Optical transmitter characteristics for GEPOF technical feasibility

Transcription

1 Optical transmitter characteristics for GE technical feasibility Rubén Pérez-Aranda IEEE 82.3 GE Study Group - May 214 Interim

2 Supporters Frank Aldinger (Mitsubishi International) Yutaka Tanida (Mitsubishi Corporation) Y.Tsukamoto (Mitsubishi Rayon) Eric Chan (Boeing) Philippe Bolle (Skylaneoptics) / Mike Cao (Dongguan ipt Industrial Co,.LTD.) John Lambkin (Firecomms) Hugh Hennessy (Firecomms) Josef Faller (Homefibre) Manabu Kagami (Toyota R&D Labs) Bas Huiszoon (Genexis) Oscar Rechou (Casacom) Naoshi Serizawa (Yazaki) Thomas Lichtenegger (Avago Tech) IEEE 82.3 GE Study Group - May 214 Interim 2

3 Agenda Objectives The optical transmitter main characteristics LED non-linear response and capacity penalties Conclusions IEEE 82.3 GE Study Group - May 214 Interim 3

4 Disclaimer Technical characteristics provided in this presentation are limited to those directly affecting the optical link budget and, therefore, the Shannon s capacity analysis. Other characteristics, like the ones related to the physical semiconductor parameters, integration, manufacturing process, etc. are intentionally left outside of the sope of this presentation IEEE 82.3 GE Study Group - May 214 Interim 4

5 Objectives This presentation provides technical characteristics of the optical transmitter used today for automotive applications as well as for consumer applications This optical transmitter is a red LED, and it is the light emitter most widely used by the industry for communications The red LED has been qualified for automotive applications, being demonstrated its reliability during the last +1 years The main objective of this presentation is to analyze the red LED from the perspective of the aspects that directly relates to the Shannon s capacity based technical feasibility assessment The results presented here will be used for Shannon s capacity analysis in [perezaranda_1_514_shannoncap] IEEE 82.3 GE Study Group - May 214 Interim 5

6 The optical transmitter main characteristics IEEE 82.3 GE Study Group - May 214 Interim

7 The optical transmitter - architecture The optical transmitter is composed by the current driver IC and the LED IC The red LED converts the electrical current into optical power In general, the I-P characteristic of LED is not linear; this topic is covered later on Electrical-to-electrical response is well approximated by a 1 st order low pass system Achievable -3dB bandwidth of LED itself is between 75 and 95 MHz, depending on the internal structure of LED Wavelength center ~65 nm; wavelength width ~3 nm Typically, the driver is a trans-conductance amplifier in charge to convert the voltage communication signal from the PHY into the adequate current to drive the LED, providing: Bias current control to ensure reliability of the LED Extinction Ratio (ER) control, to avoid switching off the LED (optical power clipping) and ensure the quantum noise from PD is low Typical target ER = 1 dbo Typical process and temperature variation of ER < ±2 dbo Frequency pre-emphasis, to enhance the bandwidth of the LED Frequency pre-emphasis gain is limited based on reliability criteria max peak current IEEE 82.3 GE Study Group - May 214 Interim 7

8 The optical transmitter - architecture PHY ER DRIVER LED + + Extinction Ratio Control Current Biasing Bias Current VDD DAC + - Preemphasis + - TCAmp + + Zin Pre-emphasis Peaking zero cut-off freq Peaking gain IEEE 82.3 GE Study Group - May 214 Interim 8

9 The optical transmitter - pre-emphasis.4 No pre-emphasis, MOST line-coding LED response for ILEDavg: 2 ma; ER: 1. db; LED Fc 3dB: 1 MHz; Preemphasis: Fz is 6 MHz, G HF is db Saturated samples ratio:.e+ LED current Rx Vout.35 ILED max pk+ LED current (A) and voltage out (arbitrary units) ILED avg.5 ILED min pk Arbitrary time unit IEEE 82.3 GE Study Group - May 214 Interim 9

10 The optical transmitter - pre-emphasis.7.6 Pre-emphasis, MOST line-coding LED response for ILEDavg: 2 ma; ER: 1. db; LED Fc 3dB: 1 MHz; Preemphasis: Fz is 6 MHz, G HF is 6 db Saturated samples ratio:.e+ LED current Rx Vout ILED pk+.5 LED current (A) and voltage out (arbitrary units) ILED max ILED avg ILED min.1.2 ILED pk Arbitrary time unit IEEE 82.3 GE Study Group - May 214 Interim 1

11 The optical transmitter - pre-emphasis.4 No pre-emphasis, high M PAM LED response for ILEDavg: 2 ma; ER: 1. db; LED Fc 3dB: 1 MHz; Preemphasis: Fz is 6 MHz, G HF is db Saturated samples ratio:.e+ LED current Rx Vout.35 ILED max pk+ LED current (A) and voltage out (arbitrary units) ILED avg.5 ILED min pk Arbitrary time unit IEEE 82.3 GE Study Group - May 214 Interim 11

12 The optical transmitter - pre-emphasis.7.6 Pre-emphasis, high M PAM LED response for ILEDavg: 2 ma; ER: 1. db; LED Fc 3dB: 1 MHz; Preemphasis: Fz is 6 MHz, G HF is 6 db Saturated samples ratio:.e+ LED current Rx Vout ILED pk+.5 LED current (A) and voltage out (arbitrary units) ILED max ILED avg ILED min.1.2 ILED pk Arbitrary time unit IEEE 82.3 GE Study Group - May 214 Interim 12

13 The optical transmitter - response 5 Driver + MOST red LED E to E response Electrical to electrical magnitude response (db) Lab measurement of real product qualified for automotive X: 15.6 Y: Frequency (MHz) IEEE 82.3 GE Study Group - May 214 Interim 13

14 Performance with temperature 1 2 AOP coupled into (lab measurements) ATX16125: real AOP vs. temperature #15 #41 #27 Avg 3 AOP (dbm) Temp ( o C) IEEE 82.3 GE Study Group - May 214 Interim 14

15 Performance with temperature 1 OMA coupled into (lab measurements) ATX16125: real OMA vs. temperature #15 #41 #27 Avg 2 OMA (dbm) Temp ( o C) IEEE 82.3 GE Study Group - May 214 Interim 15

16 Non-linear response and capacity penalties IEEE 82.3 GE Study Group - May 214 Interim

17 Non-linear distortion (-4 ºC, 15.6 MHz) fs MHz, fc MHz HD(1) MHz dbc 2 4 HD(2) MHz dbc HD(3) MHz dbc Amplitude (dbfs) 6 8 MHz dbfs MHz dbfs Frequency (Hz) RBW Hz, equiv NFFT x 1 7 Lab. measurements of a real product IEEE 82.3 GE Study Group - May 214 Interim 17

18 Non-linear distortion (+25 ºC, 15.6 MHz) fs MHz, fc MHz HD(1) MHz dbc 2 4 HD(2) MHz dbc HD(3) MHz dbc Amplitude (dbfs) MHz dbfs MHz dbfs Frequency (Hz) RBW Hz, equiv NFFT x 1 7 Lab. measurements of a real product IEEE 82.3 GE Study Group - May 214 Interim 18

19 Non-linear distortion (+15 ºC, 15.6 MHz) fs MHz, fc MHz HD(1) MHz dbc 2 4 HD(2) MHz dbc HD(3) MHz dbc Amplitude (dbfs) 6 8 MHz dbfs MHz dbfs Frequency (Hz) RBW Hz, equiv NFFT x 1 7 Lab. measurements of a real product IEEE 82.3 GE Study Group - May 214 Interim 19

20 Non-linear distortion (-4 ºC, 44.6 MHz) fs MHz, fc MHz HD(1) MHz dbc 2 HD(2) MHz dbc 4 HD(3) MHz dbc Amplitude (dbfs) MHz MHz MHz dbfs dbfs dbfs MHz dbfs Frequency (Hz) RBW Hz, equiv NFFT x 1 7 Lab. measurements of a real product IEEE 82.3 GE Study Group - May 214 Interim 2

21 Non-linear distortion (+25 ºC, 44.6 MHz) fs MHz, fc MHz HD(1) MHz dbc 2 HD(2) MHz dbc 4 HD(3) MHz dbc Amplitude (dbfs) MHz dbfs Frequency (Hz) RBW Hz, equiv NFFT x 1 7 Lab. measurements of a real product IEEE 82.3 GE Study Group - May 214 Interim 21

22 Non-linear distortion (+15 ºC, 44.6 MHz) fs MHz, fc MHz HD(1) MHz dbc 2 4 HD(2) MHz dbc HD(3) MHz dbc Amplitude (dbfs) MHz dbfs Frequency (Hz) RBW Hz, equiv NFFT x 1 7 Lab. measurements of a real product IEEE 82.3 GE Study Group - May 214 Interim 22

23 Non-linear distortion - preliminary conclusions Based on previous measurements we can do some conclusions: The non-linear response of the LED depends on the temperature The harmonic distortion measurement with input single tone depends on the frequency of the tone Based on this very basic measurements we could conclude that only low spectral efficiency modulation schemes would be feasible with the LED However, we are going to demonstrate that this conclusion is false, by analyzing the non-linear response in deeper detail The idea behind the following analysis is that the non-linear response of the LED can be adaptively compensated by the PHY in the same way the ISI is equalized in modern Ethernet PHYs to approach the channel capacity IEEE 82.3 GE Study Group - May 214 Interim 23

24 Non-linear response - the Volterra model In order to analyze the effect of LED HD in the communication system we need to develop a correct model for the non-linear response Truncated Volterra series expansion is selected to model the optical TX non-linear response Volterra series expansion is a well known technique and it have been used by the industry in a wide range of engineering fields to model non-linear systems It is attractive from the mathematical point of view linear combination of non-linear functions of the input signal It fits a large class of non-linear systems Well known adaptive filtering algorithms are suitable for Volterra series estimation L l 1 = y(k) = w o + w o1 (l 1 )x(k l 1 ) + L L l 1 = l 2 = + w o2 (l 1,l 2 )x(k l 1 )x(k l 2 ) + L L l 1 = l 2 = L l p = + w op (l 1,l 2, l P )x(k l 1 )x(k l 2 ) x(k l P ) DC offset + linear filter 2 nd order convolution Higher-order convolutions IEEE 82.3 GE Study Group - May 214 Interim 24

25 Non-linear response - the Volterra model 1 z -Δ ωoo(k) DC offset... x(k) z -Δ1 z -1 z -1 z -1 z -Δ2 ω(k) ω1(k) ω2(k) ω11(k)... 1 st order response gap... z -1 z -1 z -1 ω,(k) ω1,1(k) ω2,2(k) ω8,8(k)... y(k) gap 1 z z -1 z -1 z -1 ω,1(k) ω1,2(k) ω2,3(k) ω7,8(k) 2 nd order response... gap 2 gap n z-1 z-2... z -1 z -1 ω,2(k) ω1,3(k) ω2,4(k) ω6,8(k)... IEEE 82.3 GE Study Group - May 214 Interim 25

26 Non-linear response - the Volterra model z -Δ3 gap... z -1 z -1 z -1 ω,,(k) ω1,1,1(k) ω2,2,2(k) ω6,6,6 (k)... y(k) gap 1 z -1 ω,.1(k) ω1,1,2(k)... z -1 z -1 z -1 ω2,2,3(k)... ω5,5,6(k) 3 rd order response gap 1... z -1 z-1 z -1 z -1 ω,1,1(k) ω1,2,2(k) ω2,3,3(k) ω5,6,6(k)... The optical transmitter is well modeled by a 3 rd order Volterra system. Higher order kernels are negligible IEEE 82.3 GE Study Group - May 214 Interim 26

27 Non-linear response: Volterra DC and 1 st order -4 ºC +15 ºC.8.7 th and 1st order Volterra kernels Vol Vol1.9.8 th and 1st order Volterra kernels Vol Vol FS = MHz IEEE 82.3 GE Study Group - May 214 Interim 27

28 Non-linear response: Volterra 2 nd order -4 ºC nd order Volterra per gap Vol2 gap Vol2 gap 1 Vol2 gap 2 Vol2 gap Vol1 Vol2 gap Vol2 gap 1 Vol2 gap 2 Vol2 gap 3 Magnitude Response (db).1.2 Magnitude (db) Normalized Frequency ( π rad/sample) FS = MHz IEEE 82.3 GE Study Group - May 214 Interim 28

29 Non-linear response: Volterra 2 nd order 15 ºC.4.2 2nd order Volterra per gap Vol2 gap Vol2 gap 1 Vol2 gap 2 Vol2 gap Magnitude Response (db) Vol1 Vol2 gap Vol2 gap 1 Vol2 gap 2 Vol2 gap Magnitude (db) Normalized Frequency ( π rad/sample) FS = MHz IEEE 82.3 GE Study Group - May 214 Interim 29

30 Non-linear response: Volterra 3 rd order -4 ºC rd order Volterra per gap Vol3 gap Vol3 gap 1 Vol3 gap 2 Vol3 gap 3 Vol3 gap 1 Vol3 gap 1 1 Vol3 gap 1 2 Vol3 gap 2 Vol3 gap 2 1 Vol3 gap Magnitude (db) FS = MHz Magnitude Response (db) Vol1 Vol3 gap Vol3 gap 1 Vol3 gap 2 Vol3 gap 3 Vol3 gap 1 Vol3 gap 1 1 Vol3 gap 1 2 Vol3 gap 2 Vol3 gap 2 1 Vol3 gap Normalized Frequency ( π rad/sample) IEEE 82.3 GE Study Group - May 214 Interim 3

31 Non-linear response: Volterra 3 rd order 15 ºC rd order Volterra per gap Vol3 gap Vol3 gap 1 Vol3 gap 2 Vol3 gap 3 Vol3 gap 1 Vol3 gap 1 1 Vol3 gap 1 2 Vol3 gap 2 Vol3 gap 2 1 Vol3 gap Magnitude Response (db) Magnitude (db) FS = MHz Vol1 Vol3 gap Vol3 gap 1 Vol3 gap 2 Vol3 gap 3 Vol3 gap 1 Vol3 gap 1 1 Vol3 gap 1 2 Vol3 gap 2 Vol3 gap 2 1 Vol3 gap Normalized Frequency ( π rad/sample) IEEE 82.3 GE Study Group - May 214 Interim 31

32 Non-linear response: Volterra analysis Bandwidth of the optical TX increases with temperature, although impulse response could be considered approximately constant The magnitude of the 2nd and 3rd order Volterra kernels increases with temperature and frequency it confirms the basic single tone HD measurements It is important to note that most part of energy of 2nd and 3rd order responses is delayed respect to 1st order We can conclude that optical TX cannot be modeled as a Wiener or a Hammerstein nonlinear system The morphology of Volterra (2nd and 3rd) kernels basically does not change with temperature good from the implementation point of view IEEE 82.3 GE Study Group - May 214 Interim 32

33 Capacity penalties - channel linearization Light Source (Driver + LED) Photodiode Trans- Impedance Amplifier (TIA) Antialias Filter non-linear channel Wo Do Wo1 n(k) Do1 n (k) x(k) Wo2 y(k) z(k) Do2 v(k) x(k) Wo1(k) v(k) WoP DoP Linear Channel non-linear channel Linearizer IEEE 82.3 GE Study Group - May 214 Interim 33

34 PSD (dbm/hz) Hanning Welch, Z = 1. ohms, Avg 1 Capacity penalties - Linearizer is not implemented PHY input SNRe = 39.8 db RX signal Noise Noise + NL Ideal MMSE DFE PSD (dbm/hz) Hanning Welch, Z = 1. ohms, Avg DFE output DFE: Detector signal DFE: Detector Noise DFE: Noise bound 25.4 db 39.8 db 14.4 db Frequency (Hz) MBW Hz x 1 7 Magnitude Response (db) Frequency (Hz) MBW Hz x Magnitude (db) DFE: FFE+FBF DFE: FFE DFE: FBF Normalized Frequency ( π rad/sample) IEEE 82.3 GE Study Group - May 214 Interim 34

35 Capacity penalties - Linearizer is implemented PSD (dbm/hz) Hanning Welch, Z = 1. ohms, Avg PHY input SNRe = 39.8 db RX signal Noise Noise + NL Linearizer + Ideal MMSE DFE PSD (dbm/hz) Hanning Welch, Z = 1. ohms, Avg DFE output 36.7 db 39.8 db Linearizer + DFE: Detector signal Linearizer + DFE: Detector Noise Linearizer + DFE: Noise Bound 3.1 db Frequency (Hz) MBW Hz x 1 7 Magnitude Response (db) Frequency (Hz) MBW Hz x Magnitude (db) 4 2 DFE after linearizer: FFE+FBF DFE after linearizer: FFE DFE after linearizer: FBF Normalized Frequency ( π rad/sample) IEEE 82.3 GE Study Group - May 214 Interim 35

36 Capacity penalties Linearizer + DFE DFE Capacity penalty caused by the LED non linear response Capacity penalty in detector (db) Channel SNR e (db) Capacity loss < 1dB for SNRe < 3 db High spectral efficiency schemes are feasible IEEE 82.3 GE Study Group - May 214 Interim 36

37 Conclusions Technical characteristics of the optical transmitter used today for automotive applications as well as for consumer applications have been presented The non-linear response of I-P characteristic of LED has been analyzed in detail, concluding that high spectral efficiency modulation schemes are also feasible with low capacity penalties, opening the use of LED beyond OOK schemes The results presented here will be used for Shannon s capacity analysis in [perezaranda_1_514_shannoncap] IEEE 82.3 GE Study Group - May 214 Interim 37

38 Questions? IEEE 82.3 GE Study Group - May 214 Interim