Partial Response Signaling for Backplane Applications

Transcription

1 Partial Response Signaling for Backplane Applications IEEE 82.3ap Task Force September 24 Michael Altmann Fulvio Spagna IEEE 82.3ap Task Force - 24-Sep-4

2 Agenda Introduction Line coding alternatives for BP Ethernet S Current candidates S Partial Response codes Partial response line codes S History S Time & spectrum characteristics S Performance against other codes Slide 2 IEEE 82.3ap Task Force - 24-Sep-4

3 Introduction Low speed interfaces generally use NRZ signaling for simplicity At higher speeds, Rx data eye is closed and equalization is needed to reliably recover data Many line codes are possible to assist equalization and reduce equalizer complexity Useful line code properties S Reduce symbol rate S Reduce spectral content at high frequencies Reduce EMI Reduce Xtalk S DC Balancing Reduce baseline wander from series cap S Coding redundancy Error detection/correction possibility In-band control channel Slide 3 IEEE 82.3ap Task Force - 24-Sep-4

4 82.3ap Physical Channel ATCA - SDD SDD21 [db] Frequency [GHz].6 NRZ pulse resp.5 PR4 pulse resp.8 PAM4 pulse resp time [UI] time [UI] time [UI] Slide 4 IEEE 82.3ap Task Force - 24-Sep-4

5 Current Line Code Proposals NRZ S Obvious choice, simplest to implement for high speed links Minimum of 3 comparators to decode data & control equalizer adaption S Can be coded to alter spectrum & control disparity Requires increased bit-rate and channel BW S Requires an equalized channel BW > bit-rate/2 S Sensitive to deep channel nulls <8GHz 4-PAM (generic: M-PAM) S Multi-level signaling S Requires 2M-1 comparators to extract data and timing S 5% BW requirements from NRZ S Nyquist filtering at bit-rate/4 possible to limit Xtalk Duobinary S Current proposals considered as min design delta for NRZ solutions Slide 5 IEEE 82.3ap Task Force - 24-Sep-4

6 Partial Response Codes Long history in disk drive industry Simplest classes use short generator polynomials S PR-2 (duobinary) and PR-4 (modified duobinary) Intentionally adds ISI to shorten channel pulse response Adds levels compresses spectrum S Adds redundancy to easily enable ML/Viterbi detection Slide 6 IEEE 82.3ap Task Force - 24-Sep-4

7 Partial Response Codes Partial response codes use correlation to cancel ISI S Unit data pulses are modified with specific filter functions S PR2: Y=(1+D)X S PR4: Y=(1-D 2 )X S Ref: Partial Response Signaling, Kabal & Pasupathy, IEEE Trans on Comms, Sept 75 Duobinary (PR2) S Uses channel ISI to self-cancel S Natural spectral null at bit-rate/2 S Less critical Nyquist filtering due to spectral null Modified Duo-binary (PR4) S Better ISI cancellation than PR2 (1-D 2 polynomial) S Spectral null at DC (balanced) and bit-rate/2 Slide 7 IEEE 82.3ap Task Force - 24-Sep-4

8 Line Code Spectra PAM Codes Line codes characterized by both spectral & time-domain properties Line code=nrz Line code=pam Spectral Power [dbmw/hz] Freq [Hz] x 1 1 Spectral Power [dbmw/hz] Freq [Hz] x Filtered Eye Diag 2 Filtered Eye Diag [V] [V] Time [s] x Time [s] x 1-1 Slide 8 IEEE 82.3ap Task Force - 24-Sep-4

9 Line Code Spectra PR Codes Line code=pr2 Line code=pr4 Spectral Power [dbmw/hz] Freq [Hz] x 1 1 Spectral Power [dbmw/hz] Freq [Hz] x Filtered Eye Diag 2 Filtered Eye Diag [V] [V] Time [s] x Time [s] x 1-1 Slide 9 IEEE 82.3ap Task Force - 24-Sep-4

10 Coding Evaluation Structural systems simulation S sampled Tx S Linear channel, NEXT/FEXT, equalizer S Sampled Rx Evaluate potential line code performance in comms channel model S SNR evaluation for 4 signaling candidates (NRZ, PAM4, PR2, PR4) S NEXT, FEXT, AWGN major noise sources. Treated as statistically uncorrelated Gaussian noise S Calculate SNR-optimized equalization for each line code Slide 1 IEEE 82.3ap Task Force - 24-Sep-4

11 Evaluation Setup Transmitter Transmitter Transmitter Transmitter Transmitter Transmitter Transmitter e k fextfrsp(jω) nextfrsp(jω) + Σ - din k + txfrsp(e jω ) chfrsp(jω) rxfrsp(jω) fwefresp(e jω ) Slicer Σ Σ en(jω) T Σ - b k b' k channelinput(t) channeloutpu(t) fdefresp(e jω ) samplerinput(t) sampleroutput k detectorinput k Impulseresponse of combinedchrxtransfer functions priorto sampling Impulse responseof combinedch Rxtransfer functionsafter sampling t Slide 11 IEEE 82.3ap Task Force - 24-Sep-4

12 Channel Templates vs. NRZ and PAM4 Spectra Backplane insertion loss NRZ spectrum Near end crosstalk PAM4 spectrum Far end crosstalk f /f Slide 12 IEEE 82.3ap Task Force - 24-Sep-4

13 Equalization Targets and Optimal Corrections 2 2 [db] - 2 [db] f/f Target equalization f/f Optimal correction NRZ PR4 Duobinary PAM4 Slide 13 IEEE 82.3ap Task Force - 24-Sep-4

14 Equalization Targets and Optimal Corrections 2 2 [db] - 2 [db] f/f Target equalization f/f Optimal correction NRZ PR4 Duobinary PAM4 Slide 14 IEEE 82.3ap Task Force - 24-Sep-4

15 Coding Performance Measures PR Equalization S PR2 requires generally less signal boost then PR4 S PR4 amenable to half-rate Rx & Tx architectures Properly equalized, it has no odd-even bit interaction Data recovery S MLSE data recovery to enable error detection/correction Implementation complexity S Adds weighting range for Tx/Rx equalizer S Spectral null2 at f bit /2 reduces noise filter sensitivity Other Coding weighting considerations for line codes S Timing recovery sensitivity S ISI sensitivity Slide 15 IEEE 82.3ap Task Force - 24-Sep-4