546 IEEE TRANSACTIONS ON COMPUTERS, JUNE 1970 solved by simply requiring that people avoid printing them. This constraint apparently caused our printers Ino real difficulty, although other reports [17] indicate that people do not always make unbroken characters even when they are required to do so. Final conclusions should wait for experiments on several large data sets. It appears that algorithms T, U, and V operating on binary vectors obtained from con-tours of characters quantized as in Fig. 3 yield recognition accuracies much better than those obtainable using a partially trained optimum classifier, and that if sixpart area division--is used, reasonably good recognition accuracies can be achieved, particularly if simple tests are used to differentiate between commonly confused characters and if contextual constraints are incorporated. These conclusions support those of Bakis et al. [18 ] that curve-following features extract the significant information from handprinting. A Sonic Pen: A Digital Stylus System REFERENCES [1] J. M. Wozencraft and I. M. Jacobs, Principles of Communication Engineering. New York: Wiley, 1965, pp. 425-476. [21 L. Kanal and B. Chandrasekaran, "On the dimensionality and sample size in statistical pattern classification," 1968 Proc. NEC, vol. 24, pp. 2-7. [3] R. Alter, "Utilization of contextual constraints in automatic speech recognition," IEEE Trans. Audio and Electroacoustics, vol. AU-16, pp. 6-11, March 1968. [4] R. 0. Duda and P. E. Hart, "Experiments in the recognition of handprinted text: Part II-Context analysis," 1968 Fall Joint Computer Conf., A FIPS Proc., vol. 33, pt. 2. Washington, D. C.: Thompson, 1968, pp. 1139-1149. [51 J. H. Munson, "The recognition of hand-printed text," in Pattern Recognition, L. N. Kanal, Ed. Washington, D. C.: Thompson, 1968, pp. 109-140. [6] J. Raviv, "Decision making in Markov chains applied to the problem of pattern recognition," IEEE Trans. Information Theory, vol. IT-13, pp. 536-551, October 1967. [7] J. K. Clemens, "Optical character recognition for reading machine applications," Ph.D. dissertation, Dept. of Elec. Engrg., M.I.T., Cambridge, Mass., August 1965. "Optical character recognition for reading machine [8] applications," M.I.T. Electronics Research Lab., Cambridge, Mass., Quart. Progress Rept. 79, pp. 219-227, October 1965. [9] S. J. Mason and J. K. Clemens, "Character recognition in an experimental reading machine for the blind," in Recognizing Patterns, P. A. Kolers and M. Eden, Eds. Cambridge, Mass.: M.I.T. Press, 1968, pp. 156-167. [10] S. J. Mason, F. F. Lee, and D. E. Troxel, "Reading machine for the blind," M.I.T. Electronics Research Lab., Cambridge, Mass., Quart. Progress Rept. 89, pp. 245-248, April 1968. [11] A. L. Knoll, "Experiments with 'characteristic loci' for recognition of handprinted characters," IEEE Trans. Computers (Short Notes), vol. C-18, pp. 366-372, April 1969. [12] J. M. Wozencraft and I. M. Jacobs, Principles of Communication Engineering. New York: Wiley, 1965, pp. 459-460. [131 G. S. Sebestyen, Decision-Making Processes in Pattern Recognition. New York: Macmillan, 1962. [14] I. J. Good, The Estimation of Probabilities, Research Monograph 30. Cambridge, Mass.: M.I.T. Press, 1965. [15] W. H. Highleyman, "The design and analysis of pattern recognition experiments," Bell Sys. Tech. J., vol. 41, pp. 723-744, March 1962. [16] G. F. Hughes, "On the mean accuracy of statistical pattern recognizers," IEEE Trans. Information Theory, vol. IT-14, pp. 55-63, January 1968. [17] M. M. Chodrow, W. A. Bivona, and G. M. Walsh, "A study of handprinted character recognition techniques," Rome Air Development Center, Rome, N. Y., Tech. Rept. RADC-TR65-444, February 1966. [18] R. Bakis, N. M. Herbst, and G. Nagy, "An experimental study of machine recognition of hand-printed numerals," IEEE Trans. Systems Science and Cybernetics, vol. SSC-4, pp. 119-132, July , 1968. A. E. BRENNER AND P. de BRUYNE Abstract-An inexpensive computer graphical input device giving Cartesian coordinates with better than 1-mm resolution is described. The technique works equally well for two-or three-dimensional input variants of the device. Index Terms-Digital stylus, graphical input, sonic pen, three-dimensional input, two-dimensional input. INTRODUCTION Computer graphic displays often require interaction with a human operator which cannot be conveniently handled by a keyboard. The display program may then be written to allow the use of a light pen as an operator control. Although the hardware involved is very simple, the pen searching and tracking technique which is required complicates the computer program and consumes appreciable computer time. When time-sharing computer terminals are used, the available computer time is often insufficient for a live cycling display, and use is made of a storage tube or local display memory. Use of a light pen under these conditions becomes impractical, and a RAND [1] tablet is a natural device to perform such input functions in that it is an active device generating a computer interrupt signal, either periodically or on demand. The time required to store data following each interrupt is very small; hence efficient use may be made of the computer time. The sonic pen system [2] to be described performs functionally as a RAND tablet or the Sylvania data tablet [3], although operationally it is a much simpler device. It has the added advantages of being relatively inexpensive and having the capability of digitizing in three dimensions. Finally, the scheme directly generalizes to a three-dimensional graphical input device. In this sense, although it is similar to the Lincoln Laboratory Lincoln wand [4] in that it uses the propagation time of a sound wave as a measure of distance, the two devices differ substantially. SONIC PEN PRINCIPLE The basic sonic pen digitizer shown in Fig. 1 consists of an orthogonal pair of plane microphones positioned at two edges of the digitized area and a pen or puck capable of generating a short sonic pulse. In use the sonic pen generates a spherical sonic wavefront by causing a small spark to jump across a small gap in the tip of the pen. When the wavefront first reaches the sensing microphone planes, an output is obtained with a rise time of 1 to 2 xs. The distance from the pen to each microphone is measured by counting clock pulses during the transit time of the sound front from the spark gap to the microphone. The limiting accuracy is Manuscript received August 21, 1969; revised December 14, 1969. This work was performed at Harvard University, Cambridge, Mass., and was supported in part by AEC Contract AT(30-1)-2752. A. E. Brenner is with the Department of Physics, Harvard University, Cambridge, Mass. P. de Bruyne is with R.C.A. Laboratories, Burfington, Mass. 547 SHORT NOTES Fig. 1. A two-dimensional version of the sonic pen system. Fig. 2. One possible embodiment of a sonic pen is shown. In this case a ball point pen is used as the stylus and a hard copy sketch may simultaneously be made of the input data. Fig. 3. A three-dimensional version of the sonic pen system. TO COMPUTER -300 v. y STOP x OVERFLOW CLEAR ALL Fig. 4. A block diagram of the overall electronic logic for the sonic pen system. See the text for details. obtain equal maximum x, y, and z dimensions; if only x and y directions are used, shallow rectangular microphones are convenient and improve signal output. The leading surface of the wavefront will be parallel to a microphone over a small area when it impinges and produces an electrical impulse with a rise time of typically 1 ps. As sound travels 0.3 mm/ps in air, a resolution of about 0.2 mm is feasible at distances to about 1 meter. The two-dimensional tablet has a 30- by 30-cm plexiglass HARDWARE DESCRIPTION surface and the microphones, each with a 2.5- by 31-cm Fig. 2 shows one embodiment of the tip of the pen, which sensitive area, are located for convenience at the left and is typically pulsed at a rate of 60 times/second. If a z direc- top edges of the writing area. The frame can be placed on a tion is used, the microphones are typically square planes to table or fixed in front of a large display CRT of a computer. typically ± 0.2 mm, even at distances up to 1 meter, provided drafts can be kept to less than 0.3 meter per second and temperature variations to + 1.0°C. Greater accuracy may possibly be obtained using an automatic calibration technique utilizing a redundant microphone and by using an enclosed temperature-stabilized air space. Experience to date has given better than 1-mm accuracy without any of these special procedures. IEEE TRANSACTIONS ON COMPUTERS, JUNE 1970 548 Calibration is arranged so that in the latter case, the displayed points fall approximately under the tip of the pen. Writing may be carried out on a stack of drawings, a book, or even in space as no physical contact with any surface is required. This device has been in use for over a year with a display driven by a PDP-1 computer. Fig. 3 demonstrates a three-dimensional arrangement. The top plane is used for the z direction. ELECTRONICS The capacitor-type plane microphones consist of 25micron mylar aluminized on one side held against a perforated aluminum backing electrode by electrostatic forces. They are biased with 600 volts and produce an output voltage of typically 1 mV with 1 ,us rise time. A saturating preamplifier of gain 10 000 drives a Schmidt trigger. The output of each channel is connected to the main chassis with coaxial cables. A block schematic diagram is shown in Fig. 4. At the time of the spark, three univibrators are triggered. Two of these produced a read request of about 1 ms starting 1.5 ms after the spark. The third holds all flip-flops in the reset state for about 20 /s until all electrical disturbance from the spark has subsided. This includes overload recovery from the preamplifiers. When the reset state is released, an oscillator of about 1 MHz is allowed to scale into the x and y position scalers. When a preset count position is reached, another one shot is triggered causing a brief 2-,s reset of the x and y scalers. This allows a digitalcontrollable offset to be obtained for the distances between the microphones and the edge positions desired. The use of the scalers for this purpose enables precise and stable positioning even for large distances. After this second reset period, scaling in x and y continues until an output from the preamplifiers causes the gating flip-flops to set. The x and y scalers are stopped individually when the sound arrives and hold the data of position until reset by a following spark. The data is read out to the computer before this happens by the action of the univibrators as described earlier. If the position of the pen is outside the designated field, then a computer interrupt request is inhibited by overflow and underflow logic. Hence, no false indications are possible and only writing in the designated area is accepted. Besides the output from the x, y, and possible z scalers, a function register is read out at each interrupt request. The function switches enter control information to the program allowing for flexible use of the device. It is thus seen that there has been provided a simple, highly efficient arrangement for supplying graphic data to a computer. While only a simple embodiment has been described, other variations will suggest themselves. For instance, specific nonlinear (such as logarithmic) operation may be obtained by changing the frequency of the clock in some programmed way. ACKNOWLEDGMENT Useful discussions during the initial conception of the sonic pen idea were held with C. Zajde. Recent engineering help was performed by A. Sanderson and M. Olmstead. REFERENCES [1] M. R. Davis and T. 0. Ellis, "The RAND tablet: A man-machine graphical communication device," 1964 Fall Joint Computer Conf., AFIPS Proc., vol. 26. Baltimore, Md.: Spartan, 1964, pp. 325-331. (Such a tablet is marketed under the trade name, Grafacon.) [2] Patent pending. Details of the electronic circuitry are available from the authors. [3] J. F. Teixeira and R. P. Sallen, "The Sylvania data tablet: A new approach to graphic data input," 1968 Spring Joint Computer Conf., AFIPS Proc., vol. 32. Washington, D. C.: Thompson, 1968, pp. 315-321. [4] L. G. Roberts, MIT Lincoln Lab. Rept., Lexington, Mass., June 1966. On Information- Lossless Discrete-Time Systems KEITH L. DOTY, MEMBER, IEEE Abstract-The concept of lossless state or system is generalized by the definition of k-losslessness. If k>N(N-1)/2 for an N-state machine, k-lossless implies lossless. The series connection of a set of discrete time systems {Ai}, where Ai is ki-lossless, results in a system which is min{ki}-lossless. This result is a simple verification of the intuitive notion that a noiseless communication channel is only as lossless as the most lossy system in the channel. By means of inputoutput pair analysis, systems with the decomposition property (one of the necessary but not sufficient conditions for linearity) are shown to be information-lossless of finite order m, whenpne state has finite order m. Further, every state must have exactly the same order. For an N-state system with decomposition and zero-state linearity, information-lossless implies information-lossless of finite order m<N. As a consequence, all information-lossless linear sequential machines used as encoders allow decoding to begin after m received symbols, provided that the encoder's starting state is known in advance. Index Terms- Decomposition property, encoders, informationlossless machines, invertible, linear. I. INTRODUCTION A desirable property for a communication system is that the message received be the message sent with as high a degree of certainty as possible. Assuming no noise in the encoder and decoder, this criterion requires that no information be lost because of the encoding or decoding process. In this note we focus our attention on a specific class of "information-lossless," noiseless, finite-state encoders which meet such a requirement [1]. Where possible, statements are generalized to include discrete-time systems having a countable number of states. A system is information-lossless (IL) if, given the initial state, observed response sequence, and the final state, the excitation can always be determined or deciphered. If the Manuscript received November 27, 1968; revised December 19, 1969. Material for this paper was taken from a portion of the author's Ph.D. dissertation, University of California, Berkeley, Calif. The author is with the Department of Electrical Engineering, University of Florida, Gainesville, Fla. 32601.