TALLINNA TEHNIKAÜLIKOOL. Raadio-ja sidetehnika instituut. Mikrolainetehnika õppetool. Referaat aines. Uurimisteemakeskne individuaalõpe IXX9530

Transcription

1 TALLINNA TEHNIKAÜLIKOOL Raadio-ja sidetehnika instituut Mikrolainetehnika õppetool Referaat aines Uurimisteemakeskne individuaalõpe IXX9530 OFDM: advantages, drawbacks, and performance improvement methods Doktorant: Aleksander Brizmer, IAQDD Juhendaja: prof. Andres Taklaja TALLINN 2006

2 EESSÕNA Käesolevas referaadis kirjeldatakse mitmekandjalist modulatsioonimeetodit OFDM (Orthogonal Frequency Division Multiplexing). See meetod eeldab anmete edastust üle paralleelseid kitsaribalisi alam-kandjaid (sub-carriers), mis on ortogonaalsed üks teise suhtes ning paiknevad üks teisest lähedal. Alam-kandjad on genereeritud Fourier i pöördteisendust realiseerivate signaalide digitaltöötluselementide abil. OFDM on tänapäeval kasutamisele võetud traadita (wireless) digitaalsetes ringhäälingu süsteemides (DAB, DVB-T, DVB-H) ning traadita kohtvõrkude (wireless local area network, WLAN) uuemates standardides (HiperLAN2, IEEE802.11a, IEEE802.16). OFDM i vaadeldakse kui potentsiaalset modulatsioonitehnoloogiat tuleviku neljanda põlvkonna (4G) mobiilsidesüsteemide jaoks ning selletõttu on ta saanud paljude uurimistööde huviobjektiks. Peamised teemad, mis on käsitletud selles referaadis, on: OFDM meetodi toimimise põhimõtted; meetodi eelised- spektraalselt effektiivne, signaali mitmekiirelisele levile ja interferentsile vastupidav meetod; meehodi puudused- OFDM signaali suur tipp-ja keskmise võimsuse suhe (peak-to-average power ratio, PAPR), RF generaatori faasimüra mõju, raadiosidekanali muutuv kvaliteet, millest on tingitud sumbumine terve kandja ulatuses (flat fading); puuduste mõju minimeerimiseks kasutatavad meetodid- PAPR minimeerimismeetodid, generaatori faasimüra kompenseerimisskeemid, adaptiivsed modulatsiooniskeemi vahetamise ja kanalite väljajagamise algoritmid. Eraldi on antud OFDM meetodi tänapäeva praktiliste realisatsioonide näited. Kokkuvõttete osas on pakkutud võimalikud uurimistöö suunad, mis on seotud käesolevas referaadis käsitletud teemadega. Referaat on koostatud inglise keeles; selle koostamisel kasutatud allikad on suuremas osas IEEE publikatsioonid. Viided internetist kättesaadavatele publikatsioonidele on esitatud kasutatud allikate osas. 2

3 CONTENTS 1. INTRODUCTION RADIO PROPAGATION EFFECTS Path loss and attenuation Shadowing and slow fading Multi-path effects Short term fading Freqency selective fading PERFORMANCE AND QOS INDICATORS QoS concepts Physical level performance indicators CONTRIBUTIONS TO FURTHER CHAPTERS OFDM BASICS CONCEPTS OF ORTHOGONALITY OFDM TRANSMISSION AND RECEPTION Serial to parallel conversion Sub-carrier modulation Inverse Fast Fourier Transform Parallel to serial conversion and guard interval insertion RF Modulation EFFECT OF NOISE AND FADING ON OFDM Effect of noise on OFDM Effect of flat fading on OFDM COFDM OFDM APPLICATION NOWADAYS Digital audio broadcasting Digital video broadcasting Wireless networks (HiperLAN2 and IEEE802.11a) OFDM MAJOR DRAWBACKS AND METHODS OF THEIR COMPENSATION PEAK-TO-AVERAGE POWER RATIO Power amplifier linearization Predistortion Feedforward Other PA linearization methods PAPR reduction methods Clipping and filtering Peak windowing Phase optimization techniques Tone reservation Pulse shaping Companding schemes Reference signals subtraction Carrier interferometry codes Tone injection PHASE NOISE IMPACT AND COMPENSATION SCHEMES Phase noise: the background Phase noise compensation schemes ADAPTIVE ALGORITHMS Adaptive modulation Adaptive user allocation Existing channel allocation methods Proposed adaptive algorithms EXAMPLES OF OFDM PRACTICAL IMPLEMENTATION

4 4.1 DVB-T REALIZATIONS WLAN REALIZATIONS REALIZATIONS FOR MULTIPLE PURPOSES CONCLUSIONS REFERENCES

5 1. INTRODUCTION The rapid development of wireless communications and the trends of the integration of various communication services (such as voice, video, audio) into one system are the moving forces towards next generation of mobile digital communication systems. Subscriber mobilization, limited bandwidth and radio propagation effects (such as multi-path propagation, fading, attenuation and interference) are the factors that form the requirements for future integrated services systems: high speed, spectral efficiency, power efficiency and robustness. There are several approaches that can be used to create the communication system that could be able to match the requirements above. One of these approaches is to use multi-carrier modulation method- Orthogonal Frequency Division Multiplexing (OFDM). In case of OFDM the digital information is transmitted via an ensemble of closely spaced narrow-band sub-carriers that are mutually orthogonal. The generation of these sub-carriers is done by means of digital signal processing (DSP) algorithms that perform fast Fourier transform (FFT) operations. Nowadays OFDM is used in digital audio broadcasting (DAB), digital video broadcasting (DVB-T, DVB-H) and newer WLAN standards (IEEE802.11a and HiperLAN2). In future it is suggested to be used in the 4 th generation (4G) of mobile communications systems. The 4G systems are proposed to be broadband communication systems that offer high-speed and high-quality voice, video, audio and data services to mobile subscribers. The spectral efficiency of recently launched 3G networks is too low to support high data rate services at low cost, so that one of the main focuses of 4G systems will be to significantly improve the spectral efficiency [8]. That is where OFDM can be useful. This paper describes the principle and the main problems of OFDM transmissions, as well as proposed methods to overcome these problems. For better understanding of those factors that have their negative impact on OFDM performance, a general description of radio propagation effects (see paragraph 1.1) and the essential performance and quality of service parameters (see paragraph 1.2) are given in this chapter. Radio propagation effects Radio propagation effects, such as path loss, frequency selective fading, multi-path delay spread and Doppler spread limit the performance of wireless communication systems. In this paragraph a brief description of these effects is given for better understanding of their negative impact on OFDM performance. Path loss and attenuation During propagation, radio signals weaken with distance. This is due to the wave front of the radio signal expanding and thus reducing in power density. In free space, the propagating wave expands as a sphere and thus the power density reduces in proportion to the surface area of this sphere. If the signal is transmitted using a directional antenna, the signal still expands as a sphere, except that the energy density is concentrated to one or more areas. The area of a sphere is proportional to the radius squared, and thus in free space the RF field strength reduces proportionally with distance squared. P R = P G T T 2 λ GR 4πR 0-1 5

6 In equation 1-1 P R is the received signal power (watts), P T is the transmitted power (watts), G R is the gain of the receive antenna with respect to an isotropic antenna, G T gain of the transmitter antenna, λ wavelength of the RF carrier in meters, and R is the transmission distance in meters. This equation works with so-called line of sight (LOS) link, assuming that there are no obstacles between transmitter and receiver. In case of non-los link the power of distance-dependent λ component is increased to a higher, propagation environment-dependent value. This 4πR value is expressed in equations as α and referred as path loss exponent. Free space propagation is very predictable, and can be used to accurately model satellite communications and directional links with no obstructions, such as short-range microwave directional point-to-point links. However for most terrestrial communications such as mobile phones and wireless LAN systems, the environment is much more complex making propagation modeling much more difficult. Shadowing and slow fading In most mobile radio applications, the environment through which the communication must occur is cluttered and has many obstructions such as buildings, hills, trees, walls, etc. These objects cause reflections off their surface and attenuation of signals passing through them, resulting in shadowing. These shadows can result in large areas with high path loss, causing problems with communication. The amount of shadowing is dependent on the size of the obstructing objects, the structure of the material, and the frequency of the RF signal. Most materials are highly transparency at RF frequencies compared with visible light, making non-line Of Sight (LOS) propagation possible. Although many materials are transparent, metal objects act much like mirrors due to being good reflectors, making them opaque, resulting in wiring within walls, foil insulation, metal roofing, reinforcing in concrete, etc being opaque. In addition, large objects such as buildings and hills absorb much of the RF energy passing through, resulting in deep shadowing behind them. Under these conditions most of the received energy usually comes from reflections and diffraction around the object, rather than from the direct path. Diffraction occurs at edges of obstructing objects in the transmission path. At the diffraction edge, the signal reradiates as a spherical wave front originating from the diffracting edge. This allows it to bend partially around the object. Moving the receiver, transmitter, or objects in the environment will result in a change in the path loss due to the transmission path changing. This variation in the path loss occurs over large distances (typically wavelengths [8]) and is dependent on the size of the objects causing the shadowing, rather than the wavelength of the RF signal. Due to the slow changing nature of this variation, it is commonly referred to as slow fading. Multi-path effects In a radio link, the RF signal from the transmitter may be reflected of different objects (hills, trees, walls, etc). Some of these reflections will arrive at the receiver, effectively creating multiple transmission paths, commonly referred to as a multi-path environment [8]. The radio signal travels over a different distance for each of these paths, and thus takes a different amount of propagation time. If we were to transmit an RF pulse in a multi-path environment we would receive a signal like the one shown in Figure 1-1 (a). Each impulse corresponds to one path, with the strength of each impulse dependent on the path loss for that path. For a fixed frequency signal, (i.e. a sine wave) the propagation delay results in a phase rotation of the signal. Each of the multi-path signals will have a different propagation distance and thus a different phase rotation. These signals add at the receiver, resulting in constructive or destructive interference. Each of the multi-path signals can be represented as a phasor, which has vector 6

7 length corresponding to signal power and the angle corresponding to the phase. The received signal corresponds to the vector sum of the multi-path phasors (see Figure 1-1 (b)). Destructive interference occurs when the vector sum adds to zero. This is also referred to as a null. Constructive interference occurs when all the signals have a similar phase, reinforcing each other. Figure 0-1 Impulse response and phasor plot for multi-path channel [8] The measure of the spread in signal arrival time is referred as delay spread. It is a measure of the time dispersion of a channel, and is very important in determining how fast the symbol rate can be in digital communications. Delay spread results in time blurring, where energy from previous data symbols becomes mixed in with current symbols. This causes interference, known as Inter- Symbol Interference (ISI), because previous symbols are uncorrelated, effectively adding noise to the signal. More details about delay spread are given in [3]. There is other type of spread related to multi-path effects- Doppler spread. It is caused by rapid changes in the channel response due to movement of the receiver through a multi-path environment. It results in random frequency modulation of carrier(s), leading to signal degradation. The amount of Doppler spread is proportional to the transmission frequency and the velocity of movement. In case of OFDM transmission, the closer the sub-carriers are spaced together, the more susceptible the OFDM signal is to Doppler spread [8]. Short term fading In a multi-path environment, the received signal fades with distance due to the changing phase of the multi-path components. Short term fading is caused by the interference (constructive or destructive) that result from the combination of multiple received waves. As the receiver or transmitter are moved in space the relative phase between the different multi-path components change, causing the interference to also change, resulting in fades in the received signal power. At certain locations, the signal can suffer almost compete cancellation of the signal, resulting in a deep null in the signal. The rate of fading with distance is usually measured using the coherence distance. This is a measure of the distance over which the radio channel experiences comparable or correlated fading [8]. Freqency selective fading Multi-path also causes fading changes with frequency. This is due to the phase response of the multi-path components varying with frequency. The received phase, relative to the transmitter, of a multi-path component corresponds to the number of wavelengths the signal has traveled from the transmitter. The wavelength is inversely proportional to frequency and so for a fixed transmission path the phase will change with frequency. The path distances of each of the multipath component is different and so results in a different phase change. 7

8 Figure 0-2 Frequency selective fading for a short indoor link (measurement results) [8] Figure 1-2 shows an example of measured frequency selective fading within an indoor environment. The signal power varies by more than 25 db with frequency, showing that at certain frequencies near complete signal cancellation is occurring [8]. The frequency selective fading characteristics of a channel can be summarized by the correlation bandwidth of the channel. This is the approximate maximum bandwidth or frequency interval over which the fading is similar and correlated. The exact correlation bandwidth depends on the required level of correlation. The correlation bandwidth is inversely proportional to the channel delay spread. Frequency selective fading can be also viewed in conjunction with short term fading. This is applicable in case when receiver is moving in space. In [8] this is referred as space selective fading. Performance and QoS indicators The effects listed in the previous paragraph have negative impact on the performance and quality of service (QoS) of a wireless system. For better understanding of the main concepts of QoS and physical level performance indicators a brief review is given in the present paragraph. QoS concepts Quality of service (QoS) is defined in the ITU-T recommendation E.800 as the collective effect of service performance, which determines the degree of satisfaction of a user for this service [43]. There are four viewpoints of the QoS: customers QoS requirements / perception, and service provider QoS offering / achievement. Customer requirements are important when creating a QoS test plan to be used when estimating the QoS delivered by a service provider. These requirements are usually expressed in non-technical language and focus on user-perceived effects, rather than their causes within the network. The delivered QoS is expressed in values assigned to QoS indicators (also referred as performance indicators), which are used for tracking performance and directing optimization [44]. In [45], a three-layer model of the main QoS aspects is presented. The first layer is the Network Access, the basic requirement for all the other QoS aspects and QoS parameters. The outcome of this layer is the QoS parameter Network Accessibility. The second layer contains the other three QoS aspects: service accessibility, availability and integrity. A more detailed description of these aspects can be also found in [45]. The different services are located in the third layer. Their outcomes are the QoS parameters. They vary from service to service. From the other hand, there are some parameters that might not be always included in the QoS parameter set of one or another service, but they determine the physical level performance and, thus, have a significant 8

9 impact on the resultant QoS. In the next section three important parameters- signal-to-noise ratio (SNR), bit error rate (BER) and modulation error ratio (MER) will be briefly described. Physical level performance indicators In [46] signal-to-noise ratio (SNR) is defined as the power ratio between a signal and noise. Usually it is expressed in terms of the logarithmic decibel scale. SNR is usually taken to indicate an average signal to noise ratio, as it is possible that (near) instantaneous signal to noise ratios will be considerably different. SNR is the parameter that sets limitations for throughput and spectral efficiency (see 2.3), it also defines bit error rate (BER) in case of digital transmission. SNR is a common indicator of radio channel quality. In [47] bit error rate (BER) is defined as is the ratio of the number of bits incorrectly received to the total number of bits sent during a specified time interval. Examples of bit error ratio are transmission BER, i.e., the number of erroneous bits received divided by the total number of bits transmitted; and information BER, i.e., the number of erroneous decoded (corrected) bits divided by the total number of decoded (corrected) bits. Modulation error ratio (MER) is a quality indicator of a digitally-modulated signal. In case of digital modulation a symbol is formed using in-phase (I) and quadrature (Q) components [2]. The amount of bits in symbol is different for different modulation schemes. For example, a 16-QAM symbol consists of 4 bits and binary phase shift keying (BPSK) symbol consists of one bit only. There are certain I and Q axis values that correspond to each symbol; they form an I-Q constellation. Each location on the constellation is framed by decision boundaries. If the signal falls within these boundaries, the correct data will be received. If because of channel impairments (noise, interference) it falls in an adjacent area the data will be in error [50]. MER indicates the mean or the maximum deviation of the I and Q values from ideal signal states and thus provides a measure of signal quality. The following diagram, derived from [48], shows the vectors used for calculating the modulation error in case of quadrature amplitude modulation (QAM): Figure 0-3 Modulation error vector calculation [48] In case, illustrated in Figure 1-3, MER is a measure of the ratio of the error power to the average power in an ideal QAM signal. MER can be expressed in terms of either linear (%) or logarithmic (decibel) scale. The logarithmic scale expression is given in the equation below. 9

10 ( I j = 1 ( db) 10log { } MER = N 2 2 ( ΔI j + ΔQ j ) j = In this equation I j and Q j are the values of I and Q axis that represent the ideal center points on the I-Q constellation for j-th symbol received. Δ I j and ΔQ j correspond to the difference between the ideal center point and the actual status point for j-th symbol received. The last two parameters may also be expressed by means of error vector [49]. The further the received signal point is from the ideal locations, the poorer the MER. Poor MER is an early indicator of channel impairment [51]. N 10 Contributions to further chapters 2 j + Q This paper is organized as follows. The second chapter describes OFDM basics- the principle of orthogonality, OFDM transmission mechanisms, the effects of additive noise and flat fading on OFDM performance and the limitations on modulation schemes. A short overview of an improvement of OFDM- coded OFDM (COFDM), with the main ideas of forward error correction (FEC) coding and interleaving is given in the second chapter, as well as a brief description of wireless communication systems that use OFDM. The third chapter describes the major drawbacks of OFDM (peak-to-average power ratio, phase noise, flat fading), as well as several proposed solutions to overcome these drawbacks (PAPR reduction methods, PHN compensation schemes, adaptive modulation and user allocation algorithms). In the fourth chapter some examples of OFDM practical implementation in DVB-T and WLAN systems, as well as some multi-purpose OFDM realizations will be shown. The fifth chapter brings conclusions and subjects for future studies and research. 2 j ) 10

11 2. OFDM BASICS In this chapter the basic principles of Orthogonal Frequency Division Multiplexing (OFDM) will be described. OFDM is a special form of multi-carrier modulation that uses digital signal processing (DSP) algorithms to generate waveform that are mutually orthogonal [7]. OFDM uses the principles of conventional frequency division multiplexing (FDM) to allow multiple messages to be sent over a single radio channel, but in a much more controlled manner, allowing an improved spectral efficiency. The information is transmitted using an OFDM ensemble that is made up from a dense packing of many sub-carriers [8]. All the sub-carriers within the OFDM signal are time and frequency synchronized to each other, allowing the interference between sub-carriers to be carefully controlled. These multiple sub-carriers overlap in the frequency domain, but do not cause Inter-Carrier Interference (ICI) due to the orthogonal nature of the modulation. Typically with FDM the transmission signals need to have a large frequency guard-band between channels to prevent interference. This lowers the overall spectral efficiency. However with OFDM the orthogonal packing of the sub-carriers greatly reduces this guard band, improving the spectral efficiency [8]. The development of the concept of using parallel data transmission by means of FDM was started in 1950s. A U.S. patent was filled and issued in January 1970 [7]. The idea was to use parallel data streams and FDM with overlapping sub channels to avoid the use of high-speed equalization and to combat impulsive noise, and multi-path distortion as well as to fully use the available bandwidth. The initial applications were in the military communications. After more than thirty years of research and development carried out in different places, OFDM is now being widely implemented in high-speed digital communications. Due to the recent advancements in digital signal processing (DSP) and very large-scale integrated circuits (VLSI) technologies, the initial obstacles of OFDM implementations do not exist any more. Meanwhile, the use of fast Fourier transform (FFT) algorithms eliminates arrays of sinusoidal generators and coherent demodulation required in parallel data systems and makes the implementation of the technology cost effective. In recent years OFDM has gained a lot of interest in diverse digital communication applications [7]. The first paragraph of this chapter describes the principle of orthogonality, which is the mathematical base of OFDM spectral efficiency. In the second paragraph the mechanism of OFDM transmission is described. The third paragraph shortly describes the impact of additive noise and flat fading on OFDM performance and shows the limitations on the modulation schemes in OFDM transmission. An improvement of OFDM- Coded Orthogonal Frequency Division Multiplexing (COFDM) is briefly described in the fourth paragraph. The fifth paragraph gives a general overview of those wireless communication systems that use OFDM nowadays. 2.1 Concepts of orthogonality Orthogonal Frequency Division Multiplexing (OFDM) is simply defined as a form of multi-carrier modulation where the carrier spacing is carefully selected so that each sub carrier is orthogonal to the other sub carriers. Two signals are orthogonal if their dot product is zero [7]. That is, if you take two signals multiply them together and if their integral over an interval is zero, then two signals are orthogonal in that interval. For a set of functions the condition of orthogonality is as it is defined in the equation below [8]. t s () t s () t C dt = i j 0 0 i = i j j

12 In case of OFDM, a set of orthogonal sinusoids is used as an orthogonal basis. Equation (2-2) [8] shows a set of orthogonal sinusoids, which represent the sub-carriers for an unmodulated real OFDM signal. In this equation f0 is the carrier spacing, M is the number of carriers and T is the symbol period. s k () t ( 2πkf t) sin o 0 < t < T k = 1,2,... M = 0 otherwise 2-2 Orthogonality can be achieved by carefully selecting carrier spacing f 0, such as letting the carrier spacing be equal to the reciprocal of the useful symbol period. As the sub carriers are orthogonal, the spectrum of each carrier has a null at the center frequency of each of the other carriers in the system. This results in no interference between the carriers, allowing them to be spaced as close as theoretically possible [7]. An example of time domain representation of an OFDM signal is given in Figure 2-1. In the chart (4a) the result summation of the sub-carriers above is shown. Figure 2-1 Time domain view of an OFDM signal [8] The orthogonal nature of the transmission is a result of the peak of each sub-carrier corresponding to the nulls of all other sub-carriers. When this signal is detected using a Discrete Fourier Transform (DFT) it has discrete samples, shown as o -s in the Figure 2-2. If the DFT is time synchronized, the frequency samples of the DFT correspond to just the peaks of the subcarriers, thus the overlapping frequency region between sub-carriers does not affect the receiver. The measured peaks correspond to the nulls for all other sub-carriers, resulting in orthogonality between the sub-carriers [8]. Figure 2-2 Frequency domain view of the sub-carriers in a 5-tone OFDM signal [8] 12

13 2.2 OFDM transmission and reception A block diagram of OFDM system model is shown in Figure 2-3. Figure 2-3 OFDM system model (derived from [7]) OFDM signals are typically generated digitally due to the difficulty in creating large banks of phase lock oscillators and receivers in the analog domain [8]. The transmitter section converts digital data to be transmitted, into a mapping of sub-carrier amplitude and phase. It then transforms this spectral representation of the data into the time domain using an Inverse Discrete Fourier Transform (IDFT). The Inverse Fast Fourier Transform (IFFT) performs the same operations as an IDFT, except that it is much more computationally efficiency, and so is used in all practical systems. In order to transmit the OFDM signal the calculated time domain signal is then mixed up to the required frequency. The receiver performs the reverse operation of the transmitter, mixing the RF signal to base band for processing, then using a Fast Fourier Transform (FFT) to analyze the signal in the frequency domain. The amplitude and phase of the sub-carriers is then picked out and converted back to digital data. The IFFT and the FFT are complementary functions and the most appropriate term depends on whether the signal is being received or generated. In cases where the signal is independent of this distinction then the term FFT and IFFT is used interchangeably Serial to parallel conversion Data to be transmitted is typically in the form of a serial data stream. A serial to parallel conversion stage is needed to convert the input serial bit stream to the data to be transmitted in each OFDM symbol. This stage involves filling the data payload for each sub-carrier. The amount of data transmitted on each sub-carrier depends on the constellation, e.g. BPSK and 16QAM transmit one and four data bits per sub-carrier, respectively. At the receiver the reverse process takes place, with the data from the sub-carriers being converted back to the original serial data stream Sub-carrier modulation Once each sub-carrier has been allocated bits for transmission, they are mapped using a modulation scheme to a sub-carrier amplitude and phase, which is represented by a complex Inphase and Quadrature-phase (IQ) vector. Figure 2-4 shows I-Q constellations of different 13

14 modulation schemes, which map a certain amount of bits for each symbol. Each combination of data bits corresponds to a unique IQ vector, shown as a dot on the figure. Figure 2-4 Typical modulation schemes used in wireless communications [4] According to [4], there are several approaches to the way the constellation can be chosen: Only one constellation is included. The choice of the included constellation is a design decision, depending on the delay and multi-path propagation situation [4]. More than one constellation is included, but only one constellation is used per OFDM frame. The choice of constellation can be based on measurements of the BER [4]. More than one constellation is included, where each sub-carrier can use a different constellation. The choice of constellation is based on the frequency response in each sub-carrier [4]. The last two approaches are the base of adaptive modulation algorithms. An example of adaptive modulation algorithm proposed in [8] is given in the paragraph 3.3. In the receiver, mapping the received IQ vector back to the data word performs sub-carrier demodulation. During transmission, noise and distortion becomes added to the signal. For each received IQ vector the receiver has to estimate the most likely original transmission vector. This is achieved by finding the transmission vector that is closest to the received vector. Errors occur when the noise exceeds half the spacing between the transmission IQ points, making it cross over a decision boundary [8]. These errors can be expressed through Modulation Error Rate (MER) Inverse Fast Fourier Transform After the sub-carrier modulation stage each of the data sub-carriers is set to an amplitude and phase based on the data being sent and the modulation scheme; all unused sub-carriers are set to zero [8]. This sets up the OFDM signal in the frequency domain. An Inverse Fast Fourier Transform (IFFT) is then used to convert this signal to the time domain, allowing it to be transmitted. The IFFT is calculated as it is shown in the equation below [5]. x N 1 k = 0 ( n) = X ( k) e j2πkn N n = 0.. N 1 X is the complex symbol value on the k-th sub-carrier (frequency domain), and N is the number of output signal points calculated, and also the number of input frequency points [5]. Figure 2-5 shows the IFFT section of the OFDM transmitter. In the frequency domain, before applying the IFFT, each of the discrete samples of the IFFT corresponds to an individual sub-carrier. In equation 2-3 x ( n) is the n-th output signal complex value (time domain), ( k)

15 Figure 2-5 IFFT stage in OFDM transmission [8] At the receiver an inverse operation (Fast Fourier Transform, FFT) is used to convert the received time domain signal back to the frequency domain Parallel to serial conversion and guard interval insertion After the IFFT has been calculated, the output complex numbers are parallel to serial converted and the cyclic prefix (or guard period) is inserted. The cyclic prefix copies the complex numbers from the end of the block of output values and inserts them into the front of the block (see Figure 2-6) or from the front of the block copied to the end [5]. Figure 2-6 Copying the guard period from the end of the symbol [5] The reason why the end of the block is copied to the front is so that the delayed paths from the symbol fall within the guard period. To show why this retains orthogonality you have to consider that the OFDM signal consists of the addition of all the sub-carrier signals, which are all at different frequencies f 0 and with different values of I and Q amplitude components. Then, so long as all the multi-paths fall within the cyclic prefix duration and if samples are taken over the "useful" symbol duration (as opposed to the total symbol duration that includes the cyclic prefix) then the FFT in the receiver "integrates" over an integer number of full sine wave cycles, which is a requirement for orthogonality to hold [5]. Thus, the guard period provides protection against multi-path and inter-symbol interference (ISI). Adding a guard period lowers the symbol rate, however it does not affect the sub-carrier spacing seen by the receiver. The sub-carrier spacing is determined by the sample rate and the FFT size used to analyze the received signal [8]. Δ f = Fs N FFT 2-4 In equation 2-4 Δf is the sub-carrier spacing in Hz, F is the sample rate in Hz, and s N FFT is the size of the FFT. The guard period adds time overhead, decreasing the overall spectral efficiency of the system [8]. 15

16 There is one more factor that can lead to OFDM performance degradation at this stage. After parallel to serial conversion a complex time domain signal is formed. It is found that certain bit stream pattern results high peak values at the output of IFFT. This means high peak-to-average power ratio (PAPR) of the signal. High PAPR leads to signal distortion at the receiver side, because the power amplifier (PA) at the transmitter side has nonlinear input-output power characteristics. There are two approaches of how to avoid this distortion: power amplifier linearization and PAPR reduction methods (see 3.1) RF Modulation The output of the OFDM modulator generates a base band signal, which must be mixed up to the required transmission frequency. This can implemented using analog techniques or using a Digital Up-Converter. Both techniques are described in [8]. An example of RF modulation scheme using analog techniques [8] is shown in the figure below. Figure 2-7 RF modulation of complex base band OFDM signal, using analog techniques [8] At this stage there is one element that can seriously degrade the quality of OFDM signal at the receiver side. This is the oscillator that generates the RF carrier. The main problem about it is that there is certain range of frequency instability, which causes fluctuations of RF carrier frequency. These fluctuations result in phase noise. As a result, there will be loss of orthogonality between carriers and, thus, inter-carrier interference may occur. The impact of oscillator phase noise on the OFDM performance, as well as proposed solutions of phase noise compensation, is described in paragraph Effect of noise and fading on OFDM Effect of noise on OFDM In [8] the author defines the main sources of noise in wireless communications. They include thermal background noise, electrical noise in the receiver amplifiers, and inter-cellular interference. In addition to this noise can also be generated internally to the communications system as a result of Inter-Symbol Interference (ISI), Inter-Carrier Interference (ICI), and Inter- Modulation Distortion (IMD). These sources of noise decrease the Signal to Noise Ratio (SNR), ultimately limiting the spectral efficiency of the system. Noise, in all its forms, is the main detrimental effect in most radio communication systems [8]. It sets certain limitations on the usage of modulation schemes; hence the choice of modulation schemes for OFDM is also limited. In case of OFDM, digital modulation schemes are used for data transmission. A modulation scheme, as it was already mentioned in Chapter 1, is mapping of data words to a real (In phase) and imaginary (Quadrature) constellation, also known as an IQ constellation. Increasing the number of points in the constellation does not change the bandwidth of the transmission, thus 16

17 using a modulation scheme with a large number of constellation points, allows for improved spectral efficiency. However, the greater the number of points in the modulation constellation, the harder they are to resolve at the receiver. As the IQ locations become spaced closer together, it only requires a small amount of noise to cause errors in the transmission [8]. This results in a direct trade off between noise tolerance and the spectral efficiency of the modulation scheme and was summarized by Shannon's Information Theory [52], which states that the maximum capacity of a channel of bandwidth W, with a signal power of S, perturbed by white noise of average power N, is given by the equation below. S C = W log N 2-5 The spectral efficiency S E of a channel is a measure of the number of bits transferred per second for each Hz of bandwidth: C S S = E = log W N In equation above both the signal and noise are expressed in terms of linear scale, and the spectral efficiency is measured in b/s/hz [8]. 2-6 Limitations on modulation schemes in case of OFDM transmissions are also described by the author in [8]. They are, as well, based on the simulation results presented in the same source. The simulations were done with different modulation schemes versus additive noise. Based on the results, the author in [8] concludes that the modulation allocation is flexible in OFDM systems allowing them to be optimized to local current conditions, rather than having to always use a low modulation scheme just to ensure the system operates during worst-case conditions. For example, sub-carriers with a low SNR can be allocated to use BPSK (1 b/s/hz) or to transmit no data at all. Sub-carriers with a high SNR can transmit higher modulation schemes such as 256- QAM (8 b/s/hz) allowing a higher system throughput. Based on this flexibility, adaptive modulation algorithms can be used in OFDM transmission. More info about these algorithms is given in section 3.3 and in [8] Effect of flat fading on OFDM Frequency selective fading (see 1.1.5) seriously degrades the performance of wide-band transmission. It results in deep fades (or nulls ) in the wide-band channel spectrum. In case of OFDM transmission (which is narrow-band in its nature), this effect results into flat fading. Flat fading means, that a null covers the whole bandwidth of a single channel (OFDM sub-carrier, in this case). The channel that falls to be in a null might be not able to handle a transmission due to poor SNR. This problem looks even more severe if we consider the unstable behavior (changing SNR) of every single sub-carrier, especially in case of a moving receiver. One solution to overcome this problem is to use adaptive algorithms that estimate the current channel quality and are able to switch to a more robust modulation scheme, or to replace a carrier with poor SNR to a better one during transmission. Examples of adaptive algorithms, proposed in [8], are given in paragraph COFDM Coded OFDM (COFDM) is a further development of conventional OFDM. This method is used in digital audio broadcasting (DAB, see 2.5.1). In case of COFDM the data transmitted on the sub- 17

18 carriers is protected by forward error correction (FEC) coding [5]. The type of error correction coding used in COFDM is convolutional coding and the effect of convolutional coding is that for every one bit input to the error correction encoder, more than one bit is output depending on the "code rate" being used. For example, a code rate of 1/3 would mean that for every bit input to the error correction encoder, 3 bits will be output and these 3 bits are transmitted. Error correction coding therefore adds redundancy to the signal in order for the receiver to be able to correct any bits that are received in error. The error correction decoder used in COFDM is the Viterbi algorithm which tries to decode what bits were sent depending on the received sampled values. COFDM also allows different groups of bits to be protected with a different strength code rate because some bits are more important for the correct reproduction of the audio than some of the other bits [5]. However, the Viterbi algorithm performs poorly when it is presented with bit errors that are all bunched together in the stream, and because the sub-carriers are subject to flat fading, bit errors usually do occur in groups when a sub-carrier is in a deep fade. To protect against this, time and frequency interleaving can be used. An example of using time interleaving method, derived from [5], is shown in figure below. Figure 2-8 Time interleaving [5] The data symbols are written into the interleaving block in column order, then once the block is full, the symbols are read out in row order, so for example the symbols would be read out in the following order: 0, 8, 16, 24, 32, 1, 9, 17 and so on. At the receiver, the received symbols are written into the same sized interleaving block in row order, and once the block is full, the symbols are read out in column order to return the symbols to the original order. The effect of this is to spread out symbol errors that occur grouped together. For example, as the first few symbols transmitted in the above table would be 0, 8, 16, 24 and so on, and if a deep fade occurs which makes symbols 8 and 16 to be received in error, then because of the re-ordering carried out in the receiver the errors end up spread out in time, which allows the Viterbi decoder to have a better chance of correcting all of the symbols [5]. The idea behind frequency interleaving is the same- to spread out those sub-carriers that are in a deep fade, a similar sub-carrier allocation technique is used. The result is that no adjacent carriers will be used for a single transmission. 2.5 OFDM application nowadays Nowadays OFDM is being used in wireless broadcast systems (digital audio and video broadcasting) and in two-way multi-user communication systems (wireless networking standards HiperLAN2 and IEEE802.11a). It is also suggested to be a candidate for the 4 th generation of mobile communication systems (4G) [8]. In the present paragraph there is a short overview of those systems that use OFDM method nowadays: DAB (see 2.5.1), DVB-T and DVB-H (see 2.5.2) and wireless network standards HiperLAN2 and IEEE802.11a (see 2.5.3). Examples of practical OFDM realizations are shown in chapter 4. 18

19 2.5.1 Digital audio broadcasting DAB was the first commercial use of OFDM technology, or, to be more exact, its developed version (COFDM) [5]. Development of DAB started in 1987 and services began in U.K and Sweden in 1995 [8]. DAB is a replacement for FM audio broadcasting, by providing high quality digital audio and information services. OFDM was used for DAB due to its multi-path tolerance. DAB main parameters are shown in the table below. Parameter Transmission mode I II III IV Bandwidth 1,536 MHz 1,536 MHz 1,536 MHz 1,536 MHz Modulation DQPSK DQPSK DQPSK DQPSK Frequency range (mobile reception) 375 MHz 1,5 GHz 3 GHz 1,5 GHz Number of sub-carriers Symbol duration 1000 μs 250 μs 125 μs 500 μs Guard period duration 246 μs 62 μs 31 μs 123 μs Total symbol duration 1246 μs 312 μs 156 μs 623 μs Maximum transmitter separation for SFN 96 km 24 km 12 km 48 km Table 2-1 DAB Transmission parameters for each transmission mode [8] The high multi-path tolerance of OFDM allows the use of a Single Frequency Network (SFN), which uses transmission repeaters to provide improved coverage, and spectral efficiency. The data throughput of DAB varies from Mbps depending on the amount of Forward Error Correction (FEC) applied [8]. This data payload allows multiple channels to be broadcast as part of the one transmission ensemble. The number of audio channels is variable depending on the quality of the audio and the amount of FEC used to protect the signal. For telephone quality audio (24 kbps) up to 64 audio channels can be provided, while for CD quality audio (256 kb/s), with maximum protection, three channels are available [8] Digital video broadcasting DVB is a transmission scheme based on the MPEG-2 standard, as a method for point to multipoint delivery of high quality compressed digital audio and video. It is an enhanced replacement of the analogue television broadcast standard [8]. DVB standards specify the delivery mechanism for a wide range of applications, including satellite TV (DVB-S), cable systems (DVB-C), terrestrial transmissions (DVB-T) and transmission to handheld terminals (DVB-H). OFDM is used for the terrestrial transmission standards for DVB (DVB-T and DVB-H). The physical layer of the DVB-T transmission is similar to DAB, in that the OFDM transmission uses a large number of sub-carriers to mitigate the effects of multi-path. DVB-T allows for two transmission modes depending on the number of sub-carriers used [8]. Table 2-2 shows the basic transmission parameters for these two modes. The major difference between DAB and DVB-T is the larger bandwidth used and the use of higher modulation schemes to achieve a higher data throughput. The DVB-T allows for three sub-carrier modulation schemes: QPSK, 16- QAM (Quadrature Amplitude Modulation) and 64- QAM and a range of guard period lengths and coding rates. This allows the robustness of the transmission link to be traded at the expense of link capacity. DVB-T is a unidirectional link due to its broadcast nature. Thus any choice in data rate verses robustness affects all receivers. If the system goal is to achieve high reliability, the data rate must be lowered to meet the conditions of the worst receiver. This effect limits the usefulness of the flexible nature of the standard. However, if these same principles of a flexible transmission rate are used in bi-directional communications, the data rate can be maximized based on the current 19

20 radio conditions. Additionally for multi-user applications, it can be optimized for individual remote transceivers. DVB-H (previously known as DVB-X) stands for Digital Video Broadcasting: Handhelds and is basically an extension to older DVB-T standard. DVB-H is a terrestrial digital TV standard that uses less power in receiving client than DVB-T, and allows the receiving device to move freely while receiving the transmission, thus making it ideal for mobile phones and handheld computers to receive digital TV broadcasting over the digital TV network (without using mobile phone networks at all) [55]. As an extension to DVB-T, DVB-H uses the same specifications as DVB-T (defined by ETSI EN [54]). Video is normally encoded with MPEG-2 (but can be encoded with MPEG-1 as well, although very rarely used) and the standard, just like its other siblings DVB-C, DVB-S and DVB-T, is mostly used in Europe [55]. More information regarding DVB-H can be found in [54]. Parameter 2k Mode 8k Mode Number of sub-carriers Useful Symbol Duration (T u ) 896 μs 224 μs Carrier Spacing (1/ T u ) 1116 Hz 4464 Hz Bandwidth 7,61 MHz 7,61 MHz Table 2-2 DVB-T transmission parameters [8] Wireless networks (HiperLAN2 and IEEE802.11a) Development of the European high performance local area network (HiperLAN) standard was started in 1995, with the final standard of HiperLAN2 being defined in June HiperLAN2 pushes the performance of WLAN systems, allowing a data rate of up to 54 Mbps [53]. HiperLAN2 uses 48 data and 4 pilot sub-carriers in a 16 MHz channel, with 2 MHz on either side of the signal to allow out of band roll off. User allocation is achieved by using TDM, and subcarriers are allocated using a range of modulation schemes, from BPSK up to 64-QAM, depending on the link quality. Forward Error Correction is used to compensate for flat fading. IEEE802.11a has the same physical layer as HiperLAN2 with the main difference between the standard corresponding to the higher-level network protocols used. The main characteristics of these two technologies are shown in the table below. Parameter IEEE802.11a HiperLAN2 Spectrum 2,4 GHz 5,2 GHz Max physical rate 54 Mbps 54 Mbps Max data rate, layer 3 32 Mbps 32 Mbps Connectivity Connection-less Connection-oriented Table 2-3 Characteristics of HiperLAN2 and IEEE802.11a [8] 20

21 3. OFDM MAJOR DRAWBACKS AND METHODS OF THEIR COMPENSATION This chapter describes major drawbacks of OFDM: high peak-to-average power ratio (PAPR) that leads to the distortion of the received signal, producing inter-modulation distortion (IMD) components and the phase noise (PHN) of the oscillator at the stages of RF modulation and demodulation. Several PAPR reduction methods and PHN compensation algorithms proposed in literature. The description of some of these methods is given in this chapter; PAPR reduction methods are given in the first section and PHN compensation schemes are given in the second section. There are also adaptive algorithms, proposed in [8] to improve the system efficiency in the conditions of radio channel with its changing characteristics. A big problem in this case is flat fading of OFDM sub-carriers. Due to their narrow-band nature the sub-carriers do not experience frequency selective fading (which is natural in case of wide-band transmission), but, instead of that, they suffer from flat fading when the whole narrow-band sub-carrier falls into null (see 2.3.2). These adaptive algorithms are oriented towards flat fading of OFDM sub-carriers, as well as to a more efficient allocation of channel resources. They are described in the third paragraph. Peak-to-average power ratio One drawback of OFDM that can noticeably degrade the system performance is high Peak-to- Average Power Ratio (PAPR). High PAPR causes saturation in power amplifiers, leading to intermodulation products among the sub carriers and disturbing out of band energy [7]. Therefore, it is desirable to reduce the PAPR. To understand better the problem of high PAPR, we can refer to the OFDM system model given in Figure 2-3. At the output of the IFFT block we have time domain representation of every carrier. These time domain signals, after parallel-to-serial conversion, form a complex signal. It is found that certain bit stream pattern results high peak values at the output of IFFT. In other words several independently modulated tones at the N point IFFT exhibit large peak values in the time domain [7]. As a result the PAPR is high. By definition given in [7] we have: PAPR = 2 ( x ) E{ 2 } k x max k Where { } 1 k E x 2 k stands for expected or average value of the time domain signal. The term Crest Factor (CF), also widely used in the literature [8], is the square root of PAPR [7]. Figure 3-1 illustrates what the amplitudes x k of one symbol could look like for a particular symbol that exhibits a large peak. N 3-1 Figure 3-1 OFDM symbol in time domain with high amplitude peak [7] 21

22 This high CF could cause problems when the signal is applied to a transmitter, which contains nonlinear component such as a power amplifier. The nonlinear effects on the transmitted OFDM symbols are spectral spreading, intermodulation, and changing the signal constellation. In other words, the nonlinear distortion causes both in-band and out-of-band interference to signals. The in-band interference increases the Bit Error Rate (BER) of the received signal through warping of the signal constellation and intermodulation while the out-of-band interference causes adjacent channel interference through spectral spreading [7]. An example of distortion effect on OFDM signal spectrum is shown in the figure below. Figure 3-2 Effect of distortion on a 2 tone signal, showing harmonics and IMD [8] Harmonics can easily be removed using a relatively simple low pass filter on the output of the transmitter. IMD is much more of a problem as it results in distortion components, which are inband and out of band but close to the main transmission. These components are a result of mixing between each of the harmonics of the system, and subsequent mixing between the IMD products. In-band components result in added noise to the OFDM signal at the receiver, effectively limiting the SNR of the system, even in the absence of other sources of noise. Out of band components spread the signal in bandwidth, resulting in potential interference with other radio communications in neighboring frequency bands. Even if the signal is perfectly band-limited before going to the transmitter power amplifier, spectral spreading will occur if the power amplifier is non-linear [8]. Two approaches exist to combat the problem of high PAPR. The first approach is to use linear power amplifiers. This means applying several linearization methods to existing power amplifiers in order to extend the linear region of the amplifier input-output power characteristics. Power amplifier linearization methods are briefly described in section They are, however, more related to RF design, which is outside of the scope of the present paper. In [7] it is stated that forcing the PA to work in its linear region is not power efficient and thus not suitable especially for wireless communication systems. The second approach, referred in [7] as a better one, is to try to prevent the occurrence of such nonlinear distortion by reducing the PAPR of the transmitted signal with some manipulations of the OFDM signal itself. Several PAPR reduction techniques have appeared in the literature. However, most of these techniques introduce additional complexity, some reduction in the bandwidth efficiency of the system, and the need for side information [7]. They will be briefly described in section The comparison of these techniques is also given in [7]. 22

23 3.1.1 Power amplifier linearization There is quite a wide choice of solutions offered by several RF manufacturers and developers for PA linearization. Many of them are RF design-related, so their description is outside of the scope of the present paper. I still try to refer to general descriptions of some PA linearization methods given in [26] and [27]. For more detailed description and comparison of these methods the reader can refer to these two sources, as well Predistortion One way to reduce the effects of nonlinearities in the transmitter power amplifier is the use of predistortion [25]. This involves predistorting the signal before the power amplifier in such a way as to cancel the distortion caused by the power amplifier. This predistortion is typically done at base band by changing the amplitude and phase of the time waveform. The most general form of predistortion uses feedback from the power amplifier output to achieve an accurate linearization [8]. Predistortion can reduce distortion from nonlinearities of the transmitter power amplifiers. It however cannot prevent distortion from clipping of the signal at high power levels, due to the limited peak power of any real amplifier [8]. In [26] a similar method is described, where the functional block with transfer function reciprocal to the transfer function of PA is located next to PA. A functional block scheme of pre- and post-distortion is shown in Figure 3-3, where H is the transfer function of the PA and F 1 and F 2 are the transfer functions of predistortion and postdistortion blocks, respectively. This example is given in [26] and in this case both pre- and postdistortion are applied. The main task is here to choose F 1 and F 2 such that the function F2 ( H ( F1 ( x) )) could be a linear function of the input variable x. Figure 3-3 A functional block scheme of pre- and post-distortion [26] In [26] this method is described as requiring initial calibration and sensitive to frequency drift, as well as requiring a good PA model. Adaptive predistortion is a further development of predistortion method. In [26] it is described as a method that solves the problem of frequency drift sensitivity, but, from the other hand, requires a good PA model and introduces more complexity to the system, as well as the power overhead on a DSP chip. The functional block scheme of adaptive predistortion is shown in Figure 3-4. This method is also described in details in [27]. Figure 3-4 Functional block scheme of adaptive predistortion [26] 23

24 Feedforward In [26] the feedforward method is described as more stable than two previously mentioned methods, but, from the other hand, it is said to be susceptible to drift and aging. The functional block scheme of adaptive predistortion is shown in Figure 3-5. This method is described in details in [28]. Figure 3-5 Functional block scheme of feedforward linearization [26] Other PA linearization methods Cartesian feedback method is relatively less complex and offers reasonable IMD suppression, but stability considerations limit the bandwidth to a few hundred KHz [27]. The LINC (LInear amplification with Nonlinear Components) technique converts the input signal into two constant envelope signals that are amplified by Class C amplifiers, and then combined, before transmission. This method is referred as sensitive to drift [27] PAPR reduction methods In this section several PAPR reduction methods are described. More detailed review of PAPR reduction methods can be found in [7] Clipping and filtering One of the simplest ways to reduce the PAPR is to clip the high amplitude peaks. Several clipping techniques have been described in the literature [10]. Some clip the signal at the output of the inverse discrete Fourier transform (IDFT). But subsequent interpolation causes re-growth of the signal peaks. Other clipping techniques clip the signal after interpolation and use a filter to reduce the resulting out-of-band power. However the filters, which have been proposed, are complicated and computationally expensive. In addition they cause peak re-growth and result in significant distortion of the wanted signal [7] Peak windowing A different approach to reduce the PAPR is to multiply large signal peak with a Gaussian shaped window [7], [11]. But, in fact any window can be used, provided it has good spectral properties [7], [12]. Since the OFDM signal is multiplied with several of these windows the resulting spectrum is a convolution of the original OFDM spectrum with the spectrum of the applied window. So, ideally the window should be as narrow band as possible. On the other hand, the window should not be too long in the time domain, because that implies that many signal samples are affected, which increases the bit error ratio. Examples of suitable window functions are the Cosine, Kaiser and Hamming window. Peak windowing technique offers reasonably good reduction in PAPR 24

25 achieved independent from number of sub-carriers, at the cost of a slight increase in BER and out of band radiation [7] Phase optimization techniques Some researchers observe that by rotating the channel constellations properly, the peaks can be reduced. One method introduced the partial transmit sequence (PTS) [13] optimization scheme, in which the N-D constellation is grouped into several blocks, and the relative phases between blocks are optimized. In PTS the information bearing sub carrier block is subdivided into disjoint carrier sub blocks and introduced rotation factors for each sub block and modified sub carrier amplitude vector. Thereby, PAPR was reduced with different rotation factors for different sub blocks. This needs number of iterations to find the optimum combination of factors for sub blocks. Adaptive PTS [7], [14] was proposed to reduce the number of iteration by setting up a desired threshold and trial for different weighing factor until the PAPR drop under the threshold. With this approach we can reduce the number of iteration and the complexity of the system by only 0,1% loss in reduction of PAPR. PTS with embedded side information [7], [15] is another approach that can be combined with both conventional and adaptive PTS. This approach is embedding a combined knowledge within the transmitted data, so no extra bits are sent. But these introduce the word error at a detection of the sequence information [7]. Another method introduced a selective mapping (SLM) [13] scheme in which multiple phase rotations are applied to the constellation points, and the one that minimizes the time signal peak is used. Selective mapping involves generating a large set of data vectors all representing the same information. The data vector with the lowest resulting PAPR is selected. Information about which particular data vector was used is sent as additional carriers. However there may be potential problems with decoding the signal in the presence of noise with selective mapping. Errors in the reverse mapping would result in the data of whole symbols being lost. Another technique that combines selective mapping and cyclic coding by adding extra carriers referred to as Peak Reduction Carriers (PRC) [16]. This method shows a PAPR reduction of more than 4,5 db [8]. However such a method is also limited by the cost of additional transmit power, some bandwidth penalty and complexity limits [7] Tone reservation The basic idea of this method is to reserve a small set of tones for PAR reduction [17]. Fortunately, the problem of computing the values for these reserved tones that minimize the PAR can be formulated as a convex problem and can be solved exactly [7]. The amount of PAR reduction depends on the number of reserved tones, their locations within the frequency vector, and the amount of complexity. This method describes an additive method for reducing PAR in multi-carrier transmission, and shows that reserving a small fraction of tones leads to large reductions in PAR even with simple O(N) algorithms at the transmitter, and with no additional complexity at the receiver. When the number of tones N is small, the set of tones reserved for PAR reduction may represent a non-negligible fraction of the available bandwidth and can result in a reduction in data rate [7] Pulse shaping A new and efficient technique for reducing the PAPR of OFDM based on a proper selection of the time-limited waveforms of the different sub-carriers has been proposed [18]. In this technique each sub-carrier pulse of OFDM scheme has a different shape and all these pulse shapes are derived from the same pulse shape. The technique is referred to as pulse shaping. It has been shown that using this technique it is possible to design a set of time waveforms of OFDM systems that reduce the PAPR of the transmitted signal and improve its power spectrum simultaneously [7]. The method avoids the use of an extra Inverse Fast Fourier Transformations (IFFTs). It works 25

26 with arbitrary number of sub carriers and any type of base band modulation used. The implementation complexity of the proposed technique is by far much low compared to previously published methods. This method has the potential of reducing the PAPR of the OFDM signal without affecting the bandwidth efficiency of the system and does not require any side information [7] Companding schemes In companding [19] (compression in transmitter and expanding in receiver) method by considering the approximate Rayleigh distribution of the OFDM amplitudes, we compress the dynamic range with a memory-less transformation at the transmitter and expand the amplitude level at the receiver. This transformation essentially changes the probability distribution of the amplitude of OFDM signal and achieves the PAPR reduction by both enlarging the small amplitudes and compressing large signals [7]. The power is adaptively allocated for each subcarrier according to the distribution in each block. However, the average signal power increases after the compression and the compressed signals still exhibit non-uniform distributions [7]. A modified adaptive dynamic range companding [20] scheme with better performance has also been proposed Reference signals subtraction The concept of the PAPR reduction method by reference signal subtraction is very simple and all peaks exceeding a certain level are cancelled by subtraction of the appropriately scaled and timeshifted reference signals [7]. The basic disadvantage of this method is degradation of the system performance. The more peaks are cancelled, the more is the original signal distorted. For higher number of sub carriers this method can significantly degrade system performance. A modified method of reference signal subtraction [21] is reported to be highly effective, flexible and almost free from non-linear distortions Carrier interferometry codes Carrier-Interferometry OFDM (CI/OFDM) [22] eliminates the large power peaks and, hence, most PAPR issues, without significant rise in system complexity. CI/OFDM better exploits the frequency diversity of the fading channel, and as such, offers 10 db performance gains over traditional OFDM. This is accomplished by transmission of each bit over each of the N carriers through the novel use of Carrier-Interferometry (CI) phase codes. Specifically, the phase codes applied to the N carriers result in one bit s power reaching a maximum, when the powers of the remaining N-1 bits are at a minimum. The benefits of CI/OFDM relative to OFDM (i.e., large performance gain) are significant. The cost of added transmitter and receiver complexity in CI/OFDM, relative to current OFDM, is minimal when compared to the substantial throughput gains and PAPR reduction [7] Tone injection This is another additive method, which achieves PAR reduction of multi-carrier signals with no rate loss [7]. The basic idea is to increase the constellation size so that each of the points in the original basic constellation can be mapped into several equivalent points in the expanded constellation. The BER will not increase and the only addition to the standard receiver is a modulo-d addition after the FFT [7], [24]. Since each information unit can be mapped into several equivalent constellation points, these extra degrees of freedom can be exploited for PAR reduction. The method is called tone injection, as substituting the points in the basic constellations for the new points in the larger constellation is equivalent to injecting a tone of the appropriate frequency and phase in the multi-carrier symbol [24]. 26

27 Phase noise impact and compensation schemes There is another factor that has its negative impact on the OFDM signal quality. This is the phase noise (PHN) produced by transmitter s and receiver s local oscillator (LO), as a result of its frequency instability. This can lead to loss of orthogonality between OFDM carriers and, as a result, in drastic increase of inter-carrier interference. There are several schemes that were developed to compensate the impact of phase noise on the signal quality. Some examples of phase noise compensation schemes are presented in this paragraph. Phase noise: the background The term phase noise is widely used for describing short term random frequency fluctuations of a signal [29]. Frequency stability is a measure of the degree to which an oscillator maintains the same value of frequency over a given time. To show the main idea behind this term, I am going to refer to the description of phase noise given by Applied Radio Labs [31]. According to this description, a perfect sinusoidal oscillator would produce an ideal sine wave (see equation 3-3). s ( t) = Asin( ωt) In this equation A represents the amplitude, ω represents the cyclic frequency and ω t represents the phase of the signal. In practice, however, the signal always contains some noise. This can be represented by fluctuations in the amplitude of the signal (variations in A) and by fluctuations in the signal phase (so phase becomes ω t + phase noise) [31]. In general we can represent the noisy oscillator signal as it is defined in equation 3-3. () t = ( A + a( t) ) sin ( ωt f ( t) ) s + In this equation a () t represents the amplitude fluctuations in the signal or the amplitude noise, and f () t represents the phase fluctuations or the phase noise Figure 3-6 Signal with phase noise (time domain view) [31] There are two commonly used mathematical models of PHN: stationary PHN and Wiener PHN [37]. When a LO is phase-locked, the time-varying phase can be modeled as a stationary PHN, consisting of a constant phase difference and a zero-mean, wide sense stationary (WSS), colored Gaussian process. When the LO is only frequency-locked (e.g., a free-running oscillator), the 27

28 time-varying phase is modeled as a Wiener process, which is the integration of a white Gaussian random process. The mathematical description of these processes can be found in [37]. In [35], a different approach to phase noise definition is shown, where phase noise distortion is considered to consist of two terms, one dependent on the symbols of adjacent sub-carriers and the other on the symbol under detection, so that the performance degradation is a function of both the phase noise spectrum and the constellation of each sub-carrier. An article dedicated to DVB-T converters illustrates the impact of phase noise in a very practical way [32]. Assume that we have a DVB-T converter that has an unmodulated sinusoidal 4-GHz sinusoidal signal at the input. At the output we need to get the same kind of sinusoidal signal, but with intermediate frequency of 1,15 GHz. The frequency of the converter s heterodyne is, however, unstable and it s randomly fluctuating in some range around the heterodyne nominal frequency. So, the instant frequency of converter s output signal will change within a range, say, of 10 MHz around the nominal frequency of 1,15 GHz. As a result, there will be some uncertainty of the instant phase of the signal. If there is a QPSK-modulated signal instead of an unmodulated signal at the converter s input, the phase of the output signal will not be stable and because of that the QPSK-demodulator will not be able to demodulate the symbol correctly. This will lead to increased symbol error rate at the demodulator s output [32]. Due to the presence of phase noise, the power of sinusoidal signal is not concentrated at the carrier frequency in the pass-band, so that its spectrum cannot be modeled with a delta-pulse at the center frequency [33]. Instead, it is spread in some area around it (like it was with the 10 MHz area in the previous example). The level of phase noise shows, how rapidly is the power decreasing with the frequency offset from carrier. In other words, it expresses the slope of the output signal spectrum envelope, if there is an unmodulated sinusoidal signal at the input [32]. If the power of the signal in the 1 Hz band at the offset of 1 khz from the carrier frequency is X db less than the power of the signal at the carrier frequency (in the same band of 1 Hz), we say, that the phase noise of the tested device is in this case -X 1 khz. The term dbc means decibel per center frequency and it shows, that the power is measured with reference to the centre frequency [32]. A more general expression of phase noise is illustrated in figure 3-7. S c (f) is single sideband phase noise [29], P s is the power at center frequency and P ssb is the sideband power in 1 Hz bandwidth at an offset frequency of f Hz from the center frequency. Figure 3-7 Single sideband noise to carrier ratio [29] In the case of closely spaced sub-carriers in an OFDM system, each sub-carrier mixes with a local oscillator (LO) in a transmitter to produce RF [30]. In so doing, it carries with it the phase noise of the LO. That noise resides adjacent to the sub-carrier's RF position in the transmitted wave. Some of it overlaps adjacent and nearby sub-channels, degrading their maximum SNR and 28

29 increasing inter-carrier interference (ICI) level. LO phase noise in a receiver worsens the problem because uncorrelated noise powers add. Manufacturers of RF devices give several recommendations regarding maximum level of oscillator phase noise allowed for a certain frequency offset. An example of recommendations regarding DVB-T converter oscillator phase noise level can be found in [32]. Different sources define the nature of phase noise in different ways. In Douglas Smith article [30] phase noise is said to be caused by thermal noise in the transceiver. Thermal noise is associated with the random (Brownian) motion of atomic and subatomic particles caused by heat. In VCO datasheet [29] there is a list of factors that define the phase noise in voltage-controlled oscillator (VCO). They include Q factor of the resonator and varactor diode, the active device used for the oscillating transistor, power supply noise and external tuning voltage supply noise. The same document presents, as well, several ways to minimize phase noise. They are practical considerations that are oriented towards circuit and RF components design and, thus, are outside of the scope of the present work. To say more, these considerations are related to specific range of RF devices and may be not the most cost-effective solutions to use in wireless data communications. Some other possible methods of phase noise compensation will be presented in Phase noise compensation schemes Several schemes are developed to compensate the impact of phase noise on the performance and quality of received OFDM signal. In this paper examples of these schemes will be shown. In [34], a non-pilot PHN compensation scheme is proposed. The authors define phase noise as consisting of random Wiener process and fixed but unknown residual frequency offset (RFO). These two manifest themselves as a phase-rotation random variable which is common to all subcarriers, provided that their defining values (variance and fixed amount, respectively) are small, in which case the ICI and the per-sub-carrier amplitude distortion induced by them are both negligible [34]. In [34] the authors conclude that the combined effect of PHN and RFO (also described as Combined-Phase Impairment, CPI in [36]) on any OFDM symbol is common to all sub-carriers, but vary from symbol to symbol [36]. Assuming a known arbitrary channel profile, the authors proposed and evaluated by simulations a decision-directed (i.e., non-pilot-based) compensation scheme. In [34] this scheme is also referred as de-rotation scheme, as it is developed to compensate the phase rotation produced by CPI. The avoidance of pilots (and the increase in useful throughput and flexibility that this implies) comes at the cost of more intense receiver processing, in particular, making two sets of data decisions (one preliminary for the decision-directed mode and one final) [34]. The compensation scheme requires a one-tap equalizer [34] and works as follows: at the first step, after the FFT and per-sub-carrier channel equalization (gain division) have taken place, a set of tentative decisions for all sub-carriers is made. Thus, for every sub-carrier received complex vector, the nearest transmitted candidate symbol is located; in other words, we make a ^ tentative decision per sub-carrier. Let θ m( k) be the difference angle between the complex angle of the observed vector (for the m-th OFDM symbol of k-th sub-carrier) and the complex angle of that tentative-decision symbol defined above. Averaging m( k) get, as a result: ^ av, m θ = 1 1 N ^ θ m N k = 0 ^ θ across OFDM sub-carriers, we ( k)

30 This resulting average is then used to de-rotate all observed equalizer outputs before they enter the final decision device (the slicer ). The cumulative (over time) phase-noise estimate: θ m ^ cum, m = θ av, l l = is also used to de-rotate the total input process before the next OFDM symbol is processed. This can happen at various stages in the receiver chain, and for simplicity is denoted by authors as the carrier tracking loop arrow in figure 3-8 [34]. Figure 3-8 The PHN and RFO compensation scheme [34] The simulations conducted by the authors of [34] have shown that PHN and RFO can be compensated in a decision-directed mode without pilots, and that the resulting schemes are robust to even substantial values of the disturbances. Even if these cause some of the tentative decisions to be wrong, the averaging process smoothens these out. Pilots, in essence, take out the risk of erroneous tentative decisions, but they also absorb system utilization [34]. Another paper issued by the same authors [36] presents two algorithms for CPI compensation. As it was mentioned earlier in [34], the CPI produces an effect, which is common to all the subcarriers, but varies from symbol to symbol. Following the first compensation approach, an adaptive channel estimator can be utilized in order to estimate the equivalent channel (for the m- eq th OFDM symbol of k-th carrier) Hm ( k) = Um( 0) H m( k), which results from both the dynamics of the channel H m ( k) and the phase impairments U m( 0). The compensation for this equivalent channel is done through division of the received symbol on k-th sub-carrier with the corresponding estimate. For this estimation of the equivalent channel, a simple decision-directed Normalized Least-Mean-Squares (NLMS) adaptive equalizer per sub-carrier is utilized, which performs well in the case of no RFO and PHN. The equation that describes NLMS procedure can be found in [36]. An initial channel estimate can be obtained by transmitting a preamble (i.e., known) OFDM symbol and then dividing the corresponding received one with the transmitted: Y = P ^ 0 H 0 ( k) ( k)

31 In equation 3-6 ( k) preamble and Y 0 ( k) is the corresponding observable. P is the known (QAM or BPSK) symbol loaded on the k-th sub-carrier of the The second method (so-called 2-dimentional or 2D), described in [36] utilizes the previous method and separately compensates for the effect of the PHN and RFO (in the frequency axis) and the dynamic channel (in the time axis). Its proposed scheme is shown in the figure below. Figure 3-9 Block diagram of the proposed 2D compensation scheme [36] The algorithm starts with the equalization of the received symbols using an initial estimate of the channel, similar to the one described for NLMS. These equalized observables Y eq m ( k) feed the ^ U () 0 Estimator, which makes a first set of tentative-decisions X ( k) m, tent m for the transmitted symbols and estimates a Um() 0 as the effect of CPI. After compensating for CPI by dividing the equalized observables with the obtained estimate, a set of decisions takes place, providing the ^ final estimates of the transmitted symbols X m( k). Finally the modified adaptive equalizer is updated. The initial values for the channel estimate can be calculated with the use of preambles in the same way that in the NLMS scheme. This proposed scheme can also be viewed as a 2-D equalizer that exploits the correlation statistics both in time (by the NLMS) and frequency (by the CPI compensator). The authors in [36] note that this 2D approach can be utilized with more sophisticated adaptive algorithms, if these are required due to higher channel dynamics. As the CPI effect is estimated and compensated independently, the tracker has to follow only the channel variations. This makes this scheme more complicated but more robust [36]. The simulations in [36] have shown that NLMS scheme can compensate for PHN and RFO efficiently only when small amounts of PHN and RFO are considered and when a small QAM constellation (4 or 16 QAM) is utilized. On the other hand, the proposed 2-D scheme performs well even in the case of transmitting 64-QAM modulated symbols. It is also shown to be more computationally intensive, but also more robust [36]. Another, pilot-based method of PHN compensation is described in [33]. The estimation of PHN is performed in the frequency domain. Since phase noise is multiplied with transmitted signal in time domain then in frequency domain signal power spectral density (PSD) is convolved with phase noise PSD. Several continual pilots are added outside the information band transmitted and the distorted pilots in receiver are the convolution of pilot signals and phase noise PSD. The distorted pilots can be deconvolved for the estimation of phase noise PSD. The authors in [33] call this method per-symbol phase noise detection and compensation. The method estimates CPI (referred as Common Phase Error, CPE in [33]) and ICI based on the pilots added in the transmission spectrum and needs no feedback loops. The guard bands near the pilots avoid the convolution disturbance resulting from other data sub-carriers. The per-symbol method applying the relations between the phase noise spectral components after CPE correction lowers the computation complexity. 31

32 A novel approach to the compensation of phase noise in receivers is given in [37]. The basic concept is to modify the receiver analog front-end by adding an additional signal path directly from the oscillator, so as to provide a non-data modulated observation of the PHN. This enables the PHN to be estimated by a smoothing filter (a filter that employs both past and future information), instead of a conventional prediction filter (a filter that only uses past information). The hardware overhead of introducing this additional path in the analog front-end is modest, especially in an integrated circuit [37]. The information provided by the additional signal path allows a joint prediction and smoothing Wiener filter. The latter optimally estimates the PHN in the minimum mean-squared error (MMSE) sense. The PHN is modeled as a wide sense stationary Gaussian process or a Wiener process. Performance of the proposed scheme (see figure 3-10), in terms of the PHN estimation error variance and the signal-to-noise ratio (SNR) after PHN compensation, is analyzed for both types of PHN. Simulation results for a 64-QAM receiver employing the proposed scheme are presented, demonstrating that the proposed scheme can combat both types of PHN very effectively [37]. Figure 3-10 Block diagram of proposed receiver with PHN compensation scheme [37] Adaptive algorithms As it was mentioned before, there are several factors that have their negative impact on the wireless transmission performance. They include multi-path propagation, noise and interference between neighboring channels. Their impact on OFDM transmission results in flat fading for a single carrier. To effectively use frequency resources for multi-user applications (such as WLAN [8]) in such changing environment like radio channel, adaptive techniques that can track the channel current state (particularly- SNR [8]) are needed. In this section two proposed solutions are described. Adaptive modulation method [8] that allows switching between modulation schemes for every carrier with regards to current SNR of this carrier are described in Adaptive algorithms [8] that allow carrier assignment to users with regards to current SNR [8] of a carrier are described in Adaptive modulation In a multi-path radio channel, frequency selective fading can result in large variations in the received power of each sub-carrier. For a channel with no direct signal path this variation can be as much as 30 db in the received power resulting in a similar variation in the SNR [8]. In addition to this, interference from neighboring cells can cause the SNR to vary significantly over the system bandwidth. To cope with this large variation in SNR over the system sub-carriers, it is possible to adaptively allocate the sub-carrier modulation scheme, so that the spectral efficiency is maximized while maintaining an acceptable BER. Adaptive modulation is a powerful technique 32

33 for maximizing the data throughput of sub-carriers allocated to a user. Adaptive modulation involves measuring the SNR of each sub-carrier in the transmission, then selecting a modulation scheme that will maximize the spectral efficiency, while maintaining an acceptable BER [8]. Research results published in [8] demonstrate the effectiveness of using adaptive modulation in conjunction with different user allocation schemes. Figure 3-11 Adaptive Modulation; the choice of modulation scheme is based on the channel SNR [8]. In the figure above the concept of adaptive modulation is illustrated. For every modulation scheme, depending on the number of bits in a constellation point, a threshold value of channel SNR is defined. The more bits are in symbol, the less robust the scheme is and thus the higher SNR threshold has to be established for this scheme. The SNR must be greater than the threshold to maintain a maximum BER. Excess SNR results in the BER being lower than the BER threshold [8]. The diagram in figure 3-11 assumes that the modulation scheme is updated continuously and with no delay. Using adaptive modulation has a number of key advantages over using static modulation. In systems that use a fixed modulation scheme the sub-carrier modulation must be designed to provide an acceptable BER under the worst channel conditions. This results in most systems using BPSK or QPSK. However these modulation schemes give a poor spectral efficiency (1-2 b/s/hz) and result in an excess link margin most of the time [8]. Using adaptive modulation, the remote stations can use a much higher modulation scheme when the radio channel is good. The BER of the transmission can therefore be controlled more effectively, as sub-carriers that have a poor SNR can be allocated a low modulation scheme such as BPSK or none at all, rather than causing large amounts of errors with a fixed modulation scheme. This significantly reduces the need for Forward Error Correction [8]. There are several limitations with adaptive modulation. Overhead information needs to be transferred, as both the transmitter and receiver must know what modulation is currently being used. Also as the mobility of the remote station is increased, the adaptive modulation process with channel tracking requires regular updates, further increasing the overhead [8]. There is a trade off between power control and adaptive modulation. If a receiver has a good channel path the transmitted power can be maintained and a high modulation scheme used, or the power can be reduced and the modulation scheme reduced accordingly [8]. Distortion, frequency error and the maximum allowable power variation between users limit the maximum modulation scheme that can be used. Inter-modulation distortion results from any nonlinear components in the transmission, and causes a higher noise floor in the transmission band, limiting the maximum SNR to typically db [8]. Frequency errors in the transmission due to 33

34 synchronization errors and Doppler shift result in a loss of orthogonality between the sub-carriers. The limited SNR restricts the maximum spectral efficiency to approximately 5-10 b/s/hz [8]. Adaptive modulation requires accurate knowledge of the radio channel. Any errors in this knowledge can result in large increases in the BER, due to the small link margin used. The effective SNR of each sub-carrier can be estimated by calculating the noise vector on the received signal after channel response compensation [8]. The block diagram of the proposed signal plus noise estimation algorithm is given in figure The algorithm itself is described in details in [8]. Figure 3-12 Algorithm for estimating ((S+N)/N) for a sub-carrier [8] The simulation results given in [8] show that the adaptive modulation strongly decreases the effects of fading, allowing the SNR to be lowered by 16.5 db as compared with a fixed modulation scheme. Adaptive user allocation Existing channel allocation methods Nowadays, there are several schemes of carrier allocation for users. They include fixed frequency grouped sub-carriers, random frequency hopping, comb spread sub-carriers and TDMA [8]. The idea of fixed frequency allocation scheme is to assign each user a group of fixed frequency sub-carriers. Grouping the sub-carriers will tend to minimize the level of inter-user interference due to distortion, power level variation and frequency errors. However, having a fixed group of sub-carriers makes the transmission susceptible to fading, as the whole group of sub-carriers can be lost in a null in the spectrum [8]. The problem of flat fading can be partly overcome by randomly frequency hopping the subcarriers over a wide bandwidth [8]. The groups of sub-carriers are transmitted in short time blocks. These blocks are randomly frequency hopped to ensure that the time period spent in a null would is relatively short, approximately 11 symbols [8]. To recover data lost during a null, time interleaving and forward error correction is used. These come at the cost of reduced system data capacity and increased delay. The simulation results presented in [8] show that the performance of random hopping is better than of the fixed frequency group allocation. This is because the random allocation tended to spread the user allocation over the entire system bandwidth, minimizing the probability of all the sub-carriers suffering deep fading at the same time. 34

35 Sub-carriers can be also allocated in a fixed comb pattern, spreading them over the entire system bandwidth. This improves the frequency diversity, preventing all the sub-carriers used by a user being lost in a single null in the spectrum. However, this method requires user sub-carriers to be interleaved with one another, resulting in a large amount of overlapping energy between the users. Any slight loss of orthogonality due to frequency or timing errors can result in significant inter-user interference [8]. Another method to obtain multiple user access is to use TDMA. This is implemented by allocating each user unique time slots, during which only a single user accesses the channel. The allocation of time slots to individual users can be achieved in several ways, with the most suitable depending on the application and the type of data being transferred. However, due to bursty nature of data traffic this kind of transmission can lead to inefficient use of radio resources, as it results in clustering of data, so that most of the time no data is being transferred on the preallocated timeslot. To partly overcome this problem, dynamic timeslot allocation can be used [8] Proposed adaptive algorithms All previously described techniques allocate fixed amount of bandwidth to each user, regardless of the received signal power. This, however, leads to problems for users that have low received signal strength. The SNR of these users was insufficient to support communications even using BPSK. The SNR seen at the receiver is dependent on the signal bandwidth, and so reducing the bandwidth while using the same transmitter power increases the SNR of the signal [8]. The main aim of adaptive bandwidth allocation is to maintain communications with users that have low received signal strength. This is achieved by reducing their bandwidth to the point where the transmitted power spectral density is high enough to support communications at a low data rate. This can be used as a method for improving the quality of service [8]. This algorithm is not suitable for those applications that required a fixed data rate such as streaming video and audio. To adapt this technique to the needs of streaming applications, a joint optimization of bandwidth and modulation scheme is suggested by author in [8]. This could be achieved by allocating both the user bandwidth and modulation scheme so that the spectral efficiency multiplied by the user bandwidth results in the required data rate. This way, as the signal strength becomes weaker, the amount of bandwidth allocated to that user increases to compensate. In [8], this method is referred as a subject for further research. Another algorithm proposed in [8] is adaptive user allocation algorithm. It allocates system subcarriers to each of the users based on the bandwidth allocated to each user, and the SNR of the user channels. This algorithm, however, is not optimized. In [8] the author states that this algorithm performs relatively poorly if all of the users have a similar average SNR, as the user with the highest SNR (even if it is only 0,1 db higher than the other users) will get last pick of the sub-carriers. As a consequence it may end up with sub-carriers that are all in nulls [8]. Optimization of the algorithm is left to future research. The simulation results of joint adaptive modulation, adaptive bandwidth and adaptive user allocation algorithms presented in [8] shows that these methods can still be useful in achieving a very high quality of service and a high spectral efficiency. However, several aspects should be taken into account in case of adaptive algorithms. As it was mentioned in 3.3.1, the current SNR has to be tracked for every channel. It has to be refreshed many times per second in order to maintain a high performance [1]. This means additional computational load of appropriate elements that have to provide tracking rate at a level sufficient to maintain a certain BER. The results of simulation that has been conducted in order to define BER as a function of tracking rate are given in [8]. Tracking also means overhead, as the information about channel condition needs to be exchanged between transmitter and receiver, 35

36 resulting in user data rate reduction. Besides, a delay between the channel measurement and modulation scheme update can degrade the performance of adaptive modulation algorithm [8]. The simulation results [8] show that delay has its most negative impact on the algorithm performance in case of channel nulls. When SNR is high enough the system is more susceptible to allocation delay. This is because of the fact that the fastest channel response variations occur in the nulls [8]. 36

37 4. EXAMPLES OF OFDM PRACTICAL IMPLEMENTATION In this chapter some examples of OFDM practical implementation are shown. The descriptions of practical OFDM realizations given on this chapter are based on manufacturer s technical papers and, therefore, some of them do not include such important details as block-schemes or other data that could be a subject of manufacturer s commercial secret. The goal of this chapter is, however, not to illustrate the details of every realization, but to show, how widely OFDM is commercially implemented today. DVB-T realizations OFDM is widely used in digital video broadcasting applications (DVB-T and DVB-H). There is a wide choice of practical OFDM solutions developed for these applications by several vendors. In the present paper, some examples of these solutions are shown. An example of OFDM solution developed for DVB-T and compatible with DVB-H standard is DVB-T-7900H modulator developed by StreamTel [40]. The technical paper defines this modulator as the most advanced and reliable DVB-T / DVB-H modulator on the market, made for being on-air today and forever in any possible DVB-T application with no compromises. It is compatible with many different DVB-T transceivers. The RF output stage is equipped with the StreamTel advanced DPC (digital pre-corrector) that makes the output RF spectrum shape perfectly matching any possible existing transmitter specs, compensating for every possible transmitter input filter and response for the maximum efficiency [40]. An adjustable digital RF clipper with in-band digital noise shaper are used to maximize the transmitter power efficiency and coverage without impacting on the MER and signal quality (patented) [40]. The results of measurements, conducted by StreamTel using AGILENT VSA 89650S 80Mhz BW Vector Analyzer, refer to MER > 47,5 db. The modulation schemes used in DVB-T-7900H are QPSK, 16-QAM and 64-QAM. The size of IFFT can be chosen among the values of 2000, 4000 and 8000 and the total bandwidth can be chosen among the values of 5, 6, 7 and 8 MHz. The length of guard interval has its available values of 1/4, 1/8, 1/16 and 1/32 of the symbol length [40]. An example of DVB-T-7900H modulator realization taken from the technical paper is given in figure 4-1. Figure 4-1 Stand-alone typical connections of DVB-T-7900H modulator by StreamTel [40] 37

38 Another example of OFDM solution for DVB-T is PT5780 DVB-T modulator designed by Advanced Broadcasting Electronics (ABE) [41]. It complies with the DVB-T standard, ETS and provides the function of converting a compliant MPEG-2 transport stream into a COFDM modulated signal. This solution was developed to provide the maximum coverage area of a DVB-T network. To ensure maximum coverage area, a high MER value measured at the transmitter output, is important. If MER is low then the transmitter power has to be increased to provide coverage area. The modulator provides the best foundation to build on. The system starts with a high precision modulator giving a high initial MER value. Finally, a state of the art digital pre-corrector eliminating most of the non-linearities in the transmitter, is added [41]. As it was mentioned before, power amplifier non-linearity will cause received signal distortion and adjacent channel interference. To resist non-linearity factor, PT 5780 modulator incorporates a state of the art digital pre-corrector. The pre-corrector digitally optimizes transmitter performance, increasing available output power and coverage area. The pre-corrector offers a peak limiter in order to control PAPR [41]. The modulator supports QPSK, 16-QAM and 64-QAM modulation schemes; the bandwidth supported is 6, 7 and 8 MHz, the supported IFFT size is 2000 and 8000 and the supported length of guard interval is 1/4, 1/8, 1/16 and 1/32 of the symbol length [41]. WLAN realizations One example of OFDM solution developed for wireless Metropolitan Area Networks (WirelessMAN) is the a WirelessMAN OFDM modulator/demodulator by Commsonic [38] is a group of broadband wireless communications standards for Metropolitan Area Networks (MANs) developed by a working group of the Institute of Electrical and Electronics Engineers (IEEE) [39]. A block diagram of the demodulator part of the Commsonic solution is given in figure below. Figure 4-2 Block diagram of the demodulator part of the Commsonic WirelessMAN modulator/demodulator solution [38] The mixer in the scheme is used for carrier offset correction. Correlator is used to detect the preamble sequences and set the OFDM symbol timing. Carrier and timing loops block responds for tracking out carrier and timing errors. In order to estimate these errors Pilot Process block is used in the scheme. Channel Estimate/Correct block is used to estimate channel frequency and phase response from the known preamble sequence and to apply to subsequent OFDM symbols to correct them. 3 or 4 bit soft decision slicer with channel quality weighting is used to maximize 38

39 convolutional code performance. 2-stage permutation deinterleaver is used to spread out errors due to poor carriers [38]. The modulation schemes used in this solution are QPSK, 16QAM and 64QAM. The size of FFT is 200 carriers and the length of guard interval has its available values of 1/4, 1/8, 1/16 and 1/32 of the symbol length [38]. This solution uses forward error correction coding schemes (Viterbi and Reed-Solomon coding), interleaving and scrambling. More details about these coding algorithms can be found in [38]. Commsonic also offers FFT blocks, as well as Viterbi and Reed-Solomon codecs as separate solutions. Realizations for multiple purposes OFDM realizations designed form special purposes (WLAN, WirelessMAN, DVB-T, and DVB-H) were described in two previous sections. There are, however, OFDM realizations that are proposed to be compatible with multiple standards. Some examples of these multi-purpose realizations are given in this section. One example is a MegaCore OFDM solution developed by Altera [42]. A block scheme of OFDM transceiver with Altera intellectual property (IP) solutions is shown in figure 4-3. Figure 4-3 MegaCore OFDM transceiver solution by Altera [42] Altera signal processing forward error correction encoding (FEC) cores include high-performance encoding and decoding for Reed-Solomon, convolutional, Viterbi, and turbo codes. Interleaver/deinterleaver MegaCore function [42] provides easy customization and quick instantiation into the design. The interleaved data is then passed through a serial-to-parallel converter, which maps the symbols onto an I-Q constellation specific to the modulation scheme. The constellation mapper takes symbols as inputs and maps them to appropriate constellation points as dictated by the modulation method specified. This process generates I and Q values which are then filtered and sent to the IFFT for transformation. Altera high-performance programmable logic devices (PLDs) can implement constellation mapper functionality in user logic. A buffer is required to store the values of I and Q before they are sent to the IFFT [42]. The advantage of the Altera signal processing OFDM solution is that each of the functional blocks of the OFDM transmitter can be mapped onto dedicated, parallel hardware resources within the PLD, avoiding the difficult programming and optimization challenges of scheduling time-critical operations through a single DSP device [42]. Necessary in any wireless or wired digital communications design, digital filters help shape the signal. Altera next-generation finite impulse 39