softening which would allow the photodiode to move during bonding. A finished assembly is shown in Fig. 2. It is possible at this stage to functionally test the photodiode and to actively check the accuracy of alignment between fibre and photodiode before final assembly into the receiver circuit. Receiver assembly: The photodiode block assembly has been incorporated into a PIN/FET receiver module of the type previously used with conventional top illuminated diodes4 (Fig. 3). This consists of a thick film hybrid circuit plus the photodiode mounted in a solid-sidewall metal dual-in-line package. The fibre exits horizontally from the package and so the mounting block is attached to the gold plating with the metallised face (C) down, using either an epoxy adhesive or a low melting point solder. Wire bond connections are finally made from the block to the hybrid circuit. As the photodiode is now in the vertical plane, this final bonding is done to the gold on the face B' of the step perpendicular to the area to which the wires from the diode were attached. over the main part of the passband, the first step is to determine a nonminimum-phase lowpass prototype network from which various microwave structures, such as, for example, direct-coupled cavity waveguide filters, can be obtained. The prototype network is a folded ladder of admittance inverters and shunt capacitors with additional crosscoupling by further inverters used to realise the finite transmission zeros, see Fig. 1. Fig. 1 Lowpass prototype filter Conclusion: A quartz block containing an optical fibre has been used as a mount for a substrate illuminated photodiode. The method of construction of the block has been described, and also the method by which the diode is aligned and connected to metallised areas on the block, to produce a testable diode-to-fibre assembly. Photodiodes mounted on such blocks have been successfully incorporated into PIN/FET receivers for 1-3" jun optical communications systems. Acknowledgments: Acknowledgment is made to the Director of Research, British Telecom, for permission to publish this letter and to J. Ancell, B. Clark and R. Taylor for their assistance in preparing the blocks. B. M. MACDONALD 11th August 1981 A. G. SAUNDERS British Telecom Research Laboratories Martlesham Heath, Ipswich, Suffolk IP5 7RE, England References 1 SMITH, D . R., CHATTERJEE, A. K., REJMAN, M. A. Z., WAKE, D . , a n d WHITE, B. R.: 'p-i-n/FET hybrid optical receiver for 11-1-6 //m optical communication systems', Electron. Lett., 1980, 16, (19), pp. 750-751 2 LEE, T. p., BURRUS, c. A., DENTAI, A. c , and OGAWA, K.: 'Small area InGaAs/InP p-i-n photodiodes: Fabrication, characteristics and performance of devices in 274 Mb/s and 45 Mb/s lightware receivers at 1-31 /mi wavelength1, ibid., 1980, 16, (4), pp. 155-156 3 LEE, T. p., BURRUS, c. A., and DENTAI, A. G.: 'InGaAs/InP p-i-n photodiodes for lightwave communications at the 0-95-1-65 /an wavelength', IEEE J. Quantum Electron., 1981, QE-17, pp. 232-238 4 Several methods for the construction of the nonminimumphase lowpass functions are available from which the element values of the prototype networks can be determined, but they are not always easy to apply and require considerable computation. Even more important, however, is the fact that filters with several crosscouplings tend to be difficult to tune. Recently,1 a class of lowpass prototype functions with equiripple passband magnitude response has been introduced having only one pair of finite transmission zeros, which are located on the real axis of the complex frequency plane symmetrically about the origin at the distance ±ay. It has been shown that for the values of ox close to unity a very flat delay characteristic over the central part of the passband can be obtained, while the skirt attenuation of the resulting filter is almost identical to that for the all-pole Chebyshev filter of order n — 1. Since only one extra crosscoupling is required to realise one pair of real-axis zeros, the practical tuning of the resulting network is greatly facilitated. The magnitude-squared function of these prototypes has the form SMITH, D . R., HOOPER, R. C , AHMAD, K., JENKINS, D., MABBITT, A. W., and NICKLIN, R.: 'p-/-n/FET hybrid optical receiver for longer wavelength optical communication systems', Electron. Lett., 1980, 16, (2), pp. 69-71 0013-5194/81/220832-02$!.50/0 LEAST-SQUARES MAGNITUDE FILTERS WITH SELF-EQUALISED DELAY CHARACTERISTIC A2(co) = l)]2a>rn-3(co) - 2o\ Tn-2(co) (2) 2{o\ + co2) Uco) = In the above expression, Tn(a>) is the Chebyshev polynomial of the first kind and e2 is a constant which determines the inband loss. The aim of this letter is to demonstrate that further increase in the bandwidth coverage of delay approximation can be obtained if, instead of providing an equiripple passband magnitude response, the prototype function is determined by minimising the ratio of the reflected power to the transmitted power over the normalised passband in terms of a weighted least-mean-squares norm. Since in all cases of practical interest the reflected power must be small, this corresponds very nearly to the minimum of return loss. The weight function associated with the Chebyshev polynomial of the first kind Tn(co) is used, i.e. w(to) = (1 — co2)~' 2, and denoting the characteristic function by fn((o) = Pn(co)/(o)2 + a2), the error integral to be minimised takes the form E= P2{o>) dto (3) (1 - to2)- ' V + a\f This minimisation problem can be solved explicitly. It has been proved in a recent paper2 that the error integral reaches its minimum value if, and only if, the polynomial Pn(co) is orthogonal on the interval [0, 1] with respect to the weight function In the design of narrow bandpassfiltersat microwave frequencies with stringent magnitude and group delay specifications ELECTRONICS LETTERS 29th October 1981 (i) where the characteristic function fn(co) is defined by the rational Chebyshev function Indexing terms: Filters, Microwave filters, Transfer functions A new class of nonminimum-phase transfer functions with one pair of real-axis transmission zeros is described. These functions are suitable for determining the lowpass prototype networks in the design of narrow bandpass filters at microwave frequencies with small passband loss and very flat delay response over the central part of the passband. J Vol. 17 No. 22 = {\ -o)2 w(a>) (I-co2)12 (co2 + a2)2 (4) 833 It has also been proved that the minimising polynomial Pn{a>) can be obtained in the explicit form \ I , 4 jj I l i\ yi /J /c\ V / where Un(co) is the Chebyshev polynomial of the second kind 3 -IT -TT-. (6) -rrr: UW and (7) In this way the prototype characteristic function of the leastsquares magnitude filters is completely determined as -2z?[/ n - 2 (o;) + zt[/n(co) Jn(CO) = K ,2 CO a\ (8) where K is a normalising constant such that/ n (l) = 1. It can easily be verified that substituting Tk(co) for Uk(co) in eqn. 8, and using the recurrence relation Tn(co) = 2a>Tn-^CD)— Tn-2(co), the characteristic function with equiripple passband magnitude response (eqn. 2) is recovered. The magnitude and passband group delay responses of the least-squares lowpass prototype filter and the equiripple solution (eqn. 2) for n = 10 are presented in Fig. 2, while in Table 1 the element values for both filters are given together with some other relevant parameters. For comparison, included in Fig. 2 are also the magnitude and group delay characteristics of the standard all-pole Chebyshev filter for n = 9. The least-squares filter of the present design with one pair of real-axis zeros at a = ± 0-7372 and normalised so as to produce the bandedge attenuation of 0-8 dB at a> = 1 yields a stopband rejection almost equal to that of the equiripple passband filter designed for 0044 dB maximum passband attenuation with one pair of real-axis zeros at c = + 0-8538, but it has considerably smaller inband loss over the largest part of the passband. The bandwidth coverage of delay approximation of the least-squares filter is also considerably improved. The delay bandwidth for the least-squares filter with equiripple delay error of 0-6% is od = 0-51, which is to be compared with the delay bandwidth cod = 0-36 for the filter with equiripple passband magnitude response corresponding to the same percentage delay error. D. D. RAKOVICH 21st September 1981 Faculty of Electrical Engineering University of Belgrade, POB 816 11001 Belgrade, Yugoslavia M. Dj. RADMANOVICH Department of Electronic Engineering University of Nish, Yugoslavia References 1 LEVY, R.: 'Filters with single transmission zeros at real or imaginary frequencies', IEEE Trans., 1976, MTT-24, pp. 172-181 2 RAKOVICH, B. D., and POPOVICH, M. v.: 'Characteristic function of least-mean-square passband filters with finite attenuation poles', ibid., 1980, CAS-27, pp. 1225-1233 3 ABRAMOVITZ, M., and STEGUN, I. A.: 'Handbook of mathematical functions' (Dover Publications, New York, 1965) 0013-5194/81/220833-02$!.50/0 SELF-ALIGNED NORMALLY-OFF GaAs MESFET USING SN-DOPED SiO2 GLASS 0 05 10 20 30 normalised frequency Fig. 2 Comparison of magnitude and group delay characteristics: (LS) least squares with equalised delay for n = 10; {ER) equal ripple with equalised delay for n = 10; (Ch) Chebyshev for n = 9 Table 1 ELEMENT VALUES FOR LOWPASS PROTOTYPE (FIG. 1) c2 c3 c4 c5 ^Z' Least squares ffi = ± 0-7371567 10281706 0-8896558 1-4639700 1-5440387 1-9900473 1-9113172 An improved normally-off GaAs MESFET was fabricated by employing Sn-doped SiO 2 glass, which was used as an Sndiflusant for making the N+ layer. An N+ layer with Rs = 100 fi/D and Ns = 3x 10 13 c m " 2 was successfully obtained under 800°C, 20 min diffusion conditions. A new selfalignment technique, using doped-SiO 2 film, provided a high performance normally-off GaAs FET. Series resistance reduction is an effective way to improve performance of normally-off GaAs MESFETs used as active elements in direct coupled logic circuits.1 A typical ion gate x 1-6806248 1-8158227 21918833 21587845 0 0 0-2085087 0-2566048 drain active layer SI Ga As substrate f f Knt2 Delay error, % Delay bandwidth 834 Equal ripples ff, = ± 0-8538461 Indexing terms: Semiconductor devices and materials, Fieldeffect transistors, Glass 0-8987582 0-6 0-363 0-8241118 doped Si O2 n_ 1 1 diffused 1 1 —r 1 1 region 0-6 0-513 Fig. 1 FET cross-sections a Conventional FET b Presented FET employing doped SiO 2 ELECTRONICS LETTERS 29th October 1981 Vol.17 No. 22

1/--страниц