You are on page 1of 3

Convolution : If v{mu

..., mr)
N,-l Nr-

= S ... 2
w h e r e m , = 0, 1, 2, . . . , AT, 1, then
2*1-1 2 r-l
z

-xi,

...,mr-xr)

(5)

has e.m.f. E cos coq t and internal impedance Zs, the load has impedance ZR, Zs' and ZR' are impedances equal to the source and load impedances, respectively, and r(t) is the linear modulating (or periodically varying) resistance normally derived from rectifiers controlled by a large-amplitude symmetrical 'carrier' voltage with angular frequency cop.
r(t)

BD(i, ...,r) Bf(uu...,u,)Bl,(ul,...,ur) (6)

Relationships similar to eqn. 6 are valid for crosscorrelations and autocorrelations. Conclusions: Multidimensional b.t. is developed. Its power spectrum possesses shift-invariance and data-compression properties. The frequency composition is made up of all possible combinations of groups of odd-harmonic frequencies in each dimension. A few of the properties, such as Parsevaal's theorem, convolution and correlation, are developed.
N. AHMED R. M. BATES

ECOSCQ)

16th March 1970

Department of Electrical Engineering Kansas State University Manhatten, Kans. 66502, USA
K. R. RAO

Ecos

Department of Electrical Engineering University of Texas Arlington, Texas 76010, USA


References
1 OHNSORG, F. R.: 'Binary FOurier REpresentation'. Spectrum analysis techniques symposium, Honeywell Research Center, Hopkins, Minn., 20th-21st September 1966
2 PRATT, w. K., KANE, J., and ANDREWS, H. c : 'Hadamard transform

r + (t)

Ecos

image coding', Proc. Inst. Elect. Electron. Engrs., 1969, 57, pp. 58-68
3 CROWTHER, w. R., and RADAR, c. M.: 'Efficient coding of vocoder

channel signals using linear transformation', ibid., 1966, 54, pp. 1594-1595 4 AHMED, N., and RAO, K. R.: 'Spectral analysis of linear digital systems using BIFORE'. Proceedings of the 3rd Asilomar conference on circuits and systems, Pacific Grove, California, December 1969
5 OHNSORG, F. R., and NELSON, G. D.: 'Automatic reading of noisy

Fig. 1
a b c d

van der Graaf's equivalent modulator circuits

Series modulator Equivalent to series modulator Shunt modulator Equivalent to shunt modulator

typewritten characters'. Honeywell document 5D-B-5O, Systems and Research Division, Minneapolis, Minn., 1965 6 AHMED, N., and RAO, K. R.: 'Discrete Fourier and Hadamard transforms', Electron. Lett., 1970, 6, pp. 221-224 7 COOLEY, j . w., et ah: 'The fast Fourier transform algorithm and its applications'. Research paper RC-1743, pp. 151-152, IBM Watson Research Center, Yorktown Heights, NY, USA

In the lattice networks, r + (t) and r _ (0 are resistances identical to r(t) except that they are driven in opposite phase to each other as indicated by the subscripts. The equivalences are derived from the identity of the equations 1 ' 2 for the various circuits when expanded on the basis of a Fourier-series representation of r(t); thus
r

( 0 Yi rn cos ncop t
n= 0

0)

VAN DER GRAAF'S EQUIVALENT CIRCUITS FOR SERIES AND SHUNT MODULATORS
Indexing terms: Modulators, Equivalent circuits It is shown that the equivalent circuits used by van der Graaf, while giving perfectly correct results for the problems to which he applied them, lack general validity, even within the conditions assumed by him. In particular they appear to demonstrate that it is possible to obtain a constant input resistance (i.e. constant with respect to time) in a series or shunt modulator, whereas clearly such a result is impossible in reality. An alternative system of equivalent circuits is shown to avoid the difficulty. An explanation is given.

In an oft-quoted and apparently well known, but nevertheless unpublished, report,1 van der Graaf gives a very interesting and useful equivalent circuit for the series or shunt modulator based on a lattice or ring-modulator circuit. Effectively the equivalence is as shown in Fig. 1, where the signal source
238

and the writing out of a separate Kirchhoff voltage or current equilibrium at each frequency coQ + mcop (m an integer between a) and oc, including zero) potentially present in the system. The equivalence also assumes that the wanted output frequency is an odd-order product (i.e. m odd), usually coq + cop or a}q cop. Since in the ring modulator with the special kind of symmetry used here only odd-order products occur in the output loop and only even-order products occur in the input loop, it follows that, if the wanted product were of even order (i.e. m even), then ZR and ZR' would have to be interchanged; i.e. the output would have to be taken from a load connected in the input loop. For any quantity calculable from the set of equations referred to above, e.g. input current at frequency coq, output current at frequency coq + a>p, input impedance seen at terminals 1 and V at frequency coq etc., the same value is obtained whether the original or the equivalent circuit is used. One advantage of the equivalent circuits (indeed, the one for which van der Graaf originally designed them) is that it is very much easier to see the need for, or effect of, special tuning arrangements. For example, it is well known
ELECTRONICS LETTERS 16th April 1970 Vol.6 No. 8

that a zero-loss modulator results from the use of the ring (or lattice) modulator comprised of perfect switches in place of the r(f), and with tuning at the input such that no currents can flow other than that at coq, and tuning at the output so that no voltages can exist across the output terminals except at the wanted frequency coq + cop or coq (op (or any other single odd-order frequency). This is equivalent to saying that no even-order currents are allowed to flow, and no odd-order voltages allowed to exist, other than the wanted one. The circuit equivalent to the series modulator, for example, as in Fig. \b, immediately makes it clear that zero loss can be obtained in the series modulator only if ZS + ZR is an open circuit at all even-order frequencies (other than coq of course) and a short circuit at all odd-order frequencies. To obtain this effect with tuned circuits is impracticable in general, since all orders of frequency may exist. Thus zero-loss series and shunt modulators cannot exist in practice, although very low losses may be obtained by either (i) incorporating a few tuned circuits to give an approximation to the requirements3 or (ii) departing greatly from symmetry in the controlling waveform from which r(t) is derived,4*5 although the switches must then be lossy. It is interesting and important to look at van der Graaf's equivalent circuits from the point of view of modulating resistance functions which are specified not by their Fourierseries expansion in time, but by their resistance/voltage relationship. It appears that the circuits fail to give a proper equivalence in this respect, although the reason for this will be given later. It is known (Reference 2, pp. 131 and 141-153) that, if a lattice modulator has a pure resistance termination RR and a symmetrical carrier voltage, the input impedance of the lattice is a constant pure resistance RR if r(+Ve)r(-Ve) = RR2
(2)

eqn. 3 applies, then there would be no even-order products in the series modulator, since, when the lattice has a constant input resistance, it cannot produce any modulation products in the input loop. This conclusion, unlike that relating to the input impedance of the series modulator, is actually true (Reference 2, pp. 128-131). Corresponding conclusions apply, of course, to the shunt modulator. In the alternative treatment2 of modulator circuits in which the ring or lattice circuit is, with the assumption of a symmetrical carrier wave as before, reduced to a single-loop (or alternatively a single node pair) equivalent circuit as shown in Fig. 2, the issue does not arise. The series (or shunt) modulator is evidently also represented by exactly the same equivalent circuit, where Z = ZS + ZR (or Z s and ZR in parallel for the shunt modulator). However, the circuit now makes no pretence of retaining the 2-port character of the originals, and the terminals 1 and V cannot be identified. The problem of the input impedance of the physical circuit at those terminals therefore cannot be dealt with by this method, so that no incorrect results can be obtained. For teaching purposes, at any rate, this seems the better method to adopt. is i\ (t)

Here r( + Vc) is the value of the modulating resistance when the carrier voltage across it is +VC, andr( Vc) is the value when the carrier voltage has the same magnitude but opposite polarity. Eqn. 2 must be valid over the whole range of Vc. Rectifiers with an exponential characteristic or with a perfectswitch characteristic, meet this requirement, and it is also easy to allow for a constant term in addition (Reference 2, pp. 131 and 141-153). When this condition applies, there can obviously be no modulation products in the input loop, and thus no even-order products are produced at all. It would thus appear, from the circuit of Fig. \b, that, when the relationship corresponding to eqn. 2 applies, namely

r(t)

Fig. 2 Representation of lattice modulator as a single-loop equivalent circuit


a Lattice or ring modulator b Single-loop equivalent circuit Here Z and i take the appropriate values of Zs and is at frequencies co, + unne where in is zero or even; and of ZR and iR at frequencies where in is odd

r(+Vc)r(-Vc) = (Rs+RRy

(3)

where Zs and ZR have been taken as pure resistances Rs and RR, respectively, the series modulator has an input impedance which is a resistance Rln = RS + 2RR (4) Here Rln is defined as the ratio of voltage to current at any instant, and is not dependent on the waveform or frequency applied. This input resistance as given by eqn. 4 does not vary with time, i.e. with the instantaneous value of the carrier voltage. But this conclusion is obviously absurd, since, from Fig. \a, it is quite clear that the input resistance is high when r(t) is high and low when r(t) is low. So the equivalent circuit fails in this respect. It is interesting and significant that the 'narrowband' or 'time-smoothed' input impedance of the series modulator, as seen by a sinusoidal wave at frequency coq, i.e. as defined by the ratio of the voltage component at coq to the current component at coq, can be shown without using the equivalent circuit to be
(5)

The explanation of the difficulty with the van der Graaf equivalent circuit does, in fact, lie in its apparent retention of the 2-port configuration and in its repetition of Z s in the output and ZR in the input (n.b. the notation of Zs' and ZR which distinguishes the added impedances from the originals is due to the present authors and was not used by van der Graaf). In fact Zs' and ZR are not added impedances, but are merely components of Zs and ZR transferred to a different position, as explained below. It follows then that the apparent 2-port equivalence of the circuits of Figs. \b and d with those of Figs, la and c is quite false, and the terminals 1 and 1' of the equivalents do not correspond in all respects to those of the original circuits. In Figs. \b and d, only odd-order modulation-product currents flow through Zs' and ZR, and only even-order currents flow through Zs and ZR'. If then we distinguish Zs and ZR in the equivalent circuits from those of the series and shunt circuits by calling the former Zs" and ZR", we have ZS = ZS'+ZS" and
Z = Z ' -I- Z "
(*1}

(6)

i.e. the same as the false input resistance in eqn. 4. Note that it is still assumed that eqn. 3 applies. (This calculation is based on the work of p. 129 of Reference 2, but would take too much explanation to justify including here.) From the equivalent circuit it would be concluded that, if
ELECTRONICS LETTERS 16th April 1970 Vol.6 No. 8

where Zs' and ZR" are functions which have the impedance values of Z s and ZR at all odd-order frequencies but are zero at all even-order frequencies, and Zs" and ZR are functions which have the impedance values of Zs and ZR at all evenorder frequencies but are zero at all odd-order frequencies 239

It is now quite clear why the equivalent circuits cannot be used for calculating input impedances except at a single frequency.
D. G. TUCKER

16th March 1970

Department of Electronic & Electrical Engineering University of Birmingham Birmingham 15, England
D. P. HOWSON

Postgraduate School of Studies in Electrical & Electronic Engineering University of Bradford Bradford 7, England References
1 VAN DER GRAAF, j . j . : 'Modulators'. Report 21 CL of Netherlands PTT Research Laboratory, 1950 2 TUCKER, D. c : 'Circuits with periodically-varying parameters' (Macdonald, 1964), pp. 70-73 3 HOWSON, D. P.: 'Rectifier modulator and switched harmonic generators using a single idler circuit'. Proceedings of the conference on electrical networks, Newcastle, Sept. 1966 4 PETERSON, E., and HUSSEY, L. w.: 'Equivalent modulator circuits', Bell Syst. Tech. J., 1939, 18, p. 32 5 HOWSON, D. p.: 'Low loss shunt on series rectifier modulators', Electron. Lett., 1966, 2, pp. 225-226

SUBHARMONICALLY INJECTED PHASELOCKED IMPATT-OSCILLATOR EXPERIMENTS*


Indexing terms: Microwave Transit-time devices oscillators, Avalanche diodes, Subharmonically injected phase locking of continuously operated IMPATT oscillators has been demonstrated with frequency ratios as high as 9 : 1 . The measured lockingrange/locking-gain slope for the 2 : 1 case is about 10 dB/ decade of locking range and that for the 9 : 1 case is about 5 dB/decade of locking range.

Recently, a number of publications have appeared on the injection-locking properties of IMPATT oscillators for pulsed1 ~ 3
* The work was supported by the USAF Systems Command, Rome Air Development Center, Griffiss Air Force Base, NY, USA

as well as continuously operated4"6 modes of operation. All results described were for a 1 : 1 ratio of the oscillator frequency to the locking-signal frequency. For Gunn-effect oscillators, Oltman and Nonnemaker7 have reported successful locking with subharmonically injected signals. This letter reports similar subharmonic-locking studies which demonstrate that continuously operated IMPATT diodes can be subharmonically phase locked with ratios as high as 9 : 1. Fig. 1 is a schematic of the experimental setup used in the subharmonic-locking studies. The injected locking signal is supplied to the X band iMPATT-diode oscillator via a lowpass coaxial bias circuit which cuts off at X band frequencies. In some of the experiments performed, tuning stubs were used to match the IMP Air-oscillator bias circuit to the locking source. The iMPATT-oscillator cavity used in the study had a loaded Q factor of approximately 600. The IMPATT diodes used were of the diffused epitaxial p+-n-n+ silicon avalanche diodes developed in our laboratory by C. A. Lee. The epilayer resistivity of the diode was 1-lficm, the drift-zone thickness was 3-8//m and the area of the diode was approximately 10" 4 cm 2 . These diodes are capable of 150 mW output power at X band with an efficiency of slightly over 3 %. Fig. 2 shows a series of oscillograms showing the snap-in process of subharmonic phase locking of IMPATT oscillators as the locking power is increased.* The vertical display was logarithmic and the horizontal scale was 1 MHz/division. Oscillogram a is the spectrum of the free-running IMPATT oscillator at an output power of 100 mW and an operating frequency / 0 of 8495 MHz. The locking-source frequency fL was fixed, so that throughout the experiment 2fL = f0 + 3MHz. Oscillograms b, c and d show the measured output spectrum of the unlocked IMPATT oscillator with locking powers of 27, 23 and 21 dB, respectively, below the oscillator power. The signal shown on the extreme right of the series starting with trace b is 2fL, the second harmonic of the locking signal produced by nonlinear mixing within the IMPATT oscillator. It is seen that, as the locking power is increased, the oscillator frequency is drawn closer and closer toward the 2fL line, and additional sidebands appear on the side of the oscillator opposite to the 2fL line. The IMPATT oscillator is finally locked to the 2fL line, as shown in trace e, with a locking power 20-3 dB below the oscillator power. The asymmetric feature of the output spectrum of unlocked driven oscillators has been observed8* 9 and explained9 for frequency ratios close to 1 : 1. The fact that the results for
* These oscillograms were taken on a Tektronix type 585A oscilloscope with a type 1L30 spectrum-analyser plug-in unit

IMPATT oscillator coaxial input

slide-screw tuner .waveguide output d.c. power supply directional couplers matched load

6GHz lowpass filter

bias-insertion unit

power meter

power meter directional couplers power meter adjustable attenuator 6GHz lowpass filter locking source

oscilloscope and spectrum analyser

Oscilloscope used: Tektronix type 585A Spectrum analyser used: Tektronix type IL30

Fig. 1 Block diagram of equipment used for subharmonic locking of IMP A TT oscillator ELECTRONICS LETTERS 16th April 1970 Vol.6 No. 8

240

You might also like