Try adding a small resistor (e.g 1kΩ) in series with the reference output before the filter capacitors

Sure, I'll give that a spin, with two questions:
1. is the intent of this just to create a small RC filter to filter out the hiss? (with 1k and my cap values that's about a 3Hz cutoff point!)
2. won't this defeat the benefit of the low-impedance drive of the Reference chip? (although I suppose I'm defeating it anyway with my 50k resistors on the +inputs of the opamp!)

 

Why does the AC audio circuit have a very accurate DC reference IC that produces noise?
The accurate DC reference is used in a voltmeter or thermometer circuit, not in an audio circuit.

The - input of an opamp is biased by the negative feedback resistor.
The + input of the opamp is biased by a bypassed resistors voltage divider.

Thanks audioguru!

I suppose I'd just assumed that a nice, purpose-built voltage reference IC was just a better, more stable, quieter choice than a resistive divider. Clearly having issues with it here though, haha.

 

Audioguru beat me to the answer.
Unless they are specifically rated for low noise, don’t use a reference for the mid-bias but rather use the resistor divider as shown.
For a very stiff “virtual ground” that can drive several opamps, try the TLE2426.
https://www.ti.com/product/TLE2426

Would a virtual ground chip like this be preferential to a resistive divider? One of the reasons I tried the ref chip was to get a nice low impedance midscale voltage; seemed to me like a resistive divider would be sort of squishy, and possibly carry any noise from the power supply (even though I am regulating that!)

 

The bias voltage feeds the very low current with no signal opamp + input of the mixer opamp. Then the bias voltage is easy to filter away any hiss.

 

Hi all, just following up on this for my own understanding.

I added the earlier suggested series R after the output of the reference chip and I also added 1uF caps at the noninverting inputs of the u3 output driver, for a ~3Hz single pole lowpass filter. This worked like a charm. I'm not sure which had the greater effect, since I did both at the same time, but the circuit is very quiet now. No audible hiss at all.

What I'm wondering is, why? The REF5045 (U50) is supposed to be a low noise reference IC that produces a very stable voltage for ADCs and other precision devices, and it's supposed to have 10mA of current drive available and low noise (3uV/V, so for my 4.5V output, that's 13.5uV of noise.) Could my problem have been literally just the resistor noise from R31/R32 at U3's inputs, and those RC filters killed it? I had opted for higher value 49.9k there, thinking that looking into an opamp input's high Z, the value didn't matter much.

I've also decreased some of the input resistors into the driver stage, because I wanted more volume with certain headphones, and I think I'm going to play with the gains on some of the active input stages as well, in particular, I'd like a greater usable range on the J9/J10 inputs.

I'm considering this build my prototype, and tuning it up to work great in my system. I'd also like to make it more manufacturable, as I can see it being a valuable tool for some guitarists who have setups like mine. In particular the connector and potentiometer assembly into the case was a giant pain. I'd need to smooth that process out if I want to make more than a few of these.

Thanks for the help all!