Yerukhimovich Victor. Born in 1937 in Kharkov,Ukraina .Worked from the Electronic Engineer to Chief Designer .Ph.D. Computer Science and Signal Processing.With 1991 - in Israel. Worked as the DSP Designer, Consultant.Since July 2002 - on pension.
I have decided to open a blog and to enclose in it that, in my opinion, is of interest.I will begin with the main, from the dissertation “Working out and research of pseudostochastic computers”, defended by me in 1988. After defence has passed more than 20 years (!) And again this theme, in my opinion, became actual in connection with the researches which purpose is creation of the optical computer. Optical computer will allow to provide performance of an order of 10^15 operations in a second, that many times over will exceed performance of modern super computers.Having entered in Google words “optical processing, photonic computer, …”, it is possible to receive enough information on a situation in this area.Now efforts of developers are directed on creation of optical logic which should be integrated further into chips - elements of parallel architecture of optical computer. It is obvious, that at the first stage the manufacturing techniques of optical integrated schemes will be rather expensive in this connection an actual problem, at least the next ten years, construction optical computer is, carrying out demanded functions at comprehensible hardware expenses.In my opinion, one of possibilities of the decision of this problem consists in application of the pseudo-stochastic computing structures presented by me in named above dissertation.The first publications about stochastic computers have appeared in 1967. The name "stochastic" is connected with representation of variables by probabilities of occurrence of impulses in stochastic sequences for the purpose of use of theorems of addition and multiplication of probabilities for performance of arithmetic operations. If sequences are independents and orthogonals ,if so, the gates AND and OR become multipliers and adders, accordingly. Simplicity of equipment, parallelism of structure, noise immunity are doubtless advantages of such computers. However time of calculation of an estimation of probability and calculation error are connected by return mean-square dependence that leads in the stochastic computers to undesirable ratio between accuracy and speed. This lack, and also accumulation of errors at integration have substantially limited wide use of stochastic structures. Nevertheless, till now they find application, in particular, at realisations of neural networks.Having entered in Google words “stochastic computing”, …, it is possible to receive enough information on these subjects.I since 1968 was engaged to the offer dr. A.Shpilberg in studying of possibilities of increase of accuracy of the stochastic computers, by using properties of the shift register with a linear feedback (SRF). In the first and in the subsequent publications about the stochastic computers was examined use SRF for generating of sequences, along with other sources of noise. Authors describe various ways of elimination of the "pseudo-random" connected with periodicity generated SRF of M-sequences, for the purpose of their approach to truly stochastic binary signals, for example, considerably increasing length of the maximum period and using then its separate parts.We had been chose other way. Use of a regularity of generated sequences by us when word length (n) SRF is equal word length of operands was essentially new, opposite approach.All operations should be carried out the same as in the stochastic computers, and the result is received for (2^n -1) cloks, i.e. for maximum period SRF. Process becomes completely determined, i.e.probabilistic estimations are eliminated. In this connection, investigated devices are named "pseudo-stochastic" on purpose to underline use in structures of the stochastic computers of regular "working" sequences. The demonstration of code-sequence, additions, multiplication and calculation of a matrix for n=5 it is possible to find in demofile which can be obtained when you contact me by e-mail victoryeruh@gmail.com
About advantages pseudostochastic processing (РSР) it is possible to read when you contact me by e-mail victoryeruh@gmail.com
The laboratory for the purpose of working out of devices for military systems on the basis of application of methods of pseudo-stochastic calculations has been organized in 1973. During existence of laboratory (1973-1985) have been developed and made: the processor for classification “submarine-the surface ship”, the field analyzer of a spectrum of atmospheric noise for the purpose of revealing of starts of rockets; hardware realisation of BFA(beam forming antenna);FFT, etc. Level of integration of element base was at that time low, therefore such indicators as small hardware expenses and, hence, dimensions, a high noise stability at processing performance in real time with accuracy of an order (0,1-0,4) % - involved customers of military systems. .Since 1986, already in "the peace purposes", series of the pseudostochastic measuring devices has been created .The Ph.D. dissertation “Working out and research of pseudo-stochastic computers”, defended in 1988 was as a matter of fact already doctor's as in it the new direction in computer facilities has been theoretically proved and practically confirmed. All results resulted in the dissertation, were priority and have been protected by copyright invention certificates (22) and publications (15). Full text of the abstract thesis will be provided when you contact me by e-mail victoryeruh@gmail.com;
In 1991 I could leave "secrecy" and was repatriated to Israel. I managed to interest some firms the pseudostochastic methods of calculations. In particular, the device has been developed for processing of signals in systems GPS for which it was required approximately six times less than hardware expenses in comparison with let out. Digital models of devices DSP were offered also: spectral analyzers, digital filters, correlators, devices for image processing. However my offers connected with reduction of element base, have not found practical application. As the basic obstacle was called economic inexpediency of working out of essentially new chips while traditional were rather cheap at degree of integration to one million transistors in one chip.Opinion on "inexpediency" I consider rather disputable. Unfortunately, at me possibilities was not wide to advertise advantage pseudostochastic processors(PSP), one of which (except economic) is possibility of performance of the decision in each step at the expense of the organisation of parallel structure which represents model of a solved problem, and also noise immunity at the expense of not item coding. I believe, what exactly therefore stochastic computing are widely enough used at realisation of matrix forms in neural networks. Application in this case PSP at the set accuracy of an order (2 ^-n) will allow to accelerate computing process in 2^n time. It is necessary to notice also, that pseudostochastic integration is carried out without error accumulation that is rather important, for example, at trajectory measurements. It is obvious, that noted advantages supplement the advantages given by specialisation. Use of programming matrixes ,for example, FPGA, will essentially facilitate a problem of construction of chips in which pseudo-stochastic calculations are realised.Simplicity and economy of construction of individual operators and the structure as a whole can be effectively used to create (in the early stages) optic computers.
I have decided to open a blog and to enclose in it that, in my opinion, is of interest.I will begin with the main, from the dissertation “Working out and research of pseudostochastic computers”, defended by me in 1988. After defence has passed more than 20 years (!) And again this theme, in my opinion, became actual in connection with the researches which purpose is creation of the optical computer. Optical computer will allow to provide performance of an order of 10^15 operations in a second, that many times over will exceed performance of modern super computers.Having entered in Google words “optical processing, photonic computer, …”, it is possible to receive enough information on a situation in this area.Now efforts of developers are directed on creation of optical logic which should be integrated further into chips - elements of parallel architecture of optical computer. It is obvious, that at the first stage the manufacturing techniques of optical integrated schemes will be rather expensive in this connection an actual problem, at least the next ten years, construction optical computer is, carrying out demanded functions at comprehensible hardware expenses.In my opinion, one of possibilities of the decision of this problem consists in application of the pseudo-stochastic computing structures presented by me in named above dissertation.The first publications about stochastic computers have appeared in 1967. The name "stochastic" is connected with representation of variables by probabilities of occurrence of impulses in stochastic sequences for the purpose of use of theorems of addition and multiplication of probabilities for performance of arithmetic operations. If sequences are independents and orthogonals ,if so, the gates AND and OR become multipliers and adders, accordingly. Simplicity of equipment, parallelism of structure, noise immunity are doubtless advantages of such computers. However time of calculation of an estimation of probability and calculation error are connected by return mean-square dependence that leads in the stochastic computers to undesirable ratio between accuracy and speed. This lack, and also accumulation of errors at integration have substantially limited wide use of stochastic structures. Nevertheless, till now they find application, in particular, at realisations of neural networks.Having entered in Google words “stochastic computing”, …, it is possible to receive enough information on these subjects.I since 1968 was engaged to the offer dr. A.Shpilberg in studying of possibilities of increase of accuracy of the stochastic computers, by using properties of the shift register with a linear feedback (SRF). In the first and in the subsequent publications about the stochastic computers was examined use SRF for generating of sequences, along with other sources of noise. Authors describe various ways of elimination of the "pseudo-random" connected with periodicity generated SRF of M-sequences, for the purpose of their approach to truly stochastic binary signals, for example, considerably increasing length of the maximum period and using then its separate parts.We had been chose other way. Use of a regularity of generated sequences by us when word length (n) SRF is equal word length of operands was essentially new, opposite approach.All operations should be carried out the same as in the stochastic computers, and the result is received for (2^n -1) cloks, i.e. for maximum period SRF. Process becomes completely determined, i.e.probabilistic estimations are eliminated. In this connection, investigated devices are named "pseudo-stochastic" on purpose to underline use in structures of the stochastic computers of regular "working" sequences. The demonstration of code-sequence, additions, multiplication and calculation of a matrix for n=5 it is possible to find in demofile which can be obtained when you contact me by e-mail victoryeruh@gmail.com
About advantages pseudostochastic processing (РSР) it is possible to read when you contact me by e-mail victoryeruh@gmail.com
The laboratory for the purpose of working out of devices for military systems on the basis of application of methods of pseudo-stochastic calculations has been organized in 1973. During existence of laboratory (1973-1985) have been developed and made: the processor for classification “submarine-the surface ship”, the field analyzer of a spectrum of atmospheric noise for the purpose of revealing of starts of rockets; hardware realisation of BFA(beam forming antenna);FFT, etc. Level of integration of element base was at that time low, therefore such indicators as small hardware expenses and, hence, dimensions, a high noise stability at processing performance in real time with accuracy of an order (0,1-0,4) % - involved customers of military systems. .Since 1986, already in "the peace purposes", series of the pseudostochastic measuring devices has been created .The Ph.D. dissertation “Working out and research of pseudo-stochastic computers”, defended in 1988 was as a matter of fact already doctor's as in it the new direction in computer facilities has been theoretically proved and practically confirmed. All results resulted in the dissertation, were priority and have been protected by copyright invention certificates (22) and publications (15). Full text of the abstract thesis will be provided when you contact me by e-mail victoryeruh@gmail.com;
In 1991 I could leave "secrecy" and was repatriated to Israel. I managed to interest some firms the pseudostochastic methods of calculations. In particular, the device has been developed for processing of signals in systems GPS for which it was required approximately six times less than hardware expenses in comparison with let out. Digital models of devices DSP were offered also: spectral analyzers, digital filters, correlators, devices for image processing. However my offers connected with reduction of element base, have not found practical application. As the basic obstacle was called economic inexpediency of working out of essentially new chips while traditional were rather cheap at degree of integration to one million transistors in one chip.Opinion on "inexpediency" I consider rather disputable. Unfortunately, at me possibilities was not wide to advertise advantage pseudostochastic processors(PSP), one of which (except economic) is possibility of performance of the decision in each step at the expense of the organisation of parallel structure which represents model of a solved problem, and also noise immunity at the expense of not item coding. I believe, what exactly therefore stochastic computing are widely enough used at realisation of matrix forms in neural networks. Application in this case PSP at the set accuracy of an order (2 ^-n) will allow to accelerate computing process in 2^n time. It is necessary to notice also, that pseudostochastic integration is carried out without error accumulation that is rather important, for example, at trajectory measurements. It is obvious, that noted advantages supplement the advantages given by specialisation. Use of programming matrixes ,for example, FPGA, will essentially facilitate a problem of construction of chips in which pseudo-stochastic calculations are realised.Simplicity and economy of construction of individual operators and the structure as a whole can be effectively used to create (in the early stages) optic computers.
