Bitrate Maximizing TimeDomain Equalizer Design for DMTbased Systems

Bitrate. Maximizing Time-Domain Equalizer Design for DMT-based Systems Koen Vanbleu Promotor : Marc Moonen Coauthors: Geert Ysebaert , Gert Cuypers , Katleen. Van Acker KULeuven, ESAT SCD-SISTA, Belgium ICC 2003

Overview • ADSL Basics � What? � Transmitter/Receiver • ADSL Equalizer Design � Problem Description � Current Equalizers � Bitrate. Maximizing Equalizers • Conclusions 2

Introduction • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions • Communication at high rates towards customer � telephone wire, cable, fiber, wireless • Communication over telephone wire � Evolution: ever increasing bitrates � E. g. Time to download 10 Mbytefile Modem 56 Kbps Voice band modem 3 Time 24 minutes 128 Kbps ISDN 10 minutes 6 Mbps ADSL 13 seconds 52 Mbps VDSL 1. 5 seconds

Introduction • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Broadband communication over telephone line � ADSL (Asymmetric Digital Subscriber Line) � VDSL (Very high bit rate Digital Subscriber Line) � Bitrateis function of the line length Downstream • Conclusions Central 4 Customer Upstream Down Up Line length Frequency band ADSL 6 Mbps 640 Kbps 3 km 1. 1 MHz VDSL 52 Mbps 6. 4 Mbps 300 m 8. 8 MHz

Modulation and. Duplexing • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions • Multicarriermodulation scheme: Discrete Multitone(DMT) e. g. ADSL POTS 4 25 UP DOWN 138 POTS UP &DOWN 1104 f (k. Hz) 4 25 138 DOWN 1104 f (k. Hz) • Assign different frequency bins to up- and downstream directions � Frequency Division Duplexing (FDD) � Overlap: Echo Cancellation (EC) 5 • Traditional telephony (POTS) still available over the sa wire.

Discrete Multi Tone: Transmitter • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions bits Data symbols (QAM) Im 10 Cyclic Prefix. . . • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver 00 CP 0 Re 11 . . . 01 2 bits N-point . . . IFFT P/S Im 6 4 bits . . . Re IFFT modulation (Inverse Fast Fourier Transform)

Why Equalization? • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • Conclusions . . . • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers . . . N-point . . . IFFT CP channel P/S . . . Transmitter 7 noise Why equalization? “Invert” channel distortion while not boosting noise

Discrete Multi Tone: Receiver • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver Frequency Domain Equalizer CP noise . . . • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers Unbiased 1 tap / tone • Conclusions . . . channel h TEQ w Time Domain Equalizer S/P N-point FFT . . . Data symbols bits 10 Im 00 Re FEQ . . . 11 01 2 bits Im Re taps FFT demodulation 8 CP length + 1 4 bits

DMT Equalization: Problem Description • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers T taps TEQ w CP 1 tap/tone S/P . . . N-point. . . FEQDn. . . FFT To maximizebitrate: • Conclusions where 9 is hard withtime-domainequalizer w 10 Im{X} 00 Re{X} 11 01 2 bits

Current ADSL Equalizers (1) • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions • Channelshorteners , e. g. MMSE-based TEQ [Al-Dhahir, Cioffi] noise Channel h delay TEQ w TIR b TIR = target impulse response of (CP-length+1) 10 MMSE criterion ADSL bitratemaximization

Current ADSL Equalizers (2) • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Approximate. Bitrate. Maximizing. TEQs [Al-Dhahir], [Evans] T taps CP TEQ w S/P 1 tap/tone. . . N-point. . . FFT FEQ Dn. . . Maximize • Conclusions Approximations! based on SNR at. FFT output 11 residual ISI/ICI noise

Current ADSL Equalizers (3) • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • Approximate. Bitrate. Maximizing. TEQs (continued) • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions residual ISI/ICI noise Examples ofapproximations : • • Signal component in SNR does not only depend on `windowed’ (shortened) channel impulse response • 12 : do not forget DFT leakage (ICI)!

Bitrate. Maximizing Equalizers (1) • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers T taps CP TEQ w S/P . . . N-point. . . FFT FEQ Dn. . . Maximize • Conclusions where AND then 13 = residual ISI/ICI+noise sources (XT, RFI, …)

Bitrate. Maximizing Equalizers (2) • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions Exact bitratemaximizing (BM-)TEQ cost function • Nonlinear cost function in w only with An and Bn • tone dependent matrices • function of signal statistics • Recursive Gauss-Newton updating algorithm: • attains good local optimum • adaptivity (to track channel/noise changes) • however: high complexity 14

Bitrate. Maximizing Equalizers (3) • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions HIGHER BITRATE 15 Bitrate maximizing equalizers: • Bitrate maximizing (single) TEQ: NG = all used tones nonlinear cost function • “Per group” equalization: BM-TEQ w per group SG of NG tones nonlinear cost function • “Per tone” equalization: NG = 1 tone (advantageous) linear MMSE problem Leus, Moonen] [Vanacker,

Bitrate. Maximizing Equalizers (4) • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions Simulations Per-Tone Equalizer Bitrate. Maximizing-TEQ MMSE-based. TEQs Approx. Bitrate. Max. TEQs 16

Conclusions • ADSL Basics - Intro - DMT Transmitter - Why Equalization? - DMT Receiver • ADSL Equalizer Design - Problem Description - Current Equalizers - Bitrate Maximizing Equalizers • Conclusions 17 • ADSL Equalizer Design • Truly Bitrate. Maximizing Per-Group Equalizer • Time-Domain Equalizer (1 group of tones) • Per-Tone Equalizer (groups of 1 tone) • Recursive Gauss-Newton algorithm • Good local optimum • Complex