FI62603C - SPECIALDATAMASKIN FOER BEHANDLING AV STATISTICAL UPPGIFTER - Google Patents

Description

ΓΒΐ m, CUMULATION PUBLICATION £ 0, n, JSa [bj (”> utlAggninqsskrift 62 60 3 C Patent granted 10 01 1933 ^ rktent rioldelat ^ (51) K * .lk.3 / iBt.ci.3 G 06 F 15/36 FINLAND —FINLAND (21) llKMttlhak * mu «-PM« MM (eknlnf 771887 (22) H »k * ml» pWv «—Am6tcnln | * d» | 15.06.77 * * (23) AlkvpiM — GlMglMcadai 15.06.77 ( 41) Interpreters) ulktwk «l · - Graters offumllg

Patent} · Registry OfficesIM (44) NIhtivtIcsISMnon 1 «moonLullulwn date——

Patent * och registeratyreltan 'Ainetunuthfdochuti.tkrifunpubUcurad 30.09 82 (32) (33) (31) fjnrduey «tuollieui — W | trd priorltut (71) Gosudarstvennoe Sojuznoe Konstruktorsko-Tekhnologicheskoe Bjuro, Proektirovaniju Sch (SU) (72) Evgeny Evgenievich Vladimirov, Leningrad, Vladimir Gerasimovich Korchagin, Leningrad, Jury Borisovich Sadomov, Leningrad, Lev Mikhailovich Khokhlov, Leningrad, USSR (SU) (74) Oy Kolster Ab (54) Special computer for statistical data processing - Speciald This invention relates generally to data processing equipment, and more particularly to special digital computers for processing statistical data.

Most practical problems in the study and use of multi-parameter phenomena involve random processes characterized by random changes in certain physical quantities as a function of time.

Random quantities can be described by random functions X (t) or Y (t), which contain a number of implementation cases for these functions, namely x1 (t), x2 (t), ... xi (t) ... x ^ (t) or y1 (t), y2 (t), ... y ^ (t), ... y ^ (t), which represent a series of random numbers from 1 to k.

Today, statistical properties are calculated by statistical analyzers that use digital computational methods and are capable of performing real-time computational functions, or by digital computers specifically designed to calculate the statistical properties of random processes.

2 62603

In general, statistical analyzers are very complex devices with a high hardware redundancy, which increases the manufacturing cost of such analyzers and makes it impossible to use them for "coarse-fine-measurement" cases. Examples of such devices are the Tekelec Aitronic (France) models TE-9300, TE-9400, TE-9450, Uniscope model 7001, SAI COR model SA1-51A-54A, and Hewlett Packard (USA). ) models 3721A and 3729A. In addition, these machines have a limited operating capacity, which limits their use in certain application cases. On the other hand, digital computers are expensive and do not allow real-time computing. Solving statistical processing tasks using an IBM-360 computer, especially correlation and spectrum analysis, takes a long time, up to several hours.

A fast random analyzer for digital random signals is known from SU Inventor Certificate 402,873. This analyzer includes an analog-to-digital converter, a stochastic data rounding unit whose input is connected to the output of the analog-to-digital converter, a dynamic storage memory with one input connected to a stochastic data rounding unit and one output connected to the a comparator having one input connected to the output of the evenly distributed random number generator and the other output of the stochastic data rounding unit and having a second input connected to the corresponding output from the dynamic storage, while the output of the comparator is connected to the input of the stator to the second input and the corresponding input from the dynamic storage memory, an integrator having one input connected connected to the output of the control unit and the output is connected to the input of the digital-to-analog converter while the other input of the control unit is connected to the corresponding AND circuit input, the corresponding AND circuit output and the output of the stochastic binary part, the second input of this part being connected

3 62603

A special digital computer for processing statistical data is also already known in the art (SU inventor certificate 432.509).

The present invention is directed to providing a special purpose digital computer for processing statistical data, which allows the absolute entropy of random variables to be calculated from independent measurements, provides data collection, and reduces the cost of a piece of hardware.

The above and other objects of the present invention can be implemented in a special computer for processing statistical data including a random number generator providing an evenly distributed set of pseudo-random numbers, four stochastic data rounding units for linearly converting a code to a probability channel inputs for the first two stochastic data rounding units connected to respective input data lines, a shift register unit, three receiving registers, the multi-channel inputs of the first two receiving registers being electrically connected to the multi-channel outputs of the corresponding stochastic data rounding units and from the receiving register combined with the first receiving register a multi-channel input connected to a third stochastic data rounding unit and also connected to a multi-channel output in a shift register unit connected to its own multi-channel input, multi-channel outputs of the last two receiving registers and also a computer of the last stochastic data rounding unit includes a data storage memory for collecting and transmitting data in groups, a single-clock period multiplier for stochastically multiplying numbers and from which the inputs are respectively connected to the outputs of the last two stochastic data rounding units, which are connected to a single-clock multiplier input, which is also , the latter multi-channel output being connected to the multi-channel inputs of the first two stochastic data rounding units and the left output wires. According to the invention, the machine includes a quantization step counter for determining the quantization amplitude range from which the multi-channel outputs are connected to the multi-channel inputs of the stochastic data rounding units, a read-only memory for storing harmonic functions, "correlation window" functions and type

η = -Plog2P

the values of the available functions and store the microinstructions, one input of which is connected to the multi-channel output of the data memory, from which the multi-channel input is connected to the multi-channel outputs of the stochastic data rounding units, the electrical connections between the data rounding units is implemented through a clock whose second multi-channel output is connected to the multi-channel inputs of the stochastic data rounding units, the second multi-channel output is connected to a read-only memory multi-channel input, one output is connected to the gate outputs inputs, with the multi-channel input of the adder being read-only the multi-channel output of the memory, the multi-channel output being connected to the multi-channel input of the receiving register and being connected to the output lines, the read-only memory outputs being connected to the random number generator input, the receiving register input and the data input, the clock to the input of the adder.

It is most preferred that the read-only memory includes an input unit for receiving the data address code, decoding and transmitting it, storing constant data of the constant unit, and from which its multi-channel input is connected to the multi-channel output of the internal unit. the multi-channel input is a multi-channel input of a read-only memory, the output for receiving register data being connected from its multi-channel input to the multi-channel output of a standard unit, a microinstruction interpreter from which the multi-channel input is electrically connected to the output register output. In this case, the computer according to the invention also includes address code ports, a data mark decoder from which the multi-channel input is connected to the multi-channel output of the output register, from which the multi-channel output is connected to the multi-channel input of the microinstruction interpreter and the electrical connection between the output register and the microinstruction via both inverted and direct code, respectively, and includes gates for inverted and direct code, respectively, with some inputs from these OR circuits for inverted and direct code being connected to the corresponding output register outputs and having their outputs connected to port inputs for both inverted and direct code, the corresponding direct code ports and the inverted code ports being interconnected and connected to the multi-channel inputs of the microinstruction interpreter and the input unit; in addition, some inputs from the address code ports are connected to the multi-channel inputs of the standard unit, the outputs of these units being connected to the inputs of the respective OR circuits for both inverted and direct code, while other inputs are interconnected and connected to the corresponding read-only memory a multi-channel input to a corresponding input, wherein the other inputs are connected to the combined inputs of the respective ports for both inverted and direct code.

The present invention, implemented on the basis of stochastic data processing technology, makes it possible to substantially increase the computer's ability to process random processes and in particular to calculate their statistical properties, simplify electronic circuitry, reduce hardware costs and reduce computer size. In addition, the collection of data from the increasing flow of data to be processed and stored can be partially solved by using the proposed special purpose computer for statistical data processing. The application of a special purpose computer to the processing of statistical data will substantially increase technological developments in areas of research and technology such as hydrometeorology, geophysics, meteorology, medicine, electronics and physics.

Other objects and advantages of the present invention will become apparent from the following description of a preferred embodiment thereof, read in conjunction with the accompanying drawings, in which: Figure 1 is a block diagram of a special purpose digital computer for processing statistical data in accordance with the present invention; according to the invention.

The special purpose computer for processing statistical data includes a random number generator 1 (Fig. 1), a clock 2 having a multi-channel input 3 connected to the multi-channel output of the generator 1 and stochastic data rounding units 4, 5, 6 and 7. Multi-channel inputs 8 and 9 of the units 4 and 5, respectively, are connected to the input data lines 10 and 11.

This computer also includes a quantization step counter 12, the input 13 of which is connected to the output of clock 2, one multi-channel output is connected to the multi-channel input 14 of the stochastic data rounding unit 4 and another multi-channel output is connected to the multi-channel input 15 of the stochastic data rounding unit 5 and also gate units 16 and 17. Input 18 of gate unit 16 and input 19 of gate unit 17 are connected to the respective outputs of unit 2 while multi-channel inputs 20 and 21 of gate units 16 and 17 are connected to multi-channel outputs of units 4 and 5, respectively.

A special purpose computer for processing statistical data includes receiving registers 22, 23, 24 and a multi-channel input 25 from a receiving register 22 connected to a multi-channel output from unit 16, a multi-channel input 26 from a receiving register 23 connected to a u-Sean channel output from unit 17 and a multi-channel input The 27 receiving registers 24 are connected to the multi-channel input 25 of the unit 22, which includes the shift register 62603 unit 28, the multi-channel input 29 is connected to the multi-channel output of the unit 23 and the multi-channel output of the unit 24, while the multi-channel output of the shift register unit 28 is connected to the multi-channel to the input 29 from the same unit 28 and also to the multi-channel input 27 from the unit 24. The multi-channel output from the unit 24 is also connected to the multi-channel input 30 from the stochastic data rounding unit 7.

The computer further includes a calculator 31 having a u-Sean channel output connected to the output lines 32 and also to the multi-channel input 25 of the unit 22, the one-step multiplier 33 being connected from its input 34 to the output of the unit 6 and the output of the same multiplier 33, the multiplier 33 connected to the output of unit 7 and also to inputs 20 and 21 of multiplier 33 from units 16 and 17, and also to a read only memory 39 having one multi-channel input 40 connected to the multi-channel output of unit 2, the other multi-channel input 41 being connected to the multi-channel output of data memory 36; output lines 42, the multi-channel output being connected to the multi-channel input 43 from the calculator 31 and the output being connected to the generator 1 input 44, unit 2 input 45, receiving register 22 input 46, calculating device 31 input 47, receiving register 23 input 48, unit 28 input 40, battery memory 36 input 50 and receiving register 24 input 51, unit 2 multi-channel output connected to units 4 and 5 multi-channel inputs 14 and 15, unit 6 multi-channel input 52 and unit 7 multi-channel input 53 The second multi-channel input 54 of the unit 6 is connected to the multi-channel output of the receiving register 22.

The read only memory 39 includes an input unit 55 (Fig. 2) including an address register, an address counter and an address interpreter (the address register, the address counter and the interpreter are not shown in the figure). The input unit 55 uses a conventional known circuit.

The read-only memory 39 also includes a standard unit 56, the multi-channel inputs 57 ^, ... 57 ^ of which are connected to the multi-channel output of the input unit 55, the output register being connected from its multi-channel input 59 to the multi-channel output of the unit 56.

input, OR to circuits 60 -... 60 for inverted code, OR

X Ui to circuits 61 ^ ... 61 for direct code, of certain inputs 62T, ... 62 and 63 ,, ... 63 from said OR circuits 60T / ... 60 ^ and 61 ^ ... 61 respectively connected to the output to register 58 to outputs 64T / ... 64, 65 _, ... 65 I ω I ω

The read only memory also includes ports 66 ^, ... 66 for inverted code, ports 67 ^, ... 67 for direct code, and a microinstruction interpreter 68.

The outputs of the gates 66T, ... 66 and 67 ^, ... 67 are connected to each other and are also connected to the multi-channel inputs 69T, ... 69 of the input 69 of the interpreter 68 and also to the input unit 55 of the 1 ω a multi-channel input 70 connected to a multi-channel input 41 of the read only memory 39.

The multi-channel output of the interpreter 68 is the multi-channel output of the read only memory 39.

The read-only memory 39 also includes an information notation for the interpreter device 71, the multi-channel input 72 of which is connected to the multi-channel output of the output register 58 and the multi-channel output of which is connected to another multi-channel input of the interpreter device 68; some of the weather inputs 75 ^, ... 75 are connected to the inputs 57T, ... 57 of the multi-channel input 57 of the unit 56, while the other inputs 76I, ... 76 are connected to each other and only the read-only memory 39 is connected to the multi-channel input, to which the interconnected inputs 77T, ... 77 and 78T, ... 78 of ports 66 ^, ... 66ω and 67 ^, ... 67, respectively, are connected. Others enter- J I ω I ω lot 79τ, 80τ ha 79ω, 80ω from gates 66τ, 67- ,. and 66 x x ix ω and 67, respectively, are connected to the OR inputs of circuits 60T, 6160, and 61).

Inputs 81_, 82T and inputs 81 and 82 II ω J ω OR circuits 60 ^, 61 ^, respectively. and OR circuits 60 and 61 are interconnected and connected to the u outputs of the address code ports 74 ^, ... 74.

The operating principle of the proposed special purpose digital computer for processing statistical data illustrated in Figure 1 is as follows.

Assume now, by way of example, that hydrometeorological data from temperature indicators or water salinity transmitters are available to process this 9 62603 data, it means to calculate its statistical properties, such as: - mathematical expectation value πιχ, - autocorrelation function R (1), ΧΛ - power density of the spectrum 5χ (Ρ), - absolute entropy H (x).

Random variables describing temperature or salinity of water are obtained from transmitters as a series of N random pulses in the form of binary-decimal-coded sequences and are input to a special-purpose digital computer.

All units and circuits on this computer must be reset before operation can begin. Depending on the required calculation accuracy, the length N of the queue to be processed in unit 2 is defined for the statistical properties.

The following data is stored in read-only memory 39: values from the functions cos-γ- · lp; -Pilog P ^, log2 n £, "correlation window" functions and microinstructions.

In order to calculate statistical properties from either random process, a series of random pulses represented by random functions Y (t) or X (t) and represented by e.g. a binary decimal r-digit code are introduced via data lines 10 or 11 to the multi-channel inputs 8 and 9.

Assume now that the binary-decimal r-numeric codes come from the data lines 11 to the q-digit unit 5. At the same time, a series of independent evenly distributed pseudo-random numbers are fed from the output of generator 1 via unit 2 to the multi-channel input 15 of unit 5.

The series of r-digit numbers entered in unit 5 is stochastically rounded in this unit so that it has r-q + 1 digits and is passed through unit 17 and the receiving register 23 to the multi-channel input 29 of unit 28. Thus, the first b digits of the data string N are written to unit 28 using the special purpose computer c cycle.

When calculating the values of the mathematical expectation value m and the auto-correlation function R (1), the first digit

XX

from the output N of the multi-channel output of the unit 28 to the receiving register 22.

Information about the multi-channel output of the receiving register 22 is fed to the multi-channel input 54 of the unit 6, the single-channel input 52 of the unit 62 receiving a series of evenly distributed independent pseudo-random numbers from the output of the generator 1 fed through the unit 2. During the first operating cycle, the data from the output of the unit 6 is written to the data battery memory 36. A synchronous transfer operation is then performed to the data held in the unit 28 and the data battery memory 36. The second value of the number is then input from the multi-channel output of the unit 28 to the receiving register 24. The value of the first number is then imported from the receiving register 22 to the multi-channel input 54 of the unit 6 and the second number is imported from the receiving register 24 to the multi-channel input 30 of the unit 7. upon arrival at units 6 and 7, respectively, they are rounded to r 'bits and sent to inputs 34 and 35 of unit 33, where the unit performs a stochastic multiplication operation and writes the result to data memory 36. This process is repeated C times and all values obtained are stored in data memory 36 The calculation period is repeated depending on the desired resolution and the defined sequence of numbers.

Thus, the operation described above results in the mathematical expectation value m as well as the values of the autocorrelation function R (1)

X XX N * "I

N-1 Σ x. · X. Λ calculation. i = 1 1 11

If the mathematical expectation value m is to be squared, the data stored in the battery memory 36 is applied to the multi-channel inputs 8 and 9 of the units 4 and 5. The data of the multi-channel outputs of the units 4 and 5 are fed to the gate units 16 and 17. 22 and 23 through the multi-channel inputs 25 and 26 and the multi-channel inputs 54 and 30 of the units 6 and 7. The result obtained from the output of the multiplier element 33 is written to the data battery memory 36. The above process is repeated continuously and the number of repetition cycles depends on the required calculation accuracy.

In order to subtract the squared mathematical expectation value-2 X N-1 vo m from the sum expression r-, Σ χ. · Χ. , importing information in the form of a binary code x N-1 i = ix i + 1 62603 proportional to the magnitude rn from the output of unit x 4 through unit 16, receiving register 22, unit 6 and multiplier 33 to input 37 of data memory 36. the data is stochastically subtracted from the values of the ordinates b of the autocorrelation function R (1) stored in the same battery memory 2 36. The values of m are then again written to the data memory 36 via unit 4, unit 16, receiving register 22 and unit 6 for this process. continuing automatically until the required calculation accuracy of the values of the autocorrelation function Ρ.χχ (1) is reached.

Thus, after the transformations described above, the values of the autocorrelation function R (1) are stored in the battery memory 36.

xx Thereafter, obtained from the control signal from the operation synchronization unit 2 and input to the multi-channel input 40 of the read-only memory 39, the value of the "correlation window" function is sent from the multi-channel output of the read-only memory 39 to the multi-channel input 43 in the adder 31. 22 and through unit 6. At the same time, the value of the autocorrelation function R (1) stored in the data battery 36 is input to the input 35 of the single clock multiplier 33 via the unit 5, the gate unit 17, the receiving register 23, the shift register unit 28, the receiving register 24 and the unit 7. the autocorrelation function R (1) and the "correlation window"

XX

tio B.j is multiplied and the input is written to the data battery memory 36. This process is repeated b times for all values of the autocorrelation function R (1) and the values of R (1) are written to the unit 28 using the operation cycle of the special purpose computer C. All results calculated and written to the battery memory 36 are stored in this battery memory.

In order to calculate the power density S (P) of the spectrum,

X

the value of the expression cos (^) * T * p is obtained from the read-only memory 39 via the adder 31, the receiving register 22 and the unit 6 to the input 34 of the multiplier element 33. The autocorrelation function R (1)

XX

the value is input to the input 35 of the multiplier 33 in the same manner.

In this multiplier, the value of the function cos (^) * 1p and the autocorrelation function 62603 κχχΟ) are multiplied and the input is then written from the output of the multiplier 33 to the data memory 36. The calculation accuracy depends on the number of times the coordinate XX of the autocorrelation function R (1) (£) · 1 · ρ. Subsequent values for cos (^) · 1 · ρ are obtained in an identical manner and the process is repeated by performing a multiplication on the second ordinal of the autocorrelation function R (1).

XX

With Naata, etc. This occurs b times and the entire computation period for the spectral power density S (p), the result of which is stored in the data memory 2 of the data x 2, is performed using a 2 * c operating cycle.

In order to calculate the value of absolute entropy H (x) or H (y) in random processes, unit 2 disconnects random number generator 1 from units 4, 5 and 6 and inputs control signals, one of which enters input 13 of quantization step counter 12, while the other two arrive at units 16 and 17, respectively. inputs 18 and 19 and deactivate these units 16 and 17. When calculating the values of absolute entropy H (x) and H (y), the probability xi of the realization case of any i-state of the random process can be estimated as the density at which the values of these measurements fall into corresponding amplitude intervals in the sample sequence N.

A series of random pulses, described by the random functions X (t) or Y (t), are applied to the multi-channel inputs 8 or 9 of units 4 or 5, respectively, with the other multi-channel inputs 14 or 15 of these units receiving signals from the multi-channel outputs of the quantization step counter 12. Unit 4 or 5 and calculator 12 are used to determine the number of amplitude intervals for cases x 1, the values of which can be represented as a binary code. The values of the cases x 1 as a binary code are input from the multi-channel output of the unit 4 or 5 via the multi-channel input 38 to the data storage memory 36, which collects them and counts the number x of them that fall on each amplitude interval. When sampling N from a data set of N = 2, where k is 1, 2 ... w, i.e. an integer exponent for the base of the binary notation, to estimate the probability of measurements of the random pulse values of the division n ^ / N falling in that amplitude range, replace the comma by x ^, which corresponds to the data transfer in the battery 36.

13 62603

Address signals are then generated from the estimated probabilities stored in the battery 36 and serving as the addresses of the read only memory 39. These address signals are sent from the multi-channel output of the battery memory 36 to the multi-channel input 41 of the read-only memory 39. To the large P. log „P. the proportional information according to Equation (6) is input from the multi-channel output of the read-only memory 39 to the multi-channel input 33 of the adder 31 as the first operand. Then, based on the control signal obtained from the unit 2 to the multi-channel input 40 of the read-only memory 39, the value of the correlation factor a is obtained from the multi-channel output of the read-only memory 39. These values of the variable an come through the multi-channel input 43 of the adder 31 as a second operand. In the adder 39, the function P. log-P. the values are calculated by correlating the factor x * 2 l with the factor an and the result of the addition indicating the absolute entropy H (x) is introduced through the multi-channel output of the addition element 31 to the output lines 32 and further to the peripherals. These peripherals also receive other statistical information from the output lines 32, namely the expected value ιηχ, the autocorrelation function κχχΠ) and the spectral power density εχ (ρ), which are stored in the data memory 36.

In this way, a special purpose digital computer for processing statistical data is used to process meteorological data representing a random process, i.e., the above statistical properties are calculated for this data. Knowledge of statistical properties allows reliable short-term and long-term meteorological forecasts to be made.

The read only memory 39 illustrated in Figure 2 operates as follows.

The read-only memory works on the superposition principle. After the a-setting of the address register, calculator and address decoder (not shown) in the input unit 55 and the output register 58, the data address code is sent from the data battery memory 36 via the multi-channel input 41 of the memory memory 39 to the multi-channel input 70 of the input unit 55. The data address code is then input. 55 from the multi-channel output to the multi-channel input 57 of the constant unit 56 and to the inputs 75Ι, ... 75ω of the address code ports 74 ^, ... 74 ^. Population 14 62603

in the unit 56, the information address code is interpreted and the information corresponding to the determined address is imported from the constant unit 56 via the multi-channel input 59 and the output register 58 OR

inputs of circuits 60 -., ... 60, 61T, ... 61 62 -, ... 62, 63 _, ... c I ω 'I' ω I 'ω I

63 ^ for inverted and direct codes, respectively. The control signal f1 (t) is input from the operation synchronization unit 2 via the multi-channel input 40 of the read-only memory 39 to the inputs 76 _, ... 76 of the address code ports 74 -, ... 74 (Fig. 2).

The inputs 77 ^., ... 77 of the inverted code ports 66 ^ ... 66 accept the character f2 (t), the inputs 78 ^, ... 78 receive the direct code address from the ports 57 ^, ... 67ω again receiving the control character f ^ (t). Depending on the address code Y and the variable located in the constant unit 56 of the read only memory 39, depending on the combination of signals [f ^ f3] applied to the ports 74 -, ... of the multi-channel operation of the operation synchronization unit 2 (Fig. 1). 74, 1 (j0 66,., ... 66, 67 ,,, ... 67 (Fig. 2) contain the inputs 69 ^ ... 69 of the multi-channel I ω I ω of the decoder 68 and the input 70 of the input unit 55 variable data Q (t), the code of which is defined by measures for the variables Y and ^, as shown in Table 1.

Table 1 [fI] t (f2l k [<3] * M * 0 0 0 o 0 0 1 Ψ 0 1 0 Ψ 0 1 1 1 1 0 1 Y v Ψ 1 1 0 YV Ψ In this case, the constraints [f, · f2 · f3 V f, · f2 · 1 = 0 (1) is set to [f ^, f 2, * "·

Information about the top numbers of the output register 58 is input through the multi-channel input 72 to the data entry decoder 71 and further through the multi-channel input 73 to the decoder 68 62603, and the information from the multi-channel output of the decoder 68 is input to corresponding units in a special purpose digital computer.

Address code ports 75 ^, ... 74 ^, OR circuits 61j, ... 61 (j and ports 66. ^, ... 66 ^. 67 ^ ... 67 allow logical operations to be performed on the variables Y and Ψ, thus increasing the information capacity of the standard unit 66 without increasing its dimensions, i.e., the computer performs data compression.

The proposed invention allows the computation of random large statistical properties, e.g. the computation of absolute entropy, thus expanding the operating capacity of a computer and its field of application, enabling data compression and reducing its hardware costs.

Claims (2)

62603 16 A special purpose computer for processing statistical data, including random number generators to provide a uniformly distributed set of pseudo-random numbers, four stochastic data rounding units connected to the respective input data lines, the shift register unit, three receiving registers, the multi-channel inputs of the first two receiving registers being electrically connected to the multi-channel outputs of the respective stochastic data rounding units and further the multi-channel input of the first receiving register is connected to the third to a stochastic data rounding unit and are also connected to a multi-channel output in a shift register unit connected to its own multi-channel input, to the multi-channel outputs of the last two receiving registers and also to the multi-channel input of the last stochastic data rounding unit. for collecting and transmitting data in groups, a single-clock multiplier for stochastically multiplying numbers and from which the inputs are respectively connected to the outputs of the last two stochastic data rounding units, which are connected to the output of a single-clock multiplier, which is also connected to the data memory the output being connected to the multi-channel inputs of the first two stochastic data rounding units and to the corresponding output lines, characterized in that this includes a quantization step counter (12) for determining the quantization amplitude range from which the multi-channel outputs are connected to the multi-channel inputs (14, 15, 52, 53) of the stochastic data rounding units (4, 5, 6, 7) 62603, read-only a memory (39) for storing harmonic functions, "correlation window" functions and values of functions of type η = -Plog2P, and for storing microinstructions, one of its inputs (41) being connected to the multi-channel output of the data storage memory (36), from which the multi-channel the input (38) is connected to the multi-channel outputs of the stochastic data rounding units (4, 5), the electrical connection receiving registers (22, 23) to the stochastic data rounding units being implemented via the gate units (16, 17), while the electrical connection is generated by a random number generator (1) and between the rounding units (4, 5, 6, 7) of the stochastic data have been implemented via the clock (2), already n the second multi-channel output is connected to the multi-channel inputs (14, 15, 52, 53) of the stochastic data rounding units (4, 5, 6, 7), the second multi-channel output being connected to the multi-channel input of the read-only memory (39) ( 40), one output being connected to the input (13) of the quantization step counter (12) and the other outputs being connected to the inputs (18, 19) of the gate units (16, 17), the multi-channel input (43) of the adder (31) being connected to a read-only a multi-channel output of a memory (39), the multi-channel output being connected to the multi-channel input (25) of the receiving register (22) and connected to the output wires (32), the read-only memory (39) outputs being connected to the random number generator (1) input (44), to the inputs (46, 48, 51) of the receiving register (22, 23, 24), to the input (45) of the clock (2), a memory (36) at the input (50), an input (49) at the shift register unit (28) and an input (47) at the summing device (31). The special purpose computer for processing statistical data according to claim 1, wherein the read-only memory includes an input unit, 62603 18 for receiving a data address code, decoding and transmitting it, a constant unit for storing permanent data, and connecting its multi-channel input an input to a multi-channel output of a unit, the multi-channel input of which is a multi-channel input of a read-only memory, the output for receiving register information being connected from its multi-channel input to a multi-channel output of a constant unit, a microinstruction interpreter characterized in that it includes ports (74. ^, ... 74) of the address code, a data mark decoder (71), from which a multi-channel input (72) is connected to a multi-channel input register (58) an output, from which the multi-channel output is connected to the multi-channel input of the microinstruction interpreter (68) and that the electrical connection between the output register (58) and the microinstruction interpreter (68) is performed OR circuits (60T, ... 60 and 61T, ... 61 ) for both inverted and direct code I ω, respectively, and includes ports (66T, ... 66, 67 -, ... 67) for inverted and direct code, respectively, for some inputs (62 ^, ... 62 ^, 63j , ... 63) of these OR circuits (60 ^, ... 60, 61T ,. . .61) for inverted and direct code being connected to the corresponding outputs of the output register (58) (64 ^., ... 64 ^, 65 ^, ... 65 ^) and the outputs therefrom being connected to the ports (66T, ... 66 , 67 -, ... 67) to inputs (79T, ... 79, I ω I ω I ω 80τ, ... 80) for both inverted and direct code, from the corresponding direct code ports of output I ω (67- , ... 67) and the inverted code ports (66 _, ... 66) being interconnected and connected to the multi-channel inputs (69, 70) of the microinstruction interpreter (68) and the input unit (55), in addition, some inputs (75, ..., 75) of the address code ports (74T, ... 74) are connected to the multi-channel inputs (57) of the constant unit (56), the outputs of these units being connected to the respective OR circuits (60 -, ... 60, 61 -, ... 61) si-