A Study of Wireless Compressed Digital Transmission Andreas Floros 1, Marios Koutroubas 2, Nicolas-Alexander Tatlas 2 and John Mourjopoulos 2 1 ATMEL Hellas S.A., Patras Science Park, Platani, 26 500 Patras, Greece 2 Group, WCL, University of Patras, 26 500 Patras, Greece Email: mourjop@wcl.ee.upatras.gr ABSTRACT The paper examines the transmission of coded according to the MPEG-I Layer III standard over the Wireless Digital protocol. The study presents the effect of the transmission parameters (such as distance, packet type and length) on the achieved wireless bit rate. The paper also analyzes several aspects concerning the real-time implementation of complete -based playback setups and addresses the effects of using single-channel, stereo or multichannel streams. 0. INTRODUCTION Wireless networks are becoming a necessity these days, meeting the need for a cable-free but networked environment. Several Wireless Local Area Networks (WLANs) protocols have been developed and standardized, such as the IEEE802.11a [2] and the Hyperlan/2 [3], operating in the 5GHz band and providing hi-speed communication with data rates up to 54Mbps. Thus, WLANs are able to offer wideband wireless connectivity between Desktop PCs, Laptops, Personal Digital Assistants (PDAs), and other portable electronic devices for voice, video and data applications. In the area of Personal Area Networks (PANs) where low complexity, cost and power consumption are prime objectives, [1] represents the main wireless technology operating mainly in home environments. Although this technology was initially developed for replacing cables, it evolved into a way to create small radio LANs. is a global standard, supporting ad-hoc connectivity of various electronic devices in a minimum 10-meter range and at a theoretical maximum rate of 1 Mbps. The protocol ensures robust and secure transmissions of voice and data in point-to-point or point-to-multipoint set-ups. It is well known that the real time implementation of time-critical applications through wired or wireless networks (e.g. digital distribution/streaming for real-time playback) should ideally be performed by connection-oriented links, which provide predefined bandwidth efficiency. specification defines synchronous-connection oriented mechanisms for servicing timecritical application (for example voice and ) at a constant rate equal to 64kbps. This limited bandwidth performance represents the main drawback of the technology, relevant to the transmission of compressed quality digital. However, packetoriented (or asynchronous) connections can alternatively be employed, provided that the physical layer offers sufficient bandwidth (at least equal to the maximum encoding bit rate). This work presents an experimental assessment of different set-ups (e.g. point-to-point, point-to-multipoint, various packet types and length) as real-time compressed digital transmission paths. Section 1 outlines the protocol basic operations. In Section 2 the requirements for digital transmission via are derived and possible applications are presented. Section 3 outlines the implementation of -based applications, as well as the testing parameters and criteria for evaluating the performance of the wireless link in terms of real-time data transmissions. In Section 4 the results obtained from the different applications are presented and discussed, leading to the conclusions summarized in Section 5. 1. BLUETOOTH OVERVIEW [1] is a short-range wireless technology designed to connect a variety of electronic devices. Since its introduction, it provides a viable and promising platform for low-power, low-cost radio implementations of PAN products, allowing the replacement of cables and infrared links and connecting such devices through RF links. enabled devices are designed to operate in the Industrial Scientific Medicine (ISM) frequency band of 2.4GHz. Thanks to its frequency hopping techniques, is theoretically designed to be immune to interferences in the above frequency band, achieving a total, raw bandwidth of 1 Mbit/s within a typical operating range of 10m (for Power Class 2 and 3 devices) up to 100m (for Power Class 1 devices). Fig. 1 shows a typical connection setup between three units. One of them is the Master of the connection, while the other are the slaves, forming a piconet. Up to seven slaves can be simultaneously connected to the Master using Asynchronous, Connection-Less (ACL) links that provide a packet-switched connection between the units. The ACL links are pointto-multipoint (with broadcast mode support from the master to all slaves) and can achieve a maximum total effective data rate of 721 kbps. The data transmission is performed in a 625 µsec time slot basis, and packet retransmissions are performed if necessary, in order to assure data integrity. The number of retransmissions can significantly reduce the achieved data rate, especially in environments where electromagnetic interference in the range of 2.4GHz exists.
The establishment of SCO links creates a bandwidth overhead, which dramatically decreases the transfer capabilities of any co-existing ACL link. Slave ACL Link 2 ACL Link 1 Slave The packet-switched nature of the ACL links with the limited Master bitrate is not suitable for real-time applications. Especially in Data Data noisy environments, packet retransmissions are applied to ensure data integrity, which further reduce the effective Fig. 1. Typical connection setup bandwidth. During an ACL connection, two categories of packets can be used: the Data-Medium rate (DM) packets and the Data-High rate (DH) packets. Their difference is that the latter do not include Forward Error Correction (FEC) information. There are three DM and three DH packet types termed DM1, DM3, DM5 and DH1, DH3, DH5 respectively. A DM3 packet is an extended DM1 packet. While a DM1 packet is sent during a single transmission slot, DM3 employs three time slots etc. The same stands for all the other type of ACL packets. There is also another ACL packet type called AUX1, which is never retransmitted. On the other hand, the specification also allows Synchronous, Connection-Oriented (SCO) links, which represent point-to-point wireless paths between a Master and a specific slave, suitable for time-bounded data transmissions (e.g. voice). Each SCO link provides a maximum of 64kbit/s data rate, while each master can support up to three concurrent SCO links. Like all cable and wireless networking standards, protocol is organized following the layer stack principle. The basic operations of the protocol are implemented by four basic components: a) the radio (RF) which physically receives and transmits data b) the baseband or link control unit which processes the transmitted or received data; c) the link management software that manages the transmission and d) supporting application software. The interconnection of the device with the host application is achieved through the Host ler Interface (HCI) and the upper Logical Link and Adaptation Layer (L2CAP), which realizes higher layer protocol services such as multiplexing, packet segmentation and reassembly and the conveying of quality of service information. A major advantage of the technology is the ability of extending its basic application level functionality using specific upper software layers called profiles. For applications, several profiles are currently under standardization (e.g. the Generic /Video Distribution Profile GAVDP [4] and the Advanced Distribution Profile A2DP [5]). These profiles define procedures for setting up an /video streaming and the requirements for devices necessary for support of the compressed quality distribution. 2. BLUETOOTH AUDIO APPLICATIONS Although is an attractive wireless specification for developing cable-free PAN products, the bandwidth limitation together with the nature of the permitted wireless links between the connected devices present some major drawbacks. More specifically: On the other hand, it is well known that high quality digital reproduction is performed at bitrates in the range of Mbps (nearly 1.4Mbps for stereo CD quality). Given the above practical limitations of the standard, the following considerations are necessary for realizing high-quality, real-time reproduction through : The original data must be compressed prior to the transmission. As with many other applications, the choice of the compression ratio is a tradeoff between the desired quality and the available bandwidth. A number of ISO/MPEG standards (MPEG-1, AAC, etc.), as well as many propriety standards (Dolby Digital, ATRAC, WMA, DTS, etc.) provide a wide range of compression options for mono, stereo and multichannel applications. However, for all cases, acceptable quality is achieved for rates of 96kbps or higher. Apart from the transmitted data, many applications require the transmission of control information (e.g. volume control, timing info etc), which must be multiplexed with the compressed data. For this work, a number of possible application environments for real-time, compressed quality digital reproduction were implemented and are presented in the following paragraphs, taking into account the previous considerations. For all cases, the MPEG- 1 Layer III (MP3) format was employed for compressing the digital content (which is also included in the A2DP [5] voting profile as an optional codec), as well as the Windows Media (WMA) coding, while several connection parameters were examined in order to optimize the reproduction performance and quality. 2.1 point-to-point applications The simplest wireless digital reproduction architecture is the case of transmitting data from a enabled device to another one, using a point-to-point connection and without transmitting any other information (as shown in Fig. 2). In such a case, the slave device is receiving the compressed data through an ACL link and, after decompression, it performs the playback operation. source (master) ACL Link receiver (slave) Time-bounded applications requiring a constant throughput must be established through SCO links. The effective bandwidth in such a case is 64kbps (or 192kbps for 3 SCO links). Especially for wireless transmission, specification defines either a 64kbps logarithmic PCM format (A-law or µ-law) or a 64kbps Continuous Variable Slope Delta Modulation (CVSD) through SCO connections. Fig. 2. Point-to-point single transmission and playback Typical applications related to the above wireless environment include single transmission of data between portable devices such as PCs, laptops, portable mp3 players, PDAs etc. Fig. 3 shows an extended format of the previous reproduction setup where additional control information can be exchanged 2
between the connected devices. This information can be user control info (e.g. track selection, gain control, equalization control) as well as data representing the status of the devices (for example playback status and timing information). These two different kinds of transmitted information represent discrete logical channels, which must be multiplexed in order to be transmitted through a single ACL link. According to the specification, this multiplexing can be performed by the L2CAP layer. The above setup is typical for applications where portable players (controlled by the user) are connected to non-portable sources. Although the effective bandwidth of the pure data transmission is generally reduced (depending on the traffic of the control channels), the hardware requirements of the receiver device can also be reduced, as all the necessary processing operations on the data (filtering and equalization, volume control, noise cancellation) can be performed on user demand on the source-side, prior to compression. source (master) ACL Link receiver (slave) Fig. 3. Point-to-point and user-controlled information transmission and playback 721- b N = i ch (1) b where b is the encoding bitrate (in kbps) employed for the compression of the information, b i (kbps) is the mean traffic of the control data through all the established links and denotes floor integer truncation. For example, if MPEG-1 Layer III compression is considered at a rate of b=128kbps, then Nch=5 assuming that b i =20kbps. In practice, it is expected that the above maximum number of the supported channels will be lower than the one derived by equation (1), due to a data-traffic overhead appeared when many connections are established between a master and many slave devices. Hence, while the above point-to-multipoint application scheme is suitable for stereo reproduction, the offered bandwidth is inadequate for multichannel (e.g. 6 channel) reproduction. Another approach, which overcomes the above channel limitation, is to employ the broadcast capability of the standard and broadcast the information from the master device to all slaves. As it is illustrated in Fig. 5, in such a case, the multichannel content is compressed and broadcasted, with the channel separation taking place on each of the playback devices. Hence, it is required that channel identification information should be assigned to each receiver in order to reproduce the appropriate channel. 2.2 point-to-multipoint applications While point-to-point connections can realize single channel data transmissions through enabled devices, stereo and multichannel reproduction requires the concurrent transmission of discrete channels to two or more playback devices. Fig. 4 shows a typical example of a stereo application, where two channels (eft and ight) produced by an source (e.g. CD-Player, DVD-Video etc) are compressed and individually wirelessly transmitted to the corresponding enabled loudspeakers through two different ACL links, able to transmit both and control information. source (master) Link 1 Link 2 receiver 1 (slave)... Channel Separate receiver N (slave) Channel Separate source (master) Link 1 Link 2 Left receiver (slave) Right receiver (slave) Fig. 4. Point-to-multipoint and user-controlled information transmission for stereo playback Since the expected bandwidth performance for each ACL link in the above case is (ideally) expected to be the half of the one obtained in the case of point-to-point setup, it is obvious that the maximum number N ch of the supported channels would be equal to: Fig. 5. Wireless multichannel transmission using point-tomultipoint connection setup in broadcast mode In broadcast mode, the maximum number of the channels equals to seven (the maximum allowable connections of a master device). Hence, apart from typical stereo applications, all current multichannel playback formats can also be supported (e.g. AC-3 coding, DTS etc). 3. IMPLEMENTATION AND TESTING Following the several application environments described in the previous Section, a set of different enabled devices was implemented, in order to investigate the ability of wirelessly transmitting compressed quality data using the protocol, for different connection parameters (e.g. distance, ACL packet type). The implementation was based on propriety single chip baseband controller [6], together with Class 3/0dBm RF modules. The developed applications were executed at a PC using the supported PCMCIA physical interface for the connection with the baseband controller. Two typical Pentium III PCs were employed for all test cases. A brief 3
description of all the applications developed for this study is given in the following paragraphs. 3.1 Application implementation 3.1.1 Point-to-point transmission In order to investigate the wireless transmission of using point-to-point connection, two applications were developed following the block diagram shown in Fig. 3. The application running on the Master host (shown in Fig. 6) is able to read stereo Wave files or CD- tracks, encode them using the MPEG-1 Layer III (mp3) standard and transmit them through an ACL link to the Slave host, which performs the decoding and real-time playback. Alternatively, direct transmission of already encoded (according to mp3 or WMA formats) material is supported. Before the start of playback, pre-buffering of the data is necessary for safety reasons. The length of the pre-buffering was set to 2 seconds of. The application also provides an interface for wireless control and transmission (from the Master side) typical reproduction commands and parameters such as: a) Start/Stop/Pause playback b) volume/balance control and c) a 10 band parametric equalizer. The above information is transmitted through the control logical channel, which is also used for transmitting playback timing and status information back to the Master at regular time instances of 1 second). The multiplexing of the control information with the actual data (transmitted through the data logical channel) was performed by the L2CAP layer. encode and transmit each channel separately through the ACL links to the corresponding playback devices. The above application alternatively supports broadcasting of the data to the connected devices. In this case, during the procedure of determining the Left and Right playback devices, a channel identifier must be assigned to both playback hosts. As is shown in Fig. 5, this identifier is used for extracting the appropriate channel from the broadcasted compressed stream (mp3 or WMA), and must be known to each remote player. 3.2 Test parameters and criteria For all applications implemented, the effective bit rate achieved during the wireless transmissions was measured (as described in Appendix A), since it represents the most critical parameter for uncorrupted, real-time playback. The measurements were performed in closed rooms with no additional interference existing in the range of 2.4GHz (Normal rooms) and with such interference present (Noisy rooms). During the transmissions, typical encoded material was transmitted. The size of the compressed transmitted data was in the range of 3 to 9MB, depending on the selected compression rate (64kbps up to 320kbps). For applications where the transmission of control information was activated, apart from the periodic exchange of the necessary control data between the connected devices, typical user interaction scenarios were applied in order to simulate real usage conditions. For the above measurements, several different connection parameters were considered: a) the distance d (in meters) between the connected devices b) the ACL packet type employed for each connection (as described in Section 1) and c) the length L p (in bytes) of the packets sent to the layer stack from the application layer (here referred as application packet length). Typical values of the above parameters considered in this work are shown in Table 1. Distance d (m) ACL packet L p (bytes) 1-20 DM1, DM3, DM5 1013 55x1013 DH1, DH3, DH5 (1k 55k) Table 1. Wireless transmission connection parameters 4. RESULTS 4.1 Point-to-point connection results Fig. 6. Wireless transmission and control user interface Fig. 7 shows the effective bitrate (b e ) measured in a normal room as function of the distance between two connected devices for all possible ACL and application packets considered in Section 3.2. During these measurements, no control logical channels were established, hence all the bandwidth was available for data transmission. From this figure, the following conclusions can be drawn: 3.1.2 Point-to-multipoint transmission An extension to the previous point-to-point application was developed, allowing the transmission of two discrete channels separately, as illustrated in Fig. 4. After the successful establishment of the ACL links between the master and the slave hosts, a special initialization procedure takes place in order to define the Left and Right playback devices. After the completion of the above procedure, the source (master device) can 4
bitrate (kbps) 700 600 500 400 300 200 100 0 0 2 4 6 8 10 12 distance (m) DH5 DH3 DM5 DM3 DH1 DM1 Fig. 7. Variation of the measured bitrate with the ACL packet type, the application packet length and the distance (no interference exists) (a) (b) (c) (d) The measured application-level bitrate strongly depends on the selected ACL packet. Clearly, for all cases of application packet length L p (appeared in Fig. 6 as a group of curves under the ACL packet name) the choice of DH5 packets achieves the highest bitrate values, which are very close to maximum of 721kbps defined by the specification. On the contrary, using other packet types significantly decreases the transmission s bitrate. This is in good agreement with results obtained from existing studies [8] assessing the effects of the ACL packets on the measured channel bitrate. Based on the previous results, it is obvious that DM1 packets cannot be used for real-time compressed quality playback, as they reduce the bitrate in the range of 100kbps. The effective bitrate remains almost constant with the distance. transmission is possible even for distances higher than the 12 meters considered here. The variation of the application packet length does not affect the achieved bitrate in a systematic way. Fig. 8 shows typical results for different application packet lengths obtained in noisy and normal rooms in the case of point-to-point transmission of content together with control information, using the multiplexing procedures of the L2CAP layer. In all test cases examined, DH5 ACL packets were considered, as they were found (see Section 4.1) to provide the maximum bandwidth efficiency. From this figure, the following conclusions can be drawn: bitrate (kbps) 700 650 600 550 500 450 400 350 300 0 2 4 6 8 10 12 14 16 18 20 Distance (m) 40k, Normal 40k, Noisy 30k, Noisy 20k, Noisy Fig. 8. Variation of the measured bitrate with the distance for different application packet lengths (with the existence of interference and no interference) (a) (b) (c) (d) A general reduction on the resulting bitrate was observed, compared to the bitrate measured when no control information is transmitted. This can be explained as follows: (i) a part of the available bandwidth is used by the control information transmission (ii) the L2CAP read operations performed at the both the master and slave devices (which restore the original and control data from the transmitted multiplexed stream) block the execution of the applications (as defined in the specification), until a complete L2CAP payload packet is received. This reduces the effective bitrate, which is measured at the application level. On the other hand, the effective bitrate is constantly increasing with the application s packet length (as, in such a case, the applications execution is blocked fewer times). However, further increment of the packet length (above 40k) was found to introduce significant delays on the transmission of the control information, which are unacceptable for real-time applications. For small application packet lengths (20k, 30k), the delays introduced due to the blocking procedures of the L2CAP layer are longer than those caused by the retransmissions of the transmitted packets, thus the effective bitrate remains constant with the distance. On the other hand, when large application packets are considered (40k), and in rooms with interference, the retransmission delays are dominant and reduce the bitrate with the distance. Moreover, in rooms with interference, the measured bitrate is reduced with the distance between the devices and no connection can be performed for distances higher than 12m. On the other hand, when no interference is appeared, the transmission is successfully completed at distances greater than 20m. 4.2 Point-to-multipoint connection results Considering the case of concurrently transmitting and control data through two ACL links for stereo reproduction (as described in Section 2.2) a mean effective bitrate value was measured at both playback devices equal to 295kbps (almost the half of the corresponding value obtained for a single ACL link see Fig. 8). The distance of both devices from the master host was 1m, while only DH5 ACL and 40k application packets were employed. However, the above application setup introduces synchronization problems between the two playback devices, as different rates of retransmissions are performed on the two ACL links. The above problem can be overcome by pre-buffering an adequate portion of data (for this work, 3 seconds of was buffered before the playback started), but this does not affect a possible time-mismatch of the control (user-defined) information, which is however short and practically not noticed. In the case of broadcasting the and control information to both playback devices, the reproduction was just acceptable, due to packet losses introduced in the wireless path (in broadcast mode, data integrity is not assured as no dynamic retransmission mechanism is applied). Since the specification allows the definition of the number of retransmissions N performed in broadcast mode (which is constant even if a packet is successfully transmitted), the above losses are partially reduced with increasing N. with proportional reduction of the effective bandwidth. Typical 5
measurements of the effective bitrate and the percentage of the lost information are presented in Table 2, as a function of N. It should be noted that due to the induced packet losses, the measurements of the effective bitrate (b e ) shown in this table are not precise, as the packet losses decrease the final values. N b e (kbps) data losses 0 551 7%-9% 1 325 2%-3% Table 2. Measured effective bitrate and data losses as a function of retransmissions performed in broadcast mode (for 1m distance). Effective bitrate (b e - kbps) 700 600 500 400 320 256 192 128 112 96 64 DM1 Broadcast (No retransmissions) Broadcast (1 retransmission) Stereo - L p =40k DH1 Real-time threshold line DH5 Mono - L p =40k DH3 DM5 DM3 Although the packet losses are small (especially in the case of 1 retransmission), since they are introduced to compressed data, the final reproduction is highly distorted. Moreover, as the losses vary in length and time between the two ACL links, they additionally introduce phase distortion between the two reproduced channels. The above observations seem to represent major restrictions for the development of multichannel applications (e.g. based on AC-3 coding DTS, etc), which given the bandwidth limitations of the protocol must be realized in broadcast mode. In practice, the above problems can be overcome by developing mechanisms in both the lower layer stack and the application layer, designed to perform packet loss signaling and retrieval procedures, as well as data restoration. Future research work by the authors is targeted towards this direction. 5. CONCLUSIONS In this paper, the wireless transmission of digital data using the protocol for real-time playback applications was examined, starting from the initial conclusion that reproduction can be achieved only after compressing the original material and transmitting it through ACL links. Several application scenarios were implemented and tested, realizing point-to-point and multipoint transmission of both and control information. The conclusions drawn from the results in the previous Section can help in assessing the optimal -based transmission parameters for compressed quality, real-time playback. Fig. 9 maps the measured effective bit rate (for 1m distance) with the data bitrate, indicating the maximum allowed compressed bitrate for all test cases considered. Ideally, uncorrupted real-time reproduction requires an effective bandwidth value at least equal to the compressed data bitrate. This condition is graphically represented by the Real-time threshold line. For all transmission cases examined, the maximum allowed compression bitrate is defined by the point of intersection of the mean measured effective bitrate value with the real-time threshold line. For example, from this figure it can be deduced that stereo and control information transmission (2 discrete ACL connections) is possible when each channel is encoded at a maximum of 256kbps (Stereo L p =40k case), while this rate is increased to 320kbps or higher when a single ACL link is considered (Mono - L p =40k case). 64 96 112 128 192 256 Total compressed bitrate (kbps) 320 and higher Fig. 9. Mapping diagram of the data bitrate and the maximum allowed compressed bitrate for all test cases examined (for 1m distance) The above maximum compressed rates can somehow be relaxed if sufficient pre-buffering of the transmitted data is performed prior the playback. However, this increases the memory requirements on the playback-device side, while it introduces longer user-perceived delays. On the other hand, it should be noted that for longer distances, the effective bitrate is decreased and this proportionally decreases the maximum bitrates. The limited bandwidth provided by the technology was found to be adequate for mono or stereo playback. What it became evident though is that wireless multichannel applications require broadcasting of the and control information. However, as in broadcast mode data integrity is not assured, the packet losses introduced, significantly affect the perceived reproduction quality. Future research work should focus on the development of packet loss indication and restoration algorithms in both the layer stack and the application level in order to achieve better wireless transmission quality and playback performance. Acknowledgements The authors wish to thank Photis Thanos, Dionysis Papadopoulos and Spyros Kapotas (ATMEL Hellas S.A.) for their valuable contribution and suggestions concerning technology implementation. References [1] SIG, Baseband Specification Version 1.1. [2] IEEE 802.11a Standard, Part 11 Medium Access (MAC) and Physical Layer (PHY) specifications, 1999. [3] ETSI, Broadband Radio Access Networks (BRAN); HYPERLAN type 2 technical specification, August 1999. [4] SIG, Specification of the System, Profiles, version 1.1, Generic /Video Distribution Profile. [5] SIG, Advanced Distribution Voting Profile version 0.9, 2001 (Confidential). [6] ATMEL AT76C551 Single Chip ler Datasheet, Aug. 2001. 6
[7] Performance Aspects of Scatternet Formation, Gy. Miklos, A Racz, Z. Turanyi, A Valko, P. Johansson, IEEE 2000. [8] OK?, Mark Rison and Like D Arcy CSR, Incisor Newsletter, Issue 34, August 2001 APPENDIX A: Effective bit rate measurement The effective bit rate b e of a transmission is measured in the application level using the equation: nlp be = 8 (bit/sec) (2) T where n is the number of the application packets received in a T seconds duration and L p is the length (in bytes) of the selected application packet. Obviously, the above rate b e measured in application level is different than the bitrate b c (bit/sec) which can be measured on the wireless link itself. Generally, b e is related to b c by: b e = f( b c ) (3) where f() is a non-linear function depending on the mapping mechanism of each transmitted application packet to ACL packets (DM or DH). However, if the length L p is selected to be an integer multiple of the L2CAP packet payload (equal to 1013bytes), the relation of b e and b c becomes linear and depends only on the application load. 7