LINEAR PHKDICTION IN TKLKVISION 



BLOCK DIAGRAM OF EXPERIMENTAL SETUP 



77.-) 



POLARITY 

 SELECTORS 



Fig. 12 — Block diagram of experimental apparatus used to investigate methods 

 of prediction involving combinations of previous signal values along a lino with 

 those on the line directly above. 



frequency of 54.0 mc. The over-all video bandwidth is flat (±0.1 db) to 

 5.0 mc. Nonlinear distortion is approximately one percent when the 

 peak-to-peak signal to r.m.s. noise is 58 db. To give an idea of its com- 

 plexity, two such units with their associated power supplies require a 

 seven-foot relay rack for housing. The signal at Ji represents the input 

 picture signal, even though it may be delayed by a small fi-action of a 

 Nyquist interval from the actual input signal. The signal at J^ is the 

 same signal as found at Ji but delayed by one line time. Each of these 

 signals is fed into terminated delay lines to enable additional signal 

 samples to be obtained. The geometrical location of these signal sam- 

 ples is illustrated in the upper right section of Fig. 12. Here, six signals 

 are obtained instead of three as were rec^uired for "previous value" and 

 "slope" prechction. These signals are weighted and polarized in the same 

 manner as the three signals shown in Fig. 11. The output is the sum of 

 these weighted signals and is given by 



C = Cfo,0*JO,0 ~l~ Ql,0<Si,o + 02,0*02,0 + ffo, 1**^0,1 + Cfl,10i,i + CLo.l^i.X . 



The coefficients may assume positive or negative values. 



MEASUREMENTS 



It is ob\'ious that if we are aljle to predict the value of most signal 

 samples closely (which we will be able to do if there is a large amount of 

 correlation in the picture), then the average amplitude of our mistakes 



