Device and method for approximating the square root of a number
专利汇可以提供Device and method for approximating the square root of a number专利检索，专利查询，专利分析的服务。并且A device and method for approximating the square root of a binary number N. The device includes hardware for storing N, and a result register for storing x.sub.n, where x.sub.n is a successive approximation of r. The device also includes hardware for iteratively replacing x.sub.n with x.sub.n+1, starting with n=0, where x.sub.n+1 =(Ax.sub.n +N-x.sub.n.sup.2)/A, and x.sub.0 is within a predetermined range. "A" is a multiple of 2 so that an operation involving a product or quotient with "A" is effected by a shift operation.，下面是Device and method for approximating the square root of a number专利的具体信息内容。
What is claimed is:1. A device for approximating the square root r of a binary number N comprising:a result register for storing x.sub.n, where x.sub.n is a successive approximation of r;a first shift register connected to an output of said result register for shifting x.sub.n a predetermined number of places in the direction of its most significant bit, to form a number Ax.sub.n ;a first adder connected to an output of said shift register for adding Ax.sub.n to N, to form a number Ax.sub.n +N;a multiplier connected to the output of said result register, for multiplying x.sub.n by itself, to form a number x.sub.n.sup.2 ;a second adder connected to outputs of said multiplier and first adder for subtracting x.sub.n.sup.2 from Ax.sub.n +N, to form a number Ax.sub.n +N-x.sub.n.sup.2 ; anda second shift register connected between an output of said second adder and input of said result register for shifting Ax.sub.n +N-x.sub.n.sup.2 a predetermined number of places in the direction of its least significant bit.2. The device of claim 1 wherein N is between 0.5 and 2.3. The device of claim 2 wherein said predetermined number of places for said first and second shift registers is 2 when N is between 1 and 2.4. The device of claim 2 wherein said predetermined number of places for said first and second shift registers is 1 when N is between 0.5 and 1.5. A device for approximating the square root r of a binary number B, expressed in scientific notation, comprising:an exponent register for storing the exponent of B;a mantissa register for storing the mantissa of B;a first shift register connected to an output of said mantissa register;a first multiplexer connected to an output of said shift register and the output of said mantissa register for selectively providing a number N at an output thereof;a result register for storing x.sub.n, where x.sub.n is a successive approximation of r;a second shift register connected to an output of said result register for shifting x.sub.n a predetermined number of places in the direction of its most significant bit, to form a number Ax.sub.n ;a first adder connected to respective outputs of said second shift register and multiplexer for adding Ax.sub.n to N, to form a number Ax.sub.n +N;a multiplier connected to the output of said result register, for multiplying x.sub.n by itself, to form a number x.sub.n.sup.2 ;a second adder connected to outputs of said multiplier and first adder for subtracting x.sub.n.sup.2 from Ax.sub.n +N, to form a number Ax.sub.n +N-x.sub.n.sup.2 ; anda third shift register connected between an output of said second adder and input of said result register for shifting Ax.sub.n +N-x.sub.n.sup.2 a predetermined number of places in the direction of its least significant bit.6. The device of claim 5 further comprising:a third adder connected to an output of said exponent register for adding "1" to said exponent;a second multiplexer connected to the output of said exponent register and an output of said third adder for selectively providing a number E at an output thereof; anda fourth shift register connected to an output of said second multiplexer for shifting E one place in the direction of its least significant bit.7. The device of claim 6 wherein said first shift register shifts the mantissa of B one place.8. The device of claim 7 wherein said first shift register shifts the mantissa of B in the direction of its least significant bit.9. The device of claim 8 further comprising:a control line connected between said exponent register and said first and second mutliplexers to provide the least significant bit of the exponent of N as a control signal thereto.10. The device of claim 9 further comprising:an exponent result register for receiving the output of said fourth shift register.
The present invention relates to electronic devices and methods for making numerical approximations. More particularly, it relates to a device and method for approximating the square root of a binary number.
BACKGROUND OF THE INVENTION
Electronic devices for calculating the square root of a binary number are widely available. Such devices may employ either dedicated hardware, software or microcode. Irrespective of structure employed, most devices tend to use the direct calculational method, the Newton-Raphson iteration method or some variant of the Newton-Raphson method.
The direct method is the binary version of the decimal pencil-and-paper method. The direct method is slow due to the large number of iterations required to obtain the result. For example, the IEEE double precision data format with 53 significant bits requires 53 iterations to complete the calculation. In addition, each iterations requires many machine cycles.
The Newton-Raphson iteration method for the square root of a number N is based on the formula
xn+1 =(xn +N/xn)/2, (1)
where xn is the square root of the nth iteration and xn +1 is the square root of the (n+1)st iteration. This method has an advantage over the direct method in that a fewer number of iterations are required. However, like the direct method, the Newton-Raphson method requires a relatively large number of machine cycles. This is due to the rather lengthy calculation for the quotient N/xn.
A modified versions of the Newton-Raphson method is based on the formula
xn+1 =xn (3-Nx2)/2 (2)
This modified version is much faster because it only requires multiplications and a subtraction and no long division. It should be noted that division by 2 in a binary operation is implemented by a shift operation, and therefore requires much less time than a long division. However, the modified equation converges to the reciprocal of the square root of N, thus requiring one long division at the end of the computation. The modified Newton-Raphson method also requires three (3) multiplication steps for each iteration. Attempts to improve on this method have focused on speeding up the multiplication operation. This is generally expensive because it necessitates additional hardware. Another problem with both the original and modified Newton-Raphson methods is that the number of iterations needed to obtain a stated precision is unknown. This prevents accurate prediction of the calculation time for a given problem.
OBJECTS OF THE INVENTION
It is therefore an object of the present invention to provide a new and improved device and method for approximating the square root of a binary number.
It is another object of the present invention to provide a device and method for approximating the square root of a binary number which requires no long divisions.
It is a further object of the present invention to provide a device and method of successively approximating the square root of a binary number with only one multiplication per iteration.
It is yet another object of the present invention to provide a device and method for successively approximating the square root of a binary number in only three machine cycles per iteration.
It is yet a further object of the present invention to provide a device and method for successively approximating the square root of a binary number with a computational time of only one multiplication and one subtraction step per iteration.
It is still another object of the present invention to provide a device and method for approximating the square root of a binary number with a predictable calculation time for a given accuracy.
SUMMARY OF THE INVENTION
One form of the present invention is a device for approximating the square root r of a binary number N. The device includes means for storing N, a result register for storing xn, where xn is a successive approximation of r, and means for iteratively replacing xn with xn+1, starting with n=0. According to the invention, xn+1 =(Axn +N-xn 2)/A, where x0 is within a predetermined range and A is a multiple of 2.
Another form of the present invention is a method for approximating the square root r of a number L in a computer, L being a binary number between a and b. The equation xn+1 =(Axn +L-xn 2)/A is iteratively solved, starting with n=0; where x0 is within a predetermined range, A is a multiple of 2, and xn+1 is a successive approximation of r. Numeric operations involving A and another quantity are achieved by shifting bits in such quantity.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a graph of f(x)=x2 -N.
FIG. 2 is an enlarged view of a portion of the graph of FIG. 1.
FIGS. 3A and 3B show a block diagram of a device according to one form of the present invention.
FIG. 4 is a graph of the number of machine cycles required to approximate the square root of a binary number N for different resolutions.
DESCRIPTION OF A PREFERRED EMBODIMENT
FIG. 1 shows a graph of f(x)=x2 -N which is a convenient way of visualizing an aspect of the present invention. f(x) will be recognized as a parabola which crosses the y axis at the point -N. Of particular interest is the point r at which f(x) crosses the x axis. At point r, f(r)=0, so by substituting in f(x)=x2 -N
0=r2 -N
r2 =N and
r=N1/2
Therefore, the point r at which f(x) crosses the x-axis is the square root of N.
For purposes of the present description N is selected between 0.5 and 2. More specifically, 0.5<=N<2.
Examining f(x), it can be shown that the greatest value of its slope A occurs at x=2.
A=f'(x)=2x
A=f'(2)=4
Thus, the value of the slope at x=2 is 4
FIG. 2 shows more detail of the region of the curve f(x) between 0.5 and 2. A first point x0 is selected to serve as a first approximation of r. For example, x0 may be selected so that 0.5<=x0 <=2, but preferably x0 =0.5.
A second approximation of r is obtained by constructing a line with a slope of A, from (x0, f(x0)) to the x axis (x1, 0), where x1 is the next approximation of r. The equation of the line is:
A=[0-f(x0)]/[x1-x0 ] or
x1 =[Ax0 -f(x0)]/A
since f(x0)=x0 2 -N
x1=[ Ax0 +N-x0 2 ]/A
Subsequent approximations of r can be obtained by the equation:
xn+1 =[Axn +N-xn 2 ]/A (3)
In a preferred embodiment A=4. Because 4 is the largest slope of the curve in the region between 0.5 and 2, this assures that the root r is not "overshot" in each successive approximation. By selecting A=4 the binary product of A and xn in equation "(3)" is accomplished in the present invention by a shifting xn two places in the direction of the most significant bit. Furthermore, the binary quotient of [Axn +N-x1 2 ] and A is accomplished in the present invention by shifting the quantity [Ax1 +N-x1 2 ] two places in the direction of its least significant bit.
FIGS. 3A and 3B show a block diagram of a device 10 for approximating the square root of a binary number B. Binary number B is expressed in scientific notation by B=M×2m, where M is the mantissa of B and m is the exponent of B. M is a normalized value of B between 1 and 2, and m is an integer. Device 10 includes an exponent register 12 for storing m and a mantissa register 14 for storing M.
An adder 16 for adding 1 to m is connected to an output of exponent register 12. A multiplexer 18 is connected to the outputs of exponent register 12 and adder 16 for providing a number E at its output. A shift register 20 is connected between the output of multiplexer 18 and the input of exponent result register 22 for shifting E one place in the direction of its least significant bit.
A control line 24 is connected between the place of the least significant bit (LSB) of exponent register 12 and multiplexer 18. If the value of the LSB is "0", the output (m) of exponent register 12 is selected by multiplexer 18. If the value of the LSB is "1", the output (m+1) of adder 16 is selected by multiplexer 18.
A shift register 26 is connected to the output of mantissa register 14 for shifting mantissa M one place in the direction of its least significant bit. A multiplexer 28 is connected to the outputs of mantissa register 14 and shift register 26 and provides a number N at its output. Control line 24 is connected to multiplexer 28. If the value of the LSB in exponent register 12 is "0", the output (M) of mantissa register 14 is selected by multiplexer 28. If the value of the LSB is "1", the output (M-1) of shift register 26 is selected by multiplexer 28.
Device 10 also comprises apparatus 30 for providing a successive approximation for the square root of the number N provided at the output of multiplexer 28. Apparatus 30 includes a result register 32 for storing xn, where xn is a successive approximation of r. A shift register 34 is connected to an output of result register 32. Shift register 34 shifts xn two places in the direction of its most significant bit to form a number Axn when A=4. An adder 36 is connected to respective outputs of shift register 34 and multiplexer 28 for adding Axn to N, to form a number Axn +N. A multiplier 38 is connected to the output of result register 32, for multiplying xn by itself, to form a number xn 2. An adder 40 is connected to outputs of adder 36 and multiplier 38 for subtracting xn 2 from Axn +N, to form a number Axn +N-xn 2. A shift register 42 is connected between the output of adder 40 and the input of result register 32 for shifting the number Axn +N-xn 2 two places in the direction of its least significant bit.
Control logic 44 provides timing signals to the various components of device 10. Signals provided by logic 44 determine, among other things, the number of iterations apparatus 30 will incrementally update xn in result register 32.
The operation of the present invention provides successive approximations for the square root of a binary number B, where B is expressed the product of a binary mantissa M, 1<=M<2, and a binary integer exponent m. Mantissa M is stored in register 14 and exponent m is stored in register 12. Because device 10 is designed to perform the square root operation quickly, the exponent m must be an even number. Thus, if m is odd (LSB of m=1), the exponent m of B is increased by one and its mantissa M is reduced by a power of 2 in shifter 26. Thus, the number N provided to apparatus 30 is a binary number between 0.5 and 2.
As noted previously, the largest slope for the curve f(x)=x2 -N in the region between 0.5 and 2 is 4. Thus, the value of A is selected to be 4 and shift registers 34 and 42 both provide a two position shift of the respective number input thereto. More specifically, shift register 34 provides a multiply operation by shifting xn two places in the direction of its most significant bit, and shift register 42 provides a divide operation by shifting Axn +N-xn 2 two places in the direction of its least significant bit.
A somewhat faster operation can be achieved by noting that when N is between 0.5 and 1, the maximum slope in the xn region between 0.5 and 1 is 2. Thus, control logic 44 can provide a signal to shift registers 34 and 42, to only shift the incoming value one place when N is between 0.5 and 1.
Irrespective of whether shift registers 34 and 42 selectively change the number of shift places, apparatus 30 is effective for iteratively replacing xn with xn+1. This replacement occurs in the loop formed by apparatus 30 between the output of register 32 and its input. The value selected for x0, i.e. the first approximation to the square root of N is somewhat arbitrary. In a preferred embodiment, x0 =1 when 1<=N<2 and x0 =0.5 when 0.5<=N<1. A satisfactory result should be obtained as long as 0<=x0 <=r. The approximation of r is the product of xn in result register 32 and the exponent in result register 22.
In a preferred embodiment, the range for N is preselected between 0.5 and 2. By selecting a range for N in which the upper value of the range is four times greater than the lower value of the range, m can always be an even number. This is convenient because the square root of an even power of 2 can be obtained by shifting the decimal point in m one place in the direction of the most significant bit. Alternative ranges for M may be selected between any two binary numbers that differ by a factor of 4. For example, M could be selected between 1 and 4. In this event, adder 16 would subtract 1 and shifter 26 would shift M one place in the direction of its most significant bit. For 1<=N<4, A=8 (the slope at x=4).
Equation "(3)", xn+1 =[Axn +N-xn 2 ]/A, requires only three machine cycles for each iteration. FIG. 4 is a graph of the number of machine cycles required for different degrees of resolution. For example, 12 bit resolution requires 24 machine cycles or 8 iterations. The relationship is basically linear. An advantage of knowing the number of machine cycles for a given resolution includes the ability to preprogram control logic 44 with the number of iterations required of apparatus 30. Furthermore, if device 10 is assigned the task of approximating a square root, the time required will be known in advance.
It will be clear to those skilled in the art that the present invention is not limited to the specific embodiment disclosed and illustrated herein. For example, software and/or firmware embodiments are within the scope of the present invention.
Numerous modifications, variations, and full and partial equivalents can be undertaken without departing from the invention as limited only by the spirit and scope of the appended claims.
What is desired to be secured by Letters Patent of the United States is as follows.
| 标题 | 发布/更新时间 | 阅读量 |
|---|---|---|
| 一种用于逐次逼近型模数转换器的片外校准方法及系统 | 2020-05-13 | 505 |
| 用于微测热辐射计检测器的片上偏置校准和读出集成电路 | 2020-05-13 | 466 |
| 一种应用于CMOS图像传感器的高精度阵列模数转换器 | 2020-05-08 | 980 |
| 用于在不具有附加有源电路的SAR ADC中启用宽输入共模范围的方法和装置 | 2020-05-08 | 103 |
| 一种混合型模数转换电路装置及转换方法 | 2020-05-17 | 600 |
| 用于模数转换器的基于双比较器的误差校正方案 | 2020-05-17 | 495 |
| 图像传感器与图像读出方法 | 2020-05-17 | 291 |
| 一种逐次逼近型模数转换器及校准方法 | 2020-05-12 | 138 |
| 具有反冲线性化的逐次逼近寄存器模/数转换器 | 2020-05-16 | 700 |
| 基于压电陶瓷的动态断煤信号报警装置及其逻辑设计 | 2020-05-16 | 177 |
高效检索全球专利专利汇是专利免费检索,专利查询,专利分析-国家发明专利查询检索分析平台，是提供专利分析，专利查询，专利检索等数据服务功能的知识产权数据服务商。
我们的产品包含105个国家的1.26亿组数据，免费查、免费专利分析。
专利汇分析报告产品可以对行业情报数据进行梳理分析，涉及维度包括行业专利基本状况分析、地域分析、技术分析、发明人分析、申请人分析、专利权人分析、失效分析、核心专利分析、法律分析、研发重点分析、企业专利处境分析、技术处境分析、专利寿命分析、企业定位分析、引证分析等超过60个分析角度，系统通过AI智能系统对图表进行解读，只需1分钟，一键生成行业专利分析报告。
