Floating Point vs. Fixed Point

I'm reading discussions that say multiple instances of plugins degrades the overall audio quality.
Is this a possible limitation of floating-point ?

One Ohm wrote:
I'm reading discussions that say multiple instances of plugins degrades the overall audio quality.
Is this a possible limitation of floating-point ?
LOL!!!
I don't know what forums you're reading this from, but that is an absolute lie.
Cheers

Similar Messages

IHow to force display of trailing zeros for fixed point numbers?

I have an 8 bit unsigned fixed point number, with 7 integer bits and 1 fractional bit, so the desired delta is 0.5. I want it to always display the fractional bit, even when that bit is 0. In other words, as this number is incremented, I want to see:
0.0
0.5
1.0
1.5
etc.
But instead I'm seeing:
0
0.5
1
1.5
etc.
I set the display format to Floating point, with 1 digit of precision. The "Hide trailing zeros" checkbox is NOT checked. What am I doing wrong?
I realize I could convert to single precision for display purposes to make this happen, but I'd rather make this work with fixed point.
Thanks,
-Ron

I just rustled up a VI with what I understand your input to be - unsigned 7 bit mantissa and 1 bit exponent input, and have it displaying 2 decimal places.... Is this what you're after or did I miss it?
- Cheers, Ed
Attachments:
zeros.vi ‏7 KB

Issue with fixed point number output from 9219 module for full bridge measurement (cRIO/FPGA)

Hi,
I have a question regarding the fixed point output acquired from a 9219 module (in FPGA on a cRIO) when setup to acquire a strain measurement (full bridge).
Software: Labview 2009
Hardware: cRIO-9012 (NI-RIO 3.2.1), NI-9219 module
The 9219 module is configured in the project as follow:
Ch0: Full-Bridge +/- 7.8mV/V
Ch1: Voltage +/- 4V
Ch2: Voltage +/- 15V
Ch3: Voltage +/- 60V
The calibration mode is 'Calibrated' so the FPGA outputs fixed point numbers.
My issue is that the precision of the fixed point number for Ch0 (strain) is "(fixed point <+/-32,-1>[-2.5e1,2.5e1]:1.16e-10)", which indicates that the fixed point number is a value between +/- 0.25 and not the +7.8mV/V as I expected?
The fixed point number does not change in precision whether the range for the strain measurement is set at +/- 7.8mV/V or +/-64mV/V (the two available options).
As the fixed point number doesn't change precision I'm assuming that changing the range of the strain measurement changes the resolution of the acquired number? And as such I will need to perform additional scaling on the fixed point number to convert it to the expected range?
There is no mention of scaling of the voltage or strain measurements mentioned in any documentation or examples, with the only scaling example provided for the thermocouple measurements.
Any help/clarification is much appreciated.
Regards,
Mike

Hello Mike,
Hopefully I can help clarify some of the behavior you are seeing.
My issue is that the precision of the fixed point number for Ch0 (strain) is "(fixed point <+/-32,-1>[-2.5e1,2.5e1]:1.16e-10)", which indicates that the fixed point number is a value between +/- 0.25 and not the +7.8mV/V as I expected?
For calibrated values on the FPGA VI, the returned data is a Voltage measurement, not a directly calculated strain value. Based on the specified ranges for the 9219 in a Full bridge configuration, 250 mV will encompass all possible input values, at the module provided excitation (2-2.7V dependent on the sensor gage resistance).
The fixed point number does not change in precision whether the range for the strain measurement is set at +/- 7.8mV/V or +/-64mV/V (the two available options).
As you may notice from the fix-point definition, the fixed point data contains 32-bit precision which is larger than the acquired precision of 24-bits provided by the 9219. The fixed point data-type is coded to accept input encompassing both the range and precision of the instrument; such that, no additional coercion of the input data values is required based on user-defined software settings, i.e. the bridge sensitivities +/- 7.8mV/V or +/-64mV/V.
As the fixed point number doesn't change precision I'm assuming that changing the range of the strain measurement changes the resolution of the acquired number? And as such I will need to perform additional scaling on the fixed point number to convert it to the expected range?
The documentation does not clearly define that by varying the discrete levels of strain input (+/- 7.8mV/V or +/-64mV/V) the range of the ADC on the module is also adjusted. I am working to follow up further on this topic, to provide a clarification on the module documentation. As for scaling, the voltage values acquired regardless of the ADC resolution will still be related strain via the bridge sensitivity. The resolution of the ADC will simply define the smallest measurable change in the strain value.
There is no mention of scaling of the voltage or strain measurements mentioned in any documentation or examples, with the only scaling example provided for the thermocouple measurements.
For converting the acquired voltage values to a strain measurement, I would recommend the documentation linked here for a detailed explanation on strain calculation. Often, users will forward the acquired voltage data as fixed point values through a DMA FIFO to the RT controller on the Compact RIO; such that, they may handle conversion from voltage to strain using floating point math in real-time.
I hopes these responses provided a bit of clarity. I will continue to work to provide additional information on the 9219 specification information. Please post back any further questions.
Cheers!
Edit: Forgot to add the link.
Message Edited by Pcorcs on 04-14-2010 04:55 PM
Patrick Corcoran
Application Engineering Specialist | Control
National Instruments

Exponential to fixed point number convert

Hello All,
I am getting below result for the double value.
double d = 7777777.77;
System.out.println(d); O/P >> 7777777.77
double d1 = 88888888.88;
System.out.println(d1); O/P >> 8.888888888E7
Actually I dont want the value to be in Exponential format. I found that when the decimal is before 7 digits , am getting correct value (without exponential format). But when decimal is after 7 digits then value gets in exponential format.
If this is the default behaviour which we cant change then how to convert the exponential format number to fixed point number.
Thanks
Prasad.

Don't use the default conversion on double numbers except for debugging. Convert floating point numbers to String format either with String.format, or with a DecimalFormat object. Either gives you proper control over the format.

Division without Fixed Point Arithmetic

Hello,
     I am trying to divide 1 by a decimal, i.e. 1 / 7.75027, in Include RV63A900 of Function Group V61A. Since the ' Fixed Point Arithmetic' checkbox is not checked, I am unable to get this calculation to work.
     I have tried different scenarios of using a floating point field to receive the quotient but nothing has worked.
Any suggestions?
Thank you,
Chris Mowl

I use floating variable some thing like this :
data: v_f TYPE f,
         v_c TYPE char12.
      v_f = ( wa_komp_o-netwr / o_komp-mglme ).
* Due to decimal issues had to use Floating point value
* for calculation of condition base value
      CALL FUNCTION 'FLTP_CHAR_CONVERSION'
        EXPORTING
          decim = 5
          input = v_f
        IMPORTING
          flstr = v_c.
      v_amount = v_c * 10.

How do I find all the fixed points of a...

How do I find the number of fixed points in a randomly increasing sequence of integers in O(log n) time?
(1 1 1 1 5 7) as input will return 1. Because 5 is the fifth element.

Consider this example instead of the erroneous example given above.
Input: {-1, 0, 2, 4, 5, 6, 7}
Output: 4
I need an algorithm to find the number of fixed points of an increasing sequence in O(log n) time.
Yes. The sequence is known beforehand.
I was considering something like a binary search, but it would take more than O(log n) to find the number of fixed points.
So currently I am thinking of using something like an averaging scheme, but I haven't found any solution yet.

In need of Fixed-point algorithm

hi peeps,
Im in need ( a very urgent need ) of a fix-point algorithm , with explanations if possible. Can anybody point me to the right direction?
thanks....

See if this helps you,
To find a solution to p=g(p)given an initial approximation p0.
Input: Initial approximation p0; tolerance TOL; maximum number of
iterations N0.
Output: Approximate solution p or message of failure.
Step 1: Set i=1;
Step 2: While i<=N0 do Steps 3-6.
Step 3: Set p=g(p0). (Compute p.)
Step 4: If |p-p0| < TOL then
OUTPUT (p); (Procedure completed successfully.)
STOP.
Step 5: Set i=i+1.
Step 6: Set p0=p. (Update p0.)
Step 7: OUTPUT('Method failed after N0 iterations, N0=', N0);
(procedure completed unsuccessfully.)
STOP.
Fixed-Point Iteration Code in C Language
* Fixed-Point Iteration Code in C Language
* To find a solution to p = g(p) given an
* initial approximation p0
* INPUT: initial approximation; tolerance TOL;
* maximum number of iterations NO.
* OUTPUT: approximate solution p or
* a message that the method fails.
#include<stdio.h>
#include<math.h>
#define true 1
#define false 0
main()
double TOL,P0,P;
int I,NO,FLAG,OK;
FILE *OUP[1];
void INPUT(int *, double *, double *, int *);
void OUTPUT(FILE **, int *);
double absval(double);
double G(double );
INPUT(&OK, &P0, &TOL, &NO);
if (OK)
OUTPUT(OUP, &FLAG);
/* STEP 1 */
I = 1; OK = true;
/* STEP 2 */
while((I<=NO) && OK)
/* STEP 3 */
/* compute P(I) */
P = G(P0);
if (FLAG == 2)
fprintf(*OUP, "%3d %15.8e\n", I, P);
/* STEP 4 */
if (absval(P-P0) < TOL)
/* procedure completed successfully */
fprintf(*OUP, "\nApproximate solution P = %12.8f\n", P);
fprintf(*OUP, "Number of iterations = %3d", I);
fprintf(*OUP, " Tolerance = %14.8e\n",TOL);
OK = false;
else
/* STEP 5 */
I++;
/* STEP 6 */
/* update P0 */
P0 = P;
if (OK)
/* STEP 7 */
/* procedure completed unsuccessfully */
fprintf(*OUP, "\nIteration number %3d", NO);
fprintf(*OUP, " gave approximation %12.8f\n", P);
fprintf(*OUP, "not within tolerance %14.8e\n",TOL);
fclose(*OUP);
/* Change function G for a new problem */
double G(double X)
double g;
g = sqrt(10.0 / (4.0 + X));
return g;
void INPUT(int OK, double P0, double TOL, int NO)
char AA;
printf("This is the Fixed-Point Method.\n");
printf("Has the function G been created in the program immediately preceding\n");
printf("the INPUT function?\n");
printf("Enter Y or N\n");
scanf("%c",&AA);
if ((AA == 'Y') || (AA == 'y'))
*OK = false;
printf("Input initial approximation\n");
scanf("%lf",P0);
while(!(*OK))
printf("Input tolerance\n");
scanf("%lf", TOL);
if (*TOL <= 0.0)
printf("Tolerance must be positive\n");
else
*OK = true;
*OK = false;
while (!(*OK))
printf("Input maximum number of iterations - no decimal point\n");
scanf("%d", NO);
if (*NO <= 0)
printf("Must be positive integer\n");
else
*OK = true;
else
printf("The program will end so that the function G can be created\n");
*OK = false;
void OUTPUT(FILE **OUP, int *FLAG)
char NAME[30];
printf("Select output destination\n");
printf("1. Screen\n");
printf("2. Text file\n");
printf("Enter 1 or 2\n");
scanf("%d", FLAG);
if (*FLAG == 2)
printf("Input the file name in the form - drive:name.ext\n");
printf("For example: A:OUTPUT.DTA\n");
scanf("%s", NAME);
*OUP = fopen(NAME, "w");
else
*OUP = stdout;
printf("Select amount of output\n");
printf("1. Answer only\n");
printf("2. All intermeditate approximations\n");
printf("Enter 1 or 2\n");
scanf("%d", FLAG);
fprintf(*OUP, "FIXED-POINT METHOD\n");
if (*FLAG == 2)
fprintf(*OUP, " I P\n");
/* Absolute Value Function */
double absval(double val)
if (val >= 0)
return val;
else
return -val;

Why is "x2^0" Displayed on Fixed point numbers in hex format

I have a requirement to display a fixed point value (unsigned 20bit value) in hex format. At present when I set the numeric indicator to display this I get a "x2^0" string attached to the end.
What is this and is there anyway of turning it off?
I am trying to give a display that replicated some old style hex mechanical switches, by selecting a fixed size font and limiting the display to only be big enough to display the five hex digits (minimum field length 5 with pad with zeros) I get almost what I want (the "X2^0" is hidden of right hand side)
However when I delete characters from the number (to say enter a new number) the "X2^0" part then shows. See attached jpg which shows the format string.
This is by no means a show stopper but just annoying, is there some where to turn this formatting string off?
Dave.
Attachments:
20 bit hex.JPG ‏52 KB

It is showing how many of those bits are for the decimal places. It is kind of important information.
But since you are dealing with an Unsigned Integer (20,20 means there is no fractional part), why not just use a U32. You can set the display of the indicator for that to be %05x to just display the lower 5 hex characters.
There are only two ways to tell somebody thanks: Kudos and Marked Solutions
Unofficial Forum Rules and Guidelines

Usage of NOT having Fixed Point Arithmetic?

Hello!
SAP recommends to always do programs with Fixed Point Arithmetic attribute checked. Please don't explain here what is Fixed Point Arithmetic - I know this and it's is not my question.
The question is: why do we have it at all? Why aren't all programs simply fixed-point-arithmetic without attribute to activate or deactivate this?
I suppose that there are situations which require Fixed Point Arithmetic NOT being set, but I can't imagine any. Could anyone offer an explanation?
Thanks in advance!
Igor
Edited by: Igor Barbaric on Mar 21, 2008 9:30 AM
No replies - I'll try to bring the question to your attention once again:
Please, does anybody know the purpose of NOT having Fixed Point Arithmetic?

Please check if this helps u
This attribute defines, whether for numerical data objects of data type p (packed numbers) is the decimal point is relevant in calculations or assignments or if it is omitted. Normally, this attribute is always switched on and you might ask, why it can be switched off at all. Well, in very ancient times packed numbers had no real decimal point. It was taken into account only for presentation purposes. In order to stay downward compatible, for the introduction of fixed point arithmetic this switch was needed. In fact there are still programs that switch fixed point arithmetic off ¨C for example financial programs that calculate on the basis of cents instead of dollars.
Thanks Arjun

Since upgrading to Lion on start multiple programs open (word, excel, mail, power point) any fixes?

Since upgrading to Lion every time I start my Mac multiple programs open (word, excel, mail, power point) any fixes?

CJLinst wrote:
Yeah, but when it CRASHES and you have to hold the power button, all the apps restart. I have 8GB and it's swap-city.
To make matters worse, it appears there's a bug in Lion Mail. It opens a compose window for every email I sent in the previous session. We're talking dozens and dozens of windows.
Maybe try posting your own question on this as it is unrelated to this post. Never seen it before, so if you require help, then open a new topic.
Cheers

Fixed point division on FPGA doesn't show the fraction part

Hi,
I'm trying to normalize a fixed-point complex vector. My problem is in the part of fixed point division. Assuming I am dividing x by y (x/y), then I multiply first x by 2^26 and y by 2^9 then divide both of them using high throughput function then multiply the result by 2^-17 and here the problem appears. The result is zero, and the fraction part doesn't appear. My question is: does the "scale by power of 2" block discards the fraction part and only keeps the integer part, even if the input is a fixed point input with fraction part of 20 bits length? If so then how can I get the result of dividing a number by a bigger one? I mnean how can I display a result < 1? If not, then why isn't the fraction part displayed? The numbers I used as an example are: x=10, y=182.
The indicator is adjusted to have 64 bit word length and 44 integer length. I also write this data into a memory element and then read it through a FIFO to transfer it to the RT vi. Both of the memory and the FIFO are configured with the same word and integer lengths above. I am using 7965R series FPGA.
Thanks in advance.

It's worth noting that I tried this same procedure on RT vi itself and the fraction part was displayed.

How do you zoom out from a fixed point?

Okay, I'm not sure if this is possible, but...
My first video clip in a scene is pretty boring, since there's no action, just a guy typing at his keyboard. To add in some motion, I'd like to have it start out so that all you can see are his eyes, but then it zooms out to show the whole scene.
So what I'd want to do is somehow set his eyes as a fixed point and then zoom out from there. However, I have no idea how to go about doing this, since I don't even know how to zoom out. I know how to zoom in, but not out.
Could someone please help me?

Adjust position and scale using keyframes.
Fixed effects
Animation and keyframes
Animate motion in the Program Monitor

User Exit Fixed Point Arithmetic

Dear all,
I have developed some code on the user exit RV61AFZB and tried to change the value on the xkomv-kbetr. But, i found that the division result is wrong.
Example:
DATA: temp like xkomv-kbetr,
           c_test LIKE xkomv-kbetr VALUE '152000.00',
           c_qty like xkomv-kbetr value '3.00'.
   temp = c_test / c_qty.
The value of temp became 506.67. Since, this is a user exit and i can't change program attribute "Fixed Point Arithmetic" .
Regards,
Kit

Hello
Try this:
DATA: temp like xkomv-kbetr,
           c_test LIKE xkomv-kbetr VALUE '152000.00',
           c_qty like xkomv-kbetr value '3.00'.
data: temp1 type i.
   temp = c_test / c_qty.
   temp1 = temp.

Need arithmetic operation result in Fixed point arithmetic disabled prg

Hi,
I am writing an enhancement which is in standard SAP program which is not enabled for 'Fixed arithmetic' calculations.
I need to do some actual arithmetic operation involving packed numbers in my enhancement.
Say
p1 = '2.00'
p2 = '12.00'.
result = p1 * p2.
result coming is 2400.00. This is because Fixed point arithmetic is disabled in standard program.
I need a result of 24.00 . Is there any solution to this apart from dividing it with 100?
Regards,
Dhana

based on P1 and P2 values you need to divide the value with 100 or 1000 as per the number of decimals on P1 and P2

Setting a White Point and Black Point

I would like to suggest for future versions of lIghtroom to have an ability to set the white point and black points similar to levels and curves in Photoshop. This would be in addition to setting middle grey in the white balance settings currently in Lightroom. I find that the method of using Threshold to find the closest white and black pixels and then using these points to set the black and white points in curves or levels provides a better way to fix a color cast than the methods available in Lightroom.

MB,
There is no real need for Black and White point eyedropper type tools in Lightroom. By holding down the Alt key while changing the Blacks or Exposure sliders the display will show the areas that are being clipped. This is much faster than looking around for the darkest and lightest areas manually. If you just like to click to taste (like using the White balance dropper) then forget the Alt key and just move the sliders until you get the desired look.
Remember that the problem with Photoshop is that there are so many ways to do the same thing that it has become difficult and bloated. Lightroom strives to keep the interface as clear as possible.

How can I find a turning point or inflection point on the curve?

I have encountered a problem, is seeking a turning point or inflection point on the curve.
The curve is monotonically increasing (slope is essentially the same); the slope tends to 0 after the turning point. Because of the curve is not actually a straight line, so the derivative of point by point seems comparison difficult.
First, I intercepted two new curves from the curve. One with monotonically increasing slope and another with remained constant slope. And I transform the two new curves to two absolutely straight with “regression” filter. Then find the intersection point of two absolutely straight. Through the X coordinate of the intersection point to find the corresponding Y value on the curve.Turning point is (x, y).
However, this method can only be used for a fixed curve. Because the range of the intercept values need to give the range, if the new changes in the overall length of the curve, the value of this range may no longer apply.
My question is: Is there a better way, or Diadem functions associated itself can solve this problem?
Any of your ideas and suggestions is important to me, thanks very much.
Best Regards
Jie
Solved!
Go to Solution.
Attachments:
inflection point.JPG ‏131 KB

Hi J. Huang,
I would recommend the following generic approach, based on the pictures you submitted:
1) Create a smoothed channel from the raw data curve to remove random variations
2) Calculate a derivative/slope channel from the smoothed channel
3) Use the ChnFind() function to find the points where...
       the slope is close to zero
        the slope is positive
        the raw data is positive
        the raw data is close to the maximum channel value
Brad Turpin
DIAdem Product Support Engineer
National Instruments

Floating Point vs. Fixed Point

Similar Messages

Maybe you are looking for