Projects in Wireless Communication Lecture 1

Transcription

1 Projects in Wireless Communication Lecture 1 Fredrik Tufvesson/Fredrik Rusek Department of Electrical and Information Technology Lund University, Sweden Lund, Sept 2018

2 Outline Introduction to the course Basics of digital communications Discrete-time implementations Carrier transmission

3 Introduction Lecturer and course responsible: Fredrik Tufvesson, E:2361A 5-6 scheduled lectures Teaching assistants: Guoda Tian, E:2367 MinKeun Chung, E:2334 Computer help sessions in the lab.

4 Introduction Ultimate goals for PWC: 1) Two computers should communicate via speaker/microphones 2) Two computers should communicate using software defined radios The projects should be performed in groups of TWO students (HARD LIMIT)

5 PWC 1 In the first part of PWC we only work in software. For a passing grade you must solve three tasks: 1. A digital baseband BPSK system should be implemented in C++ and its performance should be measured and verified against theoretical results ( ) P e = Q d 2 E b min N 0 2. Later in PWC you will encounter physical passband signals at the input of the microphone. In the first part, we will provide each group with one such signal; the bits carried by the signals correspond to the ASCII code of a secret password. If you can decode the signals and provide me with the password, you have passed task Same as 2 but with OFDM transmission and convolutional code.

6 Recommended reading Software-Defined Radio for Engineers, by Travis F. Collins, Robin Getz, Di Pu, and Alexander M. Wyglinski, 2018, ISBN-13: We will use some chapters in the second half of the course, and it covers many of the aspects in the first half as well. There is a free pdf of the book available, see

7 Example Assume that you receive the following noisy signal You must remove the noise...done! Decode the bits: Convert to ASCII: You have passed PWC1, congratulations...

8 Example Assume that you receive the following noisy signal You must remove the noise...done! Decode the bits: Convert to ASCII: You have passed PWC1, congratulations...

9 Example Assume that you receive the following noisy signal You must remove the noise...done! Decode the bits: Convert to ASCII: You have passed PWC1, congratulations...

10 Example Assume that you receive the following noisy signal You must remove the noise...done! Decode the bits: Convert to ASCII: You have passed PWC1, congratulations...

11 Example Assume that you receive the following noisy signal You must remove the noise...done! Decode the bits: Convert to ASCII:You have passed PWC1, congratulations...

12 Example Assume that you receive the following noisy signal You must remove the noise...done! Decode the bits: Convert to ASCII: You have passed PWC1, congratulations...

13 Introduction Formal descriptions of the tasks can be found online.

14 Basics of Digital Communications This is a recall of baseband digital communications... We need to transmit a bit sequence {u k } = Map to symbols {a k } { 1, u k = 0 BPSK : a k = 1, u k = 1 1, u 2k u 2k+1 = 00 i, u 2k u 2k+1 = 01 QPSK : a k = 1, u 2k u 2k+1 = 10 i, u 2k u 2k+1 = 11

15 Basics of Digital Communications Each symbol is carried by a base pulse p(t) of length T, e.g. the half-cycle sinus p(t) t/t

16 Basics of Digital Communications So the transmission of bits generates the pulse train y(t) Transmission of t/t Mathematically we have y(t) = k a k p(t kt s ) Note that T s is the symbol time while T is the duration of the pulse p(t). How does T and T s relate in this example?t = T s

17 Basics of Digital Communications So the transmission of bits generates the pulse train y(t) Transmission of t/t Mathematically we have y(t) = k a k p(t kt s ) Note that T s is the symbol time while T is the duration of the pulse p(t). How does T and T s relate in this example? T = T s

18 Basics of Digital Communications So the transmission of bits generates the pulse train y(t) Transmission of t/t Mathematically we have y(t) = k a k p(t kt s ) Note that T s is the symbol time while T is the duration of the pulse p(t). How does T and T s relate in this example? T = T s

19 Basics of Digital Communications To avoid intersymbol interference one can use T < T s t/t In this example we have T = T s /2

20 Basics of Digital Communications The channel model assumed in this review is a pure AWGN channel AWGN y(t) r(t) Where the noise n(t) satisfies E{n (t)n(t+τ)} = δ(τ)n 0 /2; such a noise process must have power spectral density R(f) N 0 2 f

21 Basics of Digital Communications What does WGN look like? Can we show an example? Consider the power of the process P = R(f)df n(t) has infinite power! Thus, not possible to show an example of WGN

22 Basics of Digital Communications What does WGN look like? Can we show an example? Consider the power of the process P = R(f)df n(t) has infinite power! Thus, not possible to show an example of WGN

23 Basics of Digital Communications What does WGN look like? Can we show an example? Consider the power of the process P = R(f)df n(t) has infinite power! Thus, not possible to show an example of WGN

24 Basics of Digital Communications Explanation: Every signal we ever see in reality has been filtered by some low-pass filter.

25 Basics of Digital Communications Mathematically, in what way should the receiver process the received signal r(t). In other words...?â =...?arg min x r(t) k x k p(t kt s ) 2 dt

26 Basics of Digital Communications Mathematically, in what way should the receiver process the received signal r(t). Maximum-likelihood detection is the answer! â = arg max a P rob{r(t) a}

27 Basics of Digital Communications Mathematically, in what way should the receiver process the received signal r(t). Maximum-likelihood is equivalent to minimum Euclidean distance detection â = arg min a r(t) k a k p(t kt s ) 2 dt

28 Basics of Digital Communications To decode the (complex valued) signal r(t), we pass r(t) through a matched filter z(t) z(t) = p( t) For symmetric pulses p(t), we get Let z(t) = p(t) x(t) = r(t) p(t) = k a k g(t kt s ) + η(t) where η(t) is n(t) p(t) and g(t) = p(t) z(t). Take samples every T s seconds: x k = x(kt s ). Then x k = E p a k + η k where η k is a complex Gaussian random variable with variance E p N 0, that is E p N 0 /2 per dimension!

29 Basics of Digital Communications Energy computations and error probability: The energy per transmitted symbol E s is given by: E s = p 2 (t)dt while } {{ } E p the energy per transmitted bit is { E b = E s, BPSK E s /2, QPSK The physical minimum Euclidean distance is { Dmin 2 4E p, BPSK = 2E p, QPSK In both cases we end up with a normalized distance d 2 min probability is given by ( ) P e Q 2 E b N 0 = 2. The error

30 Discrete-Time Implementations In a computer-based package such as Matlab or C/C++, we cannot represent the signals y(t) as continuous time signals. Hence we must work with sampled versions. Let f s be the sample rate in samples/second and N be the number of samples per symbol. In PWC2, f s = samples/second We get that T s = N f s The symbol rate becomes R s = f s N

31 Discrete-Time Implementations We must sample the base pulse p(t). Assume a sample interval of T s /N seconds T s This is wrong!

32 We must sample the base pulse p(t). Assume a sample interval of T s /N seconds Ts/N T s This is wrong!

33 Discrete-Time Implementations We must sample the base pulse p(t). N+1 samples per symbol implies sample interval of T s /N seconds Ts/N T s This is wrong!

34 Discrete-Time Implementations Explanation: Plot two consecutive pulses Should be one point! 0 There should only be one point.

35 Discrete-Time Implementations Correct sampling! Represent the samples in a vector p = [ ]

36 Discrete-Time Implementations A sampled transmission signal of t/t s Slightly harder mathematical representation. Let {b k } be a zero-padded version of {a k } Then, b = [a }{{} N 1 y k = l b l p k l a }{{} N 1 a }{{} N 1 or simply y = b p a 4...]

37 Discrete-Time Implementations Convolutions in discrete-time: A convolution of x(t) and y(t) in continuous time is carried out as x(τ)y(t τ)dτ (1) Let x and y be sampled version of x(t) and y(t); the sampling rate is f s. The discrete time version of (1) is 1 x l y k l f s l The discrete time convolution must be scaled by the sampling rate!. 1/f s works as dτ in (1). The energy of the pulse p(t) must be approximated as E p = 1 f s k p 2 k

38 Discrete-Time Implementations Matched filters in discrete-time: The pulse train p should be filtered by a discrete-time matched filter. For symmetric pulses, we can take this mathched filter as z = p where p includes the last sample!, i.e. the length of p is N + 1. (This is however not crucial.) Then the output of the matched filter is (N = 20) Peak at sample samples The number of samples in y is N number of symbols and the length of the filter output is N + N number of symbols. The peak occurs at samples 1 + kn, k = 1, 2, 3...

39 Discrete-Time Implementations If there is a guard band (T < T s ), then the pulse is not symmetric and we can not take z = p. We must then use z k = p N+2 k, k = 1...N + 1 It is still true that the peaks occur at samples 1 + kn, k = 1, 2, 3...

40 Discrete-Time Implementations Implementation of discrete-time AWGN: Until now we have constructed a modulation signal y in discrete time. We now seek a noise vector n to be added to y that represents continuous time AWGN (that has inf power). We have that both the real and the imaginary parts of the samples of η(t) = n(t) z(t) are zero-mean and have variance E p N 0 /2. In discrete-time, a sample of the filtered noise process is given by η k = 1 n l z k l f s l

41 Discrete-Time Implementations Assume that the variance of each n k is σ 2 n. From probability theory it follows that η k has variance σ 2 n z 2 k /f 2 s. Since we get that σ 2 n k z 2 k/f 2 s = E p N 0 2 σ 2 N 0 = E p f s 2 = N 0 2 k z2 k 2 f s Thus, The sampling rate affects the variance of the discrete time representation of continuous AWGN