DIYHiFi.org

Again, x-axis is less than FFT range.



See what I mean about the 1/f corner?

Jocko

First, this something that I have run into the ground over the last...........I dunno.......20 years. Ask Phred. He hates it. Almost quiet enough, but it has some 1/f problems.



Next is another emitter follower-based design. I see this used a lot in low-noise designs. It has a RC filter before the follower, but you wouldn't know it by looking at this. And it uses the mystery op-amp!



Now that I have a working '797 preamp, I can see better why these designs don't cut it. (Hint: 1/f may be related to regulation.)

Jocko

Here is the baseline noise floor:



The regulator, mystery op-amp"



Jocko

Just remembered that I left the 3 kHz LPF on the input of the FFT. Above plots don't mean much.

Well, anyway, here is something a bit more useful. I think. Log plots, 20-20 kHz. Sans LPF. (Extra gain from working preamp may alleviate need for LPF.)

First, the regulator we are discussing:



Looks believable. Numbers higher than before, so even more believable.

Now, here one for you to figure out. The emitter follower style, log plot.



Now someone has to explain the noise peaking. OK, it has a RC filter, 1 Hz or so, right before the emitter follower. I imagine if you do complete transfer function, since the RC filter is inside the feedback loop, that the answer becomes clear. Just like when you figure out the response of a servo, when you complete the circuit from the amp input to the output.

I guess. I hate designing with feedback, as I am lazy and don't want to do Bode plots, root locus or any other of that stability criteria stuff.

(The fact that I prefer the sound of no global feedback is a bonus. But sometimes you have to use it for a regulator.)

And one last plot, the old Jocko-style, in log form:



This has the same RC filter, but it is outside of any feedback loop. Still looks good, but that 1/f problem that does not show up here.........

Jocko

Regulator with '5534. Looks believable, with very low noise higher up. Higher 1/f corner than "misery" op-amp.



All for now. I think.

Jocko

Wow, you've been busy tonight..
quite a progress.

With the working 797 preamp you should be able to see the
thermic noise of a 50 Ohm resistor vs. input short circuit.
1KOhm = 4nV/sqrtHz, so 250R = 2 nV/sqrtHz, 62R5 = 1 nV/sqrtHz


btw. I like the bootstrapping trick in fig. 17 of
http://waltjung.org/PDFs/Getting_the_Mo ... rences.pdf

It allows deploying a lot of filter capacitance without paying
a price in terms of leakage.

R3 = 100 Ohm already adds more noise than the 797 itself
but the 797 needs it probably to survive. I was concerned with C3.
If we short the output to gnd, this would kick the opamps
inverting input to -5V or even -10V for a moment.

I used a circuit from the Wenzel web site as a preamp with 2 * 2sk389
that gave me some 700 pV/sqrt Hz. Then I made a differential version
with 2 * 8 FETs for better DC stability.

2->4 to get even less noise, 2 * 4 for going differential
and then 2*8 to make up for the 2 differential noise sources in series.

Was an awful lot of work matching JFETs. All in all I'm not happy with it.
(There must be a photo of this board somewhere on this site)
Given the high 1/f corners of the FETs I'll probably end up with
MAT-03s as in:

http://www.femto-st.fr/~rubiola/#journal-articles

and there in particular:

http://www.arxiv.org/abs/physics/0503012

Less ado and probably better at low frequencies.

regards, Gerhard

_________________
Everything has been said already - but not yet by everyone. (Karl Valentin)

Hey Gang,

Jocko called and informed of the recent progress. A little history... about 7 months ago I was talking to the Crystek engineers about their 0.5ps jitter specs. I asked how they tested and what they used to power the damn thing to get that kind of spec. The engineer was pretty proud of the power supply. He said by far this is the most important requirement to get the specs and the noise of the regulator needs to be below 20nV to even get close to achieving that figure. He of course would not really talk about how they got there.... anyways I know I need another 7 months and I will be designing oscillators... baby steps.

I tried literally 100's of designs. Jocko and I also tried some of these new RF units made for cell phones claiming lower than 200nv. Most of these things sucked... I don't know what they were drinking but we could not get them under 3uV and if we used the damn bypass cap it was sometimes allot worse.

CG suggested to me a Maxim app note:

http://www.maxim-ic.com/appnotes.cfm/an_pk/3656

Charlie Hansen sent his regulator over and I tried that. It's pretty cool but I respect his wishes and will not divulge.

Jocko and I burned up allot of talk time and we tried just about everything. I have 3 proto boards, each one having like 5 regulators on them and a place for the oscillator to do testing.

One thing I realized was that depending on the oscillator the needs changed. The Crystek really like to see feedback applied. I have not tried Jocko's but hope to see how well it works. One of my problems is that all my input voltages are preregulated at 8V's.

I settled for now on a variation of the Maxim appnote. I did not start with a 150uV noise unit and try and get out the stuff but a 1uV unit with a simple 2N5089.

BTW the GPIB thingy is from Prologix's they sell on ebay for like $150. Nice little unit.

Most of my measurements were originally done with a 3561A with the bubble memory option. I would avg rms over 32 cycles from 0-50hz save that off then short the inputs and run it again and subtract the two.

I got rid of the 3561a and bought a calibrated 35660a and it's killing me. For some reason when I subtract the two the measurement is going up.

I can say though from looking at these screens for like 4 months that it's coming in about 10nv from 25hz up and it's at about 22nv at between 1-5hz and then goes up to about 1uv at 0.125mHz.

I will try and get some better results with the 35660a and post them. But I'm a mac guy and don't have any of my pc's on the net for obvious reasons. So I will transfer some measurements and when I get Jocko's unit check it out.

I don't put a preamp on the unit like he does. I worry then too much if the damn preamp is throwing off the measurements. This stuff is too silly as it is.

~~~~~~~~

In regards to smd or not.... man I wish Jocko would have shown me these results months ago. SMD proto sucks!!!!!

Anyways mine is all smd and it looks pretty good. None of the resistors are really pulling any current. I will quote Charlie Hansen here...

Charlie had suggest to me to stay away from the thick film SMD resistors. So I have been checking out the thin film units from Vishay.

For leaded resistors I use the Shinko Tantulum. The tempco on these is great and they are non magnetic and still pretty easy to get values but they are pricey.

Jocko tell me what values you are using and I can send you some.

~~~~~~~~~

Caps suck... You can finish what looks to be a great layout and find out the caps just can turn the unit from on to WOW! I have both the Wolfson and ESS DAC's working in my units now. The Wolfson because of voltage output requires some caps for biasing and has 3 analog supplies and a digital supply. The proto of the Wolfson basically had ceramics on the board and it sounded pretty darn well. When I started fooling around with caps and had a frankenstein dac running it was really amazing what it sounded like.

I lucked out on my RC filter in my regulator. I needed a 100uF at 6.3 and Digikey was out of the UCC Polymer units so I bought some of their low esr units. They worked great... but when I did get the polymers in there, I thought my readings all went to hell. The noise in the 10hz and under went up a good 20dB.

Well I will build up some 24 bit modules today and test the power supplies and see what I get.

Thanks
Gordon

_________________
Gordon
Chief Scientist
Wavelength Audio

20nV. I'm trying to get my head around that number. I'm assuming this is <2Hz. I almost have to believe you're talking about a thermal chamber, or maybe a peltier device attached to the active parts to get a supply that quiet. Then again, there are tricks out there...

Isn't pcb leakage and solder joints going to affect this? I've had a heck of a time to get a noise floor below 500nV due to parasitics; solder, cables, blah, blah.

Mike,

Jocko is getting less than me with leaded components. If you look at the Y axis at the top of the graph you can see he is right around 1.3nV of noise.

I don't think 20nv is nothing big, I would actually like to see it lower.

~~~~~~~

Jocko,

On the feedback resistor thingy... are you bypassing that with a cap? As I remember the sulzer/jung stuff I remember they use to bypass the feedback resistor with something huge.

Thanks
Gordon

_________________
Gordon
Chief Scientist
Wavelength Audio

Fair enough. I was thinking the top of the scale was 100n. Duh.

That's exactly the capacitor in Fig.17 of the Jung reference I gave in my post above.

Gerhard

_________________
Everything has been said already - but not yet by everyone. (Karl Valentin)

There is no feedback resistor. Unity gain. Reference voltage is adjusted down by a divider.

The graph with the peaking is the Maxim app note version. i may try some other plots with different caps on the output.

Oh, one thing that I forgot to mention...........

The '5534 is not unity-gain stable. You have to stick a cap from pin 5 to pin 8. There is no pad on the pcb for it. I solder a leaded cap onto the pins. 100 pF in my case, as it is the smallest Wima FKP. But I don't intend to use a '5534, so I left it off. For those of you who will, this is still DIY. You can easily stick a cap on where and when needed.

Glad to know that the guys who make this are willing to admit what most of us have suspected all along: you have to have a super quiet, super secret PSU to get the jitter numbers that they claim.

Translation: you won't get those numbers with a '317, 7805, or any other 3- or 8-legged chip.

Jocko

Mama don't let your babies grow up to use Bode plots...

Most oscillator companies seem to be overly optimistic on their phase noise performance to begin with, and I don't imagine most of their customers want to hear that they have to really try in order to get good results.

Trying a 9 V battery as an experiment might be illuminating as well. I do that at work when making a spread-out-over-the-bench breadboard. If nothing else, it scares the willies out of the management guys who might pass through.

Jocko,

Well I am ready to throw the HP out the window.... For some reason it tried to subtract time measurements from spectral and freaked out. I think there is something wrong with the save and recall state. I am going to have to go through it again.

When I did get it too work I had 22.197nv of noise at 1hz. I then tried to plot it and the Prologix unit errored out with an error #2 serial steam error.

I don't get the Hump you get in your maxim thingy... I also checked it with the Prism and nothing up too more than 150kHz. Actually pretty steady all the way out. My RC is set at 0.318hz so a little higher than the maxim. But as all app notes they are missing some critical issues. Like the output even being on... I think that maxim regulator requires at least a few mills to turn on and yet they don't even talk about that. If the regulator is off then willy nilly it's not a happy puppy.

CG, with regards to the osc companies... I think the main problem is their dependence on our power supplies. It would be really hard for them to give all kinds of specs without first giving a primer on power supply design.

Later, beer time for me.
Thanks
Gordon

_________________
Gordon
Chief Scientist
Wavelength Audio

Clarke said it goes away with 220 uF. I used 180 uF. And 0.1 uF on the output.

(But it is one of those evil emitter follower types.............so who cares?)

Jocko