描述

Why do the Translate function outputs appear to be incorrect?

解决方案

Several examples and improved information on the data formats are available in the data sheet for the CORDIC v4.0 and later.

This is typically because of a misunderstanding of the output format.

The following are several example calculations of the Translate function. These examples are based on a CORDIC configuration of a 20-bit input and output:

Example 1:

This example is a modified version of the example design that can be found in the data sheet for the CORDIC on pages 9-10.

B = 0.707+0.25j;

mag = abs(B) % Magnitude

0.7498993265765745

theta = angle(B) % Phase Angle

0.3398843741305139

CORDIC SIMULATION

x_in <= "00101101001111111000" = 00.101101001111111000 = 0.707 (Q1.N

Format)

y_in <= "00010000000000000000" = 00.010000000000000000 = 0.25 (Q1.N

Format)

x_out <= "00101101010000000000" = 00.101101010000000000 =

0.7499275207519531 (Q1.N Format)

P_out <= "00001010111000000100" = 000.01010111000000100 =

0.3398742675781250 (Q2.N Format)

Example 2:

This example uses integer input values:

C = 10+5j

mag = abs(C) % Magnitude

11.18033988749895

theta = angle(C) % Phase Angle

0.4636476090008061

This is where it becomes a bit tricky; the format and range are such that you might need to scale your data before inputting it into the core.

First, this is what the 20-bit input would look like if the input were all integer bits:

Decimal 10 = 00000000000000001010

Decimal 5 = 00000000000000000101

Scale by N-2, which in this case is 20-2 = 18 (this is only to reinterpret the numbers and show that the calculated output from the CORDIC is correct).

10 / 2^18 = 0.000038146972656250

5 / 2^18 = 0.000019073486328125

SCALED CALCULATIONS

D = 0.000038146972656250+0.000019073486328125j

mag = abs(D) % Magnitude

0.00004264961199760036

theta = angle(D) % Phase Angle

0.4636476090008061

Following is what the core would do if those values were put straight into the core and the data reinterpreted:

CORDIC SIMULATION

x_in <= "00000000000000001010" = 00.000000000000001010 =

0.00003814697265625000 (Q1.N Format)

y_in <= "00000000000000000101" = 00.000000000000000101 =

0.00001907348632812500 (Q1.N Format)

x_out <= "00000000000000001100" = 00.000000000000001100 =

0.0000457763671875 (Q1.N Format)

P_out <= "00001110101100110010" = 000.01110101100110010 =

0.4593658447265625 (Q2.N Format)

If you want to reinterpret the output in the original format, you will need to scale the Magnitude output. If you scale the output value by 2^18, which in this case is how the input was scaled, then you will find that the output does equal the expected value.

0.0000457763671875 * 2^18 = 12

Example 3:

Another example, using larger integer numbers:

E = 1024+512j

mag = abs(E) % Magnitude

1144.866804479892

theta = angle(E) % Phase Angle

0.4636476090008061

Decimal 1024 = 00000000010000000000

Decimal 512 = 00000000001000000000

Scale by N-2, which in this case is 20-2 = 18 (this is only to reinterpret the numbers and show that the calculated output from the CORDIC is correct).

1024 / 2^18 = 0.003906250

512 / 2^18 = 0.001953125

SCALED CALCULATIONS

F = 0.003906250+0.001953125j

mag = abs(F) % Magnitude

0.004367320268554277

theta = angle(F) % Phase Angle

0.4636476090008061

Following is what the core would do if those values were put straight into the core, and the data reinterpreted:

CORDIC SIMULATION

x_in <= "00000000010000000000" = 00.000000010000000000 =

0.003906250000000000 (Q1.N Format)

y_in <= "00000000001000000000" = 00.000000001000000000 =

0.001953125000000000 (Q1.N Format)

x_out <= "00000000010001111001" = 00.000000010001111001 =

0.004367828369140625 (Q1.N Format)

P_out <= "00001110110101011010" = 000.01110110101011010 =

0.4635772705078125 (Q2.N Format)

If you want to reinterpret the output in the original format, you will need to scale the Magnitude output. If you scale the output value by 2^18, which in this case is how the input was scaled, you will find that the output does equal the expected value.

0.004367828369140625 * 2^18 = 1145

For a detailed list of LogiCORE CORDIC Release Notes and Known Issues, see (Xilinx Answer 29570).