10GBASE-T. length of precoding response, and PMA training

Transcription

1 1GBASE-T TxFE solutios, dpsnr vs legth of precodig respose, ad PMA traiig IEEE P82.3a Task Force Austi, May 18-2, 25 Gottfried Ugerboeck 1

2 Cotets Study of trasmit frot-ed solutios Simple : o digital filterig, 8 Ms/s DAC, simple R//C sigal smoothig (f 3dB = 3 MHz), 1:1 trasformer trasmit PSD depeds o iaccurate aalog compoets, o desiged spectral ulls at dc ad 1/2T, poor retur loss. Baselie : o digital filterig, 8 Ms/s DAC, sigal smoothig by RLC frot-ed filter (f 3dB = 3 MHz) with costat output impedace, 1:1 trasformer trasmit PSD depeds o iaccurate aalog compoets, o desiged spectral ulls at dc ad 1/2T, good retur loss. Oversampled : digital filterig ad iterpolatio, 16 Ms/s DAC, simple R//C sigal smoothig (f 3dB = 1 GHz), 1:1 trasformer trasmit PSD exhibits well defied shape with spectral ulls at dc ad 1/2T, good retur loss. 2

3 Cotets Decisio-poit SNR vs. legth of precodig respose Aalysis: fiite-legth precodig respose + ifiite-legth FFE optimized i MMSE sese. Results obtaied for baselie ad oversampled trasmit froted, showig advatages of oversampled solutio. For worst case lik characteristics, a programmable FIR precodig respose of legth L = 32 is foud to be sufficiet; L = 16 leads to small, but oticeable performace degradatio. PMA traiig issues Geeratio of PMA traiig sequeces: proposal for cocise ad uambiguous descriptio. other poits i preparatio. 3

4 Study of trasmit frot-ed solutios 4

5 Trasmitter frot-ed: simple 8 Ms/s DAC Trivial AFE filter f 3dB = 3 MHz 1:1 TH precoder Z I C R V CT Z R Adaptive echo caceller Hybrid fuctio R = 1 Ω C = 1.6 pf 8 Ms/s ADC Receive filter RF 5

6 Trasmitter frot-ed: baselie AFE filter: 1 st -order LPF with costat output impedace, f 3dB =3 MHz 8 Ms/s DAC L/2 1:1 TH precoder Z I C R R/2 R/2 R V CT Z R L/2 R = 1 Ω, C = 5.3 pf Adaptive echo caceller Hybrid fuctio L = 53.1 H (R 2 = L / C) 8 Ms/s ADC Receive filter RF 6

7 Trasmitter frot-ed: oversampled TH precoder T-spaced TX filter Iterpolator 1:2 16 Ms/s DAC Z I Trivial AFE filter f 3dB = 1 MHz C R V CT 1:1 Z R R = 1 Ω C = 3.2 pf Adaptive echo caceller Hybrid fuctio 8 Ms/s ADC Receive filter RF 7

8 Measuremets of sample 1G quad-trasformer 1:1 PHY Lie S parameters Z 1 = 1 Ω = S + S a 1 a Z 2 =1 Ω b1 b 2 b2 = S21 a1 + S22 a2 b a a S 21 : isertio loss PHY to Lie (S 12 is similar) S 22 : retur loss Lie to Lie (S 11 is similar) 8

9 Trasmit PSD: simple Perfectly implemeted AFE filter + trasformer pair 3 1 spectral uit-symbol respose: DAC curret (scaled) trasfer fuctio: DAC curret to lie voltage (scaled) -1-2 [db] -3 Trasmit PSD (scaled) sesitive to aalog filter compoets ad trasformer Frequecy [MHz] 9

10 Trasmit PSD: simple 1 Perfectly implemeted AFE filter + trasformer pair 4 spectral uit-symbol respose: DAC curret (scaled) trasfer fuctio: DAC curret to lie voltage (scaled) -1-2 [db] -3 Trasmit PSD (scaled) sesitive to aalog filter compoets ad trasformer Frequecy [MHz] 1

11 Trasmit PSD: baselie 1 Perfectly implemeted AFE filter + trasformer pair 3 spectral uit-symbol respose: DAC curret (scaled) trasfer fuctio: DAC curret to lie voltage (scaled) -1-2 [db] -3 Trasmit PSD (scaled) sesitive to aalog filter compoets ad trasformer Frequecy [MHz] 11

12 Trasmit PSD: baselie 1 Perfectly implemeted AFE filter + trasformer pair 4 spectral uit-symbol respose: DAC curret (scaled) trasfer fuctio: DAC curret to lie voltage (scaled) -1-2 [db] -3 Trasmit PSD (scaled) sesitive to aalog filter compoets ad trasformer Frequecy [MHz] 12

13 Trasmit PSD: oversampled 1 Perfectly implemeted AFE filter + trasformer pair 3 spectral uit-symbol respose: DAC curret (scaled) trasfer fuctio: DAC curret to lie voltage (scaled) -1-2 [db] -3-4 Trasmit PSD (scaled) sesitive to aalog filter compoets ad trasformer Frequecy [MHz] 13

14 Trasmit PSD: oversampled 1 Perfectly implemeted AFE filter + trasformer pair 4 spectral uit-symbol respose: DAC curret (scaled) trasfer fuctio: DAC curret to lie voltage (scaled) -1-2 [db] -3-4 Trasmit PSD (scaled) sesitive to aalog filter compoets ad trasformer Frequecy [MHz] 14

15 MDI retur loss Perfectly implemeted AFE filter + trasformer pair 3-5 Miimum MDI retur loss (Sectio ) Retur loss simple -1 [db] -15 Retur loss oversampled -2 Retur loss baselie Frequecy [MHz] Simple solutio does ot meet MDI retur loss spec: ruled out 15

16 2.5 Output voltages vs time: baselie Simulated voltages at T/4-spaced time istaces; rms ad peak values determied i 1'' T 2 Differeti al voltage at DAC output : 82mV,1969 (PAR = 7.8dB) rms mv peak 1.5 Differetial Voltage [Volt] Differeti al voltage at trasformer output : mv,1322mv (PAR = 7.43dB) 562 rms peak Perfectly implemeted AFE filter + trasformer pair 3 P = 5 dbm ito 1 Ω 562 T mv rms 16

17 Output voltages vs time: oversampled Simulated voltages at T/4-spaced time istaces; rms ad peak values determied i 1'' T Differeti al voltage at DAC output : mv, 249mV (PAR = 1.86dB) 587 rms peak Differetial Voltage [Volt] Differeti al voltage at trasformer output : mv,1891mv (PAR = 1.54dB) 562 rms peak Perfectly implemeted AFE filter + trasformer pair 3 P = 5 dbm ito 1 Ω 562 T mv rms 17

18 Voltage distributios: baselie Simulatio 1'' T Differetial voltage at DAC output 82mVrms,PAR = 7.8 db : Differetial voltage at trasformer output : 562mVrms,PAR = 7.43dB Differetial voltage [Volt] 18

19 Voltage distributios: oversampled Simulatio 1'' T Differetial voltage at DAC output 587mVrms,PAR = 1.86 db : Differetial voltage at trasformer output : 562mVrms,PAR = 1.54 db Differetial voltage [Volt] 19

20 Sigle-eded eded voltage distributios at DAC output 2.5 Simulatio 1'' T 2 Distributio of DAC output voltage "oversampled" "baselie" Required miimum Ceter - tap DAC output voltage? voltage.5 V CT 1 V + 1 CT V CT Sigle-eded voltage [Volt] 2

21 Study of trasmit-frot frot-ed solutios: summary Simple solutio ruled out: fails to meet MDI retur loss spec. Digital filters DAC AFE filter rms ad peak voltage at DAC output Excess badwidth Cotrolled spectral ulls Retur loss Trasmit PSD shape Baselie solutio oe 8 Ms/s 1-st order RLC LPF, f 3dB =3 MHz higher rms, peak similar to oversampled substatial ( samplig phase depedecy i receiver) oe OK depeds o aalog compoets Oversampled solutio (1-D 2 )/(1-.75 D 2 ) + iterpolator 16 Ms/s Trivial R//C lower rms, peak similar to baselie sharp badwidth limitatio (EMI advatage) dc ad 1/2T OK digitally defied 21

22 Study of trasmit-frot frot-ed solutios: coclusio Peak voltages at DAC output for oversampled ad baselie solutios are similar; higher PAR of oversampled is compesated for by lower rms voltage. Cost of digital filterig ad oversamplig DAC outweighs disadvatages of baselie solutio o RLC AFE filter: two coils, cocers about balace, etc. o PSD shape: substatial excess badwidth, depedecy o aalog compoets, o cotrolled spectral ulls at dc ad 1/2T o hybrid fuctio requires image impedaces matchig frequecydepedet iput impedace of AFE filter. Proposal: adopt well defied trasmit PSD shape with sharp badwidth limitatio ad spectral ulls at dc ad 1/2T, as eabled by a oversampled trasmit frot-ed solutio. 22

23 Decisio-poit SNR vs. legth of precodig respose 23

24 Optimum precodig respose ad decisio-poit SNR a(d) Precoder h -1 (D) & mod 2M x(d) TX frot ed D A C H T (f) G C (f) Chael PSD N(f) RX frot ed G R (f) A D C y(d) optimal mmse FFF g E (D) z(d) = x(d)h(d) + e(d) = a(d) + 2Mk(D) + e(d) E x = E{x k 2 } kt E PSD S = T x 2 T (f) HT(f ) H T (f)g c N(f)G (f)g R (f) R 2 (f) = H = N A A (f) kt + τ ( f ) where * ( α(f) ) = α(f + 2 j2πfτ * ( H (f)e ) /( N (f ) * E SNR T * τ worst case : SNR A ( ± 1/2T) = * x A (f) = A A ) m m/ T) Decisio SNR mmse - poit SNR for give H = T 1/2T 1/2T h(d = e A j2πft (f), N ) 2 A / (f), τ, ad precodig respose h(d) 1 L * ( SNR A (f) + 1) df, h(d) = 1+ l= h l D l * For give SNR A (f) + 1, determie (arg) max h 1,h 2, L h L SNR mmse. 24

25 Folded spectral SNR fuctio +1 Class E cable, P T = 5 dbm, l = 1m, AWGN = -14 dbm/hz PS_ANEXT ad PS_AFEXT from P T = 5 dbm, l = 1m Fixed receive filter: 3 rd -order BWF, f 3dB =3 MHz; worst-case samplig phase SNRAs1f.dat 5 TX frot-ed baselie 4 3 SNR * A (f ) + 1 [db] 2 1 TX frot-ed oversampled Frequecy [MHz] 25

26 Fiite-legth h(d) ) + ifiite FFE, MMSE optimized Class E cable, P T = 5 dbm, l = 1m, AWGN = -14 dbm/hz, PS_ANEXT ad PS_AFEXT from P T = 5 dbm, l = 1m Fixed receive filter: 3 rd -order BWF, f 3dB = 3 MHz; worst-case samplig phase TX frot-ed: baselie TX frot-ed: oversampled L SNR mmse L l= h l L SNR mmse L l= h l db db db db db 24.4 db db db db db db db db db db db db db db db db db db db

27 Fiite-legth h(d) ) + ifiite FFE, MMSE optimized Class E cable, P T = 5 dbm, l = 1m, AWGN = -14 dbm/hz, PS_ANEXT ad PS_AFEXT from P T = 5 dbm, l = 1m TX frot-ed: baselie ; fixed receive filter: 3 rd -order BWF, f 3dB = 3 MHz; worst-case samplig phase Coeffs hdl.dat h(d), L=128 : SNR mmse = db Time [T] L l= h l = Mag. spec. [db] Normalized Frequecy 27

28 Fiite-legth h(d) ) + ifiite FFE, MMSE optimized Class E cable, P T = 5 dbm, l = 1m, AWGN = -14 dbm/hz, PS_ANEXT ad PS_AFEXT from P T = 5 dbm, l = 1m TX frot-ed: oversampled ; fixed receive filter: 3 rd -order BWF, f 3dB = 3 MHz; worst-case samplig phase Coeffs hdl.dat h(d), L=128 : SNR mmse = db Time [T] L l= h l = 2.52 Mag. spec. [db] Normalized Frequecy 28

29 Fiite-legth h(d) ) + ifiite FFE, MMSE optimized Class E cable, P T = 5 dbm, l = 1m, AWGN = -14 dbm/hz, PS_ANEXT ad PS_AFEXT from P T = 5 dbm, l = 1m TX frot-ed: oversampled ; fixed receive filter: 3 rd -order BWF, f 3dB = 3 MHz; worst-case samplig phase 3 hdl.dat h(d), L=32 : SNR mmse = db Coeffs 1.5 L l= h l = Time [T] Mag. spec. [db] Normalized Frequecy 29

30 Fiite-legth h(d) ) + ifiite FFE, MMSE optimized Class E cable, P T = 5 dbm, l = 1m, AWGN = -14 dbm/hz, PS_ANEXT ad PS_AFEXT from P T = 5 dbm, l = 1m TX frot-ed: oversampled ; fixed receive filter: 3 rd -order BWF, f 3dB = 3 MHz; worst-case samplig phase 3 hdl.dat 2.5 Coeffs h(d), L=16 : SNR mmse = db L l= h l = Time [T] -6-7 Mag. spec. [db] Normalized Frequecy 3

31 DP-SNR vs legth of precodig respose: coclusio Decisio-poit SNR is isesitive to legth L of precodig respose; a programmable FIR precodig respose with L = 32 is adequate. L = 16 leads to small, but oticeable performace degradatio. L = 32 provides some headroom for dealig with o-smooth SNR(f). I additio, the results illustrate the beefits of the oversampled TX frot-ed: o Higher decisio-poit SNR due to better PSD shape o SNR performace always isesitive to samplig phase due to stricter badwidth limitatio ad spectral ull at 1/2T o Spectral ull at dc reduces costellatio expasio. Proposal: adopt programmable FIR precodig with L = 32 Same respose for all pairs, or four idividual resposes? 31

32 PMA traiig issues 32

33 Uambiguous geeratio of PMA traiig sequeces L IF L -bit IfoField IF added whe ( mod 16384) Modulatio symbol couter Geeratio of Mai PN Sequece ad Derived Sequeces Sa Sb Sc Sd + : +9 1 : -9 : +9 1 : -9 : +9 1 : -9 : +9 1 : -9 TA TB TC TD Mai PN sequece mod = : Scr [ : 32] = 33 lsbs of x15979a422 (periodic iitializa tio) mod : Derived sequeces Scr [1 : 33] = Scr Scr Scr[] = Scr Sa Sb Sc Sd Scr [] 1 if mod 256 = = Scr [] otherwise = Scr [3] Scr [8] = Scr [6] Scr [16] = Scr [9] Scr [14] Scr [19] Scr [24] 1 [2] Scr [13] Scr [ : 32] [33] [33] if PMA_CONFIG if PMA_CONFIG = MASTER = SLAVE 33