QAM-Based Transceiver Solutions for Full-Duplex Gigabit Ethernet Over 4 Pairs of UTP-5 Cable Henry Samueli, Jeffrey Putnam, Mehdi Hatamian Broadcom Corporation 16251 Laguna Canyon Road Irvine, CA 92618 Tel: 714-45-87 Fax: 714-45-871 Motivation for Using QAM Passband scheme - no baseline wander effects Mature and well-understood technology Widely deployed in voiceband modems, digital cable-tv set-top boxes, cable modems Public domain technology
QAM Transmitter 3 Square-Root Nyquist Filter Bits Scrambler QAM Mapper 1,,-1,,1,,-1 D/A Lowpass Filter + Line Driver Hybrid Analog Output 3 Square-Root Nyquist Filter F S = 3F B F S = Sampling Rate F B = Symbol Rate QAM Receiver Square-Root Nyquist Filter 3 Analog Input Hybrid + Receive Filter A/D 1,,-1,,1,,-1 Feed-Forward Equalizer Decision Slicer QAM Decoder + Descrambler Bits F S = 3F B Square-Root Nyquist Filter 3 Timing Recovery Decision Feedback Equalizer Echo Canceller TX 1 F S = Sampling Rate F B = Symbol Rate 3 NEXT Cancellers TX 2 TX 3 TX 4
System Assumptions 4 pairs of UTP-5 cable up to 1 meters 25 Mb/s full-duplex per pair Broadcom measured attenuation characteristics scaled to worst-case EIA/TIA models Worst-case NEXT and echo curves from 82.3z reflector No echo attenuation in the analog hybrid Bit accurate simulations Candidate QAM Systems Throughput Goal = 25 Mb/s Constellation Symbol Rate (MBaud) Data Bits Per Symbol Extra Points (for signaling) Sampling Rate (MHz) Center Frequency (MHz) Required (BER = 1-1 ) 6x6 5 5 4 15 37.5 27. 5x5 62.5 4 9 187.5 46.875 25.3 3x3 83.3 3 1 25 62.5 2.5 Key Trade-Offs - Performance margin for BER=1-1 - Implementation complexity - Data converter precision
QAM Spectra 6x6 5 MBaud 5x5 3x3 H(f) H(f) H(f) f f 7.5 37.5 67.5 9.4 46.9 84.4 12.5 62.5 112.5 f System Comparisons 5 MBaud 6x6 System - Lower signal bandwidth and center frequency => lower channel loss and decreased susceptibility to high frequency noise - Lower symbol rate => shorter adaptive filters to cover the same time span - Higher-order modulation => increased precision requirements 5x5 System - Excessive signaling points reduce SNR margin 3x3 System - Higher signal bandwidth and center frequency => higher channel loss and increased susceptibility to high frequency noise - Higher symbol rate => longer adaptive filters to cover the same time span - Lower-order modulation => decreased precision requirements
Shaping Filters 1-1 -2-3 -4-5 -6..2.4.6.8 1. Normalized Frequency 31-tap, 2% excess bandwidth Channel Models h(t).25.2.15.1.5. -.5 -.1-2 2 4 6 8 Time (ns) Measured 1m UTP-5 loss characteristic - Includes attenuation roughness Scaled down by.5 db to match worst-case envelope Impulse response spans ~15 ns -1-2 measured EIA/TIA-568 envelope -3 2 4 6 8 1
Self-NEXT Models h(t).5.4.3.2.1. -.1 -.2 -.3 -.4 -.5-2 2 4 6 8 1 Time (ns) Worst-case UTP-5 self-next from 82.3z reflector Different worst-case models used for each of 3 channels Impulse response spans ~5 ns -1-2 EIA/TIA-568 envelope -3-4 -5-6 -7-8 NEXT models -9-1 2 4 6 8 1 Echo Models h(t).15.1.5. -.5 -.1-1 2 4 6 8 Time (ns) -1-2 -3 echo models -4 2 4 6 8 1 Worst-case characteristics from 82.3z reflector Impulse response spans ~5 ns
Received Spectra -2 signal 5 MBaud signal -2-4 -6-8 noise -4-6 -8 noise -1 25 5 75 1 125 Noise power = echo power + NEXT power High baud rate system appears more susceptible to high-pass noise, but required SNR (BER = 1-1 ) is 6.5 db lower -1 25 5 75 1 125-2 -4-6 -8 signal noise -1 25 5 75 1 125 Equalizer Tap Trade-Off 5 MBaud 32 3 28 26 1 8 6 DFE Taps T-spaced feed-forward equalizer 4 4 5 6 7 8 9 1 FFE Taps Feed-forward taps necessary to compensate precursor ISI plus pulse shaping and analog filtering 5 MBaud and systems require complete post-cursor ISI cancellation for reasonable margin - system can trade-off margin for taps 32 3 28 26 2 15 1 5 DFE Taps 28 27 26 25 24 23 14 12 1 8 DFE Taps 6 4 6 8 1 12 14 16 FFE Taps 4 6 8 1 12 14 FFE Taps
NEXT Canceller Tap Trade-Off 8. 34 SNR Margin for BER=1-1 (db) 6. 4. 2. 5 MBaud 32 3 28 26 5 MBaud 24. 8 16 24 32 4 48 NEXT Canceller Taps 22 8 12 16 2 24 28 32 36 4 44 48 NEXT Canceller Taps Echo Canceller Tap Trade-Off 6. 34 SNR Margin for BER=1-1 (db) 4. 2.. -2. -4. 5 MBaud 3 26 22 18 5 MBaud -6. 16 24 32 4 48 56 64 72 8 Echo Canceller Taps 14 16 24 32 4 48 56 64 72 8 Echo Canceller Taps
A/D Precision 6 35 SNR Margin for BER=1-1 (db) 4 2-2 5 MBaud 3 25 2 5 MBaud -4 4 5 6 7 8 9 1 11 12 13 A/D Bits 15 4 5 6 7 8 9 1 11 12 13 A/D Bits Simulation Summary System 5 MBaud (36-QAM) (25-QAM) (9-QAM) FFE Taps 8 8 8 DFE Taps 8 1 8 NEXT Canceller Taps 2 28 36 Echo Canceller Taps 36 4 56 A/D Precision / Rate 7 bits / 15 MHz 7 bits / 187.5 MHz 6 bits / 25 MHz Relative Hardware Complexity (digital) 1.1 1 1 SNR 3.2 db 28.1 db 24.5 db Margin (BER = 1-1 ) 3.2 db 2.8 db 4. db
Conclusions QAM line coding is well-suited for Gigabit Ethernet Smaller constellation sizes achieve slightly higher SNR margins - Higher speed data converters are required (6-bit 25MHz vs. 7-bit 15MHz) Accurate comparisons of various line codes requires consensus on a common set of simulation models - Echo return loss characteristic is a major factor in determining system complexity