Bit Error Rate Performance of Multiple-Channel OTDM Demultiplexer Employing A Chained Symmetric Mach-Zehnder Switch H. Le Minh, Student Member, IEEE, Z. Ghassemlooy, Senior Member, IEEE, Wai Pang Ng, Member, IEEE and M. F. Chiang Optical Communications Research Group, School of Computing, Engineering and Information Sciences Northumbria University, Newcastle upon Tyne, NE1 8ST, UK Email: {h.le-minh, fary.ghassemlooy,, ming-feng.chiang} Abstract— This paper presents a high-speed multiple-channel optical time division multiplexed (OTDM) demultiplexer based on a chained symmetric Mach-Zehnder (CSMZ) switch. In CSMZ switch semiconductor optical amplifier (SOA) are shared between two chained arms in contrast to the MZ based multiple-channel demultiplexers where there is an SOA in each arm. The analysis for bit error rate (BER) and the average received power penalty is presented and the results are compared with the simulated data. It is shown that the power penalty incurred is highly dependent on the value of coupling ratio and the channel number. Index Terms— OTDM demultiplexing, chained symmetric MachZehnder, coupling factor, bit error rate, received power penalty.




APIDLY GROWING Internet traffic volume is the major driving force for the deployment of low cost ultrahigh capacity optically-transparent photonic communication networks. In order to increase traffic capacity, photonic networks could utilize one (or both) of the multiplexing schemes: wavelength division multiplexing (WDM) and optical time division multiplexing (OTDM). WDM (and its variants) scheme has been widely researched and adopted in many practical systems, offering a huge capacity using a large number of wavelengths. Whereas OTDM offers equivalent capacity compared to WDM, but only using a single wavelength. In OTDM scheme, a data packet is generated by interleaving time-delayed ultra-short optical pulse carriers (at a single wavelength) where each carrier is modulated by a baseband data signal. Nevertheless, simultaneous demultiplexing of multiple high-speed OTDM channels is a challenging task due to the increased number of channels and the complexity of optoelectronic and optical devices operating at high bit rate (≥ 100 Gbit/s), thus there is a need for a solution in the optical domain. A number of optical configurations for multiple-channel OTDM demultiplexing have been proposed including the photonic serial-to-parallel converter (PSPC) based on the Lithium Niobate electro-opto modulator [1], surface-emitted

second-harmonic generation (SESHG) employing hybridized semiconductor/silica waveguide circuits [2] and an array of optical single-channel SOA-based demultiplexers [3-5]. Among these structures, the latter is the most promising approach owing to its extremely narrow demultiplexing window (DW) in the order of picosecond [6], which is suitable for current and future ultrahigh-capacity OTDM demultiplexing. The SOA-based demultiplexer includes diverse schemes such as terahertz optical asymmetric demuliplexer (TOAD) [3], ultrafast nonlinear interferometer (UNI) [4] and symmetric Mach-Zehnder (SMZ) [5]. The latter is the most widely used scheme compared to TOAD and UNI offering seamless circuit-integration property and using an external control signals to set up the DW with delays [6]. The SMZ is composed of two equal arms with an SOA located at the same location on each arm. Demultiplexing (switching) of data channel is carried out by applying delayed control signals to saturate the SOAs, thus resulting in non-linear gain and phase changes between the two arms [5]. Single-channel demultiplexing by SMZ is relatively straightforward, since only two SOAs are required. However, demultiplexing of multi-channels requires an array of SMZs, which is relatively complex requiring a large number of SOAs. In this paper, we report, what we believe for the first time, a new multiplechannel demultiplexing using the chained SMZ configuration requiring only half the number of SOAs compared to an array of SMZs. We investigate the BER performance and the optical power penalty of the proposed scheme at data rate of 100 Gbit/s and compared it with the back-to-back (B-B) system. The paper is organized as follows: after introduction, Section II describes the operation principle and theoretically investigates the BER performance of the proposed CSMZ. Results are presented and discussed in Section III and finally, Section IV concludes the paper. II. CHAINED SYMMETRIC MACH-ZEHNDER DEMULTIPLEXER An asynchronous OTDM system shown in Fig. 1(a) comprises of an OTDM transmitter and multiplexer (OTDMMUX) an optical fiber link, a CSMZ demultiplexer, an optical

1930-529X/07/$25.00 © 2007 IEEE This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.


10 Gbit/s baseband electrical RZ data


Clk Ch.1


OTDM packets



CP1 Control signal


CP1 CI,1

Control pulse

OTDM packet

BER estimator





OTDM packets










Output 3








Output 2






Clock recovery




Output 1




10 Gbit/s pspulse-width light source















Output M


Figure 1. (a) Schematic diagram of asynchronous OTDM system using M-CSMZ and (b) optical circuit diagram of M-CSMZ demultiplexer TABLE I TRANSFER FUNCTIONS OF COUPLER, COMBINER AND SPLITTER

ECu,x I1u,x I1d,x

Schematic 2×2 coupler

2×1 combiner

ECd,x I0d,x I0u,x


1×M splitter

Ein_1 Ein_2 Ein_1 Ein_2 Ein



Transfer function


ª E o _ 1 º ª(1 − α ) 2 «E » = « 1 ¬ o _ 2 ¼ ¬ jα 2


Eo =


Eo_1 Eo_2 Eo_M




(1 − α )



º ª Ein _ 1 º » «E » ¼ ¬ in _ 2 ¼

1 (Ein _ 1 + Ein _ 2 ) 2

Eo _ 1 = Eo _ 2 = ... = Eo _ M =

Ein M

Figure 2. Eye pattern of xth channel

receiver and a BER estimator. The transmitted OTDM packet is composed of a clock channel, a single bit “1” for packet synchronization, and M modulated 10-Gbit/s data channels (single bits). At the OTDM demultiplexer, the clock bit is recovered using an asynchronous self-clock-recovery unit [7], which drives the CSMZ. The output of the demultiplexer is applied to the optical receiver. The demultiplexed signal is passed through an optical bandpass filter to reduce the SOAs spontaneous-spontaneous noise prior to signal detection. Note that the control and data signals are set in orthogonal polarization by the polarization controller and separated at the output of CSMZ by the polarization beam splitter. Finally, a BER estimator is utilized for estimating the BER performance of demultiplexed signal and the power penalty induced by the CSMZ. A. Operation Principles of M-channel CSMZ Demultiplexer A M-channel CSMZ demultiplexer, depicted in Fig. 1(b), comprises of an 1×M input splitter, M of 3-dB 2×2 input CI,x and output CO,x couplers (x is channel number), 2×1

combiners, 1×2 splitters (see Table I) and identical SOAs in M+1 CSMZ arms (A1 to AM+1). The inclusion of 3-dB attenuators, located at each arm, will ensure identical optical powers at the SOAs and CO,x couplers, thereby ensuring a balance state between each pair of adjacent arms. Thus, in the absence of a control pulse (CP), no input signal emerges at the CSMZ output ports. For demultiplexing purposes, the extracted clock signal is split into M+1 high-powered CPs with equal intensities. CP1 applied to SOA1, just prior to arrival of data channel 1 at the interferometer I1,2, sets I1,2 to an imbalance state in order to demultiplex the data channel 1 to the output 1. Note that in an imbalance state SOA1 and SOA2 will have different gain and phase profiles, a requirement for switching. Since all other interferometers Ik,k+1 are still in the balance state, then no signal should emerge from the other output ports. CP2 (delayed by 1×Tb) applied to SOA2 restore a balance state in I1,2 and sets I2,3 to an imbalance state (i.e. gain and phase difference between SOA2 and SOA3). By doing so, it demultiplexes the data channel 2 to the CSMZ output 2. Similarly, by applying the CPs (delayed by (x–1)×Tb) to the appropriate SOAsx the corresponding xth channel are

1930-529X/07/$25.00 © 2007 IEEE This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.

demuliplexed. Note that, to complete demultiplexing of Mth channel, CPM+1 is applied to AM+1 to restore the balance state in IM,M+1. Assume that the electrical field of the input signal is Ein [8], the signals at the output of the SOAs are computed by:


E1 =

where EC is the eye closure illustrated in Fig. 2 at tsampling [10] and is given by:


2M Ein


EC x , u/d = K Pin [DWx ,1 (α I/0, u/d ) − DW x, 0 (α I/O,u/d )]


Eo _ m =

E in 2 M E in




g 1 e − jφ1 − K 1, 2 g 2 e − jφ2

m ,1

ηin and ηout are the SOA input and output coupling efficiencies, respectively, G is the SOA gain, Lf is the optical filter loss and Rp is the photodetector responsivity. The subscript 0 or 1 in DWx denotes the OFF/ON states of the CSMZ output xth. σx,0/1 is the standard deviation of the total noise power for bit 0 or 1 (0/1) with the corresponding parameters given in Table II and [11]. u and d represent the undistorted and distorted EC for α = 0.5 (ideal) and α ≠ 0.5, respectively. The noise powers σ x2,1 and σ x2, 0 for the received bit 1 (mark)


g m +1 e − jφm +1 − K m , 2 g m e − jφm


2 M E in E o _ m +1 = K m +1,1 g m +1 e − jφm +1 − K m +1, 2 g m + 2 e − jφ m + 2 2 M E in Eo _ M = K M ,1 g M +1 e − jφM +1 − K M , 2 g M e − jφ M 2 M



and bit 0 (space), respectively, including the SOA amplified spontaneous emission (ASE) noise from CSMZ and the receiver noise could be computed by [9]:



2 2 σ x2, 0 / 1 = σ SOA , 0 / 1 + σ receiver , 0 / 1


2 noise ( σ receiver ) are given by [9]: ,0 / 1 2 2 2 σ SOA , 0 / 1 = 4I x, 0 / 1 I ASE Be / B0 + I ASE (2 B0 − Be ) / B0



K m +1,1 =



1 (1 − α O,m+1 )12 (1 − α I ,m )1 2 2

[ [

I x, 0 /1 = K Pin DWx ,0 /1 (α I/O,u/d )


I ASE = 0.5 NFSOA Gη out qBo L f




1 1 1 1 1 1 1 1 § 12 1 1 1 ¨ α I ,1 + α I ,22 ·¸α O ,21 ; K m,1 = (1 − α O ,m ) 2 (1 − α I ,m ) 2 + (1 − α I ,m+1 ) 2 ; K m ,2 = α O ,2m §¨ α I ,2m−1 + α I ,2m ·¸ ¹ © ¹ 2© 2 2 1 1 1 1 1 1 12 § 12 1 12 § 12 · 2 2 2 2 + (1 − α I ,m+1 ) ; K m +1,2 = α O ,m +1 ¨ α I ,m +1 + α I ,m +2 ¸ ; K M ,1 = (1 − α I ,M ) (1 − α O ,M ) ; K M ,2 = α O ,M ¨ α I ,M −1 + α I ,2M ·¸ © ¹ © ¹ 2 2

K 1,1 = (1 − α I ,1 ) 2 (1 − α O ,1 ) 2 ; K1,2 = 1


where NFSOA is the SOA noise figure, q is the electron charge and Bo is the optical bandwidth. Note from (3) – (12), the BER performance and hence the average power penalty at the receiver depend on the input/output coupling ratios, which is investigated in the next

The BER of the demultiplexed xth data channel is estimated by [9]:



The average received powers for mark/space and the ASE power are:

B. BER Performance and Coupling Factor Variation



2 2 σ receiver , 0 / 1 = 2 q I x , 0 / 1 + I ASE Be + [4kTk / RL + i a ]Be

power gain G relates to g as G = g 2 defined in [8].



2 ) and receiver where the powers of SOA-ASE noise ( σ SOA ,0 / 1

where gk and phase φk are the temporal-variant field gain and phase, respectively, of a complex gain of SOAk induced on the electrical field of signal propagating through [8] (1 ≤ k ≤ M+1). m is even and the coefficients Kij are computed by (2) where αI,x and αO,x are the coupling factors of the CI,x and CO,x couplers, respectively. The demultiplexing window gains, therefore, can be computed from (2) with DWx = Po _ x Pin = (E o _ x E o* _ x ) (Ein Ein* ) and the temporal

BERx ,u/d = 0.5 erfc Qx ,u/d







1 1 K 12,1 G1 + K 12, 2 G 2 − 2 K 1,1 K 1, 2 G 1 G 2 cos ∆φ1, 2 ; DW m +1 = K m2 +1,1G m +1 + K m2 +1, 2 G m + 2 − 2 K m +1,1 K m +1, 2 G m +1 G m + 2 cos ∆φ m +1,m + 2 4M 4M 1 1 DW m = K m2 ,1 G m +1 + K m2 , 2 G m − 2 K m,1 K m ,2 G m G m +1 cos ∆φ m , m +1 ; DW M = K M2 ,1G M +1 + K M2 , 2 G M − 2 K M ,1 K M , 2 G M G M +1 cos ∆φ M , M +1 4M 4M where ∆φ i , j = −0.5α LEF ln (G i G j ) DW1 =


Pin is the average CSMZ input power, K = η in Gη out L f R p ,


Note that the k is even. The fields at the M CSMZ output ports are therefore determined by:

Eo _1 =


Qx ,u/d = EC x,u/d /(σ x, 0 + σ x ,1 )

g1e − jφ1 (1 − α I ,1 ) 2

1 1 g k e − jφk §¨α I ,k2 −1 + α I ,2k ·¸ © ¹ 2 2M Ein 1 1 Ek +1 = g k +1 e − jφk +1 (1 − α I ,k ) 2 + (1 − α I ,k +1 ) 2 2 2M E 1 EM +1 = in g M +1 e − jφM +1 (1 − α I ,M ) 2 2M

Ek = j

where the Q factor is defined by [10]:




1930-529X/07/$25.00 © 2007 IEEE This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.



Back-to-Back, prediction Back-to-Back, simulation Demux 100-to-10Gbps, prediction Demux 100-to-10Gbps, simulation







-10 -38


-36 -35 -34 -33 Received power (dBm)

OTDM packet




Clock pulse

Data channel 5

Data channel 1

Data channel 6

Power penalty (dB)

3 2.5 2 1.5 1 0.5 0 1









Data channel number

Data channel 2

Data channel 3

Data channel 4

Data channel 7

Data channel 8

Data channel 9

Figure 3. Simulated time waveform of 100-to-10 Gbit/s OTDM demuliplexed data channels and extracted clock pulse

section. III. RESULTS AND DISCUSSIONS The proposed scheme adopted for a parallel 100-to-10 Gbit/s demultiplexing of 9 data channels is investigated theoretically and the results obtained are compared with the simulation data (in VPITM). The OTDM packets are generated from nine 10-Gbit/s sources with the pseudo random bit sequence PRBS of (215 – 1) in addition to a clock-pulse channel. A guard-slot of 0.5 ns is inserted between consecutive packets for gain recovery in the clock extraction module. Packets are transmitted over a 70 km optical fiber span including the SMF and DCF. Figure 3 illustrates the time waveforms of demultiplexed data channels and the extracted

(b) Figure 4. (a) BER vs. received power of channel 3 and (b) power penalty vs. channels

clock pulse (Note that the clock pulse is separately extracted by a self-clock-recovery unit with a dedicated clock extraction delay [7]). The amplified extracted clock pulse is used as the CPs, sequentially applied to the CSMZ. A delay of 10 ps is used between each CPs. Figure 4(a) shows the predicted and simulated results for the BER against the received optical power for a demultiplexed channel (the 3rd channel) taking into account the noise sources induced by the SOAs. The predicted and simulated results are slightly different since the prediction/theoretical model oversimplies the switch residual crosstalk [11], in which TSW (i.e. Tb) is relatively small compared to the SOA gain recovery [12]. Also shown for comparison are the results for the B-B system. At BER of 10-9, the predicted and simulated power penalties are 1.2 and 1.7 dB, respectively compared with the B-B case. Depicted in Fig. 4(b) are the power penalties for different demultiplexed data channels, showing an average value of ~2 dB. Power penalties incurred also depend on the input and output CSMZ coupling ratios α as illustrated in Figs. 5(a) and (b) for an even and an odd channels, respectively. The power penalties (predicted and simulated) increase by a few dB for αO ≠ 0.5 and by 1 dB for αI ≠ 0.5. This power penalty variations can be explained from (2) where the dominant term is αO,x, and as a result DW, EC and hence BER becomes more sensitive to αO,x than to αI,x.

1930-529X/07/$25.00 © 2007 IEEE This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.


8 Theorical, CO4

Power penalty (dB)


Theorical, CI4


Simulation, CO4


Simulation, CI4

4 3 2 1 0



0.5 0.55 Coupling ratio


The paper introduced and characterized an ultrafast parallelchannel OTDM demultiplexer based on the CSMZ. BER was investigated analytically, and results were compared with the simulated data showing a good agreement up to BER of 10-8. The power penalty incurred is highly dependent on the value of coupling ratio and the channel number. It was shown that the offset in coupling factor of output coupler would introduce more power penalty to the performance than the offset in input coupler. CSMZ has shown a high potential for parallel highspeed and multiple-channel OTDM demultiplexing applications in future ultrahigh-capacity photonic networks





Theorical, CO5

Power penalty (dB)


Theorical, CI5


Simulation, CO5


Simulation, CI5




3 2


1 0



0.5 0.55 Coupling ratio


(b) Figure 5. Power penalty vs. coupling ratio (α) for (a) even (4th) and (b) odd (5th) channels

The incurred power penalties against the coupling ratio values also depend on the data channel being demultiplexed. For even channels, the power penalty is lower for αI/O,x < 0.5 compared to αI/O,x > 0.5, whereas the opposite is true for the odd channels, see Figs. 4(c) and (d), respectively. The reason for this characteristic is explained as follows. When demultiplexing the xth channel, SOAs are sequentially excited by CPx and CPx+1, thereby the data signal in the lower arm Ax+1 experiences more amplification than the upper arm Ax, because of the unsaturated

G x+1− SOAx+1 > saturated Gx − SOAx , see Fig.

1(b) and Table II. For odd channels, the demultiplexed signal intensity I1d,x (Fig. 2) is higher for αI/O,x > 0.5 compared to αI/O,x < 0.5 (due to greater signal contribution from the lower arm). As a result ECx,d(αI/O,x > 0.5) > ECx,d(αI/O,x < 0.5), which contributes to a lower BER and power penalty. However, for even channels, ECx,d(αI/O,x > 0.5) < ECx,d(αI/O,x < 0.5), resulting in a higher power penalty for αI/O,x > 0.5 compared to αI/O,x < 0.5.




[8] [9] [10] [11]


M. L. Dennis, W. I. Kaechele, W. K. Burns, T. F. Carruthers, and I. N. Duling, "Photonic Serial-Parallel Conversion of High-Speed OTDM Data," IEEE Pho. Tech. Lett., vol. 12, pp. 1561-1563, 2000. T. G. Ulmer, M. C. Gross, K. M. Patel, J. T. Simmons, P. W. Juodawlkis, B. R. Washburn, W. S. Astar, A. J. SpringThorpe, R. P. Kenan, C. M. Verber, and S. E. Ralph, "160-Gb/s Optically Time-Division Multiplexed Link with All-Optical Demultiplexing," IEEE Light. Tech., vol. 18, pp. 1964-1977, 2000. J. P. Sokoloff, P. R. Prucnal, I. Glesk, and M. Kane, "A Terahertz optical asymmetric demultiplexer (TOAD)," IEEE Photon. Technol. Lett., vol. 5, pp. 787-790, 1993. S. A. Hamilton, B. S. Robinson, T. E. Murphy, S. J. Savage, and E. P. Ippen, "100 Gbps optical time-division multiplexed network," IEEE Light. Tech., vol. 20, pp. 2086-2100, 2002. T. Tekin, C. Schubert, J. Berger, M. Schlak, B. Maul, W. Brinker, R. Molt, H. Ehlers, M. Gravert, and H.-P. Nolting, "160 Gbit/s error-free all-optical demultiplexing using monolithically integrated band gap shifted Mach-Zehnder interferometer (GS-MZI)," proc. IPR 2002, Vancouver, Canada, pp. IWC4, 2002. C. Schubert, J. Berger, S. Diez, H. J. Ehrke, R. Ludwig, U. Feiste, C. Schmidt, H. G. Weber, G. Toptchiyski, S. Randel, and K. Petermann, "Comparison of interferometric all-optical switches for demultiplexing applications in high-speed OTDM systems," IEEE Light. Tech., pp. 1-7, 2002. H. Le-Minh, Z. Ghassemlooy, and W. P. Ng, "Ultrafast all-optical self clock extraction based on two inline symmetric Mach-Zehnder Switches " proc. ICTON 2006, Nottingham, UK, vol. 4, pp. 64-67, 2006. M. Eiselt, W. Pieper, and H. G. Weber, "SLALOM: Semiconductor Laser Amplifier in a Loop Mirror," IEEE Light. Tech., vol. 13, pp. 2099-2112, 1995. N. A. Olsson, "Lightwave Systems With Optical Amplifiers," Light. Tech., vol. 7, pp. 1071-1082, 1989. J. D. Downie, "Relationship of Q penalty to eye-closure penalty for NRZ and RZ signals with signal-dependent noise," IEEE Light. Tech., vol. 23, pp. 2031-2038, 2005. Z. Ghassemlooy, W. P. Ng., and H. Le-Minh, "BER performance analysis of 100 and 200 Gbit/s all-optical OTDM node using symmetric Mach-Zehnder switches," IEE Proc. Circ. Devi. Syst., vol. 153, pp. 361-369, 2006. T. S. El-Bawab, Optical switching, ch. 7, Springer, 2006.

1930-529X/07/$25.00 © 2007 IEEE This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE GLOBECOM 2007 proceedings.

Bit Error Rate Performance of Multiple-Channel OTDM ...

APIDLY GROWING Internet traffic volume is the major driving force for the deployment of low cost ultrahigh capacity optically-transparent photonic ... MUX) an optical fiber link, a CSMZ demultiplexer, an optical. R. 1930-529X/07/$25.00 © 2007 ...

298KB Sizes 4 Downloads 238 Views

Recommend Documents

Bit Error Rate Performance Comparison of Equal ...
Channel Spacing and Repeated Unequal Spacing in. Reducing the ... Email:[email protected]. †. Department of ... (DSFs) in optical WDM systems enable a very high bit rate by keeping ..... dispersion schemes,” Opt. Eng., vol. 42, pp.

Bit Error Rate Performance Comparison of Equal ...
Abstract—The bit error rate (BER) performance comparison of a wavelength ... decreases due to a number of fiber nonlinearity effects, like. FWM, self and cross ...

Evaluation of Watermarking Low Bit-rate MPEG-4 Bit ... - CiteSeerX
other digital networks are means for freely and widely distributing high fidelity duplicates of digital media, which is a boon for authorized content distribution but ...

Evaluation of Watermarking Low Bit-rate MPEG-4 Bit ... - CiteSeerX
using a spatial-domain or transform-domain watermarking technique, and then ..... Average subjective test scores of each watermarked bit-stream and bit-rate.

An Exact Closed-Form Expression for Bit Error Rate of ...
prove the performance of wireless communications over fading channels without the need ... error rate of DF relaying systems that employ MRC at the des- tination are ...... PCS mobile phone systems and LG chairman's office planning future ...

Digital watermarking of low bit-rate advanced simple ...
THE INTERNET and other digital networks offer free and wide distribution of ... providers track, monitor, and enforce usage rights in both digital and analog form. ... 12 Mb/s, which are more suitable for DVD and digital TV broadcast than for low ...

Digital watermarking of low bit-rate advanced simple ...
providers track, monitor, and enforce usage rights in both digital and ..... of the spread-spectrum message signal for coping with host signal interference and sub-.

High-bit-rate continuous-variable quantum key ...
Oct 22, 2014 - High-bit-rate continuous-variable quantum key distribution. Paul Jouguet,1 David ..... is the loss coefficient of the optical fiber. For distances ...

Jointly optimized bit-rate/delay control policy for ...
Wireless Packet Networks With Fading Channels. Javad Razavilar ... K. J. R. Liu and S. I. Marcus are with the Electrical and Computer Engineering. Department and Institute ...... B.S. degree from the National Taiwan University,. Taiwan, R.O.C. ...

Efficient Minimum Error Rate Training and Minimum Bayes-Risk ...
Aug 2, 2009 - Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, ..... operation and it is identical to the algorithm de-.

Optimizing expected word error rate via ... - Research at Google
However expected word edit distance over a lattice is harder to compute than the expected frame-level loss .... to exactly compute the expected number of word errors over a lattice by expanding the lattice using error .... 20 million anonymized utter

Symbol Error Rate Expression for Decode-and-Forward ...
Apr 4, 2009 - prove the performance of wireless communications over fading ... The advantage of this scheme is that it not only allows us to optimize the.

Lattice-based Minimum Error Rate Training for ... - Research at Google
Compared to N-best MERT, the number of ... and moderate BLEU score gains over N-best. MERT. ..... in-degree is zero are combined into a single source.

Optimizing expected word error rate via ... - Research at Google
of a score FST S(z) and a decoder graph FST C ◦ L ◦ G. The decoder graph is itself a .... Figure 1: Example python implementation of sampled MBR training. ... We compared the performance of sMBR and EMBR training .... VS 0 noisified. 13k.

Efficient Minimum Error Rate Training and ... - Research at Google
39.2. 3.7. Lattice MBR. FSAMBR. 54.9. 65.2. 40.6. 39.5. 3.7. LatMBR. 54.8. 65.2. 40.7. 39.4. 0.2. Table 3: Lattice MBR for a phrase-based system. BLEU (%). Avg.

High Bit Rate Continuous-Variable Quantum Key ...
Introduction. Slice reconciliation. High bit rate CVQKD. Summary. Quantum Key Distribution. Quantum Channel. IAB. IAE. IBE. Exchange of quantum states. Induced correlations. arXiv:1406.1050. Slide 2/13 ...

Exact Bit Error Probability of Multi-hop Decode-and ...
sion for M-PSK is presented for multi-hop Decode-and-Forward Relaying. (MDFR) scheme, in ... and transmit jointly as a virtual antenna array. This enables them ... (MRC) or selection combining (SC) have been published [2-9]. In particular, in.

On the bit error probabilities of GMSK in the Rayleigh ... - IEEE Xplore
detector, also known as MSK-type receiver, can be employed instead of a rather complicated ... literature, a number of studies have been reported focusing.

Deterministic Performance Bounds on the Mean Square Error for Near ...
the most popular tool [11]. However ... Date of publication November 27, 2012; ... of this manuscript and approving it for publication was Dr. Benoit Champagne.

Deterministic Performance Bounds on the Mean Square Error for Near ...
mean square error applied to the passive near field source localization. More precisely, we focus on the ... Index Terms—Deterministic lower bounds, mean square error, near field source localization, performance analysis ..... contained in the samp