Results: The algorithm described above was applied to test sequences containing a significant amount of undesirable global motion. Interlaced-to-progressiveconversion was implemented by dropping every other field and vertically upsampling the retained fields by 2:l to restore ITUR Rec. 601 spatial resolution. No antialiasing pre-filtering was applied in this case. Blocks of 64 x 64 pixels were used yielding a field of 11 x 9 = 99 motion vectors and a total of 10 x 9 + 11 x 8 = 178 block pairs to be processed. The accuracy of estimation was 11100 of a degree for the angle of rotation, 1i1000 for the magnitude of scale changes and 114 pixel for translation. Undesirable global motion in the test sequences under consideration is due to vibrations resulting from the use of a portable camera. Fig. 3 shows the luminance component of the first frame of test sequence ‘rescue’. In the middle of the Figure, the difference between the first and the second frame shows evidence of significant amounts of global motion. At the bottom of the Figure, the result of motion compensating the second frame with respect to the first, according to the estimated global motion, is shown. The estimated parameters for this pair of frames were 4 = - 0.86 degrees, a = 1.003, zx = 8.000 pixels and z, = -5.5 pixels. Using the PSNR as a measure of frame difference, this was measured at 19.25dB for the original pair and at 30.85dB after motion compensation. To prevent edge effects affecting the PSNR measurements, only the centred 640 x 500 pixels were used in this computation. The residual frame difference after motion compensation is due to scene motion as well as global motion containing components other than pure rotation, change of scale and translation. Conclusion: A simple algorithm for the estimation of global motion parameters from sparse translational vector fields was described. An important feature of the proposed technique is its low complexity which makes it suitable for real-time implementations. Another key advantage is compatibility with motion estimation tools that already exist within internationally standardised video compression algorithms employing motion-compensated prediction. Our results show that accurate estimates of global motion parameters can be obtained from sequences which contain complex scene motion as well as global motion due to hand-held camera vibrations. In this case, global motion compensation using the computed estimates offers significant benefits in both objective and subjective quality owing to the drastic reduction of camera unsteadiness. Finally, the proposed technique will provide reliable estimates only when the majority of raw motion measurements correspond to areas of the picture whose depth map is relatively smooth (such as non-complex backgrounds). This condition is approximately met by many typical scenes. Fully differential current operational amplifier Sibum Jun and Dae Mann Kim A fully differential, wide bandwidth and good CMRR current operational amplifier is presented. The proposed circuit employs a tapped cascode current mirror as an input stage and simple common source stages as an outpnt stage. To enhance the common-mode rejection ratio (CMRR), a hgh impedence node is implemented by two cross-coupled simple current mirrors. Introduction: Current-mode operational amplifier (COA) has advantages over its voltage-mode counterpart, such as higher operating frequency and dynamic range, etc. Various designs were reported in literature [l - 51; among them, the current steering approach [1] is novel and shows a superior performance, but was based on single-ended input operation. The advantages of the fully differential designs over single-ended versions are obvious. In this Letter, we propose a very simple, fully differential design of a COA having good common-mode rejection ratio (CMRR), wide bandwidth and low power consumption. “dd Fig. 1 Proposed circuit Proposed circuit: The proposed circuit shown in Fig. 1 is organised and Mp21Mp4), two as two tapped cascode current mirror (Mp,lMp3 and two simcross-coupled simple mirrors (Mn,/Mdand Mn3iMn4), ple common-source stages (Mnsand MnJ. The input signals, Ap and An,are injected into the tapped internal nodes of the cascode current mirrors. Since the tapped cascode current mirrors are basically common-gate amplifiers with constant bias current I,,,,,the injected signals are delivered, nearly intact, to the following crosscoupled simple mirrors. The input resistance r,,, and output resistance r,,, of the input stage are: n 0 IEE 1998 8 October 1997 Electronics Letters Online No: 19980064 T. Vlachos (BBC Research and Development, Kingswood Warren, Tadworth, Surrey KT20 6NP, United Kingdom) The author is currently with Centre for Vision, Speech and Signal Processing, University of Surrey, Guildford, Surrey GU2 5XH, United Kingdom References JOTAWA, H., KAMIKURA, K., SAGATA, A., KOTERA, H , and WATANABE, H.: ‘Two-stage motion compensation using adaptive global MC and local affine MC’, IEEE Trans. Circuits Syst. Video Technol., 1997, 7, (l), pp. 75-85 VLACHOS, T., and THOMAS, G.: ‘Motion estimation for the correction of twin-lens telecine flicker’. Proc. IEEE-ICIP, 1996, Vol. 1, pp. 109-1 12 UOMORI, K., MORIMURA, A., and ISHII, H.: ‘Electronic image stabilisation system for video cameras and VCRS’, S M P T E J., 1992, pp. 6 6 7 5 HOETTER, M.: ‘Differential estimation of the global motion parameters zoom and pan’, Signal Process., 1989, 16, pp. 249-265 wu, s.F., and KITTLER, J.: ‘A differential method for simultaneous estimation of rotation, change of scale and translation’, Signal Process.: Image Commun., 1990, 2, pp. 69-80 Rec. H.262, ‘Generic coding of moving pictures and associated audio: Video’. MPEG-2 Video Compression Specification, ISO/ IEC 13818-2, 1994 THOMAS, G.A.: ‘Television motion measurement for DATV and other applications’. BBC Res. Dept. Rep., No. 1987/11 62 L r,, (1) ~ Smp + + rout 2[rop r o p ( l gmprop)] (2) where gmpand rOpare the small-signal transconductance and drain resistance of PMOS transistors M,, to Mp4,respectively. The cross-coupled smple mirrors are adopted for increasing the CMRR. This circuit provides a high incremental resistance (rh) for the differential-mode signals, while having a small resistance (rc,) for the common-mode ones. Assuming matched identical devices, r,, and r, can be shown to be equal to (3) rdm 21 T o n r,, I 2: (4) ~ Smn where g,, and v,, are the small-signal transconductance and drain resistance of NMOS transistors M,, to M,,, respectively. Hence, the injected differential input current leads to a large incremental voltage swing, while the common-mode input current has a much smaller amplitude compared to the differential one. It should be mentioned here that the proposed COA operates very similarly to the steered COA [I], at least in a differential manner, where the current mirrors are excluded intentionally to achieve a higher speed. Since the cross-coupled simple mirrors cancel each other’s transconductance values, no current mirroring operation exists in the differential-mode signal path. Current-mirroring is valid only on the common-mode signal path. If the normal current source ELECTRONICS LETTERS 8th January 1998 Vol. 34 No. I which is not cross-coupled is employed, r, has a similar order of magnitude to rd,. This leads to a poor CMRR, unless an additional common-mode feedback loop is introduced. ydm can be further increased by slightly oversizing M,, and M,,, compared to M,, and Mn4.However, it depends on process tolerance, which is difficult to predict and is therefore excluded in this Letter. Finally, the voltages on the high impedence nodes are converted again to the current by the small-signal transconductance (g") of Mn5and Mn6.The overall differential current gain AIdmand the common-mode current gain A,, of the proposed circuit are: &dm pensation capacitors. The CMRR obtained is also shown in Fig. 2. As mentioned, the CMRR has a magnitude nearly equal to the differential-modeopen loop gain. A further enhanced CMRR will be obtained if an additional common-mode feedback control is introduced. The proposed COA was simulated in a closed loop unity gain buffer. Fig. 4 shows the transient response to a step input of flOpA. The 1% settling time is found to be 7.lns. The input and output resistances of the COA are 21 and 597kQ respectively. Finally the static power dissipation was < 860pW. (5) gkn(rdm//rout) Acdm gknrcm (6) Since the common-mode gain is about one if gmm-- gmn, the CMRR of the proposed COA is nearly equal to A,,,. The dominant pole, f , is determined by the high impedence node and equal to: where C, is the half of the total capacitance connected to each high impedence node, including compensation capacitance if it exists. Hence, the gain-bandwidth product (GBW) of the proposed COA becomes g',,/(2:nCP). Simulation: A SPICE simulation has been performed using the ~ level 28 model parameters provided for LG Semicon 0 . 6 n-well CMOS technology with V,, = 0.771, vihp= 4.868, k, = 1 3 1 . 1 w Vz and k, = 41.8pNV2. The supply voltage V, is 3V and the bias is 40pA. All the devices used have the same length of current IbLas 1p.The widths are: l o p for M,, to Mn4;2 0 p for Mn5to Mn8; 30pm for M,, to Mp8.PMOS transistors Mpsand Mp6are added to bias Mns and Mn6,respectively. NMOS transistors Mns and Mn6 are auxiliary output devices. All body terminals of PMOS and NMOS transistors are connected to V, and ground, respectively. 20 0 40 ns 60 80 100 Fig. 4 Simulated transient response Conclusions: A fully differential, wide bandwidth and good CMRR current operational amplifier has been presented. The proposed circuit employs a tapped cascode current mirror as an input stage and two cross-coupled simple current mirrors to enhance CMRR. A SPICE simulation shows a 53dB differential-mode gain, a 52dB CMRR, a 300MHz unity gain bandwidth, and a 50" phase margin without compensation. 3 November 1997 0 IEE 1998 Electronics Letters Online No: 19980106 Sibum Jun and Dae Mann Kim (Department of Electrical Engineering, Pohang University of Science and Technology, San 31 Hyoja-dong, Nam-gu, Pohang, 790-784, Republic of Korea) References .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 1 2 ......... 0 -1 10 3 1 10 10 2 10 3 4 10 MHz 5 Fig. 2 Simulated dijjfej*ential-modeopen loop gain and C M R R and EL-MASRY, E.I.: 'A 200MHz steered current operational amplifier in 1 . 2 CMOS ~ technology', ZEEE J. SolidState Circuits, 1997, 32, (2), pp. 245-249 ABOU-ALLAM, E , and EL-MASRY, E.I.: 'High CMRR CMOS current operational amplifier', Electron. Lett., 1994, 30, pp. 1042-1043 ZELE, R.H., LEE, s., and ALLSTOT, D.J.: 'A high gain current-mode operational amplifier'. Proc. IEEE ISCAS, 1992, pp. 2852-2855 KAULBERG, T.: 'A CMOS current-mode operational amplifier', IEEE J. Solid-state Circuits, 1993, 28, pp. 849-852 TOUMAZOU, c., LIDGEY, F.J., and JAIGH, D.J.: 'Analogue IC design: The current-mode approach' (Peregrinus, London, 1990) ABOU-ALLAM, E., gain CMRR __-0 Design of turbo-code interleaver using Hungarian method 3 0) A.K. Khandani ...,..., . -90 A method is presented for optimising the structure of the turbocode interleaver using the Hungarian method ginear sum assignment problem). N u m e n d results are presented which show a substantial improvement with respect to a random interleaver. .*. ... .. .. ......... . . I ., -180 10 ' . n 10- 7 IO' 10- MHz 10- m Fig. 3 Simulated phase characterisics The frequency characteristics of the proposed COA are shown in Figs. 2 and 3. It is seen that the proposed COA exhibits an open loop gain of 53dB. The 3dB bandwidth of the COA is seen to be 1MHz. Also, the unity gain bandwidth is found to be 300MHz, at which the phase margin is 50" without inserting com- €LE CTRONKS LETTERS 8th January 7998 Vol. 34 Introduction: Consider a turbo code in which the underlying recursive convolutional code (RCC) is generated by the transfer function G(d) = N(d)/D(d).We know that the impulse response of G(d) is periodic with period p 5 2 - 1, where r is the constraint length of the code [l]. If p = 2 - 1, the resulting impulse response is called a maximum length sequence (MLS). This is the case for the transfer functions used in turbo codes. If we look at the impulse response of G(d) as a periodic sequence, we obtain K = 2 - 1 non-zero sequences which are time shifts of each other. We refer to these sequences as different No. 1 63

1/--страниц