10GE Link Design ForScrambled Encode
David W. Martin
September 27-29, 1999
York, UK
Page 2September 27-29, 1999
Agenda
• Terminology
• Relationships
• Data Recovery
• Clock Recovery
• Jitter Tolerance
• Encode Complexity
• Performance
Page 3September 27-29, 1999
Terminology
• Digital Sum Variance (DSV)— the pk-pk difference between the min & max running sum of the data,
over the sequence length
• Baseline Wander— the variation, in the amplitude axis over time, of the data eye center
• DC Balanced— whether or not the sum of the data, over the sequence length, is zero
(where a data value of zero is counted as -1)
• Run Length— the number of consecutive one’s or zero’s (CID) in the data stream
• Transition Density— the number of data transitions over the sequence length
Page 4September 27-29, 1999
Relationships
• Digital Sum Variance -> Data Recovery— a data pattern with a large DSV can potentially cause baseline wander at
the RX data recovery block if the lower cutoff of the RX passband doesnot capture enough of the low frequency content in the data
• Run Length -> Clock Recovery/Jitter Tolerance— the longer the run of continuous identical digits (CID) in a data pattern
the higher the Q required for the clock recovery filter, for a higher Q(i.e. narrower bandwidth) clock recovery filter the lower the input jitterthat can be tracked by the RX
Page 5September 27-29, 1999
Data Recovery
• Data passband filter...
xPIN
TZA AGC
FrequencyDoubler Clock
RecoveryFilter
LimitingAmp
PhaseShifter
FF
Data passband filter is comprised of the net effect of decoupling cap’s(between circuits with different DC bias points or different technologies)and the upper cutoff of the AGC (or passband filter if no AGC).
RXFrontEnd
Page 6September 27-29, 1999
• The lower cutoff frequency required to avoid baselinewander induced data errors for a SONET signal is inthe 10s of kHz range
• Thus for 10G the capacitors selected need to be wellbehaved over ~6 orders of magnitude
—not a cost issue, just a design choice
• Note that if the RX front-end is integrated, with noAGC/decoupling cap’s, then the required lowfrequency cutoff is achieved inherently (i.e. near-DC)
Page 7September 27-29, 1999
Baseline Wander Power Penalty
4.5
0
SONET fp
FC fp
0.11 10
7�
fp
1 107
1 106
1 105
1 104
1 103
0.01 0.1
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
f1/f0
dB
f1 - lower cutoff frequencyf0 - baud rate
operating pt of SONET data recovery circuitry
The SONET curve is plotted for the 72 CID pattern, which has an average interval between occurrences of ~7500 yrs.
Page 8September 27-29, 1999
• The upper cutoff frequency required for optimumsignal/noise is at about half the NRZ data rate
• Thus for 10G the upper cutoff frequency of theAGC/decoupling cap’s is ~5G
—not a cost issue, just a design choice
Scrambled encode not a cost issue for data recovery
Page 9September 27-29, 1999
Clock Recovery
• The longer the string of CID, the more signal energy is shiftedto harmonics of the fundamental
• The clock recovery filter bandwidth must then be narrower toreject those harmonics and the partner limiting amp must bemore sensitive to pick up the lower energy of the fundamental
fo
fo fo/3 fo/5
0.6
<<0.6
Page 10September 27-29, 1999
• The filter Q=f clk/BWflt
• For 10G SONET a filter with a 3dB BW of 15-20MHz orQ of ~600 is used (2nd order Butterworth)
• A filter with a Q of this order is a commodity
• The partner limiting amp typically has a gain of ~20dB,again a commodity part
Scrambled encode not a cost issue for clock recovery
Page 11September 27-29, 1999
Jitter Tolerance
• The clock recovery filter BW determines the RX inputjitter tolerance high frequency knee
fo
fo fo/3 fo/5
0.6
<<0.6
The wider the clockrecovery filter BW,the higher the kneefrequency of the RXjitter tolerance
Page 12September 27-29, 1999
• Input jitter tolerance (TP3)
Source: GR-1377-CORE (12/98), G.825 (02/99)
Frequency (Hz)
UIpp
10 2.4k 400k24k 4M
0.15
1.5
15
80M
The clock recovery filter BWdetermines the knee frequency
Page 13September 27-29, 1999
• The input jitter tolerance amplitude, in the knee region,from SONET OC-192 long reach can be relaxed for the10GE shorter reach environment by reallocating some ofthe optical impairment jitter budget
Topen - Jbounded - Jtrackable
σRHS + σRHS = Qtarget
where, Jbounded = Jpattern + Jpropagation + Rcvrimpl
Page 14September 27-29, 1999
• Example: Topen= 100ps
Jpattern= 15ps
Jpropagation= 2ps*
Rcvrimpl= 33ps
σRHS=σRHS= 1ps
Qtarget= 10
• Yields… Jtrackable= 30ps, twice the current mask value,thereby relaxing TX jitter generation/easing design
* IR, 10km, cooled DFB 0.3nm, laser tolerance +/-3nm, fiber zero tolerance +/-5nm, fiber dispersion slope 0.07ps/nm/km/nm
Page 15September 27-29, 1999
• Example 10GE jittergeneration (TP2)
Freq(Hz)
UIpp
10 80 k 4M
0.28
~15
80M
Relaxed jitter tolerance possible with scrambled encode
• Example 10GE inputjitter tolerance (TP3)
Freq(Hz)
UIpp
10 80 k 4M
0.30
15
80M
Page 16September 27-29, 1999
Encode Complexity
7 1 64 - 1518 12
IPG Preamble SFD Ethernet MAC frame
Idle frame
X43 + 1 Scrambling
12
Length, Type, HEC
1 2 64 - 1518 2
Type HEC scr Ethernet MAC frameLength
5
unused
10
Type HECLength unused
STS-192cSPE
HH
H
16,640 Octets
STS-192c Payload CapacityFixedStuff
63
9
3
Map into SONET
PCS
PMAPOH
Page 17September 27-29, 1999
4
STS-192c Envelope Capacity
9
90x192 octets3x192
TransportOverhead
H1 H3
A1 A2
HH
H
STS-192c Payload CapacityFixedStuff
POH
PMA
3
Map into SONET
X7 + X6 + 1 Scrambling
Page 18September 27-29, 1999
Note: Processing is at 155MHz, 64-bit wide.
Gate Count Comparison
8B/10B PHY STS-192c PHY Full SONET OH Processing8B/10B encoder: 4080 x43 + 1 scrambler: 90
length generation: 150HEC generation: 200POH generation: 500 full POH generation: 700pointer generation: 0 full pointer generation: 700LOH generation: 300 full LOH generation: 1300SOH generation: 500 full SOH generation: 700x7 + x6 + 1 scrambler: 75
8B/10B decoder: 2730 framer: 500byte align: 300x7 + x6 + 1 descrambler: 75SOH extraction: 500 full SOH extraction: 800LOH extraction: 300 full LOH extraction: 1200pointer processing: 800 full pointer processing: 1700POH extraction: 500 full POH extraction: 1000HEC delineation: 1000x43 + 1 descrambler: 90
8B/10B Total 6810 STS-192c Total 5880 Full OH proc increment 4700
Page 19September 27-29, 1999
Sublayer OH Byte STS-1#1 Notes STS-1#2-192 NotesA1 F6 F6A2 28 28J0 provision 00 (Z0)B1 calculate 00E1 00 00F1 00 00
SOH
D1-3 00 00
H1 62 93H2 0A
NDF idle,522 ptr offset FF
concatenationindicator
H3 00 00B2 00 unused 00 unusedK1 01 NR, CHID=1 00K2 10 CHID=1 00D4-12 00 00S1 F0 “Don’t Use” 00 (Z1)M1 00 unused 00 (Z2)
LOH
E2 00 00
J1 provisionB3 calculateC2 new codeG1 calculate P-REI, P-RDIF2 00H4 00
POH forSTS-192c
Z3-5 00
SummaryIn the TX direction, all OH values are hardwired defaults except:� J0: provisionable� B1: calculated� J1: provisionable� B3: calculated� G1: calculatedIn the RX direction, all OH values ignored except:� A1-2: used by framer� J0: compared against provisioned value� B1: compared against calculated value� H1-3: used by pointer processor� J1: compared against provisioned value� B3: compared against calculated value� G1: extracted
Proposed Minimal SONET OH Usage
Page 20September 27-29, 1999
• The gate count complexity for an 8B/10B or STS-192c PHYis very similar (actually slightly less for STS-192c)
Scrambled encode not a cost issue for digital processing
Page 21September 27-29, 1999
Performance
Compare 8B/10B vs SONET mapping delineation performance
MAC delineation
8B/10B delineation
encap delineation
SONET framer
ptr processor
(Recall no issue with error detection reduction, per Norival Figueira presentation Mtl-July’99)
Page 22September 27-29, 1999
SONET Encap FL due to:• > 1 error in header
SONET Framer FL due to:• > 1 error in frm pattern
SONET Ptr FL due to:• ptr corrupt, data emulation
SON ET com p onent frame loss
1.E -45
1.E -43
1.E -41
1.E -39
1.E -37
1.E -35
1.E -33
1.E -31
1.E -29
1.E -27
1.E -25
1.E -23
1.E -21
1.E -19
1.E -17
1.E -15
1.E -13
1.E -11
1.E -09
1.E -07
1.E -05
1.E -03
1.E -01
1.E + 01
1.E -15 1.E -14 1.E -13 1.E -12 1.E -11 1.E -10 1.E -09 1.E -08 1.E -07 1.E -06 1.E -05 1.E -04 1.E -03 1.E -02 1.E -01 1.E + 00
BER
Pro
ba
bil
ity
of
Fra
me
Fram er_Los s P ointer_Los s E nc ap_Los s
Page 23September 27-29, 1999
8B/10B layer FL due to:• error in /S/, /T/, /R/ characters, data emulation
SONET total vs 8B /10B frame loss
1.E -20
1.E -19
1.E -18
1.E -17
1.E -16
1.E -15
1.E -14
1.E -13
1.E -12
1.E -11
1.E -10
1.E -09
1.E -08
1.E -07
1.E -06
1.E -05
1.E -04
1.E -03
1.E -02
1.E -01
1.E + 00
1.E -15 1.E -14 1.E -13 1.E -12 1.E -11 1.E -10 1.E -09 1.E -08 1.E -07 1.E -06 1.E -05 1.E -04 1.E -03 1.E -02 1.E -01 1.E + 00
BER
Pro
ba
bil
ity
of
Fra
me
L
Frm _P tr_E nc ap_Los s 8B 10B _Los s
Page 24September 27-29, 1999
• Compare performance of proposed SONET mapping (10G) vs8B/10B (12.5G) at 1E-12 BER:
SONET MTTFL=1/2.00E-17/10E09=58 days
8B/10B MTTFL=1/3.17E-11/12.5E09=2.5 sec
SONET mapping MTTFL better than 8B/10B by ~ 2E06 at 1E-12
Page 25September 27-29, 1999
Summary
• Scrambled encode not a cost issue for:— data recovery— clock recovery— jitter tolerance— digital processing (similar to 8B/10B)
• The encode proposal has significant performance advantages:— robust delineation— extended reach/relaxed link budget
• Scrambling at 10G has been field proven since 1995— parts readily available now
• Scrambling can operate over serial/parallel optics, MMF/SMF