Transcript Slide 1
Burst-mode receivers for GPON and LRPON applications J.J. Lepley and S.D. Walker University of Essex Overview Burst-mode receivers are a key component in passive optical networks Solutions now off-the-shelf for Ethernet based PONs (such as EPON and GEPON), but GPON and the future LRPON standards are proving difficult This paper discusses some of the issues involved and presents a possible solution based on edge-triggered receivers NOC conference, Berlin June 2006 Overview Burst Mode Receiver for LR-PON AC coupling DC coupling Edge detection NOC conference, Berlin June 2006 AC coupling – what are the requirements? AC coupled front end C R Settle in less time than preamble with acceptable BER Hold signal for longer than CID with acceptable BER At 2.5 Gbps • Guard time (64 bits) = 25.6 ns • Preamble (108 bits) = 43.2 ns • CID (72 bits) = 28.8 ns NOC conference, Berlin June 2006 AC coupling Data lost Data received within BER limits AC threshold set at the midpoint assuming even mark-space ratio Large change in burst amplitude requires finite settling time during which data will not be received NOC conference, Berlin June 2006 AC coupling – upper/lower limits Maximum time constant determined by settling time between loudest and softest bursts Guard + Preamble (68.8 ns at 2.5Gbps) Settle to within the upper threshold level Loudest Softest Minimum time constant determined by maximum CID period CID (ones) (72 bits or 28.8ns at 2.5Gbps) NOC conference, Berlin June 2006 Remain within the upper threshold level AC coupling calculations Assumptions: LSR = 20 dB ER = 10 dB R=50 ohm Maximum TC Minimum TC* change from loud to soft within guard + preamble (<0.01458 V after 68.8 ns) requirement to remain within limits during a max CID period (>0.44 V after 28.8 ns) V V Calculation complicated by changing target levels with onset of burst after 25.6 ns 0.44 0.01458 68.8 ns => R=50 ohms, C<120 pF 28.8 ns => R=50 ohms, C>701 pF AC coupling will not meet the GPON/LRPON specifications…this conclusion will scale to any data rate! * An additional factor is that shorter TCs will result in some level drifting which may impair the performance of the CDR, therefore the CDR may impose some limitation on min TC NOC conference, Berlin June 2006 Overview Burst Mode Receiver for LR-PON AC coupling DC coupling Edge detection NOC conference, Berlin June 2006 DC coupling – basic designs Feedback front end designs Feedback control Pre-amp Amplitude recovery done in preamp Requires differential preamp – making the design more complicated and expensive Feedback inherently more stable than feedforward therefore more reliable Slower settling time between bursts than FFW – important for reset circuitry + Peak detector Feedforward control Feedforward front end designs Pre-amp Can use a conventional DC coupled preamp - amplitude recovery in post-amp Less stable - more prone to oscillation + Peak detector NOC conference, Berlin June 2006 DC coupled - Main design considerations Peak detector may need to detect both high and low levels to prevent mark-space distortion, especially when the extinction ratio is poor Fast resets needed for recovery from bursts within guard period • Feedforward type favoured here as feedback settling seen as too slow • less of a problem with fixed packet length formats (ATM) but particularly important with variable burst length standards such as Ethernet. Nobody has yet implemented a PON compatible DC coupled BMR capable of operating at 2.5 or 10 Gbps NOC conference, Berlin June 2006 Overview Burst Mode Receiver for LR-PON AC coupling DC coupling Edge/Impulse detection NOC conference, Berlin June 2006 Edge response vs AC coupling V Tbit 0 t V TC>>Tbit Conventional AC coupling 0 t V TC<<Tbit 0 NOC conference, Berlin t June 2006 Edge detector – impulse characteristics Use a very short time constant (less than duration of 1 bit) V Tr(t) Tf(t) Vc(t) 0 Tbit NOC conference, Berlin June 2006 t Requirements 1. The pulse must exceed the comparator threshold voltage prior to the decision time 2. The noise present on the signal must not trigger a false reading following the first trigger and prior to the decision time 3. The statistical noise present on the comparator must be considered for a full analysis – although this is relatively insignificant c.f. amplified RX noise 4. Sensitivity determined when Vpeak = Vth for a noise contribution that results in an error probability of 10-4 (abs min values of ~-30dBm using typical data and assuming thermal noise limited) 5. PE necessarily improves for larger signals – maximum input power limited by TIA overload (typ. Few mA) NOC conference, Berlin June 2006 Noise/sensitivity analysis V tpeak Vpeak f(V) Vc(tdec) Vth+ t Vthtdec NOC conference, Berlin June 2006 Pe Bit error probability calculations Tdec increasing Plot of bit error probability against input SNR (Pin/N0) for increasing decision time NOC conference, Berlin June 2006 CID treatment A full BER analysis requires special treatment of multiple bits Multiple bits provide a special problem for this receiver design because if bit m from a sequence of n bits triggers a false level on the comparator then the subsequent n-m bits will be in error as well (unless another positive error is triggered) Can be reduced to a ER enhancement factor taking the first bit as a special case This work should conclude by the end of June ready for publication NOC conference, Berlin June 2006 Conclusions AC with long time constants not compatible with GPON due to the close pre-amble and CID durations DC coupled receivers still striving to exceed the GEPON 1.25 Gbps level Edge detection looking very promising and preliminary models are indicating it is capable of several Gbps operation NOC conference, Berlin June 2006 END NOC conference, Berlin June 2006