GB2032152A - Travelling Wave Digitizer - Google Patents

Abstract

A digitizer shifts long and short wavelength square waves across x and y grids in a platen to perform upon a cursor signal time measurements that represent coarse and fine cursor position data. The periodic staircase cursor signal is filtered before use to permit time measurements that resolve cursor positions to locations between lines in the grids. The contribution of the filter to the coarse and fine time measurements is obtained in a special reference measurement, so that it may be removed. A processor combines these data into Cartesian coordinates, and implements procedures for error reduction. These include averaging of consecutive time measurements, dynamic adjustment of a timing reference to remove ambiguities inherent in measuring very small or very large cyclic quantities, and correction of coordinates for the effects of cursor motion during measurement and for any non- perpendicularity of the x and y grids. <IMAGE>

Description

SPECIFICATION Travelling Wave Digitizer This invention relates in general to electronic means for gathering data for processing by computers or other data handling or storage devices. More particularly, the invention pertains to electronic devices that convert graphic information such as charts, maps, graphs and images on film into Cartesian coordinate data through the use of an operator positioned cursor whose location on a platen is obtainable in terms of Cartesian coordinates. Such devices are commonly called digitizers.

To be useful with the computers, plotters and data storage media currently available, a digitizer must be capable of supplying at least several tens of coordinate pairs per second, as well as be accurate and otherwise easy to use. The accuracy anXase of use criteria have practically eliminated mechanical linkages from digitizers used for electronic data gathering and processing.

Most digitizers used for such purposes employ some means of electronic interaction between the cursor, or stylus, and a digitizing surface, called a platen.

Digitizers may be classified as either incremental or absolute. An incremental digitizing technique is one wherein it is inherently the effects of cursor motion that are detectable, rather than the actual location of the cursor, per se. An incremental digitizing system must continually accumulate changes in cursor position, starting with its original position. In addition, the original position must be identified as such, and assigned coordinate values (usually as the origin). Such digitizing techniques suffer from the disadvantage that the cursor cannot ordinarily be lifted away from the platen (say, to change the document being digitized). For to do so severs the electronic interaction between the cursor and the platen.Since an incremental technique requires the digitizer to constantly keep track of changes in the cursor's position in order to know where the cursor is, and since it cannot do this once the cursor has been lifted away from the platen, once lifted the cursor cannot then be replaced with the intent of digitizing further with the same point on the platen serving as the origin.

A digitizer employing an absolute technique does not have this difficulty. In such a digitizer the technique itself provides direct determination of the cursor's position on the platen. Each such determination is totally independent of the one before it, and is capable of determining the cursor's location anywhere within the active area of the platen.

A common general approach to absolute digitizing may be described as follows. Find some physical or electrical phenomenon having a parameter that varies, or can be made to vary, with distance. Apply the phenomenon along a dimension or axis of the platen in such a way that there is a one-to-one correspondence between the displacement of the cursor along the axis and the variation in the chosen parameter. Cyclic phenomena may be use as long as the correspondence is restricted to less than one cycle. The cursor detects the chosen parameter, whose variation is then measured and converted into distance units.

Examples of this approach include digitizing techniques utilizing voltage dividers along the axis of the platen. The dividers, for instance, can be used to divide a dc potential or to provide steps of phase shift in an ac signal.

Such a general approach involves the following difficulty. As the degree of desired accuracy increases (say, from .01 inch to .001 inch), or as the size of the platen increases while maintaining a desired accuracy, or both, the resolution of the technique (i.e., the ability to detect or measure smaller and smaller percentages of full scale deviation of the chosen parameter) must increase correspondingly. But beyond a certain point increases in resolution can be quite difficult. An additional aggravating factor is the case where it is desired that substantially the same digitizer circuitry work with both large and small platens with no change in the basic accuracy.

A solution to the problem of increasing resolution is to devise a technique that lends itself to the incorporation of coarse and fine modes of measurement. One such scheme is presented by Dym (U.S. Patent No. 3,921,165 issued November 1 8, 1975). His scheme involves the segmenting of the active area of the platen into sections. Successive sections along a dimension are tested until a section is found to contain the cursor. Then a fine measurement is made within that section. There are two disadvantages to the notion of "sectionalizing" the platen. First, additional circuitry must be employed to select the section of the platen to which the measuring phenomenon is to be applied, and to then apply it.

Second, there are difficulties connected with attaining linearity at the boundaries of the individual sections. Whereas techniques having a platen comprising a single measurement section along each dimension have such boundary problems only at two extremes for each dimension, a sectionalized platen has two extremes for each section. This greatly increases the difficulty of attaining linear measurement.

A common feature of systems that use coarse and fine measurement involves a "multiplicity ambiguity". That is, the fine measurement provides high resolution within a relatively short interval (one of many into which the platen may be considered to be divided during fine measurement), but is unable to distinguish between the cursor's being in a certain place in one interval and its being in the same relative place in an entirely different interval. The task of coarse measurement then becomes one of identifying the fine interval. There have been a number of systems along these lines.

One such is the technique of Centner et al., (U.K. Patent No. 3,735,044). It involves the use of two slightly different frequencies whose difference is a frequency small in comparison to either of the first two frequencies. Thus, the period of the difference frequency is long compared to either of the other two frequencies.

Also, the amplitude of the difference frequency is related to the cursor's position, and may be taken as an approximate (coarse) location. The major disadvantage to this technique is that the coarse measurement circuitry does not readily lend itself to also performing the fine measurement. In the example cited two different instances of a digitizing technique were used; one set of circuitry for coarse and one for fine. This gives rise to more varied circuitry than would be found in a digitizer whose fine measurement technique was an extension in principle of the coarse technique.

As a final example of the approaches to coarse and fine measurements, consider the technique of Cotter (U.S. Patent No. 3,732,369). In Cotter's technique the platen contains coarse and fine grids in each dimension. The obvious disadvantage here is the need for extra grids, with the attendant problem of increased difficulty of grid alignment during platen assembly.

Most previous digitizers have not incorporated microprocessors into their hardware architecture.

The incorporation of a processor can result in a reduction in the amount of circuitry needed for measurement purposes, as well as in the application of sophisticated error reduction techniques (such as will be disclosed). As an example, consider the effect of cursor motion during coordinate measurement. Unless the digitizer has separate circuitry for both the x and y axes, such that the x and y coordinates are always sampled simultaneously, it may happen that cursor motion causes the x and y values obtained to correspond to two different points. By employing a processor that can correct for cursor motion it is possible to reduce the amount of circuitry by using one common set of circuitry to alternately sample x and y values. Also, the inclusion of a processor allows sophisticated averaging and noise recognition techniques that enhance high accuracy digitizing.

The primary object of the invention is to provide a high accuracy digitizer that performs absolute measurements incorporating coarse and fine modes each of which uses the same basic technique, so as to achieve a reduction in the necessary amount of circuitry.

The present invention provides apparatus for digitizing the coordinate of a point along a measurement axis on a platen, the apparatus comprising: a platen having a plurality of parallel conductors generally perpendicular to the measurement axis; circuit means for impressing a voltage gradient between selected conductors of the platen and for propagating said voltage gradient across the conductors of the platen at a uniform rate; capacitive pickup means being capacitively coupled to the conductors of the platen for sensing the propagated voltage gradient as it is propagated across the conductors proximate to the location of the capacitive pickup means; measurement means for sensing a first point in time at which said voltage gradient begins to propagate across the conductors of the platen and for sensing a second point in time at which said propagated voltage gradient is sensed by said capacitive pickup means and for determining the difference between said first and second points in time to obtain a digital representation of the coordinate of the point along the measurement axis on the platen over which said capacitive pickup means is positioned.

A coordinate digitizer for determining the location of a freely moveable cursor on a platen is disclosed. The digitizer alternately employs x and y axis shift registers to sweep various symmetrical square waveforms of differing wavelengths across each axis of the platen. The platen comprises a flat surface within which is embedded two orthogonal grids each of which consists of uniformly spaced individual conductors. The cursor is a capacitive pickup into which is coupled each separately swept signal in the platen. The cursor signal is amplified, filtered, squared-up and compared to a reference signal. Cursor position is reflected in the length of time required for the swept waves to leave the edge of the active area of the platen and subsequently pass under the cursor.This transit time reveals itself as a time difference between corresponding leading edges of the reference signal and treated cursor signal.

A timer measures this difference, which can then be converted to coordinate data.

Since these are differences between periodic waveforms, as a given difference increases it reaches a maximum and then abruptly passes through zero, and then continues to increase as before. That is, the degree of misalignment (i.e., "phase shift") between two periodic things is itself periodic. This gives rise to a certain degree of uncertainty when the measured difference is either very large or very small; noise and other uncontrollable variables may have affected the measured difference so that it differs from the actual difference by an entire period. This is avoided by detecting the case of either a near maximum or near zero difference and temporarily offsetting the reference signal by a ha If- wavelength. The difference obtained is then adjusted arithmetically by an amount corresponding to the half-wavelength. Noise will still affect the value of the reading, but will not cause a difference that should be near the maximum to appear as very small, or vice versa.

The x and y axes are swept alternately. Within the activity for each axis there occurs alternate sweeping of that axis by long and short wavelengths. The long wavelength reveals a coarse position which is then resolved into an integral number of "fine" distance units, each of which corresponds to a short wavelength. A fine position is obtained by sweeping with the short wavelength. By itself, the fine measurement determines where in any of the possible fine distance units the cursor is positioned, but does not determine which such unit.

The time difference measurements described include an element of delay caused by amplification and filtering. The delay through the filter and its associated circuitry is regularly measured in a special measurement cycle. The filter delay is then subtracted from the coarse and fine measurement data to obtain time data that actually represents the cursor's position. The fine position is then combined with the coarse position to produce the actual raw coordinate.

These data may be converted to any desired unit of linear measure.

The platen incorporates a configuration involving redundant drive lines at its extremes in each dimension to reduce measurement errors at these extremes. These redundant lines reduce the number of active lines needed in the platen to produce an accurate active measuring area of a given size.

An internal processor for controlling the various aspects of the digitizer's activity also obtains cursor velocity information. The processor corrects the digitized coordinates for the effects of cursor motion that may occur during the alternating measurement of the x and y coordinates. This is done by noticing the velocity in the x direction and adjusting the x coordinate to correspond with the supplied y coordinate. The processor also corrects the coordinates for an individually predetermined and yet arbitrary amount of non-perpendicularity between the grids within the platen.

There now follows a detailed description which is to be read with reference to the accompanying drawings of apparatus according to the present invention; it is to be clearly understood that this apparatus has been selected for description to illustrate the invention by way of example and not by way of limitation.

In the accompanying drawings: Figure 1 is a perspective view of a digitizer whose measurement technique incorporates the use of swept square waves and a transparent capacitive pickup; also shown is a desktop computer exemplary of any of a number of devices suitable for controlling the digitizer; Figure 2 is simplified block diagram of the digitizer of Figure 1:: Figure 3 is an illustration of the general physical and electrical configuration of the platen or digitizing surface used in conjunction with the digitizer of Figure 1; Figure 4 is an illustration of the general physical and electrical configuration of the cursor or pick-up used in conjunction with the digitizer of Figure 1; Figure 5A-B is a detailed block diagram of the digitizer of Figure 1; Figure 6 is an illustration of the basic relationships between the shift registers, the swept coarse and fine square waves, and the grids within the platen; Figure 7 is a waveform diagram depicting coarse position measurement along either of the x or y axes with a swept square wave whose wavelength is 1.8 meters;; Figure 8 is a waveform diagram depicting fine position measurement along either of the x or y axes with a swept square wave whose wavelength is 40 millimeters; Figure 9 is a simplified block diagram illustrating the internal configuration of the digitizer of Figure 1 during the reference measurement for the x dimension; Figure 10 is a waveform diagram illustrating the relationship between coarse and reference measurements, for either of the x or y dimensions; Figure 11 is a waveform diagram illustrating the relationship between fine and reference measurements, for either of the x or y dimensions; Figure 12 is a waveform diagram illustrating how a reference measurement for either axis is handled when its value exceeds the coarse or fine value that it is associated with;; Figure 13 A-C is an illustration of why and how the internal timing reference signal is dynamically adjusted by a half-cycle during the various measurements performed by the digitizer of Figure 1; Figure 14 is an illustration of the method used to combine the coarse and fine values for either of the x or y axes into a single value; the method incorporates adjustments to the coarse value to render it in agreement with the fine value; Figure 1 5 A-D is a diagram illustrating the basic sequence of operations used by the digitizer to obtain a pair of coordinates corresponding to the location of the cursor on the platen; Figure 16 A-B is a simplified flow chart illustrating the relationship of the programming executed by the processor in the block diagram of Figure 5A to the basic measurement sequence of Figure 1 5; Figure 17 A-C is a more detailed yet still simplified flowchart of the interrupt service subroutine of Figure 1 so;; Figure 1 8 is a more detailed yet still simplified flowchart of the "Basic Point Processor" portion of the mainline programming of Figure 1 6B; Figure 1 9 is a diagram depicting the fundamental reason why cursor velocity-related coordinate correction is needed; Figure 20 is a diagram depicting the general nature of an algorithm used to correct for the problem shown in Figure 19; Figure 21 is a diagram depicting some details of the algorithm of Figure 20; Figure 22 is a diagram depicting the nature of an algorithm to correct for the error induced in the digitized coordinates by non-perpendicularity of the grids in the platen; Figure 23 is a diagram illustrating how the non-perpendicularity error angle of Figure 22 is encoded into the switch register of Figure 5A;; Figure 24 is a diagram illustrating the relationship between the actual sequence of measurement and computational activity, and the logical levels of the error reduction computations; Figure 25 is a diagram illustrating the reason for the inherent non-linearity of measurements taken at the extreme edges of the active area of the platen; Figure 26 is a diagram illustrating the "brute force" solution to the non-linearity problem of Figure 25; Figure 27 is a diagram illustrating the use of "redundant" drive lines in the solution of the nonlinearity problem of Figure 25; and Figure 28 is a simplified block diagram of an alternative embodiment of the digitizer of Figure 1, incorporating a magnetic pickup; Figure 1 shows a digitizer 1 incorporating a swept square wave measurement technique connected to a controller 2.Any of a number of different devices could serve as the controller. In general, the controller would be a computer of either the mini or desktop variety, or it could be a programmable calculator. Other controlling devices are possible; with a suitable interface the digitizer can be controlled with a teletype.

The digitizer 1 comprises a digitizing surface or platen 3, a freely movable pick-up or cursor 4, a keypad 5 and a multi-character dispiay 6. The general description of the digitizer use is as follows. The document to be digitized is secured to the platen by tape, weights, or any other suitable manner. With the document resting flat on the platen, the cursor is placed on the document. The cursor has a transparent region containing a reticle, or cross hairs, whose intersection represents the location whose coordinates are to be provided. Digitizing is accomplished by the operator's first activating the controller and then successively positioning the cross hairs over the locations on the document whose coordinates are desired. The controller receives the coordinate data from the digitizer, and manipulates it in whatever manner that has been provided for.

A number of additional points should be made.

To begin with, the document itself is not necessary for the digitizer's electrical operation.

The digitizer digitizes the location of the cursor cross hairs on the platen, not the position of a point on the document that just happens to be under the cross-hairs. Ideally, there is no electrical interaction between the document and the digitizer.

There are two basic modes that the digitizer may operate within. These are Single, wherein the digitizer supplies to the controller the coordinates of a single point, once each time the operator instructs the digitizer to do so, and Continuous, wherein the digitizer repeatedly and automatically supplies coordinates in some uniform or regular way. In the Single mode the operator specifies each point to be digitized by activating a Data switch on the cursor. In the Continuous mode the digitizer supplies a new point either at regular time intervals, or upon the condition that the new point is sufficiently far removed in distance from the previous one.

Both the time interval and the distance can be arbitrarily chosen, within a predetermined range, by the operator. The operator initiates and terminates continuous mode sampling through use of the same Data switch on the cursor. Other controls on the digitizer's front panel specify whether the digitizer is in the Single or Continuous mode.

Another fact of consequence is that the digitizer never transmits coordinates to the controller except as part of a response to a request for coordinates by the controller. Now, internally the digitizer is always determining the coordinates of the cursor. The notions of Single and Continuous sampling involve escapement mechanisms triggered by either the Data switch in The Single mode, or by the Data switch in conjunction with the time or distance criteria in the Continuous mode. The escapement mechanisms isolate and save the coordinates corresponding to the points the operator intended to digitize. The saved coordinate pair is sent to the controller when it is requested. Meanwhile, the internal digitizing process has never ceased.

There is a sample light on the cursor that indicates the transfer of data from the digitizer to the controller. When a point has been engaged by the escapement mechanism (i.e., when the operator has "digitized" a point) the light goes out. The light comes back on when the controller asks for and then subsequently receives the coordinates.

The keypad contains a number of control keys for such purposes as establishing the Single and Continuous modes, as well as digit entry keys.

The digit entry keys can serve to annotatepoints as they are sent to the controller. That is, what is sent to the controller is not simply a pair of numbers, but a sequence of four numbers. One of those numbers can be made to represent any floatingpoint numerical value the operator desires, through the use of the keypad. In this manner the coordinates sent to the controller may be given numerical "labels' that establish a correspondence between the digitized locations on the document and the coordinates stored in the memory of the controller. Alternatively, the annotation may be used to manually supply an elevation, or z axis, to the digitized points.

The digitizer's display is responsive to a number of digitizer operations, such as annotation of points, and can also be controlled by the controller.

Virtually all of the controls on the digitizer's front panel, and on the cursor, that pertain to actual digitizing can be activated by the controller, as well as by the operator.

The cursor may be replaced with a stylus similar in shape to, and incorporating, a ball point pen.

The digitizer incorporates an electronic beeper whose duration and frequency of tone can be specified by the controller. The digitizer also incorporates a vacuum pump. The vacuum thus obtained is applied through an electrically controlled valve and subsequently through a hose to the cursor. A vacuum port on the underside of the cursor allows the vacuum to cause the cursor to adhere to the platen or document when the electrically controlled valve is open. The valve can be controlled either by the Air switch on the cursor, or by commands from the controller.

Referring to the simplified block diagram of Figure 2, there now follows a brief description outlining the functions of the major elements of the digitizer. Each topic mentioned will be discussed at greater length in subsequent sections.

Consider measurement in the x dimension. A platen 3 containing uniformly spaced (5 mm) parallel conductors 46, collectively called the x grid, provides a physical embodiment across which a symmetrically varying electrostatic field is impressed. The field is generated by the outputs from the x shift register 9, and corresponds to the instantaneous bit pattern in the shift register. The bit pattern is generated by applying a 5 KHz square wave signal 72, named IREF (Internal Reference), to the serial data input of the x shift register as it is shifting. If, for example, the shift rate were 1.8 MHz (as it is during coarse position measurements) the resulting bit pattern in the x shift register would be 1 80 bits set followed by 180 bits cleared.

A bit pattern in the x shift register causes a corresponding electro-static field to appear at the platen. Since the conductors 46 in the x grid are parallel to the y axis, the electro-static field is uniform in the y direction. The value of the field in the x direction is a function of time and the bit pattern in the x shift register. The regular shifting of the x shift register causes the corresponding electro-static field to change its position on the platen; that is, it moves in the direction defined by the x axis.

The moving electro-static field passes under the cursor, whose essential element is a capacitive pick-up 24. The capacitive coupling between the pick-up 24 and the wires 46 of the x grid couples a signal to the cursor. The signal is a 5 KHz ac voltage which is subsequently amplified and applied to a 5 KHz filter 7. (The cursor signal is 5 KHz because that is the rate at which the signal originating the bit pattern at the input end of the x shift register is changing. Since the shift rate is a synchronized multiple of that 5 KHz, that is also then the rate at which the electro-static field in the platen varies as it moves past a given point). The filter 7 removes the staircase nature of the coupled cursor signal and renders it purely sinusoidal. The output of the filter is converted back into a square wave 43, named SIGNAL, by a zero crossing detector 8.

The position along the x axis of the pick-up 24 is reflected in the time required for a given leading edge of the electro-static square wave on the platen to traverse the distance from the edge of the platen to the location of the pick-up.

Now, the length of the active measuring area of the platen is limited, in each of the x and y dimensions, to a length somewhat less than the distance traversed by a leading edge of the moving electro-static wave when shifted at the coarse shifting frequency of 1.8 MHz, and allowed to displace itself by one wavelength. That is, the maximum length of the platen is such that there will always be less than one complete cycle of the coarse electro-static wave impressed upon the platen. A consequence of this is that any leading edge of IREF might be used to start a timer, which would then be turned off by the next leading edge detected at the pick-up. IREF could perform this duty since it is the immediate source of the 5 KHz signal that is applied to the very edge of the platen, and that is then subsequently shifted.

However, as explained below, a need exists to be able to sometimes offset, by exactly one halfwavelength, the signal that starts the phase counter. A signal 13 that can have this property is REF (Phase Reference), which can equal either IREF or IREF. For the purposes of the above explanation assume Q)REF=IREF. The need for the case ofREF=lREF is explained later.

A time interval corresponding to the one described above is measured by the phase counter 1 2 and its controller 1 The phase counter controller 11 is a means of starting and stopping the binary phase counter 12 according to the leading edges ofREF and SIGNAL. A leading edge of REF starts the counter, which then counts cycles of an 1 8 MHz clock. The next leading edge of SIGNAL causes the counting to cease. A DATA READY line signals a micro processor 14 that the phase counter has data available.

At this point the binary count in the phase counter represents the sum of two values. These are: (1) the time required for a leading edge of the swept coarse electro-static square wave to depart the edge of the platen and subsequently pass under the pick-up; and (2) a value related to the amount of time the cursor signal is delayed by the amplifier and the filter. Now, since the swept square waves traversing the platen are doing so in a continuous and uninterrupted fashion (at least for the duration of each type of measurement), and since the filter delay is several times as long as the largest measurable time interval, there is no way to identify exactly which leading edges of REF are to correspond to the various subsequent leading edges of SIGNAL. It doesn't matter; the periodicity of the waveforms removes the need for such correspondence.

However, the existence of the filter delay presents a difficulty that is solved by obtaining an additional measurement, called the x reference measurement. It is a measurement whose value is related to the delay through the filter, and which may be subtracted from the value obtained by the x coarse measurement to produce a usable x coarse component of the ultimate x coordinate.

The x reference measurement does not obtain the actual total filter delay, which is typically of the order of five cycles of the 5 KHz cursor signal.

What is measured is the "remainder", or fractional portion of the delay in excess of an integral number when the delay is expressed in terms of cycles of 5 KHz. For example, if the delay through the filter was 1.050 msec, that would correspond to 5.25 cycles of 5 KHz, and the x reference measurement would produce a value representing the quarter-cycle in excess of the five cycles. That value would be the number of cycles of 18 MHz that are contained in 50 ysec.

Electrically, an x reference measurement is made in the following way. With the y shift register cleared, the x shift register is made to parallel load IREF at a frequency (the coarse shift rate -1.8MHz) that is both an integral multiple of IREF and is synchronous with it. (This is so that the edges of IREF are not "chopped off' as they appear at the outputs of the shift register). The effect of this is to apply IREF simultaneously and uniformly to each line 46 in the x grid. Thus, the cursor sees the same 5 KHz signal no matter where it is positioned on the platen. The phase counter will now measure the "remainder" of the delay through the filter, as well as including in that measurement any other delays distributed through the system, except delays that are a function of cursor position during coarse or fine measurement.

Once the x reference value has been obtained it can be arithmetically combined with the measured x coarse value to remove the effect of filter delay from that value. The necessary arithmetic is done by a processor rather than by dedicated circuitry.

The determination of the final value of the x coordinate requires an additional type of measurement, viz., x fine. To perform the x fine measurement the shift rate of the x shift register is changed from 1.8 MHz to 40 KHz. The use of IREF as a 5 KHz source of the serially shifted waveform/bit pattern remains unchanged. The slower shift rate produces a shorter bit pattern to represent the 5 KHz IREF. In particular, the bit pattern will be four bits set followed by four bits cleared.

The shorter bit pattern produces correspondingly shorter moving electro-static square waves on the platen. Whereas in the case of coarse measurement one wavelength exceeded the maximum length of the platen. The wavelength of a fine wave is considerably shorter than the piaten. That is, there will be several cycles of fine waves on the platen, and any of several cursor positions (i.e., at intervals of one fine wavelength) will produce the same result in the phase counter.

This does not cause a problem, and does result in an overall increase in the resolution of the x coordinate. The multiplicity is resolved as follows.

The coarse measurement can be examined to determine which of the fine cycles contains the cursor location. Then the coarse measurement can be converted to an equivalent integral number of fine cycles, which is then augmented by the fine measurement.

The increase in resolution comes about in the following way. The phase counter cannot tell the difference between coarse and fine measurement.

SIGNAL and REF are 5 KHz in either case, and the maximum difference between their leading edges is 200 ,usec. In each case the phase counter measures the interval by counting cycles of 18 MHz. Now, there are 3600 cycles of 18 MHz in 200 ,usec. Thus, for either coarse or fine measurement a count of, say, 1 800 in the phase counter represents a distance amounting to a half-wavelength of the swept electro-static wave.

But, in the case of coarse measurement, that wavelength is 1.8 meters (180 bits set+ 180 bits cleared=360 shiftsx5 mm per shift=1.8 M) while for fine it is only 40 millimeters. Accordingly, the distance represented by one "fine count" is 1/45 the distance represented by one "coarse count".

Of course, the measured xfine value must be arithmetically adjusted in accordance with the value of the x reference measurement, just as in the case of the x coarse measurement. Once that is done the coarse and fine components can be combined to produce an actual x coordinate.

The procedure outlined for x dimension measurement is analogous to the one used for measurement in the y dimension. Here is how the two activities are coordinated. The series of x measurements is performed first. To do this, both the x and y shift registers are first cleared. Then the x shift register is allowed to begin shifting at the coarse rate. The y shift register remains cleared. This is important, because it prevents any y-direction cursor motion from crossing an electro-static gradient presented by the y shift register, and thereby coupling an unwanted component into the cursor signal. After the x coarse measurement is taken, the x reference measurement is made. That in turn is followed by xfine, which is performed by changing the shift frequency of the x shift register.

Prior to beginning the y dimension measurements, both shift registers are again cleared. Now the x shift register remains cleared while the y shift register begins shifting at the fine rate. After the yfine measurement there follow y reference and y coarse in turn. Then the entire coordinate measurement sequence starts over.

The x coarse, x reference and x fine, etc., measurement procedure outlined above does not employ simply single instances of phase counter activity to obtain each of the successive values.

Instead, they are each averaged from successive measurements allowed to accumulate in the phase counter. Also, due to the cyclic nature of the quantity represented by the count in the phase counter, a situation involving near coincidence of the leading edges of REF and SIGNAL may result in either a count near zero or one near 3600. Various sources of unavoidable noise, drift in the threshold of the zero crossing detector, as well as cursor motion during measurement, can cause this 3600-count ambiguity in very small or very large counts.

The need to accumulate consecutive individual measurements for later averaging, coupled with the phenomenon of the 3600-count ambiguity (which can upset the arithmetic of the averaging) gives rise to a problem. The solution is to eliminate the 3600-count ambiguity whenever its presence is possible, by temporarily shifting REF by a half-cycle or half wavelength and subsequently adjusting the measured value by an amount equal to a half-wavelength of REF (1800 counts). The effects of noise will still be present, but now they can be subjected to proper averaging.

REF is offset by a half-cycle simply by using its complement in its place. The decision to do this is made by the processor 14 that directs all internal digitizer activity. Generally, the ambiguous measurement is discarded, REF offset, and a new measurement taken. The processor then adjusts the value obtained before any other arithmetic use is made of that value.

As mentioned above, the digitizer incorporates a processor 14 that is responsible for directing all internal digitizer activity. Associated with the processor is both Read/Write memory 48 (R/W) and Read-Only-Memory 49 (ROM). The primary means by which the processor exercises its control is through the mode control latches 1 5.

Through means described in detail in subsequent sections, the processor continuously manages the values of these latches. The latches, in turn, affect the interconnections between the various circuit elements described with reference to Figure 2 thus far.

A detailed exposition of the collective use to which the latches are put is provided in the section concerning Figure 1 5. At this point, however, it is appropriate to briefly explain each of the individual latches.

The START latch is used to inform the phase counter controller 11 that a new measurement is to begin. It also causes the phase counter 12 and its controller to be cleared. The START latch is automatically reset after a suitable period of time.

The DELTA latch is used to trigger a delay one shot that withholds the start of phase counter activity until various circuit conditions have had time to stabilize. DELTA ensures that the bit pattern in the active shift register has achieved uniformity, and that the filter has had time to settle after seeing a change in the cursor signal.

DELTA is issued at the start of each different type of measurement.

The AV TWO latch tells the phase counter controller 11 to accumulate two consecutive individual measurements into the phase counter 12. AV TWO is not utilized in the present embodiment.

The AV FOUR latch tells the phase counter controller 11 to accumulate four consecutive individual measurements into the phase counter 12. AV FOUR is used in obtaining values that are later averaged by the processor.

The absence of both AV TWO and AV FOUR at the time START is issued implies an AV ONE to the phase counter controller 11. An AV ONE tells the phase counter controller to perform a single individual phase counter measurement.

The IREF/IREF latch controls the half-phase offset ofREF, and is dynamically adjusted by the processor 14 at the start of individual AV ONE's or AV FOUR's, as required.

The X1Y latch designates which of the x shift register 9 or y shift register 10 will be the active shift register.

The CLEAR latch clears both the x and y shift registers.

The SERIAL/PARALLEL latch controls whether the shift register selected by the X/Y latch will shift IREF in a serial data for either of coarse or fine measurement, or parallel load IREF for a reference measurement.

The COARSE/FINE latch specifies the shift rate of the active shift register identified by the X/Y latch.

The processor controls the values of these latches during the coordinate measurement sequence according to programmed algorithms encoded in the ROM 49.

The physical and electrical nature of the platen 3 may be understood with reference to Figure 3.

As shown there the platen consists of the pieces of (float) glass 51 and 52 having printed circuit traces upon them. Glass is shown as exemplary, but other materials could be used. Glass is preferred in the present embodiment because it allows back-lighting of transparent documents, as well as rear projection of images onto the platen.

The upper side of the upper piece of glass 51 is frosted to diffuse illumination for back-lighting purposes, and to act as a screen for rear projection. It is also the actual digitizing surface.

The upper piece of glass 51 has the y grid 17 on its underside. The lower piece of glass 52 has the x grid 1 8 on its upper side. The edges of each piece of glass 51 and 52 are ground to render the pieces rectangular. The traces 1 7 of the y grid are parallel to the long, or "length", dimension of the upper piece of glass 51. The traces 18 of the x grid are parallel to the short, or "width", dimension of the lower piece of glass 52. Ribbon connectors 20 are soldered to the traces along one end of the x grid and along one end of the y grid. Finally, the two pieces are laminated together with polyvinyl butyral resin to become a flat piece of laminated safety glass containing orthogonal x and y grids. Great care is taken during lamination to ensure proper registration of the glass pieces so that there will be the least possible error in the perpendicularity of the x and y grids.

Also shown in Figure 3 are the locations of the four sets of redundant lines 19. The purpose and function of these lines is discussed in detail in a subsequent section; briefly, they reduce the nonlinearity of measurements made at the edges of the usable area of the platen.

That usable area is identified by and contained within a rectangular line 53 etched into the top of the upper piece of glass 51. Digitizing is restricted to points on or within the region 1 6 circumscribed by the boundary 53.

Aside from the four sets of redundant lines 19, all of the conductors in the platen are separate and insulated from each other, and connected to the shift registers at only one end; the other end of each such conductor is unterminated. The x and y grids are separated and insulated from each other by the resin that laminates the glass; lines within a grid are separated and insulated by their horizontal spacing. Non-redundant lines within a grid are regularly and uniformly spaced 5 mm centre to centre. In the present example all traces are approximately .015 inches in width.

The physical and electrical nature of the cursor 4 may be understood with reference to Figure 4.

As shown in the figure, the electrical nature of the cursor is that of a shielded plate 24 of a capacitor exposed to the conductors in the platen through a circular aperture 23 in the underside of the shield 54. The aperture allows capacitive coupling between the actual pick-up 24 and conductors of the x and y grids. The diameter of the aperture is approximately equal to a half-wavelength of the smallest (i.e., the fine) swept wavelength that the cursor is exposed to. This is an optimum value, since smaller diameters would reduce the amplitude of the signal coupled into the pick-up, while larger diameters (up to one wavelength) would also reduce the signal due to selfcancellation of the out-of-phase portions of the coupled signal. The circular nature of the aperture provides a degree of preliminary sinusoidal filtering of the signal coupled into the pickup 24.

The fully shielded pick-up provides not only a well defined aperture, but protection from the effects of variable stray capacitances; e.g. to the hand of the operator.

Physically, the shielded pick-up is formed with transparent conductive coatings of indium oxide on pieces of glass that are subsequently laminated together. Both the shield 54 and the pick-up 24 are conductors formed by indium oxide coatings on the pieces of glass.

Also shown in Figure 4 are the locations of the data switch 22 and sample indicator 21.

Vacuum generated in the mainframe of the digitizer is conveyed to the cursor through a holiow cable. The vacuum is available at a narrow circular opening in the underside of the cursor.

The opening surrounds the glass reticle.

A more complete understanding of the operation of the digitizer may be had with reference to the detailed block diagram of Figures 5A and 5B. Recalling for a moment the simplified block diagram of Figure 2, the latter is essentially a subset of Figures 5A and 5B. What was said relating to Figure 2 is applicable to that part of Figure 5B that is the same as Figure 2, and will not be repeated.

As previously mentioned, the digitizer contains a processor that directs the internal operation of the digitizer's measurement apparatus; it also performs various error reduction procedures.

Now, with respect to both of these functions, no particular processor or type of processor need be used. In particular, the processor need not be of the 1 6-bit parallel variety, although such a processor is of great utility in the internal operation of the present digitizer. In a theoretical sense, and in connection with the task of controlling the internal measurement operation of the digitizer, one could conceive of the digitizer as having no processor at all, or at least of not using a processor for that purpose. In such an embodiment a state machine or other collection of logic circuitry would implement the necessary interaction between the clocks, x and shift registers, platen, pick-up, amplifiers, filter, zero crossing detector, and the phase counter.The basic digitizing technique resides in such interaction, not in programming executed by a processor.

However, with respect to the function of implementing error reduction procedures of the type to be disclosed, a processor of some sort is all but absolutely necessary. These procedures require a fair amount of number crunching and logic to implement.

What has been said is that the basic digitizing technique does not in principle require the use of a processor. But, certain error reduction procedures that are useful adjuncts to the basic digitizing technique do, as a practical matter, require the use of a processor. Hence, the embodiment disclosed involves the use of a processor to its best advantage for both the functions of internal digitizer control and error reduction.

A fairly thorough description of the attributes of the actual processor whose use is described in this embodiment may be found in co-pending application serial number 837,771, filed 9/29/77.

TheXication is in the name of Jack M. Walden et al., assigned to the Hewlitt-Packard Company, and is titled "Programmable Calculator". Of particular interest are the various Figures and their associated text among Figures 44 to 127 that describe an entity called the "BPC". The BPC is the processor used in the present digitizer.

Referring now to Figure 5A, a 1 6-bit micro processor 14 operates in conjunction with a memory that is divided into aRead/write (R/W) address space 48, a register address space 50, and a Read-Only-Memory (ROM) address space 49. Of particular interest is the nature of the register address space 50. It exists between addresses 0 and 37i, inclusive.

What is special about the register address space is that the processor automatically generates an inhibiting signal that prevents all of R/W or ROM from responding to memory cycles directed to those addresses; some element of that "bulk" memory would otherwise try to respond to addresses within that range. Such prevention allows agencies other than the bulk memory to selectively respond as memory for certain addresses within the range of the register address space. For instance, the processor itself contains four addressable registers that respond as memory underthe aegis of this scheme.

To provide a means for the rest of the digitizer and the processor to communicate with each other, the digitizer incorporates hardware that responds as memory to selected addresses within the register address space. Under this scheme the processor performs an input operation when it reads one of those "memory locations", and performs an output operation when it writes to one of those "memory locations". The processor controls internal digitizer operation in part by inputting data from the phase counter, as well as status information concerning the digitizer's controls, such as from the keypad and the cursor.

It further controls the digitizer by outputting information that sets up the mode control latches, as well as outputting other information for other purposes.

The various input operations of interest are briefly summarized below: A read-from-memory operation directed to address 1 2a obtains fifteen bits of phase counter data from the phase counter 12, and one bit of low signal (or not) information from the filter 7.

A read-from-memory operation directed to address 1 68 obtains status information concerning the digitizer's local controls, such as the keys on the cursor and on the keypad. This information is encoded as shown in Figure 5A.

A read-from-memory operation directed to address 138 obtains data from a 1 2-bit binary counter 35. The counter runs at a 1.8 MHz rate and is employed as a hardware timer in conjunction with cursor velocity measurements.

This is part of the mechanism for the error reduction procedure that corrects coordinates for the effects of cursor motion.

A special mechanism is used to control the counter 35. A timer control circuit 56 detects a write operation to address 138 (the timer) and interprets this as a command to clear and stop the counter. It does this by issuing a CLR to the counter and disconnecting its CLK IN line from the 1.8 MHz clock. The next read operation to address 138 restarts the counter. The timer control circuit does this simply by reconnecting the CLK In line of the counter to the 1.8 MHz clock on the next positive edge of the clock. The data obtained by this read is ignored. The next read, however, will stop the counter so that its count may be transferred as data to the processor. The counter does not restart.

Subsequent reads from 138 will obtain the same value, until the counter is cleared and then restarted.

A read-from-memory operation directed to address 118 1, obtains an encoded value of the axes skew angle from the 1 6-bit switch register 34.

The skew angle is the difference between the actual angle formed by the x and y grids and a true right angle (900). The skew angle represents the degree of non-perpendicularlity between the electrical axes of the digitizer's platen. The processor uses the skew angle (whose value is determined solely by the properties of the platen and is subsequently encoded into the switch register upon digitizer assembly) as input data to an error reduction procedure that corrects the coordinates for the non-perpendicularity of the axes.

A read-from-memory operation directed to address 1 08 is part of a mechanism used by the processor itself to determine the conditions surrounding an interrupt to the processor. When an interrupt occurs the processor will read the contents of location 1 0s and branch to a location based on the value read. An interrupt control circuit 57 arranges that the various different conditions causing an interrupt are each simultaneously reflected by differing values of certain bits in location 108. In this way the processor automatically branches to a subroutine that corresponds to the nature of the interrupt. An interrupt control line 39 informs the processor that an interrupt is required.

There are two general conditions that cause interrupts. The first of these is the accumulation of a complete measurement in the phase counter 12. The phase counter controller signals this event with DATA READY, The other interrupt is basically a time signal. It occurs once every millisecond, and is used by the routines that control Continuous mode sampling to monitor how long it has been since the "last point" was "digitized" (i.e., in terms of what the operator thinks is going on).

In the event that both interrupts occur simultaneously, the one millisecond interrupt is serviced first, followed immediately by service for the DATA READY. No interrupt is lost.

A read-from-memory operation directed to address 178 reads a byte of data sent to the digitizer from its controlling device.

The various output operations of interest are briefly summarized below: A write-to-memory operation directed to address 128 establishes the values of the mode control latches 1 5. The particular sequence of values sent to the latches is explained in detail in a subsequent section. At present it is sufficient to say that the various latch values appear to the processor as individual bits in the word to be written to location 128.

A write-to-memory operation directed to address 138 stops and clears the velocity timer 35, as previously explained.

A write-to-memory operation directed to address 148 controls whether the beeper is on or off, and if on, at what frequency it operates.

A write-to-memory operation directed to address 158 controls the contents of the digitizer's display.

A write-to-memory operation directed to address 1 68 controls the illumination of the sample indicator 21 on the cursor, and also controls the application of vacuum to the cursor.

A write-to-memory operation directed to address 178 sends a byte of data to the device controlling the digitizer.

Recalling the function and purpose of the phase counter controller 11 of Figure 2, this circuitry is shown in expanded form in Figure 5B.

Referring now to Figure 5B, the phase counter controller incorporates a flip-flop 32 to control whether or not the phase counter 1 2 actually counts (at the 1 8 MHz rate). When the flip-flop 32 is set counting may proceed. If conditions are met for taking a measurement, or accumulating an additional measurement, the signal OKGO will be true and the next rising edge of REF will then trigger a one shot 69, which sets flip-flop 32 and starts the phase counter. The next rising edge of SIGNAL 43 (the output of the zero crossing detector) will trigger a one shot 68 which resets flip-flop 32 and halts the phase counter.

The basic sequence of events just described is used as part of an AV ONE, AV TWO or AV FOUR related measurement. The mode control latches will have previously specified which type of operation is to be performed. What is needed is circuitry to implement the specified number of accumulations to the phase counter. This is achieved as follows.

Each time flip-flop 32 terminates either, (fl an on-going new count or (ifl an accumulation to an old count in the phase counter, the signal COUNT OVER increments a binary counter 29 that keeps track of the number of count intervals thus far accumulated. When a count detection circuit 30 produces an output that matches either AV TWO or AV FOUR, or matches their "NOR" (which specifies an AV ONE), the signal DATA READY is generated. This signals the processor 14 that the desired number of time interval measurements have been accumulated in the phase counter 12.

DATA READY also rests a flip-flop 31, which ensures that OKGO becomes false. Without OKGO REF cannot restart the phase counter. The count in the phase counter remains unchanged until the processor sets the mode control latch called START. START appears at the beginning of each AV ONE, AV TWO or AV FOUR related measurement. What START does is clear the 2-bit binary counter 29 that registers the number of counts thus far accumulated, clear the phase counter, and set the flip-flop 31. Setting flip-flop 31 generates OKGO, provided WAIT is false.

Assuming WAIT to be false, the next REF will start the next instance of the entire phase counter sequence described thus far. The START mode control latch is automatically reset by the hardware a short time after it is set.

WAIT is the output of a 1.2 msec one shot 33 (see Figure 5A) that is triggered by the mode control latch called DELTA. DELTA is issued whenever the filter may need settling time. That occurs at the beginning of each major type of measurement such as x coarse, x reference, etc.

What WAIT does is hold OKGO false until the settling time is over. Once OKGO is again true (i.e.. after WAIT goes false) the next REF will initiate the measurement.

Operating in conjunction with the DELTA|WAIT/OKGO mechanism just described is a collection of circuitry that generates a signal called PRE-LOAD (see Figure 5B). Its function is as follows. To begin an x fine or y fine measurement a new bit pattern must propagate its way down the active shift register. The same is true of coarse measurement also, but the coarse shift rate of 1.8 MHz is sufficiently rapid that no appreciable delay is seen in shift register set-up.

However, the normal fine shift rate is only 40 KHz.

This means that there can be an appreciable amount of time required to get the proper fourbits-set/four-bits-cleared bit pattern propagated all the way down to the end of the active shift register, especially if a large platen (with its correspondingly long shift register) is in use. To minimize the time required to shift in the fine bit pattern, and thereby keep the coordinate sampling rate at its maximum, the following procedure is employed.

At the beginning of a fine measurement the normal fine shift rate of 40 KHz is temporarily replaced with 1.8 MHz. In order to generate the proper bit pattern of four-bits-setifour-bits- cleared, the serial data input is simultaneously switched from 5 kHz to 225 kHz. This condition is maintained until the longest permitted shift register would be filled with the newly propagated bit pattern. It has been determined that a length equal to forty fine wavelengths will be the longest active dimension of any platen. Since each cycle of the 225 kHz now corresponds to a fine wavelength of four-bits-set/four-bits-cleared, it is sufficient to maintain the pre-load condition for forty cycles of the 225 kHz.

The above described pre-load technique is implemented as follows. A coarse-to-fine transition detection circuit 25 detects the onset of a fine measurement. Its output is used to reset a 6-bit binary counter 26 and set a flip-flop 28.

While flip-flop 28 is set it generates PRE-LOAD.

PRE-LOAD prevents, for its duration, the change in shift rate from 1.8 MHz (as it always is unless already doing a fine measurement) to 40 kHz. This amounts to a delay in instituting the 40 kHz shift frequency until the pre-load cycle is over. PRE LOAD also switches the serial data input of the active shift register from 5 kHz to 225 kHz.

PRE-LOAD is true until the 6-bit binary counter 26 counts forty cycles of the 225 kHz. A D-type latch 70 ensures that the cycles of the 225 kHz preload data frequency are counted in synchronization with the 1.8 MHz pre-load shift frequency. A count detection circuit 27 detects the count of forty and eventually makes PRE LOAD false by allowing the resetting of flip-flop 28 via the D-type latch 71. What the latch 71 does is this. First, it prevents the reset of flip-flop 28 until the count is the counter 26 has reached forty. Second, it delays the reset of flip-flop 28 until the normal shift frequency and the normal fine data frequency (which is IREF) are synchronized. Furthermore, this synchronization is of the same type that occurred between the 1.8 MHz and the 225 kHz. By performing such synchronization the change from PRE-LOAD to NORMAL will not disturb the four-bits-set/fpur- bits-cleared sequence being shifted into the shift register. To do so would defeat the very purpose of the pre-load operation.

During the interval between when the count in counter 26 reaches forty and when PRE-LOAD goes false the signal CLUTCH is true. What CLUTCH does is disconnect the active shift register from the 1.8 MHz pre-load shift frequency. This preserves the pre-loaded bit pattern in the active shift register until PRE-LOAD goes false and normal 40 kHz shifting is resumed.

Normal fine shift operation resumes as soon as PRE-LOAD goes false. Approximately 178,u sec are required for the counter 26 to count to forty.

For the sake of extra margin, PRE-LOAD is assumed to last approximately 200 ysec.

Note the DELTA always accompanies a preload condition. Now consider the 1.2 msec period (WAIT) of the one shot 33 triggered by DELTA.

The first 200 msec of that 1.2 msec are to allow time for the pre-load cycle just described to occur.

The remaining one millisecond is filter settling time for the new 5 kHz fine signal being coupled into the cursor.

The operational aspects of the phase counter control circuitry 11, and of the pre-load circuitry described above, could be obtained through a number of different methods. They could, for instance, be obtained through the use of a state machine. Any scheme that preserves the attributes, as described, of the following signals would function in the place of the example circuits shown in Figures 5A and 5B. The important signals are: START, AVTWO, AV FOUR, AV ONE (implied by the absence of both AV TWO and AV FOUR), WAIT, IREF, REF, SIGNAL, DATA READY, CLUTCH, and PRE-LOAD.

The relationship between the active shift register, the platen, and the coarse and fine swept electro-static square wave may be better understood with reference to Figure 6. Figure 6 shows that the outputs of a shift register 42 are applied to the conductors of a grid within the platen 3. Each active (i.e., non-redundant) line within the grid is 5 mm from its adjacent lines.

There must be one shift register cell for each active line in the platen.

The generation of a coarse measurement wave 62 is reflected in Figure 6. One cycle of the 5 kHz IREF has a period of 200 ,usec. In that time occur 360 cycles of the coarse shift frequency of 1.8 MHz. Hence, there are 360 shifts of the shift register 42 during each cycle of IREF. Since IREF is a symmetrical square wave, 1 80 of those shifts will occur with IREF true, and 1 80 shifts will occur with IREF false This generates a coarse bit pattern of 1 80-bits-set/I 80-bits-cleared. At 5 mm per line in the grid this generates a moving (coarse) electrostatic wave in the platen with a wavelength of 1.8 meters.

The generation of a fine measurement wave 63 is also reflected in Figure 6. During the 200 usec period of IREF there occur eight cycles of the fine shift frequency of 40 kHz. This results in a shift register bit pattern of four-bits-set/four-bits cleared. In a manner analogous to the case for coarse, this results in a consecutive series of moving fine cycles of an electrostatic wave having a wavelength of 40 mm. Note that the fine wavelength is 1/45 the coarse wavelength.

Figure 6 also shows how the redundant lines 1 9 are appended to the last active line at each end of the electrical axis of the platen.

The details of obtaining a coarse measurement may be better understood with reference to Figure 7. Therein is shown a coarse measurement wave in two different positions as it is swept across the platen. The wave 40a is shown as it exists at the time its leading edge begins to sweep across the active area. The wave 40b is the same wave as it exists twenty-two shifts later.

It must be understood that waves 40a and 40b are not conventional waveforms such as can be displayed on an oscilloscope. Waves 40a and 40b are spatial representations of a regularly shifting electro-static field impressed upon the platen.

Each shift of the shift register moves the entire wave 5 mm to the right (as viewed in the Figure).

Furthermore, the horizontal axis of waves 40a and 40b are most usefully understood as representing distance, rather than as representing time.

A short explanation of the proper way to interpret the material of Figure 7 will alleviate possible confusion when following chains of cause and effect that link IREF to REF, and link IREF, through the active shift register, through the position of the cursor, and finally through the filter and zero crossing detector, to SIGNAL.

To begin with, these spatial representations of the moving electro-static wave are "snapshots" of moving thing having length, taken at two different points in time. The motion of the wave 40a can only be shown by indicating its successive positions at successive instances in time. These instances in time might well be arbitrary; 40b is a useful such instance.

As the wave 40 moves to the right the transition from low to high passes under the cursor. This causes a conventional waveform (i.e., its horizontal axis is time) to appear at the cursor, and other conventional waveforms to subsequently appear in adjacent or subsequent circuits. If drawn against the same time axis, these conventional waveforms may be compared with each other over the entire length.

However, a spatial represenation such as 40b may be compared with its associated conventional waveforms only at one point in time, and then only at the location of the cursor. This follows from the fact that at a given point in time some location or feature of the moving electrostatic wave is passing beneath the cursor and causing some particular circuit activity, as indicated by the instantaneous values of the various waveforms. In the next and succeeding instants the moving electrostatic wave propagates to the right, so that the chosen location or feature on it no longer is "above" the phenomena it caused in the waveforms.

For example, consider an imaginary glitch 58 on wave 40b, specially added for illustrating the point of the above explanation. As the glitch 58 moves under the cursor it produces a corresponding glitch 59 on the waveform labelled "INPUT TO FILTER". At that point in time the two glitches 58 and 59 would be aligned in time, and would be at the cursor. They would appear "over each other" in the drawing if such a snapshot of wave 40b were a useful one and were shown in the Figure. But wave 40b can represent only one snapshot, and since it is more useful to depict one involving the arrival of the leading edge under the cursor, glitches 58 and 59 appear displaced in opposite directions from the line drawn beneath the centre of the cursor.As drawn, the further to the right something is on wave 40b, the earlier it will have passed under the cursor and the further to the left on the conventional waveforms will be the location of the corresponding event.

Returning now to Figure 7, what is shown is a coarse measurement wave whose leading edge propagates twenty-two shifts after leaving the edge of the grid before passing under the cursor.

Wave 40 generates waveform 60, which is the resulting signal coupled into the pick-up of the cursor. Waveform 60 has a staircase shape produced by the increases in coupling as the wave 40 periodically and abruptly moves another 5 mm under the cursor.

A complete description of all the variables affecting the shape of the cursor waveform 60 would be quite complex, and will not be attempted. However, such an analysis would take into account the following. The rounding of the edges of the staircase is brought about in part by capacitive loading on the pick-up 24 by its surrounding shield, and in part by the filtering effect produced by the circular nature of the aperture. Also, the capacitive loading on the pickup 24 is not constant. When the transition in the wave is under the cursor, part of the pick-up's coupling capacitance is to ground, and part is to the voltage level presented by the wave. The proportion changes as the wave moves beneath the cursor.

Another important aspect of the cursor waveform 60 is indicated by the notation, on Figure 7, of "A+B". This indicates the nonsymmetrical nature of stair steps above and below the zero-volts axis. This is due to the cursor's not being centred over a line in the platen (i.e., a line in the grid for the dimension being considered). As the cursor's position changes slightly, so does the degree of symmetry, as well as the sizes of the individual steps in the staircase.

These aspects just mentioned are important because they allow the filter 7 to, in essence, interpolate the cursor's position when its centre is not directly over a line in the platen.

Figure 7 also shows the output 61.of the filter.

It was assumed, for the sake of clarity, that the excess, or remainder, of the filter's delay over and above an integral number of 200 Msec periods was small; say, of the order of 3 uses.

The importance of interpolation by the filter can be seen from the following. If waveform 61 were shifted to the left by an amount corresponding to the remainder of the filter delay, the zero crossing of waveform 61 would occur just as the leading edge of the wave 40b "passes" under the centre of the cursor. The amount of offset between the centre of the cursor and the righthand adjacent drive line is marked "X" in the Figure. If the interpolation effect were not available from filtering, the zero crossing detector 8 would be unable to generate a transition in its output 44 (SIGNAL) that falls inbetween the steps in the staircase of waveform 60. That is, in such a system, the cursor's position could be resolved to the nearest drive line only. Filtering allows additional resolution by interpolating the cursor's position.

The output of the filter is used by the zero crossing detector to produce SIGNAL. This is shown as waveform 44. When SIGNAL is compared with REF 13 the length of time that the phase counter counts can be seen. If the filter delay is removed the remaining time can be divided into groups of eight shifts, plus some shifts left over. Each complete group of eight shifts represents an entire fine measurement wavelength contained in the cursor's displacement from the edge of the platen. To determine the number of such complete fine wavelengths is the primary object of coarse measurement, and the result is called the coarse component of the measurement. The fine component is the fractional portion of a fine wavelength, and is found during fine measurement.

The details of obtaining a fine measurement may be better understood with reference to Figure 8. Therein is shown a fine measurement wave in two different positions as it is swept across the platen. The wave 41 is shown as it exists as one of its leading edge begins to sweep across the active area. The wave 41 b is the same wave as it exists twenty-two shifts later. All that was said in connection with Figure 7 concerning the specialized interpretation of spatial representations of moving electro-static waves, when drawn in conjunction with conventional waveforms, also applied to Figure 8.

The main difference between Figures 7 and 8 is that in Figure 8 the measuring electro-static wave has a wave-length of eight shifts rather than of 360. Accordingly, what is measured is the fine component. The amount of filter delay assumed in each of the Figures is the same. The remarks concerning the filter's interpplation of cursor position in Figure 7 are applicable to Figure 8 also.

One important thing that Figure 8 shows is that fine measurement can be taken by noting the length of any of a multiplicity of intervals established by Q)REF and SIGNAL, even if it does not "correspond" to the one in the drawing where the cursor is located.

To obtain the coarse and fine components, as shown in Figures 7 and 8, it is necessary to know the filter delay. This is obtained during a reference measurement, and may be better understood with reference to Figures 9 to 12.

Figure 9 illustrates the equivalent internal configuration of the digitizer during the x reference measurement. The y shift register will have previously been cleared, and will remain that way since the x shift register 9 will be the active shift register. Clearing the y shift register grounds the y grid in the platen. Further, the x shift register no longer shifts a wave across the platen. Instead, the parallel load ability of the x shift register is used to drive all lines the x grid up and down at a 5 kHz rate (using IREF as the source of the 5 kHz.

Thus, no matter where the cursor is in the x (or y) dimension, it sees the same signal.

Furthermore, there is little or no delay between what the cursor sees and REF (which triggers the phase counter). So the delay between the leading edge ofREF and the leading edge of SIGNAL represents the delay induced by the filter and amplifiers. (Shift register delay is actually included also, but is usually insignificant. However, it is properly taken into account as long as the interstage shift delay equals the parallel load delay.

As will be discussed below, what the phase counter actually measure in the reference mode is the "remainder", mentioned earlier, of the filter delay. Typically, the delay through the filter amounts to several cycles of 5 kHz. Now, if there were no delay through the filter, then obviously it could be ignored. Likewise, if the delay were exactly an integral number of period of 5 kHz, it could also be ignored, since with periodic waveforms there is no way to tell the difference.

Indeed, the phase counter cannot tell the difference, and responds only to that portion of the delay that remains after all the integral multiples of the period of 5 kHz have been removed.

Referring now to Figures 10 and 1 the above analysis is illustrated in details for coarse and fine measurements, respectively. The amount of delay assumed is the same for both figures: approximately 5 1/9 cycles of 5 kHz. In each case the Figures show graphically how the periodicity of the measurement waveforms absorb all but the remainder of the delay. They also show graphically how the measured value of the delay can be removed to obtain the coarse and fine measurement components for a cursor displacement in the measured dimension. That involves subtraction of the reference value from the measured coarse and measured fine values.

All three of these values originate as counts in the phase counter, whose (averaged if averaging is the case) values can range from zero to 3599 counts. And for all practical purposes, each of coarse and fine are arbitrarily related to the reference value. That is, reference can be a iarger value than either or both of the coarse or fine values. The problem then arises of how to handle the case of a large-valued reference measurement that must be subtracted from a small-valued coarse or fine measurement. The explanation of this case may be understood with reference to Figure 12. As shown graphically in Figure 12, the solution is to "complement" the resulting negative value by adding 3600 counts to it.

Figures 10 to 12 are general purpose in the sense that they do not differentiate between obtaining or using a reference value for x measurements and obtaining or using a reference value for y measurements. In fact, no such differentiation is required, other than an x reference value be used with x measurements, and a y reference value with y measurements. A y reference value is obtained in the same manner as for x, but the x shift register is cleared while the y shift register is parallel loaded with 5 kHz.

The x reference value will, in practice, be very similar to the y reference value. However, due to the very sensitive and critical role that a reference value plays in the formation of actual coordinates, and since there is some separate x and y circuitry involved (the shift registers, the x and y grids), and since the cursor may move about on the platen as measurements are taken, it is desirable to take separate x and y reference measurements. A system employing a single reference measurement would work, albeit with somewhat less accuracy.

The accuracy of the digitizer's operation is improved by incorporating a means for averaging a number of consecutive measurements taken by the phase counter. In part, this is done by providing the type of phase counter operation involving AV TWO's and AV FOUR's. Recall that in these operations the phase counter automatically accumulates a number of consecutive measurements. However, averaging also employs the repeated use of AV FOUR's. For example, coarse measurements involve a single AV FOUR that is averaged, while reference and fine measurements involve an "AV 21" that is formed form an AV ONE followed by five AV FOUR's. In the latter case the processor's firmware accumulates the total value as the AV ONE and AV FOUR's proceed. There is more to averaging, however, than simply accumulating the result of n consecutive measurements and dividing it by n.

Because the time interval being measured is a difference between two periodic waveforms, the difference itself is also periodic. Thus, if a small difference becomes "markedly smaller", it abruptly becomes "large". Likewise, if a large difference gets "too large", it abruptly becomes "small". The presence of noise and drift in the threshold of the zero crossing detector can cause what should be small counts in the phase counter to become "markedly smaller", and end up "large". Such noise and drift can also cause counts in the phase counter that should be large to become "too large", and end up "small". Such behaviour seriously interferes with the accumulation of numbers for averaging, and is called the problem of 3600-count ambiguity.

The problem of 3600-count ambiguity, and its solution by dynamically controlling a halfwavelength offset in REF, may be understood with reference to Figure 1 3 (A-C).

Section 1 of Figure 13 (A-C) illustrates three general cases where the count in the phase counter is incorrect. In two of the three cases shown (1 and 11 of section 1 of Figure 13 (A C)) the resulting count differs from what it should have been by approximately 3600 counts. These two are each cases where a small change in the relationship between REF and SIGNAL produced an abrupt change in the value registered in the phase counter. In the third case (iii) the count is larger than 3599. This is the only case of the three where the effects of noise and threshold offsets can be explicitly discerned. For the other two cases (i and ii) there is no reliable way (e.g., statistical inference from prior values) to determine if a large or small value is actually beset by the 3600-count error.All that can be stated with assurance is that a sufficiently large or sufficiently small count might be off by approximately 3600 counts.

Section 2 of Figure 13 (A-C) is an example showing that undetected case i or case ii 3600count ambiguities are fatal to averaging.

As shown in section 4 of Figure 13 (A-C), the general solution is to arrange that the phase counter not be used to accumulate measurements for actual use if the count is very small or very large. This is done by first testing the range of the count, and "if necessary", shifting REF by a half-wavelength.

The "if necessary" is the following condition: "Is 4001m < 3200?" where m is the value returned by the phase counter. If the condition is met, then adjustment of REF is "not necessary"; if the condition is not met, then adjust "is necessary". The exact choice of boundaries of the condition are somewhat arbitrary. The values shown work and are convenient.

After REF is offset, the old value of m is discarded and a new one obtained. The new measurement will differ from the previous one by approximately 1800 counts. That is, it will be a "medium-sized" count. It will still exhibit minor variations in value due to noise, and will need to be arithmetically adjusted to take out the 1800count change induced in its value. But after the 1800-count offset is adjusted, the resulting value can successfully be averaged to reduce the effects of the noise; the 3600-count ambiguity that interferes with averaging never appears.

The means used to control the half-wavelength offset of(bREF is the processor's ability to inspect the value returned from the phase counter, and if necessarily, change the value presented by the IREF/IREF mode control latch. This is followed by repeating the measurement; additional measurements may also be made. The processor sets flags in its programming so that it will know how to arithmetically adjust the values of m's subsequently obtained.

As shown in section 5 of Figure 13 (A-C), a newly returned value will need to be decreased by 1800 counts if its original value was small (case ii), or, increased by 1 800 counts if its original value was large (case iii).

Figure 13 (A-C) has assumed that dynamic control of the half-wavelength offset of REF is always used in conjunction with the results of AV ONE's. This need not be, nor is it, the case. The general technique can be applied to measurements made under the auspices of an AV FOUR. In that case the adjustment is by 4x 1800=7200 counts when 1,600 < m < 12,800.

A detailed description of how the half wavelength offset of REF error reduction procedure is incorporated into the overall coordinate measurement and preparation procedure is presented in a subsequent section.

The manner in which the coarse and fine components are combined to produce a coordinate value in a given dimension is explained by Figure 14. Section 1 of Figure 14 shows how the total distance is formed as a sum of a coarse component and a fine component.

Sections 2 and 3 of Figure 14 compare the ways in which the coarse and fine measurements are made, to show that the dimensional weight of a coarse count is 45-times that of a fine count.

This follows from the ratio of their swept wavelengths, and from the fact that each is resolved by the phase counter into a fixed 3600count range. Observe also that each fine count represents a distance of .0111 mm.

Accordingly, and as shown in section 4, a distance corresponding to one entire fine measurement cycle is represented by a coarse count of 80. This follows from the fact that 1/45 of the 360 shifts of a coarse wave is 8 coarse shifts. Since there are 3600 counts/360 shifts, or 10 counts per shift, each fine cycle corresponds to 8 shifts at 10 counts per shift, or 80 coarse counts.

As shown in section 5, what is done is to determine how many times 80-counts is contained in the coarse count. That is, (1) c/80=N+r/80 where: C is the coarse count, N is an integer, r is the remainder.

At this point it would be possible to convert N to an equivalent number of fine counts, add the measured fine count, and take the sum to be the raw dimensional value of the coordinate However, the presence of "noise" in the counts, and the effects of cursor motion, render that approach inadequate for high accuracy digitizing.

An error reduction technique utilizing previously unused resolution in the coarse count solves this problem.

Continuing with section 5 of Figure 14, a statistical argument is advanced to show that random noise is much more likely to produce a significant error in the coarse count than in the fine. That is, one extra count due to noise in the coarse count is much more significant than one extra count due to noise in the fine count. It is also assumed that noise (in counts) is equally likely in both the coarse and fine measurements.

Now, the coarse measurement contains more significance or resolution than has been used by simply finding the highest multiple of eighty that it contains. If this were not so, there would be no remainder r in equation (1) above. That is, the coarse count would exactly indicate, times some integral multiple, the number of fine cycles it represents. The remainder r can be used as a rough predictor of the general size of the fine count accompanying the coarse count. A remainder of 78 (out of a possible 79) does not agree with a fine count of, say, 50 (out of a possible 3599). Under such conditions we are justified in assuming that either noise has absorbed some legitimate coarse counts, or that cursor motion during measurement has changed the fine cycle associated with the coarse measurement.In either case the correct result can be restored by adjusting the coarse value to reflect agreement between the remainder r and the fine value. Section 5 of Figure 14 illustrates graphically the expected correspondence between the remainder r and the fine count.

Section 6 illustrates graphically the method of adjusting the coarse count to ensure "coarse/fine agreement", and is explained below.

Section 6 of Figure 14, in cases A, B, and C, shows that no adjustment to the coarse count is required when any of those three types of agreement occur. Cases D and E iliustrate the required adjustment of the coarse count whens either of the following two conditions occur If a low remainder is accompanied by a igh fine count, the assumption is that the coarse count was made larger by extra counts due to noise, or that there was cursor motion toward smaller values in the dimension under consideration. The correct action in either case is to decrement the value of N obtained in equation (1) above (when the coarse count was divided by 80). This is the correct action in the case of noise, because of the statistical likelihood argument alluded to above.It is the best choice in the case of cursor motion because it provides the answer that would have been obtained had the digitizer taken a coarse measurement at the same time it took the fine measurement. It is, after all, with the fine measurement that the digitizer's accuracy and resolution ultimately reside; to adjust the fine to agree with the coarse would be absurd.

If a high remainder is accompanied by a low fine count, the assumption is that the coarse count was made lower by a loss of counts due to noise, or that there was cursor motion toward larger values in the dimension under consideration. The correct action is to increment the value of N.

Once the coarse count has been adjusted, the total distance (in millimeters) is found this: DISTANCE=(45 xADJUSTED COARSE+fine)x.O1 11.

A complete description of the sequence of values assumed by the mode control latches in performing the coordinate pair measurement activity is given in Figures 1 SA to 1 sod. The first thing that is done is to clear both shift registers.

Next, the x coarse measurement is performed.

This involves an AV ONE to set the half wavelength offset of Q)REF, followed by an AV FOUR. The x reference measurement is performed next, followed byxfine. Each of these two measurements begins with a preliminary AV ONE, to set REF, followed by five AV FOUR's there can actually be more than five, since each AV FOUR can detect the need to change REF. This could happen if the cursor were moving. If such a change were necessary the AV FOUR that revealed the need would be ignored, and repeated after the change. The y measurements are performed in a slightly different order, but in a manner otherwise entirely analogous to the x measurements.

The general order of the measurements is as follows.

CLEAR X COARSE Averaged by 4 X REFERENCE Averaged by 21 X FINE Averaged by 21 CLEAR Y FINE Averaged by 21 Y REFERENCE Averaged by 21 Y COARSE Averaged by 4 Note that this order puts the two fine measurements as close together as possible. That is an optimum choice, as it minimizes the degree to which the coordinates must be corrected to compensate for the effects of cursor motion. The effects of cursor motion are examined in a subsequent section.

The figure shows that DELTA is issued once at the beginning of each different type of measurement coarse, reference, or fine. START is issued once for each AV ONE or AV FOUR to be performed.

The general nature of the activity performed by the processor 14 may be understood with reference to the overview of the processor's activity shown in Figures 1 6A and B. Figure 1 6A shows the general structure of an interrupt service subroutine entered each time the phase counter controller 11 generates DATA READY (recall the DATA READY generates an interrupt).

The main purpose of the interrupt service subroutine is to control the contents of the mode control latches 15, and to generate the "A" and "B" copies of the Sample Table and of the Velocity Correction Timers. The "A" copy of the Sample Table holds the latest data gathered by the measurement sequence of Figure 1 5. The firmware controlled timers necessary for cursor velocity-related correction of the coordinates are active during the time taken to generate the "A" copy of the Sample Table, and are themselves the "A" copy of the Velocity Correction Timers. As soon as the "A" copies are filled, ttey are transferred to the "B" copies for safe keeping and further use; the interrupt service subroutine will begin to refill the "A" copies with fresh data just as soon as possible.If the "B" copies are "busy" (i.e., in use and must not be changed) the newly completed "A" copies are discarded, and the measurement process begins again.

In addition to the "A" and "B" copies of the Sample Table, there are also "C" and "D" copies.

The structure of these various copies is shown in Figure 1 6A. Generally speaking, each copy is derived from the copy associated with the previous letter, and represents another stage in the preparation of the final coordinates. There are no further copies of the timers, however, the "B" copy of the timers is used in the preparation of the "C" copy of the Sample Table, and after that is of no further interest. The structure of the timers is also shown in Figure 1 6A.

A detailed explanation of the operation of the interrupt service subroutine, and of the use to which the Sample Table and the Velocity Correction Timers are put is given in a subsequent section.

Referring now to Figure 1 6B, shown therein is an overview of the processor's mainline firmware activity. The processor pursues this activity whenever it is not executing the interrupt service subroutine of figure 1 6A. The mainline activity shown in Figure 1 6B can be considered as a sequence of very general activities. First, a check is made to see if a new key on the keypad 5 has been pressed. This involves a read-frommemory operation directed to address 168. If there is a new key down it is entered and processed.

Next, a check is made to see if the "B" copies of the Sample Table and Velocity Correction Timers contain fresh data that has not yet been operated upon. If they do, the data is operated upon by a section of firmware called the Basic Point Processor. The Basic Point Processor will be discussed in connection with Figure 18, but its main function is to combine the coarse and fine components so that both coordinates can be expressed in terms of (fine) counts. That is also where velocity-related coordinate correction occurs. At the conclusion of the basic point processor the combined and velocity corrected coordinates are in the "C" copy of the Sample Table.

After the activity of the Basic Point Processor is complete the current digitizing mode (Single or Continuous sampling) is taken into account to decide if the "C" copy of the Sample Table should be saved in the "D" copy for later output to the digitizer's controller.

Next, the mainline firmware attends to any required communication between the digitizer and its controlling device. This communication can take two forms. First, the controller can request that the digitizer supply some type of data. In this case the digitizer sends the contents of a previously loaded and formatted output buffer to the controller. Second, the controller can send commands to the digitizer.

Such commands are two-letter ACSII memonics. The digitizer responds to over forty different such commands, many of which are accompanied by sequences of parameters. The activities of all but a few of a digitizer's commands are outside the scope of this disclosure. Three commands will be briefly discussed so that the general nature of the digitizer's operation with respect to the transmission of coordinate values to the controller may be understood.

The first command of interest is OS (Output Status). This causes the digitizer to put various kinds of status information into the output buffer.

The status information reflects a variety of conditions, such as, "New data is available", "There has been an outof-bounds or low signal error", etc. After sending OS to the digitizer the controller must then request the digitizer to output the contents of the output buffer. That actually transmits the status information to the controller.

The next command of interest is OD. This command instructs the digitizer to format the "D" copy of the Sample Table into the output buffer, in anticipation of the controller's impending request for the transmission of the output buffer's contents. OD is the normal means used to prepare coordinate data for transmission to the controller.

By using the "D" copy the escapement mechanisms associated with Single and Continuous sampling are effective. Note that the "D" copy does not necessarily reflect the current location of the cursor. It reflects the last point the operator thinks was "digitized".

The final command of interest is OC. This command by-passes the escapement mechanisms associated with the Single and continuous modes that regulate the flow of coordinate data into the "D" copy of the sample Table. OC causes the "C" copy of the Sample Table to be formatted into the output buffer, in anticipation of the controller's impending request for the transmission of the outupt buffer's contents. The OC command provides a way for the controller to learn the location of the cursor even though the operator is not "digitizing".

A more detailed understanding of the interrupt service subroutine may be gained with reference to Figures 1 7A to C. Referring now to Figure 1 7A, the first thing done upon an interrupt caused by DATA READY is to capture the data presented by the phase counter 12. This is done with a readfrom-memory operation referencing location 128.

Next, for each AV ONE or AV FOUR measurement, the data (m) is inspected in accordance with the principles outlined in Figure 13 to determine if a half-wavelength offset of REF should be induced. If so, the measurement just obtained is abandoned, the. IREF/IREF control bit complemented, and the measurement repeated by resending the same (except for IREF/iti) mode control latch values as for the measurement just abandoned.

The use of AV FOUR's complicates the task of managing the IREF/IREF control bit. The example in section 2 Figure 13 shows that an AV FOUR cannot, by itself, be relied upon to reveal the actual presence of a 4x3600-count ambiguity.

That is, the result of an AV FOUR is a sum. If one of the counts contributing to the sum is near zero, while all the others are near 3599, the result is not a number "near" 4x3599=14,396. Thus, the presence of the ambiguity remains undetected, which is, of course, not permissible.

The solution is to precede each sing Is AV FOUR or group of consecutive AV FOUR's with an AV ONE, and to impose a velocity limit on cursor motion. The AV ONE correctly reveals the possibility of 3600-count ambiguity; the subsequent AV FOUR's are taken with REF offset, or not, as determined by the AV ONE. Now, assuming that the AV ONE revealed that REF need not be offset, cursor motion will not be so great that, by the end of the next AV FOUR, the cursor will penetrate more than "part way" into the "region of uncertainty".In fact, the "region of uncertainty" (the 400 counts at each end of the measurement interval, as discussed in Figure 13) is sufficiently large that the cursor can penetrate that "part way" without actually incurring the possibility of ambiguity, even if it was just outside that region at the start of the AV FOUR. This gives the AV FOUR the opportunity to anticipate that the next AV FOUR might indeed contain an ambiguity. If that possibility occurs, the AV FOUR just completed "Discards itself", and is repeated with REF offset. And, in the other case, if the preceding AV ONE reveals the need for an immediate offset, once made, the offset provides at least 1 800-400=1 200 counts of "running room" for the cursor to travel in during the subsequent AV FOUR's. That amounts to several times the distance of "part way" mentioned above.The velocity restriction is chosen to work the smaller distance; it is therefore automatically compatible with the larger distance.

Note that an AV FOUR preceded by a known "good" AV FOUR is as "protected" as if it had been preceded by an AV ONE. Coarse measurements comprise an AV ONE followed by a single data gathering AV FOUR, whereas fine and reference measurements comprise an AV ONE (which also gathers data) followed by a series of five data gathering AV FOUR's. Each preceding AV FOUR protects the one that follows it. Thus, the velocity restriction, in conjunction with the knowledge that the previous measurement was wholly outside the "region of uncertainty", combine to render AV FOUR's absolutely reliable, with respect to 3600-count ambiguities.

The velocity restriction referred to above is obtained by the following analysis. First, it is necessary to determine how long a worst case AV FOUR might last. And second, it is necessary to decide how far is "part way" into the 400-count "region of uncertainty". Testing of the actual system revealed that a conservative value for "part-way" would be halfway; that is, 200 counts into the 400-count "region of uncertainty". What this means is that actual region where genuine uncertainty might occur is really only something less than the first 200 or last 200 counts of the measurement interval. Presence of the cursor in the second 200-count interval or in the next to last 200-count interval is used to warn of possible impending cursor presence in the "region of genuine uncertainty".The cursor velocity restriction arises from considering a maximum 200 (fine) counts of cursor motion during the maximum length of time required to perform the measurement involved (an AV FOUR).

Fine counts are chosen above because they represent a shorter distance than 200 coarse counts, and therefore result in a tighter restriction which works for both coarse and fine. 200 fine counts represents a distance of 2.222... mm.

The longest time for an AV FOUR is determined as follows. First, it is noted that no DELTA (refer to the explanation of Figure 5A) ever precedes an AV FOUR. Therefore, the 1.2 msec duration of the one shot 33 need not be added to the time being sought. Next, it is noted that the request for an AV FOUR is totally asynchronous to REF. The phase counter will not actually begin measuring until the leading edge ofREF. So there is at most a 200 Msec (one cycle of 5 kHz) delay on this account.

Next, each of the next three measurements will appear to take a full 200 ysec each, regardless of the actual number of counts accumulated. The fourth measurement could required counting as high as 3199 (3599-400). That is 88.886% of a possible 200 ,usec duration, or 177.78 Msec. For the sake of margin, and to keep the numbers nice, call it 200 usec anyway. Thus, there could be five 200 ,usec intervals, or one msec, required for the completion of an AV FOUR.

Thus, the velocity restriction is 2.222... mm/1 msec=2.222... M/sec, or approximately 87.5 inches/second.

Referring once again to Figure 17A, the flowchart is arranged to drop the returned value associated with an AV ONE performed as part of a coarse measurement. (Referring briefly to Figure 15, it can be seen that such an AV ONE is done only to learn whether or not REF need be offset, and not as part of an accumulation of data. The data is taken by an immedately subsequent AV FOUR).

If the count in the phase counter is usable, a check is then made to see if a present halfwavelength offset ofREF must be accounted for by arithmetically adjusting the measured value.

This is also in accordance with the principles outlined in Figure 13.

Finally, the data is accumulated into an accumulator called the "m accumuiator".

Referring now to Figure 178, in the case that the measurement just completed is an AV FOUR a test is made to see if more AV FOUR's need be done to complete an ongoing "AV 21" pursuant to a fine or reference measurement. If so, then the current collection of settings for the mode control latches is used to repeat the AV FOUR (up to four more times-assuming no changes to REF) until the "AV 21" is fully accumulated. This could require more than four repetitions if cursor motion moved the cursor into a region of 3600-count ambiguity, and one of the AV FOUR's had to be repeated.

At the conclusion of an AV FOUR or of an "AV 21" the accumulated entry (in the m accumulator) is put into the appropriate location of the "A" copy of the sample Table. (No solitary AV ONE's are ever used for data gathering purposes). Then the m accumulator is cleared so that it will be ready for the next measurement activity.

At this point the next measurement activity will involve a different type of measurement than just completed. That is, referring briefly to Figure 15, it can be seen that it shows fourteen different combinations (excluding IREF/IREF) of the mode control latches. What is meant here is that the next one of the fourteen combinations is to be implemented next. Note that this does not necessarily mean a change from x to y n;easurement, or from coarse to fine, etc., although it could. It might only mean a change from using an AV ONE to using a series of AV FOUR's.

Returning now to Figure 1 7B, the next activity is to select the next word from the measurement control table (shown in Figure 1 7C). The measurement control table contains in ROM a representation of each of the fourteen different necessary combinations of the mode control latches. At this point the next entry in the table is merely selected; it will be sent out to the latches sooner or later.

Next, a check is made to see if the entire coordinate measurement cycle is complete. If it is, then the "A" copies of the Sample Table and Velocity Correction Timers are transferred to the "B" copies, unless the "B" copies are busy. If that should be the case, to avoid a deadlock the "A" copies are abandoned and the "B" copies are left alone; the measurement process continues, eventually resulting in new "A" copies. In either event the IREF/IREF control bit is set to indicate IREF, in preparation of the beginning of a new coordinate measurement cycle. Also, at this point the two firmware controlled Velocity Correction Timers (TT and TE) are cleared.

Referring now to Figure 1 7C, the next activity is to extract the newly identified next word in the measurement control table, and "or" into it the current value of the IREF/IREF control latches.

This involves a write-to-memory operation referencing location 128. This actually starts the next measurement activity, unless that activity is CLEAR.

A CLEAR does not involve a measurement by the phase counter. Hence, it does not produce a subsequent DATA READY that will cause re-entry into the interrupt service subroutine to begin the next measurement. Such a situation would cause a deadlock. Hence, the flowchart traps a CLEAR just after it has been issued, and rather than leaving the subroutine, the subroutine is started over as if there had been an interrupt The CLEAR operation is very fast anyway; CLEAR goes directly to the CLR inputs of the shift registers where it is acted upon with out delay.

In the event that the upcoming operation was not a clear the two timers (TT and TE) are updated, before the interrupt service subroutine is terminated to await the next DATA READY.

Updating the timers is accomplished as follows.

The timers are each accumulators that represent the duration of various aspects of measurement. The timer TT represents the total time required to perform the entire measurement cycle, and is updated each time the interrupt service subroutine is used. The other timer TE represents only the time required to perform the y fine measurement.

Two genuine hardware timers could be used to some advantage in implementing TT and TE.

However, other considerations resulted in the use of a single hardware timer that can time an interval of up to approximately 2.25 msec. The name of this hardware timer, as it appears in Figure 17, is TI (Time Interval). TI is the timer 35 described in connection with Figure 5A. Clearing, starting and reading TI involve both write-tomemory and read-from-memory operations referencing location 138.

What is done is this. Each time the interrupt service subroutine is called upon the value in TI is used to augment the contents of the time accumulator TT, and if appropriate, TE also Then TI is cleared and restarted in preparation for its use the next time the interrupt service subroutine is entered.

The specific use to which the time accumulators TT and TE are put is explained in detail in a subsequent section.

The Basic Point Processor, mentioned above in connection with Figure 17, may be understood with reference to Figure 18. That figure is a simplified flowchart of the Basic Point Processor.

The first thing the Basic Point Processor does is average the values in the-"B" copy of the Sample Table. This involves dividing the coarse values by four, and all others by twenty-one.

Next, the x coarse and xfine components, as identified in Figures 7 and 8, are combined. This is done as follows. First, the x reference value is removed from the measured x coarse and measured x fine values, in accordance with the principles explained in connection with Figures 10, 1 and 12. Then the coarse value is adjusted, if necessary, to be in agreement with the fine value, as explained in connection with Figure 14.

At this point the two values are actually combined as shown in Figure 14, and the resulting sum is placed into the appropriate entry of the "C" copy of the Sample Table.

Following the computations for the x dimension, the identical procedures are followed for the values.

The next activity is to correct the x coordinate for the effects of cursor motion during the measurement process that produced the values just used. The formula used to correct the coordinates is explained in connection with Figures 19, 20, and 21 below. Before proceeding to that explanation, however, it should be noted that such correction is only performed by the Basic Point Processor when there is a "good set" of immediately preceding coordinates. That is, velocity correction requires the use of the previous point, and a low signal or other catastrophic event (such as turn-on} must not occur between the current point and the previous one. Accordingly, a check is made to see if the current point is the first point following a low signal condition. If it is, velocity-related correction is not performed.In either case, the new x coordinate is saved for use as the old x during the next pass through the Basic Point Processor.

Referring now to Figure 19, the basic reason why velocity-related coordinate correction is needed may be understood. That reason is that since x and y measurements are made separately, and involve repeated measurements for later averaging, cursor motion can cause the resulting x coordinate to correspond to one location on the line being digitized, while the resulting y coordinate corresponds to another. Taken together, that (x,y) pair does not represent a point on the line being digitized.

Figure 20 is a general description of a solution to this problem. The solution is to correct the x coordinate, according to the cursor's velocity in the x dimension, so that the supplied (x,y) coordinate pair does represent a point on the line being digitized. The solution assumes that the velocities of the cursor in both diensions are constant.

The velocity of the cursor in the x direction is found by dividing the difference between the x values of two consecutively digitized points by the time interval between taking the two points. This time is the value of TT mentioned in connection with Figure 1 7. What TT represents, however, is not the time from the measurement of one x coordinate to the measurement of the next one.

Instead, it is the time required for the entire (x,y) measurement sequence for the point to be corrected. This was easier to implement, and in most cases is sufficiently close to the x-to-x time to be used in its place.

It is recognized that the two times are not, in all cases identical. The interrupt service subroutine's ability two repeat AV ONE's and AV FOUR's may result in differences between the two times, as may the following consideration. Not all AV ONES or AV FOUR's take the same amount of time, even if the shifting of REF is disallowed.

The processor's request for an AV ONE or AV FOUR is totally asynchronous to REF. Thus, there is an unpredictable amount of delay in the start of each measurement, and their is absolutely no guarantee that the length of time for one particular AV ONE is the same as for any other AV ONE. On the average, however, the differences over the nominal six AV ONE's and the twentytwo AV FOUR's required for a complete coordinate measurement, tend to cancel.

Once the x velocity is obtained, it is used to determine how much to adjust the x coordinate.

This is done by noting the time taken to make the y (fine) measurement. For it is during this time, called TE in Figure 17, that the x coordinate and y coordinate is adjusted is found by multiplying TE by the x velocity. The x coordinate is then adjusted by that amount, as shown in Figure 1 8.

A more complete understanding of the relationship between TE and TT may be gained with reference to Figure 21. Shown there is the sequence of measurement activity corresponding to the point whose x coordinate is to be corrected.

TE is taken as the time required to do the yfine measurement only, and not as the entire y measurement time, for the following reason. It is assumed that the cursor velocity in the y direction is not so large that it will have an appreciable effect upon the y coarse measurement. That is, it is only the y fine measurement whose value changes when cursor motion is present, as compared to the same measurement with no cursor motion. The coarse measurement, because of the restriction on y-direction velocity, and because of rounding to an equivalent number of fine cycles (see Figure 14) will not change (except to maintain coarse/fine agreement as described in connection with Figure 14). Now, by adjusting x, we are essentially making x track what variation occurred in y. But the above argument shows that it is only during the y fine measurement item that y varies.Hence the yfine measurement is taken as TE.

As mentioned earlier, the digitizer incorporates a means to correct coordinates for the effects of non-perpendicularity of the x and y grids in the platen. Referring to Figure 22, it may be seen that for any skew (i.e., non-perpendicularity) of the axes, it can be assumed that the y coordinate is correct, and that only the x coordinate need be adjusted. Further, referring to part three of Figure 22, it can be seen that the corrected x coordinate Xc is: Xc=Xm/cosO+YmTan 0 where: Xc is the corrected x, Xmis the measured x, 0 is the skew angle.

The embodiment is as follows. The skew angle is encoded in a 1 6-bit switch register 34 whose value the processor can obtain by doing a readfrommemory operation directed to location 1 18 (see Figure 5A). The encoded value obtained is placed into a floating point representation of the skew angle that is then used by that portion of Figure 1 6B labelled "Coordinate Finalization".

What "Coordinate Finalization" does is employ the first three terms of the appropriate Taylor Series approximations to compute the terms necessary to fine Xc The skew angle, which is determined at the time the platen is assembled and tested, is subsequently encoded into the switch register 34 as shown in Figure 23. The angle is encoded, in radians, as a mantissa portion of a floating point number whose exponent is already chosen.

A summary of the various error reduction procedures incorporating arithmetic or logic performed by the processor is depicted in Figure 24. Some of these error reduction procedures require corresponding hardware (e.g., the halfphase offset of REF or the 16-bit skew angle switch register). Figure 24 shows both the actual sequence used and the logical hierarchy of the procedures. By logical heirarchy is meant the necessary order in which the various procedures must be performed if they are to be valid. All of the error reduction concepts shown in Figure 24 have been previously explained. They are shown collectively in Figure 24 so that their relationships to one another may be better understood.

The physical and electrical nature of the platen incorporates an error reduction technique of its own, namely, the incorporation of redundant lines at the extremes of the platen. The location and physical nature of the redundant lines 1 9 is shown in Figures 3 and 6. The basic reason they are necessary is to solve the non-linearity problem shown in Figure 25.

Figure 25 shows why there is non-linearity at the ends of an N-line active area. What happens is that when the cursor's centre is brought to the edge of the active area of the platen the amplified staircase signal 64 derived from the cursor becomes severely distorted, This produces a spurious phase shift 65 in the filter output 66.

This phase shift induces an error in the resulting count in the phase counter.

The distortion in waveform 64 comes about because of the lack of drive lines under a portion of the cursor. This causes the measurement wave to appear abruptly to the pick-up 24.

A "brute force" solution to the problem of the abrupt appearance of the measurement wave is shown in Figure 26. There, the active area is extended well beyond either end of the desired usable area. Thus, as far as the pick-up 24 can tell, it might as well be in the middle of the active area, even when it is on the edge of the usable area. The coupling capacitance from the pick-up to the driven lines in the grid is sufficient to eliminate all distortion in the cursor signal, and remove the subsequent phase shift from the filter output. The only trouble with this solution is that it is wasteful of circuitry; each of the extra driven lines beyond the edge of usable area must have a shift register cell.

Figure 27 illustrates how the use of redundant lines 19 may be used to gain the performance advantages of the method of Figure 26, without incurring the need for as much extra circuitry. The redundant lines, when used in conjunction with a small number of extra regular lines 67, provide the needed coupling capacitance to keep the distortion in the cursor signal down to an acceptable amount. Compared with Figure 26, however, fewer shift register cells are required; or, for a given number of shift register cells available, the size of the usable area of the platen may be larger.

In the present embodiment the spacing between the adjacent redundant lines is 4 mm, as opposed to the 5 mm spacing between regular drive lines. This was done to save space at the edges of the platen. What is important is the amount of capacitance between the pick-up and the redundant lines, not the spacing between the lines, per se.

While the present invention has been described in connection with a particular preferred embodiment thereof, it will become apparent to those persons skilled in the art that various modifications and applications are possible. It is desired, therefore, that the invention not be limited to the details of construction illustrated and described herein above, and it is intended by the appended claims to cover all modifications which fall within the spirit and scope of the invention.

For example, it is obvious that a digitizer could be constructed that uses the swept wave technique and wherein the x and y shift registers were simultaneously operated at different wavelengths. The resulting composite cursor signal would be separated by two filters, each of whose output would be sent to two separate phase counters, or to a single phase counter that selectively measures the x or y phase count. This would provide simultaneous digitizing of the x and y coordinates. Another obvious extension or modification of this invention is the replacement of the electro-static, or capacitive pick-up, with a magnetic one. This would be possible in the following way. The unterminated end of each of the driven lines in the x and y grids within the platen would be either grounded or connected to a voltage source capable of supplying an appropriate amount of current.The x and y shift registers would then be modified, if necessary, to act as either current sources or current sinks, as appropriate. The essential element of the pick-up would then become a magnetically coupled coil of wire, rather than a capacitively coupled disc.

Figure 28 is a simplified block diagram illustrating one possible way of implementing the alternate magnetic embodiment. The shift register 9 and 10 are now connected to the x drive lines 79 and the y drive lines 80 in the platen through current drivers 73. One current driver is employed for each conductor to be driven in the modified platen 81. Additionally, the drive lines 79 and 80 in the platen are no longer unterminated on the non-driven end. Instead, they are each grounded, so as to serve as a return path for the current supplied by the drivers 73. An equivalent alternate arrangement would be to run those ends to a current source, instead of to ground, and let the current drivers act as ground.

A good diameter for the magnetic pick-up coil 74 would be half the wavelength of the fine measurement wave.

Some additional circuit changes are required to implement the magnetic embodiment, particularly in regards to the reference measurement.

Because of the nature of the magnetic coupling between the pick-up coil 74 and the grids in the platen, simultaneously driving an entire grid with IREF 72 will not induce a signal into the pick-up coil 74. That requires a relative motion between the coil and a gradient in the incident magnetic field produced by the grid, and further, that such gradient exist within the space interior to the circumference of the coil. Simultaneous driving of all lines in a grid will not produce such a gradient.

Reference measurements of the amplifier and filter can still be made, however, by means of an additional signal REF/RUN 75, and additional switching means 76 and 77. What REF/RUN does is cause switch 76 to suspend shifting of the active shift register. This is not absolutely necessary, but is a good practice to reduce noise in the system during the reference measurement.

REF/RUN also causes, via switch 77, the amplifier/filter combination to receive the signal IREF via a matching network 78 consisting primarily of an attenuator, rather than receive the signal from the pick-up. In this way, it is as if the pick-up did have induced into it a reference signal, and the reference measurement can proceed.

Another variation of the invention concerns the method used to combine the coarse and fine components. In the preferred embodiment disclosed above the combining is done as follows.

The coarse count is adjusted for agreement with the fine count, rounded and scaled to an equivalent fine count, and finally added to the fine count, after which the resulting sum is converted to a distance. All the combining arithmetic is done in terms of "counts" of the phase counter. Other embodiments are possible, wherein the coarse and fine components are first converted to either time or distance units before the coarse component is adjusted or combined with the fine component.

As can be appreciated from the foregoing description, a digitizing apparatus according to the present invention can provide for ease of changing the size of the platen without substantially altering the remaining digitizer circuitry or changing the accuracy specification.

Apparatus according to the invention can also provide a digitizing technique incorporating coarse and fine measurements without segmenting the grids in the platen nor requiring extra grids in the platen.

Additionally, apparatus according to the invention can provide a digitizing technique whose associated platen incorporates a means of reducing non-linearity at the extremes of the platen while also minimizing the amount of circuitry required to implement such a nonlinearity reduction.

Furthermore apparatus according to the invention can provide means for correction of errors caused by cursor motion during coordinate measurement, so that further reductions in the amount of circuitry may be achieved by alternate sampling of the x and y coordinates by a substantially common set of measurement circuitry.

An apparatus according to the invention can provide a means for internal compensating for lack of perfect orthogonality between the grids in the platen.

Claims (2)

Claims

1. Apparatus for digitizing the coordinate of a point along a measurement axis on a platen, the apparatus comprising: a platen having a plurality of parallel conductors generally perpendicular to the measurement axis; circuit means for impressing a voltage gradient between selected conductors of the platen and for propagating said voltage gradient across the conductors of the platen at a uniform rate; capacitive pick-up means, positionable over the point on said platen whose coordinate is to be digitized, said capacitive pick-up means being capacitively coupled to the conductors of the platen for sensing the propagated voltage gradient as it is propagated across the conductors proximate to the location of the capacitive pick-up means; measurement means for sensing a first point in time at which said voltage gradient begins to propagate across the conductors of the platen and for sensing a second point in time at which said propagated voltage gradient is sensed by said capacitive pick-up means and for determining the difference between said first and second points in time to obtain a digital representation of the coordinate of the point along the measurement axis on the platen over which said capacitive pick-up means is positioned.

2. Apparatus for digitizing the coordinate of a point along a measurement axis on a platen substantially as hereinbefore described with reference to the accompanying drawings.