IEEE Transactions on Consumer Electronics, Vol. CE-27, No. 4, November 1981 TV GHOST CANCELLING BY LMS-RAT DIGITAL FILTER Mohammad Reza Asharif*, Kazuo Murano**, and Mitsutoshi Hatori* *Department of Electrical Engineering, Faculty of Engineeirng, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113, JAPAN **Transmission Systems Laboratory, Fujitsu Laboratories Ltd. 1015 Kamikodanaka, Nakahara-ku, Kawasaki 211, JAPAN I. INTRODUCTION Recently several methods for automatic TV ghost cancelling are under active research. Generally speaking, these methods are Categorized into two substantial groups: 1) TV ghost cancelling at RF stages including antenna method. 2) Cancellation at video stage. In both of the mentioned above methods, there are many articles[P-33. As we know, there are two parameters for ghost evaluation i.e. the attenuation R and the delay r . The ghost parameters are dependent on TV channel, location of the receiver antenna, and the weather condition which is a function of time and season. Therefore, the ghost parameters are variable and the method of cancellation should be capable of following these variations. Ghost may be produced by many multi-directional reflected signals, in this case, multi-ghosts appear on TV picture. It is difficult to treat multi-ghosts by the method of antenna or the method which is applied at RF stages"2J. Therefore, the method of cancelling at video stage is thought to be more prospective, not only for its ability to process multi-ghosts or to construct a digital filter which is convenient for LSI, but also for its adaptive control nature. Therefore, this research is focused on the utilization of the adaptive digital filters for TV ghost cancelling. There are two types of digital filter: The feedback and the feedforward types. The feedforward type generates the "grand-child" ghost if the filter does not have sufficient number of taps, while the feedback type does not have this kind of ghosts. In order to adjust the taps of the filter, the adaptive control technique is preferable for the simplicity of the control circuits. The Least Mean Square (LMS) algorithm is the original concept for the adaptation of the filter in the present work. The joint optimizations of ghost and white noise is also considered. If one uses the inverse type of filter for TV ghost cancelling the input S/N will decrease that is the noise power will increase. While, as we discuss later, by using the suitable test signal in the adaptive digital filter the joint optimizations of ghost and white noise will be obtained that is the noise power will not increase. Specially, the feedforward path in the digital filter is necessary for noise minimization[4]. While the feedback path is needed to avoid the "grand- child" ghost. In the realization of the adaptive recursive LMS digital filter for TV ghost cancellation the error samples have delay with respect to instantaneous input signal samples therefore the system may be over compensated. Here we have proposed a new method in order to avoid over compensation. This method is based on Random Access Tapping (RAT) control. It will be shown that the RAT method also prevents the system to enter into any pseudo optimum condition. In order to guarantee the stability of the RAT control we define a new function for signature operation in the LMS algorithm. The computer simulation results also confirm the capability of the recursive LMS algorithm and its improvement by RAT method for TV ghost cancellation. The hardware design of the adaptive recursive LMS-RAT digital filter for TV ghost cancellation is described in detail. At the end the experimental results are given. [5, 6, 7, 8] II ADAPTIVE RECURSIVE LMS FILTER Assuming the tapped delay line (TDL) filter with both the feedforward and the feedback paths as is shown in Fig. 1. If A and B are the vertical vectors of non-recursive and recursive tap coefficients, respectively and X, Y are column vector of the input, output signal samples series in the filter then we have: A(n) [aO(n)al (n)a2(n) aNF (n)] (1-a) B(n) = [b (n)b2 (n)b3 (n) . bNB(n) ] (1-b) 0098-3063/81/0588-0604 $00.75 © 1981 IEEE 588 Contributed Paper