Internet Access to High Speed Trains using the IEEE a System

Transcription

1 Internet Access to High Speed Trains using the IEEE 82.11a System DANNIEL BYBERG KRISTIAN CALAIS PAUL JONSSON Department of Signals and Systems CHALMERS UNIVERSITY OF TECHNOLOGY Göteborg, Sweden 23 EX26/24

2

3 Acknowledgements First of all, we thank professor Arne Svensson at the department of Signals and Systems for creating the opportunity to write this thesis, being our supervisor and guiding us when needed. We also want to thank Anders Ljung at UTAM-IT AB who came up with the idea of Internet Access to High Speed Trains using the IEEE 82.11a System. Further, we thank all the people who we have discussed our work, including professor Tony Ottosson and Ph.D. student Matts-Ola Wessman, at the department of Signals and Systems. Last, but not least, we want to thank our friends and families who have accepted our dedication to this thesis and supported us through the hard work. Front page: A 3D view over time and frequency showing the inverse gain of the multipath channel.

4

5 Abstract Internet Access to High Speed Trains using the IEEE 82.11a System concerns the possible use of a Wireless Local Area Network (WLAN) system, according to the IEEE 82.11a standard, as a high-speed Internet connection to railroad trains with a speed of up to 4 km/h. The 82.11a system, based on the Orthogonal Frequency Division Multiplexing (OFDM) technique, is first analyzed and described to the reader. Then system performance simulations are run to show the possibilities and difficulties to use 82.11a as a data link to a fast moving train. The data communication is made between closely spaced access points along the railroad and a terminal on the train. Due to the fact that the transceiver is moving very fast, Doppler effects will cause severe problems and the received signal will be highly distorted compared to the transmitted sequence. This phenomenon complicates the decoding process of the received signal. This thesis proposes some channel estimation algorithms and detection techniques in order to make detection possible for the data link in this environment. To do this, channel estimations based on the least square method and different interpolation techniques will be used. Normally, the channel estimation is based on training symbols that are transmitted in the beginning of each data packet, but due to the high speed of the vehicle and the multipath channel, this will not be enough. For this reason, pilot symbols that are transmitted at four frequencies have also been implemented into the estimation algorithms. A comparison of the bit-error-rate (BER) performances for the different techniques at a relatively speed of 4 km/h is finally generated and presented for the reader. Furthermore, relevant background theory describing how the IEEE 82.11a system is designed in a detailed block view is given. Necessary knowledge like properties of the physical channel will also be explained. Keywords OFDM, Doppler shifts, multipath channel, intersymbol interference (ISI), intercarrier interference (ICI), coherence time, coherence bandwidth, interpolation, fast Fourier transform (FFT), cyclic prefix (CP), BER.

6

7 Contents Chapter 1 - Introduction Background information Outline of the thesis... 9 Chapter 2 - An overview of the system... 1 Chapter 3 - Orthogonal Frequency Division Multiplexing Introduction to OFDM Basic principles of OFDM Cyclic prefix Channel estimation in OFDM based systems A mathematical description of OFDM Chapter 4 - The IEEE 82.11a standard Parameters related to the system Main fields of the 82.11a frame Preamble Pilots Signal Data The Inverse Fast Fourier Transform (IFFT) block Requirements Chapter 5 - Channel modeling The COST27 standard Changes from COST27 6-RA Doppler spectrum Rayleigh fading distribution Ricean fading distribution Echoes Path loss and log-normal fading Simulation method - The Spectrum method channel.m - MATLAB coding Channel properties Delay spread Coherence bandwidth Coherence time Phase shift Special channel properties in this thesis Verificationsand problems Chapter 6 - Channel estimation and detection techniques System model Channel estimation algorithm based on the long training sequence Channel estimation algorithms based on pilot symbols Optimal channel estimation Differential detection Diversity detection Chapter 7 - Simulations General simulation properties and goals Parameters used to evaluate the performance of a system System verifications System verification using an AWGN channel System verification using a single Rayleigh channel System verification using the COST27 channel model Link budget Initial analysis Performance analysis Chapter 8 - Conclusion and future work Conclusion Future work References...I Appendix A - BPSK, QPSK, 16-QAM and 64-QAM constellations...iii Appendix B - Calculation of the number of pad bits in the data field...iv Appendix C - Acronyms... V Appendix D - Noise calculations in a baseband system...vi

8

9 chapter one - introduction 1.1 Background information Chapter 1 - Introduction Wireless communication equipment as cellular phones and notebook computers are today used by almost everyone, at least that is the case in many of the industrial countries. The applications supported by these devices, like on-line meetings with live cameras, downloading of music and remote control software, demands a high-speed data connection. High-speed connections as these are not available to users at all times. Consider travelling by train for instance, imagine the convenient to switch the laptop on and continue working on the way home from a meeting several hours away. Assuming that the work demands an Internet connection of course, as the case often is nowadays. Others might want to be entertained by watching a movie to pass the time more pleasantly. Anyhow, train companies supporting a relatively good connection thus enabling applications like those mentioned above would be welcomed. There is a lot of wasted time to be saved out there. Indeed, it is enjoyable to read a book as well, but more opportunities suits more people that might consider the train on their next trip. Today there exists several Wireless Local Area Network (WLAN) standards but non is designed to be used in an environment where the information is being sent over open terrain to a fast moving train. It would be interesting if such a system could be used under more demanding situations than mainly indoors and between slow moving access points. This idea is exactly what will be tried to be accomplished in this thesis. The IEEE 82.11a WLAN system, operating at the 5.8 GHz unlicensed frequency band will be used here. Within this band, it is allowed to transmit with the relatively high power of 1 watt. Data will be transmitted over a simulated channel, assumed to be a good model for the physical channel that the receiver/transmitter 1 will face during transmission to a train. At the receiver side, algorithms will be designed and tested for the possible use in the system as it is constructed today. The goal is to have a maximum range of 8-13 km between the antennas. With efficient hand-over between the base stations, this would be accomplished by placing antennas on existing telephone posts at every 16 km close to the railroad and on an access point at the train. These telephone posts do already exist in Sweden, but that is not the case in all countries. Finally a hint of future work to get a satisfying data transmission performance based on a modified 82.11a system is given. Those whom only like to know about OFDM systems in general, and the 82.11a system in particular can also read this thesis. 1.2 Outline of the thesis The outline of this thesis is: Chapter 2 gives an overview of the system and why it is chosen as a candidate to investigate in this project. Chapter 3 introduces the reader to the basic principles of an OFDM system. Advantages and disadvantages connected to OFDM will be mentioned. Chapter 4 describes the 82.11a standard. Timing and rate dependent parameters are discussed and all parts of the system, like the convolutional encoder and the inverse fast Fourier transform block, is carefully explained. Chapter 5 specifies the channel characteristics and properties used for the simulations. Chapter 6 explains the channel estimation algorithms. The system model is also described. Chapter 7 is about the simulations made in the thesis. The result in form of BER curves and decision boundaries are shown. Chapter 8 contains the discussions about the results found in this thesis. It also introduces some ideas to future work, in order to improve the system. 1. Note that the linguistic use of receiver and transmitter in this thesis refers to the direction of the signal. Both the train and the base stations along the rails transmits and receives signals. They should correctly be referred to as transceivers, but then the directional information of the signal is lost for the reader 9

10 Internet Access to High Speed Trains using the IEEE 82.11a System Chapter 2 - An overview of the system An important factor when determining the viability of using the IEEE 82.11a standard in an application like this is how it measures up to existing technologies and if there are any other possible alternatives. Where are we standing today regarding trains offering high-speed Internet connections? There are some ways that we know today but none that really can claim to offer a stable broadband connection as in offices and other more fixed locations. One of these systems is the one soon to be offered by Linx - a Scandinavian railway company. As they claim they will be the first in the world to offer Internet in some of their trains. The technology is based on a satellite downlink and multiple GSM system uplinks. This system does not measure up to the data rates considered in this thesis. One easily identifiable complication using the 82.11a system is that it was originally designed to operate in an office environment and of course is adapted for it. Compared to many outdoor applications, the 5.8 GHz carrier frequency imposes a serious limitation to the range of the signal. The power loss is proportional to the carrier frequency raised to a factor of typically The more recent proposal of using OFDM modulation in the 2.4 GHz band (82.11g system) is of interest if it had not been for the fact that it will not meet the power considerations. The 82.11g system is offering the same data rate as a in the 2.4 GHz band, but a lower transmitted power must be used. The facts that the 5.8 GHz band is unlicensed and the high allowed power transmissions are the main reasons to why the 5.8 GHz band will be used in this thesis. As for the use of an OFDM based modulation it is well suited for this type of application, i.e. point-topoint communication. It has had great breakthroughs in recent history being adopted in ADSL modems, WLAN standards and in HDTV systems. One of its main characteristics is its spectral efficiency and its ability to compensate for multipath in a cheap way - using Fourier transforms to isolate and compensate for frequency selective fading. A more in-depth explanation of the OFDM is given in chapter 3.. In total, OFDM seems like a good technology to use but there are some issues that need to be addressed to continue the research in this direction. What are the noise characteristics of the train? Can we use the 5.8 GHz band, considering both the legal issues and the range problem? Is the cost of raising base stations and maintaining them worth the effort? How many modifications, except of course the required power amplification, is required to get this system to work in relation to stock IEEE 82.11a parts? Some of these are: Maximum length of the data packets? Will more known training symbols be needed? Maximum distance between the transmitter and the receiver? 1

11 chapter three - ofdm Chapter 3 - Orthogonal Frequency Division Multiplexing This chapter introduces the reader to some principles of how an OFDM system functions. Both an intuitive and a more mathematical point of view will be used in order to explain the system. 3.1 Introduction to OFDM During the last decade, the use of Orthogonal Frequency Division Multiplexing (OFDM) has become a very popular modulation technique for wireless communications. This technique is nowadays used in several applications like the European broadcasting standards Digital Audio/Video Broadcasting (DAB/DVB) [1] and for Wireless Local Area Networks like HiperLAN type 2 and IEEE 82.11a [2]. The numbers of high data-rate applications are growing continuously and OFDM can offer the required high rates, for example IEEE 82.11a that supports rates up to 54 Mbits/sec [12]. Some often-discussed reasons to why OFDM has been successful are the robustness against multipath channels and its spectral efficiency [3]. OFDM also has good properties when it comes to the hardware production since it uses the fast Fourier transform (FFT) algorithm, which easily can be implemented in an integrated circuit [4]. 3.2 Basic principles of OFDM In OFDM, a high-rate serial data stream is divided into lower-rate parallel streams that are transmitted simultaneously over several subchannels, i.e. OFDM is a multicarrier system. When the symbol duration increases, as the data rate decreases 2, the problem caused by a multipath channel may be reduced or even eliminated [5]. The conditions to eliminate the effect of the multipath propagation will be discussed later. As the name OFDM may suggest, the transmitted signal is formed in the frequency domain. The data stream is divided onto a number of parallel streams that are transmitted simultaneously over several narrow subchannels. Then, an inverse Fourier transform (IFT) translates the signal to the time domain and the signal is returned to a serial stream again before it is transmitted over the channel. The reverse operation is done when the signal has reached the receiver, which means that the received serial stream once again is divided into several parallel streams and the Fourier transform (FT) algorithm transforms the signal back to the frequency domain and the decoding procedure may begin. 2. Due to serial to parallel conversion. 11

12 Internet Access to High Speed Trains using the IEEE 82.11a System sin( 2 πf c ) Data symbols S/P + sin( 2 π f c + (N 1) f ) Frequency I F F T Time domain F F T Frequency Transmitted (QAM) Received Figure 3.1: A simple OFDM system. The symbols are constructed in the frequency domain, an IFT transforms the signal into the time domain and it is transmitted over the channel. The reverse operation is made at the receiver side. Figure 3.1 shows a very simple OFDM system, where f is the subcarrier spacing and S P is the serial to parallel conversion. The signal may look a little bit random in the time domain, but since the interesting information is stored the signals frequency domain it does not really matter how the signals acts in the time domain. There are however some restrictions regarding the time domain signal; it cannot consist of too large amplitudes and some other, more practical, restrictive aspects. Obviously, many details have been left out in the figure, but it is a good introduction to a first OFDM system. In a practical system the output from the IFFT block, with an added guard time, form an OFDM symbol. The reason for the guard time, which is also called cyclic prefix, will be explained in the next chapter, In contrast to the time domain properties, the frequency properties are of huge interest and will be discussed in the next paragraphs. An OFDM system is a special case of a multicarrier system. Each subchannel frequency is orthogonal to another in order to form an orthogonal signal set [6]. This is much more bandwidth efficient, compared to an ordinary multicarrier system where the different frequency bands are divided into nonoverlapping subchannels. (a) Frequency Saved bandwidth (b) Frequency Figure 3.2: Channel allocation of ordinary FDM- (a) and OFDM system (b). There is a lot of bandwidth to gain by using OFDM [5]. In figure 3.2, the image at the top (a) shows an ordinary multicarrier system with frequency division multiplexing (FDM), and the figure at the bottom (b) shows an OFDM system. One can see that there is much bandwidth to gain by using OFDM instead of FDM. On the other hand, one may also wonder how robust the OFDM system is against changes in the received signal spectrum. If the orthogonality of the system is lost intercarrier interference (ICI) is introduced to the system. ICI means that there exists crosstalk between some of the subcarriers [5]. In figure 3.2 it is clear that the system is free from ICI since the spectrum from all other subchannels are zero at the centre frequency of an arbitrary studied channel. 12

13 chapter three - ofdm Cyclic prefix In most wireless systems, the multipath channel causes many problems which the receiver must deal with in order to reach a good performance. The multipath channel arises from the fact that the transmitted signal travels different paths on its way from the transmitter to the receiver. This means that delayed versions of the signal will arrive and sum up at the receiver. The receiver in this system will have a high relative speed compared to the transmitter. This relative speed will introduce frequency shifts, called Doppler shifts, which are proportional to the relative speed v and the wavelength λ of the system like v λ. The multipath channel causes two main problems in OFDM based systems, namely intersymbol interference (ISI) and intercarrier interference (ICI) [6]. Intersymbol interference occurs when the received symbol is affected by the previous transmitted symbol. In single carrier systems, where the transmitted symbol is made shorter, the ISI may be caused by several previous symbols [6]. The other problem, ICI, is due to leakage between the different subchannels and is only a problem for multicarrier systems. The crosstalk will make the subchannels to be non-orthogonal to each other [5]. In order to remove the ISI, a guard interval can be used between two OFDM symbols, which means that there is a small delay between adjacent symbols. Unfortunately this quite simple solution does not remove the ICI. To remove both the ISI and ICI in the same operation, the guard interval is replaced by a cyclic prefix [1]. The cyclic prefix is a special kind of guard interval between two OFDM symbols. It is an exact replica of the last part of the symbol, which is copied and placed in the beginning of the symbol. This is illustrated in figure 3.3 where the figure at the top (a) shows transmitted signal with a guard interval and the next figure (b) introduces the cyclic prefix. Guard time filled with a silent period Symbol 1 Symbol 2 (a) Symbol 3 Guad time filled with a cyclic prefix Symbol 1 Symbol 2 Symbol 3 Figure 3.3: A silent interval between two symbols (a) can be used to avoid ISI. In order to remove both ICI and ISI is the cyclic prefix introduced (b). The duration of the cyclic prefix depends on the time differences between the first and last arrived signal in the multipath channel, i.e. the length of the channel impulse response. The cyclic prefix must be at least longer than this time difference in order to avoid ISI and ICI, or in other words; the duration of the cyclic prefix must be longer than the channel impulse response [19]. In the previous paragraphs it has become obvious that the use of a cyclic prefix is a very important component in OFDM systems. As so often, good things also bring some bad sides with them. When a guard interval is introduced the efficiency of the system decreases as new data has to wait for the redundant cyclic prefix to be sent. Therefore, it is of interest to minimize the length of the guard time/ cyclic prefix. This can be done if the last arriving signal paths contains little energy compared to the other paths, which often is the case in real systems [1], as well as in the one used here Channel estimation in OFDM based systems (b) Regarding the OFDM system, it is not only bandwidth efficient 3 but also very robust against frequency selective channels. This is due to the fact that each subchannel has a small bandwidth compared to the total channel bandwidth. Hence each subchannel will face a channel that can be seen as flat over the total 3. See figure

14 Internet Access to High Speed Trains using the IEEE 82.11a System subchannel bandwidth. Frequency Selective Channel Almost Flat Fading Channel Subchannel Bandwidth Total Bandwidth Figure 3.4: Orthogonal subchannels with a narrow bandwidth relative the frequency selective channel, such that each subchannel is affected by a flat fading channel. (Copyright Maxime Flament, 22 [4]). As each subchannel is affected by a almost flat channel, as in figure 3.4, is of great importance. This property can be used when the receiver begins to compensate for the effects that the physical channel does to the transmitted signal. To explain how this can be exploited an important mathematical operator, the convolution, must be explained. The operator is used to calculate how the multipath channel affects the transmitted signal. The convolutional operator for continuous time is defined as [9] f x( t) h( t) = x( u)h( t u) du (3.1) where the sign denotes the convolution operator. Recall the well-known fact that a convolution in time is equivalent to a pure multiplication in the frequency domain [7], which can be written as y( t) = x( t) h( t) X ( f )H ( f ) = Y ( f ) (3.2) where y( t) and x( t) represent the received and transmitted sequences, h( t) is the channel impulse response. Capitals are used to illustrate the system in the frequency domain, i.e. the Fourier transform of the corresponding time domain functions represented by small letters. The convolution is a complex operation and it can be understood that the use of multiplications instead must be more trivial and less computational complex to use when a system is studied. This is also the case in the second part of equation 3.2, where the frequency properties of the system is studied. It is well know how multiplication and the reverse operation, division, shall be performed. To understand how to undo the convolution operation demands more powers of reasoning. Now back to some practical aspects to why the convolution theorem in equation 3.2 has been discussed in such detail. Convolution is introduced to facilitate for the receiver when it shall decide what the transmitter has transmitted. Assume a flat channel for each subchannel, which means that data transmitted on that subchannel is affected in exactly the same way by the physical channel as in figure 3.4. A not far-fetched truth since any OFDM system is constructed to fulfil that criterion. The OFDM system can be seen as parallel AWGN 4 channels facing different attenuations. Thus, the transmitted signal can easily be reconstructed by a simple operation as X k ( f ) = Y k ( f ) H k ( f ) (3.3) where X k ( f ) is an estimation of the transmitted signal, Y k ( f ) is the received signal and H k ( f ) is the impact of the channel. Note that they are expressed in the frequency domain. In equation 3.3 there is a 4. Additive White Gaussian noise. Normally distributed random numbers, with zero mean, variance one and standard deviation one. 14

15 chapter three - ofdm perfect reconstruction of the transmitted signal, assuming no noise in the system and that the frequency response of the channel may be estimated exactly. The subscript k in the equation indexes the subchannel number. The number of subchannels may vary between different systems and in the IEEE 82.11a system there are 52 subchannels. Estimation techniques will be described in detail in chapter six, where the estimation of the channel frequency response is based of known transmitted symbols called training symbols and pilots. 1.8 True channel (H) Estimated channel (H est ) Absolute value Hz x /H est Absolute value Hz x 1 7 Figure 3.5: A frequency selective channel and the one tap equalizer response for each of the subcarriers. The equalizer taps is simply the inverse of the physical channel. To roughly illustrate how the estimations can be done and how to undo the channel effects, see figure 3.5. The figure shows the frequency response of a channel together with the estimated channel value for the subcarriers (top figure). The lower figure shows the inverse of the estimated channel values, which can be said to be the equalizer response at the receiver side. The equalizer for this system simply consists of one tap per subcarrier. If the channel response is large the equalizer response becomes low and vice verse when the channel response is low. The equalization is, according to equation 3.3, done by a simple multiplication with the inverse of estimated channel frequency response value, i.e. a multiplication with a complex exponent of the form i Eq k H k ( f ) e H k ( f ) = (3.4) where the sign denotes the phase operator. The reason for the complex number is due to the fact that both the phase and the absolute value of the channel must be known. In a standard single carrier communication system, a more demanding equalizer must be used which normally increases the complexity of the receiver. As mentioned, it can hardly be a more trivial calculation then a pure multiplication as in a simple OFDM system. This is a very nice property of OFDM systems and one of the reasons to why OFDM has become popular. The complexity in an OFDM system mainly depends on the number of subchannels; these directly determine how many taps it must be in the equalizer. To construct and reconstruct a signal it is necessary to use the inverse Fourier transform and its reverse operation and those transforms also introduce a certain complexity to the system. The more subchannels that is to be used, the more computations must be performed by the Fourier transform as the inputs to the Fourier transform is proportional to the number of subchannels. The Fourier transform has been well studied throughout a long period of time and there exists several algorithms in order to make it more efficient. In this thesis the Fourier transform is computed using the fast Fourier transform algorithm (FFT), which is well suited for transforms when the number of inputs is equal to a power of two [9]. The usage of FFT will come back and be discussed when the 82.11a system is explained. 15

16 Internet Access to High Speed Trains using the IEEE 82.11a System 3.3 A mathematical description of OFDM Assume that the OFDM system operates at a carrier frequency f c, a symbol duration equal to T and the number of subcarriers are N s. Then, one OFDM symbol starting at the arbitrary time t s can be written in discrete time as N s N s i +.5 s( t) Re d i j2π f 2 exp c ( t ts ) T =, t s t t s + T N i = s 2 (3.5) s( t) =, t < t s t > t s + T where d i represents the data symbols in the frequency domain [5]. In this thesis the simulations have been made in the baseband, which leads to a representation of the signal by complex values. The equivalent baseband formula is [5] N s 1 2 N s s( t) = d i j2π i 2 exp T --- ( t t s), t s t t s + T N i = s 2 (3.6) s( t) =, t < t s t > t s + T Complex values does not exist in reality, but they are nice to use from a mathematical point of view. When a real system is used in practice, information carried by the real-part values can be transmitted on one channel and the values carried by the imaginary part on another. Exactly how this is done will be explained in chapter It has been mentioned earlier but it cannot be said to many times; the orthogonality between the subcarriers is a very important aspect in OFDM systems. Since this aspect is of great importance, it will be illustrated how orthogonality can be seen, not only in the frequency domain but also in the time domain. First the time domain aspect is considered by showing three unmodulated carriers, i.e. three pure sinusoids with different frequencies, which can be seen in figure 3.6. Normally, the subcarriers are modulated with some symbols like phase shift keying (PSK) or quadrature amplitude modulation (QAM) See Appendix A. 16

17 chapter three - ofdm f = 1*3125 Hz amplitude f = 2*3125 Hz f = 3*3125 Hz time [microseconds] amplitude time [microseconds] Figure 3.6: A time domain view that shows the orthogonality between the subcarriers. They are orthogonal due to the fact that they are pure sinusoids and they sum up to an integer number of cycles within the interval. It can be seen that the carriers are not only mathematically orthogonal to each other, they also sum up to an integer number of cycles within a certain time. It is based on this time that the Fourier transform is calculated, which is made in order to turn the signal from the time- to the frequency domain. Note that the frequency spacing between the carriers in the figure is khz. This is the same spacing as defined in the in the IEEE 82.11a standard. The definition for mathematical orthogonality between two signals is [6] T s s 1 t )s t) 2 = (3.7) where s 1 ( t) and s 2 ( t) are signals with duration T s seconds. The equation shows that it is impossible to create the first signal as a linear transformation of the second and vice versa [11]. The fact that the different carrier frequencies sum up to an integer number of cycles is not a mere accident. This leads to the fact that the subcarriers also will be orthogonal to each other, even when the frequency properties are investigated [5]. Another very important aspect regarding the integer number of cycles is that ICI will be eliminated when a cyclic prefix is added to each symbol. This thesis will not dig deeper into the mathematics in this matter. In this thesis we are satisfied with that information and will not go any further with the mathematics beyond it. The interested reader may read [5] for that kind of knowledge. The spectrum of the individual subcarriers are found in figure 3.7. If a similar definition is used to describe the orthogonality in the frequency domain as in the time domain, the concept of orthogonality in that sense will not hold. But if the figure is studied carefully, it is clear that all individual subchannels are zero at the centre frequency of each other. It is this orthogonality that is of interest for an OFDM system. 17

18 Internet Access to High Speed Trains using the IEEE 82.11a System Figure 3.7: A frequency domain point of view that shows the orthogonality between the subcarriers. Symbols are transmitted simultaneously on the subchannels. (Copyright Maxime Flament, 22 [4]). In order to transmit as much data as possible, many subchannels should be used. In most cases a system has a limited total bandwidth allocated for usage. As the number of subchannels is increased in the system, each subchannel gets a more limited bandwidth to not exceed the total bandwidth or leak to adjacent subchannels. When the subchannels becomes narrower they also get very sensitive to frequency shifts that will appear when the receiver is moving fast relative the transmitter. The motion will introduce frequency shifts due to the Doppler effect. If the Doppler frequency shift is large relatively the subchannel bandwidth, the orthogonality will be lost [4]. Such is the case in this thesis, where the receiver is placed on a train moving with a speed of up to 4 km/h. This will introduce large frequency shifts and thus considerable complicate the reception procedure and the total performance of the system will be affected to the worse. 18

19 chapter four - ieee 82.11a Chapter 4 - The IEEE 82.11a standard In the following sections the IEEE 82.11a standard a will be described together with some fundamental theory that is necessary to know in order to understand the function of the different blocks in the system. A packet transmitted by an IEEE 82.11a system does not only include pure informational data, it also contains some overhead that the receiver must utilize to synchronize to the received signal and to estimate the effects of the channel on the received signal. Otherwise, the error rate of the decoded bits will be too big. All parts in the packet will be carefully explained. 4.1 Parameters related to the system From this point, everything will be explained according to a baseband model, which means that the centre frequency of the system will be the direct current (DC) component, equal to Hz. As the signal is designed to fulfil the specifications in the IEEE 82.11a standard it will, in practical systems, be transformed from the baseband up to some higher frequency range. See figure 4.1 for an illustration. There figure (a) shows the signal spectrum in the baseband. Figure (b) shows the transformation up to a higher centre frequency named f. 2w 2w w w Frequency f w f f+w Frequency (a) (b) Figure 4.1: Transformation from the baseband (centred around DC) to a higher centre frequency. With this figure fresh in mind, hopefully it is clarified why, for example, a subchannel can be indexed with either a plus or a minus sign, which will appear later on. The positive and negative subchannel closest to Hz will be indexed +1 and -1 respectively with increasing indexes outwards. There are several parameters included with the IEEE 82.11a standard. Different data rates, modulations and coding rates can be used and the receiver must know these parameters in order to perform the decoding process. Each data rate is related to a unique combination of modulation and coding rate. If the receiver knows the data rate, it also knows what kind of coding and modulation that is used by using the information in table 4.1. Data Rate (Mbits/s) Modulation Coding rate (R) Coded bits per subcarrier (N BPSC ) Coded bits per OFDM symbol (N CBPS ) Data bits per OFDM symbol (N DBPS ) 6 BPSK 1/ BPSK 3/ QPSK 1/ QPSK 3/ QAM 1/ QAM 3/ QAM 2/ QAM 3/ Table 4.1: Rate dependent parameters in the IEEE 82.11a standard. (From IEEE 82.11a-1999 Copyright 1999 IEEE. All rights reserved [12]). Table 4.1 shows the relationship between data rate, modulation and coding rate. When the modulation 19

20 Internet Access to High Speed Trains using the IEEE 82.11a System and coding rates are known, the last three columns in table 4.1 can be easily calculated. The number of data carriers is equal to 48 and the relationships between the parameters become N CBPS = 48 N BPSC N DBPS = N CBPS R = 48 N BPSC R There are three rates that are mandatory according to the standard for the system to support and these are 6, 12 and 24 Mbits/s. However, as mentioned in the introduction, this standard can support rates up to 54 Mbits/s. Parameter Explanation Value F s Baseband sampling frequency 2 MHz N FT FFT/IFFT integration period 64 N SD Number of data subcarriers 48 N SP Number of pilot subcarriers 4 N ST Number of total subcarriers 52 (N SD +N SP ) F Subcarrier freq. spacing.3125 MHz (F s /N FT ) T FFT FFT/IFFT integration time 3.2 µs (1/F s x N FT ) T PREAMBLE PLCP Preamble duration 16 µs (T SHORT + T LONG ) T SIGNAL Duration of Signal symbol 4. µs (T GI + T FFT ) T GI Guard interval duration.8 µs (T FFT /4) T GI2 Training symbol GI duration 1.6 µs (T FFT /2) T SYM Symbol interval 4. µs (T GI + T FFT ) T SHORT T LONG Short training sequence duration Long training sequence duration Table 4.2: Timing dependent parameters in the IEEE 82.11a standard. (From IEEE 82.11a-1999 Copyright 1999 IEEE. All rights reserved [12]). Table 4.2 gives an overview of the timing related parameters. Some of the parameters will be explained in more detail later. Still it is often quite trivial to understand what they mean. Briefly, the FT period time seen in figure 3.6 is equal to 3.2 µs which corresponds to 64 sample values assuming the baseband sampling frequency of 2 MHz. A cyclic prefix of.8 µs will be added to this signal and together they form an OFDM symbol of length 4. µs. The observant reader can see that the total number of subcarriers is equal to 52 but there are 64 inputs to the IFFT/FFT. This is not a typing mistake and it was carefully considered when the parameters were specified. The reasons to this are the properties of the frequency spectrum and will be explained in chapter Main fields of the IEEE 82.11a frame 8. µs (1xT FFT /4) 8. µs (T GI2 + 2 x T FFT ) An IEEE 82.11a frame can be divided into three main fields, preamble, signal and data shown in figure 4.2. In order to describe the frame in more details, the three main blocks are split into smaller parts. This will be explained in the next coming chapters. 2

21 chapter four - ieee 82.11a Figure 4.2: A frame consists of three main blocks; Preamble (training symbols), Signal (information about the rate) and Data. (From IEEE 82.11a-1999 Copyright 1999 IEEE. All rights reserved [12]). Roughly, the frame can be said to consist of one part (preamble) that provides the receiver knowledge of the physical channel, a second part (signal) that tells the receiver how the data should be decoded and a last part, which is the actual data and represents the most interesting part for the user Preamble Every frame starts with the preamble field. This consists of ten identical symbols, called the short training sequence, followed by two repetitions of a long training symbol. All together this makes 12 training symbols. The short training symbols are used for signal detection, performing automatic gain control (AGC) and coarse frequency offset estimation. For a signal that is repetitive of length T, it is possible to perform a coarse frequency offset of 1 ( 2T ) Hz. As the symbol time decreases, the maximum measurable frequency offset increases. Since each short training symbol is 8/1 µs, a frequency offset up to 625 khz can be estimated [5]. There are also some other reasons to why the symbols have been made short. When performing signal detection, it is convenient to correlate one short training symbol with the next following symbol. The more symbols there are, the more correlation calculations can be made. The short training sequence is transmitted on 12 of 52 subcarriers, according to the defined as S 26, 26 vector S 26, 26 = 13 6 (,, 1 + j,,,, 1 j,,,, 1 + j,,,, 1 j,,,, 1, j,,,, 1 + j,,,,,,,, 1 j,,,, 1 j,,,, 1 + j,,,, 1 + j,,,, 1 + j,,,, 1 + j,, ) (4.1) The indexes are defined according to the baseband subcarrier number and the content of the vector is the actual training symbols to be sent. The lower the index is, the lower will the centre frequency of the subchannel be 6. To transmit the short training sequence, an inverse Fourier transform of size 64 on the S 26, 26 vector is first made. Then it is periodically extended to a length of 8. µs, and the final result is 1 identical short training symbols of length 8 ns each. The two long training symbols are unlike the short training sequence transmitted on all of the 52 subcarriers. These are mainly used for channel estimation. Since these symbols have a longer period than the short ones, also fine frequency offset estimation can be made. How to perform the channel estimation may vary between different implementations, but some useful algorithms will be suggested later on in this paper (chapter 6). 6. See figure

22 Internet Access to High Speed Trains using the IEEE 82.11a System The long training sequence is constructed according to the L 26, 26 vector defined as L 26, 26 = ( 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) (4.2) To transmit this kind of training symbol, the IFFT operator is applied on the vector found in equation 4.2 generating one long training symbol of length 3.2 µs. The first long training symbol is periodically extended to a length of 8. µs, where the first 2.8 µs represent the cyclic prefix and the rest µs represent the first and second long training symbol. Two identical symbols are used in order to obtain well trusted references (amplitudes and phases) for the detection. By taking the average value of the two received long training symbols, the impact of the noise can be reduced with a factor of a 3 db [5]. The total training structure in the time domain is shown in figure 4.3. Figure 4.3: The training sequence starts with ten identical short symbols and two long training symbols follow them. The short and long training symbols are mainly used for signal detection and channel estimation, respectively. (From IEEE 82.11a-1999 Copyright 1999 IEEE. All rights reserved [12]). The training symbols are not the only sequences that are known to the receiver. Also symbols called pilots are transmitted during the packet to estimate the channel variations that may occur during the packet transmission Pilots As mentioned earlier, each packet starts with known training symbols called short and long training symbols and these are only transmitted in the beginning of a packet. Pilot symbols, on the other hand, are transmitted during the whole frame. Pilots are transmitted directly after the long training symbols, simultaneously with the signal field. Recall that information is transmitted simultaneously over several, but not all, subchannels. The unused subchannels are dedicated to the pilots. The pilots are transmitted on four subcarriers according to table 4.2. The main reason for the pilots is to estimate fine frequency offsets that appears after the training symbols [12]. The frequency offset is supposed to affect the subcarriers close to the pilot subcarriers in a similar way and therefore only four subcarriers are used [5]. Keep in mind that the system originally was intended for indoors office use, and a larger coherence bandwidth was expected than in the environment considered here. Of course, the pilots may be used to estimate the channel as well if some good interpolation technique is used. Interpolation must be used in order to estimate the effect on the non-pilot subcarriers, where the actual data is transmitted. This is done in some estimation strategies in this thesis and these will be explained in chapter 6. The pilots are modulated and constructed in the Fourier domain according to the two vectors and p v, where P 26, 26 is defined as P 26, Refer to [12] for further information. 22

23 chapter four - ieee 82.11a P 26, 26 = (,,,,, 1,,,,,,,,,,,,,, 1,,,,,,,,,,,,,, 1,,,,,,,,,,,,,, 1,,,,, ) where the indexes of P 26, 26 represent different subcarriers. The non-zero elements are the BPSK modulated symbols at the four pilot subcarriers. The p..126v vector gives the sign of the pilot symbols, which means that the vector only consists of ±1. To create the first four pilot symbols, simply multiply P 26, 26 with the first number in p..126v and the same procedure for the next four pilots and so on. The pilots are then inserted at the pilot subcarriers and together with the inputs on the data subcarriers an IFFT translates the signal from frequency- to the time-domain. In figure 4.4 all training sequences can be viewed in a time and frequency diagram, where known symbols are filled. First comes the short training sequence transmitted on twelve subcarriers, followed by the long training sequence that is transmitted on all of the 52 data subcarriers and finally the pilots transmitted on the four pilot subcarriers during the rest of the packet. Frequency (baseband) (4.3) Carrier 26 to 1 Carrier 1 to 26 Figure 4.4: A time and frequency plot over an IEEE 82.11a frame. Known tones, i.e. training- and pilot symbols are filled (blue). First, short training symbols that are transmitted on 12 subcarriers, then the long training symbols on all subcarriers and finally the four pilot tones. When known tones are transmitted at all of the subcarriers, it is often referred in the literature as block type pilot transmission. The second case is called comb type pilot transmission and uses only a few subcarriers acts as pilot tones. The main reason for using block type pilots is that the channel is assumed to be constant during the whole packet and it is enough to do one channel estimation and assume that it is accurate for the rest of the packet. If the channel does change during the transmission it is necessary to track the variations. This can be done by using comb type pilots on some subcarriers and interpolate between them to find the total channel changes on all subcarriers [13]. There are a lot of research papers regarding optimal pilot placement, one is [2]. In this thesis the pilot pattern has not been modified. The intension was to keep it as specified in the standard [12] Signal Time Several smaller blocks, the rate, reserved, length, parity and tail block, build up the signal field. This field is needed to give the receiver knowledge of the modulation scheme and coding rate that is used in the remaining part of the packet. Thus, it is very important that this field is decoded correctly at the receiver. The signal field is hereby transmitted with the lowest rate; BPSK modulation and rate 1/2 convolutional 23

24 Internet Access to High Speed Trains using the IEEE 82.11a System encoding. Rate (4 bits): Contains information about the data rate. Reserved (1 bit): Reserved for future use. Length (12 bits): Number of octets in PSDU 8 (maximum = 495 octets). Parity (1 bit): Positive parity. Tail (6 bits): In order to return the convolutional encoder to the zero state. That is a total of 24 bits. With rate 1/2 convolutional encoding and BPSK modulation these 24 bits becomes 48 BPSK channel symbols which are transmitted on the 48 data carriers. Accordingly, the signal field is equal to one OFDM symbol of length 4. µs, including the cyclic prefix. Convolutional encoding is an often-used technique to decrease the number of bit errors at the receiver. The encoder will be explained in more detail when the data part is described Data Like the previous preamble and signal field, the data field does also consist of different parts, the service, PSDU, tail and pad bits blocks. The rate and the length parts in the signal field determine the coding rate and the number of octets in this field. Service (16 bits): Scrambler synchronization, and reserved bits for future work. PSDU (variable): The coded data bits. Tail (6 bits): Same as for the signal field. Pad bits 9 (variable): To make the data field an integer number of N CBPS. 1 When the uncoded data bits in the data field are generated, they go through the following operations at the transmitter. Scrambling. Convolutional encoding. Interleaving. Mapping from coded bits into symbols. These steps, called the baseband modulation, will now be explained in detail. The inverse operations are then done at the receiver Scrambler It has been described that an OFDM signal is a summation of several modulated signals that are generated by using the IFT. When these signals sum up it can lead to a high peak-to-average power (PAP) ratio [5]. There are some practical disadvantages with a high PAP ratio. One main drawback is the reduced efficiency of the power amplifier when the signal is transmitted. In any practical transmitter there are unwanted non-linearities in the amplifier. OFDM systems suffer from high probabilities of a high PAP which makes this problem especially important to address [14]. The more money that is spend on the amplifier, the better the amplifier gets, but the PAP problem will remain. Scrambling is one solution to reduce the probabilities of high PAP ratios and a scrambler is also incorporated into the 82.11a system [5]. Each sequence is scrambled according to a predefined generator polynomial S( x) = x 7 + x (4.4) which is a modulo-2 summation of the data bits (zeroes and ones) in a certain order. How to perform the summation gets clearer if figure 4.5 is studied. Before transmitting, the initial state of the scrambler will be set to a pseudo random state (non-zero) based on the bits in the Service field. 8. Physical Sublayer service Data Units 9. How to calculate the number of pad bits, see Appendix B 1. See table

25 chapter four - ieee 82.11a In this work, the scrambler is cancelled because random bits are transmitted in order to compensate for the absent of a scrambler. This can be done since the scrambler also introduces some pseudo- random pattern to the transmitted sequence. Of course, if a system is constructed in hardware and for real usage together with other equipment, the scrambler must be included in order to fulfil the specifications of the standard. Figure 4.5: The scrambler defined in the standard. It introduces a pseudo-random pattern to the transmitted data stream. (From IEEE 82.11a-1999 Copyright 1999 IEEE. All rights reserved [12]). The content of the registers is shifted one step to the left for every new incoming data bit Convolutional encoder By using a convolutional encoder, a better bit-error performance can be obtained. This type of code introduces memory, or redundancy, in the transmitted bit sequence [15]. The introduction of the memory has the effect that the coded bits depend on some of the previous transmitted coded bits. The convolutional encoder can be described with just a few properties. These are for the specific encoder used in this thesis defined as Constraint length: L = 7, 6 memory states + 1 input that is the uncoded bit. Generator polynomials: g = = , g 1 = = Coding rate: R = 1 2 (number of input bits k/n output bits). These three properties yields a convolutional encoder according to figure 4.6. Figure 4.6: The convolutional encoder defined in the standard, coding rate 1/2. Its outputs (coded bits) are dependent on the content of the memory states and the input bit that is to be coded. (From IEEE 82.11a-1999 Copyright 1999 IEEE. All rights reserved [12]). When a new uncoded bit has entered the encoder, the content of the memories will consequently be shifted one step to the right. The constraint length is defined as the number of memory states in the encoder plus the incoming bit. This definition may vary in different literatures, but this is probably the most common way to define it. Next, the generator polynomials tells which of the memory contents that shall be summed up (modulo-2 summation) in order to form the output of the corresponding polynomial. A generator polynomial defined as, for example, means that the summation is made by using the first incoming bit, together with the content of the first and last memory box. From figure 25