Results and discussion: To evaluate the performance of the rrequency selective surfaces, measurements were made in the T E and T M planes of incidence at 22" using a Martin Puplett interferometer with a swept frequency range of 80200GHz. The diplexers were positioned at the waist of a Gaussian beam which produced an edge taper lower than -40dB at 89 GHz. The measurement technique was designed to simulate the normal operation of the FSS in the feed chain. Fig. 2 shows the measured and predicted swept frequency transmission response of the two FSS for T M incidence at 22". Good agreement is demonstrated in both cases. For the waveguide array the measured loss in the transmission band varied between -0.6 and -0.3dB a t 148.5 and 151.5GHz, respectively. Because the effective bandwidth is narrow (- 3%), and the rolloff very steep, a machining accuracy considerably better than 10pm would be required t o achieve a symmetric low loss performance at the band edges. In contrast the transmission band of the double square FSS is broad ( - 1lo/"), and even though the centre frequency was measured to be -154GHz, the loss over the prescribed range of frequencies was less than 0.2dB. The reflection losses for both FSSs were less than 0.2dB, and again computed results showed good agreement with the experimental values. The sensitivity of the passband to the angle of incidence is summarised in Table 1 using computed results at 0, 22 and 45". Both FSSs satisfy our design criteria except at 45' T E where the passband of the waveguide array is shifted below the 148.5-151.5 GHz range. For applications requiring linear polarisation only, the T M orientation is preferable because broader relatively stable passbands are generated. Table 1 shows that the waveguide array has a bandwidth that varies between 3 and 6.5% at 0 and 45" incidence, respectively. For the double square FSS the bandwidth varies between 9 and 12% over the same angular range. For operation in the circular polarisation mode the performance of the FSS is limited by divergence of the transmission bands in the orthogonal planes and/or narrowing in the T E plane. Table I shows for IWO L a example that at 45" incidence the waveguide array is unusable whereas the double square FSS gives a bandwidth of -6%. The rejection band (87.5-91.5GHz) of the waveguide array is unaffected by the incident angle because the individual waveguides comprising the array were designed to he far below cutoff. For the double square FSS the common reflection band is determined by that generated in the T M plane. Our results show however that virtually no degradation occurs over the angular range W 5 " due to the relatively broad band over which the elements reflect the incident energy. Conclusions: The performances of a waveguide and a double square printed FSS have been compared at large incident angles in the millimetric band. It is shown that the passband of the printed element FSS is much broader and relatively insensitive to the incident angle in both the T E and TM planes. The selection of a device for a particular application will additionally be influenced by the required rejection and mechanical characteristics. Acknowledgments: We thank W. J. Hall and R. J. Martin for useful discussions, A. C. de C. Lima for running some of the software, and M. J. Archer for writing one of the computer models used in this study. The printed element FSS was fabricated by the Materials Science Department at the BAe Sowerby Research Centre. This work was supported by a British Aerospace, Space Systems research grant. 3rd March 1992 R Cahill (Systems and Payloads Department, British Aerospace (Space Systems) PLC, FPC 320, PO Box 5, Filton, Bristol BSI2 7QW. United Kingdom) E. A. Parker ( T h e Electronic Engineering Laboratories, T h e Unioersify o f K e n t , Canterbury, Kent C T 2 7 N T , United Kingdom) References R., and PARKER, E. A ' 'Crosspolar levels of ring arrays in reflection at 45 incidence: influence of lattice spacing', Electron. Lett., 1982. 18, pp. 1060-1061 PAKKEK, E. A., LANGLEY, R. J., CAHILL, R., and VAKDAXAGLOU, J. r.: 'Frequency selective surfaces'. ICAP 83, IEE Conf. Publ. 219, 1983, pp. 459-463 C ~ N c. , c : 'Transmission of microwaves through perforated flat plates of finite thickness', IEEE Trans., 1973, MTT-21, pp. 1-6 HAMDY, s. M. A., and PARKER, E. A.:'Current distribution on the elements of a square loop frequency selective surface', Electron. Lett., 1982, 18, pp. 624-626 rAHiLL, - BC 2CC 15C frequency, GHr MULTIMODE OSCILLATION AND M O D E LOCKING OF MAGNETOSTATIC WAVE DELAY LINE OSCILLATOR S. h' Dun,te\ and j'. K Fetisw Indexing terms Magnetostatic waves, Microwave oscillators Multimode oscillation at up to 15 frequencies and a modelocking regime in a magnetostatic backward volume wave delay line oscillator have been realised At 3 53GHz central frequency the mode-locked oscillator generated 1-10 ns long coherent pulses with 53 ns pulse repetition period. . , 1 250 L Fig. 2 Sweptfrequency transmission response berween 80 and 200GHz U Waveguide FSS: TM incident at 22 b Double square FSS: TM incidence at 22 ~ measured A calculated ELECTRONICS LETTERS 9th April 1992 Vol. 28 No. 8 Introduction: Solid state microwave oscillators, containing a magnetostatic wave (MSW) delay line in the feedback loop, have been intensively studied [I, 21. The possibility of electrical tuning over a wide frequency range, low phase noise and planar structure make them very attractive for applications in microwave devices and communication systems [3]. In previous research attention has mainly focused on singlemode oscillators. Also the wideband nature of MSWs allows a multimode regime to he realised [l, 4, 51, which may be used for frequency net oscillation or for coherent radio pulse oscillation. 789 We present a more detailed investigation of multimode oscillation and, for the first time, realisation of active synchronisation of the mode phases in an MSW delay line oscillator. Oscillator design: The oscillator consisted of a GaAsmonolithic transistor amplifier (frequency range f = 3-4 GHz, voltage controlled: gain coefficient G = 18-25dB and output power saturation level P, = 3-5 mW), planar directional coupler (B = -4dB) and wideband magnetostatic backward volume wave (MSBVW) delay line in the feedback loop. The delay line was realised on yttrium-iron-garnet (YIG) film (dimensions 25 x 3 x 0.02mm3, saturation magnetisation 1750Gs, FMR linewidth AH = 0.5Oe) prepared by LPE technology on a nonmagnetic substrate. Two microstrip transducers of 50pm width and 3mm length deposited on the film surface at a distance y = 4mm apart were used to excite and to receive the MSWs. The DC magnetic field of H = 7150e strength created hy an SmCo, permanent magnet was applied tangentially and perpendicularly to the transducers. For harmonic field modulation with a magnitude up to h - 200e and frequency F = 2-30 MHz we used an electromagnetic coil wound on the structure. The MSBVW delay line exhibited an amplitude response bandwidth of Af = 380 MHz at - 30dB level with a minimum insertion loss of -13dB and output signal phase changing up to 32 mad. Experimental results and discussion: For gain coefficient G < 20dB, only a singlemode oscillation of the frequency f = 3535MHz with output power P,, = 2mW existed. For G = 20-23dB, stable multimode oscillation at up to 15 frequencies (see Fig. l a ) was observed. The distance between modes increased from 17 to 22 MHz with an increase of frequency. Total output power reached Po., = 2.4-3 mW. VI I $4 a f r e q u e n c y , GHz Fig. 1 Measured and calculated multimode oscillation spectra a Measured b Calculated The oscillation frequencies may be found from the phasebalance condition: qIcfN) + ( p 2 c f N ) + q3cfN) = 2nN, where 'pl = k ( f ) y , ", q 3 are the phase shifts in the delay line, in the connection lines and in the amplifier, respectively, N = 1, 2 .. . is the mode number, and &(f)is the MSBVW wavenumber. Phase shifts q , ( f ) and q 3 ( f )have been measured and approximated within the 3.3-3.8 GHz frequency range by where E = 2.2, I = 5cm are the dielectric penetration and the length of the connection lines, and c = 3 x 108m/s. The dependence q , ( f )is given by the well known MSBVW dispersion equation. Fig. Ib demonstrates good agreement between the measured and calculated oscillation spectra. Note that nonequidistance of the spectrum is mainly caused by the nonlinearity of the MSBVW delay line phase response. The real distribution of the mode intensities arises from the mode competition process and is determined by the real amplitude and phase responses of the oscillator components. The mode-locking regime in the oscillator described was obtained by harmonic modulation of the field strength with 790 frequency F = 19 MHz. After modulation, the spectrum spread to 330 MHz, the intensities of central modes decreased by 1-2dB and the distance between them became equal to the modulation frequency. With that the oscillator generated shot pulses shown in Fig. 2 a The pulse repetition period was equal to T = 53ns, the pulse lengths were 6t = 7-1011s and mean oscillation power reached P, = 2mW. a b lmnri Fig. 2 Observed oscillator output pulses and calculated pulse envelope Output pulses b Pulse envelope U The mode-locking principle, we believe, is based on the dependence of specific MSW properties on the magnetic field strength. As shown in Reference 6,the harmonic field modulation H = H, + h cos (2nFt) resulted in the MSW phase modulation and gave rise to the appearance of harmonics of frequencies f N f nF ( n = 1, 2 . .. is the mode number) at the delay line output. In the multimode oscillation regime the harmonics arise near each eigenmode. Provided nF is equal to the intermode interval AfN the harmonics act as an external force and carry out the mode-locking. The number of locked modes is determined by the magnitude of the modulation field h. The envelope of the pulses is given by where N , < N < N,, and A: and q; are the amplitudes and the phases of locked modes of the numbers from NI to N,. With that it should be taken into account that external pumping results in the establishment of exactly defined nonzero initial phases qg of the modes under consideration. These phases may be calculated using the equations for amplitude A(/) and phase Q ( f ) responses for oscillator loop: A ( f ) = [I + ezu(f)Y - 2eQ'f)Ycos {k(f)Y}l- a(/) = 2ny AH/v, are the group velocity where vg = 271 and attenuation coefficient of MSW, respectively, y = 2.8 MHz/Oe, the gyromagnetic ratio. Fig. 26 shows the theoretical pulse envelope for measured amplitudes A: and calculated phases q: for five central modes. It can be seen that the MSW phase response nonlinearity explains well the nonsymmetry of the radio pulse envelope. Note, that the central frequency of coherent radio pulses may be tuned within amplifier range by changing the field strength. The pulse length is controlled by adjusting the delay line parameters and modulation field strength. A more wideband nondispersive MSW delay line should be used for shorter-pulse oscillation. Passive mode-locking may be realised by including an MSW signal-to-noise enhancer [3] in the oscillator loop. 5th February 1992 S . N.Dunaev and Y. K. Fetisov (Moscow Institute ofRudio Engineering, Electronics and Automution, Vernadskogo 78, Moscow 117454, Russia) ELECTRONICS LETTERS 9th April 1992 Vol. 28 No. 8 References MILLER, N., and BROWN, 0.:‘Tunable magnetostatic surface wave oscillator’,Electron. Lett., 1976, 12, pp. 209-210 ISHAK, w.: ‘4-20GHz magnetostatic-wave delay-line oscillator’, Electron. Lett., 1983, 19, pp. 930-931 ADAM, I. 0.: ‘Analog signal processing with microwave magnetics’, Proc. I E E E , 1988,76, pp. 159-171 CASTERA, 1. P . : ‘Tunable magnetostatic surface-wave-oscillators’, l E E E Trans., 1978, M A G 1 4 , pp. 826-828 S E T H A m , 1. c., and STIGLITZ, M. R.: ‘Magnetostatic wave oscillator frequencies’,J . Appl. Phys., 1981,52, pp. 2273-2275 FETISOV, Y. K.: ‘Microwave signal processing and forming using dynamic magnetostatic wave devices’. Proc. 20th Microwave European Conf.,Budapest, 1990, Vol. 2, pp. 1142-1 144 decomposition as each subsequent decimation on a filtered image generates a batch of bandpass DCT coefficients. The pyramids of the decimated images are each filtered by a weighting function based on a cosine summation hence it is called the ‘cosine pyramid’. The hierarchy set of error signals which are essentially the remaining DCT coeficients is referred hereafter as the ‘DCT pyramid’ which epitomises the idea of encapsulating DCT within a pyramidal data structure. The image coding technique introduced here results in data compression by encoding the DCT pyramid as illustrated in Fig. 1. Notice that the apex level, the final decimated image, is simply DCT coded. reconstructed reconstructed c o s ~ wpyramid DC T pyramid apex level DCT pyramid cosine pyramid coefficient (/ccefflclent extraction quantisation Padding-- COMPACT IMAGE CODING USING TWO-DIMENSIONAL DCT PYRAMID K. H. Tan a n d M. G h a n b a r i Indexing terms: Image processing, Codes and coding, Signal Drocessino ~ ~ An image coding technique using the discrete cosine transform within a pyramidal data structure is described. A distinct feature is the in built capability to correct blocking effects at low bit rates employing noninteger decimation. I n t r o d u c t i o n : The property of the discrete cosine transform (DCT) is that when it is applied to natural images most of the energy is concentrated at lower frequencies. This property can be exploited by retaining only M coeficients from an original of N . The technique of manifesting DCT coding within a pyramidal data structure to be described here relies on this ability to perform decimation in the transform domain. Decimation of a sample block of size N by a ratio of M/N, where M < N, is achieved by performing an N-point DCT analysis (forward transform) followed by a n M-point DCT synthesis (reverse transform) using the first M coeficients, denoted hereafter by N + M. Obviously both integer and noninteger ratios are possible in just one operation. The synthesised picture elements, pixels, are equivalently derived by subsampling a filtered version of the original sequence at intervals of N/M samples. This filtered sequence is derived by convolving a periodic symmetrically extended original N-point data block with a weighting function that is based on a cosine summation [l]. The spread of the filter function broadens as the decimation factor increases. This is consistent with the requirement for a narrower passband as the sampling rate is reduced t o avoid aliasing effects. DCT decimation inherently provides the corresponding lowpass filtering required according to the decimation rate. This makes it a potentially powerful tool. C o n c e p t : A DCT decimated image can be interpolated, again through the DCT domain, to recover a n image that closely resembles the original. This is achieved by performing an M-point DCT transform on the decimated image followed by padding with zero coeficients from M to N - 1 positions before a n N-point reverse transform. A two-dimensional DCT domain decimation technique can be realised from the onedimensional case by performing an N x M block transform in the horizontal direction first followed by the vertical. Decimation is then obtained by extracting the lower P x Q coeficients from the M x N coellicients for a P x Q reverse transform this time initiating from the vertical direction. The decimation process effectively decomposes an image into two distinct components of differing characteristics. They are a lowpass filtered image reconstructed from the low frequency transform coeficients and a set of highpass DCT coefficients. A pyramidal data structure is then generated by successive decimation that implicitly embodies sub-band ELECTRONlCS LEJTERS 9th April 1992 Vol. 28 No. 8 k32, Fig. I Summary of procedure of using three level DCT pyramid as image coder Two significant advantages of coding the DCT pyramid as opposed to the classical Laplacian pyramid [Z] are identified. First, the error signal elements are in the DCT domain rather than the pixel domain for the Laplacian pyramid. The high compaction efficiency of the two-dimensional DCT would force most of the picture information content into the decimated image, especially when the pixel-to-pixel correlation is high. This in turn reduces the importance of the DCT pyramid and coarser quantisation can be applied to achieve a lower entropy. Secondly, assuming the area of the original image to be A, the total image area of the DCT pyramid is always equal to A, regardless of the rate of decimation or the number of levels in the pyramid. For the Laplacian pyramid, the error area approximates 4/3A, when n , the number of levels in the pyramid, is greater than 4. Clearly, this makes the DCT pyramid attractive for compact coding. A key feature of the coding scheme just outlined is the drastic reduction of blocking effects at low bit rates; a degradation which is inherent in DCT. The mechanism involved is the overlapping effects of the DCT blocks during the decimation and interpolation processes at the various levels of the pyramid. If noninteger decimation with different block sizes and decimation factors at each level of the pyramid is employed, then the interpolated blocks would have different block lengths and the edges of the blocks are nonaligned. Should blocking occur in the reconstructed picture at a level of the pyramid, at the next stage of interpolation, due to the nonalignment of the blocks, the edges are smoothed by the filtering of the basis vectors of the DCT. As a consequence edges appear smoother and block boundaries are removed. Experiment results und f i n d i n g s ; The entropy for each level of the DCT pyramid after quantisation was calculated. The bit rate for the image is the sum of these entropies weighted by their sample density. Fig. 2b shows an image of ‘Trevor’ coded at 0,20bit/pixel using noninteger decimation scheme at three levels of the cosine pyramid. The decimation ratios used for the horizontal and vertical directions were 8 + 4; 11 + 6; 8 + 4 and 8 + 3; 9 + 5 ; 12 + 6, respectively. The image is much less blocky than that coded via the conventional DCT at the same bit rate as shown in Fig. 2u. The selective quantisation of the sub-bands has preserved a significant amount of detail which is clearly evident from the background. The ringing around the edges of the ‘Trevor’ image demonstrates the lowpass filtering effects of DCT decimation as described previously. 791

1/--страниц