Keysight Technologies 2018.01.31 Greg LeCheminant / Robert Sleigh
Introduction Why use Pulse Amplitude Modulation 4-Level (PAM4)? Review Standards using PAM4 Output (Transmitter) Characterization Key Optical PAM4 Measurements Key Electrical PAM4 Measurements Keysight Solutions Summary Q&A 2
E N A B L E S H I G H E R D ATA T H R O U G H P U T NRZ > 28 Gb/s limits trace length or increases cost PAM4 packs 2 bits / symbol Halves the channel BW needs Allows designers to develop products to fit cost structure of available channel technologies 4 amplitude levels 2 bits of information in every symbol ~ 2x throughput for the same Baud rate 28 GBaud PAM4 = 56 Gb/s Lower SNR, more susceptible to noise More complex TX/RX design, higher cost 3
D E S I G N A N D M E A S U R E M E N T Packing 4 levels into amplitude swing of 2 lose 9.6 db SNR It is not just about timing jitter budgets anymore! Better management of noise and return loss Finite rise time creates inherent DDJ How to implement clock recovery? Closed eyes with lower SNR FEC often required How is mask testing done on such an eye diagram? Or what replaces eye mask testing? What are the implications of using FEC on how components and systems are tested?... 4
400G Class (8 lanes of 26 GBd, 4 lanes of 53 GBd) Status IEEE 802.3bs 200/400 GbE (Draft 3.5) Medium reach SMF + C2C (chip-to-chip), C2M (chip-to-module) PAM4 @ 26.6 GBd and 53.1 GBd OIF CEI-56G (OIF-CEI 4.0) 5 reaches (USR, XSR, VSR, MR, LR) Electrical PAM4 up to 29 GBd, NRZ up to 58 Gb/s IEEE 802.3cd: 50/100/200 GbE (Draft 3.0) Short reach MMF, 50G/100G using SMF, C2C,C2M, backplanes & cables. PAM4 @ 26.6 GBd 64G Fibre Channel Short & medium reach in MMF and SMF PAM4 @ 28.9 GBd IEEE 50/100/200/400G >10 km SMF 600G Class (1, 2, 4, 8 and 12 lanes of 26.56 GBd PAM4) Status InfiniBand High Data Rate (HDR) 50 Gb/s to 600 Gb/s 1, 2, 4, 8 and 12 lanes Technical work complete, will be published in spring 2018. Technical work complete. Published Dec 29, 2017. http://www.oiforum.com/wp-content/uploads/oif-cei-04.0.pdf Note: CEI-56G-XSR-PAM4 needs more technical discussion and was not included in 4.0 Entering 3rd major draft cycle, firming up (Fixing some leftover problems in 802.3bs) Second complete draft (some time to go) Study group concluding, nearing project start Active development (Rev 1.4). Highly leveraged from IEEE 802.3bs/cd and OIF CEI-56G-VSR-PAM4. 800G Class (Next generation-100g lane rate) Status OIF CEI-112G 4 reaches of 56GBd PAM4, CNRZ-5 Project starts for 3 reaches, C2M complete draft 5
T Y P I C A L I M P L E M E N TAT I O N Ethernet Switch using 400GBASE-FR8 Optical Link. Both IEEE and OIF-CEI are used. Switch Card Backplane Line Card 400G-FR8 Module Switch ASIC n Retimer n Retimer n Host ASIC 8 Retimer ROSA TOSA 400G-FR8 Module 8 Retimer ROSA TOSA CEI-56G-VSR PAM-4 or NRZ CEI-56G-LR PAM- 4 or enrz CEI-56G-MR PAM-4 or NRZ 400GAUI-8 8 x 26 GBd PAM-4 (8 x 56 Gb/s) 400GBASE-FR8 8 λ WDM in SMF 6
7
L E T S R E V I E W N R Z F I R S T! Extinction Ratio Ratio of eye one level to zero level (how efficiently is available laser power used as modulation power) Optical Modulation Amplitude (OMA) Difference of the one and zero level (A measure of the actual modulation power) Eye-Mask A measure of the relative opening of the transmitter eye diagram Transmitter Dispersion Penalty (TDP) Compare the system BER versus an ideal transmitter (a power penalty metric) 8
I N S O M E C A S E S I T W O R K S, A N D I N O T H E R S I T D O E S N T Extinction ratio and OMA for PAM4 NOT measured on the aggregated PAM4 eye diagram Measured on the zero level and the three level Measured on specific bit sequences, so the scope MUST be pattern locked Long patterns like PRBS31 are no longer allowed (PRBSQ13 and SSPRQ are used) New measurement names to distinguish PAM4 method Outer OMA Outer Extinction ratio 9
Eye closure is expected to be severe, requiring receiver equalization, reducing the value of an eye mask It is certainly possible to have a PAM4 eye-mask, but the standards committee felt that it would not be a very good predictor of performance in an actual system Emphasis was placed on the transmitter dispersion measurement, as with both NRZ and PAM4 it is a better predictor of system level performance 10
T R A N S M I T T E R D I S P E R S I O N A N D E Y E C L O S U R E Q U AT E R N A R Y Tells you the performance of your transmitter relative to an ideal transmitter For NRZ TDP, we literally measured the BER performance of the transmitter compared to an actual golden transmitter How much extra power was required at the receiver to compensate for non-ideal performance For TDECQ we indirectly measure SER (symbol error ratio) 11
Takes advantage of the very high target SER (2e-4) Statistically determine the SER directly on the signal captured with an oscilloscope Rather than attenuate the signal to force errors, mathematically add noise --------- to create errors. Increase the added noise (---------) until the target SER is observed Repeat for the transmitter virtual ideal transmitter The amount of noise that must be added to the ideal reference signal to reach the target SER will generally be larger than the noise added to the signal being tested. The db difference in noise levels represents the power penalty for the transmitter under test Added noise = TDECQ Transmitter Keysight under test Confidential Virtual ideal transmitter 12
Early versions Optimized virtual equalizer to minimize the spread of the eye levels (e.g. open up the eye) T/2 spaced virtual equalizer Classic Bessel-Thomson scope BW (BW = 75% of baud rate) Measurement made at two time slices at the eye center Final version Optimized to minimize the TDECQ penalty T-spaced virtual equalizer Nyquist scope bandwidth (50% of baud rate) Measurement time position allowed to be optimized for minimum TDECQ penalty In each case, the changes were made to better represent the typical system that the transmitter would operate in. In most cases this is relative to how real system receivers will operate 13
Consider anything that could possibly degrade the SER (think like a receiver!) Linearity: If the signal levels are not proportionally distributed, the decision thresholds at the receiver may not be optimally set Skew: If the three eyes are not aligned in time, the receiver may not be making decisions at the optimal time Noise margin: The absolute noise that can be added to the signal to reach the target SER (an absolute value, not relative to an ideal transmitter) Advanced TDECQ analysis helps to isolate what signal impairments are dominating the overall TDECQ result Partial TDECQ: The TDECQ contribution for each eye Partial SER: The effective SER for each eye Partial Noise margin: the noise margins for each eye 14
15 15
Samples that lead to errors are shown in red 0/1 and 2/3 eye 16 16
Total TDECQ at 1.87 db Dominated by right side of 0/1 and 2/3 eyes 1.98 and 2.15 db 17 17
M Y S Y S T E M W O R K S B U T T H E T R A N S M I T T E R F A I L S T H E T D E C Q S P E C The TDECQ specification is set to ensure that if the transmitter is paired with the worst case allowed channel and receiver, the system SER will be achieved If receivers perform much better than required, the transmitter performance conceivably could be worse than allowed by the standard and still achieve system level requirements As the TDECQ measurement is used, discussions continue in IEEE 802.3cd (50/100/200 Gb/s links) on whether the current TDECQ virtual receiver definition correctly emulates what will be used in real systems. 18
19
I E E E 8 0 2. 3 B S A N N E X 1 2 0 D 2 0 0 G A U I - 4 A N D 4 0 0 G A U I - 8 What are the key PAM-4 TX parameters that get measured? Output waveform o Level Separation Mismatch Ratio Signal-to-noise-and-distortion ratio (SNDR) (Note there is still some debate over SNDR specs in IEEE 802.3cd) Output Jitter o J RMS o J4u o Even-Odd Jitter (EOJ) IEEE 802.3 participants can access the latest draft specs on the 802.3 website. Reference: IEEE P802.3bs /D3.5, 10th October 2017, page 352. 20
120D.3.1.2 LEVEL SEPARATION MISMATCH RATIO, R LM ( 2 0 0 G A U I - 4 A N D 4 0 0 G A U I - 8) How evenly spaced are the PAM-4 levels? Defined as a function of the mean signal level transmitted for each PAM4 symbol level (Annex 120D) Significant change in definition from IEEE 802.3bj (100GBase-KP4) Clause 94 Vmid V 3 V 2 V 1 V 0 Measurement Setup: Receiver: 4 th Order Bessel-Thomson low-pass filter with 33 GHz BW CR PLL BW 4 MHz and a slope of 20 db/decade Tested using PRBS13Q (no longer uses the stair-step pattern) ES = Effective Symbol Level R LM = min ((3 x ES1), (3 x ES2), (2 3 x ES1), (2 3 x ES2) 21
I E E E P 8 0 2. 3 B S / D 3. 5, 1 0 T H O C T O B E R 2 0 1 7, C L A U S E 1 2 0 D. 3. 1. 6 Procedure: 1. Measured at the output of TX with all lanes enabled 2. Capture PRBS13Q waveform (lane under test). 3. Import into math program and perform matrix math. a. Compute: i. Determine Linear fit pulse response, p(k). Measure Pmax ii. iii. Determine Linear fit error waveform, e(k). Measure σ e, the standard deviation of e(k). Measure RMS deviation from mean voltage on the flattest portion of at least 6 consecutive PAM4 symbols. Compute σ n b. Calculate SNDR using formula. Measurement Setup: Receiver: 4 th Order Bessel-Thomson low-pass filter with 33 GHz BW CR PLL BW 4 MHz and a slope of 20 db/decade Note - SNDR is very sensitive to noise measurement (ensure to use a low noise scope) = 22
J4u, J RMS, and EOJ you may recognize some of these acronyms from other (older) Standards So they should be pretty straight forward to measure, right? While the IEEE Output Jitter names may sound familiar, they are measured very differently! o Traditional Jn (e.g. J5, J9) and EOJ parameters were measured using all edges of an NRZ pattern. o IEEE 802.3bs now measures J4u, J RMS, and EOJ on 12 specific edges of a PRBS13Q (PAM4) pattern! Reference: IEEE P802.3bs /D3.5, 10th October 2017, page 357. So don t just pull out your scope and press the J4 or EOJ button in Jitter Mode! 23
Different TX Architectures are used to generate PAM4 signals Some TX designs may use different clock buffers for MSB and LSB; this can result in different uncorrelated jitter appearing on different edges. Measuring jitter only on JP03 (clock) patterns (original method) could miss potential issues. PAM4 TX PAM4 24
200GAUI-4 and 400GAUI-8 transmitter characteristics at TP0a (120D.3.1.1) Acquire data for 12 specific edges of a PRBS13Q (12 sets of data should be of equal size) Initially very time consuming to make measurements since Each histogram should include at least 10 6 hits. o o In a Jan 2017 IEEE ad hoc meeting, Keysight presented advanced methods to speed up test times while maintaining accuracy. Reference: http://grouper.ieee.org/groups/802/3/bs/public/adhoc/elect/30jan_17/lecheminan t_01_013017_elect.pdf o o RT Scopes can acquire entire patterns very quickly, but only a few samples of each edge are valid in a waveform, so many pattern acqs are required Sampling scopes can target edges very efficiently, but sampling speeds are relatively slow (ksa/s vs GSa/s) Histogram size changed: Size of all sets should be chosen to enable calculation of J4u with sufficient accuracy. Typically > 300k samples works well. New acquisition methods were accepted (edge model technique) Amplitude Edge Model Amplitude-to-Time (Jitter) transfer function ~ 100% efficiency Sample 1 Jitter Sample 2 25
200GAUI-4 and 400GAUI-8 transmitter characteristics at TP0a (120D.3.1.1) New 12 edge jitter method in FlexDCA reduced test time from hours to << 1 minute. J4u and JRMS jitter o Measure RJ/PJ on 12 specific transitions using a PRBS13Q pattern (exclude correlated jitter). o Data from all edges is combined and analyzed Even-Odd Jitter (EOJ) o Measured on PRBS13Q (3 repeats) o Max from measurements on all 12 edges FlexDCA (Option 200 Jitter with Option 9FP PAM4 Analysis) reports : J4u, J RMS, EOJ ALL measurement (per the Standard) Plus individual results for each of the 12 edges o Rise: 0 to 3, 1 to 2, 0 to 1, 2 to 3, 0 to 2, 1 to 3 o Fall: 3 to 0, 2 to 1, 1 to 0, 3 to 2, 2 to 0, 3 to 1 Measurement Setup: Receiver: 4 th Order Bessel-Thomson low-pass filter with 33 GHz BW CR PLL BW 4 MHz and a slope of 20 db/decade Contact Keysight for info on 12 edge jitter measurement capabilities for RT scopes. 26
System Board Level Chip Validation / Verification Compliance Troubleshooting Sampling Scopes (SS), DCA Best for validating/characterizing PAM4 designs For applications that place top priority on waveform precision. Highest Fidelity Low noise Ultra-low jitter High Bandwidth Highest Resolution (14-16 bits) Modular Platform Electrical Optical TDR/TDT Lowest price for same BW 86100D-9FP PAM4 Analysis SW N1085A PAM4 Pre-Compliance SW Real-time Scopes (RT) Best for troubleshooting PAM4 designs The most versatile tool for all areas of high-speed digital communications Best for troubleshooting Captures one-time (glitch) events No explicit trigger required Does not require repetitive signals for pattern waveform measurements. N7004A Optical-to-Electrical Converter N8827A/B PAM4 Analysis SW N8836A PAM4 Pre-Compliance SW PAM4 SER/BER Error Capture and Decode capabilities 27
28
U S E Y O U R E X I S T I N G H A R D W A R E S O L U T I O N S T O T E S T PA M 4 D E S I G N S New Electrical Sampling Scope (includes built-in clock recovery and precision timebase) Keysight 86100D DCA-X with 86108B Channels: 2 Bandwidth: 50 GHz Jitter: <45 fs rms typ. Electrical Clock Recovery integrated HW Clock Recovery works with PAM-N signals up to 32 Gbaud 86100D-9FP PAM-4 Analysis SW (works with any DCA module, optical or electrical) N1085A PAM-4 Pre-Compliance SW Note - Remote head modules with 110 GHz bandwidth are also available. NEW 64 GBd electrical clock recovery with internal variable EQ now available! Stop by our booth! New Electrical Real-time Scope Keysight DSO Z-Series Channels 2-4 Bandwidth: up to 63 GHz Sample Rate: Up to 160 GSa/s PAM-4 Serial Data Analysis Wizard Software Clock Recovery N8827A/B PAM-4 Analysis SW N8836A PAM-4 Pre-Compliance SW PAM-4 SER/BER Error Capture and Decode capabilities Use as a PAM4 Error Detector with M8040A BERT software 29
F U L L C O V E R A G E F O R A L L O P T I C A L PA M 4 A P P L I C AT I O N S ( U S I N G E X I S T I N G K E Y S I G H T O S C I L L O S C O P E S O L U T I O N S ) N1092A/B/D DCA-M 86100D DCA-X with 86105D/86115D module 86100D DCA-X with 86116C module N7004A 33 GHz O/E Converter for RT scopes 20-28 GBd optical channels 53 GBd TDECQ Ref Receiver Multimode and Single-Mode 1, 2 or 4 channels High sensitivity receiver design (low noise receiver) Fastest sampling combined with 160fs typical trigger jitter Lowest cost 4 channel solution Ideal frequency response (SIRC) PAM-4 analysis with TDECQ (Option 9FP/9TP) 8.5 GBd to 28 GBd Multimode and Single-Mode 1 or 2 @ 34 GHz optical channels (1 to 4 optical per mainframe) 1 @ 50 GHz Electrical Channel Ideal frequency response (SIRC) < 100 fs rms timebase jitter (86100D-PTB) PAM-4 analysis with TDECQ (Option 9FP/9TP) 25/26/28GBd or 53/56 GBd Single-Mode 1 @ optical channel per module 1 @ 80 GHz Electrical Channel Ideal frequency response (SIRC) < 100 fs rms timebase jitter (86100D-PTB) PAM-4 analysis with TDECQ (Option 9FP/9TP) Fully-integrated optical-to-electrical (O/E) converter, compatible with Infiniium V-Series, 90000X/Q, and Z Series scopes. Up to 28 GBd, multimode and single-mode Ideal for system level R&D debug and troubleshooting Includes built-in measurements for optical signals Stop by our booth and ask us about compliant clock recovery solutions for 26/53 GBd applications! 30
Perform basic PAM4 measurements (Eye Width/Height, Linearity, Skew, etc.) using measurement capability built into the baseline FW: 86100D-9FP for the 86100D DCA-X N8827A/B PAM-4 Analysis Software for RT scopes 31
P R E - C O M P L I A N C E S W A P P S F O R E M E R G I N G S TA N D A R D S U S I N G PA M 4 N1085A PAM-4 Measurement App for Ethernet and OIF-CEI (for the 86100D DCA-X) N8836A PAM-4 Measurement App for Ethernet and OIF-CEI (for Infiniium real-time scopes) We are actively updating PAM4 algorithms as the Standards evolve. Updates will be included in future SW releases. 32
Transition from NRZ to PAM4 is revolutionary Many new challenges in both electrical and optical links Many new measurements, including: Optical: TDECQ, Outer OMA, Outer ER Electrical: SNDR, Linearity on PAM4 signals, Output Jitter (J4u, J RMS, EOJ) Keysight provides industry leading tools for fast and accurate characterization of PAM4 designs Tx Characterization Oscilloscopes New Keysight 86100D DCA-X sampling oscilloscope 86100D-9FP PAM4 Analysis Software N1085A PAM4 Measurement Application for Ethernet and OIF-CEI N1076B Electrical Clock Recovery for 125 MBd to 64 GBd applications Keysight N1092X DCA-M sampling oscilloscope Keysight DSAX 63 GHz real-time oscilloscope N8827A PAM4 Measurement Tool With BER N8836A PAM4 Measurement Application for Ethernet and OIF-CEI 33