kdpof demo v1

Demonstration of 1Gbps over 50m of low cost SI-POF with KDPOF technology

June 2010

AbstractThis paper presents the implementation of a prototype based on KDPOF physical layer technology that provides more than 1 Gbps error-free communication over more than 50 m of low cost standard 1 mm Step Index Plastic Optical Fiber (SI-POF). The prototype is based on generic Arbitrary Wave Generator and Digitizer boards as well as a high performance computer that carries out all the digital signal processing and communication tasks implemented in software. A general description of the technology as well as detailed studies and results are presented in the paper.

IntroductionCurrent POF solutions use simple Non Return Zero (NRZ) modulation to communicate. This is the modulation tech-nique implemented either in home network, industrial or car infotainment applications. NRZ modulation has been used for years in the Glass Optical Fiber (GOF) world with big success.

GOF can be considered as an almost perfect transmission media due to its low distortion, high bandwidth and very low attenuation. However, POF has a frequency and time response much different than glass fiber, besides a high attenuation. Current glass fiber modulations use NRZ 8b/10b or NRZI 4b/5b line coding, requiring a baud rate of 1.25 GHz and 125 MHz, for 1Gbps and 100 Mbps solu-tion, respectively. Looking to the POF frequency response we realize that even 100 or 150 Mbps are possible with good link budget indeed with direct detection (by limiting amplifier), the 1 Gbps solution is not possible without a more advanced modulation system.

Following figure shows the variation of POF bandwidth (black) in function of the fiber length, as well as the bandwidth-length product (red). As can be seen a good flat response for the required 1.25 GHz baud rate in only possible in the very first meters.

February 2010 STRICTLY CONFIDENTIAL

!"#

$%&'()*+,&-.&'-&/&,*01&)$'

Optics simulation (samples January 2007) (part 3 of 3)

25

Frequency domain response: magnitude vs. frequency 3dB Bandwidth vs. length

0 500 1000 1500 2000 250090

80

70

60

50

40

30

20

10

0Frequency response of SI POF

Frequency (MHz)

Nor

mal

ized

pow

er

0 10 20 30 40 50 60 70 80 90 1000

500

1000

1500Bandwidth ( 3dB) of SI POF

Length (m)

MH

z

0 10 20 30 40 50 60 70 80 90 10020

25

30

35

40

Length (m)

MH

z * 1

00m

293 MHz

40 MHz68 MHz

RCLED @ T = 25 ºC

KDPOF technologyKDPOF proposes the use of advanced telecommunication techniques in the POF media. These techniques are widely used in cable and wireless communication systems. They are proven due to the fact they are used in standards like 10G base-T or WiFi.

KDPOF takes a look to the POF media no as a perfect glass fiber communication system where simple NRZ gives a good tradeoff between performance and cost, but as a system closer to cable or air, where the communication channels are much more complex.

KDPOF technology defines four parts of the communica-tion system physical layer:

• Modulation: KDPOF proposes to use multiple level Pulse Amplitude Modulation (PAM) technique. The number of levels is defined by the bandwidth, the required bit rate and the coding. Deep studies carried out on link budget maximization and presented later, show that 16-PAM and bandwidth of 156.5 MHz are good choices to provide 1 Gbps. This exact frequency is considered because it is easily obtained from the GMII interface.

• Equalization: KDPOF technology uses an advanced channel and equalization system allowing the accurate channel and noise estimation and compensation. This technique optimally cancels the distortion created by the fiber and the optoelectronic devices like the LED, photodiode and trans-impedance amplifier. Linear and non-linear components are compensated.

• Channel coding: this is a challenging part of the communication system. Current state of the art coding as LDPC and Turbo-Codes are very expensive in terms of silicon area and power consumption when managing a bit rate of 1 Gbps. KDPOF technology implements a low power multi-level coset coding which minimizes the silicon area and power while keeping very good error correction performance.

• Frame building: the encapsulation of the higher layers information in the physical layer requires a flexible and efficient frame building, with preambles and correct size

Knowledge Development for POF

www.kdpof.com © KDPOF 2010www.kdpof.com © KDPOF 2010

http://www.kdpof.com


for synchronization and equalization, headers for announcing the basic physical layer parameters, and to encapsulate higher layers data, as well as a required CRC to guarantee packet integrity.

Chasing the Shannon limit of POFKDPOF innovates in the methodology for POF communi-cation systems design, applying the most powerful crite-rion: to approach the Shannon capacity limit as much as possible through the use of tools belonging to the Informa-tion Theory.

Information Theory gives answers to several important questions like, what elements (i.e. optoelectronics, fiber, amplifiers, etc.) compose the POF communication chan-nel?, where is the capacity (Shannon’s limit) for this com-munication channel? how can we calculate the capacity? or, what are the most suitable telecom techniques to mini-mize the gap to capacity?

Usually the transceivers for optical fibers are designed ac-cording to the optimization of the eye-diagram, searching for a tradeoff between bandwidth and Signal-to-Noise Ra-tio (SNR). In order to achieve higher data-rates, current developments of POF communication systems consider the optimization of digital modulation, light emitters, pho-todiodes and analogue circuits as isolated problems. This produces as result over-engineered solutions for each problem with excessive costs and without achieving the main goal: to approach capacity. Or, in other words, to provide 1 Gbps with the highest link budget in order to maximize the market success.

On the other hand, KDPOF introduces a novel criterion in POF systems design, optimizing in a coupled manner the digital circuit and the analogue transceiver. This is carried out by accurate models, which are verified in laboratory, and advanced numerical algorithms that maximize the capacity (i.e. entropy) from the point of view of the Informa-tion Theory.

Link budget maximizationDeep studies has beed carried out in order to find the communication techniques that lead to the link budget maximization. State of art of optoelectronic devices cur-rently available in the market has been considered, since the KDPOF objective is to make possible 1 Gbps over 50 m of low cost SI-POF without requiring the development of new photonics.

Several modulation, coding and equalization techniques has been evaluated together with the KDPOF proposal for comparison, searching in all of them the optimum baud-rate and level to optimize the link margin. On the one hand, binary coded M-PAM is considered using Linear Equalizer (LE) or Decision Feedback Equalizer (DFE) in the receiver side. Algebraic coding as Reed-Solomon (RS) with high code-rate is considered by its low implementation cost. On the other hand, complete KDPOF solution based on multi-level coset coded M-PAM and non-linear Tomlinson-

Harashima Precoding (THP). DMT and OFDM techniques were studied and discarded in the past so they are not considered. The main argument was that POF is a peak power limited non-linear communication channel and this kind of modulations have a huge cress factor, reducing a lot the total amount of power injected in the fiber. Next figure shows the link margin results (dBo) for 1 Gbps over 50 m of fiber in function of the baud rate (MHz).

200 300 400 500 600 700 800 900 1000 1100 12000

2

4

6

8

10

12

14

16

18

20

Link Budget for L=50.00m, AFE A3PICS dcn(1), CH pof optic ESKA GH4001 LED RED 030 PD SI,

realimpl 1, realgap 0, uncoded 0

Fs (MHz)

dBo

Ref. 0 dBoLE RS: LED 50.00 MHzLE RS: LED 100.00 MHzLE RS: LED 350.00 MHzDFE RS: LED 50.00 MHzDFE RS: LED 100.00 MHzDFE RS: LED 350.00 MHzKDPOF: LED 50.00 MHzKDPOF: LED 100.00 MHzKDPOF: LED 350.00 MHz

As can be seen, the curves are given by discrete points, corresponding each one to a realistic working point for each proposal. Realistic performance is obtained by sto-chastic simulations for each technique that is shown; for example, performance losses in DFE by error burst propa-gations and precoding loss in THP are taken in considera-tion. Three different bandwidths for the LED are evaluated: 50 (blue), 100 (green) and 350 MHz (red), so taking into account from the lowest to the highest cost current drivers and LED in the market. The average power injected in POF is -3 dBm with Optical Modulation Amplitude (OMA) of 0 dBm. Link budget, given in optical dB, includes the at-tenuation of the fiber.

RS-PAM with linear equalization results are depicted with crosses, increasing from right to left side of the plot the PAM level in 1 bit basis. So, the most right point corre-sponds to 2-PAM, which is similar to NRZ, but with binary coding and equalization. As can be seen, the optimum baud-rate for this proposal depends on the LED band-width. For 50 and 100 MHz the optimum is near to 300 MHz with 16-PAM, and for 350 MHz is near to 400 MHz with 8-PAM.

In the same way RS-PAM with DFE is depicted with trian-gles. The optimum is from 16-PAM, for 50 MHz LED, to 8-PAM for more than 100 MHz bandwidth. It is important to realize the DFE performance loss respect to LE when PAM level is increased due to the burst errors propagation.

Finally, KDPOF performance is analyzed. As can be seen, the number of discrete points is twice of the previous pro-posals, because PAM level is increased from 1 bit in 0.5 bits basis. This is done based on the multi-level coset cod-


www.kdpof.com © KDPOF 2010




ing structure that works with two dimensional lattices and enables for a more accurate bit rate selection. KDPOF technology provides a 4 dBo (8 electrical dB) advantage, keeping a lower cost and power consumption thanks to its smart coding. It is common knowledge that THP equaliza-tion approaches the ideal (no error propagation) DFE per-formance for high spectral efficiencies, due to the facts the precoding loss rapidly decreases to zero and the crest factor loss approaches that of simple PAM without precod-ing. THP and coding are the main keys for the advantage given by KDPOF.

All this analysis brings the conclusion that a baud-rate around 300 MHz (150 MHz bandwidth) with 16-PAM are the optimum modulation and coding parameters for link margin maximization. Therefore, selection of 312.5 MHz, easily obtained from GMII, is justified.

Prototype setupThe prototype consists of one high performance personal computer, generic high speed AWG and Digitizer, Analog Front Ends (AFE) for the transmitter and the receiver and 50 m of standard SI-POF. Next figure shows an scheme for the KDPOF prototype.

May 2010

!"#

$%&'()*+,&-.&'-&/&,*01&)$'

CONFIDENTIAL

Demonstrator - 1st version

3

FEATURESCG11G: 4 MegaSamples of on-board memory

CG11G2: 16 MegaSamples of on-board memory

12 bit resolution

1 billion conversions per second

Free software for importing and generating waveforms

Software Development Kits for C/C++, LabVIEW and MATLAB

•

•

•

•

•

•

Arbitrary waveform generator and function generator card capable of providing aperiodic analog waveforms.

APPLICATIONS Stimulus-response test systems

RF signal generation

Wireless communications

Manufacturing test

Radar signal simulation

Optical and magnetic storage media testing

Advanced ultrasonic signal generation

Video signal generation

Network analysis

www.gage-applied.com

CompuGen 11G/11G2High-speed arbitrary waveform generator card

We offer the widest range

of high-speed digitizers and

instrumentation cards available on

the market today. Our powerful

PC-based instrumentation products

allow you to create reliable,

solutions quickly and easily.

Reduce development time

applications such as radar, wireless

communications, spectroscopy,

etc. by using our Windows-based

application software or our SDKs.

AWG:•1 GS/s•12 bits•16MS

ADC:•1.1 GS/s•12 bits•170MS

USB 2.0

PCI

Personal Computer

AFE RX AFE TX

Simplexstd SI-POF

LEDPhotodiode

The PC is in charged to implement the TX and RX tasks as modulation, coding, DSP, equalization, timing-recovery and bit-rate adapt. The system is not real-time, because the PC and hardware are not able to manage the required computational charge. However, the system manage sepa-rated frames that are real-time transmitted to the channel providing the target bit-rate.

The high speed AWG is connected through PCI bus to the PC, working at a fixed sampling rate of 1 GHz and vertical resolution of 12 bits. It carries out the tasks of the Digital-Analog Converter (DAC) and Zero Order Hold (ZOH). Be-cause sampling rate cannot be configured, the PC per-forms simple ZOH sampling interpolation to adequate the system to the required baud-rate. The system works at 333.33 Mbaud/s repeating 3 times each TX signal sample, preserving the crest factor on the LED. This sampling rate is very close to the proposed frequency of 312.5 Mbaud/s.

On the other hand, the digitizer consists of a 12 bits ADC and works at 1.1 GHz, so additional multi-rate filtering as

well as decimal decimation is implemented by PC in order to accommodate the signal to the baud rate and perform the timing-recovery (clock phase and frequency compen-sation). In final ASIC this blocks will not be needed be-cause DAC and ADC will work at baud rate.

The AFE for TX and RX is schematized in next figure.

May 2010

!"#

$%&'()*+,&-.&'-&/&,*01&)$'

CONFIDENTIAL

AFE RX

AFE - DFE blocks

5

AWG Driver LEDSimplex

std SI-POF

ADC

FDAC = 1 GHzDAC 12 bits

FADC = 1.1 GHz12 bits

TXL1

DAC12 bits

ZOH

FS->FDAC

AFETX

Simplexstd SI-POF

AFERX

AFE TX

Anti-aliasDFLT

DAGC SAH RXL1

FC = FS/2FADC->FS

Timing Recv

FADC FADC FADC FS

FS FS FDAC

PD TIA VGAAnti-aliasFLT

As can be seen, the transmission part is composed by a current driver, based on Operational Amplifier, and an ana-log Firecomms RC-LED. The reception part consists of a Hamamatsu S5971 photodiode, a discrete trans-impedance amplifier based on OPA LMH6624, a TI VCA824 based Variable Gain Amplifier (VGA) and a passive 5th order Butterworth anti-alias filter. The AFE RX also in-cludes a DC-offset active compensation circuit in order to maximize the dynamic range of ADC.

The system performs the startup in three stages:

• Synchronization: in this step the receiver estimates and compensates the sampling frequency deviation with the transmitter, as well as determines the best sampling phase regarding to the channel delay group to maximize the SNR. The equalization is entirely performed by the receiver using a non-linear DFE structure and minimum bit-rate and guaranteeing the communication capacity for the physical layer headers.

• THP enable: once synchronization is achieved, the system automatically changes to work with THP equalization and performs the SNR measurements required to determine the final system bit-rate.

• Link: in function of SNR and channel estimations the transmitter selects the coding and PAM level that provide the maximum possible bit-rate guaranteeing a bit error ratio (BER) less than 10-12. At this point the full l ink is stablished and continuous tracking of synchronization, channel and noise is done.

The system achieves the complete link state in less than 10 frames, which represents less than 7 ms. It is important to note that KDPOF technology is bit-rate adaptive. For a final 1 Gbps system the maximum spectral efficiency could be limited to that needed for this bit-rate; in case of poor installations, overpassed link margins due to ambient con-ditions or extra long fiber, the system is able to provide link, reducing the target bit rate, e.g. 800 Mbps at 80 m of POF.

ResultsThe prototype has been evaluated with the previously de-scribed setup obtaining error free 1 Gbps over 50 m of SI-POF. However, this result represents an implementation loss for more than 9 dB regarding to the predicted results by simulation models. The causes has been identified, being the main ones the excess noise in TIA and non-





linearities in current driver feeding the LED. They will be improved in future implementations enabling the evaluation of KDPOF prototype with state of art of optoelectronics in market.

Regarding to the non-linear response of the electrical-optical conversion in LED, KDPOF is performing detailed studies and models based on Hammerstein, Wiener and complete Volterra systems that will bring good knowledge about this subject. The presented prototype includes non-

linear THP equalization based on automatic Hammerstein model estimation of the POF communication channel, that allows us to compensate the main non-linearity source of the LED device.

The following screen dumps show the graphical user inter-face of the system for the three stages of the startup. The first one shows a received frame where the fine clock fre-quency and phase recoveries have not finished in the syn-chronization stage.

The five plots in the top show, from left to right, the chan-nel impulse response sampled at baud rate, the two filters (feedforward and feedback) that compose the DFE and future THP equalizer, the frequency domain magnitude response of channel and equalizer and the two polynomi-als that compose the Hammerstein system. P(x) is the non-linearity of the transmitter and Q(x) the compensation polynomial involved together with P(x) in several parts of the equalizer. In the bottom, the equalized symbols as well as the symbols belonging to each level of the multilevel structure are plotted. As can be seen, due to the fact the clock has not been recovered yet, the system performs with a high BER, very low SNR in detector and symbols cannot be distinguished. As was explained, the symbols are represented in two dimensions, since they belong to

two dimensional lattices. The right side shows several values related with the current state of the system: BER, SNR in channel, SNR in detector, estimated deviation in ppm, etc. It is important to observe that, indeed in this conditions the physical layer header is received valid thanks to a robust design.

The second screenshot shows the synchronization stage once the timing recovery has finished. As can be seen the symbols are well distinguished and there is not BER. Dur-ing this stage the system is configured with the minimum spectra efficiency only using the first level of the multilevel coding. It is also observed that the noise variance is very low regarding to the constellation energy; this indicates






that much more information can be transmitted over the channel.

The third one shows the system in THP enabled stage. As can be seen, the equalized symbols experiment lattice multiplication due to the involved modulo operation and SNR in detector has decreased ~1.25 dB. This number well corresponds with theoretically predicted precoding loss for 2-PAM (QPSK) constellation. For larger constella-tions it quickly decreases as can be seen in the link stage.

Finally the fourth screenshot shows the system after link achievement. The SNR in detector has been recovered with almost zero precoding loss and complete multilevel coding structure is operating. The first level is in charged to split the complete constellation (16-PAM, rotated 128-QAM in 2D) in the 4 cosets with minimum constellation distance. The noise variance regarding to constellation volume is maximum -minimum Volume-to-Noise Ratio (VNR)- in this first level, so the correction capacity of the corresponding component code is the highest. Based on the first level decoding, the system splits the left constella-tion in other 4 cosets. The VNR has been increased 6 dB so that the component code can be much lighter to

achieve the same BER. Finally the third level is hard de-coded with a slicer working with 12 dB more VNR respect to the first level. The physical layer bit rate is 1096 Mbps.

ConclusionsA general description of KDPOF physical layer technology has been provided, including detailed studies about link margin maximization and modulation parameters selection. Experimental results are provided based on the fist version of KDPOF hardware/software based prototype, which achieves 1 Gbps over 50 m standard SI-POF.

Carlos Pardo: Co-Founder and CEO of KDPOF. He has been working on the past developing silicon for DVB broadband communications as well as for the power line communications. He has more than 15 years experience on developing and commercializing advanced communications silicon.

Rubén Pérez de Aranda: Co-Founder and CTO of KDPOF. He has been working in the past developing high end DSL and power line communica-tions. He has more than 8 years experience in the management of research and development of digital signal processing and communication systems.

KDPOF: Knowledge Development for POF S.A. is a Fabless Semiconductor Company located in Madrid-Spain. The main objective of the company is the development of silicon for the POF market. KDPOF has Ethernet POF silicon for the industrial, professional and home networking market.





kdpof demo v1

Documents