Temporal/spatial optical CDMA networks-design, demonstration, and comparison with temporal networks

1160

IEEE PHOTONICS TECHNOLOGY LETIERS, VOL. 4, NO. 10, OCTOBER 1992

Temporal/ Spatial Optical CDMA Networks-Design, Demonstration, and Comparison with Temporal Networks Eugene Park Antonio J. Mendez, a n d Elsa M. Garmire Abstract-We present the design and experimental results of a temporal / spatial (T / S) noncoherent optical code-division multiple access (CDMA) network, based on matrix codes, using a breadboard of passive multimode fiber-optic couplers and delay lines. We show that a T / S CDMA network allows for shorter bit times, given a set laser pulse width, compared to a temporal CDMA network. Also, T / S codes result in a reduction of autocorrelation sidelobes and cross-correlation peaks, and the T / S network has lower losses. Finally, we show that the couplers are critical components in maintaining code integrity.

In this letter we present the network architecture used to implement P O T / S matrix codes and show experimental results of autocorrelation and cross-correlations using a simple passive breadboard of multimode (MM) fiberoptic couplers and delay lines. The design and results are compared to a temporal CDh4A network using PO (0, 1) pulse sequences, where the maximum cross-correlation and autocorrelation sidelobes of these codes equal one [6]. We will show that coupler splitting uniformity was important in maintaining code integrity in both the temporal and T/S cases. The codes used in our experiments ETHODS have been proposed to use optical code- were developed at the University of Southern California’s division multiple access (CDMA) techniques to Communication Sciences Institute [6], [SI. Because the multiplex data among many users by encoding bits with goal of our experiments was to demonstrate optical CDh4A unique optical pulse signature sequences [11-[6]. These in the context of a short-haul environment, e.g., LAN, the systems and codes are one-dimensional, utilizing only the issue of pulse dispersion was not considered. time domain. In using one-dimensional temporal codes The optical T/S CDMA network is shown in Fig. 1. for a specified bit time, lengthy codes require very short The number of users is denoted by NI (the number of pulse lasers; or, for a specified laser pulse width, the bit codes in the PO set), w is the number of encoder/detime increases linearly with code length, decreasing the coder delay lines (the code weight), and f is the number data rate. A remedy to these problems is to code in two of space channels (the number of rows in the matrix : dimensions, namely time and space, using a hybrid CDMA code). The shaded regions of the encoders/decoders repnetwork [7]-[9]; we call this a temporal/spatial (T/S) resent optical interconnections that are set by the T/S network. One can use two-dimensional codes (matrices) codes. The code and network parameters are interdepenwhich, in comparison to the temporal codes and for the dent, that is, one can either choose the network paramesame number of users and code weight, offer a dramati- ters (4, w , and f) and then derive the codes, or use a cally decreased effective temporal code length [71, [SI, given code set and design the network from it. resulting in an increased data rate or more lenient restricThe T / S matrix codes have dimensions of f X L, where tions on the source pulse width. Also, using a T/S net- L , is the number of columns, related to the temporal work exploits the lower cost of fiber ribbons or bundles length. The ith code in the PO set is denoted by { M ( N , , compared to short pulse laser sources. The disadvantages w,f ) J where M signifies a “matrix” code-and i goes from of this scheme are that the network fabric must include a 1 to NI. Our experiments have focused on the T/S single distribution star for each space channel, and the optical pulse-per row (SPR) network, which uses matrices with path lengths among all the channels between each en- the property of having at most a single “1”(pulse) per row coder/decoder and the distribution star must be equal for [SI, because it is simple to implement. Fig. 2 shows an the code to maintain its integrity. Generally, CDMA codes example of a T/S SPR encoder/decoder scheme where have the property of being pseudo-orthogonal (PO): their w = f = 4; in general, w does not have to equal f , as autocorrelation maximum equals the weight w of the shown in the boxed area. Notice that, unlike the temporal code, and their cross-correlation maximum is less than w case, there are no sidelobes in the autocorrelation. This is but greater than zero [2], [3]. Codes with low cross-corre- unique to T/S SPR codes because of the time-space lation maxima are well suited for optical CDMA networks correlation process. A n important fact to notice is that employing threshold detection because of the high distin- T/S SPR encoders/decoders require only one 1 X w guishability between the maximum of the autocorrelation coupler, as opposed to temporal encoders, which require and the cross-correlation. All codes must fit within a bit two (see [5]), consequently reducing the single-path losses. time Tb, and the optical pulse must fit within a chip time In comparison, for the temporal case the ith code is T, where T, = T,/(codelength). 1041-1135/92$03.00

0 1992 IEEE

PARK et al.: TEMPORAL/SPATIAL OPTICAL CDMA NETWORKS

1161

length, and that the path losses within each encoder, as well as among the encoders, are uniform. A chip time of T, = 400 ps, corresponding to a fiber length of 8 cm, was chosen using the approximate l/e2 width of the transmitter pulse. Delay lines were made using 62.5/125 gradedindex connectorized fiber. The length tolerance was +2 mm, corresponding to a temporal tolerance of k10 ps. Unequal coupler pigtail lengths were “compensated” by the delay lines, which became dedicated to that branch of a coupler pair. Using MM couplers introduced the problem of nonuniFig. 1. Temporal/spatial CDMA encoders, decoders, and network fab- form mode coupling among branches. To counter this, a ric. Optical interconnections in shaded areas depend on the code. power balancing or “equalization” procedure was performed in which losses were induced in the delay-line fibers of branches with lower losses compared to the others by introducing small radius ( 2 mm) s-shape bends. The breadboard was set up for the T/S SPR network following Figs. 1 and 2. In our experiments, only one decoder, which was chosen to be the matched filter of Encoder 1, was necessary. The setup for the temporal CDMA network is well known and so we do not repeat it From here (see [5]). The code sets used to design the encoders Stars as well as the output of Encoder 1 are shown in Fig. 3 for #I+ #2+ both networks. For the temporal code, L = 25 and T,, = #3+ 10 ns. Notice that for L , = 4 columns in the T/S SPR #4+ Encoder Encoder‘Output Mismatched Decoder I code, T,, = 1.6 ns, a factor of approximately six times less Fig. 2. Example of a T/S SPR CDMA encoder implementing an ( M ( 4 , than the temporal bit time. Auto- and cross-correlations were obtained by launch4, 4)) matrix code along with the matched-decoder autocorrelation and the misrnatched-decoder cross-correlation. Boxed area shows an ( M ( 8 , 4, ing the OTDR pulse into Encoders 1-4 individually. Figs. 8)) encoder as a comparison. Delay lines are shown with corresponding 4 and 5 show the decoder outputs for the T/S SPR and chip-time delays. temporal networks, respectively. Note the time scale difference between the temporal and T/S cases. As exdenoted by { L ( N ,w ) J where the length L is a function of pected, no sidelobes exist in the T/S SPR autocorrelathe number of codes in the PO set N and the code weight tion. Also notice that although the cross-correlations in w [6], and i goes from 1 to N. Details of the (0, 1) Figs. 403) and (d) are temporally identical, each pulse sequence code construction can be found in [61, while originates from a different space channel as indicated. construction methods for T/S codes can be found in [SI. After equalization, the maximum measured single-path We have designed a breadboard to implement either loss for the T/S SPR network was 31 dB, and for the {M(4, 4, 4)} T/S SPR or (L(4, 4)) temporal codes in a temporal network the loss was 46 dB. As alluded to noncoherent optical CDMA network. The breadboard earlier, this 15 dB difference is due to two less 1 X 4 enclosure housed Amphenol 62.5/ 125 graded-index MM couplers in the T/S network. fiber couplers: ten 1 X 4 bidirectional trees and four The task of maintaining peak pulse uniformity was time 4 X 4 stars. A n 850 nm laser source, with a FWHM of less consuming and not completely successful as illustrated in than 100 ps, and an APD detector, both part of an Fig. 4(b)-(d) and Fig. 5(b)-(d) where ideally the pulse Opto-Electronics optical time-domain reflectometer peak magnitudes in the cross-correlations should be equal. (OTDR) system, were used as the pulse transmitter and The predominant source of this problem was nonuniform receiver, respectively. The source had a peak pulse power mode coupling, or splitting, of the light in the MM fiber of approximately 200 mW, and the detector was sensitive couplers. If uniformity is not maintained, improper autodown to -27 dBm. The OTDR sampling unit and proces- and cross-correlations would result. Mode scrambling sor improved the response by 24 dB, using sweep averag- would alleviate this problem, but for our simple demoning, compared to the real time detector response. stration we chose the previously described equalization Proper optical implementation and correlation of the procedure. codes impose the requirement that the temporal placeIn summary, we have presented the design and experiment of the pulses must correspond to the sequence and mental results of a temporal/spatial CDMA network that the pulses in all sequences must be of equivalent based on weight four, four user, single pulse-per-row, peak magnitude. This physically translates into ensuring pseudo-orthogonal matrix codes. We compared the netthat the delay lines (we refer to “delay lines” as being work design and experimental results with temporal separate from the coupler pigtails) are of the proper CDMA architecture and codes. We have shown that using

-

IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 4, NO. 10, OCTOBER 1992

1162

40Ops

o

1

2

3

TIME (ns)

4

5 TIME (ns)

Fig. 4. T/S SPR decoder outputs (OTDR detector voltage as a function of time). (a) Autocorrelation, (b)-(d) cross-correlations from Encoders 2-4, respectively. Space channel origin of each pulse is indicated.

a T/S CDh4A network allows for shorter bit times given a set laser pulse width; the T/S network has lower losses compared to a temporal network; and we have indicated that coupler splitting uniformity is critical for proper correlation. Also, using T/S SPR codes resulted in an autocorrelation with no sidelobes. Although we have implemented the two-dimensional codes in a time-space hybrid network, it is conceivable to use these codes in other hybrid schemes, e.g., time-wavelength.

111 S. Tamura, S. Nakano, and K. Okazaki, “Optical code-multiplex transmission by Gold sequences,” J. Lightwurv Technol., vol. LT-3, pp. 121-127, Feb. 1985. [21 P. R. Prucnal, M. A. Santoro, and T. R. Fan, “Spread spectrum fiber-optic local area network using optical processing,” J. Lightwave Technol., vol. LT-4, pp. 547-554, May 1986. J. A. Salehi, “Code division multiple-access techniques in optical fiber networks-part I: Fundamental principles,” ZEEE Trans. Commun., vol. 37, pp. 824-833, Aug. 1989. [41 A. J. Mendez, S. Kuroda, R. Gagliardi, and E. Garmire, “Generalized temporal code division multiple access (CDMA) for optical communications,” SPZE Proc., vol. 1175, pp. 208-217, Sept. 1989. H. S. Hinton, “Photonic switching fabrics,” ZEEE Commun. Mug., vol. 28, no. 4, pp. 71-89, Apr. 1990. F. Khansefid, “Sets of (0, 1) Sequences with application to optical fiber networks,” Ph.D. dissertation, Dep. Elec. Eng.-Systems, Univ. Southern California, (also Communication Sciences Institute report 88-08-02), Aug. 1988. [71 R. M. Gagliardi and A. J. Mendez, “Pulse combining and timespace coding for multiple access with fiber. arrays,” presented at Summer Top. Meet. Opt. Multiple Access Networks OMthl4, Monterey, CA, July 1990. R. M. Gagliardi and H. Taylor, “Code sequences and code matrices for CDMA fiber optic systems,” ZEEE Trans. on Commun., Dec. 1991, (also Communication Sciences Institute report 91-08-01, Dep. Elec. Eng. Systems, Univ. Southern California). [91 J. Y . Hui, “Pattern code modulation and optical decoding-A novel code-division multiplexing technique for multifiber networks,” ZEEE J. Select. Areas Commun., vol. SAC-3, pp. 916-927, Nov. 1985.