output isolator losses, the actual saturated output was +17 and +16-7 dBm for HEDFAs 1 and 2, respectively. Both satu- rated output power and noise figure (NF) exhibited a near flat bandwidth response from 1531 nm to 1568 nm. Results and discussion: We performed several transmission tests with an HEDFA 1 stand-alone unit as a booster ampli- fier module in a 622Mbit/s Longspan system. Fig. 4 shows several bit error ratio measurements and our experimental setup arrangement as an inset. On the transmitter side, a low 1530 1540 1550 1560 signal wavelength,nm 1570 1616731 Fig. 3 Dependence of HEDFA performance on input signal wavelength • HEDFA 1 A HEDFA2 -48 -Ul -46 -45 -44 -43 -42 -41 -40 receiver sensitivity,dBm J616/A] Fig. 4 Bit error ratio measurements with HEDFA 1 as power booster • 0 km without HEDFA • 0 km with HEDFA T 304-3 km with HEDFA, P OM = 14-8 dBm • 294-3 km with HEDFA, P out = 110 dBm chirp DFB laser with a centre wavelength of 1549nm and a modulated linewidth of less than 0-1 nm delivered a 0 dBm signal input to the HEDFA module. The HEDFA 1 module provided an output saturated power of +15.1 dBm. Splicing of a 5% monitoring tap coupler at the output optical isolator reduced the available P M output by 0-3 dB to +14-8 dBm. On the receiver side (R), a novel APD-HEMT front end receiver [4] provided a receiver sensitivity of —45 dBm at a BER of 10~ 10 at a 622Mbit/s NRZ modulation with a pseudorandom 2 23 -l format. The system arrangement had an overall power budget (T-R) of +59-8 dBm. In the back-to-back tests, the system yielded the same performance with or without the HEDFA and whether the HEDFA was pumped simulta- neously with both pumps or with either pump alone. These results indicate that there was no detectable sensitivity degra- dation caused by pump leakage orfibreamplifier spontaneous emission. In the fibre experiments, repeaterless data transmis- sion took place over 304 and 296 km with the HEDFA double pumped (980 and 1480 nm) and single-pumped (980 nm), respectively. Although the system suffered a 1-2 dB penalty due tofibredispersion effects, experimental system data reveal no evidence of an error floor at a BER of 10" 14 . Acknowledgments: The authors wish to thank the entire Long- span team, K. Ogawa and B. Owen for their support. 7th August 1992 J.-M. P. Delavaux, C. F. Flores, R. E. Tench, T. C. Pleiss and T. W. Cline (AT&T Bell Laboratories, Solid State Technology Center, 9999 Hamilton Boulevard, Breinigsville, PA 18031-9359, USA) D. J. DiGiovanni and J. Federici {AT&T Bell Laboratories, 600 Mountain Avenue, Murray Hill, NJ 07974-0636, USA) C. R. Giles and H. Presby (AT&T Bell Laboratories, 791 Holmdel- Keyport Road, Holmdel, NJ 07733-0400, USA) J. S. Major and W. J. Gignac (Spectra Diode Laboratories, 80, Rose Orchard Way San Jose, Ca 95134-1356, USA) References 1 M1KKELSEN, B., JOERGENSEN, C. G., VAN DEN BERK, J. A., PEDERSEN, C. F., LARSEN, c. c, DURHUUS, T., and STUBJAER, K. E.: 'High sensitivity laser pumped amplifier-receiver for 2-5Gb/s multichannel trans- mission'. Tech. Dig. ECOC 91, paper, Paris, September 1991 2 AOKI, Y., SAITO, T., FUHAGAI, K., SUNOHARA, Y., ISHIGAWA, S., a n d FUJITA, s.: 'Low noise and high saturation output erbium doped fiber amplifiers pumped at 0-98/im and 1-48/an LDs for long distance optical communications'. Tech. Dig. paper, Paris, Septem- ber 1991 3 LAMING, R. I., TOWNSEND, J. E., PAYNE, D. N., MELI, F., GRASSO, G., a n d TARBOX, E. j.: 'High power erbium-doped amplifiers operating in the saturated regime', Photonics Lett., 1991, 3, (3), pp. 253-254 4 FLORES, c. F., PLEISS, T. c , and DELAVAUX, J.-M. P.: 'Over 300km repeaterless transmission at 622Mbps using the Longspan type fibre amplifer booster module'. AT&T-Bell Technical memoran- dum NEW FUNCTIONAL-LINK BASED EQUALISER W. S. Gan, J. J. Soraghan and T. S. Durrani Indexing terms: Adaptive filters, Neural networks, Equalisers A new nonlinear adaptive functional-link (FL) channel equal- iser is presented. Results are presented for the performance of the FL equaliser compared to a multilayer perceptron (MLP) equaliser and a conventional linear taped-delay line equal- iser. The main attractions in using FL networks over MLP networks for equalisation are its speed of convergence and its integral linear combiner structure. Introduction: Channel equalisation can be treated as a region classification problem [1] using a special class of adaptive neural network called a multilayer perceptron (MLP). However one of the main drawbacks of MLP equalisers is the excessive training period required [2]. In this Letter, a new functional-link (FL) equaliser is presented and a comparative study to an MLP based equaliser is conducted. The per- formance of the new FL equaliser compared with the MLP equaliser using numerical simulation of the channel equal- isation problem is given. Advantages of using the nonlinear structures over the linear structure are also highlighted. New FL equaliser: The new FL equaliser is shown in Fig. 1. It consists of two main parts, the FL expander and the linear combiner with a sigmoidal thresholder. The FL expander per- forms a nonlinear transformation on the input data such that the dimension of the output data is much larger than that of ELECTRONICS LETTERS 13th August 1992 Vol.28 No. 17 1643