TLT-1406 Introduction to Communications Engineering Spectrum, waveforms, and data transmission 1. General view to communication systems 2. The idea of modulation 3. Spectrum concepts 4. Modulation, bandpass transmission and complex signals 5. Digital transmission system model 6. Channel capacity Core content: • Understanding signal spectrum • Understanding the idea and purpose of modulation • Main elements of a digital transmission link The material is a selected subset of the slides of the TLT-5206 Communication theory. Recommended textbook for further information: A.B. Carlson, Communication Systems, Fourth Edition, McGraw-Hill, 2001. TLT-5206 / 1 Introduction Information and messages Telecommunications is transferring information from one place to another in an electric or electromagnetic form. In the following we refer to an information bearing message, which can be a waveform, a bit sequence etc. The main task of a communications system is to make a copy (as good as possible) of the transmitted message in the receiver. INFORMATION SOURCE TRANSMISSION SYSTEM INFORMATION WELL Original message can be analog (e.g. speech) or digital (e.g. text). Regardless of the source format, the message can be transferred either in analog or digital form. · Division between analog and digital transmission is mostly based on how source information is presented from transmission system’s point of view. For example, speech transmission (i.e., original message is analog): · Analog transmission system: target is to transfer the original message as little interfered and distorted as possible, e.g., using appropriate carrier modulation. · Digital transmission system: representing the original message by a bit sequence (source coding, speech encoder), and transferring the bits effectively using some specific communications waveforms In analog transmission there are certain quality criterions (Signal-tonoise ratio, distortion, etc.). In digital transmission, instead, the objective is to minimize the bit error probability. TLT-5206 / 2 Information and messages (continuing) All (at least most) physical transmission channels, and therefore, also used transmission signals (i.e. waveforms), are analog and continuous · E.g. voltage or current signal in a cable, electromagnetic wave in radio path, etc. INFORMATION SOURCE TRANSMITTER TRANSMISSION SYSTEM INFORMATION WELL TRANSMISSION CHANNEL RECEIVER Thus, telecommunications is fundamentally: Producing effective message bearing waveform that performs as well as possible in the used communications channel (transmission media). TLT-5206 / 3 Elements of a communications system Transmitter adapts the signal from the source to suit the transmission channel. The most essential signal processing operations are, e.g., · Modulation (almost all communications systems) · Coding (digital source and/or digital communications system) Two fundamental resources are · Transmission power and energy [mW,W,…] · Bandwidth [kHz,MHz] Note! Physical and/or official regulations given by authorities Transmission channel generates different undesired changes in the message signal. In practice these changes appear also in the receiver, but in the model they are all associated with the channel. For example: · Attenuation · Noise: additive random nature interference signal · Interference: additive interference signal from some recognizable source (e.g., adjacent channel transmission) · Distortion: distortion affected by the non-ideal response of the communications system (linear filtering, non-idealities, etc.) · Transmission can be “one-way” or “two-way”: o Simplex (one-way) o Full-duplex (transmission and reception can be done “at the same time”) o Half-duplex (two-way transmission but not at the same time) Receiver restores the message signal into the original form so that the non-idealities created by the channel are minimized (amplification, filtering, demodulation, decoding). TLT-5206 / 4 Transmission channels Electric wire · Pair of wires (e.g., regular telephone connections) · Coaxial cable Radio path · broadcasting (transmissions of broadcasting companies) · microwave links · satellite transmission · cell networks Optical fiber Magnetic tape, etc. TLT-5206 / 5 Physical limitations The most essential physics law originated limitations in communications systems are power/energy and bandwidth. In nature there is always, e.g., thermal noise which may become as a quality limiting factor (such as when increasing the transmission distance → transmission loss increases). On the other hand, all physical systems have a limited bandwidth. Bandwidth will eventually become as a limiting factor, since it directly affects the achievable information transmission rate. Especially systems utilizing the radio path are often limited by the bandwidth. For a channel whose bandwidth B and signal-to-noise power ratio (S / N ) is known, we are able to calculate the largest possible capacity C < B log2 (1 ∗ S / N ) This is known as the Hartley-Shannon law and it is one of the most famous results in information theory related to communications systems. The law will be more carefully studied in the end of the course (information theory part). The above mentioned fundamental resources (power/energy and bandwidth) are anyhow those physical keystones, whose around the contents of the course, e.g., regarding to wave form studies, are based on (basically a physical layer of any communications system). TLT-5206 / 7 Modulation In modulation two separate waveforms are connected: a modulating message signal and a properly chosen carrier. In below there are examples of amplitude modulation using sine wave and square wave as a carrier. In both cases the message signal is found in the envelope of the modulated signal. In the receiver the envelope can be easily distinguished from the received signal (demodulation). Usually the carrier frequency is much larger than the highest existing frequency in the message signal. Moreover, the frequency band of the modulated signal is commonly located around the carrier frequency. As a result, there is a frequency shift in modulation. TLT-5206 / 162 Examples of analog modulation methods Unmodulated carrier: Low frequency signal (information signal): Amplitude modulated carrier (information lies in amplitude variations): Frequency modulated carrier (information lies in frequency variations): TLT-5206 / 8 Advantages of the modulation Transmission channel requirements For instance, in radio transmissions the required antenna size is proportional to the used wave length (at least 1/10 of the wave length). E.g., with 100 Hz frequency a 300 km long antenna would be necessary whereas with 100 MHz frequency a practical antenna size would be 1 m. Separation of different transmission stations By setting an individual carrier frequency for each transmission, it is possible to transmit several messages without interfering with each other. Radio frequencies are divided into separate radio bands by international agreements (different communications services on each band). Multiplexing When messages are modulated using separated carrier frequencies, multiple messages can be transmitted simultaneously in the same channel. Mitigation of noise and interference When using appropriate modulation methods and increasing the bandwidth the effects of noise can be reduced in the demodulated signal (by not changing the transmission power, of course). Implementation originated limitations From communications electronics point of view, the relative bandwidth (transmission bandwidth/carrier frequency) should be about 1-10% → modulation method has to be selected depending on the carrier frequency and transmission bandwidth. TLT-5206 / 6 Frequency allocation of electromagnetic spectrum TLT-5206 / 9 Coding Coding is basically digital transmission related message signal modification. Decoding means the inverse operation performed in the receiver. Examples: 1) ASCII-code: coding alphanumeric signs into binary data 2) When binary data is transmitted in a good quality channel, transmission capacity can be increased by using 2M-level symbols (each correspond to M binary data bits). 3) In error control coding, redundancy is added in the original information message. This enables correction of some errors occurred in the channel. Error correction capability depends on the used code structure. In real life digital transmission all these can be used along with some other coding schemes. NB! Introduction to digital transmission is presented in the last part of the course. TLT-5206 / 10 SIGNALS AND SPECTRA Electric/electromagnetic signal: electric or electromagnetic time dependable quantity. For example, voltage between the ends of the wire pair as a function of time, or a electromagnetic wave in the radio path. In the following we concentrate on representing signals in time domain and frequency domain and study how these two are connected. In the frequency representation, a signal is divided into sinusoidal components with specific frequencies. The frequency representation (i.e. spectrum) is mathematically defined by the Fourier transform (note that the plural of spectrum is spectra). In case of periodic signals, spectrum can also be found by using Fourier series. TLT-5206 / 11 Sinusoidal signals Sinusoidal signal can always be presented as follows v(t ) < A cos(w0t ∗ f) < A cos(2p f0t ∗ f) § A is amplitude, w0 < 2p f0 is angular frequency ( f0 is frequency) and f is phase angle (or just phase). This is a periodic signal whose period is T0 < 2p / w0 < 1/ f0 One of the signal maximums is found from t < ,f / w0 TLT-5206 / 12 Spectrum of a sinusoidal signal One-sided spectrum of a sinusoidal signal is depicted in the following figure: Amplitude spectrum Phase spectrum Amplitude spectrum and phase spectrum include a peak at frequency f0 . From spectrum it is possible to observe the essential parameters of a sinusoidal signal: frequency, amplitude and phase. TLT-5206 / 13 Linear combination of sinusoidal signals w(t ) < 7 , 10 cos(40pt , 60↓) ∗ 4 sin 120pt This can be modified as: w(t ) < 7 cos 2p 0t ∗ 10 cos(2p20t ∗ 120↓) ∗ 4 cos(2p60t , 90↓) from which one-sided line spectrum can be drawn: Amplitude spectrum Phase spectrum TLT-5206 / 14 Complex representation of sinusoidal signal Real life signals are always real valued. However, a concept of complex signals is often a useful and necessary tool in communications engineering. Although in some of the studied cases one would manage using only real valued signals, it is common in spectrum analysis to use formalism that enables also utilization of complex signals. Complex signals are closely related to so-called bandpass signals (more carefully studied later in the course). For now, without a specific physical background, a complex signal is defined as a signal or wave form whose value (at any moment of time) is a complex number. · In practice: two parallel real valued signals (Re and Im parts) In this context Euler’s equations are frequently needed: e ° j q < cos q ° j sin q The way to another direction can be found by e j q ∗ e,j q cos q < < Re éë e j q ùû 2 e jq , e,jq sin q < < Im éë e j q ùû 2j Notations in the future: 1) In spectrum representations the free variable is f (unit Hz). Angular frequency is defined as w < 2p f . Certain fixed frequencies are denoted as f0, f1, fi , etc. 2) Phase angle is measured with respect to a cosine wave. 3) Amplitude is always positive: ,A cos wt < A cos(wt ° 180↓) 4) The unit for the phase angle is degree (° ), although it is usually defined as radians. TLT-5206 / 15 Two-sided spectrum For real and periodic signals one can use a one-sided line spectrum. Instead in future, we concentrate on two-sided spectrum which enables also studying of complex signals. In two-sided spectrum representation the base functions are complex exponential functions (instead of real oscillators). In case of real valued signal, two-sided spectrum is obtained by substitution (Euler) A cos(w0t ∗ f) < A j f j w0t A , jf ,j w0t e e ∗ e e 2 2 Considering the earlier example, the two-sided spectrum is now given as: w(t ) < 7 cos 2p 0t ∗ 10 cos(2p20t ∗ 120↓) ∗ 4 cos(2p60t , 90↓) < ... TLT-5206 / 19 Fourier-series Objective: Finding out the frequency content (spectrum) of an arbitrary periodic signal. A periodic signal can be represented as an exponential Fourier series development: v(t ) < ⁄ å n <,⁄ cne j 2pnf0t where T0 is the period (cycle time), f0 < frequency, and cn < c(nf0 ) < 1 T0 ò v(t )e T0 , j 2 pnf0t 1 is the fundamental T0 dt Coefficients cn are complex quantities which are usually represented in the polar form: cn < cn e j arg cn Exponential Fourier series defines the two-sided (line) spectrum for a periodic signal. It consists of multiples of f0 (can you see why?). Number cn indicates the value of the amplitude spectrum at frequency nf0 whereas arg(cn ) indicates the value of the corresponding phase spectrum. TLT-5206 / 21 Sinc-function In future, we repeatedly use the sinc-function given as: ìï sin pl ï sinc l < ïí pl ïï 1 ïî kun l ÷ 0 kun l < 0 Where does this come from? In spectrum analysis following type of integrals come up frequently: 1 T T /2 ò ,T / 2 e j 2p ftdt < ... The result of the integral can be conveniently represented by the sincfunction (see the next example). TLT-5206 / 22 Example: Square wave or pulse train The coefficients for the Fourier series can now be given as: 1 cn < T0 < T0 / 2 ò ,T0 / 2 1 v(t )e , j 2pnf0tdt < T0 t /2 ò ,t / 2 Ae , j 2pnf0tdt A A sin pnf0t At sinc nf0t < e , j pnf0t , e j pnf0 t ( < ∋ ,j 2pnf0T0 T0 pnf0 T0 In the following figure the amplitude and phase spectra are shown in case of pulse ratio (or duty cycle) t /T0 < 1/ 4 . The sinc shaped function sinc f t can be found from the envelope of the amplitude spectrum. The amplitude of the DC-component is c0 < At /T0 , which can easily be derived also from the time domain function (how?). TLT-5206 / 26 Fourier transform and continuous spectrum Objective: Defining the frequency content (spectrum) for an arbitrary non-periodic signal. Let’s study signal whose energy (compare to the power) E < ⁄ ò ,⁄ v(t ) 2 dt is finite. Practically this means that the signal is restricted in some relatively small time span. For this type of a signal the Fourier transform is defined as: V (f ) < F Ζ v(t ) ∴ < ⁄ ò ,⁄ v(t )e ,j 2p ftdt where V ( f ) is the spectrum of v(t ) . With non-periodic signals the spectrum is continuous (meaning what?). The spectrum has the following properties: 1) V (f ) is a complex function. V ( f ) is the amplitude spectrum and argV ( f ) is the phase spectrum. 2) V (0) < ⁄ ò ,⁄ v(t )dt 3) If v ( t ) is real, then V (,f ) < V )( f ) , i.e., V (,f ) < V ( f ) and argV (,f ) < , argV ( f ) If needed, time function v(t ) can be obtained from V ( f ) with the inverse Fourier transform: v(t ) < F ,1 Ζ V ( f ) ∴ < ⁄ ò V (f )e j 2pftdf ,⁄ (cf. Fourier series) TLT-5206 / 27 Example: Rectangular pulse In future, a rectangular pulse is denoted as Ο(t / t ): ì ïï 1 Ο(t / t ) < í ïï 0 î t ; t /2 t = t /2 (draw by yourself) Let’s now study signal v(t ) < AΟ(t / t ) . Its Fourier transform is given as V (f ) < t /2 ò ,t / 2 Ae ,j 2p ftdt < At sin p f t < At sinc f t pf t It can be seen that the spectrum of a rectangular pulse correspond with the envelope shown in the previous example with the line spectrum of a square wave. In addition, it is shown that most of the spectrum energy is located between the frequencies f ; 1/ t . This means that the spectrum of short (or narrow) pulses is wide. · Reciprocal spreading phenomenon (universally valid) TLT-5206 / 32 Calculating Fourier transforms in practice Straightforward integration is basically possible only in few cases (text book examples, etc.) Other ways: 1) Tables & transform theorems (duality, frequency and time translation, etc.) · The most common elementary operations and their mapping into the transform domain 2) Approximation. If function z%(t ) approximates function z (t ) , then ⁄ ò ,⁄ Z (f ) , Z%(f ) df < 2 ⁄ ò ,⁄ z (t ) , z%(t ) 2 dt It follows that the approximation in frequency domain can become as accurate as desired by improving the approximation accuracy in time domain. Proof: try Rayleigh’s energy theorem … 3) Numeric methods: Discrete Fourier transform (DFT) · Can be calculated, e.g., with Matlab software (see Matlab exercises!). · The differences between the continuous Fourier transform and the DFT must be recognized! TLT-5206 / 33 About the DFT Background: Using the same methods as with continuous signals, it is possible to define the Fourier transform and spectrum also for a discrete signal (sequence of numbers) x (k ) . Fourier transform of an N -point sequence x (k ) is defined as X (e j 2 p fTS )< N ,1 å x(k )e,j 2p fkT S k <0 (cf. continuous signal transform) Here TS refers to the distance between two adjacent numbers in time domain and notation X (e j 2p fTS ) tries to emphasize (i) the difference compared to the continuous transform, (ii) and second, the periodicity of the sequence transforms · Above function X (e j 2p fTS ) repeats itself at integer 1/Ts multiples (Why? Verify yourself…) It follows that if the sequence x (k ) is a sample sequence taken from a continuous signal x (t ), i.e., x (k ) < x (kTs ), k < 0, K, N , 1 , and if X ( f ) does not include larger frequency components than 1/ 2Ts (so there is no aliasing), then X (e j 2p fTS ) < TS X ( f ) ; , 1/(2TS ) ′ f ′ 1/(2TS ) (A lot more about this will be discussed later…) TLT-5206 / 34 About the DFT (continuing) Now the actual discrete Fourier transform (DFT) of an N -point sequence x (k ) is defined as: X (n ) < DFT Ζ x (k ) ∴ < N ,1 å x(k )e,j 2pkn / N k <0 ; n < 0,1, K, N , 1 When compared to the previous one, it is shown that DFT consists of samples from Fourier transform of a sequence at frequency points nf0 where f0 < 1/(NTs ) (this is DFT’s resolution): X (n ) < X (e j 2p fkTS ) f <nf 0 f0 <1/(NTS ) ; n < 0,1, K, N , 1 Note! FFT (Fast Fourier transform) is, instead, just a set of methods to calculate the DFT in an effective manner (see Matlab: “help fft”). TLT-5206 / 35 Calculating the spectrum with the DFT So, with DFT it is possible to define the spectrum of a signal by using a sampled sequence of it. According to previous pages, the frequency resolution of the DFT is f0 < 1/(NTs ) and it includes the frequency band ,1/(2Ts ) ; f ; 1/(2Ts ) . Negative frequencies can also be incorporated, since X(n ) can be understood periodic so that X (N , n ) < X (,n ) < Ts X (,nf0 ) . · Frequency band can be increased by reducing the sampling interval. · Frequency resolution can be improved by increasing the sampling interval or by adding the number of samples (e.g. by adding 0samples in the end of the sequence) · If x (t ) ÷ 0 when t ; 0 or t ″ NTs signal must be cut in the sampling. This induces distortion in the signal. The effects of distortion can be mitigated by using appropriate window functions. TLT-5206 / 148 A general bandpass signal Let’s study a bandpass signal vbp (t ) whose spectrum Vbp ( f ) is centralized around a specific centre frequency fc : Vbp ( f ) < 0 when f ; fc , W ja f = fc ∗ W (In principal the centre frequency can be arbitrary chosen within this frequency band. In practical cases the selection is somewhat obvious.) A sinusoid of frequency fc can be spotted in the waveform. The narrower is the band, the more the signal resembles the pure sinusoid. Small variation in the envelope A(t ) and/or in the phase f(t ) can be seen in the signal. Mathematically this can be represented as (intuition) vbp (t ) < A(t ) cos ∋ wct ∗ f(t ) ( It is worth of noticing that the envelope is generally considered as nonnegative, A(t ) ″ 0 . Consequently, the possible changes in the sign correspond to ±180° phase rotation. TLT-5206 / 150 Representation of a bandbass signal (continuing) According to the previous, there are two different representation forms for the bandpass signal. Both of them are defined by two separate time functions: 5) A(t ), f(t ) Envelope and Phase 6) vi (t ), vq (t ) In-phase and Quadrature (I/Q) components Both of the representation forms are used a lot in future. Besides, both of the forms have their own unique strengths (e.g., waveform behavior, spectral analysis). The connection between these representation forms is found as follows: vi (t ) < A(t ) cos f(t ) vq (t ) < A(t ) sin f(t ) ...and the same in reverse direction: A(t ) < vi2 (t ) ∗ vq2 (t ) f(t ) < arctan vq (t ) vi (t ) Instead of using terms in-phase and quadrature component, they are often referred as I and Q components. TLT-5206 / 151 Lowpass equivalent signal A spectrum of a bandpass signal can be represented using spectra of in-phase and quadrature components: vbp (t ) < vi (t ) cos(wct ) , vq (t ) sin(wct ) ↔ 1 j Vbp ( f ) < ∋Vi ( f , fc ) ∗ Vi ( f ∗ fc ) ( ∗ ∋Vq ( f , fc ) , Vq ( f ∗ fc ) ( 2 2 To assure that the bandpass signal is bounded within frequencies fc , W ′ f ′ fc ∗ W , in-phase and quadrature components must be lowpass signals (why?): Vq (f ) < Vi (f ) < 0 when f =W Now, it is possible to formulate a lowpass equivalent signal: vlp (t ) < 12 ∋ vi (t ) ∗ jvq (t ) ( < 12 A(t )e j f(t ) Generally, this is a complex signal, whose interpretation directly corresponds with the earlier phasor representation. Its spectrum is Vlp ( f ) < 21 ∋Vi ( f ) ∗ jVq ( f ) ( < Vbp ( f ∗ fc )u( f ∗ fc ) Thus, this is the positive part of the bandpass signal’s spectrum relocated around the zero frequency. TLT-5206 / 152 Spectrum of a lowpass equivalent signal A passband signal/system can be effectively modelled using the lowpass equivalent model, both in analytical studies and simulations. * Complex signal models are a necessary tools for doing this! Modulation: baseband => passband Demodulation: passband => baseband TLT-5206 / 155 About complex signals Based on the previous, a physical real valued bandpass signal vbp (t ) < vi (t ) cos(wct ) , vq (t ) sin(wct ) < A(t ) cos(wct ∗ f(t )) < Re Ζ vlp (t )e j wct ∴ can be described using a complex valued lowpass equivalent vlp (t ) < vi (t ) ∗ jvq (t ) < A(t )e j f(t ) whose · Real and imaginary parts at any time instant express the I and Q components of a physical bandpass signal · Amplitude and phase at any time instant express the envelope and phase of a physical bandpass signal In general, complex signals do not involve any strange “mystique” properties · Simply two parallel real valued signals representing the real and imaginary parts of a complex signal · Processing and modification performed by using complex arithmetic NB! This is one of the topics that will be discussed more detailed in the 7cr course extension TLT-5206 / 338 INTRODUCTION TO DIGITAL TRANSMISSION Digital transmission has become more and more popular in every fields of communications during the last 0 years. GSM introduced around 1992 All new communications systems under development are based on digital transmission. Contents: 1) The elements of digital transmission system 2) The advantages of digital transmission 3) Introduction to baseband digital transmission · Digital PAM system 4) Introduction to digital carrier modulation · I/Q modulated PAM/PSK/QAM; digital frequency modulation 5) Introduction to the information theory In course TLT-5406 Digital transmission this topic is studied more extensively · will be lectured in periods 3 and 4 · http://www.cs.tut.fi/kurssit/TLT-5406/ Furthermore, more detailed issues regarding specific techniques, such as CDMA and OFDM(A), are discussed in courses · TLT-5606 Spread spectrum techniques · TLT-5706 Multicarrier techniques TLT-5206 / 340 The elements of digital transmission system The system includes possibly a conversion of an analog message signal (e.g. speech) into the digital form (sampling and quantization) and vice versa. The transmitting end of the actual transmission system converts the digital signal into an analog waveform which is then transmitted into the channel. The receiver converts the analog waveform again into the digital form. The transmission chain includes: Source coding/decoding: To reduce the bit rate of the digital message signal by removing some existing redundancy (“compressing”) One of the most essential results of the information theory is that the source coding and the channel coding can be performed independently between each other . Channel coding/decoding: To reduce the effects of errors produced in the transmission channel (error control coding) Almost in any reasonable channel, an arbitrary small bit error probability can be achieved by adding redundancy in the transmitted signal. Modulation/demodulation: Converting a digital signal into an analog waveform and vice versa. Channel: Here harmful noise, interference and distortion are encountered. While designing the system, one target could be minimizing the required bandwidth or/and the transmission power/energy. TLT-5206 / 341 Block diagram of digital transmission system In the following, when referring to the digital transmission system we mean the part of the chain, in which the interfaces are the input of the channel coding and the output of the channel decoding. Therefore, source coding and decoding are not included here (compression of information, speech/audio/video coding, etc.) Based on this definition we are able to design and analyze these types of systems independent of the nature of the transmitted information TLT-5206 / 342 Fundamental parameters of a digital transmission system The external operation of the previously defined system can be described with the following parameters: · Transmission rate (bits/s) · Error probability · Propagation delay and the delay caused by the signal processing and other processing From system’s external operation point of view only these parameters have significance · The fact that how, e.g., a certain transmission rate is achieved, is not important for the end user => more degrees of freedom / flexibility when compared with the pure analog transmission Thus, the internal operation of the system (coding, waveforms,…) can be optimized with respect to to the properties of the used transmission medium (bandwidth, transmission power,..) so that the requirements for the external operation are fulfilled! Notice that also the requirements for the external operation parameters are strongly application specific · E.g., intelligible speech vs. file transfer TLT-5206 / 345 BASEBAND DIGITAL TRANSMISSION Bits, Symbols, and waveforms The starting point in the baseband digital transmission is the utilization of pulse amplitude modulation (PAM) for transmission of binary bit sequences or generally multilevel symbol sequences. A multilevel symbol is produced when, e.g., 4 bits are combined into one symbol. In this case, 24 < 16 different symbol levels are required to represent all possible bit combinations. Generally, B bits can be represented using 2B levels. The number of used symbol levels (i.e., the number of bits / symbol) is selected based on the requirements of the application and transmission channel so that different symbol levels can be reliably distinguished in the receiver. When several bits are combined into one symbol, the used symbol rate (baud rate) can be decreased. This affects directly the required bandwidth as we will see later on. Simple example: bit sequence : 01442443 1 0 0 1144244 1 0 03 K symbol sequence : -3A 9A K bits/s symb/s binary signal: 16-level signal: NB! Bit rate, symbol rate and symbol alphabet size (M) are related as fbit < log2 (M ) ≥ fsym TLT-5206 / 346 Pulse shapes A digital PAM signal is composed of pulses scaled (weighted) with the transmitted symbol values (from here the designation “PAM”). Therefore, the waveform transmitted to the channel is given as x (t ) < å ak p(t , kT ) k < ... ∗ a 0 p(t ) ∗ a1p(t , T ) ∗ ... Here T is the symbol interval (the symbol rate is fsym < 1/T ) and p(t ) is the fundamental pulse shape whose amplitude is scaled according to the transmitted symbol value ak . It is important that pulses representing consecutive symbols do not interfere with each other in the reception. In ideal case the following condition is fulfilled: 1 when t < 0 So called Nyquist criterion to ì ï ï p(t ) < í avoid inter-symbol ï t T T 0 when < ° , ° 2 ,... ï interference. î This can be practically implemented in two different ways: ) Using short pulses that are not superimposed in time o Simple implementation, however, the bandwidth is not the smallest possible o Example: a rectangular pulse with the symbol interval duration ) Using pulses that are superimposed in time, but we take care that the above condition is still fulfilled o The used bandwidth can be minimized, however, the implementation is more complex (e.g., symbol synchronization in the receiver o This is called as Nyquist pulse shaping and it is an essential element in developing modern communications systems TLT-5206 / 347 Pulse shapes (continuing) Example: Illustration of PAM signal composed of individual pulses using the rectangular pulse and a little bit smoother (and longer) pulse. · Short rectangular pulse: · Smoother and longer pulse: TLT-5206 / 349 Spectral contents of a digital PAM signal Spectral contents of a digital transmission signal depend on the symbol sequence properties and the used pulse shape (quite obviously). Now, let’s assume that · Symbol sequence ak is a discrete random signal with power spectral density of Ga (e j 2p fT ) · The Fourier transform of the used pulse shape is P ( f ) In this case, the power spectral density of a digital PAM signal is 1 Gx ( f ) < P (f ) 2 Ga (e j 2p fT ) T The latter term is commonly constant => pulse shape determines the spectrum. Continuous and discrete time signals appear here, and therefore, the mathematical derivation of the above result is not trivial. On the other hand, the result is quite reasonable from the filter interpretation point of view that will be discussed later on · PAM signal generation using a transmitter filter Here the general starting point is obviously that the spectrum of the transmitted signal must be fitted into the properties of the channel · E.g., in cables the attenuation is not constant in the used frequency band (it is increased in high frequencies). That is why the signal power is desired to be centered in the low frequencies where the attenuation is the lowest. This will also decrease the crosstalk issues in cable systems and radio frequency interference. · On the other hand, in AC coupled systems the spectrum is desired to be zero at the zero frequency. · Thus, the above result offers tools to process the PAM signal spectrum (i.e., functions Ga (e j 2p fT ) and P ( f ) ). TLT-5206 / 364 Pulse design Time domain Requirement: inter-symbol interference (ISI) is zero p(t ) p(0) < 1 p(mT ) < 0 , m < °1, °2, K Pulse form is otherwise free. Frequency domain The ideal bandlimited sinc-pulse is not a practical solution. Practical pulses include typically about 10 -100% excess bandwidth compared to the sinc-pulse. Consequently, the overall bandwidth is W < (1 ∗ a) 1 2T where a < 0.1, ... , 1 is so called roll-off factor. In practice, the pulse shape filter P ( f ) has a symmetric transition band with respect to the minimum bandwidth 1/(2T ) . The Nyquist criterion guarantees zero ISI and leads to the specific shape of the spectrum. Raised-cosine filter is a common way to generate such pulse shapes. TLT-5206 / 368 Raised-cosine pulses The idea: Controlling the length of an ideal sinc-pulse using a window function. Here the pulse duration is decreased by increasing the excess bandwidth a . In this case the oscillation of the pulse attenuates faster but the bandwidth increases. é cos(apt /T ) ù é sin(pt /T ) ù é cos(apt /T ) ù ú ê ú p(t ) < ê < sinc( t / T ) úê êë 1 , (2at /T )2 úû êë pt /T úû êë 1 , (2at /T )2 úû The Fourier transform is: ìï ïïT ïï ïT é pT P ( f ) < ïí 1 , cos ê ïï 2 ë a ïï ï0 ïîï ζ ∋ f , (| 1∗a ù ú 2T û ; f ′ 1,a 2T 1,a 1∗a ′ f ′ 2T 2T 1∗a ; f ″ 2T ; p(t ) P (f ) , 1 2T 1 2T f TLT-5206 / 357 BASEBAND DIGITAL TRANSMISSION SYSTEM BASED ON THE NYQUIST PULSE SHAPING Channel Typically modeled with a linear filter (describes the linear distortion) and additive noise (usually normally distributed). · In mobile systems the channel varies as a function of time, and therefore, also the filter models have to be time-dependent. Coder Converts the incoming bit stream into a symbol sequence where the symbols are taken from a specific alphabet. Some examples: 1) Binary alphabet, the symbol sequence is basically the same as the bit sequence 2) Two consecutive bits are described using the alphabet {,3, , 1, ∗ 1, ∗ 3} (or more generally using the complex alphabet {,1, , j, ∗ 1, ∗ j } ) TLT-5206 / 358 About the transmitter elements Transmitter filter Converts the discrete symbol sequence into a continuous-time signal. The impulse response of this filter g(t ) is now the transmitted pulse shape. Thus, the transmitted waveform is given as S (t ) < ⁄ å m <,⁄ Am g (t , mT ) where Am is the transmitted symbol at time instant mT , and 1/T is the symbol rate. Consequently, the waveform is composed of pulses scaled with the symbol values. Depending on the used pulse shape, the pulses may overlap in time. Below there is a simple example, in which the pulse shape g(t ) is a rectangular pulse of length T : TLT-5206 / 359 Channel The received waveform can be given as R(t ) < b(t ) ) S (t ) ∗ N (t ) < < < ⁄ ò b(t )S (t , t )d t ∗ N (t ) ,⁄ ⁄ ⁄ Am g(t , mT , t ) d t ∗ N (t ) ò b(t )må <,⁄ ,⁄ ⁄ å m <,⁄ Amh(t , mT ) ∗ N (t ) Here h(t ) is the received pulse shape h(t ) < ⁄ ò ,⁄ b(t )g(t , t ) d t < b(t ) ) g (t ) Example: If the channel is steeply bandlimited 1 ì ï ï B( f ) < í ï 0 ï î f ; BW f ″ BW then the rectangular pulse shape (in previous page) is not feasible, since it would be strongly distorted in the channel · Magnitude of the distortion depends on the ratio between the bandwidth BW and the symbol duration T (why?) In future optimizing the pulse shape within the given band limitations is anyway one of the most important objectives. TLT-5206 / 360 Elements of the receiver Development and functioning of the receiver is generally more critical · Detection of transmitted bits reliably from the received signal (noisy and distorted) Basic receiver functionality is depicted below. Timing recovery Defines the correct timing and sampling time for the received pulses. There are often specific components in the transmitted signal that make the synchronization easier. However, this is not always necessary (in carrier modulated system also the carrier synchronization is important). Receiver filter 1) Attenuates noise and interference outside the transmission band 2) Affects the pulse shape 3) If the transmission channel is known, the receiver filter can compensate linear distortion caused by the channel (equalization using the inverse transfer function) · Channel’s transfer function is not usually known, and therefore, adaptive techniques are important · In practice part of the receiver filtering (e.g., channel equalization) is performed using discrete time filters (after sampling) NB! The signal after the receiver filter f (t ) is given as Q(t ) < f (t ) ) R(t ) < ... < ⁄ å m <,⁄ Am p(t , mT ) ∗ N ϒ(t ) where p(t ) < g(t ) ) b(t ) ) f (t ) is the overall pulse shape and N ϒ(t ) < f (t ) ) N (t ) describes the filtered (colored) noise. TLT-5206 / 361 Elements of the receiver (continuing) Sampling In sampling the continuous-time signal is sampled to create the corresponding discrete time signal. In ideal case the sample is taken at the moment in which the sample corresponds best with the transmitted symbol (effects of other symbols is minimized) Detection (decision making) µ k of the transmitted symbol sequence Ak is In detection an estimate A created using the received sample sequence Qk . Conventionally this is based on some decision thresholds. Example: Let’s consider a ternary alphabet {,1, 0, ∗ 1} . In this case the decisions could be done according to the following figure: Here we have applied so called minimum distance principle which will maximize the probability of correct decisions with certain assumptions (e.g., noise distribution type). · Basically, statistical decision making (detection theory) is a large field of applied mathematics which will be more discussed in the TLT-506 Digital transmission course. Decoding Describes the detected symbol sequence back into the bit stream according to the used alphabet. TLT-5206 / 366 Pulse shaping in baseband PAM system Inter-symbol interference is essential in the sampling process in the receiver. The observed pulse shape p(t ) < g(t ) ) b(t ) ) f (t ) depends on the transmitter filter g(t ) , the receiver filter f (t ), and the channel b(t ) . Now the corresponding transfer function is P ( f ) < G (f )B( f )F (f ) . The objective is that the cascade of the three transfer function P ( f ) < G ( f )B( f )F ( f ) fulfils the Nyquist criterion. This is so called zero-forcing criterion, since it forces the inter-symbol interference to zero. However, when the channel noise is included, it does not necessarily offer the optimal solution (as seen later on) The transfer function of the channel is usually fixed or it cannot be affected. Transmission and receiver filters are designed together. Here the following solutions are available: 4) Pulse shaping in the transmitter, the receiver filter approximates the ideal lowpass filter whose bandwidth is (1 ∗ a)/ 2T . Simple from implementation point of view. 5) Matched filter pair is theoretically the optimal solution. In this case the impulse responses are mirror images and the amplitude responses are the same. If the channel does not considerably affect the pulse shape, this gives the optimal solution. 6) Transmitter filter is designed according to the item (1). The receiver filter strives adaptively to minimize ISI with certain criterions. Feasible in the sense that the channel’s transfer function is rarely known (at least while designing the filter). On the other hand, implementing an adaptive filter, instead of a fixed one, is more complicated. TLT-5206 / 369 Raised-cosine pulses (continuing) Many of the Nyquist filters, also the raised-cosine filters, have also so called square root Nyquist versions · Here the idea is that an individual pulse does not fulfill the Nyquist criterion but two consecutive filters (in cascade) do. · I.e. one square root filter in the transmitter and one in the receiver. Expressions: p (t ) < 4a t t t cos([1 ∗ a ]p ) ∗ sin([1 , a ]p ) T T T t t p [1 , (4a )2 ] T T ì ïïï T ïï ï T é pT P ( f ) < ïí 1 , cos ê ïï 2 ë a ïï ïï 0 ï î ζ ∋ f , (| 1∗a ù ú 2T û ; f ′ 1,a 2T 1,a 1∗a ′ f ′ 2T 2T 1∗a ; f ″ 2T ; Design and analysis in Matlab, e.g., using the “rcosine” function In case of an ideal channel, square-root raised-cosine filters in both transmitter and receiver is the ideal solution providing 0 ISI. TLT-5206 / 375 CARRIER MODULATION IN DIGITAL TRANSMISSION In carrier modulation the baseband signal is relocated around a desired carrier frequency (...in frequency domain). ...or…from digital transmission point of view: In carrier modulated digital transmission, we develop effectively bit carrying waveforms, whose power/energy is located in the desired part of the spectrum. Contents: · Complex quadrature (I/Q) modulation and complex constellations: QAM, PSK TLT-5206 / 377 Complex (I/Q) modulation Here s(t ) < x (t ) < ⁄ å m <,⁄ am g(t , mT ) é jw t ⁄ ù 2 Re ê e c å am g(t , mT ) ú êë úû m <,⁄ and wc < 2p fc where fc is the carrier frequency. (scaling factor 2 is included to match the powers of the transmitted signal x (t ) and the baseband signal s(t ) ). Spectrum illustration is given below: Modulating signal spectrum Modulated signal spectrum X (f ) WLP < (1 ∗ a) 1 2T NB! Bandwidth of the modulated signal WBP < (1 ∗ a) 1 T WBP < (1 ∗ a) 1 T TLT-5206 / 378 I/Q modulation Practically g ( t ) (transmit filter) is always a real pulse: ⁄ 2 cos(wct ) å Re[am ]g(t , mT ) , 2 sin(wct ) å Im[am ]g(t , mT ) x (t ) < m <,⁄ ⁄ m <,⁄ Hence, the real and imaginary parts of the baseband signal modulate the cosine and sine components of the carrier. Although this is the practical implementation principle, complex notation is used in future (since it is much simpler). NB: The term bandpass PAM is quite often used in case of the I/Q modulation. TLT-5206 / 379 Bandpass PAM receiver Two equivalent structures: 1) Based on complex signals: 2) Based on real signals: Spectra interpretation using complex signals: TLT-5206 / 381 Constellations In I/Q modulated PAM, the symbol values are complex numbers. If we assume that the transmit filter g(t ) is real (practically always true) and we denote the complex symbol using its amplitude and phase as am < rme jfm , the modulated signal can be further written as x (t ) < < < é jw t ⁄ ù 2 Re ê e c å amg (t , mT ) ú êë úû m <,⁄ é ⁄ j w t jf ù 2 Re ê å e c rme m g (t , mT ) ú êë m <,⁄ úû 2 ⁄ å m <,⁄ rm cos(wct ∗ fm )g(t , mT ) Consequently, it is possible to think the modulated waveform so that · The amplitude and phase in a certain symbol period defines the amplitude and phase of the modulated carrier · Changes between consecutive symbols depend on pulse shaping The utilized complex symbols (the symbol alphabet) can be illustratively presented with a constellation figure. Often used constellations are, e.g., · QAM: constellation points are distributed in uniformly spaced grid · PSK: constellation points are distributed uniformly in a circle Also many other constellations are available. TLT-5406/103 CONSTELLATIONS Symbol values are complex in I/Q modulation. They determine the amplitude and phase of the modulated carrier at the sampling instants. The transitions between adjacent symbols depend on the used pulse shaping. Example: Square pulse shaping, symbol sequence 1 -1 3j -5j 1 6 4 2 0 -2 -4 -6 0 0.5 1 1.5 2 2.5 3 3.5 Time in symbol intervals 4 4.5 5 The complex symbol values (alphabet) can be represented by constellation: • QAM: points are situated on a regular rectangular grid • PSK: points are situated on a circle • Also other constellations are available, but not in wide use. TLT-5206 / 382 Example constellations 4-PSK (QPSK) · Alphabet size 2B < 4 , each symbol represents B < 2 bits. · Symbols: Am < be j fm ; fm Î {0, p 2 , p, 3p 2} · Information is in the phase of the modulated waveform 16-QAM · Alphabet size 2B < 16 , each symbol represents B < 4 bits. · Symbols: Am < am,I ∗ jam,Q ; am,I , am ,Q Î {°c, °3c} · Information is in the amplitude and phase of the modulated waveform More general: · M-PSK: 2B constellation points uniformly distributed in a circle · M-QAM: 2B square formed (2B / 2 ≥ 2B / 2 ) constellation (uniformly spaced “grid”) TLT-5206 / 383 Noisy constellation Because of the noise, the received samples are not perfectly matching the constellation points. Now, if we draw noisy samples from the complex domain, the following types of results can be seen: Symbol error rate as Im{Qk } a function of SNR is an important performance metric. Re{Qk } It is calculated based on the properties of Gaussian distribution. Im{Qk } Re{Qk } If the physical channel noise is Gaussian distributed, also the deviation around the ideal constellation points is distributed in the same way (why?). Subsequently, Gaussian “clouds” are seen around the points. Generally, the objective in decision making is to select that specific symbol (from the used alphabet) which the received complex sample Qk would most probably be representing. Intuitively the most µ k which minimizes the distance reasonable selection is the symbol A 2 µ k . It can be shown that with certain presumptions this is also Qk , A the optimal selection (that maximizes the probability of the correct decision). By this way a decision region can be formed around every constellation point. This region includes all those points that are closer to that specific symbol compared to any other symbols in the alphabet. Here are examples of decision regions for the 4-PSK and 16-QAM: Im{Qk } Re{Qk } Im{Qk } Re{Qk } TLT-5206 / 392 SHORT INTRODUCTION TO INFORMATION THEORY In this section we touch the very basics of such concepts as information, entropy, and channel capacity. In the general electrical communication context, by using these ideas, it is possible to determine the largest possible information transmission rate through a given channel · This is called the channel capacity · Formulated by the so called Shannon-Hartley law Even though it is usually not possible to achieve the channel capacity in a practical system, it is an important reference point when evaluating the performance of practical systems. In fact, the Shannon-Hartley law is one of the most important fundamental laws of nature in the field of communication theory, and it is quite useful also in practical engineering work. In general, the main idea here is to shortly introduce these concepts. Much more detailed treatment will be given in the following course TLT-5400/5406 Digital Transmission Sources and references: · E. A. Lee and D.G. Messerschmitt, “Digital Communication”, Kluwer, 1988/1994/2003. · S. Benedetto and E. Biglieri, “Principles of Digital Transmission”, Kluwer, 1999. · B. Sklar, “Digital Communications”, Prentice-Hall 2001. TLT-5206 / 401 CHANNEL CAPACITY Addressing the fundamental physical limits for the amount of information (per time unit or channel use), which in theory can be communicated error-free over a given channel. SOURCE X CHANNEL Y SINK The source is here modeled as a sequence of independent observations of a source random variable X . The observation consists of another random variable Y . Based on the earlier discussions, the average amount of information at source per unit time is H (X ). The question is now: How much of this information can “pass” through the channel ? And in general: What’s the maximum amount of information transfer rate that a given channel can support ? These are addressed in the following. TLT-5206 / 404 Example 1: Binary Symmetric Channel 1–p x=0 p y=0 Y X p x=1 1–p y=1 · channel bit flip probability p · input probabilities PX (1) < q , PX (0) < 1 , q · an example of discrete-input discrete-output system model The maximum of mutual information and thus the capacity is obtained for q < 1/ 2 which yields (…) C s < 1 ∗ p log2 p ∗ (1 , p) log2 (1 , p) g p < 1/2 g p=0 or p < 1 Þ Cs < 0 (input and output are independent of each other) Þ C s < 1 (error-free binary channel) TLT-5206 / 405 Example 2: Capacity of Continuous-Time AWGN Channels Physical system bandwidth W [Hz] and the channel adds white Gaussian noise (AWGN) to the transmitted waveform. Then it turns out that the maximum mutual information and thus the capacity is of the form (bits per second here) C < W log2 (1 ∗ S ) N where S denotes the received signal power and N the additive noise power within the signal band. This is generally known as ShannonHartley law. When deriving this result (TLT-5400/5406), it is assumed that also the information bearing signal is Gaussian distributed (this is indeed needed to maximize the mutual information). Obviously none of the practical digital transmission systems transmit Gaussian signals but are typically based on discrete symbol alphabets (binary, QPSK, 16QAM, etc.) Then one crucial question is how much of the available maximum capacity is lost when using the discrete alphabets. The answer is: Not much, when designed and implemented properly. This (among others) will be addressed in much more details in the Digital Transmission course ... :o)