-
FEC) is a suitable additional error control strategy for
We use a time-varying channel model as described in [8] H.264 video transmission [17]. AP-FEC is commonly
to replicate the time correlated nature of the observed implemented using Reed Solomon (RS) or Raptor codes,
instantaneous signal power at the target station. The PER the latter being a type of fountain code that is recommended
for consecutive packets is not independent, due to the time- as part of the 3GPP broadcast specification [18] and
correlated characteristics of the mobile channel. This implemented in [7]. When AP-FEC is applied at the
implies that the probability of receiving a packet in error at receiver it is able to correct packet erasures according to the
the PHY is correlated in time. The channel model replicates error correcting capability of the code used, which depends
multipath fading as a function of terminal velocity, carrier on the code rate. Due to hardware limitations, AP-FEC was
frequency and Doppler spectrum. Here we use the classical not used in the WRC trial; however analysis is included in
Jakes Doppler Spectrum [15]. This allows us to model the the following simulations. The AP-FEC mechanism is
spaced-time autocorrelation of the fast fading envelope. based on a generic erasure code, with code rates of 0.75
The fading can be modeled as either a Rayleigh or Rician and 0.875, applied across 8 NALUs (i.e. a depth of 8), for
process. The severity of the Rician fading is controlled via equal frame protection. In our simulations we assume that a
the K-factor. The resulting spaced-time autocorrelation is fixed 256 kbps IP stream is generated, with or without AP-
imposed onto a set of i.i.d. Rayleigh fading samples using a FEC. This means the encoded video quality decreases (as a
Doppler filter [13, 14]. result of the lower video bit rate) when AP-FEC is added.
The instantaneous signal power is simulated at the However, as will be seen later, when received over a
receiver for the duration of the video stream. Hence, the broadcast WiFi channel the benefits of AP-FEC far
instantaneous signal to noise ratio (SNR) on a per packet outweigh the reduction in encoded video quality.
basis varies with time (and the time correlation depends on Alternatively, if fixed video encoder quality is desired, the
the Doppler spread). Given knowledge of the noise floor use of AP-FEC simply increases the IP transmission rate
and the average received signal power over the entire video (which is a problem for fixed rate cellular standards but not
sequence, the instantaneous SNR per packet is computed for variable rate wireless LANs).To further enhance the
for any desired average SNR. received video quality we also study the effect of NALU
interleaving, as correlated error bursts degrade the video
IV. VIDEO ENCODING AND ERROR CONTROL performance more significantly than randomly distributed
errors. This is because cross packet error correction and
H.264/AVC encoding: Our study of multicast WiFi video concealment fail for long sequences of lost packets.
received video quality is based on encoded video sequences In particular, we experiment with the interleaver block size
taken from the World Rally Championship 2007 (WRC). to explore the trade-off between improved performance and
The video was encoded using the H.264/AVC standard [16]
2
to produce a 256 kbps video stream in 320x240 pixel ProVu encoder from ProVision Communications

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.
V. Sgardoni et al.: Robust Video Broadcasting over 802.11a/g in Time-Correlated Fading Channels 71

T r a n s m itte r
NALUs FEC In t e r l e a v e d
p a ck ets .p a c k e t s

VLC /NAL
FEC I n te r le a v e r M AC
H .2 6 4 N PHY
E ncod er d ,r
V id e o fr a m es
C hann el

D e -F E C D e-
V id e o fr a m e s H .2 6 4 M AC PH Y
d ,r I n te r le a v e r
D ecoder N p a ck ets
NALUs R e c e iv e r
FEC
p a c k e ts M A C -P H Y
s im u la t o r

U ser
d e v ic e

Fig. 1 Block diagram of the cross-layer simulator with AP-FEC (depth d, rate r) and external interleaver (block size N)

fixed delay. As mentioned above, the use of higher layer quality of the received video sequence in terms of PSNR
interleaving introduces a fixed delay, and this is undesirable per frame and video FLR.
in live or interactive video applications. Since the data In the simulation results the average received SNR over
source rate is 256 kbps (32 kBytes/sec) and the packet size the wireless channel varies between 5 and 25dB. All of the
is 750 Bytes, the number of packets transmitted per second 802.11a/g link-speeds are simulated. Mobility in the
is 42.67. For an interleaver block size of N packets, the wireless channel was modeled for walking spectators. The
delay is given by N/42.67. It is obvious that a large block model is also capable of simulating results for vehicular
size introduces significant delay. The level of acceptable use. In order to study the video performance statistically, a
delay will vary depending on the application. number of video transmission simulations were performed.
Peak Signal to Noise Ratio (PSNR) is the statistical In particular, the following results were produced by
metric generally used to assess video quality. The PSNR of averaging the video PSNR per frame over seven separate
a picture frame is based on the Mean Square Error (MSE) simulations (each assuming an uncorrelated and
of the received frame, compared with a reference frame (i.e. independent radio channel).
that generated at the encoder). In order to investigate the effect of packet interleaving on
the received video quality, an external interleaver is
V. THE CROSS-LAYER SIMULATOR applied. The interleaved error pattern is then used to
An integrated 802.11a/g MAC-PHY simulator is used to generate the sequence of received video packets. Since both
model the MAC frame loss process. The simulator is the simulated and measured error bursts can last for several
described in detail in [8]. A time sequence of MAC frames hundred milliseconds, a large interleaver block size is
is passed into the simulator. Outputs include i) MAC layer required.
FLR, ii) MAC-to-MAC frame delay, and iii) throughput.
These are evaluated as a function of the channel’s average VI. EXPERIMENTAL SET UP
SNR, the Power Spectral Density (PSD) and the selected During the WRC trials the video IP stream was sent over
PHY layer link-speed. Importantly, the PER for contiguous an 802.11b/g wireless network. The broadcast video
packets is not independent and this has a significant effect transmission measurements presented here were taken using
on the performance of AP-FEC and video error outdoor terminals in a cricket stadium as shown in fig. 2.
concealment. To replicate the bursty nature of the packet
error process an accurate time-correlated channel model is
implemented based on the PSD of a typical radio channel.
The video transmission simulator (Fig. 1) can simulate
the transmission of an arbitrary H.264 video sequence from
the transmitter to the receiver through the MAC and PHY
2
layers of WiFi. The encoder translates each video frame
1
into a number of fixed length NALUs. For broadcast
transmission an RTP/BCT/IP stack is assumed. In the
absence of AP-FEC there is 1:1 correspondence of video
Fig. 2 View of routes followed at cricket field for measurements at
NALUs to IP packets and MAC frames. If AP-FEC is
Cricket Festival 2008
applied, the packets created with FEC have the same length
as the NALUs, and are then mapped 1:1 to IP packets and These measurements used a WiFi Access Point
MAC frames. The MAC-PHY simulator provides an error (connected to a desktop computer) and two WiFi cards
modeling tool that predicts the error/loss pattern for a (each connected to a notebook computer). The hardware
sequence of NALUs. This information is then used to was configured in a client/server arrangement. The server
create the sequence of video packets arriving at the was used to broadcast a video sequence over 802.11g on a
receiver. The video transmission simulator evaluates the selected link-speed. Different channel scenarios were

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.
72 IEEE Transactions on Consumer Electronics, Vol. 55, No. 1, FEBRUARY 2009

evaluated by walking along three different routes (up to 10m, different channel realizations to obtain an average PSNR as
40m and 100m from the Access Point respectively). The a function of mean SNR. This was performed for all link-
measurements were taken using the logging software speeds and for Doppler shifts of 4Hz and 65Hz.
developed in [19]. The following data was recorded: i) the PSNR for video1, SNR=15dB, link speed=1, Doppler 4Hz
received packet sequence number, ii) the received signal 48

46
strength (RSSI), iii) the PER and iv) the link-speed. The
44
experimental data was processed to supply measured packet 42
error sequences (patterns) to the video transmission 40

PSNR (dB)
simulator, as described in section V. The video transmission 38

simulator is then used to evaluate the received video quality 36

in terms of PSNR and NAL unit Loss Rate (NLR). 34

32
error-free PSNR at TX

VII. ANALYSIS OF RESULTS 30 mean PSNR at RX no FEC


mean PSNR at RX FEC=0.875
28
A. Simulation Results 0 100 200
Frame
300 400 500

(a)
The video transmission simulator evaluates the PSNR
NAL loss rate over time - mode 1 packet= 750 Doppler=4Hz SNR= 15dB
per received video frame for different link-speeds, mean 0.7
NAL loos
channel SNR values and Doppler spreads. The following 0.6 mean

results were generated for the ‘ArtCar’ video sequence, no FEC


0.5
which consists of 410 frames at 12.5fps.
Fig. 3a shows the PSNR per frame. The video is 0.4

NLR
broadcast over a WiFi link using link-speed 1 (BPSK, ½ 0.3

rate). The mean SNR at the receiver was 15dB and the 0.2
maximum Doppler shift was 4Hz. The blue plot (no
marker) shows the error-free PSNR per frame, as computed
0.1

at the video encoder. This is used as an upper bound of 0


0 50 100 150 200 250 300 350 400 450

video quality at the receiver. The mean received PSNR for time slot
(b)
video encoded without AP-FEC is shown using the red plot Fig. 3 a) PSNR per frame with/without FEC, b) NLR per frame
with the ‘x’ marker. The green plot with the circle marker
average PSNR vs. SNR - all link speeds - Doppler=4Hz
shows the received PSNR for video protected with a code 50
BPSK 1/2
rate 0.875 AP-FEC. Clearly the use of cross packet AP- 45 QPSK 1/2
FEC improves the mean received PSNR since a number of QPSK 3/4
average PSNR [dB]

40 16QAM 1/2
lost packets can be recovered prior to video decoding. With 16QAM 3/4
cross packet AP-FEC the received PSNR per frame can be 35 64QAM 3/4

seen to improve (approaching the error free value). It 30


should be noted that this test is extremely severe since the 25
average SNR is 15dB and in Rayleigh fading channels the
20
deep fades cause severe error bursts. The simulator outputs 5 10 15 20 25
the number of lost MAC frames (and hence NALUs) per SNR [dB]
(a)
video frame at the receiver prior to the application of cross average PSNR vs. SNR - all link speeds - Doppler=65Hz
packet AP-FEC. 50
BPSK 1/2
Fig. 3b shows the NLR over the time corresponding to 45 QPSK 1/2
QPSK 3/4
the duration of the video sequence (410 frames). Each NLR 16QAM 1/2
average PSNR [dB]

40
16QAM 3/4
value is averaged over a time window of 132ms. The bursty 64QAM 3/4

nature of the NLR is clearly visible. Figs 3a and b are time- 35

aligned to allow a direct comparison between the NLR and 30

the video PSNR. Regions of high NLR (typically greater 25


than 20%) result in low video PSNR. Errors in a P-frame
without AP-FEC propagate into the following frames, thus 20
5 10 15 20 25
SNR [dB]
decreasing the PSNR even at times when the NLR is low. (b)
With cross packet AP-FEC the received PSNR per frame Fig. 4a, b. Average PSNR per video sequence versus average channel
can be seen to improve (approaching the error free value). SNR for all link-speeds (no FEC). Top: max Doppler 4Hz, Bottom:
To compare video quality at each of the 802.11a/g link- max Doppler 65Hz

speeds the average PSNR of the entire video sequence can Figs. 4a and 4b show the results in the case without AP-
be computed as the sum of the PSNR per frame over the FEC. When the mean received SNR exceeds 15dB we
entire sequence. We compute the average PSNR over seven observe that link-speeds 1-4 (BPSK and QPSK) achieve an

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.
V. Sgardoni et al.: Robust Video Broadcasting over 802.11a/g in Time-Correlated Fading Channels 73

average PSNR ≥30dB for the 65Hz channel. For the 4Hz the cross marker shows the mean PSNR without any AP-FEC,
channel only link-speeds 1-2 (BPSK) meet this target. the green line with the star marker shows the mean PSNR with
Furthermore, at the higher Doppler shifts link-speed 1 0.75 rate AP-FEC and the yellow line with the circle marker
achieves a higher average PSNR for SNR ≤15dB. At high shows the mean PSNR with 0.875 rate AP-FEC.
Doppler values the error burst length is significantly TABLE I
reduced, thus making lost NALUs easier to conceal. Delta average PSNR values from Figs. 5a, b
In Fig. 5a and 5b we show the difference in the average
SNR=15dB SNR=20dB
PSNR (delta PSNR) between the error-free video encoded
BPSK 1/2 5.471 1.517
at 256 kbps and the received video, as a function of mean No
QPSK 1/2 8.521 3.559
SNR for all link-speeds. These results, and all subsequent FEC
QPSK 3/4 12.37 5.93
results, focus on the pedestrian case with a maximum
Doppler shift of 4Hz. Fig. 5a shows the delta PSNR of the BPSK 1/2 1.897 0.544 AP-
QPSK 1/2 4.369 1.256 FEC
received video when no cross packet AP-FEC is applied,
QPSK 3/4 12.61 2.229 0.875
whereas fig. 5b shows the delta PSNR when AP-FEC with
a code rate of 0.875 is used. We observe that when AP-
FEC is applied in this slow fading channel the PSNR It can be seen that when the 0.75 rate AP-FEC is applied
difference drops by approximately 2-4dB, resulting in the video quality drops at some locations compared to the
significant video quality improvement for link-speeds 1-4 higher rate. This observation is also supported by our
(BPSK, QPSK and 16QAM). Table I shows specific values experimental results.
of delta average PSNR for link-speeds 1-3 and mean SNR 48
PSNR for video1, SNR=15dB, link speed 3 Doppler 4Hz
error-free PSNR at TX
values of 15dB and 20dB for the cases without and with 46
mean PSNR at RX FEC=0.75
AP-FEC (code rate = 0.875). To achieve near perfect video mean PSNR at RX FEC=0.875
mean simulated PSNR at RX no FEC
44
quality the received SNR needs to exceed 20dB.
42
d elta ave r ag e P S NR d iff e r en c e fr o m r e f. vs. S NR - D opp le r = 4 Hz
30 40
PSNR (dB)
B PS K 1 /2
Q PS K 1 /2 38
25
average PSNR difference [dB]

Q PS K 3 /4
1 6QA M 1 /2 36
20 1 6QA M 3 /4
6 4QA M 3 /4 34
15 32

10 30

28
5 0 100 200 300 400 500
no FE C Frame
0
5 10 15 20 25 Fig. 6 PSNR per frame with/without FEC, max Doppler 4Hz, mean
S NR [dB ] SNR 15dB, link-speed 3.
(a)
delta averag e P SNR diff eren c e from ref. vs. SNR - D opp ler= 4 Hz
30
The degradation with lower code rate AP-FEC only occurs
B PS K 1/2
Q PS K 1/ 2
when residual errors remain after the decoding process.
25 Q PS K 3/ 4 Ultimately the PSNR depends on the location of the missing
average PSNR difference [dB]

1 6QA M 1 /2
20 1 6QA M 3 /4 NALUs and how they propagate over time. It is possible for a
6 4QA M 3 /4
lower PER to result in a higher PSNR. Furthermore, to
15
maintain a constant transmission rate, when the code rate is
10 reduced, we also have to lower the encoded video rate.
5
Clearly, based on the rate-distortion curve, this reduces the
c rossFEC = 0.87 5 encoded PSNR. Low AP-FEC code rates should only be used
0
5 10 15 20 25 over channels where a high PER is expected. With a low
S NR [dB ]
PER the received video quality with a low rate code can be
(b)
worse than a high code rate. The same situation is also seen
Fig. 5a, b Delta average PSNR vs SNR for all link-speeds, max
Doppler 4Hz a) no FEC, b) FEC with code rate 0.875 in our later experimental results and observed in [7].
Finally, Fig. 7 clearly demonstrates a case where the
Simulations have shown that improvement in video quality mean PSNR approaches the error-free PSNR when AP-
is not guaranteed in severe channel conditions when AP-FEC FEC is applied. The video is broadcast over a WiFi link
is applied. Fig. 6 shows the PSNR per frame for the ‘ArtCar’ using link-speed 3 (QPSK, ½ rate), the mean SNR at the
video sequence broadcast over WiFi using link-speed 3 receiver is 25dB and the maximum Doppler shift is 4Hz.
(QPSK, ½ rate). The mean SNR at the receiver was just 15dB The received video quality improves with AP-FEC,
and the maximum Doppler shift was 4Hz. The blue plot with reaching its best for code rate 0.75, where the PSNR is
the dotted marker shows the error-free PSNR, the red line with improved by approximately 3-4dB.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.
74 IEEE Transactions on Consumer Electronics, Vol. 55, No. 1, FEBRUARY 2009

48
PSNR for video1, SNR=25dB link speed 3 Doppler 4Hz
varying channel has a relatively large coherence time
(hundreds of milliseconds). Therefore, if a large interleaver
46
block size is used to match the channel coherence time, the
44
resulting decoded PER will improve significantly since the
42 bursty PER prior to decoding will be smoothed. When
PSNR (dB)

40 determining the interleaver block size another key


38
consideration is the delay it introduces. If the interleaver
error-free PSNR at TX block size is set to 200 packets (each of 750 bytes) then the
36
mean PSNR at RX no FEC
mean PSNR at RX FEC=0.75
video delay is approximately 4.8 seconds (using the
34
equation from section IV). Assuming that a 5 second delay
32
0 100 200 300 400 500
is acceptable and that a suitable buffer can be implemented
Frame at the receiver, we now investigate the received video
Fig. 7 PSNR per frame with/without FEC, max Doppler 4Hz, mean
quality improvement based on a block size of 200 packets.
SNR 25dB, link-speed 3
PSNR for video 1 at cricket pitch - li nk speed 3 a nd route 2
50
B. Experimental Results
Our simulation results are supported by experimental 45
measurements taken during VISUALISE trials at a cricket
40
stadium. Fig. 8 shows the SNR at the mobile WiFi device

PSNR (dB)
as the user moves along ‘route 2’ using link-speed 3 35
(QPSK, ½ rate). The SNR was calculated from the received
RSSI assuming a receiver noise floor of -93dBm. 30

50
SNR at receiver - mode 3 and route 2
25 erro r fre e PS NR at TX
mea n PS NR at Rx FE C=0 .75
45 mea n PS NR at Rx no FE C
20
40 0 100 2 00 300 400 5 00
F rame
35
SNR (dB)

30
Fig. 9 PSNR per frame with/without FEC based on measured PER at
cricket match
25
Packet Error Rate over packet window sizes - link speed 3 & route 2
20 35
window size 500
15 window size 200
30
window size 50
10
0 20 40 60 80 100 25
time of packet reception (sec)

Fig. 8 Measured channel SNR during video reception at cricket match 20


PER (%)

Fig. 9 shows the PSNR per frame for the ‘ArtCar’ video 15

sequence broadcast over a WiFi link to a mobile receiver as


10
it moves along ‘route 2’ using link-speed 3. This data
corresponds (and is time-aligned) to the SNR data shown in 5

Fig. 8. In fig. 9 the blue plot (with no marker) represents 0


0 500 1000 1500 2000 2500
the error-free PSNR, the red line with the cross marker the number of packets

mean PSNR without AP-FEC and the green line with the
Fig. 10 Measured PER averaged over a window of 50 packets
star marker the mean PSNR with 0.75 rate AP-FEC.
Fig. 10 depicts the mean PER over time. This was Fig. 11 shows the packet error pattern at different stages of
evaluated using different packet interleaver block sizes, as the video transmission system: the top figure shows the error
shown here for 50, 200 and 500 packets. It can be observed pattern without AP-FEC or interleaving, the upper middle
that for small block sizes the troughs are deeper and the figure shows the error pattern after AP-FEC decoding (0.875
peaks are higher, whereas for larger block sizes the PER is code rate), the lower middle figure shows the error pattern
much smoother over time. This figure demonstrates that a after the combined application of interleaving (with a 200
large packet interleaver lowers the peak PER and reduces packet block size) and AP-FEC (0.875 code rate) and the
the bursty nature of the packet loss mechanism. bottom figure shows the error pattern after the combined
To improve the received video quality over 802.11a/g we application of interleaving (with a 200 packet block size) and
now study the impact of varying the interleaver block size, AP-FEC at a code rate of 0.75. In the graphs a value of zero
assuming the AP-FEC spans a depth of 8 NALUs. Given represents a correctly received packet while a value of one
that the measurement data was captured using a slow represents a missing packet. It can be seen that the addition of
moving (walking pace) receiver, we observe that the slow interleaving improves the percentage of correctable missing

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.
V. Sgardoni et al.: Robust Video Broadcasting over 802.11a/g in Time-Correlated Fading Channels 75

packets compared to the case where only AP-FEC with a transmission simulator. The mean PER achieved for each code
depth of 8 NALUs is applied. The process of interleaving rate and block code size, based on the measurements taken for
rearranges the missing packet sequence over a large window, link-speed 3 and route 2, is shown in Fig. 14. It is observed that
thus randomizing the burst errors and smoothing out the for higher code rates the PER increases. For larger block sizes
PER. In cases where the interleaver block size is longer than lower PER is seen, especially for higher code rates. For the
that AP-FEC depth this combination improves the AP-FEC channel encountered in our experiment, a block size of between
decoder performance. The lower code rate of 0.75 further 200-500 packets reduces the mean PER to less than 1% with
reduces the PER. In harsh channel conditions (such as those AP-FEC rates of between 0.70 and 0.75.
shown here) there is still a residual missing packet rate, and
video quality is maximized in this case via error concealment 50
PSNR for video 1 at cricket field - link speed 3 and route=2

in the video decoder.


error patterns for link speed 3 and route 2 - FEC 0.875 - block 200 45
1
40

PSNR (dB)
0 35
0 500 1000 1500 2000 2500 3000
1
30
error free PSNR at TX
25 mean PSNR at Rx FEC=0.875
0 mean PSNR at Rx no FEC
0 500 1000 1500 2000 2500 3000
mean PSNR with interleaver +FEC
1 20
0 100 200 300 400 500
Frame
Fig. 12 PSNR per frame with/without FEC and interleaving as
0
0 500 1000 1500 2000 2500 3000 measured at cricket match for link-speed 3 and route 2
number of packets
PSNR for video 1 at cricket field - link speed 5 and route 1
FEC 0.75-block 200 50
1
45

40

0 35
0 500 1000 1500 2000 2500 3000
number of packets 30
PSNR (dB)

25
Fig. 11 Packet error pattern - top: without FEC, upper middle: with
20
FEC at 0.875, lower middle: with interleaving (200 block size) and
FEC at 0.875, bottom with interleaving (200 block size) and FEC 0.75. 15
error free PSNR at TX
10 mean PSNR at Rx FEC=0.75
Fig. 12 shows the mean PSNR of the received video 5 mean PSNR at Rx no FEC
mean PSNR with interleaver+FEC
sequence based on the PER measured at the sports stadium. The 0
0 100 200 300 400 500
blue curve with no marker shows the error-free PSNR. The red Frame
curve with the star marker shows the mean PSNR per received Fig. 13 PSNR per frame with/without FEC and interleaving as
frame without AP-FEC. The green curve, with the star marker, measured at cricket match for link-speed 5 and route 1
shows the PSNR per received frame, when 0.875 AP-FEC is 6
video PER vs. code rate - link speed 3 and route 2

applied. The light blue curve, with the dotted marker, shows the block size=50
block size=200
PSNR per received frame, when 0.875 AP-FEC and 200 packet 5
block size=500

interleaving are applied. It can be seen that the combination of 4


interleaving with a 200 packet block size and AP-FEC decoding
PER (%)

with a 0.875 rate significantly improves the PSNR. 3

Fig. 13 shows an improvement in PSNR when a combination 2


of interleaving (with an interleaver block size of 200 packets)
1
and AP-FEC with code rate 0.75 is applied. However, in this
figure we also note that the PSNR with AP-FEC only is actually 0
0.5 0.6 0.7 0.8 0.9
worse than the case without AP-FEC. This confirms the code rate

observations made from our simulations (Fig. 6). Fig. 14 Evaluated PER vs FEC code rate for different block code sizes
Further investigation of the video quality for a variety of code
rates was conducted through the development of a generic VIII. CONCLUSIONS
erasure block code emulator for AP-FEC protection. Using the For broadcast transmissions, lost MAC frames (or
error patterns that were recorded during the measurements taken NALUs) are inevitable. However, with the use of cross
at the cricket stadium, this emulator generates new error patterns packet AP-FEC the NLR can be reduced (often to zero)
using two parameters, the block code size n and the code rate r. and the video quality improved. Results have shown that
The resulting error patterns were then fed into the video the degree of motion in the radio channel can affect the

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.
76 IEEE Transactions on Consumer Electronics, Vol. 55, No. 1, FEBRUARY 2009

error burst length, and hence the decoded video quality. [17] M. van der Schaar, S. Krishnamachari, S. Choi, X. Xu, ‘Adaptive
cross-layer protection strategies for robust scalable video
In combination with error concealment, good quality transmission over 802.11 WLANs’, IEEE Journal on Selected
broadcast reception over 802.11a/g can be achieved for Areas in Communications, Vol. 21, No. 10, Dec. 2003.
average SNR values in excess of 20dB (this assumes a [18] 3GPP TS 26.346 V6.1.0, Technical Specification Group Services
worst case Rayleigh fading channel). Further and System Aspects; Multimedia Broadcast/Multicast Service;
Protocols and Codecs, June 2005.
improvements in received video quality can be achieved if
[19] T.K. Chiew, P. Ferre, D. Agrafiotis, A. Molina, A. Nix, and D.
the AP-FEC approach is combined with packet level Bull, ‘Cross-Layer WLAN Measurement and Link Analysis for
interleaving. Selection of the most appropriate interleaver Low Latency Error Resilient Wireless Video Transmission’, IEEE
block size must be taken into account as a trade-off ICCE 2005
between latency and video performance. A wireless video
Victoria Sgardoni received the degree Dip. Ing. in
transmission system based on the concepts described in Electrical Engineering from the University of Patras,
this paper was developed as part of the VISUALISE Greece in 1985 and her MSc in Communications
project. Engineering from Imperial College, University of
London in 1988. She then worked on speech
recognition at Logica Cambridge, UK and
ACKNOWLEDGMENT subsequently for a number of companies including
Reuters and the Greek PTT OTE. She is currently a lecturer at the
This work was partly funded by the Technology Strategy Technological Educational Institute of Chalkis in Greece and is pursuing a
Board as part of the VISUALISE project. PhD at the Centre for Communications Research, University of Bristol.
Her research interests include broadband wireless communications, cross
REFERENCES layer video delivery, robust error control coding and signal processing.

[1] http://www.3cresearch.co.uk/item/8 Mantalena Sarafianou received her undergraduate


[2] S. Lee and K. Chung, "Joint Quality and Rate Adaptation Scheme degree in Electrical and Computer Engineering from
for Wireless Video Streaming,", IEEE AINA 2008, pp.311-318 Aristotle University of Thessaloniki, Greece in 2007
and continued with an MSc degree in Communication
[3] I. Haratcherev et al., "Optimized Video Streaming over 802.11 by Systems and Signal Processing from the University of
cross-layer signaling", IEEE Communications Magazine, Jan 2006. Bristol, UK in 2008. She is currently working towards
[4] M. van der Schaar and D. Turaga, "Cross-layer optimization and her PhD degree at the University of Bristol. Her
retransmission strategies for delay sensitive multimedia research interests include antennas, wireless communications, wireless
transmission", IEEE Trans. on Multimedia, vol. 9, No.1, Jan. 2007. video streaming and microwave breast cancer detection.
[5] C-M. Chen, C-W. Lin and Y-C. Chen, "Adaptive error-resilience
transcoding and fairness grouping for video multicast over Pierre Ferré received his engineering degree from
wireless networks," IEEE Conference ICC 2007, pp.1661-1666. ENST Bretagne, France in 1999, jointly with his
M.Sc. in Communiction Systems and Signal
[6] J. Ma, X. Feng, Y. Liu and B. Tang, “Video multicast over
Processing from the University of Bristol. He joined
WLAN”, IEEE ISCIT 2005, vol. 2, pp. 1400-1403, Oct. 2005.
the Centre for Communications Research at the
[7] J. Afzal, T. Stockhammer, T. Gasiba, W. Xu, ‘System Design University of Bristol in 2001 where he graduated with
Options for Video Broadcasting over Wireless Networks’, CCNC his Ph.D. in 2006. From 2004 he has worked as a
06 Research Associate in the University of Bristol on a number of European
[8] V. Sgardoni, P. Ferré, A. Doufexi, A. Nix and D. Bull, "Frame projects in the area of wireless video transmission. His main research
Delay and Loss Analysis for Video Transmission over time- interests include robust and error resilient video coding, distributed video
correlated 802.11a/g channels", IEEE Conf.PIMRC’07, pp.1-5, coding, wireless LANs and MANs (PHY/MAC), cross layer
Sept. 2007. communications for improved video delivery and error modelling.
[9] A. Willig, M. Kubisch, C. Hoene, and A. Wolisz, ‘Measurements
Andrew R. Nix received BEng and PhD degrees from
of a Wireless Link in an Industrial Environment using an IEEE
the University of Bristol in 1989 and 1993
802.11-Compliant Physical Layer’, IEEE Transactions on
respectively. He joined the Centre for
Industrial Electronics, volume 49-6, Dec. 2002.
Communications Research at the University of Bristol
[10] P. Ferre, A. Doufexi, A. Nix and D. Bull, ‘Throughput Analysis of as a member of lecturing staff in 1993. He is currently
IEEE 802.11 and IEEE 802.11e MAC’, IEEE WCNC 2004 Professor of Wireless Communication Systems. His
[11] IEEE Std 802.11g; Part 11: Wireless LAN Medium Access Control main research interests include broadband wireless
(MAC) and Physical Layer (PHY) Specifications: Further High- communications, radiowave propagation modelling, cellular network
Speed Physical Layer in the 2.4 GHz Band, d1.1, 2001. optimisation and advanced digital modulation/reception techniques. He has
[12] A. Doufexi, S. Armour, M. Butler, A. Nix, D. Bull, & J. published in excess of 300 Journal and Conference papers.
McGeehan, ‘A Comparison of the HIPERLAN/2 and IEEE
David Bull (M’94, SM’07) holds the Chair in Signal
802.11a Wireless LAN Standards, IEEE Communications Mag.,
Processing at the University of Bristol. He also leads
pp. 172-180, May 2002
the Signal Processing activities in the Centre for
[13] C. Komninakis, ‘A Fast and Accurate Rayleigh Fading Simulator’, Communications Research where he is Deputy
IEEE GLOBECOM 2003. Director. He is co-founder and Chairman of ProVision
[14] IEEE 802.16 standards, IEEE 802.16.3c-01/2 9r4, ‘Channel Communication Technologies Ltd. Prior to his current
Models for Fixed Wireless Applications’ appointments he was a Systems Engineer at Rolls
[15] W.C. Jakes, ‘Microwave Mobile Communications’, IEEE Press, Royce and subsequently a Lecturer at Cardiff University. He has also acted
New York, 1974. as an independent consultant to numerous international organisations in the
[16] Joint Video Team (JVT) of ISO/IEC MPEG and ITU-T VCEG. fields of video coding and signal analysis. Prof. Bull is a Fellow of the IEE
ITU-T Rec. and Final Draft Intern. Standard of Joint Video and a Chartered Engineer, U.K.
Specification (ITU-T Rec. H.264-ISO/IEC 14496-10 AVC), March
2003.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on May 19, 2009 at 04:35 from IEEE Xplore. Restrictions apply.