This page was kindly provided by Andrew Fremont.
1.0 INTRODUCTION
This section is intended to cover the more technical aspects that must be understood when designing a telecine system. It relates to the work carried out at Marconi from 1967 to 1985. Some of the opening paragraphs may be well known to some and could be skipped.
2.0 BASIC COLOUR REPRODUCTION
Before diving into the subject of film reproduction, it seems sensible to talk a little about the 
optimisation of studio cameras. Most people with a TV background will know that standard colour TV systems attempt to reproduce all possible colours using a display that emits only red, green and blue light in various proportions. The light is mixed additively whereas film attempts the same task using yellow, cyan and magenta dyes in a subtractive system; two very different things! Figure 1 shows the relative spectral sensitivity of the cone receptors of the human eye. Immediately we can see that the red and green sensitivities have a significant overlap. It is may be surprising that we can differentiate colours in the yellow / orange range. If we look at the CRT phosphor spectrum shown in figure 2 
we can see that switching on the green gun is not only going to stimulate the green cones but also will stimulate the red ones as well. The blue and red phosphors do a somewhat better job of exciting the blue and red cones respectively. If we now consider the task of reproducing a spectrally pure cyan, say at 500nm, we would drive the blue and green guns. Unfortunately the eye will see a considerable amount of red in the result, giving a desaturated reproduction. If we look at the CIE Color Matching Functions in figure 3 we can see what the camera spectral sensitivity would ideally be for best reproduction. The negative lobe of the red characteristic can be created by taking a little of the blue and green signals from the red signal using a correction matrix. 
Of course this is fine provided that there is some red signal to subtract from. Hence cyan is always a bit of a challenge. In fact you end up adjusting all three colours. The correction values are calculated using the actual camera spectral response (colour splitter block, tubes and any filters) and a set of defined colours. The matrix is optimised to give the least errors overall. (Do you think the human neural system has an in-built colour correction matrix? We know it has auto-colour balance). The Mark VII camera control unit had a set of video In/Outs where a matrix could be fitted as noted in the B3404 article but the studio camera had no matrix until the Mark VIII came along.
3.0 FILM
3.1 Colour Film Basics
With almost all home photography using digital techniques there will be a decreasing number of people who have actually held a piece of film and even less that understand the processes involved in producing a film image. It therefore seems useful to give a brief description of a colour film and the processes. Figure 4 shows a simplified cross section of a film.
The top layer responds to a fairly broad spectral band of blue light. The next layer is a yellow filter that cuts all blue sensitivity from the layers below. Next is the green sensitive layer followed by the red sensitive layer. Modern material is more complex but what has just been described is the film that was available in period that is considered here. What happens when the film is processed will depend upon whether the material is negative, a print film or a reversal film (as used for home slides). The path for negative and print film is much the same because both reverse the tone of the exposing light. A Colour Development, using special chemicals, is performed first. As each of the three layers develop a negative silver image is formed (as for black and white film) but a negative (negative relative to the exposing light) colour image is also formed. This gives (top to bottom) a yellow, magenta and cyan image whose density is proportional to the density of the silver image produced at each layer. The silver is then bleached away leaving just the dye image.
For reversal material a black and white development is applied first, resulting in a just a black and white negative image at each layer. The film is then exposed to a flat white light so that any undeveloped silver halide crystals are now sensitised. A colour development is then undertaken followed by bleaching as before. The strength of the dye images produced will be proportional to the number of crystals that developed in the second development. For a bright subject area most crystals will have been developed in the first process, leaving relatively few for the second process. Thus a well exposed area will result in a light image i.e. a positive image.
Some may be familiar with the orange colour of negatives; this will be discussed elsewhere. There are in fact a number of different film types that are made for specific purposes. For example a master negative from a major production will be printed to a number of Inter-positive sub-masters for safety apart from anything else. These might then be printed onto a number of Inter-negatives before the final release prints are made. News film would almost always be reversal material.
We should note that Technicolor printing is not a photographic process. These prints are produced in what could be described as a mini printing press with the printing rollers replaced by three films to carry the yellow, magenta and cyan inks.
3.2 Film Density Characteristic Details
Having noted how film images are produced we now can look at the actual transfer characteristics. The film industry always plots film characteristics on a log / log basis. Actually it will be log exposure against density but density is given by D = log10(1/transmission).
Figure 5 shows a plot for a negative colour film. The three lines give the density as measured in red, green and blue using filters with a narrow spectral bandwidth. The three curves do not overlay due to the orange mask colour mentioned above. You will note that the curves for this particular film are substantially straight over a log exposure range of about 2.5; this corresponds to an exposure range of about 300:1. 
The straight line tells us that the gamma is constant and in this case it is about 0.65 (actually – 0.65 in TV terms); the long linear portion means that we stand a better chance of accurate correction. Figure 6 shows reversal film. We can see that whereas the negative gave density ranges of about 2 (100:1) the range of the reversal material is about 3, i.e. 1000:1. In the late 60s and 70s we were lucky to be able to display more than about 60:1 with a shadow mask tube; hence a lot of black stretch was needed to see a reasonable amount of black detail. Here there is little of the characteristic that is all that linear, indicating a varying gamma. The almost linear part in the central range of the curve shows a gamma of about 1.8. In general reversal material has more contrast than a release print. If we look again at figure 6 we can see that the shape of the three curves is much the same but the reversal film curves are noticeably different in the high density area. This is because the maximum achievable density for the individual film dyes is different. This lack of black tracking can be taken out by the telecine black balance control. The foregoing description was may be a little brief but it is important to know this sort of data when you are deciding what gamma ranges to design in.
3.3 Reading Film
3.3.1 What are we trying to do?
The telecine designer’s task is to build a machine that achieves the best reproduction of a given film.
One question is: are we trying to reproduce the projected film image or the original scene or may be something else? So we find ourselves with both artistic and technical points to consider. A first attempt to cover the issues that follow here was made during a TV symposium in Sydney in the late ‘60s. It caused a considerable stir amongst the delegates. “You cannot do that, you are interfering with what the director intended” some said. The fact is that if you do not do something to modify the basic image that the telecine sees it would not be not make good television, particularly if the film has faded. In fact we at Marconi basically did not attempt to reproduce the projected image and went towards getting back to the initial scene. This was important for drama inserts anyway.
3.3.2 Achieving the Best Reading
It is often impossible to know how the film passing through the telecine was made and it is certainly impossible to attempt to correct for all the processes if you do not know (a personal opinion!). Moreover different film manufactures employed their unique chemical ‘tricks’ to make their film the best. This sort of thing did not help the task of matching film to studio for drama inserts.
If we look at a typical set of film spectral sensitivities (figure 7) we will see that they are nowhere near the CIE curves of figure 3; this is a potential problem in itself. Should we put in a linear matrix to make them a better match to the CIE requirements? We decided that the answer is no, it is not that simple. This is mainly because we do not have access to the original exposure data (the equivalent of the camera signal in the studio case). What we do have is the result of an ‘analysis’ of the yellow / cyan / magenta dye images. Figure 8 shows a typical set of dye spectral curves.
The curves are shown as Density (defined above). Ideally the dyes would have a block profile with, say, the cyan dye stopping red light only and passing all green and blue light; clearly we are somewhat distant from that position. The total density at a given wavelength is the sum of all three densities. As we may remember from the distant past, adding log data (density is a log function as already noted) is the same a multiplying. Hence any individual spectral reading gives a product related to the initial red, green and blue exposures (and other factors if it is not the original film). Getting a bit difficult? It gets worse! The readings obtained will include a gamma factor that varies as you move along the curves. Say you read the magenta dye on its peak at 540nm, you might read with a gamma of 1.5 (i.e. 1.5*log(k/exposure)). If you read the same curve at about 580nm (where the density is half the peak) the reading will have a gamma of 0.75. We can see that the 540nm reading also included a cyan dye reading at about 0.25 gamma plus a yellow reading at around 0.35 gamma (I said it got worse). As a result of this the RGB signals will represent a summation of a complex range of YCM dye transmissions at different gammas, further modified by each colour channel’s spectral sensitivity. Luckily the relative shape of the density curves does not change with the amount of dye present; e.g. the relationship of densities at 540 and 580nm for the magenta dye will always be 2:1 irrespective of the actual value of the peak density. Naturally the complex nature of the camera signal makes it extremely difficult to know what the initial film RGB exposures were. Obviously we need a way of making life a little simpler. An important step along this road is to make the spectral bandwidth of the RGB sensors as narrow as possible, consistent with getting a reasonable amount of light through, say 20nm at half amplitude. The peaks of the spectral lobes are centred as closely as possible to the peak dye densities. In this way we can get as near as possible to reading the initial film RGB exposures. Unfortunately lack of camera tube red sensitivity forced us to use a peak at about 615nm for the red channel, which was certainly not ideal and the level of ‘crosstalk’ from the G and B exposures was increased. This meant that the main red gamma was less than that for the other two channels. There is still more work to do on the signals but the accuracy of what we can eventually achieve is improved by the narrow band philosophy. However, on the down side, the narrow band approach does take away a lot of sensitivity. While the signals we obtain still have a complex mix of data from all three exposures, the gamma of the components does not vary so much over the narrow spectral ranges. Another downside of the narrow lobe method is that the gamma read from the film is going to be higher and more black stretch (and increased noise as a result) will be needed. By now you will no doubt have realised that reading film accurately can be difficult but just one more complication should be noted before moving on. The film manufacturers ‘tricks’ were mentioned earlier on; it is possible to design a film where the development of one layer has an effect on a neighbouring layer. This is called Inter Image Effect. Agfa film of the day was an example where this was used to provide a very strong red. May be it looked great for direct projection but in the context of telecine reproduction the effect had to be more or less removed to get a match to studio images. The Inter Image Effect is a little like having an additional set of dye curves added to the mix. So to sum up what has just been covered, the aim is to get to a situation where we can almost write:
VR = (red exposure) γrc * (green exposure) γrm * (blue exposure) γry
Where γrc is the gamma of the red reading of the cyan dye, γrm is the gamma of the red reading of the magenta dye etc.
There are similar equations for green and blue. Obviously we do not really have single values for each gamma but our narrow lobes help us to approach that condition.
Now having got to this situation we still have the problem that the ‘cross-talk’ signals (green and blue for the red signal) multiply the main signal. If we pass the signal through a logarithmic converter the signals are then effectively added. We then pass the three colour signals to a nine-term linear matrix to remove the cross-talk factors. This done we can then convert the signal back to normal via an exponential conversion process. By adjusting the terms of the matrix we can perform the gamma correction process at the same time. Arriving at a good set of data for the matrix was not that easy but an initial start point was obtained by taking dye reading at the appropriate reading wavelength and then calculating an inverse matrix from this. The resultant correction was usually to strong and it was reduced. For the B3410 settings were calculated for Eastman Color dyes (with a negative and positive film values) and Technicolor inks. There was also a fully variable set of controls. In addition there was an ‘Amount’ control. Special settings were also supplied to some customers where their stock film was not Eastman Color based.
3.3.3 Negative Mask Layers
A comment was made earlier about the orange appearance of negative film. In order to improve the reading of the negative film during subsequent photographic printing, additional dye layers are provided that are depleted as neighbouring layers form images. Sadly which layers were involved is now for the writer lost in the mist of time but it was possibly cyan and magenta. Anyway, these additional dye images acted much like the electronic masking. In fact they were so effective that they had to be partially cancelled to avoid excessive colour saturation when scanning negative film.
4.0 Conclusion
The foregoing has been a rather rapid trip through film reproduction via Telecine but, hopefully, it has shown the basic thought process to be followed.
In B3402 / B3404 days the difficulty of producing accurate processing electronics made the task more difficult. Unwanted tube effects such as image shading also contributed to significant degradation, particularly with negative reproduction. With the digital system of the B3410 telecine, the accuracy of the logarithmic and exponential conversion was assured by ROM look-up tables and, naturally, the stability was excellent. With only the single TV line provided by the line array sensor to correct it was possible to remove all shading effects. This allowed us to approach the ‘Looking through an open window’ level of reproduction that the folks in Hollywood used to talk about.