Abstract The Ubiquitous Communications project aims at

Adaptive Error Correction for Wireless Communicationposition paperKoen Langendoen Maarten Ditzel Johan PouwelseFaculty of Information Technology and SystemsDelft University of Technology,The Netherlandskoen@pds.twi.tudelft.nlAbstractThe Ubiquitous Communications project aims at delivering video to a mobile user over a high-speed wireless link both indoor and outdoor.The high data rates(150Mb/s)require the use of4th generation transceivers that operate at high fre-quencies(e.g.,17GHz).We anticipate that the quality of the wireless link varies frequently due to changes in the environment(e.g.,moving po-sition).We propose to use adaptive forward er-ror correction in the communication protocols to guarantee a certain Quality of Service(QoS)level to the application.The key ideas are1)to dy-namically match the coding rate to the QoS re-quirements on a per stream basis,and2)to exploit the(dynamic)characteristics of the individual sub carriers of our OFDM modulation scheme.The responsiveness of our adaptive error correction depends on the feed-back loop implemented in software,which will be well within the100ms range of user perception.1Mobile multimediaThe advances in mobile communication indicate that in the near future high-demanding multime-dia applications can be offered to a mobile user in a wireless environment.Consequently,signif-icant research is being performed in the area of mobile multimedia as witnessed by the interest in workshops like MoMuC.The overall drive is to match the high demands in throughput and la-tency of multimedia to the limited,but rapidly increasing,performance of wireless communica-tion systems.Multimedia applications,such as video conferencing,typically require the trans-mission of real-time video streams;to obtain rea-sonable image quality,large amounts of data have to be communicated in(soft)real-time requiring small delays and jitter.Wireless systems,on the other hand,provide an unreliable communication medium with a high error-rate in comparison to wired networks.2UbicomThe Ubiquitous Communications project(Ubi-com)at Delft University of Technology[3]aims at developing a campus-wide system for wireless communication that is capable of supporting mul-timedia applications.By using an embedded sys-tems approach we can integrate important func-tionality such as compression and error correc-tion in hardware to realize the performance level required by the application.Our target is a vi-sual geographic information system that uses aug-mented reality techniques to display information on a mobile user’s headset;information is super imposed on the user’s view using a retinal scan-ning display.Augmented reality requires imme-diate response to changes in thefield of interest of a user yielding strong real-time constraints on the system.As an example,any head movement is a fast change in thefield of interest of the user to which the system has to respond by displaying the corresponding augmented reality information at the correct position.To minimize power con-Figure1:Ubicom architecture. sumption in the mobile unit,the main processing power is located in the backbone network of the Ubicom architecture(see Figure1).Since the main computational tasks of the mo-bile application will run on the Ubicom backbone, the wireless communication subsystem must be able to handle large data volumes with small de-lay so the imposed real-time constraints can be met.To handle the anticipated data rate of about 10Mb/s per user,Ubicom will use a4th gener-ation transceiver that operates at17GHz using OFDM modulation providing a raw data rate of 150Mb/s.It’s difficult to estimate the error rate at17GHz,but we assume that it will be higher than the rates reported for equipment using lower frequency bands such as waveLan[5].3Wireless communication Communication protocols for wired networks such as TCP/IP do not handle the errors of wire-less transmission well,since they interpret packet loss as a result of congestion and,consequently, slow down the transmission rate.Enhancing re-liable transport protocols provides some perfor-mance improvement[2],but the use of retrans-missions to handle corrupted packets compro-mises the tight latency requirements of multime-dia applications.Augmented reality applications impose even more stringent real time constraints. Many wireless communication systems employ forward error correction(FEC)to handle trans-mission errors in real-time.To overcome the over-heads of adding and removing the redundant in-formation needed for error correction,most FEC schemes are implemented in hardware.The disadvantage of implementing FEC in hardware is the lack offlexibility.The qual-ity of the wireless channel depends on the dis-tance between the transmitter and receiver,hence, for optimal performance the coding rate must be adapted to the local conditions.Fortunately,soft-ware FEC implementations have recently been shown to achieve acceptable throughput(100 Mb/s)on stock PC hardware[7].With software FEC it will be possible to adapt dynamically to the observed channel quality.4Adaptive error correction Several researchers have advocated the use of adaptive error correction codes(e.g.,[4,6]).The typical approach is to use a block code like Reed-Solomon that can correct a certain number of er-rors,say,in a packet.When the rate of un-correctable packets,which contain more than errors,rises above a certain threshold,the cod-ing rate is reduced by adding more redundancy to the data stream.Conversely,when the error rate drops,the coding rate is increased to take ad-vantage of the improved link quality.Note that a feedback loop is required to allow the transmit-ter to adapt the coding rate according to the error rate observed at the receiver.This limits the re-sponsiveness to tens or hundreds of milliseconds, which is fast enough to adapt to changes caused by human action such as motion.In the Ubicom project,we want to improve the general approach to adaptive error correction in two ways.First,we take the multimedia applica-tion’s requirements into account by applying dif-ferent coding schemes to different data streams. Second,instead of regarding the channel as hav-ing a single error characteristic,we want to take advantage of the different characteristics of the sub carriers used by our OFDM modulation tech-nique.4.1Adaptable streamsWe feel that multimedia applications running on the Ubicom architecture must be aware of the unreliable nature of the wireless channel if they want to achieve maximum performance.Simply relying on the basic communication software to transparently hide all errors,through error cor-rection and retransmits,will compromise perfor-mance too much.Reliable transport protocols add overhead to start and stop timers,create a copy of each data packet in case a retransmission is required,and cannot give any real-time guaran-tees since packets may be retransmitted multiple times.Moreover,a significant amount of power is consumed by these type of protocols.By giving up transparency and making applica-tions aware of unreliable communication,we can take advantage of the different data streams com-municated over the wireless link.Multimedia ap-plications typically communicate multiple types of data(video,audio,files,control)with different quality of service requirements between the Ubi-com backbone and mobile user.The throughput over the wireless link can be optimized by select-ing an appropriate FEC per data stream based on the specific QoS requirements.For example,an uncompressed video stream must be transmitted with the smallest possible delay and will use a light FEC with a high coding rate.Afile trans-fer,in contrast,requires reliable transfer and uses a heavy FEC and retransmissions.The adaption of the FEC to the QoS of a data stream must be done dynamically to account for the changes in quality of the wireless link.4.2Adaptable carriersThe OFDM modulation technique employed in the Ubicom transceiver divides the17GHz fre-quency band into a number(e.g.,512)of sub bands carrying independent data signals.Since these sub carriers differ in frequency,error sources such as reflection and diffraction will im-pact the various sub carriers differently.Although the differences between neighboring carriers will be small,differences between sub carriers spaced further apart will be noticeable.For example,in the Median project[1]the sub carriers at both ends of the60GHz band are of poor quality com-pared to the ones in the middle.The general approach with OFDM modulation hardware is to run a single,fixed FEC across the sub carriers to compensate the bad carriers with the good ones.In Ubicom,however,we pro-pose to take advantage of the individual sub car-rier characteristics by mapping the different QoS streams of the multimedia application to the ap-propriate carriers.By allocating the lowest QoS requirements to the worst sub carriers,the coding overhead of the FECs can be minimized yielding maximal throughput over the wireless link.5ImplementationTo achieve the adaptive error correction policy for Ubicom,we need aflexibility that can only be reached through careful hardware software code-sign.Given the high data rates(150Mb/s),how-ever,it is impossible to have the adaptive er-ror correction on the main CPUs in the mobile station that must be shared with the compute-intensive augmented-reality software driving the user’s head set.Therefore we propose to use a dedicated co-processor in the mobile station as shown in Figure2.The incoming signal is converted from the ana-log to the digital domain,where it is fed through a fast Fourier transform(FFT)yielding the OFDM symbols.These symbols are then mapped to(par-allel)bit streams,which are fed into the error cor-rection unit.Finally,the corrected data stream is placed in memory where the main CPUs can access it.By implementing the error correction functionality on a co-processor,the incoming data can be decoded in parallel with the application running on the main CPU.For thefirst prototype implementation we will probably use an off the shelve low-power DSP or FPGA chip,which pro-vides an acceptable balance between programma-bility and speed.The mapping of data streams onto OFDM sub carriers and selection of the appropriate FEC is aFigure2:Structure of the mobile station.non-trivial problem.The following(incomplete) list of research questions must be addressed by Ubicom to establish a reasonable policy that uti-lizes the wireless link effectively:Is it feasible to consider carrier mapping and FEC selection as two orthogonal issues,or must both factors be considered together?For a given unreliable data stream,how should we balance coding overhead,error rate,and throughput?For reliable communication,should we usea strong FEC with few retransmissions,or alight FEC with many?What kind of fairness must be guaranteed be-tween multiple data streams?How to implement the feed-back loop?One option is to dedicate a few OFDM carrier to control signals,another option is to layer on top of a data-stream.Which is best?6ConclusionsTo maximize the usage of the scarce through-put on a wireless link,adaptive error correction schemes adjust the coding rate depending on the current quality of the channel.In this position pa-per we outline how the general adaptive scheme can be improved in two ways:1.exploit the different QoS requirements of theindividual data streams in a multimedia appli-cation by selecting a code rate per stream, 2.take advantage of the individual OFDM subcarrier characteristics by mapping reliabledata streams to the best carriers,and bulk data streams to the worst carriers.Both improvements allow a lighter FEC to be used than in the traditional approach,so the effec-tive throughput over the link increases.We will need to refine our ideas and address a number of open issues before we can actually realize an im-plementation within Ubicom. References[1]Advanced Communication Technologies andServices(ACTS).The Median Project./ACTS/.[2]R.Caceres and L.Iftode.Improving the perfor-mance of reliable transport protocols in mo bile computing environments.IEEE JSAC Special Is-sue on Mobile Computing Network,1994.[3]Delft University of Technology.TheUbiquitous Communications Project.http://ubicom.twi.tudelft.nl/.[4]D.Eckhardt.A Case for Adaptive Error Correc-tion in a Wireless Local Area Network.Carnegie Mellon University,1997.[5]D.Eckhardt and P.Steenkiste.Measurementand analysis of the error characteristics of an in-building wireless network.In Computer Commu-nication Review(SIGCOMM’96),volume26(4), pages243–254,October1996.[6]D.L.Goeckel.Robust adaptive coded modulationfor time-varying channels with delayed feedback.In35th Allerton Conf.on Communication,Con-trol,and Computing,October1997.[7]L.Rizzo.Effective erasure codes for reliablecomputer communication protocols.In Computer Communication Review,vol27(2),April1997.。