Combatting PT2399 noise: A Discrete TL431 Regulator?

[niektb]

Hi all! As some of you might remember (Super Buffer Topic?), I've been toying lately with hifi-inspired alternatives to create better-sounding guitar pedals. If only I had the time to try all those ideas haha. The post-student life is busy haha *sigh*
As many of you know, the PT2399 is a notoriously noisy chip which I've been looking to tame. I know from my HiFi experience that the TL431 is a popular choice for audio to improve the sound.
https://www.tnt-audio.com/clinica/regulators_noise3_e.html also shows that it as a lower noise than f.e. a LM317.

Does anybody have tried something like this? Did it work in reducing the noise that's coming from the PT2399?

P.S.:I'm looking to use this in tandem with the output filtering from the Dark Rift Delay (PedalPCB). This one should also reduce noise significantly.

[PRR]

Is the PT2399's noise coming from the power supply?

[Rob Strand]

Yes, there's plenty of sources:
- the power supply  ; the PSRR isn't that great.
- noise on Vref (pin 2)
- the grounds
- the method of A/D and D/A
- even the opamps

You can see some good info about the Vref pin here.  It used a 6k + 6k divider but you don't know what internal resistance path there are or things that cause spikes on Vref.

https://www.electrosmash.com/pt2399-analysis

You would have to build a totally bullet proof set-up possibly with separate opamps.

If the noise hasn't changed you know it's the method.  If you have some reduction then you would have to swap stuff in and out to find what part is actually causing the noise.   A fairly long process.

Another angle is to use two devices then mix the outputs back together.  If the noise is random is should give a 3dB signal to noise improvement.   AFAIK, there's no way to synchronize the clocks of two devices.  Two devices with near equal clocks might cause some weird pips, whining or beating.

[PRR]

Stock up on turd polish.

The PT strikes me as "the very least we can do" after three decades of BBDs.

There ought to be FAR better ways to do delay today.

[Rob Strand]

I haven't seen anyone trying to lower the cut-off of the two filters.   That might help stop glitches from the A/D D/A.    It wouldn't be hard to just double the values of the low-pass filters caps around the opamps (not the high-pass/decoupling stuff and not the 2x82n integration caps).

[Radical CJ]

This may be a solution to a different problem, but running multiple PT2399s in series so you can keep the delay times shorter on each individual PT2399 improves the sound quality. 

Stock up on turd polish.

The PT strikes me as "the very least we can do" after three decades of BBDs.

There ought to be FAR better ways to do delay today.

Is there anything even similar in remotely the same price bracket? I can get PT2399s for $2.50 but a MN3007 is $25 (Australian dollars and suppliers).

[deadastronaut]

i run 2 x pt2399's current mirrored for accumulated longer cleaner delays, no noise issues at all...and ive built a few.. (of course if you push the time pot bigger than 100k it will still go mental)...

single pt2399 delays are inherently noisey. which is why i went this route.

dreamtime delay.

https://www.diystompboxes.com/smfforum/index.php?topic=120165.0

[niektb]

This may be a solution to a different problem, but running multiple PT2399s in series so you can keep the delay times shorter on each individual PT2399 improves the sound quality. 

I've tried current-controlled delay times (on a single PT2399), but it felt more distorted and not so pleasant. I'm a little anxious to use them as a result  https://www.diystompboxes.com/smfforum/index.php?topic=125598.0 for the full story

Is there anything even similar in remotely the same price bracket? I can get PT2399s for $2.50 but a MN3007 is $25 (Australian dollars and suppliers).

Of course I can go DSP easily, I have the knowledge to do so but I find it too much effort and to expensive. I can get a PT2399 for $0.80 (Tayda) and it's extremely easy to use, compared to full-blown DSPs...

[amptramp]

I have built a regulator in the past of a remote sensing application where I had to have the lowest noise and used a pair of LM113 reference diodes as the voltage reference.  Looking back, I could have got lower noise with an LM399.  This reference was used to bias an op amp that drove a series transistor regulator with feedback resistors to set the voltage.  The output voltage was 15 volts and the noise was 33 nV/(SQRT Hz).  In this design, I had the reference driven by the output for minimum noise (since the load generated almost no noise), and to get the regulator to start, I used a resistor from the input side with a zener in series so when the output came up, the zener would shut off the current source from the input.

In your case, there is probably more noise on the output side so you don't need to use the trick I used.  It is possible to design a regulator for lower noise than a packaged device will give you.

[Rob Strand]

Given two PT2399 in series works OK you would think that using a single device working at half the bandwidth should also be OK.   So halving the cut-off frequency of the filters ie. doubling the caps should achieve a similar amount of conversion noise.  The A/D D/A converters track the signal slope.   If it can't track you will get evil artifacts.  Higher frequencies have a higher slope.  The lower the band-width the lower the slope and the less chance of artifacts.

Pondering a bit further it might even be worthwhile tweaking the 82n integration caps, but only after the filters are changed.

[niektb]

I have 100nF as integration cap between pin 9 and 10, 330nF between 10 and 11 already. It darkens a bit, but that's something I like

The Valve Wizard advises a Zener Follower Regulator to reduce noise:
http://www.valvewizard.co.uk/PT2399_Data_Notes.pdf

[Rob Strand]

I have 100nF as integration cap between pin 9 and 10, 330nF between 10 and 11 already. It darkens a bit, but that's something I like

I guess the question is does it improve the noise or make it worse?

The Valve Wizard advises a Zener Follower Regulator to reduce noise:
http://www.valvewizard.co.uk/PT2399_Data_Notes.pdf

What Merlin mentions at the end makes sense.  The fact he makes a point of writing it up means it's something that should be moved up the list of things to try.

Something to keep in mind is there's often many sources of noise.   If your circuit has noise problems one of those is often the main cause.   When you remove the dominant source of noise the other sources of noise are then noticeable.    In these cases things you tried earlier might actually work.

[anotherjim]

Seems to me that auto variable input and output filters (low-pass VCF) that follow delay time could be an answer. That would mean that short delays don't have to be darker just to help long settings. And once those are there, signal amplitude could also control the cutoff of the output filter so noise isn't so noticeable. The onboard opamps can be eliminated to an extent. The input amp (pins 15,16) at unity with, say, 1k resistors fed from the input VCF. The "spare" amp on pins 13,14 not used at all and those pins are linked to keep it quiet.
The frequency of the chips VCO output can be used to derive the VCF control, but it could be as simple as having a dual gang delay time pot. Indeed, instead of VCF, simple LDR vactrol control of the filtering.

The extra complication is fun to think about, but it might all be lipstick on a pig. I happen to think this is the case with BBD delays with pre-emphasis and companding and the good thing about them is what all those extras are trying to hide!

[ElectricDruid]

Stock up on turd polish.

The PT strikes me as "the very least we can do" after three decades of BBDs.

There ought to be FAR better ways to do delay today.

+1 agree. It *is* pretty remarkable that there isn't a proper high-quality single-chip delay available in this day and age. It's not like it's impossible. Something like the PT2399 but with better convertors (or a better rate) and more memory is hardly a complicated chip to build.

[PRR]

My $29 cellphone will record or play hours of speech with far better fidelity than a BBD. Plus I hear I can do BookFace on it, and maybe call people. It does not record and play at the same time in the same app, but that's just code.

I bet the BluBerryPi chips which run a unix and a browser can (perhaps with hardware ADC/DAC) do as much in a more open device.

[niektb]

I have 100nF as integration cap between pin 9 and 10, 330nF between 10 and 11 already. It darkens a bit, but that's something I like

I guess the question is does it improve the noise or make it worse?

To be fair, I didn't A/B it. My reference schematic already had them

Stock up on turd polish.

The PT strikes me as "the very least we can do" after three decades of BBDs.

There ought to be FAR better ways to do delay today.

+1 agree. It *is* pretty remarkable that there isn't a proper high-quality single-chip delay available in this day and age. It's not like it's impossible. Something like the PT2399 but with better convertors (or a better rate) and more memory is hardly a complicated chip to build.

I think the major contribution to this problem is that DSPs are really accessible nowadays. For just a few bucks you can get a powerful programmable device. However, smaller packages sizes are also a trend. Meaning that there áre cheap DSPs, but more than often they're not in a DIY-friendly package (such as QFNs or BGAs)

[Rob Strand]

I guess the question is does it improve the noise or make it worse?

To be fair, I didn't A/B it. My reference schematic already had them

I was thinking if they were too large the rate the signal can ramp up and down is limited by the large caps.  If the filters were still high bandwidth it could promote conversion noise since the slow ramp means the A/D D/A converters are always struggling to track the signal slope.    If on the other hand the filters are set to a low cut-off then the A/D D/A are only trying to track the lower frequencies and will have a lot less conversion noise.  That's the theory/thinking anyway.