I believe you specify an input source, and the simulator computes the gain from input to output, and uses this to scale the output noise, resulting in input-referred noise.

It calculates the small-signal gain over the same frequency range that you did the noise sweep. So the gain is frequency dependent, as you say. Noise analyis computes the magnitude of each noise source, and then multiplies that by the transfer function from that noise source to the output, The same method of calculating transfer function is used for the signal input.

So you end up with a sum of all the noise powers at the output (over frequency). You can then divide this sum by the signal gain (again over frequency) to get the input referred noise (over frequency). Often the output-referred and input-referred noise are then integrated over a particular bandwidth to give you the total noise power in the bandwidth of interest.

Regards,

Andrew.

Well, it's certainly not a single gain in spectre, nor in other simulators that I've used. If computed that way, the input-referred noise would not be very useful. I can't really see any computational benefit in doing it that way either, since you're already computing frequency-swept transfer functions for all the noise sources, so adding one more for the input source is not exactly going to cost you much.

Can you post a reference to the papers that talk about the input-referred noise being computed using a mid-band gain?

Regards,

Andrew.

Geoffrey_Coram wrote on Aug 2^nd, 2006, 4:56am:


Yes, it is indeeded to guide the designer to measure SNR,  i.e.,in computing the IRN, using single mid-gain or DC gain.

Thanks Gerffrey, Andrew, the replies are very helpful to me.