current mirror ratio V.S. noise

If noise is important, should the current mirror ratio set to scale down, instead of scaling up?


Case 1: Bias current is 1uA, mirrored up by X10, then the current is 1uA*10=10uA.

Case 2: Bias current is 100uA, mirrored down by ÷10, then the current is 100uA÷10=10uA.


So case 2 has smaller noise, why?

If the current is used for the tail current of VCO, the noise is important. Should I use case 2 bias? Thanks.

The current source noise is related to its transconductance and by minimizing Gm you are able to minimize the noise of current source. Of course, current mirror works as current amplifier. By amplifying a current (with ratio > 1) you amplifying a noise of biasing device also.

Using a large capacitor at the gate of the transistors, you can bypass the noise from the diode connected device. So you only have to minimize the noise of the mirroring device which is related to its transconductance. Depending on the W/L and the overdrive voltage you can find the design with smaller transconductance.

-Mahdi

Should I add a resistor and cap to form an RC filter, instead of only the capacitor? Thanks.

You only need to place a capacitor between the gate and ground if you are using an nmos current mirror. For pmos current mirror put the capacitor between gate and vdd.

Distinguish between noise gain by the mirror rack, and noise generated
within the mirror rack. You say the noise is smaller but you don't say
why, where the noise comes from or what kind. Maybe it's that the
master is the main source and making it 100X area instead of 1X, is
the whole deal. Or maybe it's in a larger circuit where noise is passed
in on the reference current line, and less gain = better.

Shunt C works for HF noise but isn't much help for 1/f noise.

dick_freebird said:

Shunt C works for HF noise but isn't much help for 1/f noise.

Click to expand...

Reducing the flicker noise requires a large W and L.

mahdi3999 said:

Reducing the flicker noise requires a large W and L.

Click to expand...

100uA with 200mV over the threshold gives 0.5...1mS. I will use the last one. Using 1pF cap you can get 160MHz 3dB bandwidth. To reach say 160kHz, you need 1nF. It is hard to integrate. Not to mention the gate leakage of the cap! It can deteriorate the mirror performance.

- - - Updated - - -

shanmei said:

Should I add a resistor and cap to form an RC filter, instead of only the capacitor? Thanks.

Click to expand...

The resistor will add more noise, so you won't gain much.

- - - Updated - - -

shanmei said:

If noise is important, should the current mirror ratio set to scale down, instead of scaling up?
Case 1: Bias current is 1uA, mirrored up by X10, then the current is 1uA*10=10uA.
Case 2: Bias current is 100uA, mirrored down by ÷10, then the current is 100uA÷10=10uA.
So case 2 has smaller noise, why?
If the current is used for the tail current of VCO, the noise is important. Should I use case 2 bias? Thanks.

Click to expand...

The output device noise will stay the same. The input device's contribution to the output noise is approx. 4kT*gm_out/gm_in. You can only modify the input device's gm and thus its noise. Therefore case 2 will result less noise, but you pay for it with lots of power, which might be better spent elsewhere.

In a VCO the tail current source is rarely the most critical noise source. If it is true in your case, then I suspect the design is far from optimal.