1. OFDM – Orthogonal Frequency Division Multiplex Naftali Chayat CTO, BreezeCOM ©BreezeCOM, 2000

2. Overview Motivation – the Multipath and its effects OFDM principles Error Correction Coding OFDM-based standards –Broadcast standards – DAB and DVB –LANs & Access – 802.11a and HIPERLAN/2 –Baseband – ADSL

3. High Speed Digital Comm – the curse of multipath The traditional way of sending information is serially This type of communication is affected by multipath time

4. Multipath time frequency

5. Effect of Multipath: Inter-Symbol Interference (ISI) Each bit becomes distorted by echoes The symbols disturb each other Sent data Data and the echoes Resulting waveform

6. Solution 1 - Equalization Equalization is building an “inverse filter” If channel has nulls, you cannot inverse Decision Feedback Equalizer (DFE) can handle also channels with nulls –Uses past decisions In coded systems past decisions may be unreliable In long channels – complexity problem

7. Solution 2 – Parallel Channels Send several long symbols in parallel Only the edges are corrupted Sent signal Received signal

8. How to send in parallel? Use signals at different frequencies Both sine and cosine – complex exponential

9. Frequency domain view ORTOGONALITY –The peak of each signal coincides with nulls of other signals

10. FFT period and Guard Time Equispaced in frequency  periodic in time Send slightly more than one period Frequency domain Time domain Guard time Waveform sent FFT period

11. Guard time and Multipath The multipath corrupts the Guard Interval The FFT region remains undistorted Sent signal Received signal FFT period symbol GI

12. Multipath effect on OFDM Each subcarrier is scaled, but they still do not interfere with each other Sent signal Received signal

13. QAM constellations For multiple bits/symbol – QAM constellations Gray coding is typically used – neighbors differ by one data bit only 0111 1000 11 QPSK=4QAM 2 bits/sym 16QAM 4 bits/sym 64QAM 6 bits/sym Q I Q I Q I

14. Modulating the subcarriers The sine and cosine are multiplied by I and Q and added. 16QAM 4 bits/sym I Q

15. QAM in OFDM environment Larger constellations require higher SNR Some subcarriers are received stronger, while others are received weaker –Some may be completely faded Some form of mutual protection is needed  Error Correction Coding

16. Convolutional Codes and the Viterbi Algorithm Data bits are passed through a shift register The XOR outputs are sent. Trellis representation –States are labeled according to shift register contents DD 00 01 10 11 00 01 10 11 data x0 x1 0 000 00 0 110 11 1 011 01 1 101 10 time ntime n+1

17. Convolutional Codes and the Viterbi Algorithm Noisy versions of the sent bits are received. Each transition in a trellis is assigned a “likelihood” Viterbi Decoder finds out what’s the most likely path through the trellis, yielding the sequence of data bits 00 10 00 01 10 11 00 01 10 11 00 01 10 11 00 01 00 101100

18. Viterbi Algorithm (1) How do we get from Metula to Eilat in the shortest way? A lot of paths to check Solution – do it stage by stage Metula Haifa Tveria Tel Aviv Jerusalem Ashkelon Beer Sheva Eilat 100 120 140180 65 120 11090 70 120 210 260

19. Viterbi Algorithm (2) Start with Haifa and Tveria – this is trivial: –120 Km to Haifa –70 Km to Tveria Metula Haifa Tveria Tel Aviv Jerusalem Ashkelon Beer Sheva Eilat 100 120 140180 65 120 11090 70 120 210 260

20. Viterbi Algorithm (3) The shortest route from Metula to Tel Aviv: –Through Haifa: 120+100=220 Km –Through Tveria: 70+180=250 Km –Choose through Haifa! From Metula to Jerusalem –Through Haifa – 260 Km –Through Tveria – 190 Km Metula Haifa Tveria Tel Aviv Jerusalem Ashkelon Beer Sheva Eilat 100 120 140180 65 120 11090 70 120 210 260 220190

21. Viterbi Algorithm (4) The shortest route from Metula to Ashkelon: –If through Tel Aviv, then also through Haifa: 220+65=285 Km –If through Jerusalem, then also through Tveria: 190+90=280 Km –Choose through Jerusalem! From Metula to Beer-Sheva –Through Tel Aviv – 330 Km –Through Jerusalem – 310 Km Jerusalem is better in both cases Metula Haifa Tveria Tel Aviv Jerusalem Ashkelon Beer Sheva Eilat 100 120 140180 65 120 11090 70 120 210 260 280 310

22. Viterbi Algorithm (5) Finally, to Eilat: –If through Ashkelon, then also through Tveria and Jerusalem: 280+260=540 Km –If through Beer-Sheva, then also through Tveria and Jerusalem : 310+210=520 Km –Choose through Jerusalem! Conclusion: the sortest route is Metula – Tveria – Jerusalem – Beer-Sheva – Eilat Shortest Distance – 520 Km Metula Haifa Tveria Tel Aviv Jerusalem Ashkelon Beer Sheva Eilat 100 120 140180 65 120 11090 70 120 210 260 520

23. QAM to metrics conversion For each bit a metric (sign+likelihood) is extracted The strength of the subcarrier is weighted into the likelihood estimation MSB LSB Middle bit

24. Interleaving Convolutional Codes work well with scattered errors, and perform badly with clustered errors Adjacent subcarriers typically fade together The coded bits are interleaved (reordered) prior to transmission Upon reception, metrics are deinterleaved (reordered back) prior to Viterbi decoding

25. OFDM Advantages For a long multipath – relatively low computational complexity –The FFT algorithm has log(N) ops/sample complexity Integrates well with Error Correction Coding Extreme robustness in multipath

26. OFDM Disadvantages The time domain waveform is noise-like –Large peak-to-average ratio (crest factor) Dictates large Power Amplifier backoff Long symbols impose higher sensitivity to oscillator instabilities – offsets and phase noise

27. OFDM in Broadcasting Standards Digital Audio Broadcasting –192-1536 subcarriers –QPSK modulation, Convolutional ECC –1.536 MHz total bandwidth Digital Video Broadcasting –1705 or 6817 subcarriers –QPSK, 16QAM or 64QAM –Convolutional +Reed-Solomon ECC

28. Single Frequency Network All broadcasting transmitters operate at the same frequency and transmit same data Received signals have “artificial multipath” Receiver Tx 1 Tx 2 Tx 3

29. ADSL: Asymmetric Digital Subscriber Line – OFDM over copper The copper twisted pairs exhibit a response with long tail in time domain. Static channel – power and constellation is negotiated per subcarrier according to SNR The OFDM is at baseband, not at radio frequencies 512 subcarriers, up to 32K-QAM Trellis coded modulation for ECC

30. The 802.11a +HIPERLAN/2 High Speed Physical Layer for the 5 GHz band

31. Frequency Allocations USA –5.15-5.25 GHz (50 mW, indoor) –5.25-5.35 GHz (250 mW) –5.725-5.825 GHz (1 W) Europe –5.15-5.35 GHz (100 mW) –5.47-5.725 GHz (1 W) –Only for the HIPERLAN devices Japan –5.15-5.25 GHz (?) –MMAC W-Eth WG will adopt 802.11a

32. Main Parameters 20 MHz channel spacing –16.6 MHz signal bandwidth –5 MHz grid for various regulatory domains Multiple data rates- 6 to 54 Mbit/s –support of 6, 12 and 24 Mbit/s rates is mandatory OFDM modulation –BPSK, QPSK, 16QAM or 64QAM on each subcarrier –pilot assisted coherent detection 802.11 multirate mechanism support

33. Channelization in US

34. Data and Pilot subcarriers 52 non zero subcarriers, spaced 312.5 KHz –48 data subcarriers –4 pilot subcarriers Center frequency subcarrier not used –leakage in quadrature modulators may corrupt the data

35. OFDM Frame Structure Carrier spacing is 312.5 KHz Fourier transform performed over 3.2 μsec 0.8 microsecond Guard Interval for ISI rejection

36. Error Correction Coding ECC is a must - some subcarriers may fade Bit Interleaved Convolutional Coding used –more robust than trellis in Rayleigh fading Industry standard K=7, R=1/2 code –higher coding rates (2/3, 3/4) derived by puncturing –tail zero bits added to message (trellis termination) Interleaver spans one OFDM symbol –latency and complexity considerations

37. Supported Rates and Modulations Modulation of the data subcarriers by either –BPSK, QPSK, 16QAM or 64QAM –1, 2, 4, or 6 bits/subcarrier, correspondingly

38. Preamble Structure Short sequences in the beginning –Signal Detection, AGC convergence, Diversity resolution, Timing estimation, Coarse frequency estimation Long sequences with Guard Interval –Fine frequency estimation, Channel Estimation t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 Short training sequence GI2 T1 T2 Long training sequence GI SIGNAL GI DATA1 GI DATA2 10*0.8  sec=8.0  sec1.6+2*3.2  sec =8.0  sec 0.8+3.2  sec = 4.0  sec 4.0  sec Signal detection AGC convergence Diversity selection Coarse freq. offset estimate Fine timing acquisition Fine freq. offset estimation Channel estimation RATE and LENGTH Received at 6 Mbit/s DATA is received at RATE indicated in the SIGNAL field

39. Multirate Mechanism Support Each Frame has a SIGNAL field in the beginning –Contains RATE and LENGTH The SIGNAL is transmitted at the most robust data rate (6 Mbit) –The rest of the packet is at the RATE indicated Even if the receiver does not support the RATE of the signal is too weak, it can predict how long the packet will last AP can communicate with different stations at different rates

40. Transmitter Performance Specs

41. Timing Related Specs Short InterFrame Space (SIFS) is 16 usec –Finish decoding, decide you need to reply and put a signal on air Slot Time is 9 usec –Listen, decide that there is no signal and only then transmit DataACK SIFS Data DIFS+n*Slot DIFS=SIFS+2*Slot Second station detects signal and does not transmit Short IFS for ACK gives it priority - second station yields

42. SUMMARY OFDM is an excellent choice for channels with long multipath OFDM has disadvantages with power efficiency and with phase noise tolerance OFDM finds applications in many areas: –Broadcasting (DAB, DVB-T) –WLANs (802.11a, HIPERLAN/2) –Baseband over copper (ADSL)