## [Qucs-devel] RE: Noise wave analysis

 [Qucs-devel] RE: Noise wave analysis From: Stefan Jahn - 2005-11-24 07:36:18 ```On Mi, 23.11.2005, 17:08, Andy Koz wrote: > Stefan Hi Andy! > I have looked at the code and I now think I understand the operations, I > still think that the formulae > > res = (k * n * adjoint (k) - kelvin (T) / T0 * fabs (1 - norm (g)) * > d * adjoint (d)) * norm (1 / (1 - g)); > > might cause an error since g is set to -1 and fabs(1-norm(g)) is zero > (if my interpretation of the norm function is correct), looks like a > divide by zero may happen somewhere. 'fabs(1-norm (g))' is nowhere in the denominator, but in the numerator only. so i dont see a problem here. > Also, what is the purpose of expanding the noise correlation matrix? the purpose is to have an additional reference port, so you can check what happens if you put some kind of impedance at your device's source port (e.g. if s-parameters are measured mos/mes data). cheers, stefan. ```

 [Qucs-devel] Re: Noise wave analysis From: Stefan Jahn - 2005-11-23 09:57:34 ```On So, 20.11.2005, 00:09, Andy Koz wrote: > Hi Stefan Hello! > I have been looking for information on how to use Noise data in > S-parameter files and came across your web pages. Your document shows > Noise correlation matrices but some of the calculations don't seem to > make sense. > > The S-parameter container with additional reference port (noise wave > section) shows an equation with matrices K and D and the description > shows K' and D' as the adjoint. But the additional info shows these > matrices are not rectangular. In fact K' is *not* the transposed of K (same for D) in the description. This would have been indicated by a K^T. K as well as K' are square. D and D' are 1 column matrices. The A^+ matrices are adjoint. > How do you do a transpose of say the D > matrix which has only one column. The transpose of a one column matrix is a one row matrix. > Also the gamma may be short circuit so this would be -1, the part > |1-|Gm|^2| would result in infinity. It never occurs in the poles, only in the zeros. Or did I oversee something? > Also the equation would result in the subtraction of 2 different sized > matrices, this is also illegal. Which one? > Are you sure these equations are correct? I am quite sure, because it works in fact. Qucs can use the noise data in these S-parameter files correctly. But probably there is still an error in the documentation... we will see, right? Cheers, Stefan. ```
 [Qucs-devel] RE: Noise wave analysis From: Stefan Jahn - 2005-11-24 07:36:18 ```On Mi, 23.11.2005, 17:08, Andy Koz wrote: > Stefan Hi Andy! > I have looked at the code and I now think I understand the operations, I > still think that the formulae > > res = (k * n * adjoint (k) - kelvin (T) / T0 * fabs (1 - norm (g)) * > d * adjoint (d)) * norm (1 / (1 - g)); > > might cause an error since g is set to -1 and fabs(1-norm(g)) is zero > (if my interpretation of the norm function is correct), looks like a > divide by zero may happen somewhere. 'fabs(1-norm (g))' is nowhere in the denominator, but in the numerator only. so i dont see a problem here. > Also, what is the purpose of expanding the noise correlation matrix? the purpose is to have an additional reference port, so you can check what happens if you put some kind of impedance at your device's source port (e.g. if s-parameters are measured mos/mes data). cheers, stefan. ```