to k if cd = c, [I]. This value is used to select the binary digit r d , - > required to determine which output has to be used for routing the packet [I]. The packet ‘ad’ of the header comprises the first four bits representing the value cd and the last four bits representing rd. The intermediate eight bits are Os and are present in order to avoid interactions in the optical routing operation. T o obtain all-optical routing, cd is conveniently represented in unary coding (c,,,). A signal in unary form can be represented by a string of ‘0’s of length cd followed by a ‘1’. The quantity ( k + c d - c c , ) mod k can be computed by means of sum and mod k operations with the device shown in Fig. IC. The sum can be optically performed feeding the pulse representing cdl into a delay line DL,. The length of D,, is equal to the value (k - c,)T/8 and depends only on the r NIU position. To perform the operation mod k, the output of the delay line is forward to a loop L, producing a delay kT/8, which allows us to add the first k bits of the results to the second k bits, obtaining the correct operation mod k expressed by the second four bits. In this manner a pulse is produced in the time position equal to xT/8. By means of DL, the pulse is delayed by 2k - 1 in order to check, by means of the A N D gate, the value of the (x - 1)th bit of the rd part of the address. Based on the schematic diagrams of the devices in Fig. I we propose in Fig. 2 the schematic diagram of an NIU that is designed to operate with all-optical processing. Each NIU is subdivided in two equal parts, part a and part b. This switch allows the packet to be retransmitted by one of the two outputs (0,. 0,). Collisions among the packets coming from part a and part b are avoided, delaying the packets by means of the buffer loops L , and L , whose electro-optic switches S, and S , are driven by the central unit that chooses the priority. The gain of the optical amplifiers depends on their output saturation power which has to be lower than 100-200mW, and on the energy level at the optical gates. In particular for pulses of 1 ps duration, the XOR gate requires pulses with an energy of the order of 1 pJ [4], whereas AND gates require that the two input pulses have to present a product of their energy greater than 1 pJ2 [SI. Starting from all these energy requirements the gain of the amplifiers has to be chosen as follows: A, = 12 dB, A, = 6 dB, A, = 9dB, A, = 27dB, A, = 21 dB, A, = lOdB, while all the couplers but C, present a coupling ratio of 3dB. C , has to ‘spill’ a very low fraction of energy (lo/), which is detected from the photodiode to determine the arrival of a packet. According to all these considerations the total throughput of the network is equal to 0.048 x 0.43 1 1Tbit/s, where 0.43 is the throughput eficiency per NIU, in a multihop network with k = 4, p = 2 [3]. 22ndJune 1992 F. Matera and M. Settembre (Fondazione Ugo Bordoni, Via B. Castigfione 59,00142, Roma, Italy) E. Ripani (CNR, Via E . Castigfione 59,00142, Roma, Italy) clock References 1 ACAMPORA, A. s., and KAROL, M. J.: ‘An overview of lightwave packet network’, IEEE Network, January 1989, pp. 29-41 2 SOCCOLICH, c. E., ISLAM, M.N., HONG, B. I., CHBAT, M.,and SAUW,J. R.: ‘Application to ultrafast gates to a soliton ring network‘. Proc. of Nonlinear Guide-Wave Phenomena, Cambridge, 2nd-4th September 1991, pp. 366-369 3 MATFRA, I.,ROMAGNOLI, M.,SETTEMERE, M.,TAMBURRINI, M., and DE MARCHIS, G.: ‘Proposal of an all optical TDMA network. Proc. LAN, EFOCiLAN 91, London, UK, pp. 132-137 4 WOOD, D.: ‘Constraints on the bit error rates in direct detection optical communication systems using linear or soliton pulses’, J. Lightwaue Technol., 1990, LT-4, (7),pp. 1097-1106 kj part b M E M O R Y ARQ SCHEME W I T H LOW RELIABILITY ZONE DETECTOR C. Lau Indexing terms: Digital communication systems, Automatic repeat request The performance improvement of a memory ARQ scheme using majority voting decision and reliability zone detectors is analysed. Introduction: In a conventional ARQ system [l, 21, the receiver checks each received packet for possible errors. Any packet detected in error is discarded and a request for retransmission of the erroneous packet is made. In a memory ARQ system [3-51, the receiver stores the erroneous packet. It is then combined with subsequent retransmissions for possible error correction. In this Letter, we consider a memory ARQ system as shown in Fig. 1 in which a message to be transmitted is split into groups of maximum length K bits. Each group is then assembled into a packet of length n = K r bits where r represents the overhead required for addressing, error detection, etc. It is assumed that the error detection code used can detect all errors. Each packet denoted by (bi, b,, .. ., b,) where b, E (0, l}, i = 1, 2 , . .., n, is then sent over an additive white Gaussian noise channel using binary antipodal signalling. The + ELECTRONICS LETTERS r 13th August 1992 Vol 28 No. 17 . - 1571 input to the vector channel is x if bi = 0; else xi = -A. = (xl, x2,. . ., xn) where xi = A If Y, is a codeword, transmission is completed; otherwise a request is made for retransmission. T = 0 corresponds to the one-threshold memory ARQ scheme in Reference 4. Eoaluation ofrepeat request probability: Assume that an all '0' codeword is transmitted. The bit error probability a t the hard decision detector output is P, = f erfc [A/oJ(2)], where erfc (a) is the complementary error function of a. After J transmissions, the probability that the decision accumulator has value s is given by PBJS) = T decision ldetector x I - PJ-1 I s=j-2/ / = 0 , 1 , 2,..., j reliobility packet detector Fig. 1 Block diagram of memory ARQ system where Pb&) is the probability that the value s falls in the low reliability zone At the receiver, the received vector corresponding to the j t h transmission of x is y j = (yj, yj, 2 r . .., yj,J where yj, = xi nj, and {n,,;} represent outcomes of independent Gaussian random variables, each with variance aZ.The hard decision detector then carries out a hard estimate on the received vector y j . A weight wj. = 1 is assigned if the amplitude of the received signal yj,i is positive; otherwise, w j , i = -1 is assigned. A decision accumulator B j , i is maintained for each bit in the packet. After the jth reception of the packet, the accumulators are updated using Bo, = 0 and B,, = B j - 1. w j ,;. The outputs of the accumulators are examined by the reliability zone detector. The range of the weight stored in the accumulator is divided into a low reliability zone, - z j 5 B,, I zj, and high reliability zone B j , , < - z j and B j ,z z j , where z j is a predetermined zone threshold. This is illustrated in Fig. 2. + + + I i Zl-1 i -21.. high reliability zone (2) Because bit errors are independent, the repeat request probability is decision accumulator updating F k (;>.:(I I I -2 -1 i i 0 1 I 2 ... i ZI low reliability zone ,. PJj) is the probability that s is in the positive high reliability zone (5) and Pb&) is the probability that s is in the positive low reliability zone PbrcCi) = tPBi(0) c + s = 1P B J S ) (6) The repeat request probability, P,G),is plotted against the signal-to-noise ratio (SNR) 20 log,,(A/o) for n = 512, T = 5, j = 1, 2 , . . ., 5 and different values of z, in Fig. 3. The I i 21.1 (4) J high reliability zone Fig. 2 Weights associated with reliability zones Let t be the number of accumulators of which the weights fall within the low reliability zone, i.e. - z j IB j , i , , B j , i z ,..., B j , i ,5 zj. If f I T , where T is an integer number which is small compared to the maximum number of errors that the code can detect, the decoder input v, is determined from the following: (1) for those Bj, i , i # i,, i 2 . . .. , i,, which fall within the high reliability zone, the corresponding uj, I is set to '0'if B j , I > z j or to '1' if B , < -zi. (2),for those B j . i , i = i,, i,, ..., i,, which fall within the low reliability zone, the corresponding decoder inputs {uj. uj,i2, .. .,t i j , i , } form the set of 2' binary combinations in which one combination is equal t o {b;,, biz,..., bz,}. In this case (t IT), there are 2' received words. If those B, , falling within the high reliability zone are correctly received, then one of the 2' received words is a codeword and transmission of the packet is completed; otherwise a request is made for packet retransmission. For t z T, the decoder input is given by 1" o j , I= 1 1 or 0 with equal probability 1572 for B , z 0 for B j , < 0 for B j , = 0 (1) 3 4 5 6 7 8 9 1011 1213 SNR Fig. 3 Repeat request probability against SNR curve for j = 1 corresponds to the repeat request probability 1 - (1 - PJ'. After j = 2 transmissions, the repeat request probability drops substantially for SNR > 8 dB. We note that in memory ARQ systems without the reliability zone detector [4], there is no reduction in the repeat request probability when j increases from 21 1 to 21 + 2, 1 = 0, 1, 2, ... . For systems in which the number of packet transmissions, J, is to be chosen to achieve a certain repeat request probability, significant savings in SNR can be obtained by increasing j. For example, increasing; from 1 to 2 results in a saving of 3 dB for a target repeat request probability of ELECTRONICS LETTERS + 13th August 1992 Vol. 28 No. 17 Fig. 4 shows how the repeat request probability decreases with various values of j and T for S N R = 9dB. For a given j , the value of zj that gives the lowest P,(j) was used. From the Figure, we note that odd values of j should be used for T = 0 [4]. It can be seen that there is a substantial reduction in P,(j) for j 2 4 when T is increased to 1. Further increases in T result in smaller reductions. However, for j = 2 and 3, significant reductions in P,G) can only be obtained when T > 3. of coherent detection optical receivers, the direct detection optical receiver has the advantage of simplicity. FSK is an attractive modulation scheme because the transmitter laser frequency can be directly modulated by the current. FSK signals can be optically demodulated by converting them to baseband ASK signals. This can be carried out using a solid Fabry-Perot interferometer etalon as the FSK discriminator. Recently, the attractive capabilities of the DFB and DBR laser as a multifunction device have been investigated. The laser can be used for light emission or optical filtering and photodetection. Recent work has demonstrated FSK transmission using multielectrode DFB and DBR lasers for both transmitter and receiver [2, 31. The laser receiver acts as a filter, an FSK discriminator and a photodetector. More recently, using a two-electrode DFB laser for both transmitter and receiver, a 1.5 Gbit/s transmission with -20dBm sensitivities for bit error rate has been achieved [4]. In this Letter, we describe the ability to use a Fabry-Perot laser, with one antireflection coated facet, as an FSK discriminator/photodetector. We report FSK transmission up to 1.5Gbit/s with -30dBm sensitivities for bit error rate. Experiment: The experimental setup for the different measurements is depicted in Fig. 1. At the transmitting end, a number of transmissions J !f P Fig. 4 Repeat request probability against j V e Conclusions: The repeat request probability of a memory ARQ system can be significantly reduced by including a reliability zone detector in the receiver. This technique can also be applied to a memory ARQ system with soft decision detectors. I Dattern 1 23rd June 1992 C . Lau (School of Applied Science, Nanyang Technological University, Nanyang Avenue, Singapore 2263, Singapore) m References Fig. 1 Experimental configuration of transmission system s. R., and LIN, s.: ‘Selective-repeat-ARQ schemes for broadcast links’, IEEE Trans., 1992, COM-40, pp. 12-19 BENICE, R. I., and m y , I. A. H.: ‘An analysis of retransmission systems’, IEEE Trans., 1964, CS-11, pp. 135-145 3 BENELLI, G.: ‘An ARQ scheme with memory and soft error detectors’, IEEE Trans., 1985, COM-33, pp. 285-288 4 LAU,c., and LEUNG, c.: ‘Performance analysis of a memory ARQ scheme with soft decision detectors’, IEEE Trans., 1986, COM-34, pp. 827-832 R. : ‘Performance evaluation of efficient continuous 5 FANTACCI, ARQ protocols’, IEEE Trans., 1990, COM-38, pp. 773-781 1.524pm, two electrode GaInAsP/InP DFB buried ridge structure (BRS) laser was used. This DFB laser was fabricated on material grown by MOCVD, with a second-order corrugation. One of the two facets of the DFB laser is coated with a I44 layer of silicon monoxide, using electron beam evaporation, giving a residual reflectivity of 2%. This laser has two equal electrodes of 300pm each and presents a continuous tuning range of over 90GHz with a minimum spectral linewidth of 5 MHZ [SI. At the receiving end, a single-electrode 350pm long BH F P laser was used. One of the two facets of the FP laser is coated with an AR coating of 2.5 x reflectivity at 1.52pm. The electrodes of the FP laser were matched with a 47 R resistance and a bias current connection. The two lasers were connected bv a variable outical attenuator (VA) and a polarisation controller (PC). An optical spectrum analyser with O.lnm resolution was used to observe the spectral output of the two lasers. 1 CHANDRAN, 2 l.5Gbit/s TRANSMISSION SYSTEM USING FP LASER AS FSK DISCR I M INATORlPHOTO D ETECTOR ~ Results: Fig. 2 shows the spectral output of the F P laser biased at three times the lasing threshold current (Ifp= 53 mA). From Fig. 2, it can be seen that the F P laser presents M. J. Chawki, L. Le Guiner, D. Dumay and J. C. Keromnes -25 Indexing terms Optical receivers, Frequency shrft keying, Semiconductor lasers An optical FM receiver based on a monoelectrode FabryPerot laser, with one antireflection coated facet is reported The device, biased far above threshold, acts as an FSK discnminator/photodetector An FSK transmission system, up to 15Gbit/s with -30dBm sensitivities for BER, has been demonstrated using this laser. The speed is lmited by the transmitter FM response and not by the receiver response 2 U - 45 -39 9 4 -65 Introduction Performance analysis of direct detection optical receivers has been the subject of interest for many researchers over the past several years [l]. Compared to the complexity ELECTRONICS LETTERS 13th August 1992 15224 15274 w 15324 138612 Fig. 2 Spectral output of FP laser Vol. 28 No. 17 1573

1/--страниц