How to realize subtraction on differential signals?

[diego.fan]

Hi all, could you please give me some hint?
I'd like to realise subtraction on differential signals. My purpose is to cancel common mode signal and to amplify differential mode signal.
How to make subtraction on voltage?

1. Because the frequency is lower than 100Mhz, I cannot use inductors.
2. Because the peak to peak of differential signals is from 0 to Vdd. It means after subtraction, the range will be from Vdd to -Vdd. Which kind of circuit can realise this large output range?

Thanks.

 

[crutschow]

How large is the common-mode signal?

 

[betwixt]

If the peak to peak is 0 to VDD it doesn't mean you get a negative result. Differential signals are normally two identical waveforms but with one inverted with respect to the other. If you add them you get zero or at least any common offset they share. For example if 'A' is zero, 'B' would be VDD and when 'A' is VDD then 'B' would be zero.

The whole idea of canceling common mode signals is that they exist equally and in the same polarity on both wires. If you follow normal practice of inverting one of the signals and then adding them, the common mode is canceled (+CM) - (+CM) = 0, but the differential signal is doubled (+DS) - (-DS) = 2DS.

At 100MHz you can easily use an amplifier with differential inputs to do this.

Brian.

 

[diego.fan]

How large is the common-mode signal?

There is no dedicated value. To be honest, I just want to cancel the noise from Vdd

 

[mtwieg]

1. Because the frequency is lower than 100Mhz, I cannot use inductors.

Why not? Some baluns can work down in the <1MHz range.

 

[BigBoss]

There is no dedicated value. To be honest, I just want to cancel the noise from Vdd

The noise signal is not coherent on each diff signal line so you cannot extract the noise from them easily.

 

[FvM]

The noise signal is not coherent on each diff signal

The ambiguity of "noise". Common mode interferences can be expected to be coherent on both lines, that's the nature of common mode.

Seriously speaking, the question is rather vague. Both points 1 and 2 don't hold, as already mentioned.

1. Common mode chokes can be well used to reduce common mode noise at various frequencies. Its effect reduces for lower frequencies of the common mode interferences, but it does not depend on the frequency range of the useful signal.
2. Consider that a differential receiver/amplifier has a gain. It can be smaller than unity if appropriate. You can always adjust it so that the output signal fits the available swing.

How does VDD come into play?

Is your primary objective high SNR or suppression of strong CM signals?

 

[diego.fan]

Why not? Some baluns can work down in the <1MHz range.

Really? But balun needs inductors which has big loss in low frequency.

- - - Updated - - -

The noise signal is not coherent on each diff signal line so you cannot extract the noise from them easily.

My idea is: suppose the noise from Vdd(Vdd is supply voltage. The noise should be common mode) can affect outputPlus and outputMinus signal. If I want to cancel supply noise. There are two ways:
1. I can use one of "output signal-Vdd" to cancel Vdd noise. Because their noise are correlated, I think it can cancel the supply noise.
2. Why not use "outputPlus-outputMinus" to cancel supply noise? Because at every time, the supply noise on outputPlus and on outputMinus should be correlated. I think this way is better. Because I can enlarge output swing from "0~Vdd" to "-Vdd~Vdd", which means I enlarge signal range. It can help improve SNR.

- - - Updated - - -

How does VDD come into play?

Is your primary objective high SNR or suppression of strong CM signals?

VDD is the supply voltage. At the same time, because I use differential inverters. The signal swing is also "0 to VDD".
My primary objective is to cancel the noise from supply voltage.

 

[betwixt]

I'm curious to know what signal with 0V to VDD swing is prone to noise. Is it a logic signal, in which case there are digital solutions or is it analog in which case it is impossible to remove VDD noise when the signal is at VDD level. By definition, any noise on VDD will also be on a signal at the same voltage.

Can you explain what your circuit actually does, what is it's purpose?

Brian.

 

[diego.fan]

I'm curious to know what signal with 0V to VDD swing is prone to noise. Is it a logic signal, in which case there are digital solutions or is it analog in which case it is impossible to remove VDD noise when the signal is at VDD level. By definition, any noise on VDD will also be on a signal at the same voltage.

Can you explain what your circuit actually does, what is it's purpose?

Brian.

Hi Brian， this is a differential ring oscillator. I want to reduce phase noise further. Because the noise of supply voltage can be counted into 1/f noise I think. As I think, because the supply noise on two output are coherent, I think maybe I can use this way to cancel supply noise.

 

[betwixt]

I'm not sure subtracting differential signals will improve phase noise. It really depends on how you generate the differential signal. It may be that two out of phase signals are already present in the ring oscillator but unless they are exactly inverted the differential method wont work. Differential signals are used where there is a risk of common mode noise being introduced between source and destination and it can effectively cancel it, but, if the noise is already present in the source I doubt it will have any effect.

Brian.

 

[KlausST]

Hi,

I agree.

An example:
Imagine the two differential signals.
One is close to VCC, the ither is close to GND.
If the VCC is noisy .. it may influence the signal that is close to VCC,
but I doubt that it will influence synchronously the close_to_GND signal with the same magnitude.

Klaus

 

[FvM]

I presume, for a differential ring oscillator "noise" is only of interest as far it degrades the phase noise. Thus a differential comparator is the perfect device to convert the output into single ended signal.

True analog "substraction" of oscillator signals or keeping the "Vdd to -Vdd" differential swing doesn't seem to make sense.