Shunt regulator for a phono pre

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Having lived for a while with a gainclone i decided (maybe naively) that a full solid state system will be cool. Not that i don't like the sound of my valve phono and DHT (PX25) power amp, it's just that the success of the GC gave me hope that a simple but good sounding system is possible with premium quality opamps. So far the results have been encouraging, rather than exilarating, but of course i've always known that a good phono stage is a lot more difficult than a a good power amp.
I started the prototype with the intention to build several different power supply regulators and compare the sonics. Initially it was built with the 'standard' 317/337 combo and the sound was admittedly not too bad, but the usual 'greyness' and 'thickness' of the regulators was unmistakeable. My first 'improvement' was a simple tl431 based shunt fed from a ballast resistor. The midrange cleaned up remarkably but in general the sound became less balanced, with very weak bass. Next the resistor was replaced with a 317 hooked as a current source at 70mA. Each rail of the phono draws around 50mA, so i thought 20mA for the shunt transistor should be enough. The bass came back, but interestingly a lot of the 'greyness' i've learnt to associate with the 317 came back as well. So, maybe a 317 is not a perfect CS after all :).
Finally i built the simple CS shown on the diagram: it's based upon a forward diode drop. As anticipated the sound was now really better. If anything, the bass was maybe a tad too soft compared to the 317/337 regulators but everything else more than made up for it. As you can see i use very substantial transistors for both the current source and the shunt element. Initially tried BD139/BD140 but they really didn't sound as good.
So, happy as i am with the result, there is certainly room for improvement. I'll appreciate all comments and suggestions and some time next week will try to compare this to a Jung type regulator.


cheers
peter
 

Attachments

  • shuntps.jpg
    shuntps.jpg
    81.6 KB · Views: 3,118
You are some way ahead of me in your phono amp regulator study.

I had thought of trying the Jung super-reg, probably using ALW's circuit+boards, and then following that with a shunt regulator to see what difference it makes. Interestingly, ALW uses the 317/337 as a pre-regulator for the super-reg, so I am guessing that the "greyness" you hear is cleaned up by the super-reg.

You might clean up your own current source by using a LED or buried-channel zener for the voltage reference at the input.

I downloaded an article on shunt regs a few weeks ago - I don't seem to hav the URL but a google on wenzel and finesse should find the source.

I will be interested to hear how your invetsiagtion progresses.

--John
 
John

I started this journey 'backwards' with the simple shunt-reg as i wanted something small, cheap and not prone to oscillations and problems. Mostly because i intend having a set of the regs next to each op-amp/buffer. The Jung solution is not so well suited to the idea of having 2 or 3 per channel. Also i have serious doubts about the preregulators. Having never built a Jung/ALW reg i still have some experience with what Audio Research used to use when the SP10 was the best preamp ever. It is similar to Jung in using an opamp with a series-pass transistor, but it's output is floating and not referenced to a particular voltage, just used to buffer each stage and fed from a preregulator. I never got to like what it did to the sound.
Interestingly Elso Kwak seems to also dislike preregulators, i'll certainy try them, but against clear prejudice.

cheers
peter
 
Hi Rob

After reading your post i was annoyed for not having thought of this myself :) A .47 seems a bit low for a 5k1 resistor so i tried a 3.3uF Wima MKP (admittedly not my favourite). To my surpise the sound did not get any better; to the contrary - it lost dynamics. Weird?

cheers
peter
 
An interesting and thoughtful post Andy. I used tl431 primarily for the low noise and simplicity. If low noise is not a priority i'll gladly get rid of it although most of the discrete circuits i've seen typically consist of differential amp with/without cs/c mirrors, current amp,shunt element - easily 7-8 transistors per rail, not counting the CCS. Not ideal if you want 4 of these.

cheers
peter
 
Re: A few thoughts

ALW said:
As has been pointed out to me recently, the SPICE model is DC only and hence useless for AC simulation :bawling:

If you're using LTSpice, a good TL431 model is available in the files section of the Yahoo LTSpice user's group at http://groups.yahoo.com/group/LTspice. This model includes AC effects. It also solved the convergence problems I was having with the first TL431 model I downloaded from the net, which I believe was provided by TI. I think you need a Yahoo email account to access these groups though. But that's no problem. They are easy to set up.
 
HI Andy,

Great post, very informative.

I use a TL431 to control the source of a grounded gate mosfet shunt regulator. This enables 340V B+ regulation to within about 0.1V. I do not believe noise is the issue either; more the output characteristics of the regulator, particularly where tubes are used. The so-called 'tube' sound is in fact largely the interaction with the power supply (H2 generation); you don't want a rigidly regulated power supply, rather you want a fixed, ohmic source. Like you, I don't use a cap to bypass the upper divider. Causes overcorrection; just not necessary.

Cheers,

Hugh
 
TS431IZ - 1.24-6.0 volt zener shunt diode

ALW said:


One thing to be aware of is the noise of a 431 is almost always quoted with Vout = Vref.

The noise increases with Vout, since the noise gain of the circuit rises as Vout increases.

The bypass mentioned earlier fixes this, buit has other side effects.

Noise is rarely the major sonic factor anyway.

As an example, take the Jung circuit and remove the gain bypass cap, increasing the noise gain of the circuit. This, in my 24V supplies increases noise, output impedance and worsens line rejection by 11dB!

It also sounds a LOT better :)

Do you have a link to a data sheet for the TS device - all I found was an SGS-Thomson part.

Andy.
I did a search at
ELFA.se English Catalogue (prices in Euro)
Our supplier in northern europe,
as I know they have TS431


This is what I found

TS431CTTR TO92 73-000-08 Taiwan Semiconductor
TS431IZ TO92 73-053-45 TS 431 ST Microelectronics


you can find those 2 in my searchresult:

search result TS431

If you click the little blue "i"
at each product, you get the PDF datasheet, "info"

What is good about TS431, compared to TL431
is that it has ref-voltage 1.24V, can produce lower voltages
Can also be used at VERY LOW currents, 60uA -30mA
(this allows for higher value resistors, with easier filtering with good quality caps)

While TL431 works at 2.5-36V, 10 volts recommended
and needs at least some 500uA (0.5 mA) to assure proper working.
>2 mA is recommended

this is how I recall it from my memory

True also that you have to multiply noisefigure
noise at 10 volt will be 8x1.24= 8 times higher than at 1.24 V
BEFORE eventual filtering

halojoy
 
ALW said:
As an example, take the Jung circuit and remove the gain bypass cap, increasing the noise gain of the circuit. This, in my 24V supplies increases noise, output impedance and worsens line rejection by 11dB!

It also sounds a LOT better :)


Hmmm, interesting! Makes me wonder if the improvement in sound was due to having improved the stability of the regulator due to the reduction in loop gain.

Also, thanks for the reference to the regulator articles.

I've been playing around with the simulation of a Jung-like regulator putting out +/- 90V for a power amp I'm designing. It's was inspired by the circuit at http://www.glass-ware.com/tubecircuits/High_Voltage_Regulator.html Simulations showed total instability with an assumed-perfect 100 uF load cap (using the Analog Devices model for the AD797). Then I looked up in the Cornell-Dubilier electrolytic capacitor app note and they said their lead inductance was about 30 nH. This gives a series resonance at about 92 kHz for the cap. Combining this with a guesstimate of the ESR at this frequency, I plugged the result back into SPICE and found that the regulator was now marginally stable. I'm using pass transistors that are much less broad band than the ones specified in the Jung regs, because of the high-voltage requirement (and the requirement of about 400 mA output current). The two feedback resistors are equal, giving a DC gain of 2. I found that the stability improved greatly by putting a 22 pF cap from the op-amp output back to its inverting input, introducing some phase lead into the loop gain just below the unity-gain crossover frequency. Once I did this, the transient simulation using a pulse of load current really cleaned up.

One problem I'm having with the simulation is the lack of data on the series resonant frequency and series resistance of capacitors. The RF capacitor makers supply this important info, but I wish the conventional capacitor manufacturers would do it as well. I'd like to know, for example, if putting a 0.1 uF polypropylene cap on my power amp supply will send the regulator into outer space or not. I suppose that to model this accurately, I'd have to not only know the capacitor data, but also treat the trace between the regulator and this cap as a microstrip line. That is, calculate its characteristic impedance using RF design software and model it as a transmission line with that impedance (and the correct length of course) in SPICE.

All this for a regulator? Sheesh ;) . But hey, what fun would it be otherwise?
 
Having looked at the TS431 data sheet, noise is the same as for a TL431.

I was referring to the other TS431, the one by taiwan semiconductor which could be a direct replacement for the TI one.
TS431 by Taiwan Semiconductor

TL431 byTI

The noise graphs seems (Vout does not show on the TI graph) to be measured under similar conditions and if correct the TS431 should have about half the noise compared to the TL431.
 
The noise graphs seems (Vout does not show on the TI graph) to be measured under similar conditions and if correct the TS431 should have about half the noise compared to the TL431.

You're probably right, take a look at the Onsemi ones, they are identical to the Taiwan Semi units in noise terms.

URL=http://www.onsemi.com/pub/Collateral/TL431-D.PDF]Onsemi TL431[/URL]

They're very generous with samples ;)

Andy.
 
Folks, forgive me if I have overlooked something on this thread. However, this is my take on 317,337 IC regulators: BE CAREFUL! Don't put a really good cap directly at the output of one. It will cause noise ringing. There was a good article on this in 'EDN' or 'ED' some years ago. What happens is this: The regulator starts increasing its output impedance, because it has a finite bandwidth and must roll off its gain early. This RISE in output impedance looks just like a synthetic inductor at the output. Put a really good cap directly at the output, and it RINGS! This is because there is no series damping. Add a 2 ohm resistor in series with the really good cap, and you are probably OK. However, why then would we put a really good cap at the output, if we have to add a 2 ohm resistor to spoil the Q? This can make things really interesting, and confusing, if you don't have any idea of what is going on.
 
diyAudio Senior Member
Joined 2002
Hi John,

This can make things really interesting, and confusing, if you don't have any idea of what is going on.

This is one reason why Fred Dieckman, Jocko and some others in the know always say:

Horses for courses.;)

I could add to it by saying that using filmcap for PSU filtering is also not such a good idea although for other reasons.

Cheers,;)
 
BE CAREFUL! Don't put a really good cap directly at the output of one. It will cause noise ringing. There was a good article on this in 'EDN' or 'ED' some years ago. What happens is this: The regulator starts increasing its output impedance, because it has a finite bandwidth and must roll off its gain early. This RISE in output impedance looks just like a synthetic inductor at the output. Put a really good cap directly at the output, and it RINGS! This is because there is no series damping. Add a 2 ohm resistor in series with the really good cap, and you are probably OK. However, why then would we put a really good cap at the output, if we have to add a 2 ohm resistor to spoil the Q? This can make things really interesting, and confusing, if you don't have any idea of what is going on.

Yep, pretty much what I was getting at (without beign so specific) in post #6.

The comment John makes though applies to any feedback regulator, with the possible exception of some of the newer fast transient performance devices, which are optimised for low ESR caps.

I have no idea how these devices sound though.

As I mentioned almost all of the audio grade caps (elna, nich) have a very nice impedance curve for this application, being resisitive above the min Z point of their curve, this then masks the effect of ESL, until the ESL has greater impedance than the ESR element.

Adding a series R is what I currently do, but I use a lower value of 0.5R at present (with a PPS film cap, which sounds, in my specific application, way better that a 'lytic).

I am about to measure / listen to various R's here, I must admit gut feel tells me 2 ohms is a lot - the worst 'lytics only approach 500mOhm ESR, but Mr Curl is an experienced guy, so I'd never ignore free guidance from such sources ;)

This is a classic example of where SPICE modelling can trip you up - if you don't model parasitics and non-ideal components the results will not bear any resemblance whatsoever to the real world results.

BTW, I have that app note John mentions somewhere, it was an EDN article by a National Semi engineer - I'll see if I can dig it out and stick it up for download.

The basics are as John says, the o/p of the reg appears inductive by an amount that varies with load, amongst other things.

A low ESR cap produces a high-Q noise peak at the reg o/p, a low ESR one produces a smaller amplitude peak, with greater occupied bandwidth as the resonant frequency / Q would predict.

I found that I could derive output L of the reg (1 / (2.pi.Sqrt(L.C)), then change the C and predict the noise peak frequency with high accuracy, for a fixed load on an LM317 or similar.

Andy.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.