Questions about low-noise low-voltage analog design

[Digital Larry]

I just made a board using the FV-1 and it's pretty noisy.  The noise seems to be smooth white noise-y.  Not like chirps you'd get from listening to a CPU data bus.  A tremolo patch absolutely modulates the noise, so the main concern seems to be the input side of things.  It is a 4-layer board with fairly solid GND/VCC planes.  Only vias, no traces.

On the input, you'll have a non inverting buffer (to get super high input impedance).  You'll have a DC blocking cap with high impedance load resistors at the input (high pass), biasing at Vcc/2, and possible gain shaping (e.g. high/low shelf) through the feedback and shunt networks attached to the op amp.  Another DC blocker to the DSP input.  High freq shunt cap to GND at DSP input.

On the output you have pretty much the same, perhaps minus the high freq shunt.

On my first protos I just picked some op-amps out of thin air and they have 21 nV/root Hz at 1 kHz.  They are not advertised as being great for audio.  There is a unity gain buffer on the input and gain-of-2 Sallen-Key low pass filter at 12 kHz on the output.

I could run my op-amps at 5 volts since I have 2 regulators on board, though they are currently hooked to 3.3.  This also would broaden the selection of available parts vs. 3.3 volts only.  No change in gain, just use a higher Vcc.

I can get 3.3V opamps from Linear down to about 3.3 to 4.5 nV/root Hz.  And it's an important first step to try.  If I go to 5 volts parts, the best I've found so far is 1.1 nV/root Hz and those are about 7 bucks onesie twosie from Digikey for a quad part.

And I know that keeping impedances low is good for reducing the accumulation of thermal noise.  Then the concern becomes how much current can your op-amps deliver into a reduced load resistor, how much supply current is available, and then what happens to your filter cap values, because they will increase, and can you still get a reasonable size/cost cap? 

I may branch out to other DSP things soon, but I am interested in learning a strategy to optimize noise performance where you are running op-amps in buffer/low gain amplifier or filter configurations that are very likely to find themselves wrapped around a DSP regardless of what that DSP is.

Thanks,

DL

[amz-fx]

It is probably not the op amps.

Do you have the recommended 1k and 2.2n filters on the stereo outputs? They are essential and should be close to the output pins of the FV-1.

You should drive the input of the FV-1 with a strong signal since the ADC needs to use a lot of its range to keep the noise down. For example, if you only have a 1v pk-pk signal going into the input, then it probably needs to be boosted closer to the ~3v pk-pk limit to improve S/N.

regards, Jack

[Digital Larry]

Thx Jack.

No, I don't have exactly the same output circuit as the dev board schematic for example (which uses 470 and 1.0 nF, but same difference). 

http://www.spinsemi.com/Products/appnotes/spn1001-dev/SPN1001-DEVB_sch.pdf

I run the DACL output to the input resistor of a Sallen-Key LPF at 12 kHz.  I don't block DC going in as I need it to be centered around Vcc/2 anyway.  I use AC coupling and a bleeder resistor on the filter output which goes to the output jack.

1k/2.2 nF is a one-pole LPF at 72 kHz?  http://sim.okawa-denshi.jp/en/CRtool.php

The test signal is my Squier strat black with plastic mirrored pickguard psssssh  you're so jealous  that when I engage the bridge humbucker, the FV-1 clipping LED flashes.  So I think my levels are in the right ball park.

[blackieNYC]

got it in a box?
Had to ask. 

[Digital Larry]

It is not in a box.  However the problem is shhhhh noise, not hummmmm.

[anotherjim]

Has the whole analog stuff got to run off 3v3?

I would suggest treble pre-emphasis before the LPF. Post de-emphasis can probably be included in a slightly steeper output filter.
This can't completely cure it. If noise is particularly strong it could turn into a woosh rather than a shoosh.

[Digital Larry]

I'm pretty sure I can make complementary shelving EQs for the input and output.

I can run the op-amps off 5 volts, or 9 volts unregulated, which I'd rather not.

Thx for your input.

[amz-fx]

Sounds like you are doing all the right things.

Don't tie the two outputs together (not that it sounds like you are doing this, but just in case).

I have seen lots of people complain that the FV-1 is noisy. I have built a handful of FV-1 projects and did not have that experience with any of them, but enough builders complain that it makes you wonder.

regards, Jack

[PRR]

Hiss of 12AX7 or TL072 is near 14uV/Hz. So the 21uV/Hz is high, but within the range of our classic gitar input devices.

I'd like to see the ACTUAL schematic. You could have a design flaw that you are not seeing or describing.

But I agree. The delay/reverb chip is much more likely to be hissy than basic opamp stages.

[Digital Larry]

Okay here is the schematic of the relevant sections.



On the left, analog input.  The op amp is set up with a shunt network (R18/C18) to boost the gain however in this case I have left it off, so what is left is a 47k from output to (-) with a 680 pF across it.  That's a bit unorthodox and I don't recall the theory behind it, but given the high input impedance of the op-amp it is close to being hooked up as a unity gain buffer.

On the bottom, you'll see the power supply section and at the lower right, the network creating the virtual ground reference point.  2 10k resistors and a 1 uF cap.  Is it enough?  Don't know.

Upper right, the 2P gain of two low pass filter which buffers/amplifies/filters the DAC out of the DSP chip.

Various 2-pin headers go off board to a 3PDT true bypass switch.

[Rob Strand]

Single-ended designs require careful layout and even then end-up with crap getting in.
Your have to micro manage grounding paths.
I would try connecting C18 to the ground near the input socket.
(Then, maybe even the ground side of R20 & C27 to that point also.)

It wouldn't hurt to put C27 = 47uF to 100uF, at least for now.

[amz-fx]

C25 isn't do very much since the impedance driving it from the op amp is probably very low. You need some series resistance immediately after C23 (before C25). I can't see the value of C25 but a new 1k resistor and 10n (C25) will give about 16kHz corner.  http://www.muzique.com/schem/filter.htm

TSV914 is rail-rail output, so that is good. As Paul said, 21uV/Hz is okay but not what we would call low noise.

Is C21 in the right place? Should it be on pin 3 of the op amp? I'm not familiar with the filter as drawn, so it could be okay and I'm just not recognizing it.

Also, we cannot see where ADCR goes. It is okay to leave it floating as long as the program does not try to read it. You could alternately tie it to ADCL and not read it.

regards, Jack

Edit: adding ADCR note

[Ice-9]

From the partial schematics I can only see so much, other important components around the input and output to the FV-1 pins are not very visable.

It is important to have the filtering around the output pins 27-28, if you only use one output then leave the other unconnected. The FV-1 is hissy if the outputs are not well filtered and of course it is important to drive the inputs for full resolution.

A quick and simple test is to remove the FV-1 chip and connect pin 1 +2 to the output pin 27 or 28. You will most likely find that the audio with the FV-1 removed and bypassed that the audio is pretty silent of noise. In which case look at your FV-1 output filtering.

I have tried many well known manufactures FV-1 based pedals at guitar stores and guess what ? many of them do have lots of Hiss.

[Digital Larry]

ADCR goes off to a switch and eventually could be used for tap tempo input.

The 0.1 uF to ground at ADCL is probably stupid and is a remnant of the reference RC 1 pole input filter.

The LPF at output is supposed to be a Sallen Key, but I obviously put the cap to ground in the wrong place, thanks.  I didn't yet attempt to measure frequency response.

DACR goes to a transistor and LED combo which can be used to visualize the LFO.

Yeah bypassing the chip altogether is the best idea!  For getting the residual noise that is.  It will put a crimp in the functionality for sure.    

Thanks for all the comments!

[amz-fx]

I'm glad we could help you get it sorted out.

When you make the changes, please report back and tell us what you think about the noise/hiss level.

Best regards, Jack

[PRR]

The values of R18 R18 are adding about 3uV (20KHz) each, which is high for guitar interfacing.

Why so high?? That opamp can drop 20mA, and 2V/20mA is 100 Ohms. Dunno if it does that clean, but it is rated for 600r loads.

I'd go closer to 1K than 47K here. Dunno if that makes C18 an awkward value... ah, 1.5K and 10uFd seems practical.

[Digital Larry]

Thx PRR!  I have not done analog design in many years and rather than succumb to analysis paralysis, I decided to shoot first and ask questions later.  Sorry for all the cliches.  Changing values is simple.