JP2840156B2 - Digital signal processor for selectively performing CORDIC, division or square root calculations - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a programmable digital signal processor for performing CORDIC, non-restoring division, or non-restoring square root calculation procedures on digital signals. The present invention relates to the processor implemented by a serial signal.

[0002]

2. Description of the Related Art Bit serial digital signal processing is known to be effective in terms of the amount of digital hardware required to perform fixed algorithm operations including multiplication and addition operations. . However, when a programmable algorithm is to be employed, or when a significant amount of memory is included to execute the algorithm, electronics designers, such as in general purpose computers or microprocessors, have to implement bit-serial digital signal processing. Instead, bit-parallel processing has traditionally been used.

[0003]

SUMMARY OF THE INVENTION A data acquisition system for generating digital data for computation receives analog inputs from a plurality of sensors. These analog signals must then be digitized before being used by a computer as a basis for performing their operations.
What is desired is to have each analog-to-digital converter for analog output signals from various sensors within a single inexpensive monolithic integrated circuit, together with some predetermined simple initial processing circuitry. It is to include. Such data acquisition circuits can be constructed using metal oxide semiconductor (MOS) integrated circuit technology and are suitable for applications such as power measurement and internal combustion engine control.

Since the unit cost of monolithic integrated circuits tends to increase with the complexity of digital hardware to some extent, analog-to-digital converters in which the digital hardware involved is economical,
Multiplexers and digital signal processors have been specifically considered by the inventors. Bit-serial multiplexers and processors are particularly economical as digital hardware. And the interconnection of bit-serial signals requires only two lines, one line leading the serial data and the other line leading the timing (parsing) signal. It is. In data acquisition systems, the speed requirements on digital signal processors are often not so demanding, and the relevant bit-serial operation is often sufficient in speed. Oversampling analog-to-digital converters of the sigma-delta type are especially
First-order sigma-delta analog-to-digital converters are economical as digital hardware.

A bit-serial multiplier regularly laid out by a computer on a silicon substrate, known as a silicon compiler, is described by RI Ha.
U.S. Pat. No. 4,860,240 issued to Aug. 22, 1989 by rtley and SE Noujaim, entitled "LOW-LATENC".
Y TWO'S COMPLEMENT BIT-SERIAL MULTIPLIER ". A bit-serial multiplier regularly laid out by a computer on a silicon substrate, known as a silicon compiler, is also described in R.
March 20, 1990 by I. Hartley and PF Corbett
Their US Patent No. 4,910,700 "BIT-SLI"
CED DIGIT SERIAL MULTIPLIER ”; and July 3, 1990
Their US Patent No. 4,939,687 "SERIAL-
PARALLEL MULTIPLIER USINGSERIAL AS WELL AS PARALLE
L ADDITION OF PARTIAL PRODUCTS ”. The bit-serial adders by RI Hartley and PF Corbett are known as silicon compilers, which are regularly aligned by a computer on a silicon substrate, Description 1988
U.S. Patent Application No. 26 filed and granted on October 31
No.5,210 "DIGIT-SERIAL LINEAR COMBINING APPARATUS U
SEFUL IN DIVIDERS (current name) ". A description of an oversampling analog-to-digital converter using a first-order sigma-delta modulator is:
For example, U.S. Pat. No. 4,896,156 "SWITCHED-CAPACITANC" issued on Jan. 23, 1990 by SL Garverick
E COUPLING NETWORKsFOR DIFFERENTIAL-INPUT AMPLIFIE
RS, NOT REQUIRING BALANCED INPUT SIGNALS ”. PL Jacob and SL Garveric
According to U.S. Pat.No. 4,955, issued on August 21, 1990,
1,052 ”CORRECTION OF SYSTEMATIC ERROR IN AN OVER
"SAMPLED ANALOG-TO-DIGITAL CONVERTER" describes the correction of systematic errors in oversampling analog-to-digital converters.
Assigned to y and incorporated here for reference. According to what the inventor has learned,
Some integrated circuits that perform both analog-to-digital conversion and simple data processing require on-chip circuitry that can perform computational procedures other than those performed by the bit-serial multiply-add processor. There is something to say. For example, in an integrated circuit used for power metering, an on-chip digital device capable of performing CORDIC, division, and square root calculation procedures.
It has been found desirable to have a processor. Further, from the viewpoint of comparison with a bit-serial division / addition processor, it is desirable that the compensation processor also operates on a bit-serial signal. The purpose of the present invention is
It is an object of the present invention to provide a bit-serial processor for selectively performing a sequential step of performing a CORDIC, non-restoring division, or non-restoring square root calculation by sequential approximation.

[0006]

SUMMARY OF THE INVENTION The present invention provides one continuous
According to a constant i that increases by one at each step,
Hands with steps that are numbered consecutively, starting with
CORDIC, division, or square root calculation procedure calculated in order
Digital process that selects and executes a series of steps
The digital processor.
Is provided with each signal input port,
First via the signal input port for each temporary storage data
The initial step of the CORDIC procedure
Configured to receive the first vector component during the
The first accumulator, and is temporarily stored and updated.
As the first digital accumulated input obtained, the CORDIC procedure
During each subsequent step, the first accumulation result from the previous step
And the signal input of each temporary storage data
Via the power port as the first digital accumulation input
Receives the dividend input signal during the initial step of the division procedure
The first accumulator configured to:
At each subsequent step of the division procedure,
As the first updated digital accumulation input
Wherein the first dividend is received at the time of the
The first accumulator and the respective temporary storage data
A first digital accumulation input via a signal input port of the
In the initial step of the square root calculation procedure,
Receiving the input signal and performing each step of the square root calculation procedure.
At the time of backup, the temporarily stored and updated first digital
As a calculation input, the first switched number at the previous step is received.
The first accumulator configured to be continuous;
First, temporarily store such digital accumulator results
First accumulator means for performing the respective signal input ports;
Ports and their respective output ports are provided.
Via the signal input port for each temporary storage data
2 as the digital accumulation input, the initial status of the CORDIC procedure
At the step, receiving the second vector component, the CORDIC
During each subsequent step of the procedure, an updated second digital
Receives the second accumulation result of the previous step as accumulation input
Via the signal input port for the temporary storage data.
Then, as a second digital accumulation input,
At the initial step, the divisor input signal is received and the
At each subsequent step in the sequence, the temporarily stored and updated
As the digital accumulation input of 2, the divisor signal of the previous step is
Continuous second digital accumulation configured to receive
Second accumulator means for temporarily storing the result;
Each signal input port and each signal output port
Is provided as the third digital accumulation input.
Zero of any function in the CORDIC, division, square root calculation
At the time of the eye step, temporarily store arithmetic zero, and
Approximate tangent angles for each arc at each step of the IC procedure
Temporarily store the value, at each step of the division procedure,
Temporarily store the approximate value of each of the quotients;
At each step of the calculation procedure, calculate the approximate value of each square root
A third digital device configured to temporarily store
Accumulator means for temporarily storing an accumulation result
And at each step of the CORDIC procedure according to i
Means for generating a tangent radix of the arc to be formed; said CORDIC procedure
And the first accumulation result at each step of the division procedure
Before updating the first
Means for comparing the accumulation result with zero; and the square root
Updating the first accumulation result at each step of the calculation procedure
Before each of the first accumulation results
To compare the trial estimate for the number of actuations
Means; at each step of said CORDIC procedure,
Responsive to the comparator result being both zero or positive,
Making adjustments to the first accumulator means so that the first digital
Subtracts the current second accumulation result from the accumulation result,
As a result, the resulting sum is doubled and the updated first
Means for generating a digital accumulation result; said CORDIC
In each step of the procedure, at least zero or positive
In response to the comparator result, the second accumulating means is adjusted.
The second digital accumulation result and the current first
2 in the digital accumulation result of^{-(i + 1)}Multiplied by
Means for generating a new second digital accumulation result
And at least in each step of the CORDIC procedure
Responsive to the comparator result being zero or positive, the third
To the third means of digital accumulation
Add the tangent radix of the arc determined according to the result and i,
Procedure for generating an updated third digital accumulation result
Steps; less than zero at each step of the CORDIC procedure
Responsive to the comparator result being
With adjustments, the first digital accumulation result and the second accumulation
Double the sum of the results obtained by adding the calculation results and update
Means for generating a first digital accumulation result obtained;
Update to less than zero in each step of the CORDIC procedure
Responding to each of the first digital accumulation results before
The second accumulator is adjusted and its second digital
Is added to the current first digital accumulation result.^{-(i + 1)}
Is decremented by the product multiplied by
Means for generating a result of summation of the totals; said CORDIC procedure
The comparator result being less than zero in each step of
In response to the third accumulation means,
The tangent of the arc obtained from the digital accumulation result of 3 according to i
Generates updated third digital accumulation result by subtracting radix
Means for generating; in each step of said division procedure
At least in response to the comparator result being zero or positive.
In response, the first accumulation means is adjusted and the first accumulation means is adjusted.
From the digital accumulation result, 1 of the second digital accumulation result
/ 2 is subtracted to generate an updated first digital accumulation result.
Means for generating; in each step of said division procedure
At least in response to the comparator result being zero or positive.
In response, the second accumulation means is adjusted and the second accumulation means is adjusted.
2 for the digital accumulation result and the current first digital accumulation result
^{-(i + 1)}Multiplied by 2 and the updated second digital
Means for generating an accumulation result; each step of said division procedure
Said ratio being at least zero or positive in step
Responding to the comparator result and adjusting the third accumulating means.
Calculated according to the third digital accumulation result and i
A third digit updated by adding the tangent radix of said arc
Means for generating an accumulation result; each of the division procedures
Responsive to said comparator result being less than zero,
In response, the first accumulation means is adjusted and the first accumulation means is adjusted.
The result obtained by adding the digital accumulation result and the second accumulation result
Updated the first digital accumulation result by doubling the sum of the results
Means for generating a result; each step of said division procedure
The first data before being updated to less than zero.
Adjusting the second accumulation means in response to the digital accumulation result.
In addition, the second digital accumulation result is stored in the current first data.
2 for digital accumulation result^{-(i + 1)}Is decremented by the product of
To generate an updated second digital accumulation result
Means in each step of the division procedure,
Responsive to the comparator result being full, the third accumulating means
, And from the third digital accumulation result to i
Therefore, the third radix updated by subtracting the tangent radix of the arc obtained is
Means for generating a digital accumulation result of;
In each step of the root calculation procedure, the first digital accumulation
The calculation result is at least a trial estimate for the number of switches
Responsive to the comparator result being equal to or greater than
The first digital accumulation result by adjusting the calculation means
2 ^{-(i + 2)}And the sum of the third accumulation result is subtracted and updated.
Means for Generating a Generated First Digital Accumulation Result
And in each step of the square root calculation procedure, a first
Each digital accumulation result is a trial estimate for the number of switches
Responsive to the comparator result being less than
Adjusting the stage, its first digital accumulation result and 2
^{-(i + 2)}And the sum of the third accumulation result is added and updated.
Means for generating a first digital accumulation result;
In each step of the square root calculation procedure, the first digital
The cumulative total is at least a trial summary
Responding to the comparator result which is equal to or greater than the calculated value,
The third digital accumulation by adjusting the accumulation means of 3
Result and 2^{-( i + 1)}And the updated third digital accumulation
Means for generating a calculation result.

[0007]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. Throughout the accompanying drawings and the description that follows, the symbol Δ without digits is used to indicate a one-word (32-bit) delay operator. Symbol Δn (n is an integer)
Is used to indicate an n-bit delay operator. The symbol S & Hn is used to indicate the sampling and holding operation of the nth bit of each bit-serial word in the bit-serial word stream. In a bit-serial word, the meaning of the symbol 2 ^m (where m is an integer in the range from -31 to 0) is a 32-bit bit string in a bit-serial word string. (31+) in consecutive bits in a serial word
m) th bit is "1" and all other bits are "
0 ".

FIG. 1 is a block diagram showing an embodiment of the present invention. The monolithic integrated circuit 5 in FIG.
It is used to monitor three-phase power mains. The mains provided here are a conductor 1 for the first-phase alternating current into which the primary winding of the current sense transformer 11 is inserted, and a conductor 1 for the current sense transformer 12.
Conductor for second phase alternating current with secondary winding inserted
2, and a conductor 3 for a third-phase alternating current into which the primary winding of the current sense transformer 13 is inserted. The primary winding of the voltage sense transformer 14 is connected to the ground and the first phase conductor 1.
Is connected to sense a voltage between the two. The primary winding of the voltage sense transformer 15 is connected to ground and the second phase conductor.
It is connected to sense the voltage between the two. The primary winding of voltage sense transformer 16 is connected to sense the voltage between ground and third phase conductor 3.

Sigma-delta modulator 2 in integrated circuit 5
1,22 and 23 are current sense transformers 11,12 and 13
The digitalization of the analog current flowing through the main conductors 1, 2 and 3 is performed by digitizing the voltage obtained from the secondary winding of. The sigma-delta modulators 24, 25 and 26 in the integrated circuit 5 digitize the voltage applied from the secondary windings of the voltage sense transformers 14, 15 and 16 and convert the analog voltages appearing on the main conductors 1, 2 and 3 Digital representation of The streams of output bits from the sigma-delta modulators 21-26 respectively flow into the input ports of a 6-channel decimation filter 20. From the output port of this filter 20, a bit serial multiply-add processor 30
(This is described in more detail below), and cyclically supplies the decimated, bit-serialized data as follows: (1) Analog current flowing through main conductor 1, (2) analog voltage on main conductor 1, (3) analog current flowing through main conductor 2, (4) analog voltage on main conductor 2, (5) main conductor 3. And (6) the analog voltage on the main conductor 3.

The multiply-add processor 30 converts the bit-serial output signal into a bank 19 of bit-serial registers.
To supply. The bit serial input signal from the bank 19 can be captured by the CORDIC processor 40.

An analog reference voltage AGND referenced to the voltage supplied from the first end of the secondary winding of the transformer 11-16 to the sigma-delta modulator 21-26 is located within the integrated circuit 5. A generator 18 for this DC voltage is applied to the second ends of these secondary windings. The frequency of the clock signal generated by the clock generator circuit 100 located in the integrated circuit 5 is equal to the frequency of the crystal signal at a frequency of 10 MHz controlled by the crystal 101.
This is the ratio specified for the oscillation of the control-type oscillator (not separately shown). This 10 MHz
Oscillation (frequency) is divided by 1/4 in the clock generator to generate 2.5 MHz. This determines the oversampling rate for the sigma-delta modulators 21-26. The decimation filter 20 updates the data to the multiply-add processor 30 at a word rate of 4.9 kHz. Bank 19 of bit-serial registers updates its bit-serial input signal to CORDIC processor 40 at word rates slightly above 5 Hz.

[0012] Programmable read only memory
What is stored in (PROM) 9 is a multiplication-addition processor
Program instructions and coefficient data for 30 and CORD
These are program instructions for the IC processor 40. PROM
The control circuit 29 is arranged in the integrated circuit 5,
Appropriately pass program instructions and coefficient data to the multiply-add processor 30, and
Pass appropriately to DIC processor 40. PROM 9 is a separate monolithic integrated circuit (I
C), because PROM 9 is preferably of the electrically erasable type and cannot be easily combined with analog CMOS in a single IC. The inventor uses NEC μPD28C04, which is an electrically erasable PROM, as the PROM 9,
It can store 512 bytes and can read at data rates in excess of 2.5 MHz. This PRO
M has 11 address lines and 8 data lines. The coefficient stored in PROM 9 is 16
The two's complement number with the binary point of the bit followed by the most significant bit. Since the interface between the integrated circuits 9 and 5 is 17 bits wide, an appropriate amount of power is required to drive at high speed through this interface.

The parallel input / serial output (PISO) converter 31
Controlled by a ROM control circuit 29 (not shown in detail in FIG. 1), the coefficients stored in PROM 9 in a bit-parallel format are applied to a bit-serial Convert to format. In the power measurement system detailed herein, these invariant coefficients are used as multiplicands in the bit serial multiplier portion of processor 30. And the design of such a bit-serial multiplier makes it appear as if parallel-to-serial conversion is not necessary. However, the bit-serial multiplier layout on the integrated circuit board allows only standard bit-serial designs to accept only bit-serial signals through the interface between the functional elements. As determined by the silicon compiler, and as shown in FIG. 1, a SIPO converter 31 will be used.

(To lay out the program data RAM 6, the inventor used one silicon compiler program. Sigma-delta modulators 21-26.
Are arranged as macrocells, such as pulse width modulators 64 and 66. Another silicon compiler program
((Proceedings of the IEEE, vol. 75, pp. 1272-128
2, Sept. 1987, "A Silicon Compiler for Digital Si
gnal Processing: Methodology, Implementation and App
lications ") (as described by F. Yassa, et al.) was used to lay out the remaining bit-serial part of the circuit on IC5.)

In order to conserve power for the high speed drive of the random access program data memory 6 to the multiply-add processor 30, this program
Data memory 6 is located in integrated circuit 5. Multiplication
-Writing from the addition processor 30 to the memory 6, or
Reading from the memory 6 to the multiplication-addition processor 30 is performed via the RAM control circuit 35. A serial input / parallel output (SIPO) converter 32 is controlled by a RAM control circuit 35 (not shown in detail in FIG. 1) to convert the bit serial program data from the multiply-add processor 30. , Memory 6
To be stored in a bit-parallel format. The parallel input / serial output (PISO) converters 33 and 34
(Not shown in detail in FIG. 1), the first and second bit serial data to a multiply-add processor 30 for the bit-parallel program data stored in the memory 6. Is converted to two words at the same time.

For accurate power measurement, the main conductors 1, 2
A multiply-add processor 30 is required that performs an integral number of cycles of integration on the AC signals on and 3. To detect an integer number of cycles of these AC signals, a processor
For these signals that 30 receives from the decimation filter 27, the high-pass filtered digital values are passed from the multiply-add processor 30 to the zero-cross detector 36.
To be added to The zero-cross detector 36 returns a zero-cross instruction to the processor 30.

The triangular coefficient generator 37 is to exchange data with the processor 30, the processor 30 is sinc ²
Generates a coefficient that defines the kernel of the triangular filter when programmed to perform the low-pass filter operation procedure.

The CORDIC processor 40 receives a bit-serial output signal from the multiply-add processor 30 which is slightly above the 5 Hz update rate decimated by the bank 19 of bit-serial registers. The algorithm of the arctan operation used in the CORDIC processor 40 is described in IRETransaction on Electronic Compu
ters, Vol. EC-8, No. 3, pp. 330-334, Sept, 1959, "The CORDIC Trigonometric Computing Techniqu
e "by JE Volder in a dissertation and a more detailed description is
Described by JS Walther in the article "A United for Elementary Function" in the digest of the puter Conference, 1971, pages 379-385, and was formed to be bit serial. The arctan coefficient for CORDIC processor 40 is permanently stored in read only memory 38 located within integrated circuit 5. In the CORDIC processor 40, non-restoring division and non-restoring square root extraction are performed by an iterative sorting technique. This technique is described in the following literature: New York, Heidelberg and Be.
DES published in 1975 by rlin's Spring-Verlag
In pages 278-301 of the document IGN OF DIGITAL COMPUTERS-An Introduction, HW Gschwind and E.
A general explanation is given in J. McCluskey,
The difference is in bit serial adaptation. Latency through the CORDIC processor 40 is extended to a bit-serial length of two words to accommodate bit-serial operations, with one-word and one-word missing one-word comparison times. Therefore, according to the CORDIC processor 40, the operations at the even number and the odd number are alternately executed. The CORDIC processor 40 can operate the following basic functions. Note that xin and yin
Are input variables selected for the CORDIC processor 40. (A) yin / xin, ( b) yin (1/2), (c) tan -1 (yin / xin), and (d) magnitude (yin, xin) = (yin 2 + xin 2) (1 ^{/ 2)} Normalizer circuit 39 arranged in integrated circuit 5 has C
In cooperation with the ORDIC processor 40, another basic function is calculated. (E) [sign (oa2)] * (cordic_out-L) / M where L and M are CORDIC supplied from PROM 9
This is a constant included in the program instruction. Also, [si
gn (oa2)] is the polarity of the value oa2 stored in the register. The signal called cordic_out is sent to the CORDIC processor 40
, And yet another primitive is generated using the CORDIC processor 40 yin / xin function. The symbol * represents multiplication throughout this specification.

The processing through the CORDIC processor 40 is performed based on time division multiplexing, and two functions can be measured simultaneously. Thus, during a selected time period, the bit serial data of one function, digout1, is loaded into register 60 and applied to a digital meter 61, which is a substantial application. Similarly, during the selected time period, the bits of the other function digout2
The serial data is loaded into register 62 and applied to digital meter 63. Pulse width modulator 64
In response to 1, a pulse of the corresponding duration
To the D'Arsonval meter 65. Similarly, the pulse width modulator 66 responds to digout2 with another D'Arsonval meter 67 with the pulse of the corresponding duration as signal pdm2.
Add to

Shown in more detail in FIG. 2 is the portion of the PROM control circuit 29 and clock generator 100 on the power measurement IC 5 and the connection of the PROM control circuit 29 to the off-chip PROM 9. is there.

The 11-stage master counter in clock generator 100 is clocked at four times the oversampling rate in sigma-delta modulators 21-26.
The two-stage counter 102 included in the 11-stage master counter generates a lower count output composed of two parallel bits.
03 is for counting the overflow bits from the counter 102 and generating a higher count output composed of nine parallel bits. Here, the nine-stage counter 103 is of the up / down type, and the count output composed of nine parallel bits can be selectively set to 2's complement. Such selective two's complementation is performed according to the complement of the least significant bit of the count output of the counter 102, and the complement here is controlled by the logic inverter 104. And is added as a sawtooth signal for the decimation filter 20 of FIG. The filter 20 will be described in further detail with reference to FIG. This optional two's complementation provides a high count output consisting of nine parallel bits and its complement, the rising and falling coefficients of the decimation filter 20 (described in more detail with respect to FIG. 7) for each of the six channels. Used as

Each time those outputs are updated, sigma
The ds_cntl signal, generated by the delta modulators 21-26 at a rate of 2.5 MHz, is applied as a reset signal to the counter 102 to provide the multiplication-addition processor 30 and the sigma-
Synchronization with the delta modulators 21-26 is achieved.

In response to the least significant bit of the count output of the counter 102 being "1", the 9-bit parallel bit latch 105 latches the up-count portion of the count output of the counter 103. The contents of latch 105 are used to address PROM 9 to access programmed instructions to multiply-add processor 30. That is, the nine-stage counter 103 is used as a program counter for addressing the PROM 9 and continuously and cyclically multiply-add processor 30.
Are sequentially read. Latch 10
The four least significant bits temporarily stored in 5 are passed directly to PROM 9 as address bits for the access byte in 16-byte units shown in FIG. Multiplexer 106 selectively transmits the five most significant bits temporarily stored in latch 105 to PROM 9 as address bits in response to a "0" control signal received from AND gate 107. .

According to the fact that both the ctr2 and ctr4 signals are "1", the sixth, eighth and first output addresses of the PROM 9 are output.
The "1" output signal is output from the AND gate 107 only during the fourth and sixteenth periods. The ctr2 and ctr4 signals here correspond to the up-count from the second and fourth stages of the master counter composed of the counters 102 and 103, respectively, when obtained from the 9-bit latch 105. is there. A
In response to the "1" control signal received from the ND gate 107, the multiplexer 106 outputs the count output of the five parallel bits of the counter 108 as an address bit as a PRO bit.
Passed to M9. It is assumed that the PROM 9 has a wait time of 3 bits.

AND gates 109 and 1010 and cout11
pdate_wb (as occurs in the circuit of Figure 11) and cordi
The c_inc signal overflows from the counter 108 to generate "1" as the count value of the five-stage counter 108. The cordic_inc signal is a flag bit included for each program instruction to CORDIC processor 40. And while this flag bit is "1" for all instructions, since the last CORDIC instruction,
The counter 108 becomes "0" and stops. Therefore, CORDI
Until the last instruction of the program for the C processor 40, the counter 103 will increment only once every 2048 counts of the master counter consisting of the counters 102 and 103. update_wb flag is logic inverter
Complemented by 1012, this complement is applied to AND gate 1011 as one of the input signals. The other input signal to the AND gate 1011 is an overflow of the counter 103. If the counter 103 overflows when the update_wb flag is other than "1", AND
Gate 1011 adds COUT11 to zero counter 108.
Reset to count.

The clock cycle "1" indicating the transition of the output signal from the sigma-delta modulator 21-26 is delayed by one bit serial in the clock delay stage 1013, and reset to the two-stage counter 102. Signal, whereby the counters 102 and 103 are synchronized.

The final carry output cout11 of the master counter consisting of counters 102 and 103, which occurs after the count 2047 has been reached, becomes the input of the multiplexer 1015 and is connected to the output of the clock delay stage 1014 and to the bits of the bit serial word. corresponding to 2 ⁰ for outputting a control signal is "1" (bit corresponding to 2 ⁰ 1) input port is selected by. During the next 31 bits during which the control signal is "0", the multiplexer 1015 is in a latching operation and the 1-bit clock delay stage 1014
Select output for input port. The signals from the output port of the clocked delay stage 1014 are shown in FIGS.
Is the cordic_start signal used in. This signal is "1" during the first cycle of the 64 sets of MAP instructions provided to the multiply-add processor 30.

FIG. 3 shows 16-byte program information stored in the NEC μPD28C04 as an example of the PROM 9. According to the circuit of FIG. 2, 16 bytes of stored program information and constant information are continuously extracted at the timing shown in the timing chart of FIG. Reading data from the PROM 9 will be described in more detail with reference to FIG.

The 48-bit latch 90 shown in FIG.
Receives the output signals from the 8-bit latches 91, 92, 93, 94, 96 and 97 as its input signals. (Because the timing of each control signal is controlled independently by the silicon compiler, this 48-bit latch 90 is actually built into the monolithic IC and is not real, but it understands the operation of the IC. Addresses 0 and 8 of the 16 8-bit bytes read sequentially and cyclically from PROM 9 are the 32-bit serial words 2 ^-28 Is latched by the 8-bit latch 91 at the timing shown in FIG. Addresses 1 and 9 of the 16 8-bit bytes read sequentially and cyclically from PROM 9 are latched into 8-bit latches 92 at the timing 2 to ²⁴ of the 32-bit serial word. Is done. Addresses 2 and 10 of the 16 8-bit bytes sequentially and cyclically read from PROM 9 are latched by 8-bit latches 93 at timings ^2-20 of the 32-bit serial word.
Latched. The address 3 and address 11 of the 16 8-bit bytes sequentially and cyclically read from PROM 9 are 8-bit data at the timing ^2-16 of the 32-bit serial word. Latched by latch 94.

Address number 5 in 16 8-bit bytes sequentially and cyclically read from PROM 9 is:
The data is latched by the 8-bit latch 96 at the timing of the latch instruction. Then, at the timing of another latch instruction eight bits later, address 7 in the 16 8-bit bytes sequentially and cyclically read from PROM 9 is latched by 8-bit latch 97. The latch instructions for latches 96 and 97 are the outputs from AND gates 961 and 971, respectively. 32 when both bit-serial word of 2 ^-8 output signal from the timing signals and logic inverter 962 is "1", the output signal of the AND gate 961 becomes "1". 32-bit timing signals and logic inverter 97 2 ⁰ serial word
When both the output signals from 2 are "1", the output signal of the AND gate 971 becomes "1". When ctr5 goes to "0", the output signal from logic inverters 962 and 972 goes to "1". The ctr5 signal here is output to the counters 102 and 10 when latched by the latch 105.
This is the output signal of the fifth count bit (counting from 0) of the counter consisting of three.

It is assumed that 48 bits are temporarily stored in the latch 90. Bits in output signal of latch 90
0-4 is RAM 6 as its read0 address read0_adr
(Not shown in FIG. 2). Bits 5-9 in the output signal of latch 90 are read1 address read1
Added to RAM 6 as _adr. Bits 10-14 in the output signal of latch 90 are the write address
Added to RAM 6 as write_adr. Bit 15 is WR0
This is a flag, and at the timing of "1", an operation of directly writing the read0 data read from the RAM 6 again to the RAM 6 is performed. Bits 16-18 are a 3-bit control signal for the multiplexer 45 shown in FIG. Similarly, bits 19-21 are control signals for multiplexer 46. Bits 22-24 are control signals for multiplexer 47. Bits 25-28 are the load flags FA, OA1, OA2 and NP, respectively. Bit 29 is the XP flag. Bit 30 is the EZ flag. Bit 31 is not used. Bits 32-39 latched from latch 96 to latch 90 are correlated as instruction CC0.
It is passed to the DIC processor 40. Bits 40-47 latched from latch 97 to latch 90 are correlated as instruction CC1.
It is passed to the DIC processor 40.

The address number 4, address number 6, address number 12 and address number 14 of the 16 8-bit bytes sequentially and cyclically read from the PROM 9 are stored in the parallel input / serial output register 95. Loaded in parallel. The register 95 here has 8-bit temporary storage for the coef signal to be applied to the multiply-add processor 30. The data at addresses 4 and 12 in the 16 8-bit bytes read sequentially and cyclically from PROM 9 is written to PISO register 95.
The OR gate 951 outputs "1" at bit serial ^2-12 timing in order to output a command to load the data. This allows the first eight bits of the coef signal to be available to multiplexer 46 (see FIG. 9).
become. To output a command to load the data for sequential and sixth address in the 16 8-bit bytes to be read cyclically and the address number 14 from PROM 9 in PISO register 95, bit-serial 2 ^-4 timing Outputs "1" from the OR gate 951. As a result, the latter eight bits of the coef signal are
Will be available to

The data at address 13 and address 15 in the 16 8-bit bytes sequentially and cyclically read from the PROM 9 are loaded in parallel to the parallel input / serial output register 98. This load operation
This is performed in response to the ctr5 signal which becomes "1" simultaneously with the output signal of the OR gate 982, and in response to the AND gate 981 which outputs "1". Any timing bit-serial 2 ⁰ timing or bit-serial 2 ^-8 in response to become "1", the output signal of the OR gate 982 becomes "1". The bit serial signal from the serial output port of the parallel input / serial output register 98 is applied to the input port of the 16-bit clock delay line 983, but its output port returns to the serial input port of the PISO register 98. To form a circular serial memory loop 984. This circular serial memory loop 984 allows the temporarily stored 8-bit bytes at addresses 13 and 15 from PROM 9 to be read from data read directly from parallel input / serial output register 98. At the same time, they are obtained in parallel.
Two 16-bit pairs of 8-bit bytes at address 13 and address 15 from the PROM 9 are input to the shifter circuit 985 from the parallel input / serial output register 98. This shifter circuit 985 shifts those 16 bits to the higher 16 bits and fills the preceding 16 bits with "0", thereby reducing the L constant used by the normalizer 39. What happens. At the same time, each successive pair of thirteenth and fifteenth 8-bit bytes from PROM 9 goes from delay line 983 to shifter circuit 986. This shifter circuit 986 has 16
It shifts the bits to the higher 16 bits and fills the bits preceding those bits with 16 "0" s, thereby generating the M constant used by the normalizer 39. Thus, L is the constant value in the current CORDIC instruction, and M is the constant value in the CORDIC instruction that circulates and loops around loop 984.
(How to use the constant values L and M in the normalizer 39 will be described in detail later with reference to FIG. 9.)

What is shown in more detail in FIG.
Control circuit 35 and its interconnection to RAM 6, serial input / parallel output register 32, and parallel input / serial output registers 33,34. The connection of the multiply-add processor 30 to the registers 32, 33 and 34 is also shown. R
SIPO register 32 for write input port of AM 6
Is connected to WR0 in the instruction of the multiplication-addition processor 30.
Multiplexer 3 in response to the flag being "0"
It is shown to be done selectively via 51. Said W
When the R0 flag is "0", the multiplexer 351 outputs R
Selects the read0 output that was read two bits ago as its write input to AM 6 and previously stored temporarily in an 8-bit wide parallel bit latch 352.

As shown in the timing diagram for RAM 6 in FIG. 6, in each cycle of the operation of the RAM, between each of the first four consecutive counts of the counter 102, READ0, READ1, and WRITE access is executed sequentially. READ0, READ1 and WRITE
Access is preceded by the first count in four count cycles of counter 102, at which time the RAM
6 is not accessed. The signal output from the counter 103 (in FIG. 2) in response to the overflow instruction from the first stage of the counter 102 in FIG. 2 is a 2-bit, 4-bit, 6-bit, and 8-bit signal on the clock delay line 353. Delayed by time, each logic inverter 354
, A load instruction to PISO register 33 and latch 352, a load instruction to SIPO register 32 and PISO register 34, and a write_enable signal applied to RAM 6 during a WRITE access. A chip_enable signal is applied to RAM 6 from AND gate 354 during each of the READ0, READ1 and WRITE accesses. The two-input AND gate 355 outputs a "1" signal at the same time as the output signal from the logic inverter 354.
In response to the ctr0 signal being applied as one of the inverter inputs whose input signal goes to "1", it has a duration of "1" for two bits after the ctr3 signal goes to "0".

The multiplexer 356 outputs c from the counter 102.
Controlled by the tr2 signal, the ctr2 signal is delayed by two bits on clock delay line 357. Two bits later, c from counter 102
The tr2 signal becomes "1", and the multiplexer 356 outputs the latch 90
Select the 5-bit write_adr signal from the
RAM with ctr3 and ctr4 outputs taken from 103
Entered in 6. The multiplexer 356 becomes “0” in response to the ctr2 signal delayed by 2 bits on the clock delay line 357, selects the 5-bit width output signal from the multiplexer 358, and
Along with the ctr3 and ctr4 outputs from 103, they are input to RAM6 as a 7-bit (read) address. The multiplexer 358 outputs ctr1 from the counter 102 which is "1".
In response to the signal, the 5-bit read0_adr from latch 90
Select a signal for its output port. Multiplexer 358
Selects the 5-bit read1_adr signal from latch 90 for its output port in response to the ctr1 signal from counter 102 being "0".

FIG. 7 shows the 6-channel decimation filter 20 in more detail. Sigma-delta modulator 21 (Figure
The output sample ds-i ₁ from 1) and the output sample ds-v ₁ from the sigma-delta modulator 24 (FIG. 1) are decimat
The time is multiplexed and filtered by the ion filter channel 201 in a time division manner. This decimatio
n Multiplexer 2010 included in filter channel 201 provides an output from sigma-delta modulator 21 in response to a count output ctr1 that changes the upper bit output from two-stage counter 102 (of FIG. 2) more slowly. Sample ds
-i ₁ and output sample from sigma-delta modulator 24 ds-
v Choose between ₁ .

The selected signal is provided from multiplexer 2010 as a bit-serial multiplication signal to multiplier 2011. Another input of the multiplier 2011 is (of Figure 2)
This is a sawtooth wave of the bit-parallel multiplicand signal from the 9-stage counter 103. The multiplier 2011 includes a plurality of bits connected to each bit of the sawtooth signal as a first input signal of each AND gate and an output signal of the multiplexer 2010 as a second input signal of each AND gate. Consists of an AND gate. When the lower bit output ctr0 of the two-stage counter 102 is "1", this sawtooth signal corresponds to a positive up counter from the nine-stage counter 103.
On the other hand, the count output from the two-stage counter 102 ctr
When 0 is "0", this sawtooth signal corresponds to a negative down counter from the 9-stage counter 103. As a result, one input signal sample is time-division-processed, and two accumulation operations of the rising mode and the falling mode of the triangular filter are facilitated.

The cumulative operation d using the descending filter constant is performed.
ds-i, which is the cumulative operation of si ₁ with the ascending filter constant
Accumulated d with ₁ and falling filter constant
the sv _1, ds-v ₁ to be accumulated operation at elevated filter constant
The parallel bit output of the
From 2011, it is added to the parallel bit adder 2012 as its addend. The sum output from digital adder 2012 is input to cascade word latches 2013, 2014, 2015 and 2016. These word latches 2013, 2014,
2015 and 2016 are counters of counter 102 (of Figure 2)
The contents are sequentially transmitted forward by the clock. During accumulation by the time division switching method including digital adder 2012, the output signal from latch 2016
By 17 and 2018, it is selected as its augend input signal to adder 2012.

The "1" of cout11 indicating the overflow of the 9-bit counter is used as a load instruction to the parallel input / serial output stage 2019, and the contents of the latch 2014 are
That is, the filtering output of ds-v ₁ sample consisting of the complete triangular filter coefficient is output to the parallel input / serial output stage 2019.
It is taken in. Similarly, when cout11 indicating the overflow of the 9-bit counter becomes "1", the parallel input
Also used as another load instruction for the serial output stage 20110, the contents of the latch 2016, i.e., a convolution of ds-i _one sample consisting of the complete triangular filter coefficients, is applied to the parallel input / serial output stage 20110. It is captured.

In the parallel input / serial output stage 2001, the overflow instruction cout11 from the counter 103 (FIG. 2) is "1" from its parallel input port, that is, "1" indicating that the maximum count has been reached. At this timing, the parallel bit signal 0101 is loaded. A "0" input is applied to the serial bit input port of the PISO register 2001. After the overflow instruction cout11 returns to "0", P is reset until this overflow instruction cout11 next becomes "1".
ISO register 2001 outputs a 1010 bit sequence followed by "0". The output signal from the PISO register 2001 controls the multiplexer 2017.
The output signal from PISO register 2001 is delayed by one clock in one bit clock delay 2002 to generate a control signal to multiplexer 2017.

In the first clock cycle following the overflow indication cout11 being "1", the latch 201
The intermediate result of ds-i1 that has been cumulatively calculated with the rising filter coefficient stored in 5 is selected by the multiplexer 2017 for the augend input of the adder 2012, and is cumulatively calculated with the falling filter coefficient. It is added to -i1 in advance. In the second clock cycle following the overflow indication cout11 being "1", the multiplexer 2018
As a result, an arithmetic zero is selected for the input port of the adder 2012, and the accumulation operation of ds-i1 is started with the ascending filter coefficient. In the third clock cycle following the overflow indication cout11 being "1", the latch 2015
Cumulatively calculated with the ascending filter coefficient stored in
The intermediate result of ds-v1 is selected by the multiplexer 2017 with respect to the augend input of the adder 2012, and is added in advance to ds-v1 that is cumulatively calculated with the falling filter coefficient. In the fourth clock cycle following the overflow indication cout11 being "1", the multiplexer 201
According to 8, an arithmetic zero is again selected for the input port of the adder 2012 and the accumulation operation of ds-v1 with the ascending filter coefficient is started.

The output sample ds-i ₂ from the sigma delta modulator 22 and the output sample ds-v ₂ from the sigma delta modulator 25 are filtered by the decimal filter 202 by a time division switching method. decimation
Elements 2020-2029 and 20210 in filter 202 are dec
Elements 2010-2019 and 2011 in imation filter 201
0 corresponds to each. Sigma-delta modulator 23
The output sample ds-i ₃ from the sigma-delta modulator 26 and the output sample ds-v ₃ from the sigma-delta modulator 26 are filtered by the decision filter 203 based on time division multiplexing. Elements 2030-2039 and 20310 in decimation filter 203 are the same as elements 2010 in decimation filter 201.
-Supports 2019 and 20110 respectively. Multiplexers 2027 and 2037 are controlled by an output signal from PISO register 2001, similar to multiplexer 2017. Also, multiplexers 2028 and 2038
Is controlled by an output signal from a latch 2002 that is clocked, similar to the multiplexer 2018. The parallel input / serial output stages 2039, 2029 and 2019 are connected to a bit serial memory loop 204. The parallel input / serial output stages 20310, 20210 and 20110 are connected to another loop 205 of bit serial memory. Multiplexer 206 responds to the vi-select bit in the instruction of multiply-add processor 30, such as provided by PROM 9, for reading to circuit 207 from one of bit serial memory loops 204 and 205.
Select a 16-bit chunk. The circuit 207 here has 16
A "0" precedes these chunks to form a 32-bit data
A word is formed and taken as its addend input signal of the bit-serial adder 208. The adder 208 adds the bit serial 2 to the power of 0 as its augend input, and adds the received one's complement data as its addend input signal to the signed 32-bit complement 32-bit data. Convert to word.

FIG. 8 shows the multiply-add processor 30 in more detail. The select_R ₁ signal, enable_R ₁ signal, and XP signal are each called “logic signals”. The "logic signal" is normally used for control purposes, and is 32 bit cycles in the case of power meter IC5 and is constant during the bit serial word (BSL). The meaning of the constant 2 ^-15 is a bit
Only 16 is set high as a "1" and the other bits indicate a bit serial word with a "0".

The two inputs of the bit serial multiplier 301 are the bit serial coef_in
And the output of the two-input multiplexer 302 as a bit-serial multiplier signal. This two-input multiplexer 302
In the data_in signal bit-serial at one of its input ports, the left shifter 303 2 ¹⁶
These are the two inputs of the doubled data_in signal. In the left shifter 303, the 16 binary positions are shifted to higher positions, and the lower bits are filled with zeros. The bit-serial product output of the bit-serial multiplier 301 is ANDed as an addend input signal to the bit-serial adder 305.
Applied through gate 304. The bit serial difference output signal from the bit serial subtractor 306 is applied to the bit serial adder 305 as a bit serial augend input signal via an AND gate 307. The meaning of "1" output from the NAND gate 308 means that in the double precision operation, except when the logic signal XP is high and at the same time the bit 16 is high, the AND gate 30
The signal at the output port 4 is the bit
The signal at the output port of AND gate 307 is equal to the serial product and the bit serial difference output signal from subtractor 306. The logic signal XP here is
It is a part of a program instruction for the multiplication-addition processor 30. The select_R ₁ signal applied to multiplexer 309 is a "1", and the bit output from two-input multiplexer 309 is selected except when the output of AND gate 310 is selected.
The serial acc (umulator) _in signal is added to the subtractor 306 as the minuend.

Addition of bit serial from adder 305
The force R0 is the bit serial signal R ₁ = ΔR0
Input signal to the 1-word long delay line 311
It is. Bit serial from delay line 311
Sum output signal R₁ Is the two inputs to AND gate 310
One of the signals, the other input is enable_R₁ No
This is a magic signal. Bits from delay line 311
・ Serial sum output signal R₁ Is the bit serial signal
No.R_Two = ΔR₁ 1-word delay line
Input signal to the input terminal 312. Delay line 312
Bit serial output signal R_Two Is the bit
Real signal R_Three = ΔR_Two One word long
This is an input signal to the ray line 311.

The bit serial sum output signal R ₀ from adder 305 is applied to one input port of AND gate 314, the other input port being an XP logic signal. The sample and hold circuit 315 supplied from the output port of the AND gate 314 outputs a logic signal according to the bit 16 of the input signal. Here, the least significant bit is bit 0 (LSB first). The logic signal from the output port of the AND gate 314 is delayed by one bit serial word to produce a bit serial subtractor 30.
Added to the 6 decrement input ports.

Normally (all of the input logic signals are set to "0"), the multiplication-addition processor 30
Is multiplied by the data_in signal and the coef_in signal to generate a product, which is added to the acc_in signal. The bit-serial multiplier 301 is composed of 16 coefficient bit slices, and in the multiplication, only the upper 16 bits are used for the coefficient term. The output from the bit serial multiplier is 32
A bit-serial word of bits, with a data_in word of 32 bits per word and a coef_in signal of 16 bits per word (both of these signals are signed numbers in 2's complement representation) Upper 47-bit product
Equals 32 bits. The lower 15 bits of this product are
Discarded in the internal circuit of serial multiplier 301.

In the double precision mode, during the first word of a continuous two-word operation, the flag signal XP, which is used as part of a program instruction to the processor 30, is set high. The signals data_in and coef_in are
Enter the same value between two word operations. The signal acc_in and the output of the multiplier correspond to the lowest (first word) part and the highest (second word) part in a two-word length. When XP is high, it is the data input to the multiplier 301 selected by the multiplexer 302.
The data_in is pre-shifted to the left by 16 bits by the left shifter 303 so that the lower bits are filled with zero. Therefore, during the first word operation in the double precision operation, the output of the multiplier 301 is
Equivalent to discarding the upper 16 bits of a 47-bit full product instead of discarding 5 bits. However, since the upper 16 bits are inherently related to the value of the upper 16 bits of the signal data_in, only the lower 16 bits are valid. To summarize, the multiplier 301 during the first word
In the output, the lower 16 bits are valid and the upper 16 bits are invalid.

[0050] The output of the multiplier is when added to the least significant word AND gates 304 and 307 are actuated from Acc_in, since XP and 2 ^-15 bit is high at the same time, the output signal of the NAND gate 308 is " In response to "0", bit 16 of both the addend input and the augend input of adder 305 is masked. Thus, bit 16 of the sum output of adder 305 is equal to the carry bit from the addition of the lower 16 bits (ie, the valid portion of the word). This bit is sampled and held by the sample and hold circuit 315 to generate a logic signal test_out, which is delayed one word in a delay 316 to generate a logic signal acc_bak for use in the second word period.

In the second word period in double precision, the logic signal XP is zero. In fact, there is no clear difference between simple precision operations and second word operations in double precision. As mentioned earlier, data_in from the previous word
And the acc_in input signal should be repeated and the high word acc_in in the second word should be applied. When the lower part of the second word operation produces a carry, the logic signal acc_bak goes high during the operation of the upper part in double precision. This logic signal acc_bak is subtracted from acc_in. Since the high logic signal has a two's complement value of -1, the carry from the lower word will increment the upper word. A logic signal that is "0" has a two's complement value of zero. Thus, when there is no carry from the lower word, the second word is not affected.

The net effect of the double precision operation is that the full 47 bit product of the data_in and coef_in signals is preserved during accumulation. Since the least significant bit of the lower word and the upper 16 bits have no significance, the dynamic range of the accumulator is also 47
Is a bit.

FIG. 9 shows the connection of the multiply-add processor 30 to other circuits in more detail. The eight input multiplexer 45 selects the data_in signal to be applied to the multiply-add processor 30. Bit serial
The data_in signal can be selected from the following. (000) read_0 signal applied from RAM 6 via PISO register 33, (001) read_1 signal applied from RAM 6 via PISO register 34, (010) output signal from multiply-add processor 30, (011) The difference signal between the read_1 signal and the read_0 signal from the subtractor 51, (100) the output signal of the decimation filter 20, (101) the output signal of the decimation filter 20, (110) the register 192 in the register bank 19 (FIG. 1) And a (111) AND gate 50 zero crossing indication.

The multiplication is performed by another 8-input multiplexer 46.
The coef_in signal to be applied to summing processor 30 is selected. The bit serial coef_in signal can be selected from the following. (000) read_0 signal applied from RAM 6 via PISO register 33, (001) read_1 signal applied from RAM 6 via PISO register 34, (010) output signal from multiply-add processor 30, (011) The difference signal between the read_1 signal and the read_0 signal from the subtractor 51, the relatively high-frequency si from the (100) triangular coefficient generator 37
nc ² filter coefficient βnb, (101) relatively low frequency si from triangular coefficient generator 37
nc ² filter coefficient β wb, (110) coef signal applied from RAM 9 via PISO register 95, and (111) multiply-add processor obtained from adder 52
The sum of the output signal of 30 and the coef signal supplied from RAM 9 via the PISO register 95.

A four-input multiplexer 47 selects the acc_in signal to be applied to the multiply-add processor 30. The bit serial acc_in signal can be selected from the following. (00) read_0 signal applied from RAM 6 via PISO register 33, (01) read_1 signal applied from RAM 6 via PISO register 34, (10) output signal from multiply-add processor 30, and (11) The difference between the output signal of the multiply-add processor 30 and the read_0 signal applied from the RAM 6 via the PISO register 33, as determined by the subtractor 53.

In FIG. 16, the multiplication-addition processor 30
Bit serial register 19, which is a component of the register bank 19 located between the
1-196 is shown. Each of these registers 191-196 is a 32-bit bit serial memory, where the stored bits are available cyclically and read / write is bit-by-bit.

The bit serial register 191 is shown in FIG.
) Is written from the multiply-add processor 30 to the register 191 and stored as _a result of the low pass filter operation accumulated over frequency. As can be seen in Figure 16, when the 100 input of the multiplexer 401 is selected, f _a is the xin input signal for the CORDIC processor 40. And the multiplexer 402
When the 100 input is selected, f _a is the yin input signal for the CORDIC processor 40.

The bit-serial register 192 stores f, the normalized signal frequency (8f _in / f _s ), as written to the register 192 from the CORDIC processor 40 (in FIG. 16) during frequency updates. Used to The contents of this bit-serial register 192 are
Multiplexer 45 (in FIG. 9) allows reading and selection and is one of the data_in signals to multiply-add processor 30 (also in FIG. 9). As can be seen in FIG. 16, when the 101 input of multiplexer 401 is selected, f
Xin input signal to And a multiplexer
When the 101 input of 402 is selected, f becomes the yin input signal to the CORDIC processor 40.

The bit-serial register 193 is shown in FIG.
Multiplication-addition processor 30 (as shown in 6)
Is used to store n _p , the number of periods used in the kernel for low pass filtering for multiply-add processor 30, as written to register 193 from. Register from output of processor 30
The n _p loaded into 193 is responsive to the NP flag appearing in the current instruction to the processor 30. As can be seen in FIG. 16, when the 111 input of multiplexer 401 is selected, n _p
It becomes an xin input signal to the C processor 40. And
When the 111 input of multiplexer 402 is selected,
np is the yin input signal to the CORDIC processor 40.

[0060] Bit-serial register 194 is used to store the n _s. This n _s is divided by f
Equivalent to n _p and the operation is performed by CORDIC processor 40. And CORDIC processor 4
Writing from 0 to register 194 is done during operation initiation or updating (as seen in FIG. 16).
(Sometimes, especially in the drawings, "n _s" may appear as "ns".) The contents of (11 in) bit-serial register 194 is triangular filter coefficients generator TRI
Added to NC37. As can be seen in FIG. 16, when the 110 input of multiplexer 401 is selected, 6
In the bit shifter 1910, n _s multiplied by 2 to the sixth power is CORD
It becomes the xin input signal to the IC processor 40. When the 110 input of the multiplexer 402 is selected, n _s multiplied by ²⁶ in the 6-bit shifter 1910 is
It becomes a yin input signal to the DIC processor 40.

[0061] With reference to FIG. 9, the bit-serial register 195, multiplier - a low pass filter processing result of accumulating a signal obtained from the addition processor 30 o _a1
Is used to store As can be seen in FIG. 16, when the 000 input of the multiplexer 401 is selected, o _a1 is the xi to the CORDIC processor 40
n input signals. When the 000 input of the multiplexer 402 is selected, o _a1 is set to the CORDIC processor.
Yin input signal for 40.

Referring to FIG. 9, a bit serial register 196 is a low pass filter processing result oa2 that accumulates the signal obtained from the multiplication-addition processor 30.
Is used to store As can be seen in FIG. 16, when the 001 input of multiplexer 401 is selected, oa2 is xi to CORDIC processor 40.
n Input signal. When the 001 input of the multiplexer 402 is selected, oa2 is set to the CORDIC processor 4
Yin input signal for 0.

Returning to FIG. 9 only, a two-input AND
Gate 48 has one of its inputs as multiply-add processor 30
Output, and the output port of 3-input NOR gate 49 is "1"
If so, write the output to RAM 6. Register bank that is the interface between processors 30 and 40
When no load instruction is present in any of the 19 bit serial registers 191, 195 and 196, NOR gate 49
Outputs "1". NOR gate 49 outputs "0" when there is a load instruction applied to any of bit serial registers 191, 195 and 196. At this time, RAM 6
In, arithmetic zero will be written to the addressed location.

As will be described in more detail with respect to FIGS. 11 and 12, the u generated from triangular filter coefficient generator 37
pdate_wb signal, every time the number of cycles of the filter coefficients 2n _s is completed, generates a pulse to be "1" normally The "0" state. Further, Update_nb signal, every time the number of cycles of the filter coefficients 8n _s is completed, normall
y Generate a pulse from "0" to "1". u
The pdate_wb and update_nb signals are output at the delay lines 54 and 55, respectively, from the output R ₃
Delayed by 3BSL (96 bits) to synchronize with
Update_nb signal delayed and processor 30
FA flag included in program instruction for
When "1", the output of the AND gate 56 becomes "1". A load instruction for the bit serial register 192 used to store the value f _a calculated by the multiplication-addition processor 30. When the delayed update_wb signal and the OA1 flag included in the program instruction to the processor 30 are simultaneously "1", AND
The output of the gate 57 becomes "1" and the multiplication-addition processor 30
Is a load instruction for the storage bit serial register 195 having the value oa ₁ calculated by the above. The delayed update_wb signal and the OA2 flag included in the program instruction to the processor 30 are simultaneously "
When it is "1", the output of the AND gate 58 becomes "1" and the multiplication is performed.
-A load instruction for the storage bit serial register 196 of the value oa2 calculated by the addition processor 30.

Multiplexer 59 and delay line
Reference numeral 591 denotes a latch configuration, and a logic signal at an output port of the multiplexer 59 is supplied to a triangular coefficient generator 37.
Generates a pulse so that the update_wb signal that is generated from the normally “0” state changes to “1”, the current value of the update_nb signal from the triangular coefficient generator 37 is latched, and the update_freq signal is generated. Figure 12 shows the waveform
Is shown in As shown in FIG. 16, the AND gate 197 outputs the A of the update_freq signal and the cordic_start signal.
By taking ND, it becomes "1" during the first word of each cycle of the instruction of the CORDIC processor 40. The output signal from the AND gate 197 is ANDed with f_sw in the AND gate 198, and becomes "1" during the instruction of the CORDIC processor 40 instructing the load operation of the value f. And update_freq
At the time when the signal is high, the value f calculated by the CORDIC processor 40 by "1" from the AND gate 198
Is loaded into the bit serial register 192. The output signal from the AND gate 197 is also ANDed with the n _s _sw flag at the AND gate 199,
It holds “1” during the instruction of the CORDIC processor 40 that instructs the load operation of the value n _s . At the time when the update_freq signal is high, the
The value n _s calculated by the CORDIC processor 40 by "1"
Is loaded into the bit serial register 194. The left shifter 1910 multiplies the value n _s by ²⁶ and supplies it to the multiplexers 401 and 402. Thus, even if the CORDIC processor 40 performs the division twice, the output signal of the CORDIC processor 40 does not overflow.

FIG. 10 shows the structure of the zero-cross detector 36 in more detail. The operation of this zero cross detector 36 is decoded from the program instruction of the processor 30.
This is made possible by the EZ flag. Zero cross detector 36
_Holds essentially the sign bit of the input signal s _{hk for it} (selected from one of the results of the high pass filter operations of v _hk and i _hk , described with respect to FIG. 19A), When the sign bit changes, a zero-cross signal is generated. Zero cross detector 36
Normally, the output signal ZC of "0" becomes "1", indicating the time when the sign bit changes. The mask signal rflag is generated by a timer 360 included in the zero-crossing detector circuit 36 to reduce the possibility of a zero-crossing malfunction occurring in response to noise spikes or harmonic distortion. When this is "1", the detector 36 instructs the zero-crossing in which the sign bit changes only after a predetermined time has passed since the zero-crossing instruction.
Tolerated.

The timer 360 is essentially a counter which counts down from a predetermined value ZCT read from the PROM 9. The bit-serial subtractor 361 included in this counter is arranged to feed back the difference output signal through the multiplexer 362, and generates a timer signal by delaying one word on the clock delay line 363. I do. Then, the timer signal here is added as a minuend input signal to the subtractor 361. At the beginning of the down-count, the OR gate 364 generates a "1" which switches the multiplexer 362 in response to either the reset signal or the output signal ZC of the zero-crossing detector 36 going to "1", Selecting the value ZCT causes the initial word of the timer signal to be delayed by one word on the clock delay line 363. According to the circuit 365, each bit serial of the timer signal
The logic state of the 22nd bit of the word (the sign bit of the word) is sensed and a logic signal is generated. The logic signal is held at the same logic state until the logic state of the 22nd bit of the next following bit serial word is sensed by the circuit 365. When the 22nd bit of the specified value ZCT is sensed,
Is fixedly "0" because is positive. Circuit 365 responds by generating a logic signal "0". The response from circuit 365 is applied to the input port of logic inverter 366. The output port of the inverter 366 is connected to a two-input AND gate 367 so as to apply one of the input signals. AND gate
The 367 outputs an output signal "1" under the following conditions. That is, in response to its input signal being a "1" from the logic inverter 366, and at the same time, the EZ flag "1" (the other of the input signals) which enables the operation of the zero cross detector 36. The AND gate 367 is conditioned as desired. The output signal from AND gate 367 is a two-input AND
This is one input signal of the gate 368. The AND gate 368
^Takes the bit serial ^2-15 word bits as the other input signal. The output signal of the AND gate 368 is applied to the subtractor 361 as a subtraction signal. As long as "1" is added as an input from AND gate 367 to AND gate 368, AN
D-gate 368 applies the bit serial ^2-15 word bits to subtractor 361 as a subtraction signal. timer count in timer 360 until the polarity of the timer signal changes
The down operation continues, but at this point the circuit 365
Produces an output signal of "1". Inverter 366
In response to "1", add "0" to AND gate 367, and
In response, "0" is added to the AND gate 368. AND
Gate 368 responds to its "0" input signal by
Add logic zero to 1. For this reason, the down-counting operation is stopped and the circuit 365 continues to output a "1" output until the timer is reset by either reset signal or ZC signal going to "1".

EZ flag and timer-code detector circuit 36
In response to the output signals of 5 being "1" at the same time,
Gate 369 generates the rflag signal. When this rflag signal is "1", the digital differentiator 3610 is activated and the clock rflag delay line 3611 is activated.
, A bit serial signal sign bit change applied thereto from the output port is detected. The clock delay line 3611 provides a signal s _hk (signals v _h1 , i _h1 , v _h2 , i _h2 , v _h3 and i) read from the RAM 6 to the input port after a delay of one word length at its output port. _h3 and one) is repeated in, s _hk - input to the code detector circuit 3612. The s _hk - According to the code detector circuit 3612, samples the bit 31 of the delay the s _hk signal, holds the bit for one word one-bit serial word. When the rflag signal is "1", the retained bit is selected by multiplexer 3613 and directed to clock delay line 3614. s _hk
-The currently held bits from the output port of the sign detector circuit 3612 are passed to the exclusive-OR gate circuit 3615
Exclusive-ORing with the signal at the output of clock delay line 3614 is performed. rflag signal continuously "1"
As long as, the exclusive-OR operation is performed by the exclusive-OR gate circuit 3615 on the currently held sign bit and the previously held sign bit, and the _shk -sign detector 3612 Digitally differentiate the held output. The result of this digital differentiation operation is applied as one input to a two-input AND gate 3616. This A
ND gate 3616 accepts the rflag signal as the other input signal and generates a ZC signal at its output port. except when transition in the sign bit of s _hk occurs, both input signals to the exclusive -OR gate circuit 3615 "0" or "1", the response of the gate is "0"
It is made to be. And as a result, AND
The response ZC of the gate 3616 becomes "0". When transition to the sign bit of s _hk occurs exclusively -OR gate circuit 3615
One of the input signals to is "0" and the other is "
1 "so that the response of that gate is" 1 ", and consequently the response of AND gate 3616
ZC becomes "1" and gives a zero-cross instruction.

As previously described, when ZC is "1", the timer 360 is reset and conditioned by the multiplexer 362 to apply ZCT to the input port of the clocked delay line 363.
After a word delay on clocked delay line 363, the sign bit of ZCT is detected by timer-sign detector circuit 365. The rflag signal generated by AND gate 369 goes "0" in response to the output signal of timer-code detector circuit 365 being "1". rfl
When the ag signal is "0", the multiplexer 3613
Returns the state of the code that generated the zero cross from the output port of the clock delay line to its input port until the timer counts down to zero. When rflag signal is "0", it s _hk -
Any noise spikes that change the output state of the sign detector circuit 3612 are not allowed by the multiplexer 3613 to enter this feedback loop. by the way,
When the rflag signal is "0", the sign bit being fed back, s _hk - sign detector circuitry despite any difference in the code bit being currently detected by 3612, ZC signal "0" become.

FIG. 11 shows the configuration of the triangular filter coefficient generator 37 in more detail. The bit-serial counter 370 is an arithmetic one to bit-serial register.
By cyclically counting up to the value n _s stored in 194 (FIG. 16), the triangular filter coefficient generator 37 shown in FIG. 9 can be used. This _ns input signal is valid in the upper 16 bits of the 32 bits in the bit serial register 194, as shown in FIG. Bit-serial counter 370 counts and is inherently included in the upper 16-bit sub-range of the 32-bit full range for bit-serial signals in MAP processor 30. bit-serial adder 371, a second bit-serial ^- to function as an accumulator (accumulating means) for ^15. The counting operation in the counter 370 is
Since AND is taken between the bit-serial 2 ^-15 in cordic_start pulses and AND gate 372, cordic_
This is achieved by adding the augend input signal to the adder 371 when the start pulse is "1". Adder 371
The output signal from is delayed on clock delay line 373 by the duration of a 1-bit serial word. Then, this delay signal is output to the AND gate 374
Is selectively applied as an addend input signal to the bit-serial adder 371. The bit-serial comparator 375 outputs the bit-serial delay signal from the clock delay line 373 to the bit-serial
Bit serial value n _s read from register 194
Is compared to From this comparator 375, the count is n _s
Is reached, the string "1" is output. OR gate 376 generates a "1" at its output port in response to a "1" from comparator 375 or a "1" applied as a reset signal. The "1" at the output of OR gate 376 is delayed by a duration of 2 BSL on delay line 377, and at the two input AND gate 378,
AND with the next cordic_start pulse is taken. AND
The output result "1" of the normally "0" output signal from gate 378 is inverted by logic inverter 379 and normally "1" applied to AND gate 374.
The output becomes "0" as the logic signal of. As a result, the accumulation operation by adder 371 is interrupted and arithmetic zero is added to adder 371 as an addend input signal. Reset accumulated for 2 ^{-1 5} bit-serial calculated as arithmetic 1 when comparing with the n _s.

Another counter 3710 included in the triangular coefficient generator 37 of FIG. 11 is a bit parallel counter including three cascaded counter stages. The state_0 output signal generated by the first of these stages is a "0", corresponding to the counter 370 reaching a count of n _s.
The logic state and "1" logic state are inverted. The state_1 output signal generated by the second of these stages is st
The logic state "0" and the logic state "1" are inverted according to the "1"-"0" transition of the ate_0 output signal. The state_2 output signal generated by the third of these stages is state_
1 "0" logic state and "1" logic state are inverted according to "1"-"0" transition of output signal. The string of "1" output from AND gate 378 as a reset signal to bit serial counter 370 is applied as one input of a two input AND gate 3724. And this
The other input signal of the AND gate 3724 is 2 bit-serial ^- is ^31. In response, AND gate 3724 generates a reset pulse one clock cycle long. The pseudo delay element 3725 is a negative one-bit delay applied to all relevant circuits in the silicon compiler program. Needless to say,
Although it is not possible for the negative delay of the reset pulse of the counter 3710 in the pseudo delay element 3725 to actually exist, the silicon compiler program will execute all circuit paths where the pseudo delay element 3725 is shown. , A pseudo delay is generated by introducing a unit clock delay in parallel with the circuit path. A pseudo-delay was used to synchronize the "1" update_wb signal with the cordic_start signal. This relatively time-advancing one-clock-cycle-long reset pulse is applied to counter 3710 to reset each of its three counter stages, thereby causing signal st
ate_0, state_1 and state_2 are each set to the "0" value. FIG. 12 shows the ctr count output of the counter 370, and
Signals state_0, state_1 and stat from counter 3710
The relative timing of e_2 is shown.

Β _wb supplied from the multiplexer 3711
The signal is in the form of a 2 _ns sample width of a symmetric triangular filter, as shown in FIG. As shown in FIG. 11, during the count operation period when state_0 is “0”, the multiplexer 3711 is used by the state_0 signal.
Selects ctr and forms the rising part of its output signal β _wb . During the count operation period when state_0 is "1", the multiplexer 3711 selects the difference output from the bit-serial subtractor 3712 by the state_0 signal, and forms a falling part of the output signal β _wb . This subtractor 37
In 12 the minuend input signal is n _s and the subtrahend is ctr, and the difference output signal is n _s -ctr.

Β _nb supplied from multiplexer 3713
Signal, as shown in FIG. 12, those of sample width symmetrical triangular filtering operations kernel 8n _s.
As shown in FIG. 11, during the count operation cycle when state_2 is "0", as is provided by the sum output signal of the bit-serial adder 3714.
The state_2 signal conditions the multiplexer 3713 to select the rising portion of its output signal β _nb .
During the count operation cycle when state_2 is "1", the state_2 signal causes the multiplexer 3713 to cause the rising portion of its output signal β _nb to be supplied by the difference output signal of the bit-serial subtractor 3715. Conditioned to choose. By placing all bits upward by one bit, the value n _s is multiplied by two by the shifter 3716 to generate an input signal to the AND gate 3720. Furthermore, the results 2n _s from shifter 3716 is multiplied by 2 by shifting 3717, to generate a minuend signal 4n _s against subtractor 3715. Bit serial adder 371
Is used as a subtraction input signal by a subtractor 3715. After triangular filter coefficient generator is started, during the four successive ctr lamp is state_2 is "0", the adder 3714 is to generate a ramp up to 4n _s, which is the output signal β It is selected by multiplexer 3713 as the rising part of _nb . During the next four consecutive ctr ramps where state_2 is "1", adder 3714
Although generates a ramp up to 4n _s again, the lamp is subtracted from 4n _s in subtractor 3715, it is adapted to generate a complementary ramp. This complementary ramp is then selected by multiplexer 3713 as the falling part of its output signal β _nb .

[0074] When generating the respective lamp to 4n _s as its sum output signal, the adder 3714 receiving the counter output ctr as its augend input, also, another bit-serial adder as its addend input Accept the sum output from 3718. What adder 3718 accepts as its addend and addend input signals are the output signals from AND gate 3719 and AND gate 3720. At the beginning of the ramp output from adder 3714,
The "0" value state_0 and state_1 signals input to AND gate 3719 and AND gate 3720 add the arithmetic zero addend and addend input signals to adder 3718. Adder 3718 sums these arithmetic zeros, generates an arithmetic zero as the sum output signal, and supplies this to adder 3714 as its addend input signal. Therefore, this adder 3714
Will be equal to its ctr augend input signal.

In the next cycle of the ctr output from the counter 370, the “1” value state_0 signal applied as one of the input signals to the AND gate 3719 reverts to the n _s signal input as the other input signal. It is made valid and used as an addend signal by the adder 3718. The “0” value state_1 signal applied to AND gate 3720 adds the arithmetic zero augend input signal to adder 3718. Therefore, the sum output signal from adder 3718 corresponds to n _s . Since adder 3714 accepts the addend and augend ctr signals, the sum output signal is n _s + ctr.

In the next cycle of the ctr output from the counter 370, "0" is applied to the AND gate 3719.
The value state_0 signal arithmetically adds an addend signal of zero to adder 3718. The "1" value state_1 signal applied as one input signal to the AND gate 3720 is
The 2n _s signal accepted as other input signals to reenable used as augend the adder 3718. Therefore, the sum output signal from the adder 3718 corresponds to 2ctr. The adder 3714 accepts the 2n _s signal as addend, since accepting ctr signal as augend, the sum output signal becomes 2n _s + ctr.

In the next cycle of the ctr output from counter 370, the "1" value state_0 signal applied as one input signal of AND gate 3719 will cause the n _s signal to be accepted as the other input signal at its output. It is again valid and used as an addend signal in the adder 3718. State_1 signal of one is applied as an input signal "1" value to the AND gate 3720, the 2n _s signal accepted as other input signals as valid again, used as augend signal of the adder 3718. Thus, the sum output signal from the adder 3718 corresponds to 3n _s. The adder 3714 accepts 3n _s signal as addend, since accepting ctr signal as augend, the sum output signal becomes 3n _s + ctr.

When the state_0 signal is "1", AND
Gate 3721 responds to the final count of counter 370 by
This signal is called pdate_wb and normally generates a "1" pulse, which is a "0" signal.
Indicates that the wideband (ie, narrow kernel) low pass filter operation at 0 has been completed. ctr is
The duration for which the update_wb signal is "1" is equal to the program cycle of the multiply-add processor 30, i.e., 64 BSL, since it is incremented each time cordic_start occurs. AND gate 3722 accepts update_wb at one of its input ports. Every fourth time when the update_wb signal pulse becomes "1", the AND gate 3723 simultaneously sets "1"
"1" is transmitted to the other input port of the AND gate 3722 in response to both the state_1 signal and the state_2 signal. The AND gate 3722 responds by generating a "1" pulse, called an update_nb, which is a normally "0" signal, which is a relatively low bandwidth (i.e., wide) signal in the multiply-add processor 30. Indicates that the low-pass filter operation (of the kernel) has been completed. This low bandwidth low pass filter operation is used by the multiply-add processor 30 when calculating f _a . In this processor 30, a wide band low-pass filter operation is used for calculating other signals.

FIG. 13 shows the CORDIC processor 40 in more detail. According to this processor 40, yin / xi
Not only can the functions arctan and magnitude (xin, yin) of n be computed simultaneously in an iterative CORDIC procedure, but they can also perform division or square root operations, sharing the same digital hardware used in CORDIC operations. it can. The three bit serial accumulators (accumulation means) included in the CORDIC processor 40 can selectively perform addition or subtraction operations, respectively. The first of these bit-serial accumulators accumulates the xout signal and includes elements 404,413-415,429-431,409 and 410. The second of these bit-serial accumulators operates on the zout signal, and element 4
Includes 05,432-435,411 and 412. The third of these bit-serial accumulators is yout
Computes a signal and includes elements 403, 422, 424-426, 4
Includes 07 and 408.

The xout signal is the magnitu value during the CORDIC operation.
This is an output signal of the processor 40 for a function de (xin, yin). At this time, the xout component and the yout component of the vector decomposed so as to be orthogonal to each other follow a series of asymptotically rotating operations, and the yout component decreases so as to be as close to zero as possible. Increase to be as close as possible to the size. A bit serial accumulator that accumulates the xout signal is not used when performing the square root operation. Then, when performing division, the bit-serial accumulator operates as a serial memory for the magnitude of the xin signal, rather than increasing or decreasing xout.

The bit-serial accumulator that accumulates the zout signal calculates the arctan of yin / xin by adding and subtracting the characteristic coefficient of the arctan supplied from the ROM circuit 38 during the CORDIC operation. These arcta
n The intrinsic coefficient is a binary value such that this accumulation operation requires addition or subtraction of a full digital word. When bit-serial operations are performed, where the successive bits of the digital word become increasingly important, the division and square-root operation procedures use the full digital word of their respective partial quotients and partial square root results. Is required. This is because, in arithmetic operations on parallel bits or bit-serial operations where successive bits of a digital word become increasingly insignificant, these bits can be successively accumulated using a simple shift register. This is in contrast to what is done on the basis of A bit-serial accumulator that accumulates the zout signal, which is used to accumulate arctan yin / xin during CORDIC operations, uses a divide-and-square operation procedure performed by processor 40 in bit-serial arithmetic calculations. Also used to accumulate results. This saves hardware because CORDIC and the division and square root operations can be performed on the same hardware.
The zout signal is the output signal of the CORDIC processor 40 for all functions except magnitude (xin, yin).

In all iterative procedures performed by the processor 40, the yout signal is such that for the CORDIC and divide operation procedures, the magnitude of yout is asymptotic to the aforementioned value, arithmetic zero. This is selectively increased or decreased by a bit serial accumulator that accumulates the yout signal. As the value of yout decreases overall, a bit-serial accumulator for yout shifts yout up by a unit amount at each successive operation step to avoid loss of dynamic range in the operation. Type. Yout scaling is usually achieved by correction scaling for all relevant numbers related to the number required by the algorithm. When calculating yin / xin, 1
At each iteration, x is kept constant instead of being scaled by a factor of two. When calculating the square root of yin, the number of attempts is scaled by 1/2 at each iteration instead of 1/4. When performing a CORDIC rotation, y is modified with unscaled x,
Also, x is modified using y scaled by ^2-2i during iteration step i. This is
X and y before modifying the other, as is traditionally done when circuitizing a CORDIC algorithm.
Is different from scaling each of the two by 2- ⁱ . In all cases, the accumulation of zout is y
It is unaffected by the scaling of x and x.

The signal div_sw is a logic signal that is “0” except when the processor 40 performs division. And
The signal sqrt_sw is a logic signal that is “0” except when the processor 40 performs the square root operation. Signal div_sw
And the signal sqrt_sw is provided as a single bit field in each of the instructions for the CORDIC processor 40 and the signals passing through the processor 40 to selectively perform each of the three different types of operations. Controls the multiplexer used to switch the transmission of the data. ctr5 is a logic signal that is driven high only during the odd cycles of a 64-cycle CORDIC operation.
When ctr5 is "0", CORDIC (in Figure 2)
The 5-bit counter output from program counter 108 defines the PROM 9 address, one of 32 CORDIC instructions for CORDIC processor 40. ctr5 is
When "1", the count output from counter 108 is the PROM 9 address for the CORDIC coefficient for CORDIC processor 40. Each CORDIC instruction accessed when ctr5 is "1" is maintained when it is "0" next to the ctr5. This reflects that the processor 40 requires a period of two words for an operation cycle.

While the cordic_start signal is "1", multiplexer 403 selects the yin signal applied as its first input signal from multiplexer 402 in FIG. 16 as its output signal y. Similarly, multiplexer 404 is its multiplexer from multiplexer 401 in FIG.
The xin signal applied as the input signal of 1 is the output signal
Select as x. The multiplexer 405 then selects the arithmetic zero added as its first input signal as its output signal z. The ctr5 signal applied as an input signal to AND gate 406 is low,
The output signal from ND gate 406 is set low. The low output from AND gate 406 causes its output signal, selected by multiplexer 407, to be equal to y, the input signal from multiplexer 403. This output signal is input to a one-word delay line 408. The low output from AND gate 406 causes its output signal, selected by multiplexer 409, to be equal to x, the input signal from multiplexer 404. This output signal is input to a one-word long delay line 410. Further, the low output from AND gate 406 causes its output signal, selected by multiplexer 411, to be equal to z, the input signal from multiplexer 405. This output signal is input to a one-word long delay line 412. cordic_start
After the point when the signal is high, that is, the CORDIC processor 4
After a period of input decisions for the bit-serial yout, xout and zout signals by zero, multiplexer 403
Is the yout signal returned from the output of the delay line 408 as a second input signal. The output signal selected by multiplexer 404 is the xout signal returned from the output of delay line 410 as a second input signal. The output signal selected by the multiplexer 405 is output to the delay line 4
From the twelve outputs is the zout signal returned as the second input signal. The arithmetic operation proceeds only when the ctr5 signal applied as an input signal to AND gate 406 is high. Thus, in a loop through elements 408, 403 and 407, multiplexer 407 interrupts the bit-serial latch operation. Elements 410,404 and 409
In the loop through, the bit serial latch operation is interrupted by the multiplexer 409. Similarly, the element
In a loop through 412, 405 and 411, the multiplexer 411 interrupts the bit-serial latch operation.

In the operation algorithm, the dividend y
Requires a positive divisor | x | to be used with in. ctr5
Is "0", ie elements 410, 404 and 409
During a bit-serial latch operation in a loop through, the sign bit of the output signal x from the multiplexer 404 is sampled by a sample and hold circuit 413 and held for a duration of one word. The multiplexer 414 is controlled by the held sign bit, and the held sign bit becomes “0”.
, The x input signal or -x when the retained sign bit is "1"
Select an input signal as its output signal. That is, when ctr5 is "1" during the arithmetic operation, the output signal from multiplexer 414 is | x |. Multiplexer 41
The -x input signal for 4 is obtained from the bit-serial subtractor 415 as the difference between the arithmetic zero and the output signal x of the multiplexer 404.

Normally, in the CORDIC calculation, the rectangular coordinates
The vector with xout and yout follows a stepwise rotation where the magnitude of yout decreases continuously toward zero and the magnitude of xout increases toward the full vector.
The angles of each rotation included in successive rotation steps become progressively smaller and are accumulated to increase the accuracy of the rotation angle zout which reduces yout to zero. CORD
At each step of the IC operation, the value before yout is compared to arithmetic zero to determine its sign. This is to determine whether the current value of yout is smaller than the value before yout as the first cross sum by increasing or decreasing the value before xout by 1/2. The second cross sum is generated as an updated xout by either: When the previous value of yout has increased by half the previous value of xout that occurs in the first cross sum,
If the value before xout is reduced by half the value before yout, or if the value before yout is reduced by half the value before xout that occurs in the first cross sum, the value before xout Is increased by half of the previous value of yout. Determining whether the previous value of yout increases or decreases by half of the previous value of xout also determines the decrease or increase of the rotation angle zout in the step of rotation.

As executed in processor 40,
The first cross sum is performed by increasing or decreasing twice the previous value of xout by the previous value of xout. No.
The cross sum of 2 is 2 ^-2 yout in the first step of the operation,
In the second step of the operation, only 2 ^-4 yout, in the third step of the operation, 2 ^-6 yout, and so on.
Performed by increasing or decreasing the previous value of ut. These modifications in the CORDIC algorithm are
This reflects the fact that the bit-serial accumulator for t is of the type that shifts yout up by one bit in each successive operation step, and when the value of yout decreases overall, To avoid loss of dynamic range in calculation.

The control bit div_sw during the operation of CORDIC is “0”, and the multiplexer 416 is selected so as to continuously output “1” as the other input signal to the AND gate 406. Further, the output signal of the AND gate 406 is equal to its ctr5 input signal.
The multiplexer 417 selects | x | as its input signal according to the control bit div_sw having a value of “0”.

The control bit sqrt_sw during the CORDIC operation is "0" and the multiplexer 418 selects arithmetic zero as its output signal. The output signal from this multiplexer 418 is a value that is compared to yout, which is continuously modified during each of the repetitive operation steps. The arithmetic zero output signal selected by multiplexer 418 is compared to the y output signal from multiplexer 403 to determine the sign of y that indicates the direction of correction y should be made. The comparison here is made by the digital comparator 419, and the sign bit is stored in the latch 420 for one word. When y is arithmetic zero or greater, comparator 419 produces an output signal of "1";
Generates "0" output signal. Latch 420 for "1" or "0"
The output signal of the magnitude you decrease before the value of yout
whether to increase or decrease the previous value of yout when updated to t, or to decrease the previous value of xout when the previous value of xout is updated to an increased size xout Whether to increase and ROM
The arctan coefficient from 38 determines whether the accumulated rotation angle zout decreases or increases.

In the multiplexer 421, the control bit sqrt_sw which is "0" is set to | x
Select the signal and add it to the bit-serial adder 422 as the augend input signal. This "0" value control bit sqrt_sw also selects the previously selected | x | signal in multiplexer 423 and applies it to the bit serial subtractor 424 as a subtraction input signal. The addend input signal to adder 422 and the subtraction input signal to subtractor 424 are the y output signals from multiplexer 403, respectively.

In the following description of the specification,
(i-1), i, (i + 1) indicate the (i-1) -th, i-th, and (i + 1) -th repetitions of the operation steps in the processor 40.

In the multiplexer 425, when y _i is negative, the sum output signal from the adder 422 becomes y _i and | x
_i | is selected as a cross sum of the, or, when y _i is positive, the difference between the output signal from the subtractor 424 is selected as the decrement or increment. The output signal from multiplexer 425 is squared by left shifter 426, resulting in
The doubled signal is provided as a second input signal to multiplexer 407 as a repeated value y _{(i + 1)} .

The y _i signal from multiplexer 403 is scaled by δ _i = 2 ^{−2i at} scaler 427 and delayed at delay line 428, ^followed by x _i ^plus 2 ⁻² _i · y _i . Generate a cross sum. The output signal from the delay line 428 is applied as an augend input signal to a bit serial adder 429 and as a subtraction input signal to a bit serial subtractor 430. The addend input signal to adder 429 and the minuend input signal to subtractor 430 are the | x | input signals from multiplexer 414, respectively. Multiplexer 431 selects the sum output signal from adder 429 as the second input signal to multiplexer 409, in response to y _i being zero or positive, and responds to y _i being negative. Then, the difference output signal from the subtractor 430 is selected as the second input signal to the multiplexer 409. This second input signal to the multiplexer 409 is repeated in its output signal in response to the ctr5 signal being high.

The output signal z from the multiplexer 405 is
It is added as an augend input signal to a bit-serial adder 432 and
Added as the minuend input signal for three. Lead
The arctan coefficient sequentially read from the only memory 38 is used as an addend input signal by the adder 432.
Further, when the sqrt_sw control signal is "0", the multiplexer 434 selects the tandata signal from the ROM 38 as its subtrahend input signal to the subtractor 433, as during the CORDIC operation. Multiplexer
435 outputs the sum output signal from adder 432 to multiplexer 411 in response to y _i being zero or positive.
And the difference output signal from the subtractor 433 is selected as the second input signal to the multiplexer 411 in response to y _i being negative. The second input signal to the multiplexer 411 repeats its output signal in response to the ctr5 signal being high.

During the operation of yin / xin following the procedure of division,
Processor 40 operates as a fractional machine,
xin and yin consist of the sign bit of the two's complement arithmetic value in the most significant bit, followed by the binary point. this
During the operation of yin / xin, div_sw is "1". Thus, the multiplexer 416 selects the output signal from the saturation arithmetic circuit 440 as its output signal. Multiplexer 417
, The | x | / 2 signal is selected as the output signal. Further, the output signal of the scaler 427 is continuously set to “0”. Assuming that the magnitude of yin is less than xin, the string of "1" s generated by the saturation arithmetic circuit 440 causes the division procedure to proceed. As will be described in more detail with respect to FIG. 15, the div_sw signal is "1" so
In the coefficient ROM circuit 38, and supplies the 2 ^-1 bit-serial between CORDIC program instructions of the first pair, second
Provides a bit serial 2 ^-2 between consecutive pairs of CORDIC program instructions and a third consecutive pair of CORDIC
Supplies bit-serial 2 ^-3 during DIC program instructions. The process proceeds in a similar manner.

The output signal of the scaler 427, which is continuously "0", results in the arithmetic zero being input to the adder 429 as an addend input signal. x |. Also, as a result, since the arithmetic zero is input to the subtractor 430 as the subtraction input signal, the difference output thereof is also equal to the minuend input signal of | x | Therefore, when yin / xin is calculated, the output signal | x | of the multiplexer 431 becomes equal to the output signal of the multiplexer 414. When ctr5 is "1" and is circulated in the loop through elements 410, 404 and 409, after this value is selected by multiplexer 409, the output signal from multiplexer 404 will be transmitted throughout its arithmetic cycle (ie, cordic_start Until the next time that goes high) | x |.

The multiplexer 417 includes a multiplexer 404
Outputs half of the | x | output signal maintained from This | x | / 2 signal is supplied as a trial divisor in a division procedure and is added to y in adder 422 or subtracted from y in subtractor 424. The x / 2 signal selected by multiplexer 417 is shown as being applied by bit shifter 436, as shown in the schematic provided for the silicon compiler. In practice, the shifting of the lower bits is done by a silicon compiler that shifts one bit higher in every parallel pass.

During the operation of yin / xin, the control bit sqrt_sw is “0”, which causes the multiplexer 434 to output as the subtraction input signal to the subtractor 433
tandata = 2 Select ^-i . At each successive point in time when the ctr5 signal is "1", the value z from multiplexer 405 is summed by adder 432 with the value of tandata equal to 2
The value selected by the multiplexer 435 is a value increased by ^-i or a value reduced by 2 ^-i of the value of tandata in the subtractor 433. The updated value z is fed back to multiplexer 405 via delay line 412 and is used at the next point in time when the ctr5 signal is "1". The choice made by multiplexer 435 (and by multiplexers 425 and 431) is made according to the sign of the operated-up y before the successive step of comparing the dividend y or the modified dividend with zero. The above steps are performed in a division procedure as one of the attempts to reduce the size of the dividend to be modified. If y _(i-1) (the previous value of y) is zero or positive, the magnitude of y _i (the current value of y) is | x | /
2 by subtracting 2 and increasing by tandata = 2− ⁱ in the adder 432.
The value of z from 5 is selected by multiplexer 435 to update the value of z. If y _(i-1) is negative, y _i
The magnitude of (current value of y) is reduced by adding | x | / 2, and the value of z from multiplexer 405 when subtracting tandata = 2− ⁱ in subtractor 433 is To update the value of z.

Control bit sqrt_s which is a "0" value
Due to w, the multiplexer 418 selects arithmetic zero as an output signal. The arithmetic zero output signal selected by multiplexer 418 is compared with the output signal y from multiplexer 403 to determine the sign bit of y, where the comparison is performed by digital comparator 419 and latch 420.
This is done by the sign bit stored for one word within. The control bit sqrt_sw, which is a "0" value, is to add the | x | / 2 signal previously selected by the multiplexer 417 in the multiplexer 421 to the adder 422 as its augend input signal. The control bit sqrt_sw, which is a “0” value, further causes the multiplexer 423 to multiplex the | x | / 2 signal selected earlier.
That is, the subtraction signal is added to the subtractor 424 as its subtraction input signal. Addend input signal to adder 422 and subtractor 4
The minuend input signals for 24 are the y output signals from multiplexer 403, respectively. In the multiplexer 425, the sum output is selected from the adder 422 when the value before y is negative, and the difference output is selected from the subtractor 424 when the value before y is positive. In order to maintain the dynamic range during the division procedure between the steps of the trial division operation, the lower one bit without shifting the value of y- | x | or y + | x | Rather than following the procedure of shifting the fractional divisor | x |, the y- | x | or y + | x | selected by the multiplexer 425
Without shifting the trial divisor | x |
It is shifted upward by one bit. The output signal from multiplexer 425 is multiplied by 2 by left shifter 426,
The weight of the corresponding output signal is increased, and the resulting doubled signal is added as a second input signal to multiplexer 407.

The saturation arithmetic circuit 440 is used to maintain the magnitude of the quotient zout that does not become greater than one. Quotient z
| y | must be less than | x | to maintain the size of out. Only when | y | is small enough to allow the division to proceed satisfactorily,
Gives "1". When y exceeds the -x output signal from the subtractor 415, indicating that the negative value of y is not too large within the range, a "1" is generated by the bit-serial comparator 441 in the circuit 440. Occur. When y is less than x and a positive value of y is indicated not to be too large within the range, a "1" is generated as its output signal by the bit-serial comparator 442 in circuit 440. Is done. AND gate 443 in circuit 440 generates a string of "1" in response to both indications that y is in range, and is delayed by one-word before appearing as an output signal from saturation arithmetic circuit 440. For this purpose, it is added as an input signal to the delay line 444. When the positive y is out of range, comparator 419 continuously outputs "0" after the sign bit of "0" for y has passed.
1 "is output, and the accumulation of zout results in its largest positive value, and the string" 1 "becomes" 0 "
Will be followed by the sign bit of Negative y
Is outside the range, the comparator 419 outputs "0" continuously after the sign bit of "1" for y. Then, the accumulation of zout results in its least negative value, and the string of "0" will be followed by the sign bit of "1".

(Yin) The procedure of the square root operation for calculating ^(1/2) is similar in many aspects to the procedure of the division operation. However, rather than saying that y _i is the dividend when i = 1, and that it is the remainder dividend otherwise, y _i is the square root when i = 1 and otherwise the square root of the remainder. It is. The trial divisor in the division operation procedure becomes the trial square root to be divided by the remainder dividend, producing a result to be compared with the trial test root itself, and the square root or remainder dividend Determine if the square root number should be reduced. Keep track of the trial test root itself in a separate xout register, because the trial square root to be divided by the number of square roots of the remainder should be derived from the square root of zout extracted so far. do not have to. The trial square root in the first operation step is an invariant 1/4 or binary 0.01.

(Yin ^{) During} the operation of ^(1/2) , the control bit div_sw is “0”, and the multiplexer 416 selects continuous “1” as another input signal to the AND gate 406. The output signal of the AND gate 416 repeats the ctr5 input signal.

(Yin) During the operation of ^(1/2) , control
Bit sqrt_sw is "1". As described in further detail with respect to FIG.
Is conditioned, between the first pair of CORDIC program instructions is a bit serial 2 ^-1 , the second pair of CORDIC
Bit serial 2 ^-2 is supplied as a tandata signal between program instructions, bit serial 2 ^-3 , etc., between the third pair of CORDIC program instructions. sqrt_sw
Since the signal is "1", the multiplexer 421 selects arithmetic zero as an addend input signal to the adder 422. Therefore, the sum output signal is equal to the augend input signal y, and it is only possible to reduce the square root of the remainder. The sqrt_sw signal is
By being "1", in the multiplexer 423,
A trial square root corresponding to the sum output signal of the bit serial adder 437 is selected as the subtraction input signal of the subtractor 424, where the adder 437
Accept the output signal as its augend input signal,
data / 2 as its addend input signal. The tandata / 2 signal here is shown as being added to the adder 437 by the bit shifter 438, as shown in the schematic provided for the silicon compiler, but is actually The bit shift when going to is done by shifting up by one bit in the silicon compiler in all parallel paths. Since the sqrt_sw signal is "1", the multiplexer 434 selects arithmetic zero as a subtraction input signal to the subtractor 433. Thus, when that occurs, the accumulation of the square root will be fixedly incremented. Further, the sqrt_sw signal is compared to the square root number or the remainder square root number y _{i at} multiplexer 418 to determine whether y _{(i + 1)} should be reduced by being a "1" signal. Choose a trial square root.

FIG. 14 shows the operation of the scaler 427 in more detail. Signals ctr5, ctr6, ctr7, ctr8, ctr9 and ctr10 indicate the upper 6 bits of the count output of counter 103 (in FIG. 2). OR gate 4270
During the yin / xin operation, the div_sw control bit is set to "
In response to being 1 ", the multiplexer 4271 selects arithmetic zero as the other input signal of the scaler circuit 427, and the output signal becomes arithmetic zero.
Gate 4270 selects an arithmetic zero as another input signal to scaler circuit 427 in response to the ctr10 output from counter 103 reaching "1" at full count, and the resulting output signal is It is made to be zero.

Except when the output signal of the OR gate 4270 is “1”, the scaler 427 outputs ctr6 and ctr from the counter 103.
In response to 16 consecutive counting conditions of the 7, ctr8 and ctr9 signals, as referenced by the bit-serial signal y applied to the scaler 427, by twice the number of bits in the corresponding count, Shift the signal y to the right,
The bit serial signal ys from the scaler 427 is output.

FIG. 14 shows a 2-bit write shifter 4272 and a multiplexer 4273 and a ctr7 signal used for selectively dividing by a factor of 4 in response to the ctr6 signal being "1". 1 in response to being
6 4-bit write shifter 4274 and multiplexer 42 used to selectively divide by a factor of
75, ctr8 8-bit write shifter 4276 and multiplexer 4277, ctr9 used to selectively divide by a factor of 256 in response to signal being "1"
Shown are 16-bit write shifters 4278 and multiplexers 4279, respectively, used to selectively divide by a factor of 65 536 in response to the signal being a "1". In fact, shifting bits downwards is provided by a silicon compiler that introduces a delay in all parallel signal paths and shifts those parallel signal paths upward. Since the right shifter itself is not available in a bit-serial format, the left shift is required for all parallel signal paths in this procedure.
Procedures using shifters are required.

FIG. 15 shows the arctan coefficient ROM 38 in more detail. The reduced size for the actual ROM 38 was made possible by the fact that at small angles the tanjent function was almost linearly proportional to angle. Therefore, the actual ROM 38 stores the arctan coefficients of only the eight maximum angles having a tanjent having a slope of a power of two. These arctan coefficients are addressed by the ctr6, ctr7 and ctr8 outputs of counter 103. Although the stored arctan coefficients are 16-bit long, they are read through the parallel input / serial output output registers in ROM 38 by following up with the 16 trailing "0" s. Then, the serial output from the ROM 38 is delayed by a delay line 381 and shifted so that the 16 succeeding "0" appear as the leading "0". Ctr9 of counter 103 and 2
ctr10 signal, the first two ⁹ counts of the counter 103 (CORDI
As long as both are "0" for eight cycles in the C operation, the output signal of the OR gate 382 is "0",
This allows multiplexer 383 to read from ROM 380
The arctan coefficient is selected as the input signal to yet another multiplexer 384. In response to the output signal from OR gate 385 being "0", arct from multiplexer 383 is output.
the output signal repeats as the output signal of multiplexer 384. As long as the sqrt_sw and div_sw signals accepted as its input signals are both "0", that is, during the operation of the arctan of yin / xin, the output signal from the OR gate 385 is "0".

During the operation of arctan yin / xin, the tandata output signal from multiplexer 384 is a delay line
Delayed by one word at 386 is applied to multiplexer 387 as one of the input signals. The other input signal of the multiplexer 387 is half the magnitude and is generated by the bit shifter 388 from the output signal of the delay line 386. The ctr5 control signal causes the multiplexer 387 to repeat larger input signals when the ctr5 signal is "1" and to repeat smaller input signals when the ctr5 signal is "0". First of counter 103
After the 2 9th count (8 cycles of the CORDIC operation), the output signal of the OR gate 382 is "1", and the multiplexer 383 repeats the output signal from the multiplexer 387. Thus, for each successive even CORDIC operation cycle, the minimum arctan coefficient stored in ROM 380 is halved and appears as the tandata output of multiplexer 384.

During the operation of yin / xin or the square root of yin, one input signal of the OR gate 385 is “1”, and the multiplexer 384 outputs the signal of the multiplexer 383.
The output signal of the multiplexer 388 in the tandata output signal is repeated instead of the arctan output signal. cordi
When c_start signal is high, the first cycle of CORDIC instructions, multiplexer 388 selects the 2 ^-1 bit-serial as its output signal, for this, t
The andata signal is also 2 ^-1 . Then multiplexer 38
8 repeatedly outputs the output signal of the multiplexer 387, thereby providing a bit serial output for each even CORDIC instruction.
2 ^-1 is halved and tandat from multiplexer 384
a Output signal appears.

According to the CORDIC processor 40 in FIG. 16, the xin input signal selected by the multiplexer 401 is received, and the yin input signal selected by the multiplexer 402 is received. CORDIC processor 40
Is the bit serial yout output signal and the bit
The serial zout output signals are output simultaneously, but these signals are applied to the respective input ports of a two-input multiplexer 400. The bit-serial yout or zout output signal selected by multiplexer 400 is fed back as the respective input signal to multiplexers 401 and 402 for continuous operation.
Enabled by CORDIC processor 40. In the power measurement application, the output data of the CORDIC processor 40 is returned to the multiplication-addition processor 30 only when the processor 40 performs the operation of n _s or f. The output signals of multiplexer 400 cause the respective input signals to be written to registers 192 and 194 in register bank 19 under appropriate connections to feed back _ns and f to multiply-add processor 30. .

As noted above with respect to FIG.
The processing via the C processor 40 is performed based on the time division switching method, and it is effective to measure two functions at the same time. In response to both the high cordic_start signal and the high load1_sw signal during the first word period of each CORDIC instruction cycle, AND gate 68 generates a "1" output signal, and the function digo
Indicates that the bit serial instruction for ut1 is to be loaded into the bit serial output register 60.
High during the first word of each CORDIC instruction cycle
In response to both the cordic_start signal and the load2_sw signal being high, the AND gate 69 generates a "1" output signal and loads the bit-serial instruction for the function digout2 into the bit-serial output register 62. To do so.

The shared PDM counter (not specifically shown) accessed by each of pulse width modulators 64 and 66 is a bit-serial counter included in clock generator 100 (see FIG. 1). Clock generator 10
It counts the number of crystal oscillators of 10 MHz at 0 (for example, 2 ¹⁵ ). Each bit-serial comparator (not specifically shown) included in the pulse width modulator 64 compares digout1 in bit-serial form with the bit-serial count coming out of the PDM counter. Which generates its output pulse train. Similarly,
Each bit-serial comparator (not shown) included in the pulse width modulator 66 compares digout2 in bit-serial form with the bit-serial count coming out of the PDM counter. Which generates its output pulse train. The cycle of the pulse width is 1024 BSL, and if the resolution is 10 bits, the count may be executed in words (32 bits).
In order to further increase the resolution, division within a word is necessary. For example, considering the case where 2 bits are added, dig
The upper 11th and 12th bits of the out1 or digout2 signal are 0, 1/4, 1/2, 3 /, corresponding to 00, 01, 10 or 11
Time division of 4 words is performed to selectively generate a signal of "1" value. Thereby, a resolution of 12 bits can be realized in a pulse cycle of 1024 BSL.

The PDM generator 66 shown in FIG.
Are connected to a two-input exclusive OR gate 70 as an input signal. Then, except when both the load1_sw and load2_sw signals are high and at the same time the cordic_start pulse is high, the p
The dm2 output signal corresponds to the pulse from the PDM generator 66. Under these conditions, digout1 and digout2
Are output from the PDM generators 64 and 66, respectively. Multiplexer 71 is AND gates 68 and 69
Accepts two "1" s with the output signal of
In response, it generates "1" as its output signal. The output signal of multiplexer 71 is applied to exclusive OR gate 70 as its second input signal. Exclusive OR
The pdm2 signal from gate 70 is the complement of the pulse from PDM generator 66. Thus, PDM generators 64 and 66 are arranged for a push-pull drive to a center read D'Arsonval meter. The one-bit delay 72 is connected so as to latch the state of the output signal from the multiplexer 71 when the output signal of the AND gate 68 returns to “0”.

The bit serial yout or zout output signal selected by multiplexer 400 is applied as an input signal to normalizer 39. As noted in the previous description of FIG. 1, this normalizer 39 performs yet another basic function calculation in cooperation with the CORDIC processor 40. (E) [sign (oa2)] * (yin-L) / M According to this function, the meter's zero is first converted by a constant L, and then the converted reading is factor M
By scaling down with
On meters such as 3,65 and 67, extended scale readings are possible. In power measurement, this function is a fixed last operation, so yin for the last operation
The signal is equal to the yout output signal of multiplexer 400 for the penultimate operation. L and M are PRO
Constants in the CORDIC program instruction from M 9, extracted from elements 98 and 984-986 of FIG. 2 and added to normalizer 39. Multiplexer 402
Either the difference output signal from the subtractor 391 or the difference output signal from the subtractor 392 is selected by the multiplexer 390 as the output signal to be added to The yout signal from the multiplexer 400 is subtracted from L in the subtractor 391, and L is subtracted from the yout signal in the subtractor 392.
Is subtracted from the signal. M is directly applied to multiplexer 401 as its input signal, and as a divisor of xin to a dividend of (yout-L) selected from normalizer 39 to processor 40 by multiplexer 402,
(In the last operation by the CORDIC processor 40).

The polarity of oa2, sign (oa2), is determined from the oa2 bit serial data stored in the register 196 by sampling and holding the bit 31 of oa2 in the sample and hold circuit 394. The sampled and held sign (oa2) is applied from circuit 394 to a two-input AND gate 395. This AND gate 395
The other input signal is sq_sw, which is "0" during all CORDIC operations, except when it is desired to change the polarity of the final CORDIC output according to the sign (oa2) being negative. When the signal sq_sw is "0", the AND gate 395
Is "0" and this signal sq_sw is "
If it is 1 ", it is [sign (oa2)]. Exclusive OR
Gate 396 receives the output signal of nand gate 395 and the n_sw signal as input signals, and
At 390, the selection of the output signal from the subtractor 391 and the subtractor 392 is controlled. To invert the polarity of the digout1 and pdm1 output signals or the polarity of the digout2 and pdm2 output signals, set the n_sw signal to high. Exclusive OR when n_sw signal is "1"
In the gate 396, the output signal of the AND gate 395 is "0"
, The difference signal (L-yout) from the subtractor 391 is used as its output signal, and when the output signal of the AND gate 395 is "1", (yout-L) from the subtractor 392 is used.
Multiplexer so that the difference signal becomes the output signal.
Select 402. When the n_sw signal is "0", the output signal of the AND gate 395 is output by the exclusive OR gate 396.
When it is "0", the difference signal (yout-L) from the subtractor 392 is used as its output signal, and when the output signal of the AND gate 395 is "1", (L-you
t) is the output signal of the multiplexer 4
Selected at 02.

FIG. 17 shows the format of an instruction for the CORDIC processor 40. These instructions are 3
It has two main fields. That is, in the CORDIC processor 40, a 9-bit field for controlling the input and function selection, a 7-bit field for the load and control flag signals, and a 16-bit field for the CORDIC coefficient (RC-coef value). is there. The 7-bit field for the load and control flag signals contains four single bit load signals. That is, f_sw, ns_sw, l
Load1 corresponding to oad1_sw and load2 corresponding to load2_sw are included. The 7-bit field for this load and control flag signal also contains three single bit control signals. That is, s used by the normalizer 39 in FIG.
q_sw and n_sw, and inc corresponding to the cordic_inc instruction added to AND gate 1010 in FIG. 12 to control the implementation of the count in CORDIC instruction counter 108. The 9-bit field that controls input and function selection in CORDIC processor 40 has a 3-bit portion that controls the selection of yin, another 3-bit portion that controls the selection of xin, and Has a further 3-bit part that selects the function to be performed by

FIG. 18 shows the coding of inputs and functions in detail. RC is the result of a previous calculation by processor 40, such as provided by multiplexer 400 in FIG. RC-coef is the current C
Is the value of the ORDIC coefficient, and RC-coef _-1 is the current C
The value of the CORDIC coefficient immediately before the ORDIC coefficient.

FIG. 19 is a diagram showing a flow of calculations executed by the processors 30 and 40 in the power meter IC5. One instruction to the multiplication-addition processor 30 is executed from the 6-channel decimation filter 20 to the processor 30 for each input of a 32-bit word that is sequentially switched to time division. While one of the instructions to CORDIC processor 40 is being executed, a complete cycle of 64 instructions to multiply-add processor 30 is executed. The time to execute one of the instructions to CORDIC processor 40 is equal to the oversampling interval for sigma-delta modulators 21-26.

The first procedure is to compensate for the non-linearity in the current transformer 11-16 by using the decision filter 20.
Is to linearize the bit serial digital data of v ₁ , i ₁ , v ₂ , i ₂ , v ₃ and i ₃ supplied from. The current transformers 11-16 and the scale resistors used with them are mounted on a printed circuit board together with the power meter IC5, and during its manufacture, the entire printed circuit board assembly is tested as a unit and the current A coefficient of non-linearity associated with the transformer 11-16 is determined. These coefficients are written to PROM 9 included in the assembly. These nonlinearity coefficients are used by the multiply-add processor 30 to determine
For the bit serial digital data of v ₁ , i ₁ , v ₂ , i ₂ , v ₃ and i ₃ added from n filter 20,
Each is corrected using a third-order polynomial. After each of these polynomial corrections, each is high-pass filtered to remove the DC component that is an offset error in the sigma-delta modulators 21-26.

Referring now to the procedure shown in FIG. 19, in each of the steps LIN1, LIN3 and LIN5, the respective corrected value v _ck of v _k is determined by the linearization coefficient A _k read from PROM 9 And B _k and the previously calculated gain coefficient Γ _vk are used by multiply-add processor 30 for all samples of v _k . The calculation here is performed according to the following equation. v _ck = ｛[(A _k * v _k ) + B _k ] v _k + Γ _vk ｝ v _k k = 1,2,3 (1)

[0121] In each of the steps LIN2, LIN4 and LIN6, respectively corrected value i _ck of i _k are, PROM 9
Are calculated by the multiply-add processor 30 for all samples of i _k using the linearization coefficients D _k and E _k read from, and the previously calculated gain coefficient Γ _ik . The calculation here is performed according to the following equation. i _ck = ｛[(D _k * i _k ) + E _k ] i _k + Γ _ik ｝ i _k k = 1,2,3 (2)

In each of the steps HPF1, HPF3 and HPF5 following steps LIN1, LIN3 and LIN5, v _ck
Each of the high-pass filter response to the v _hk
Is calculated by the multiply-add processor 30 for all samples of v _ck using a constant P read from PROM 9. This operation is performed according to the following equation, where z ^-1 is the one-word delay operator for a 32-bit BSL. v _hk = v _ck -z ^-1 v _ck + P * z ^-1 v _hk k = 1, 2, 3 (3)

In each of steps HPF2, HPF4 and HPF6 following steps LIN2, LIN4 and LIN6, i _ck
Response i _hk of each of the high-pass filter for
Uses the repetition factor Q read from PROM 9 to calculate i _ck
Is calculated by the multiplication-addition processor 30 for all the samples of. This operation is performed according to the following equation. ^{_{i hk = i ck - z -1}} i ck + Q * z -1 i hk k = 1,2,3 (4)

As shown in FIG. 19, at the multiplication step MULT1, the multiplication-addition processor 30 executes v _h1
The multiplication of the response by the high-pass filter of itself with itself allows squaring, followed by a low-pass filter operation step LPF1. As shown in FIG. 20, the low-pass filter operation step L
The bit serial signal result from PF1 is the CORDIC input signal to the square root operation step SQRT1.
In addition to processor 40, it generates a bit-serial indication of the effective voltage on mains conductor 1.

As shown in FIG. 19, the multiplication step
In MULT2, multiplying the response of i _{h1 by} the high-pass filter with itself and performing the square operation is performed by the multiply-add processor 30, followed by
The product is input to step LPF2 of the low-pass filter operation. As shown in FIG. 20, the bit serial signal resulting from the low pass filter operation step LPF2 is applied to the CORDIC processor 40 as an input signal to the square root operation operation step SQRT2, and the Shows a bit-serial indication of the effective current through conductor 1.

As shown in FIG. 19, the multiplication step
In MULT3, computing the response of v _{h2 by} the high-pass filter and itself multiply is performed by the multiply-add processor 30, followed by
Input the product to step LPF3 of the low-pass filter operation. As shown in FIG. 20, the bit serial signal resulting from the low-pass filter operation step LPF3 is applied to the CORDIC processor 40 as an input signal to the square root operation operation step SQRT3 and Shows a bit serial indication of the effective voltage on conductor 2.

As shown in FIG. 19, the multiplication step
In MULT4, multiplying the response of i _{h2 by} the high-pass filter with itself and multiplying by itself is performed by the multiply-add processor 30, followed by
Input the product to step LPF4 of the low-pass filter operation. As shown in FIG. 20, the bit serial signal resulting from the low-pass filter operation step LPF4 is applied to the CORDIC processor 40 as an input signal to the square root operation operation step SQRT4 and Shows a bit serial indication of the effective current through conductor 2.

As shown in FIG. 19, the multiplication step
In MULT5, calculating the square of the response of v _{h3 by} the high-pass filter and itself is performed by the multiply-add processor 30, followed by
The product is input to step LPF5 of the low-pass filter operation. As shown in FIG. 20, the bit serial signal resulting from the low pass filter operation step LPF5 is applied to the CORDIC processor 40 as an input signal to the square root operation operation step SQRT5, and the Shows the bit-serial indication of the effective voltage on conductor 3.

As shown in FIG. 19, the multiplication step
In MULT6, calculating the square of the response of i _{h3 by} the high-pass filter and itself is performed by the multiply-add processor 30, followed by
Input the product to step LPF6 of the low-pass filter operation. As shown in FIG. 20, the bit serial signal resulting from the low-pass filter operation step LPF6 is applied to the CORDIC processor 40 as an input signal to the square root operation operation step SQRT6 and Shows a bit serial indication of the effective current through conductor 3.

In the SELECT step shown in FIG. 20, any one of these effective voltages and currents can be selected to appear as the digout1 output signal from IC5. Also, in a subsequent or concurrent SELECT step, any one of these effective voltages and currents can be selected to appear as the digout2 output signal from IC5. The processing of FIG. 19 performed in the multiply-add processor 30 to perform the effective voltage and current calculations for each of the phases is described by the following equation. Where v _sk is “high
Pass filtered linearized phase k voltage v
_hk squared and _isk
Is a “high-pass filtered, linearized phase
are intended to be read as the square "of k current _{i hk. v sk = v hk} * v hk (5) i sk = i hk * i hk (6)

As shown in FIG. 19, according to the multiplication-addition processor 30, in each step APF1, APF2 and APF3 of the all-pass filter, a high-pass filter is used.
_Set the phase of each of the filter responses v _h1 , v _h2 and v _h3 to 9
Shifted by 0 degrees can produce the high pass filter responses v _p1 , v _p2 and v _p3 phase shifted according to the following equation: v _pk = z ^-1 v _hk + α (z ^-1 v _pk- v _hk ) k = 1, 2, 3 (7)

As shown in FIG. 19, according to the multiplication-addition processor 30, in each interpolation (phase shift) filter step PSF1, PSF2 and PSF3,
When compared with the secondary voltages of current transformers 14, 15 and 16, respectively, the response of the high pass filter i _h1 to compensate for the respective phase differences in the secondary voltages of current transformers 11, 12 and 13 , I _h2 and i _h3 can be shifted. According to the interpolation filter steps PSF1, PSF2 and PSF3 here, it is possible to generate the responses i _p1 , i _p2 and i _p3 of the high-pass filter phase-shifted according to the following equation. i _pk = i _hk + δ _k (z ⁻¹ i _hk −i _hk ) k = 1,2,3 (8) These calculations are usually performed by program data for temporarily storing the results of these operations. Using RAM 6, v
_{p1, v p2, v p3,} i p1, i p2, is executed in the sequence of i _p3.

As shown in FIG. 19, according to the multiplication-addition processor 30, the following can be performed. That is, according to the following equation, the multiplication step MULT7 multiplies v _{h 1} and i _p1 to calculate the effective power of the first phase, and temporarily stores the result. _h2 and i _p2 are multiplied to calculate the effective power of the second phase, and the result is temporarily stored.In a multiplication step MULT9, v _h3 is multiplied by i _p3 and the third phase is multiplied. Calculating the effective power and temporarily storing the result; adding the first and second phase effective power components temporarily stored in the addition step ADD1;
The bit-serial sum signal p _s can be generated by adding the three-phase effective power components. _ps = ( _vh1 * _ip1 ) + ( _vh2 * _ip2 ) + ( _vh3 * _ip3 ) (9) According to the multiplication-addition processor 30, the sum signal _ps is low-passed. • Filter operation step An indication of the average effective power flowing through the main conductors 1, 2 and 3 can be passed to the LPF7.

As shown in FIG. 19, according to the multiplication-addition processor 30, the following is enabled. That is, according to the following equation, the multiplication step MULT10
Multiplying v _p1 by i _p1 to calculate the reactive power of the first phase, temporarily storing the result, and multiplying v _p2 by i _p2 in the multiplication step MULT11 to calculate the second Calculate the reactive power of the phase and temporarily store the result, in the multiplication step MULT12 v _p3 and i _p3
And calculate the reactive power of the third phase, and temporarily store the result.In the adding step DD2, the reactive power components of the first and second phases temporarily stored are calculated. by adding the reactive power component of the third phase can be generated sum signal q _s bit-serial. _{q s = (v p1 * i} p1) + (v p2 * i p2) + (v p3 * i p3) (10) Then, the multiplication - According to the addition processor 30, row the sum signal q _s path Pass to filter operation step LPF8 to generate an indication of average reactive power flowing through trunk conductors 1, 2 and 3.

In the function selection step shown in FIG. 20, the indication of the effective power P and the reactive power Q flowing through the main conductor can be selected to appear as digout1 and digout2 output signals from IC5.

Alternatively, as shown in FIG.
The real power and reactive power indications are calculated by the CORDIC processor in the ARCTAN step by calculating the power factor angle 、.
It also calculates the square root of the sum of those squares in the RMS step, which allows the main conductors 1, 2 and
The apparent power VA for 3 can be calculated. In another operation shown in FIG. 20, in step P / VA, the CORDIC processor calculates the cosine of the power factor angle 無効 with or without sign of the reactive power determined by step sgn (Q). be able to. You can also take the complement of the cosine determined in Step 1 -X.

The calibration coefficient is frequency dependent. 50, 60 or 400 Hz printed circuit board assembly
Because it is used in power measurement of
A preliminary procedure is to determine the frequency of the power to be measured. This preparatory procedure is performed later in the execution of a series of instructions to the multiply-add processor 30 than the steps outlined above for Equations 1-10. The result will be stored in RAM 6 for use in executing the next cycle of multiply-add processor 30 instructions. This is generally satisfactory since the power line frequency changes are very slow.

The accuracy or inaccuracy of the linearization by the third-order polynomial does not affect the determination of the frequency in any case. Then, as shown in FIG. 19, in the ZCD step executed by the zero-cross detector 36, the responses i _h1 and i i of the high-pass filter
1 selected from _h2 , _ih3 , _vh1 , _vh2 and _vh3
For each _shk , a zero-cross indication is generated. These zero-cross indications are passed to the low-pass filter operation step LPF0 by the multiply-add processor 30, as described by the following equation: ^{_{f a = z -1 f a +}} β wf * zcd (R, s hk) (11) zcd of the zero-crossing detector 36 (R, s _hk) consisting response than the maximum value 1 / 2R provisions have s _hk When it has a low frequency, it is a function of this _shk only. Where R is PR
This is a specified minimum time between zero crossings, as read from OM9. R and _shk are applied to the ZCT (see FIG. 9) and X inputs of the zero cross detector 36, respectively. When the frequency of _shk is greater than 1 / 2R, the zero-cross detector 36 generates a pulse corresponding to the highest frequency sub-harmonic of the signal frequency lower than 1 / 2R. The accumulated filter response f _a is CORDIC
Added to processor 40. Then, according to the following equation, f _a is divided by 4n _s, in which then the quotient is divided by further 4n _s, thereby calculating the frequency f of the indication of the zero-crossing. _{f = f a / 4n s /} 4n s (12) This value f multiplication - for later use by the addition processor, written into the register 192.

After the value f is updated in the register 192, the frequency-dependent constant read from the register 192 to the multiply-add processor 30 and from the PROM 9 to the multiply-add processor 30 C _0k
And multiply according to the following equation proceeding from C _1k-
The frequency-dependent gain coefficient Γ _{v k} by the addition processor 30
Is calculated. Γ _vk = C _0k + f * C _1k (13) The results of the operations of Γ _v1 , Γ _v2 and Γ _v3 are given by v ₁ , v ₁ , respectively, with respect to the linearization of the samples from current transformers 14, 15 and 16, respectively. v is used later on ₂ and v of all samples _3.

After the value f in register 192 has been updated, the frequency-dependent constants read from register 192 to multiply-add processor 30 and from PROM 9 to multiply-add processor 30 F _0k
And multiply according to the following equation proceeding from F _1k :
The addition processor 30 calculates a frequency-dependent coefficient _{ik ik} for nonlinearity. Γ _ik = F _0k + f * F _1k (14) The results of the operations of Γ _i1 , Γ _i2 and Γ _i3 are given by i ₁ , as used later on in all samples of i ₂ and i _3.

After the value f in register 192 has been updated, the frequency-dependent constant read from register 192 to multiply-add processor 30 and from PROM 9 to multiply-add processor 30 H _k
And the following equation proceeding from G _{k, the} frequency-dependent interpolation coefficient δ _k is calculated by the multiplication-addition processor 30. δ _k = H _k + f * G _k (15) The result of the calculation of δ ₁ , δ ₂ and δ ₃ is the current transformer 14
Current transformer 11 for voltage samples resulting from −16
It is later used to correct the phase of the current sample resulting from -13.

After the value f in register 192 has been updated, the frequency-dependent constant read from register 192 to multiply-add processor 30 and from PROM 9 to multiply-add processor 30 J
And the following equation progressing from K, the multiplication-addition processor 30 calculates the frequency-dependent interpolation coefficient α. α = J + f * K (16) The result of the operation of α is later used to shift the sine voltage by 90 degrees in an all-pass filter operation performed by the multiply-add processor 30.
The calculations of J and K are made to provide a precise 90 degree phase shift for 50 Hz and 60 Hz signals.

After the value f in register 192 is updated, the instruction executed by multiply-add processor 30 multiplies f read from register 192 and T read from PROM 9 together, and The binary fractional part of the resulting product is truncated in a round-down operation called “floor” to define a parameter n _p as represented by the following equation: n _p = floor (T * f) (17) where the value T is half the maximum duration allowed between pulses update_nb, and therefore,
It corresponds to the maximum time between the output signals of IC5. Also, n
_p is the maximum number of complete signal cycles that can be accommodated within this duration. The value n _p here is written to the register 193 and made available to the CORDIC processor 40. According to the CORDIC processor 40, the number of samples n _s of the sawtooth wave of each of the sawtooth wave generator in the triangular filter coefficients generator 37 is calculated after, according to equation below. n _s = floor (n _p / f) (18) where the value n _s is written to the register 194 and sent to the triangular filter coefficient generator 37 needed by the low-pass triad operation already in progress. Will be available.

Step LPF of low-pass filter operation
1 to LPF 8 are performed using the double precision mode in the multiply-add processor 30 and the narrow band sinc ² FIR filter coefficients generated by the triangular filter coefficient generator 37. Thus, two instructions are required for each filter operation. During that first instruction, the XP flag is "
Set to 1 ", β _wb is selected for coef_in, the filtered signal is selected for data_in, the least significant word of the associated accumulator is selected from RAM 6 for acc_in, Then, the result is written back to the RAM 6 to update the least significant word of the accumulator, and during the second instruction, the XP flag is restored to “0” and β _wb is set to coef_in
Again and the filtered signal is
Again selected for data_in, the most significant word of the associated accumulator is selected from RAM 6 and the result is added to the carry from the first instruction as described above and stored in RAM 6 Written back to update the most significant word of the accumulator. In the following, the sequence of these two instructions is represented by the function xp () to describe the operations from LPF1 to LPF8. ^{_{v ak = xp (z -1 v}} ak + β wb * v sk) (19) i ak = xp (z -1 i ak + β wb * i sk) (20) p a = xp (z -1 p a + Β _wb * p _s ) (21) q _a = xp (z ^-1 q _a + β _wb * q _s ) (22)

In practice, the function selection operation shown in FIG. 20 is performed by specifying only two desired functions in the MAP program and the CORDIC program. No operations are performed that are not required to form the two desired outputs, and MAP3
0 is designed to have enough memory and operation speed for the most required function combination (ie, the combination of the three-phase effective power P and the three-phase reactive power Q).

[0146]

Those skilled in the art, after reading the foregoing disclosure, should be able to design several digital processors embodying the present invention, including: -Some are used in the field of engine control outside the field of metering, and this should be noted when interpreting the scope of the appended claims.

[Brief description of the drawings]

FIG. 1 is a schematic diagram of a monolithic integrated circuit used in connection with a power metering system, which includes an oversampling analog-to-digital converter and a digital signal in a bit-serial format. A cascaded digital signal processor operating according to one of the cascaded digital signal processors embodying the present invention, comprising:
Selectively perform RDIC, non-restoring division, or non-restoring square root calculation.

FIG. 2 is a schematic diagram of a control circuit section including a circuit for interfacing with a programmable read only memory for the power measurement system of FIG. 1;

FIG. 3 is a diagram showing program information stored in a programmable read only memory.

FIG. 4 is a timing chart for an access operation of a programmable read-only memory.

FIG. 5 is a schematic diagram showing a random power supply for the power measurement system of FIG.
FIG. 2 is a schematic diagram of a control circuit section including a circuit for interfacing with an access memory.

FIG. 6 is a timing chart for an access operation of a random access memory.

FIG. 7 shows an analog of oversampling in FIG.
FIG. 2 is a schematic diagram of a multi-channel decimation filter used between a digital converter and a cascaded digital signal processor.

FIG. 8 is a schematic diagram of a bit-serial multiply-add processor that is a front processor in the cascaded digital signal processor of FIG. 1;

FIG. 9 is a schematic diagram showing the connection of the bit serial multiply-add processor of FIG. 8 with other elements in the power measurement system of FIG. 1;

FIG. 10 is a schematic diagram of a zero-cross detector used in the power measurement system of FIG.

11 is a triangular filter coefficient generator used to generate filter coefficients for performing a sinc ² digital low pass filter operation using the bit serial multiply-add processor of FIG. 8; FIG.

FIG. 12 is a timing diagram showing a waveform plotted against a time axis in relation to the triangular filter coefficient generator of FIG. 11;

FIG. 13 is a schematic diagram of a bit serial processor embodying the present invention, ie, the last processor of the cascaded digital signal processor of FIG. 1 that selectively implements CORDIC, divide, or square root calculations. .

FIG. 14 is a schematic diagram of a scaler used in the CORDIC processor of FIG.

FIG. 15 shows an arctan, as schematically shown in FIG.
FIG. 4 is a schematic diagram illustrating in more detail the interface between an arithmetic based read only memory and a CORDIC processor.

FIG. 16 is a schematic diagram showing the connection of the bit serial CORDIC processor of FIG. 13 with other elements in the power measurement system of FIG. 1;

FIG. 17 is a diagram showing an instruction format of a CORDIC processor.

FIG. 18 is a coding table for input selection and function selection as used in the format of FIG. 17 in the CORDIC processor instructions, where the instructions are executed by the monolithic integrated circuit CORDIC processor of FIG. 1; What is used.

FIG. 19 is a block diagram of a process executed in a multiplication-addition processor of the monolithic integrated circuit of FIG. 1;

FIG. 20 is a block diagram illustrating processing executed in a CORDIC processor of the monolithic integrated circuit of FIG. 1; 19 and 20 correspond to the left-hand part and the right-hand part of the composite block diagram of the processing performed in the monolithic integrated circuit of FIG. 1, and the composite block diagram here is referred to as FIG. 19 in the specification. You.

[Description of Signs] 1, 2, and 3 main conductors; 11-16 current transformer; 18 analog reference generator; 19 register bank; 20 6-channel decimation filter; 21-26 sigma-delta modulator; 29 PROM control; Multiplication-addition processor; 36 zero-cross detector; 37 triangular filter coefficient generator; 38 ARCTAN coefficient ROM; 39 normalizer; 40 CORDIC processor;

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 17/10 G06F 7/52 320 G06F 7/552 G01R 19/00 JICST file (JOIS)

Claims (3)

(57) [Claims] 1. increment by one for each successive step
According to the constant i, the numbering starts from zero and continues sequentially.
CORDIC, division, calculated in a procedure with steps
Alternatively, select a series of steps in the square root
The digital processor to run, each signal input port is provided,
The first data through the signal input port for the temporary storage data of
Initial step of the CORDIC procedure as digital accumulation input
Before being configured to receive the first vector component at times
A first accumulator, which is temporarily stored and updated
As the first digital accumulation input, each of the CORDIC procedures
During the subsequent step, the first accumulation result of the previous step is received.
And the signal input port of each temporary storage data.
Through the port as the first digital accumulation input.
Receives the dividend input signal during the initial step of the arithmetic procedure
The first accumulator configured as described above, and
At each subsequent step of the division procedure, it is temporarily stored and updated.
In the previous step, as the new first digital accumulation input
The first dividend configured to receive a first dividend of
It is an accumulator and the signal of each temporary storage data
Signal as the first digital accumulation input
At the initial step of the square root calculation procedure
Receiving the signal and each step of the square root calculation procedure
Sometimes a temporarily stored and updated first digital accumulation input
As a force, the first opened / closed number at the previous step is received.
The first accumulator configured in such a manner that the first
Temporarily store such digital accumulator results
Accumulator means for; each signal input port
And each output port is provided.
Through the signal input port for the temporary storage data of
As a digital accumulation input, the initial step of the CORDIC procedure
Receiving a second vector component during the CORDIC procedure
Updated second digital accumulation during each subsequent step of
As an input, the second accumulation result of the previous step is received,
Via the signal input port of each temporary storage data
As a second digital accumulation input, the first digital division
Receiving a divisor input signal during the
During each subsequent step of the second, the temporarily stored and updated second
Receives the divisor signal of the previous step as the digital accumulation input of
And a continuous second digital accumulation unit
Second accumulator means for temporarily storing the result;
Each signal input port and each signal output port
Provided as a third digital accumulation input,
Zeroth of any function among CORDIC, division and square root calculation
Temporarily stores arithmetic zero at the step of
At each step of the procedure, an estimate of the tangent angle of each arc
Is temporarily stored, and at each step of the division procedure, the quotient is
Temporarily store the approximate value of
At each step of the procedure, calculate the approximate value of each square root.
A third digital, configured to store in time
Third accumulator means for temporarily storing an accumulation result
And at each step of the CORDIC procedure according to i
Means for generating a tangent radix of the arc to be formed; said CORDIC procedure
And the first accumulation result at each step of the division procedure
Before updating the first
Means for comparing the accumulation result with zero; and the square root
Updating the first accumulation result at each step of the calculation procedure
Before each of the first accumulation results
To compare the trial estimate for the number of actuations
Means; at each step of said CORDIC procedure,
Responsive to the comparator result being both zero or positive,
Making adjustments to the first accumulator means so that the first digital
Subtracts the current second accumulation result from the accumulation result,
As a result, the resulting sum is doubled and the updated first
Means for generating a digital accumulation result; said CORDIC
In each step of the procedure, at least zero or positive
In response to the comparator result, the second accumulating means is adjusted.
The second digital accumulation result and the current first
2 in the digital accumulation result of^{-(i + 1)}Multiplied by
Means for generating a new second digital accumulation result
And at least in each step of the CORDIC procedure
Responsive to the comparator result being zero or positive, the third
To the third means of digital accumulation
Add the tangent radix of the arc determined according to the result and i,
Procedure for generating an updated third digital accumulation result
Steps; less than zero at each step of the CORDIC procedure
Responsive to the comparator result being
With adjustments, the first digital accumulation result and the second accumulation
Double the sum of the results obtained by adding the calculation results and update
Means for generating a first digital accumulation result obtained;
Update to less than zero in each step of the CORDIC procedure
Responding to each of the first digital accumulation results before
The second accumulator is adjusted and its second digital
Is added to the current first digital accumulation result.^{-(i + 1)}
Is decremented by the product multiplied by
Means for generating a result of summation of the totals; said CORDIC procedure
The comparator result being less than zero in each step of
In response to the third accumulation means,
The tangent of the arc obtained from the digital accumulation result of 3 according to i
Generates updated third digital accumulation result by subtracting radix
Means for generating; in each step of said division procedure
At least in response to the comparator result being zero or positive.
In response, the first accumulation means is adjusted and the first accumulation means is adjusted.
From the digital accumulation result, 1 of the second digital accumulation result
/ 2 is subtracted to generate an updated first digital accumulation result.
Means for generating; in each step of said division procedure
At least in response to the comparator result being zero or positive.
In response, the second accumulation means is adjusted and the second accumulation means is adjusted.
2 for the digital accumulation result and the current first digital accumulation result
^{-(i + 1)}Multiplied by 2 and the updated second digital
Means for generating an accumulation result; each step of said division procedure
Said ratio being at least zero or positive in step
Responding to the comparator result and adjusting the third accumulating means.
Calculated according to the third digital accumulation result and i
A third digit updated by adding the tangent radix of said arc
Means for generating an accumulation result; each of the division procedures
Responsive to said comparator result being less than zero,
In response, the first accumulation means is adjusted and the first accumulation means is adjusted.
The result obtained by adding the digital accumulation result and the second accumulation result
Updated the first digital accumulation result by doubling the sum of the results
Means for generating a result; each step of said division procedure
The first data before being updated to less than zero.
Adjusting the second accumulation means in response to the digital accumulation result.
In addition, the second digital accumulation result is stored in the current first data.
2 for digital accumulation result^{-(i + 1)}Is decremented by the product of
To generate an updated second digital accumulation result
Means in each step of the division procedure,
Responsive to the comparator result being full, the third accumulating means
, And from the third digital accumulation result to i
Therefore, the third radix updated by subtracting the tangent radix of the arc obtained is
Means for generating a digital accumulation result of;
In each step of the root calculation procedure, the first digital accumulation
The calculation result is at least a trial estimate for the number of switches
Responsive to the comparator result being equal to or greater than
The first digital accumulation result by adjusting the calculation means
2 ^{-(i + 2)}And the sum of the third accumulation result is subtracted and updated.
Means for Generating a Generated First Digital Accumulation Result
And in each step of the square root calculation procedure, a first
Each digital accumulation result is a trial estimate for the number of switches
Responsive to the comparator result being less than
Adjusting the stage, its first digital accumulation result and 2
^{-(i + 2)}And the sum of the third accumulation result is added and updated.
Means for generating a first digital accumulation result;
In each step of the square root calculation procedure, the first digital
The cumulative total is at least a trial summary
Responding to the comparator result which is equal to or greater than the calculated value,
The third digital accumulation by adjusting the accumulation means of 3
Result and 2^{-( i + 1)}And the updated third digital accumulation
Means for generating a calculation result;
Processor. 2. The method according to claim 1, wherein each of the procedural steps is performed at intervals of two consecutive words, and the comparison of the first successive accumulation results is performed in an earlier word period of the two consecutive words in each of the preceding procedural steps. , Wherein the updating of the first, second and third accumulation results is performed in a word period after the two consecutive words in the procedural steps from zero onwards. The digital processor of claim 1. 3. Increment by one for each successive step
According to the constant i, the numbering starts from zero and continues sequentially.
CORDIC, division, calculated in a procedure with steps
Alternatively, select a series of steps in the square root
The digital processor to run, each signal input port is provided,
The first data through the signal input port for the temporary storage data of
Initial step of the CORDIC procedure as digital accumulation input
Before being configured to receive the first vector component at times
A first accumulator, which is temporarily stored and updated
As the first digital accumulation input, each of the CORDIC procedures
During the subsequent step, the first accumulation result of the previous step is received.
And the signal input port of each temporary storage data.
Through the port as the first digital accumulation input.
Receives the dividend input signal during the initial step of the arithmetic procedure
The first accumulator configured as described above, and
As a stored and updated first digital accumulation input:
In each subsequent step of the division procedure,
The first dividend configured to receive a first dividend of
It is an accumulator and the signal of each temporary storage data
Signal as the first digital accumulation input
At the initial step of the square root calculation procedure
Receiving the signal and also storing the updated first
Each step of the square root calculation procedure is used as a digital accumulation input.
At the time of backup, the first
The first accumulator configured in such a manner that the first
Temporarily store such digital accumulator results
Accumulator means for; each signal input port
And each output port is provided.
Through the signal input port for the temporary storage data of
As a digital accumulation input, the initial step of the CORDIC procedure
Receiving the second vector component and updating the updated second
Each successive scan of the CORDIC procedure
At the time of the step, the second accumulation result of the previous step is received,
Via the signal input port of each temporary storage data
As a second digital accumulation input, the first digital division
During the initial step, the divisor input signal is received and temporarily stored.
As the updated second digital accumulation input.
During each subsequent step of the arithmetic procedure, the absolute value of the divisor input signal is
A series configured to receive a signal corresponding to the value
To temporarily store successive second digital accumulation results
Second accumulator means; a respective signal input port and
Each signal output port is provided and the third data
Digital accumulation input, CORDIC, division, square root calculation
Arithmetic zero at the zeroth step of any function
Temporarily memorize it and store it at each step of the CORDIC procedure.
The approximate value of the tangent angle of each arc is temporarily stored, and
At each step of the calculation procedure, temporarily calculate the approximate value of each quotient.
At each step of the square root calculation procedure.
Configured to temporarily store the approximate value of each root
Temporarily stores the third digital accumulation result
Third accumulator means for performing each step of the CORDIC procedure.
The tangent radix of the arc determined according to i
Means for performing the steps of the CORDIC procedure and the division procedure.
Before updating the first accumulation result at the time of
The first accumulation result is compared with zero
Means at each step of said square root calculation procedure
Before updating the first accumulation result,
And trials for the first accumulation results and the number of switches to be opened and closed.
Means for performing an estimate comparison; each of said CORDIC procedures
Wherein the step is at least zero or positive
Adjusting the first accumulator means in response to the comparator result.
The current second accumulation is obtained from the first digital accumulation result.
The result of the operation is subtracted, and the resulting sum is 2
Multiply to produce an updated first digital accumulation result
Means for: at each step of said CORDIC procedure,
Respond to the comparator result being at least zero or positive
Then, the second accumulation means is adjusted and the second
Digital accumulation result and current 1st digital accumulation result
^{-(i + 1)}Multiplied by 2 and the updated second digital
Means for generating an accumulation result; each of said CORDIC procedures
Wherein the step is at least zero or positive
Adjusting the third accumulating means in response to the comparator result;
Calculated according to the third digital accumulation result and i
A third digit updated by adding the tangent radix of said arc
Means for generating a cumulative result;
In each step, the comparator result that is less than zero
Responding and making adjustments to said first accumulator.
Obtained by adding the digital accumulation result and the second accumulation result
Doubled sum of results, updated first digital accumulation
Means for generating a result; each step of said CORDIC procedure.
Before updating to less than zero in the first
In response to each digital accumulation result, the second accumulation means is adjusted.
The second digital accumulation result is added to the current first
2 in the digital accumulation result of^{-(i + 1)}Multiply only by the product
To generate an updated second digital accumulation result.
Means for: at each step of said CORDIC procedure,
Responsive to the comparator result being less than zero, the third accumulation
The third digital accumulation result by adjusting the calculation means
Is updated by subtracting the tangent radix of the arc obtained according to i
Means for generating a third digital accumulated result;
The absolute value of the first accumulation result is equal to the absolute value of the second accumulation result.
In each step of the division procedure that is smaller than the log value,
The magnitude of the first accumulation result is at least zero or positive.
In response to the comparator result, the first accumulation means is adjusted.
The second digital accumulation result is added to the second digital accumulation result.
デ ジ タ ル of the digital accumulation result of
Means for generating a digital accumulation result of the division;
In each step of the procedure, at least zero or positive
In response to the comparator result, the third accumulator is adjusted.
The third digital accumulation result and 2^{-(i + 1)}To
Add to produce an updated third digital accumulation result
Means for calculating the absolute value of the first accumulation result,
Steps of the division procedure that are smaller than the absolute value of the accumulation result of step 2
Response to the comparator result being less than zero at
Then, the first accumulation means is adjusted and its first data is
Digital accumulation result and half of the second digital accumulation result
To produce an updated first digital accumulation result.
Means for; in each step of said division procedure,
Responsive to the comparator result being less than zero, the third data
Digital accumulator means to adjust its third digital accumulator
2 from the calculation result^{-(i + 1)}Only the third digit
Means for generating a tall accumulation result; said square root calculation
In each step of the procedure, each first digital accumulation result
Is at least equal to the trial estimate for the number of
Responsive to the comparator result being above, the first accumulating means
To the first digital accumulation result.
^{-(i + 2)}And the sum of the third accumulation result is subtracted, and 2
Multiply to produce an updated first digital accumulation result
Means for; calculating at each step of said square root calculation procedure
Where the first digital accumulation result is
Responding to the comparator result being less than the trial estimate,
Adjusting the accumulating means of the first and accumulating the first digital accumulating means
Double the result and update the updated first digital accumulation result
Means for generating; said division procedure or said square root
In each step of the calculation procedure, a first digital accumulation
The result is at least the same as the trial estimate for the number of switches
Responsive to the comparator result being greater than or equal to, the third accumulation
Adjusting the means, the third digital accumulation result and 2
^{-(i + 1)}, And the updated third digital accumulation result
Means for generating the digital program.
The Sessa is that each procedural step consists of two consecutive bit serial
Each of the above-mentioned procedural steps
In the comparison of the first accumulation result,
Occurs during the first word period of the serial word,
The update of the first, second, and third accumulation results is zero.
And the 2-bit serial wireless
Performed during the second word period of the code.
Digital processor.