martin clark
pinko bodger
No: you haven't read why it works - for the Naim's purposes its a very acceptable tradeoff. For yours, it might not be.
LM117/317 datasheet tracking pre-regulator
- Thread starter 13mh13
- Start date
13mh13
Pi-s Artist
Thx for your reply. I read more carefully and think I understand why those values may be application-specific.
Given than, can you suggest generic output caps (and/or cap combos) that work for apps such as powering digital ICs, opamps, etc. And, preferably, those components located remotely (i.e., TPR in a PSU box and what it's powering in, say, a CDP).
Speaking of which, what are some strategies when decoupling/bypassing at the output of the reg. board vs. decoupling/bypassing at the power pins if the IC? E.g., I noticed in the CD2 PCB images posted midcm1 that Naim does not seem to use large-value (electro) caps on power pin inputs of 1541 and 7220; but, most Philips CDPs do employ both an electro and a small-value film/ceramic cap on power pins.
martin clark
pinko bodger
To me, bypass caps serve the load, not the reg. So keep them close to the load 
I'm not going to pretend to recommend generic values - often generics are in the ICs datasheet anyway, but its only ever a 'suggestion' because there's always room for experimentation and refinement; and so much depends on the gestalt - the detailed physical reality of your layout.
I'm not going to pretend to recommend generic values - often generics are in the ICs datasheet anyway, but its only ever a 'suggestion' because there's always room for experimentation and refinement; and so much depends on the gestalt - the detailed physical reality of your layout.
13mh13
Pi-s Artist
Actually, I have blindly done this, but it was never an "informed" or "intelligent" decsion until your confirmation aboveTo me, bypass caps serve the load, not the reg. So keep them close to the load
.IAC, and OTOH, I noticed the Avondale TPR, as well as the I/V PSU -- both noted earlier in this thread -- have bypass caps on their outputs. I'm assume that if the reg. is very close to the load, as is the case with Naim, one can ditch a lot of the bypass caps??
martin clark
pinko bodger
I think you can, but have to check this for every use; and sometimes, an extra big, cheap bypass cap can be just the thing to make a 3-pin reg workwell with low-esr film or other HF bypass at the load, by acting as a snubber. This is how the CD2 is mostly laid-out BTW - small film bypasses at the load, but 10uF + tants are located back at the regs.
13mh13
Pi-s Artist
I see the tants, but not 10uF (unless the 10uF is a tant).
BTW: what's that opamp and the heavy-duty heatsink deal on the far right-hand side, next to +5 label?:
martin clark
pinko bodger
Pre-reg feeding the 5v regs, indicated by dotted line. Fitted with heatsink, it disspates about 2w.
The chip is a TC4086 (quad bilateral switch) used with the cd transport chip (I think). The heatsink is mounted on LM317 reg used to drop the 15v supply down to approx 9v before feeding the two top right hand regs for the 7220 and cd transport chip. these two chip draw a lot of current hence the heatsink. This is why a CD2 gets warm.
13mh13
Pi-s Artist
Here's an image of the TPR + gyrator I came up with -- this is a negative (337) version. Not much room for heatsinks, so this one's for low-current use only...
...Which brings me to an application. Earlier I noted I wanted to power an IV ...
...which currently uses this PS...
...with a +/- TPRs. Well, now that I've got some (!!), my course of action is to simply use the rectifier and smooting part of the orig PS (see above). That has always been located in a separate PS box -- i.e., the IV board rec'd regulated power "remotely". But given the recent discussion in the Naim CD2 thread, I'm going to relocate the TPR into the main CDP box: connected, as close as possible to the IV board.
Martin (or anyone)...do you have any specific suggestions here? What I'm most curious about is regarding output bypass/decoupling caps on the TPR. Recall, the orig IV PS had massive capacitance (220uF, see schema above) on its output -- dunno why it was so large? Also, the IV has huge input capacitance (again, see schema above). But now, since the reg and IV will be kissing each other, I'm not sure what good "ballpark" values to start with.
...Which brings me to an application. Earlier I noted I wanted to power an IV ...
...which currently uses this PS...
...with a +/- TPRs. Well, now that I've got some (!!), my course of action is to simply use the rectifier and smooting part of the orig PS (see above). That has always been located in a separate PS box -- i.e., the IV board rec'd regulated power "remotely". But given the recent discussion in the Naim CD2 thread, I'm going to relocate the TPR into the main CDP box: connected, as close as possible to the IV board.
Martin (or anyone)...do you have any specific suggestions here? What I'm most curious about is regarding output bypass/decoupling caps on the TPR. Recall, the orig IV PS had massive capacitance (220uF, see schema above) on its output -- dunno why it was so large? Also, the IV has huge input capacitance (again, see schema above). But now, since the reg and IV will be kissing each other, I'm not sure what good "ballpark" values to start with.
martin clark
pinko bodger
If it works well, leave it C15/C16 will be having marginal effect given they are followed by the RC on the IV/stage supply pins. If you are tight for space you could leave them out, but if not they certainly should do no harm.
I/V stage signal current is mostly HF (44.1Khz x whatever oversampling ratio you are using) so its C1, C4,C7a+b which will probably dominate how the PSU 'sounds'.
C15/C16 will be having marginal effect given they are followed by the RC on the IV/stage supply pins. If you are tight for space you could leave them out, but if not they certainly should do no harm. I/V stage signal current is mostly HF (44.1Khz x whatever oversampling ratio you are using) so its C1, C4,C7a+b which will probably dominate how the PSU 'sounds'.
13mh13
Pi-s Artist
Got it!
One (perhaps) final question. Those two 1K's around the first 317 of the TPR. They produce a voltage drop of 2.5V. I think your write-up on acoustica said this was okay. When would one need to tweak these values -- i.e., what "symptoms" should one watch out for that indicate that the voltage drop on the first reg needs tweaking?
martin clark
pinko bodger
All they do is alter the volatge drop across the second reg. Noramlly you want to keep this fairly small to minimise the total dropout voltage of the whole reg. The second 317 needs at least 2.0-2.5v across it (see datasheet), but because the first reg gives good control of this, you dont really need to set it at more than that. Setting the two resistors equal makes this 2.5v, which means you can get away with an ouput voltage maybe 6-7v below the raw supply (after allowing a little bit extra for ripple and sag under load)
13mh13
Pi-s Artist
Okay, I tested a + and a - TPR (with gyrator) in the IV noted above...
Before gauging sonics, I measured (with true RMS DMM) and adjusted ("optimized") voltages on the TPRs as loaded by the IV circuit. I let things warm up, and made smaller tweaks (via a voltage setting R on the 2nd 3x7, which is a Bourns pot).
Sonics:
After about 1hr warm-up playing a pink noise burn-in CD, I slipped on headphones for a listen. Immediately, I noticed a loss of image focus. The SQ was a tad richer, tho'. Other sonic metrics -- bass slam, PRaT, etc. -- were unaffected.
Note that my TPR is not a "pure" TPR as it also has:
A gyrator before the first reg.
An LED + 3.7K resitor on the output of TPR as an "idiot" light (on-indicator).
Also, as prev. noted, I moved the regs. as close as possible to the IV board, which should've been an added plus.
I can't live with the loss of image focus, but don't know what that sonic metric my be attribtable to: do new components have to "break in" (I doubt this is the case)?? As a result, I'll probably re-install the orig. single-reg. topology.
Comments/suggestions are welcome!
Before gauging sonics, I measured (with true RMS DMM) and adjusted ("optimized") voltages on the TPRs as loaded by the IV circuit. I let things warm up, and made smaller tweaks (via a voltage setting R on the 2nd 3x7, which is a Bourns pot).
Sonics:
After about 1hr warm-up playing a pink noise burn-in CD, I slipped on headphones for a listen. Immediately, I noticed a loss of image focus. The SQ was a tad richer, tho'. Other sonic metrics -- bass slam, PRaT, etc. -- were unaffected.
Note that my TPR is not a "pure" TPR as it also has:
A gyrator before the first reg.
An LED + 3.7K resitor on the output of TPR as an "idiot" light (on-indicator).
Also, as prev. noted, I moved the regs. as close as possible to the IV board, which should've been an added plus.
I can't live with the loss of image focus, but don't know what that sonic metric my be attribtable to: do new components have to "break in" (I doubt this is the case)?? As a result, I'll probably re-install the orig. single-reg. topology.
Comments/suggestions are welcome!
Simple enough to gauge their effect by a) bypassing and b) disconnecting.
The gyrator could be choking your dynamics also.
The loss of focus might be due in part to noise from the reference diode in the first lm3x7. To my mind the C3 cap actually increases its injection into the output. You could try removing C3. Doing this will I guess decrease supply line rejection, so it's a compromise.
Another idea is to connect an electrolytic cap (22uF-100uF) between the output of the first reg and 0V. The ALWSR, which uses this tracking preregulator approach does this. I found in the ALWSR that the result was quite sensitive to the size and type of cap. I would not use a film or Tant; others may have actual experience. Silmic II and Stargets are good as would most aluminium elcaps.
Don't give up without a fight.
13mh13
Pi-s Artist
Still working on this issue ...
Do you mean in addition to or instead of the 1uF tant. already there (C2):
martin clark
pinko bodger
Just get the soldering iron out and try, try, try
Sorry, I was ignoring the existing C2. The idea is to increase significantly this cap value. Start by adding. The existing Tant will still be effective at HF when the Aluminium elcap gives out. However capacitors in parallel can oscillate at HF. Listen and if in doubt remove the Tant.
I use a 47uF Silmic II (Low ESR at LF) in two of my ALWSRs, but Stargets are easier to get (RS) and pretty good. Panasonic FC and FM are good and available. I guess the effect will depend on the nature of current draw of the following stage. So as Martin said try things out.
I use a 47uF Silmic II (Low ESR at LF) in two of my ALWSRs, but Stargets are easier to get (RS) and pretty good. Panasonic FC and FM are good and available. I guess the effect will depend on the nature of current draw of the following stage. So as Martin said try things out.
13mh13
Pi-s Artist
I spent several hours in the last few days doing most of the above, and then some. None of it helped. I reinstalled the original single-3x7 ckt, and like "magic", the focused sound snapped back in.
This experience has me doubting the efficacy of "tricked-out" regulators, like TPRs. Earlier in this thread, I made the comment, “I've built and used TPRs before, but never paid attn to measurements that much -- relying on mostly SQ as a 'metric'. And TPRs always did just that -- improved SQ.” This statement, I just realized, is somewhat misleading. Basically, I “upgraded” from Philips' stock single-voltage regulation (LM7805, 7905, etc.) to TPRs. I never experimented with tried-and-true single 3x7-based regulation -- the type Naim and other “high-enders” use. Maybe the “high-enders” don’t employ “fancy” regulation judiciously? In fact, the lesson of this thread may be just that.
Not a total loss...I was able to improve on the quality of orig single-3x7 ckt by adding snubbers, lowering the output cap values and RCRCing the pre-reg smoothing caps.
In many ways, I’ve been fighting since the beginning. I don’t like fighting -- too much blood loss, and one’s dead!
John Jones
New Member
Hi there,
I have been reading this thread with interest because I have to decide what regulated supplies I am going to use in a headphone amp I am building. Ideally I want to separate all supplies, so this will require 8 supplies in all: 4 at +15v and 4 at -15v.
I have read about the Jung/ALW type super regs but these are quite expensive to build and 8 would be very pricey.
I therefore thought along the lines of 8 LM317/337 regs, or indeed any other integrated regulator out there, but tweaked for best performance.
I was disappointed to see the negative results from the tracking preregulator tests in this thread. And I have read all the tips in this article: http://www.acoustica.org.uk/t/3pin_reg_notes1.html
Therefore my question: From experience, what is the best way to tweak the LM317/337 regulator, or is there a better regulator out there and what is the best implementation of said regulator for audio supplies?
I have been reading this thread with interest because I have to decide what regulated supplies I am going to use in a headphone amp I am building. Ideally I want to separate all supplies, so this will require 8 supplies in all: 4 at +15v and 4 at -15v.
I have read about the Jung/ALW type super regs but these are quite expensive to build and 8 would be very pricey.
I therefore thought along the lines of 8 LM317/337 regs, or indeed any other integrated regulator out there, but tweaked for best performance.
I was disappointed to see the negative results from the tracking preregulator tests in this thread. And I have read all the tips in this article: http://www.acoustica.org.uk/t/3pin_reg_notes1.html
Therefore my question: From experience, what is the best way to tweak the LM317/337 regulator, or is there a better regulator out there and what is the best implementation of said regulator for audio supplies?
martin clark
pinko bodger
Yes, there are (much) better regulators, but those acoustica notes and this thread is about getting the best out of cheap jellybean parts
There are a number of ways you could get really good performance cheaply to start with for your application.
You could just use a 5th reg for each rail to provide a raw, say+/-18v rails the subsequent regs draw from. It's guaranteed to work, and the low impedance feed will suit the separate regs well.
You could just use a touch of R/C decoupling ahead of the 4regs per rail
You could do both or invent your own order of play.
Sometimes keeping things simple is a very elegant solution and more than 'good enough'.
The best result will come from working-out what the target circuit is sensitive to; for example, if it involves a lot of current sources then supply impedance probably doesn't matter so much but noise control might (so a simple RC+emitter follower or the more elaborate 'Teddyreg' might be a good choice). That kind of thing.
I remain convinced of the TPR approach, esp for good results at moderate to higher currents (100s of mA) but- as 13mh13 has found - it is a surprisingly 'subtle' thing: it has two nested gain loops and therefore demands attention in layout and parts choice and a measure of experimentation to get right for each application.
There are a number of ways you could get really good performance cheaply to start with for your application.
You could just use a 5th reg for each rail to provide a raw, say+/-18v rails the subsequent regs draw from. It's guaranteed to work, and the low impedance feed will suit the separate regs well.
You could just use a touch of R/C decoupling ahead of the 4regs per rail
You could do both or invent your own order of play.
Sometimes keeping things simple is a very elegant solution and more than 'good enough'.
The best result will come from working-out what the target circuit is sensitive to; for example, if it involves a lot of current sources then supply impedance probably doesn't matter so much but noise control might (so a simple RC+emitter follower or the more elaborate 'Teddyreg' might be a good choice). That kind of thing.
I remain convinced of the TPR approach, esp for good results at moderate to higher currents (100s of mA) but- as 13mh13 has found - it is a surprisingly 'subtle' thing: it has two nested gain loops and therefore demands attention in layout and parts choice and a measure of experimentation to get right for each application.