Re: Fwd: qfunc/qfuncinv implementation in communications package

[Richardson, Anthony]

Alec - Maybe I misunderstood your method.  I thought you were talking about computing  these functions using numerical integration.  Thats always too slow and usually too inaccurate.  Rational approximations are normally used instead. There are a couple of nice rat approx for qfunc and qfuncinv in Abromowitz and Stegun for example.  (We don't need qfunc since we have erfc.  i wrote a qfuncinv in c from a & s today, but i think one is already in the dev code). 

 

Tony Richardson

 

 

 

------ Original Message ------
From: Alec Teal
Date: 1/25/2013 4:57 PM
To: Richardson, Anthony;
Cc: Przemek Klosowski;address@hidden;
Subject: RE: Fwd: qfunc/qfuncinv implementation in communications package

 

please post some of our chat Antony

"Richardson, Anthony" <address@hidden> wrote:

>Przemek Klosowski wrote:
>> On 01/24/2013 04:33 PM, Richardson, Anthony wrote:
>>
>> >
>> > Octave 3.6.3 returns zero for both forms:
>> > [x=[0:3:18]' qfunc(x) erfc(x/sqrt(2))/2]
>> >
>> >       0.000000000000000    0.500000000000000    0.500000000000000
>> >       3.000000000000000    0.001349898031630    0.001349898031630
>> >       6.000000000000000    0.000000000986588    0.000000000986588
>> >       9.000000000000000    0.000000000000000    0.000000000000000
>> >      12.000000000000000    0.000000000000000    0.000000000000000
>> >      15.000000000000000    0.000000000000000    0.000000000000000
>> >      18.000000000000000    0.000000000000000    0.000000000000000
>> >
>> > _______________________________________________
>> > Are you using the default display "format"  (scientific notation)?
>>
>> No, I did 'format long' to show full precision.
>
>Try it with one of the exponential formats.  With 'format short  e'.   I get:
>
>octave:24> [x=[0:3:18]' qfunc(x) erfc(x/sqrt(2))/2]
>ans =
>
>  0.0000e+000  5.0000e-001  5.0000e-001
>  3.0000e+000  1.3499e-003  1.3499e-003
>  6.0000e+000  9.8659e-010  9.8659e-010
>  9.0000e+000  0.0000e+000  1.1286e-019
>  1.2000e+001  0.0000e+000  1.7765e-033
>  1.5000e+001  0.0000e+000  3.6710e-051
>  1.8000e+001  0.0000e+000  9.7409e-073
>
>'format long' is a fixed point 15 digit format.
>
>Tony
>
>
>
>_______________________________________________
>Help-octave mailing list
>address@hidden
>https://mailman.cae.wisc.edu/listinfo/help-octave
>