Sep 23, 2003 - ... presentations, point to multipoint digital slide projection, advertisement broadcasting ... 06125 Perugia, ITALY; emails: {frescura, giorni, feci,.
F. Frescura et al.: JPEG2000 and MJPEG2000 Transmission in 802.11 Wireless Local Area Networks
861
JPEG2000 and MJPEG2000 Transmission in 802.11 Wireless Local Area Networks F. Frescura, M. Giorni, C. Feci, and S. Cacopardi, member, IEEE
Abstract — In this paper the transmission of JPEG2000 and Motion JPEG2000 codestreams in 802.11 WLAN environment is analized. The issues related to the transmission of JPEG2000, using non reliable protocols as UDP-RTP, are investigated and a proper protection scheme for dealing with such an environment is presented, along with the obtained performance with different systems parameters. The proposed technique addresses a wide range of multimedia applications such as wireless PC to Digital Projector multimedia presentations, point to multipoint digital slide projection, wireless video-surveillance systems.
Obviously in wireless standards there is always the need for robust multimedia transmission. The mentioned applications require the best trade-off between Quality of Service (QoS) (i.e. image quality at the end receiver), bandwidth (i.e. transmission rate) and delay. Since JPEG2000 [[2]] provides a better quality at low compression rate when compared to other image coding standards such as JPEG, it represents a good candidate for wireless multimedia applications. Moreover JPEG2000, through its high scalability, enables a wide range of QoS strategies for operators.
Keywords: JPEG2000, UDP, RTP, WLAN, 802.11. I.
INTRODUCTION
WIRELESS Local Area Networks have evolved considerably from their first applications. The technology that was born as a way for some companies to transmit data in warehouses and on the factory floor has evolved into a cost-effective and flexible means for enterprises to connect mobile workers for increased productivity and simplified interoperability. The success of the IEEE 802.11b standard [[1]] that achieved nearly Ethernetequivalent speeds and the work of the Wireless Ethernet Compatibility Alliance (WECA) as an industry forum that pushed for Wi-Fi interoperability amongst equipment vendors, all played fundamental roles in this acceptance process. Moreover the new Wireless LAN standards as 802.11a and 802.11g are offering increased bit-rate (up to 54 Mbit/s) and additional operating frequency bands to improve system capacity. In addition to that, the fast growing spread of WLAN based hot-spots in public structures as airports, train stations and large commercial malls are becoming new, effective, technological platforms for innovative multimedia wireless applications with a shorter time to market with respect to the upcoming 3G mobile technologies. In this framework a number of applications are being studied and proposed as wireless PC to Digital Projector presentations, point to multipoint digital slide projection, advertisement broadcasting and wireless video-surveillance systems. ______________ The authors are with University of Perugia, DIEI, Via Duranti 93, 06125 Perugia, ITALY; emails: {frescura, giorni, feci, cacopardi}@diei.unipg.it
Contributed Paper Manuscript received September 23, 2003
Fig. 1 - Data Link Rate vs Range for 802.11b and 802.11a WLAN [[3]].
Fig. 2 - Throughput vs Range for 802.11b and 802.11a WLAN [[3]].
In 802.11b/a/g WLAN a significant difference between nominal bit-rate and effective throughput is often verified, mainly due to channel errors. Figure 1 and Fig. 2 shows the nominal data rate and the throughput versus range for 802.11b and 802.11a WLANs [[3]]. At boundaries of a certain radio mode a Packet Error Rate of 10-2 or even higher is not uncommon. Overcoming that boundary first produce a rapid amount of frame error rate, than it leads to a selection of a
0098 3063/00 $10.00 © 2003 IEEE
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
862
IEEE Transactions on Consumer Electronics, Vol. 49, No. 4, NOVEMBER 2003
more robust modulation mode with a consequent reduction of data rate. On the other hand some studies have analyzed the sensitivity of JPEG2000 to errors [[4]]. They have shown that JPEG2000 bitstream is very sensitive to channel errors, and that this sensitivity is different for various parts of the bitstream. In order to exploit this property, different classes may be defined, for instance headers and compressed data with different sensitivity/priority classes. This allows, for instance the application of Unequal Error Protection (UEP) techniques, as proposed by some authors [[5]], who take into account the different degrees of sensitivity of the bitstream. Although these techniques may provide adequate codestream protection in sparse error environments, they offer poor performance or even fail to protect data in environments in which errors are packet based, as in WLAN, when non reliable protocols as UDP are adopted. In this paper we propose a Reed Solomon based Unequal Error Protection scheme, combined with proper data interleaving and WLAN data partitioner.
reduces the window size before packet retransmission. This adjustment results in an unnecessary reduction of the bandwidth utilization causing significant performance degradation (poor throughput and long delays). With respect to the transmission of JPEG2000 and MJPEG2000 streams, major drawback of TCP is represented by packet retransmission, which results in significant delay and overhead in bandwidth occupation, and in a consistent delay jitter in delivering image frames. Such a drawback substantially neutralizes the positive effect due to JPEG2000 higher compression rates.
II. JPEG2000 CODESTREAM TRANSMISSION ON 802.11 WLAN – TCP VS. UDP In this paragraph, we analyze two main aspects related to the choice of the transport protocol for the transmission of JPEG2000 / MJPEG2000 stream over wireless 802.11 channel: the real time play out requirements with respect to the transmission of JPEG2000 compressed codestreams and the overhead due to headers. The mechanism implemented by the 802.11 [[1]] layer is a simple Send-and-Wait algorithm, where the transmitting station is not allowed to transmit a new fragment until it receives an ACK for the said fragment, or it decides that the fragment was retransmitted too many times and drops the whole frame. This results in a packet loss at the upper layer. Transport Control Protocol (TCP) [[6]], [[7]] is a reliable protocol designed to perform well in networks with low biterror rates, such as wired networks. TCP assumes that all errors are due to network congestion, rather than to loss. When congestion is encountered, TCP adjusts its window size and retransmits the lost packets. In wireless networks, however, packet loss is mainly caused by high bit-error rates over airlinks. Thus, the TCP window adjustment and retransmission mechanisms result in poor end-to-end performance. TCP uses a Go-back-N protocol and a timer-based retransmission mechanism. The timer period (the timeout interval) is calculated based on the estimated round-trip delay. Packets whose acknowledgements are not received before the timer expires are retransmitted. In the presence of frequent retransmissions, TCP assumes that there is a congestion and invokes its congestion control algorithm. The algorithm reduces the transmission (also called congestion) window size. As the window size is reduced, the transmission rate is reduced also. This window size adjustment technique prevents the source from overwhelming the network with an excessive number of packets. In the presence of high bit error rates in wireless links, that results in packets burst error due to the 802.11 layer, TCP reacts the same way as in a wired link: it
Fig. 3 - TCP header.
Transport protocols which do not guarantee packet delivery, such as UDP, could be the solution. UDP [[8]] provides users access to IP-like services. UDP packets are delivered just like IP packets - connection-less datagrams that may be discarded before reaching their targets. UDP is useful when TCP would be too complex, too slow, or just unnecessary. A serious difficulty in this direction is represented by the poor capability of the JPEG2000 decoder in recovering from packet loss situations.
16 bits
16 bits
Source Port
Destination Port
Length
Checksum Data
Fig. 4 - UDP header.
Our solutions take in account packets of reduced dimension down to 256 or 128 bytes. For this dimension, the overhead due to the header becomes significant. The LLC header and the 802.11 header concur to the total overhead with TCP or UDP header. In Fig. 3 the TCP header of 20 byte is depicted, while in Fig. 4 there is the lighter UDP header of 8 bytes. III.
JPEG2000 WIRELESS SYSTEM ARCHITECTURE
The proposed architecture for JPEG2000 / MJPEG2000 wireless system (JPWL) is based on a Reed Solomon Unequal
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
F. Frescura et al.: JPEG2000 and MJPEG2000 Transmission in 802.11 Wireless Local Area Networks
Error Protection scheme, combined with proper data interleaving and WLAN data partitioning. Fig. 5 shows the main functional blocks of the proposed architecture. j2k stream JPEG2000 Encoder
UEP Reed Solomon Coder and Block Interleaver
WLAN Packetizer
UEP Reed Solomon Decoder and Block Deinterleaver
WLAN Depacketizer and lost packet filler
j2k stream JPEG2000 decoder Unrecovered Error Map
802.11 WLAN
Error Sensitivity Table
Fig. 5 JPEG2000 Wireless System Architecture for WLAN Environment.
The JPEG2000 encoder provides both the codestream and an Error Sensitivity Table (EST), in order to signal to the UEP Encoder the relevant parts of the codestream and their sensitivity to errors. After the encoding, data are interleaved and packed to be sent over the WLAN. The packetizer puts markers on each packet in order to allow reordering and gap filling at the receiver side in case of packet lost. A. Error Sensitivity Table In addition to the compressed stream the JPWL encoder produces an index file of the codestream called Error Sensitivity Table (EST). This table contains values which are absolute and relate to a measure of the sensitivity. Although this table is used in this context to enable the operation of the UEP encoder, it may be also used in rate transceiver with QoS policy, intelligent retransmission techniques and so on. The table allows either packet indexing or byte range indexing of codestream. Its purpose is to carry error sensitivity levels but not how the codestream is protected. The mapping between sensitivity levels and actual protection technique is selected statically or dynamically by the UEP encoder, depending on the QoS requirements of the application and the target Packet Loss Rate (PLR) to be addressed. The EST information appears in the Main Header or in the Tile Part headers, identified by a dedicated marker segment. Each Sensitivity value (ES) is 2 bytes long in the table: S=0 corresponds to headers, S=1: 216-2, from high to low priority, S=216-1 indicates no priority. B. UEP Reed Solomon Encoding Reed Solomon (RS) codes are a special case of BCH codes. An (n, k) RS codes takes a group of k data symbols and generates n-k parity symbols; n is the codeword size, measured in symbols [[9]]. The error correcting capability of the code is directly related to the number symbols, which differ between any two different codewords. In particular, a terror correcting RS code from the Galois Field GF(2m) has the following parameters. Number of symbols: n = 2m – 1. Number of parity symbols: n – k = 2 x t.
863
Here m is the number of bits per symbol and t is the number of symbols, containing one or more bit errors, which the code can correct. In the proposed JPWL system architecture, we selected the RS codes over GF(16). This means that each symbol is a nibble, having m = 4 bits. Thus, for example, the (15, 9) code produces codeword with 15 nibbles (60 bits) each, and is able to locate and correct errors in any 3 nibbles. The choice of m = 4 bits is a good compromise between the capability of alignment with JPEG2000 codestream data structures and the number of different classes of codes with a range of protection capabilities and redundancy. C. Data interleaving and partitioning The proposed JPWL system architecture has been designed to address a packet error prone environment as WLAN, thus it implements in addition to the UEP RS Codec, an interleaver/deinterleaver system to provide error spreading at the receiver side, before the RS decoding process. Actually, the missing packets after the 802.11 MAC layer are identified and replaced by dummy data packets by the WLAN “depacketizer” and gap filler. This means that, after the gap filling process and the deinterleaving, a missing packet is converted into a number of sparsely distributed errors on the received codestream. Interleaving process is realized dividing the stream generated by the RS coder, into many group of codewords and storing them (by rows) into a matrix having a number of column that is multiple of the codeword length. Interleaving produces a sequence of packets of fixed dimension that are obtained by transmitting the symbols stored into each column of the matrix. Deinterleaving process is dual of interleaving, except that the matrix is written by column for establishing correct order of the packets. When deinterleaving is terminated, packets are sent to the RS decoder that corrects errors eventually presents into symbols and reconstructs the JP2 bit-stream. Due to an interesting feature of Reed Solomon codes, the not recovered codewords are tracked and signaled to the decoder (Fig. 5) that may use such information for implementing optional error concealment or selective retransmission techniques. D. Headers redundancy ISO/IEC 15444-1/ITU-T T.800, provides a set of error resilience tools. These tools are based on one critical hypothesis, precisely that the headers (Main Header and Tile Part header) of the codestream syntax are guaranteed to be error free. Unfortunately, although a strong Reed Solomon code is adopted to protect the headers, in the proposed JPWL system architecture, in case of a large number of lost packets, there is no guaranty that the headers are received free of errors. In the case of error within the headers, the codestream is not decodable in a proper way, and it may lead to a decoder crash. To reduce the probability of occurrence of this event, in addition to the Reed Solomon encoding, the headers may be redounded for those environment in which a large Packet Loss Rate (PLR) is expected. Header redundancy is flagged by a
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
864
IEEE Transactions on Consumer Electronics, Vol. 49, No. 4, NOVEMBER 2003
dedicated marker segment and it can be used to adopt various strategies in order to improve the decoding reliability: In the cases in which an information of uncorrected packet is available at the decoder (cfr. Section 3.5), an erroneous header may be discarded and an error free replica of the header may be used instead. Eventually the decoding process may be stopped, without a decoder crash, if an error free header replica is not available. The header replicas may be combined with a majority decoding strategy. In order to apply this strategy at least three replicas (or any odd number of replicas) of each coded header must be transmitted. The majority decoding may be used both in the cases in which no header replica is correctly received or when the information of uncorrected packet is unavailable. Some WLAN applications may adopt UDP for codestream transport and a reliable side TCP connection for commands and signaling. For those applications a selective retransmission of uncorrected headers is feasible, if an error free header replica is not available. E. Residual Error signaling After deinterleaving and Reed Solomon decoding some residual error may still affect the codestream. As said in previous section, these errors may be very harmful if located in one of the JPEG2000 / MJPEG2000 headers. In order to allow the JPEG2000 decoder to be aware of the presence and the “position” of these errors, the proposed JPWL system architecture produces a Residual Error Table (RET). This table may be embedded in the codestream with a dedicated marker segment or it may be produced as an external file that a JPWL “aware” decoder can exploit. The generic entry of the table is 8 bytes long. The first 3 bytes indicate the starting byte pointer of the error affected data block, the second 3 bytes field indicates the end byte pointer of the error affected data block and the last 2 bytes field indicates the number (if available) of errors in the data block or a generic indication of presence of errors in case the exact number of errors is not available. When adopting a Reed Solomon decoder the typical length of each data block is the one of the selected Reed Solomon codeword. IV. SIMULATION RESULTS IN WLAN ENVIRONMENT FOR SINGLE FRAME JPEG2000 APPLICATIONS Simulations were carried out using JPEG2000 Verification Model 8.6 for encoding / decoding processes and an internally developed Channel / Network simulator. The Channel / Network Simulator is inserted into the simulation environment, to realize the corruption process at the bit-level or packet level. When operating at packet-level it is possible to simulate the packets loss produced by the 802.11 MAC layer, when adopting UDP.
A.
System configuration for PGM image
A first set of simulation trials has been carried out using the grayscale PGM image format. The main simulations parameters are reported below: Image: Woman. Format: PGM (monochrome 8bpp). Image size: 2048x2560 pixels. Compression ratio: 8:1. Number of layers: 6. Number of tiles: 1. Quantization step size: 1/27. Packet size: 512 byte. VM Standard Tools: concealment, markers segmentation, resync markers. Repetitions: 1000. In Table 1 the different configurations for the k parameter of the RS Coder are shown. By varying the k parameter is possible to change the amount of redundancy for each codeword and consequently the robustness of the coded information. UEP schemes with an average resulting redundancy of 10%, 20% and 30% have been considered. Last tree rows of Table 1 show the additional redundancy introduced by network header (using UDP protocol). It is worth to point out that the network packetization process, adds redundancy proportionally with the number of packets. However number of packets grows up with the RS Redundancy, therefore overall redundancy introduced by network packetization is not monotonically increasing with the dimension of the compressed and coded data. TABLE 1 – RS/ NETWORK REDUNDANCY FOR SIX LAYER PGM IMAGE
RSRedundancy
10 %
20%
30%
Headers
RS(15,3)
RS(15,3)
RS(15,3)
Layer 0
RS(15,9)
RS(15,9)
RS(15,7)
Layer 1
RS(15,11)
RS(15,9)
RS(15,7)
Layer 2
RS(15,13)
RS(15,11)
RS(15,11)
Layer 3
RS(15,13)
RS(15,11)
RS(15,11)
Layer 4
RS(15,13)
RS(15,13)
RS(15,11)
Layer 5
RS(15,15)
RS(15,13)
RS(15,13)
Ratio
8:1
8:1
8:1
Network Overhead / Pkt
1.56%
1.56%
1.56%
Network Packets
1508
1669
1774
In conclusion we can assert that the network efficiency is reached with a compromise between RS redundancy and overall number of network packets. Fig. 6 shows average PSNR values vs. PLR and RS Redundancy amount for PGM image. In addition, for comparison purpose, Fig. 6 does also report the performance of the unprotected JPEG2000 codestream. It is worth to point out that, in case of unprotected codestream, due to the loss of packets, the corrupted codestream often leads the decoder to crash or terminate abnormally the decoding process. When
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
F. Frescura et al.: JPEG2000 and MJPEG2000 Transmission in 802.11 Wireless Local Area Networks
the number of crashes or abnormal terminations overcome the 10% of the number of transmitted frames the PSNR value on curve shown in Fig. 6 is set to zero to indicate an unreliable decoding process. The proposed protection method improves the performance considerably with respect to standard JPEG2000 error resilience tools. It allows obtaining: an effective minimization of decoder crash or abnormal termination events (since headers are almost always received correctly). An additional improvement of such a behavior may be obtained with one of the header redundancy techniques described in Section. III.D. a remarkably good performance with a minimal PSNR degradation, with respect to error free decoding also for Packet Loss Rate (PLR) of 10-2, with a still acceptable image quality for PLR of 10-1.
when the compression ratio is high and the resulting file size is small. The selection of a smaller packet length leads, however, to an increase of the overhead due to the WLAN protocol as shown in Table 2. Fig. 7 and Fig. 8 show average PSNR values vs. PLR and RS Redundancy amount for BMP image in the case of 1:20 and 1:10 compression ratio respectively. The PSNR results confirm that the system performance is mainly driven by the capability of the interleaver / deinterleaver to spread the effect of one or more packet lost on a large number of packets. When the number of packets is reduced because of the compressed image size, the system performance does not increase as expected even when using powerful UEP scheme. TABLE 2 - RS/NETWORK REDUNDANCY FOR SIX LAYER BMP IMAGE RSRedund.
10 %
20%
30%
Headers
RS(15,3)
RS(15,3)
RS(15,3)
Layer 0
RS(15,9)
RS(15,9)
RS(15,7)
Layer 1
RS(15,11)
RS(15,9)
RS(15,7)
Layer 2
RS(15,13)
RS(15,11)
RS(15,11)
Layer 3
RS(15,13)
RS(15,11)
RS(15,11)
Layer 4
RS(15,13)
RS(15,13)
RS(15,11)
Layer 5
Fig. 6 - Average PSNR (dB) for "woman" image. Compression 8:1 (512 bytes packet size).
B. System configuration for BMP image A second set of simulation trials has been carried out using the 24 bit color BMP image format. The main simulations parameters are reported below: Image: XGA text and graphic image (slides for PC presentation). Format: BMP (colors 24bpp). Image size: 1024x768 pixels. Compression ratio: 10:1, 20:1. Number of layers: 6. Number of tiles: 1. Quantization step size: 1/27. Packet size: 256 byte. VM Standard Tools: concealment, markers segmentation, resync markers. Repetitions: 1000. In Table 2 the different configurations for the k parameter of the RS Coder are presented. UEP schemes with an average resulting redundancy of 10%, 20% and 30% have been considered also. In this case the WLAN packet length is 256 bytes. The smaller packet size is been selected in order to keep the number of the WLAN packet high enough to allow a proper operation of the interleaver / deinterleaver, especially
865
RS(15,15)
RS(15,13)
RS(15,13)
Ratio
20:1
10:1
20:1
10:1
20:1
10:1
Network Overhead / Pkt
3.13%
3.13%
3.13%
3.13%
3.13%
3.13%
613
1120
669
1234
707
1310
Network Packets
Since packet size cannot be reduced below a certain extent, in order to keep the overhead low, this means that in case of small images, the interleaver memory must be sized to manage several images together, in order to keep the number of packet high. In MJPEG2000 this implies a delay that has to be considered and adequately dimensioned, particularly in real time applications.
Fig. 7 - Average PSNR (dB) for bitmap image, ratio 20:1 (256 bytes packet size).
Figure 7 and Figure 8 show how the number of packets impact on the system performance. Our intuition suggests that the interleaver operation is what is really driving the
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
866
IEEE Transactions on Consumer Electronics, Vol. 49, No. 4, NOVEMBER 2003
performance. In fact for compression ratio of 20:1 the number of packets it’s not sufficiently high to allow an effective spreading of the effects of many “large” missed packets. In these case the use of more powerful Reed Solomon codes does not help, as proved from the fact that the performance for average redundancy of 20% and 30% are very similar. When the number of packets increases (Figure 8) the interleaver can work better, the performance globally increases and there is a more significant advantage in using more powerful UEP configurations.
Fig. 8 - Average PSNR (dB) for bitmap image, ratio 10:1 (256 bytes packet size).
BMP case do apply. When the number of packets is low the interleaving capabilities drive the overall system performance. The use of powerful UEP schemes is almost useless, at least for practical level of redundancy (below 30% average). The difference in the shape between Figure 7 and Figure 9 is due to the nature of the image. The PGM is a natural image (woman) while the BMP is a PC based picture with text and graphics, the latter being richer of high-frequency components and more sensitive in terms of PSNR to uncorrected errors.
Fig. 9 - Average PSNR (dB) for PGM image, ratio 20:1 (566 bytes packet size).
In order to better clarify the interleaving/deinterleaving operation with respect to the number of packets variations, an additional set of simulation trials has been carried out. In this framework the same PGM and BMP images of the previous set of simulations have been selected. However, in this new simulation scenario, the compression ratio values of 20:1 and 10:1 have been considered and the packet size have been set to proper values in order to produce approximately the same number of packets in both the PGM and BMP cases, as shown in Table 3. TABLE 3 - RS/NETWORK REDUNDANCY FOR SIX LAYER BMP IMAGE RS RS RS Compression Packet Packets Packets Packets Format Ratio Size 10% 20% 30% 1:20 PGM
1:10 1:20 1:10 1:20
BMP
1:10 1:20 1:10
611
668
706
1119
1236
1312
283 byte
1121
1235
1312
2138
2372
2523
256 byte
613
669
707
1120
1234
1310
1126
1237
1313
2140
2368
2520
566 byte
128 byte
Figure 9 and Figure 10 show the performance of the PGM image for compression ratios of 20:1 and 10:1 and a packet size that leads the encoder to produce approximately the same number of packets as the BMP case. This means that Figures 9 and 10 are directly comparable with figures 7 and 8 respectively. Also in this case the comments provided for the
Fig. 10 - Average PSNR (dB) for PGM image, ratio 10:1 (566 bytes packet size).
Figure 11 shows the performance of both PGM and BMP images with the smallest packet size. For the PGM this size is 283 bytes, while for the BMP is 128 bytes. These values lead to a number of packet ranging from 1120 to 2520 for both PGM and BMP images, as shown in Table 3. The reduction of the packets size, while increasing the overhead due to the network, up to 6.25 % for 128 bytes packets, it allows the interleaver to operate in a very effective way. This leads to very good performance with very low PSNR degradation for PLR values up to 10-2. It’s now evident, how the interleaving performance is related almost entirely to the packet size with respect to the image size and how the system performance may be adequately tuned by a proper combination of packet size, interleaving size and UEP coding rate. Further work is required and planned to relate all these parameters into a global mathematical model that could allow to calculate “on the fly” the optimal combination of parameters for a certain WLAN environment.
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
F. Frescura et al.: JPEG2000 and MJPEG2000 Transmission in 802.11 Wireless Local Area Networks
867
V. SIMULATION RESULTS IN WLAN ENVIRONMENT FOR MOTION JPEG200 APPLICATIONS In this section the results of the simulation of the transmission of sequences of Motion JPEG2000 frames are reported. The selected frame formats are HDTV 720p (1280x720, 30fps) and SDTV PAL (720x576, 25 fps). The simulation environment is the same described above, for single images. The main purpose of this set of simulation trials is to validate the proposed protection technique for MJPEG2000 codestream and, most of all, to verify the performance of the multiple frame buffering/interleaving technique. Fig. 11 - Average PSNR (dB) for PGM and BMP images, compression ratios 20:1 and 10:1.
A. System configuration for BMP image in HDTV format The first set of simulation trials covering MJPEG2000 has been carried out using the HDTV 720p image format. The main simulations parameters are reported below: Image: HDTV generic image. Format: BMP (colors 24bpp). Image size: 1280x720 pixels. Frame rate: 30 frame per second. Compression ratio: 20:1. Number of layers: 6. Number of tiles: 1. Quantization step size: 1/27. Packet dimension: 256 byte. Repetitions: 1000. The network redundancy values, the number of interleaved frames and the corresponding number of generated packets for various UEP schemes are reported in Table 4.
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
868
IEEE Transactions on Consumer Electronics, Vol. 49, No. 4, NOVEMBER 2003
Compression ratio: 20:1. Number of layers: 6. Number of tiles: 1. Quantization step size: 1/27. Packet dimension: 256 byte. Repetitions: 100. For this scenario the network redundancy values, the number of interleaved frames and the corresponding number of generated packets for various UEP schemes are reported in Table 5. TABLE 5 - RS/NETWORK REDUNDANCY FOR PAL FRAMES RS-Redundancy 10% 20% 30% Compression Ratio
1:20
1:20
1:20
Network Redundacy
3.12%
3.12%
3.12%
Interleaved frames Network Packets
Fig. 12 - Average PSNR (dB) for multiple buffered-interleaved HDTV 720p frames, compression ratio 20:1.
1
5
1
5
1
5
373
1457
402
1609
423
1706
Figures 13 show the average PSNR vs. the PLR when interleaving is performed inside a single frame and when adopting multiple frame interleaving on 5 frames. The delay introduced by this buffering / interleaving is 0.4 seconds (endto-end), being still acceptable in many real-time applications and almost negligible in video streaming or video surveillance applications.
TABLE 4 - RS/NETWORK REDUNDANCY FOR HDTV FRAMES RS-Red.
10%
20%
30%
Ratio
1:20
1:20
1:20
3.12%
3.12%
3.12%
Network Red. Interleaved images Network Packets
1
2
4
1
2
4
1
2
4
698
1296
2491
768
1434
2764
812
1521
2939
Figures 12 show the average PSNR vs. the PLR when interleaving is performed inside a single frame and when adopting multiple frame interleaving. The use of a bigger interleaver, able to process more than one frame (specifically two and four in the simulated scenarios) allows to produce a larger number of packets without the need of a reduction of the packets size. As said in previous section, a large number of packets to play with, is the real driving factor for improvement of system performance. Figures 12 show how moving from 1 to 4 interleaved frames leads the PSNR performance to improve considerably for high values of PLR (up to 10-2) and to scale better when more powerful UEP schemes are adopted. B. System configuration for BMP images in PAL format The last set of simulation trials has been carried out using the SDTV PAL image format. The main simulations parameters are reported below: Image: PAL generic images. Format: BMP (colors 24bpp). Image size: 720x576 pixels. Frame rate: 25 frame per second.
Fig. 13 - Average PSNR (dB) for multiple buffered-interleaved PAL SDTV frames, compression ratio 20:1.
It is evident how the multiple frame interleaving leads to significantly improved performance with respect to single frame interleaving for every value of PLR and every UEP scheme.
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
F. Frescura et al.: JPEG2000 and MJPEG2000 Transmission in 802.11 Wireless Local Area Networks
C. Visual inspection results This section shows from a visual inspection perspective the effects of the proposed error protection techniques. Figure 14 shows the original HDTV frame.
869
Figures 16 (a-d) show 4 frames of the original PAL sequence used for both PSNR and visual evaluation. Figures 17, 18 and 19 show the decoded first frame of the sequence at a PLR value of 10-2 with (a) single frame interleaving and (b) 5 frames interleaving, for UEP redundancy values of 10%, 20%, 30% respectively. It’s evident how the multiple frame interleaving allows to obtain good performance also for high values of PLR and low redundancy UEP schemes. VI. CONCLUSIONS
Fig. 14 – Original HDTV image.
Figure 15a shows the decoded HDTV image at a PLR value of 10-2, a RS redundancy of 10% and single frame interleaving (PSNR = 26.74 dB). Figure 15b shows the HDTV image decoded at the same UEP conditions and PLR, when using 4 frame interleaving (PSNR = 32.74 dB). The performance advantage of the multiple frame interleaving scheme is clear.
In this paper the issues related to the transmission of JPEG2000, using non reliable protocols as UDP-RTP, in WLAN have been investigated and a proper protection scheme for dealing with such an environment has been presented, along with the obtained performance with different systems parameters. The results show the effectiveness of the proposed technique and identify crucial system parameters, as the interleaver / deinterleaver size for the optimization of system performance in case of transmission of small (or highly compressed) JPEG2000 or MJPEG2000 frames.
(a)
(a)
(b)
(b) Fig. 15 – (a) decoded HDTV image, PLR 1e-2, RS redundancy 10%, single buffered (PSNR = 26.74 dB); (b) decoded HDTV image, PLR 1e-2, RS redundancy 10%, four buffered (PSNR = 32.74 dB).
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
870
IEEE Transactions on Consumer Electronics, Vol. 49, No. 4, NOVEMBER 2003
(c)
(a)
(d)
(b) Fig. 18 – (a) decoded PAL image, PLR 1e-2, RS redundancy 20%, single frame interleaving (PSNR = 18.38 dB); (b) decoded PAL image, PLR 1e-2, RS redundancy 20%, five frame interleaving (PSNR = 42.08 dB).
Fig. 16 – (a), (b), (c), (d), 4 frames of the original PAL sequence.
(a)
(b) Fig. 17 – (a) decoded PAL image, PLR 1e-2, RS redundancy 10%, single frame interleaving (PSNR = 16.80 dB); (b) decoded PAL image, PLR 1e-2, RS redundancy 10%, five frame interleaving (PSNR = 36.45 dB).
(a)
(b) Fig. 19 – (a) decoded PAL image, PLR 1e-2, RS redundancy 30%, single frame interleaving (PSNR = 25.38 dB); (b) decoded PAL image, PLR 1e-2, RS redundancy 30%, five frame interleaving (PSNR = 45.70 dB).
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.
F. Frescura et al.: JPEG2000 and MJPEG2000 Transmission in 802.11 Wireless Local Area Networks
VII. REFERENCES [1] [2] [3] [4] [5]
[6] [7] [8] [9]
ISO/IEC 8802-11 – ANSI/IEEE Std 802.11, Information Technology – Part 11: Wireless Lan Medium Access Control (MAC) and Physical Layer (PHY) Specifications - First Edition, 1999-08-20. ISO/IEC 15444-1 / ITU-T T.800 JPEG 2000 Part 1 J. C. Chen, J.M. Gilbert, “Measured Performance of 5-GHz 802.11a Wireless LAN Systems”, Atheros Communications, Inc. Poulliat C., Nicholson D., “Impact and efficiency of error resilience tools for mobile applications”, ISO/IEC JTC 1/SC 29/WG 1, N2176, June 2001. Moccagatta I., Soudagar S., Liang J., and Chen. H., ”Error-Resilient Coding in JPEG-2000 and MPEG-4”, IEEE Journal on Selected Areas in Communications , Vol. 18, No. 6, pp. 899-914, June 2000. W. R. Stevens, “TCP/IP Illustrated”, Volume 1. Reading, MA: Addison Wesley, Professional Computing Series, 1984. S. Tanenbaum, Computer Networks, Third Edition. Englewood Cliffs, NJ: Prentice-Hall Press, 1996. J. Postel, RFC768, User Datagram Protocol, 28 August 1980 J. G. Proakis, “Digital Communications”, McGraw-Hill, Third Edition 1995.
Fabrizio Frescura (M’92 - Ph.D’97) was born in Perugia, Italy, in 1966. He received the laurea degree in Electronic Engineering from the University of Perugia in 1992 and the Ph.D. in 1997. He has been involved in personal communications, OFDM technique, applied to Wireless LANs, Digital Audio and Video Broadcasting, and DSP algorithms and architectures for Digital Modems. His current research activity is Wireless Transmission of Multimedia and DSP architecture for multimedia processing.
871
Cesare Feci was born in Chiusi, Italy, in 1979. He received the laurea degree in Electronic Engineering from the University of Perugia in 2003. His current research activity is Wireless Transmission of Multimedia and DSP architecture for multimedia processing. Since 2003 he has joined the DSPlab of the University of Perugia. Saverio Cacopardi received his degree in Electrical Engineering in 1970, from the University of Rome, Italy. From 1971 to 1975 he was employed by SIP (Italian Telephone Operating Company), working in the PCM transmissions and CATV fields. In 1975 he joined the University of Rome as an assistant professor. From 1979 to 1985 he has been Associate Professor of Electrical Communications at University of Ancona. From 1986 to 1991 he as been associate professor of Radio Aids to navigation at the University 'La Sapienza' of Rome. In 1991 he has joined the University of Perugia as associated professor and since 2000 he is Full Professor of Electrical Communications. His research activity is in personal communications and OFDM technique. He is a member of the IEEE and the AEI.
Mauro Giorni was born in Lecco, Italy, in 1978. He received the laurea degree in Electronic Engineering from the University of Perugia in 2002. His current research activity is Wireless Transmission of Multimedia and DSP architecture for multimedia processing. Since 2003 he has joined the DSPlab of the University of Perugia.
Authorized licensed use limited to: UNIV OF ALABAMA-TUSCALOOSA. Downloaded on May 13,2010 at 01:39:01 UTC from IEEE Xplore. Restrictions apply.