-110dB Low Noise Ripple Filter

The output motorboats at about 1Vppk.

RL

Sorry - 400mVppk. 1V before the RC.

RL

Thanks. Delete C4, 220 uF. It was added to correct a small peaking at 455Hz and I forgot to check the .tran afterwards.

Deleting C4 has minimal effect on the overall response.

I was a bit concerned about adding this cap since it was in parallel with the unknown and variable impedance of the LED, so the effect was a bit unpredictable. The peaking shows there is a bit of loss of phase margin, so variations in the overall loop gain could cause more peaking or outright oscillations. This is a common problem with feedback loops running close to the limit, so there is nothing new there.

However, the -110 dB ripple rejection at DC makes it worthwhile to address any stability problems in real life. This can be checked by adding a step function and looking at the response. Any ringing or damped oscillation is a sign of trouble.

I'll have to see if there is another way to kill the peaking without causing oscillations. Changing R2 from 10K to 1K has the amazing effect of dropping the DC attenuation from -110 dB to -133 dB without causing oscillations, so I'll have to examine that further. Why that should have any effect at DC is mystifying.

It is easy to modify the circuit to monitor the open loop gain and phase, so I will have to do that and see what it tells us.

Thanks again for the feedback.

--
The best designs occur in the theta state. - sw

I modified the circuit to plot the open loop gain and got the most bizarre result ever. You can see this by downloading

To start, run 438B0376.ASC to see a normal Open Loop Gain plot. It shows

150 dB of gain at DC, falling to 0 dB at 1 MHz. The gain slope is a straight line.

Next, run 45216608.ASC. This plot is simply bizarre. The gain is 21 dB at DC up to 100 Hz, then a huge hump to 45 dB at 10 KHz. It does not look like the plot of an op amp with feedback.

I tried substituting a LT1028 which has been quite reliable in other circuits. The gain hump went up to 60 dB at 25 KHz.

I really don't know what is wrong, or how to fix it. This is going to take some work, but I'm sure there is a good lesson to learn here.

Thanks very much for the feedback. I would probably have gone on for the rest of my life oblivious of this problem.

--
The best designs occur in the theta state. - sw

This sort of thing doesn't have general purpose test applications, so it's hard to know what to look at.

As a power source, you can always apply simply input and output pulse stimulation to check stability and dynamic characteristics that are plainly visible in transient analysis. This can also be a check on the control loop nodes, biasing and start-up/ shutdown behavior.

RL

missing 2n7002 symbol

Under some circumstances, with some component locations, a simulation may show that this type of filter doesn't turn on. This gives really impressive noise rejection. ;-)

RL

RL

Thanks for the file. Interesting web site.

I have given up on that configuration. The open loop gain and phase plots can be bizarre.

I have focused instead on what it takes to develop good attenuation over the frequency band of interest. This is approximately as follows:

DC : 1mHz to 10 Hz midband : 10 Hz to 1 MHz RF : above 1 MHz.

DC requires some sort of reference voltage. Bangaps are out. They generate too much noise. Zeners also generate noise but it can be reduced by filtering. However, filtering has little effect on flicker noise (1/f).

LEDs offer lower noise than zeners but are not very good as voltage references.

Constant current diodes have the potential to give good noise performance, but they are not available in LTspice. Some examples are the Microsemi

1N5283 to 1N5314 series at

Capacitance multipliers can handle midband very well. However, they must be cascaded to reduce leakage through the device. They introduce phase shift which makes it difficult to use in feedback circuits.

Capacitor ESR and ESL affect the midband and RF response. Filtering above

1MHz is difficult.

Good shielding and grounding are required for high values of attenuation.

Here is a test circuit that uses the above guidelines and achieves the following performance:

1 mHz : -50 dB 1 Hz : -98 dB 10 Hz : -140 dB 100 Hz : -176 dB 1 KHz : -172 dB 10 KHz : -186 dB 100 KHz : -197 dB 1MHz : -174 dB

Shielding and grounding are likely to be the limiting factors.

--
The best designs occur in the theta state. - sw

It should be in your \lib\sym\misc directory, or maybe it's gone in LTspice XVII. I run IV and have no access to XVII files.

Rob had no problems loading the file, but I no longer use that approach. The open loop gain and phase can be bizarre.

--
The best designs occur in the theta state. - sw

It is there, but the looong list is not well sorted.

Insert a NMOS, and click it to edit the type. Click the Part No header field first to sort it, and pick

2N7002 from the start of the list.

--

-TV

I spotted two obvious wiring problems that I should have corrected before posting the original. Fixing them produces the following output:

Freq. Old New 1 mHz : -50 dB -47 dB 1 Hz : -98 dB -135 dB 10 Hz : -140 dB -203 dB 100 Hz : -176 dB -255 dB 1 KHz : -172 dB -269 dB 10 KHz : -186 dB -285 dB 100 KHz : -197 dB -287 dB 1MHz : -174 dB -256 dB

Everything above 1 Hz is essentially in the noise.

With this level of performance, the 2,000 uF caps are overkill. Smaller caps could be used that are much cheaper. The value could be adjusted to suit your needs.

Polymer caps usually have the lowest ESR. Radial caps normally have the lowest ESL.

--
The best designs occur in the theta state. - sw

One thing you don't seem to have gotten from the previous exercise, is that it's bad practice to power your linear regulator's reference from its own output.

Who's on first?

You're counting on leakage or luck to get any output voltage at all.

RL

Bite my tongue. Reference power is from preceding stage. No complaint.

RL

Thanks, Tauno. That worked and was easy. Steve must have a custom part in the .\misc lib using LTspice IV. Given the U1 designator, it is a sub circuit.

SYMBOL misc\\2n7002 112 96 R270 WINDOW 0 68 20 VRight 2 WINDOW 3 93 12 VRight 2 SYMATTR InstName U1

Thanks for the help on the 2N7002. My file is dated August, 2006, so it has been there a long time.

Please note: I have given up on that approach. The new filter crushes the LT3042/3045. See

--
The best designs occur in the theta state. - sw

Thanks for checking.

--
The best designs occur in the theta state. - sw

Am 03.03.21 um 20:00 schrieb Steve Wilson:

And below 1 Hz it is the noise.

1.6 uV/rt Hz @ 1 mHz if the spice models reflect reality.

But watch their leakage. I had a disaster when I wanted a source at AC GND but enforce the DC source current from a current mirror. On the drain side, you see more or less every electron that defects through the capacitor. Standard Aluminium is much better, and wet tantalum even more, but in the end it is a bad idea. You don't get rid of the effect.

-----------------------

• interesting if you need a just little bit of negative gate bias

• without DC/DC converter in an otherwise only +VCC system.

• (happened to be in my include file just above the HP LED).

• LED / Photogenerator pair Vishay .subckt VOM1271 A K P N D1 A K CAP D2 A 1 LED R1 A 1 2.4e6 R2 1 K 10m tc=-3m5, -18u C1 V A 0p1 C2 N K 2p G1 N V 1 K 155m D3 V N PV n=11 I1 V N TBL(0 0 0.3 80n 0.6 0u8 18 3u) D4 V P Do Q1 N V P Qo .model CAP d Rs=1 Cjo=100p M=1.2 Vj=0.4

• Is=1e-30 N=20 ; disable diode .model LED d Is=0n6 Rs=30m N=3.17 Xti=41 Eg=0.8 .model PV d Is=60p Rs=0.3 N=2.5 Cjo=20p Xti=45 Eg=0.5 .model Do d Is=5f Rs=8 N=1.7 Cjo=1p .model Qo pnp Is=1f Bf=200 Xtb=1.5 Vaf=100 BVbe=6.2

• Rb=200 Rc=10 Re=8 Cjc=0p5 Cje=1p .ends VOM1271

• the low noise HP/Avago/Broadcom LED. I did not check the model.

• Optically, it is quite a dim bulb. If you want light, take sth. else.

• I have the belly feeling: the brighter, the more noise. GHF

.model HLMP-6000 d (IS=93.2P RS=42M N=3.73 BV=4 IBV=10U CJO=2.97P VJ=.75 M=.333 Iave=0.04 mfg=Avago type=LED)

I think the Osram LEDs are in the LT wiki.

Gerhard

My LTspice shows 1uV/root(Hz) at 1mHz. This is flicker noise which every circuit has.

The noise drops to 33nV/root(Hz) at 1Hz and 14nV/root(Hz) at 10Hz, with a peak of 20nV/root(Hz) at about 1.6KHz.

It shows 1.08nV/root(Hz) at 10KHz and drops from there.

All my caps go to ground so leakage is not a problem.

You need my DC coupled preamp. It eliminates the leakage in an AC coupled circuit:

You can use two in a cross-correlation circuit:

Polymers are dry tantalum but do not explode and burn. They are self healing.

Thanks. I'll try that.

--
The best designs occur in the theta state. - sw