JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 21, NO. 11, NOVEMBER 2003 2455 Optical CDMA Random Access Protocols With and Without Pretransmission Coordination Hossam M. H. Shalaby, Senior Member, IEEE Abstract—The link layer of an optical direct-detection code-di- vision multiple-access (CDMA) packet network is considered. Two different protocols that need pretransmission coordination are proposed. A variation of the second protocol that does not need pretransmission coordination is discussed. Both system throughput and average packet delay are derived and investigated for two different receiver models: the correlation and chip-level re- ceivers. Both multiple-access interference and the photodetector’s shot noise are taken into account in the analysis. The case where the number of users exceeds the available number of CDMA codes is numerically investigated. Our results reveal that the proposed protocols yield competitive system throughputs when used with the correlation receivers. Further, significant improvement in the throughput is achieved when using chip-level receivers along with the second protocol. Index Terms—Chip-level receivers, code-division multiple access (CDMA), correlation receivers, direct detection optical channel, on–off keying, optical CDMA, optical link layer, optical networks. I. INTRODUCTION O PTICAL fibers offer a large bandwidth in the order of ter- ahertz, making it the best candidate for current and future communication and computer networks. Optical code-division multiple-access (CDMA) systems [1]–[16] have been shown to be competitive candidates in order to mine this terahertz band- width when combined with wavelength-division multiplexing (WDM) techniques. In optical CDMA techniques, a user is normally given a signature code that satisfies good auto- and cross-correlation properties [1] to help in its data transmission and identifying itself. Several receiver detection models have been proposed in literature. Some of them are summarized in [14]. The most traditional ones are the correlation receiver [2], correlation receiver with double hardlimiters [8], [9], and chip-level receiver [10]. The main difference between the correlation receivers and chip-level receivers is that in the latter, the bit decision rule depends on the received optical power in each mark chip of the signature code, whereas in the former, it depends on the total optical power in all underlined mark chips. A comparison between chip-level and double-optical-hardlimiters correlation receivers can be found in [16]. It was shown that although chip-level receivers are much simpler and more practical than double-optical-hardlimiters correlation receivers, the bit error probabilities of both of them are almost similar to each other, even under ideal conditions for the optical hard Manuscript received October 31, 2002; revised June 23, 2003. The author is with the Department of Electrical Engineering, University of Alexandria, 21544 Alexandria, Egypt (e-mail: shalaby@ieee.org). Digital Object Identifier 10.1109/JLT.2003.819533 limiters. Other powerful, but rather complex, receiver models have also appeared in the literature. These include multiuser detection receiver [6], interference estimation and cancellation [11] receiver, etc. Basically two types of optical direct-detection CDMA signal formats have been studied in the literature: binary on–off keying (OOK) [2]–[6] and -ary pulse-position modulation (PPM) schemes [4], [7], [12]. It has been shown that for fixed data rate and chip duration, there is no advantage in using PPM in place of OOK, but PPM becomes superior to OOK if the average power rather than the chip time is the constraining factor. Although there is a lot of research in the field of optical WDM that has been done at the level of network layer [17]–[23], most of the research in the field of optical CDMA has focused on the physical layer of the network [2]–[16]. There are, however, a few authors [26]–[32] that have examined the network or link layer of optical CDMA communication systems. In this paper, we propose two different protocols for slotted optical CDMA packet networks. These protocols, called Pro 1 and Pro 2, need pretransmission coordination; and a control packet is sent by a transmitter before launching its data. Of course in order to implement Pro 1 and Pro 2, we need both the transmitter and the receiver be tunable. That is they should be able to tune their signature codes to the one assigned in the control packet. Furthermore, we suggest a variation of Pro 2 that does not need pretransmission coordination. Of course the im- plementation of this variant protocol does not require any re- ceiver tunability, and is thus simpler. With the aid of cyclic redundancy check (CRC) codes, a receiver can determine whether a received packet is correctly detected or not. If not it will ask for retransmission. This of course would increase the channel traffic and interference. A transmitter asked for data retransmission is not allowed to generate new packets; rather it keeps retransmitting the same packet (after random delay time slots) until it receives a successful acknowledgment from destination. Since under normal situations the network users send their data in a burst mode, i.e., they are not all active at the same time, we will allow the total number of users to exceed the number of available codes. Two types of performance measures are examined in this paper. The first one is the average system (or network) throughput in packets per slot, which tells how many packets on the average are received successfully per time slot. The other one is the average packet delay in time slots, which tells after how many slots (from transmission) on the average a packet will be received successfully. Our second aim in this paper is to figure out which of the two proposed protocols leads to a better performance in terms of average throughput (in packets 0733-8724/03$17.00 © 2003 IEEE