Randy Slone's "Fig 11.4" (Self's "Blameless"?): PCB layout

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I've been trying to see if I could make a PCB layout for Randy Slone's "Fig 11.4" from his book "High Power Audio Amp Construction Manual." I'm told that his Fig 11.4 is an almost exact replica of Doug Self's "Blameless", with the addition of one or two small improvements.

I took this "Fig 11.4", and attempted to fit everything into one Eagle-Lite PCB (3.2" x 4"), and the result is shown below. I also added a few small "featurettes", based on my readings and on discussions in the "Fig 11.6" thread here. These include a diode and RC filter on each supply rail to shield the VAS and input stages from current sourcing and sinking in the OPS, and an input signal conditioning circuit. Thought I'd share all this and see what you guys think:

An externally hosted image should be here but it was not working when we last tested it.


An externally hosted image should be here but it was not working when we last tested it.


I have almost zero experience in designing power amps or power amp PCBs, hence all help I get is badly needed.
 
Ex-Moderator
Joined 2002
Sorry, but there are lots of things wrong with this board, but two obvious ones for starters, Q7, the bias trannie, should be on the same heatsink as the output devices for stable operation, and the feedback path is far too confused.

I appreciate how difficult it is to get going with this, I've been there. But you need to block out your layout better, try and roughly follow the layout of the schematic, rather that just putting parts in where they fit on the board. Follow the logical signal path, and arrange components around it.
 
pinkmouse said:
Sorry, but there are lots of things wrong with this board, but two obvious ones for starters, Q7, the bias trannie, should be on the same heatsink as the output devices for stable operation, and the feedback path is far too confused.
Can you please elaborate? I thought thermal coupling of the Vbias transistor with one of the pre-driver transistors is good enough in this case, because it's a CF OPS. I thought coupling with the OPS devices themselves is a strict necessity only in the case of BJT EF OPS.

And the feedback path seems to be a straight, fat track coming from the top of the board straight down to C6, R10, R11, to the base of Q2. Can it be any straighter than this in a congested PCB?

When I appear to contradict you, please don't misunderstand my intentions as an attempt to show how smart I am... I'm just trying to learn as much as I can from you.

I appreciate how difficult it is to get going with this, I've been there. But you need to block out your layout better, try and roughly follow the layout of the schematic, rather that just putting parts in where they fit on the board. Follow the logical signal path, and arrange components around it.
I thought I'd done that, in fact. The input stage is lower down in the board, the VAS comes after that, roughly, and the OPS is at the back. And I've tried to place devices "logically close" to their neighbours, so that track lengths are on the whole as short as I could get them.

I think I'll need more details. :(
 
For CFP outputs (Sziklai) the VBE multiplier should track the driver's thermally, not the outputs.

Beware when using the CFP output. These must be well laid out or they will oscillate. If this is your first amp that you're designing youreself, stick to using an EF output stage. Personally I prefer the triple Darlington but that is just me :)

Why use a big 5W resistor for the output resistor/inductor ? A good 2W device is fine here, and you can wind the inductor around it.

For a low power amp (you'll only get about 100W with 3821/1302 outputs), you could use 3W collector resistors. Use vertical mounting to make it more compact.

Another change, put a 1K resistor in the base of Q8. Otherwise, if the VAS saturates due to clipping (such as the protection circuits activating) it will pull all the current from the voltage reference, and screw up the current source for the LTP. The result is usually rail sticking or oscillation. Better still... use a seperate current source.
 
jaycee said:
For CFP outputs (Sziklai) the VBE multiplier should track the driver's thermally, not the outputs.
Thanks. This is what I'd read in Slone's instructions too.

BTW, my reading is limited to a little bit of Self and two complete books of Slone. I've also read online Web pages on the JLH and of course, some of these threads here.

Beware when using the CFP output. These must be well laid out or they will oscillate.
Slone says the CF oscillation tendency can be largely taken care of by using the appropriate driver devices, and since this is his design, I'm hoping he's made it stable enough for it to be forgiving of a novice PCB design. :) Anyway, I'm hoping to build it and see where it leads. I'm treating this as a learning experiment, not a world-beating amp.

Why use a big 5W resistor for the output resistor/inductor ? A good 2W device is fine here, and you can wind the inductor around it.
Actually that 5W resistor's part is supposed to be the size of my final inductor. Basically, I'm hoping I'll put a 2W resistor inside it, and wind the copper around it, and it'll finally come out as thick as the 5W resistor's layout.

For a low power amp (you'll only get about 100W with 3821/1302 outputs), you could use 3W collector resistors. Use vertical mounting to make it more compact.
If you say 3W is adequate, I'll use 5W, because I'm not very confident about the heat dissipation ratings of the resistors I get here (I'm buying from the local Bombay market). If I got a good international brand of resistors, I'd risk going by tighter ratings. I've been told by senior constructors that these resistors sometimes give up (i.e. get destroyed) due to heating.

Another change, put a 1K resistor in the base of Q8. Otherwise, if the VAS saturates due to clipping (such as the protection circuits activating) it will pull all the current from the voltage reference, and screw up the current source for the LTP. The result is usually rail sticking or oscillation. Better still... use a seperate current source.
Aha... the base of Q8.

If you don't make a face or fall asleep, I want to tell you about my story of that resistor. :D

Basically, I started working on the Fig 11.4 (BJT CF OPS) and Fig 11.6 (L-MOSFET hybrid CF OPS) amps in parallel. Since Slone says they are identical in all respects other than OPS, I entered the schematic from just the MOSFET circuit, and made two copies. I altered the OPS of one of them, and that became BJT OPS "Blameless". I'm still editing both in parallel, keeping them in sync.

Much later, while doing a detailed comparison of my BJT schematic with the one in the book, I realised that Slone had made changes in various little bits of the VAS and input stage after all. Some of those changes I could figure out, hence I made them consistent across both. For instance, in the resistors behind the Vbias transistor, he had just one resistor in the upper leg and one pot in lower leg for the MOSFET design, but he had a second resistor in series with the pot for the BJT design. I put that second resistor in for both circuits... it won't do any harm and will increase both sensitivity and safety. (AndrewT helped me a lot in that other thread on the MOSFET design). In the process, I learned quite a bit about how the Vbias setup operates.

One place where there was a difference was this resistor in the base of Q8. In the BJT design, he's put base resistors for both Q8 and Q3 (IIRC they're 1K each). In the MOSFET design, they're not there. Since I'd started copying with the MOSFET schematic, they're not present here too. I was discussing this only yesterday with one of my senior-designer friends and we couldn't figure out the purpose of the resistors. Now, with your explanation, I know. I'll put them in.

Should I put them in for the MOSFET design too?

About a separate current source, even AndrewT had said this is better. I am running too tight on space, so I'm dragging my feet. :( It'll require an extra transistor, three or four resistors and an electrolytic cap.
 
Is R22 too large?

AndrewT was suggesting that 47 for R22 is beginning the VAS protection triggering at only twice the CCS rating. He thought a larger latitude would be better to avoid protection-induced distortions on transients... maybe 22 instead of 47. Since I have no clue about these things, and his suggestions made sense, I thought I could always proceed with the topology and PCB now, fix the values later.

What do you think about that resistor?

Incidentally, it's 47 for both the MOSFET circuit and this one.
 
Your resistor/inductor won't be that big, i think. On my own circuits it's only 12.7mm long and about 5mm diameter total. The inductor passes most of the current in the audio band, so the resistor normally isn't dissipating much power at all.

Fair comment on the collector resistors. The main thing that destroys them though I think, is how inductive they are. The best kind of resistors to use here are low inductance, which isn't normally the case with most cheap wirewounds.

The two resistors and capacitor that you see on the current source are a "bootstrap" to allow the current source to come up to stability quickly. I think this is to eliminate the turnon thump. You might find it is fine without these parts. Quite a lot of amps don't use them.

Another alternative current source is to use a diode based voltage reference instead of another transistor. A 3mm green LED makes a good low noise reference. Here is a good page showing a number of different types.

The turnon thump is not much of an issue. If you are putting a "DC Protect" circuit in to disconnect the speakers via relays (for if the amp goes faulty), it is a simple matter to add a 2-3 second delay on powerup here to counteract the thump.

I wouldn't call myself an expert on these matters, in fact a lot of it I have just learnt by messing about and looking at lots of schematics. I should probably get myself some books and learn the hardcore theory and maths behind it all. While it's not a sure fire way to get a working design, I have done a lot of messing around in SPICE and have learnt from there whats bad and whats good :)
 
Added two base resistors R39 and R40 for the two CCS transistors, as per feedback from you and because they were there in the original schematic anyway.

An externally hosted image should be here but it was not working when we last tested it.


An externally hosted image should be here but it was not working when we last tested it.


I'd like to separate out the two CCS, specially if it means just adding one more transistor and resistor, but I'm still studying the board for available space.
 
jaycee said:
Your resistor/inductor won't be that big, i think. On my own circuits it's only 12.7mm long and about 5mm diameter total.
Oh, okay, that's cool. How many turns and what gauge of wire do you use? Slone's books always talked about 18-20 turns of 14-16 AWG wire, wound on a half-inch-dia former.

Fair comment on the collector resistors. The main thing that destroys them though I think, is how inductive they are. The best kind of resistors to use here are low inductance, which isn't normally the case with most cheap wirewounds.
Absolutely. I don't know where to get those good non-inductive resistors. And I don't want to buy boutique "audiophile grade" resistors for two or three dollars each, I guess. I'm a learner... I'm not a world-beating amp designer.

The two resistors and capacitor that you see on the current source are a "bootstrap" to allow the current source to come up to stability quickly. I think this is to eliminate the turnon thump. You might find it is fine without these parts.
This I didn't know. I'm still looking through that Web page you pointed me to, on current sources and mirrors. I believe Rod Elliott uses a green-eyed CCS for his P3A. I guess one can just use two 1N4148 in series, no?

The turnon thump is not much of an issue. If you are putting a "DC Protect" circuit in to disconnect the speakers via relays (for if the amp goes faulty), it is a simple matter to add a 2-3 second delay on powerup here to counteract the thump.
I may not always put in a delay circuit... I'm toying with the idea of embedding these amps into active speakers at some date.

While it's not a sure fire way to get a working design, I have done a lot of messing around in SPICE and have learnt from there whats bad and whats good :)
Other than lots more reading, this SPICE part is the part I need to do. I don't have a simulator set up on my PC, and I haven't yet learned how to use those tools. I downloaded Simetrix once (it works on Linux, and I use Linux on my laptop), and simulated a phono preamp, but that's where it ended. I'll also have to go hunting for the SPICE models of all the transistors. But I think the effort will help me a lot.
 
Ex-Moderator
Joined 2002
tcpip said:
I think I'll need more details. :(

Good morning, and Happy New Year tcpip!

Okay, I misinterpreted the topology of the output stage. I blame beer. :)

But look at the feedback routing, it loops all over the board to get to the stage around Q16. Not good. I would loose those oversized copper floods around the output, you don't need them, and that would allow much better routing in the central areas. Maybe swap the driver/bias sections around in that area so you can get Q15/16 much closer together. Feedback also needs to come from a point directly in the middle of the two output resistors to reduce distortion.

I'm also confused by your grounding, you seem to have lots of different grounds on the schematic, (good!), but I can only see two powergrounds on the board?
 
I would recommand to make first tests with a simple pole compensantion, not the two poles one.

A good indicator for oscillations is the heating of zobel resistance R31.

I've seen decoupling resistors like R36 and R37 to be burnt when aging, they were slowly degraded due the in-rsuh current at every startup.
 
pinkmouse said:
Good morning, and Happy New Year tcpip!
Thanks, and fully reciprocated. :) I was wondering if you'd ever return to this thread. :D

But look at the feedback routing, it loops all over the board to get to the stage around Q16. Not good. I would loose those oversized copper floods around the output, you don't need them, and that would allow much better routing in the central areas. Maybe swap the driver/bias sections around in that area so you can get Q15/16 much closer together. Feedback also needs to come from a point directly in the middle of the two output resistors to reduce distortion.
I'm following what you're saying, but I'll need more inputs to really understand its importance. I thought that Q15 and Q16 were just OPS protection circuits, hence the quality of distortion and so on was secondary for the tracks that led to them... i.e. I could take more liberties with those parts. The part I was trying to be careful about was the track which reached from the top of the PCB to the bottom to reach the base of Q2.

And about the feedback routing "looping all over the board to get to the stage around Q16", I tried re-tracing that track, and found that it's a single L-shaped red track starting from one end of the 5W resistor R24, bending at right angles below R26, and coming down straight towards Q16. Is this supposed to be too much looping all over the board? How straight does this track need to be?

Do I need to keep Q15 and Q16 close to each other? I was trying to follow the schematic and keep them on either side of the central copper, which implies they'd be far away from each other.

I'm also confused by your grounding, you seem to have lots of different grounds on the schematic, (good!), but I can only see two powergrounds on the board?
If you see the ground symbols on the schematic, all the ground points connect to either a symbol labelled "GNDA" or another symbol called "PE". These are special supply symbols in Eagle; any number of nets connected to the same supply symbols are physically treated as a single net. So physically there are just two nets in the schematic which go to ground. On the PCB, each of these nets goes to one solder-pad, labelled PGND1 and PGND2. Basically, I've tried to keep all grounds "close to" the upper supply rail running from PGND1, and all grounds "close to" the lower supply rail running from PGND2.

So the PCB has three points which will eventually connect to the off-board star-ground point. The two power grounds are PGND1 and PGND2, and the signal (and feedback) ground points go to INGND.
 
forr said:
I would recommand to make first tests with a simple pole compensantion, not the two poles one.
Since the schematic is of a tried-and-tested amp which Randy Slone has built many times, I'm assuming that I don't have to do anything with those parts. Even if I had to, I wouldn't be competent to make changes there.

A good indicator for oscillations is the heating of zobel resistance R31.
There will also be a resistor in parallel with L1 in the output. I guess that too will heat up.

I've seen decoupling resistors like R36 and R37 to be burnt when aging, they were slowly degraded due the in-rsuh current at every startup.
Yes, that occurred to me too, and I was wondering what to do about that. Will it be any better if I use 5W resistors? Whatever be the wattage, the sudden thermal stress of heating and cooling due to the inrush current will certainly take its toll.
 
Hi TCPIP,
Considerning the two pole compensantion, I strongly advise you not to use it first until you amp is fully well working. In fact, you have not to make any modification, just do not mount R13. It will be then just as all Self's amps with a simple Miller compensation. As you are making your own PCB, you may experience differently from Randy Sloane.

--- There will also be a resistor in parallel with L1 in the output. I guess that too will heat up.---
Not in my experience. Self reports that L1 can sometimes heat up due to very high currents in high power applications.

I found burnt R36 and R37 in twenty years equipment which was probabaly often switched on. 5 W resistors may be be a good idea.
 
tcpip said:
Oh, okay, that's cool. How many turns and what gauge of wire do you use? Slone's books always talked about 18-20 turns of 14-16 AWG wire, wound on a half-inch-dia former.

To be honest I don't think it's all that critical. I'm not sure what wire gauge I used, it comes from power chokes salvaged from dead PC power supplies usually :) I think it's about 1.5mm dia. I normally just wind this around the resistor body, as many turns as I can fit onto it.

tcpip said:
Absolutely. I don't know where to get those good non-inductive resistors. And I don't want to buy boutique "audiophile grade" resistors for two or three dollars each, I guess. I'm a learner... I'm not a world-beating amp designer.

Audiophile grade is definitely not worth spending the extra for, IMHO. Having said that I've not had any problems with cheap 3W wirewounds either.

tcpip said:
This I didn't know. I'm still looking through that Web page you pointed me to, on current sources and mirrors. I believe Rod Elliott uses a green-eyed CCS for his P3A. I guess one can just use two 1N4148 in series, no?

He does use the LED referenced current source, yes. You can also use two 1N4148's but they are noisier and more subject to temperature drift. I'd say use the LED.

tcpip said:
Other than lots more reading, this SPICE part is the part I need to do. I don't have a simulator set up on my PC, and I haven't yet learned how to use those tools. I downloaded Simetrix once (it works on Linux, and I use Linux on my laptop), and simulated a phono preamp, but that's where it ended. I'll also have to go hunting for the SPICE models of all the transistors. But I think the effort will help me a lot.

Check out LTSpice. This is rather good, free, and apparently runs very well on WINE. You can get it here. There is a good Yahoo! Group as well for it
 
Hi tcpip

The downside of using resistors in the base of the VAS current source is that this slows it down! I haven't seen too many Miller compensated amps. go unstable when the VAS saturates, but my preference is to add anti-saturation diodes. You can use 1N4148's up to about +/- 35V but higher voltages will need a higher voltage diode (still needs to have very low capacitance).

One reason not many people seem to favour the CFP is because it can only let the output transistors switch off slowly becuase the base is discharging through a 100 ohm or thereabouts resistor. The capacitance of a forward biased base junction is substantial, and really could do with some help from the drivers. The fully-complementary configuration enables the drivers to pull some of the charge out instead of letting it discharge through a resistor.

But you have chosen to use high speed 2SA 1302 etc outputs. These should n't have so much problem as the older slower devices. On the other hand, in a CFP they may tend to oscillate.

I am not a big fan of Miller compensation even with two pole. You've put a 2k load on the VAS at high frequencies. If you are going to make sure your VAS can drive this it needs at least 15 mA (depengin on your supply voltage) just to dump current into this load. This is the reason I think many amplifiers are designed with too little VAS current. Indeed, give Self some credit because he mentioned buffering the VAS with an emitter follower (but his VAS stage only ran at about 6 mA). I think an emitter follower is OTT, when just a tad more current in the VAS will fix the problem. But on the down side (there are always trade offs) the VAS dissipation increases. To my view, this is a small price to pay for the convenience of a relatively simple design.

I'd recommend that you set the VAS current to about 25-30 mA.

You are right in that with a CFP you only need to thermally join the driver and bias stabiliser.

Two pole compensation often puts a spike in the frequency response around the upper -3dB point. Have you checked the effect on your design?

Cheers
John
 
forr said:
Considerning the two pole compensantion, I strongly advise you not to use it first until you amp is fully well working. In fact, you have not to make any modification, just do not mount R13.
Okay, I'll keep this in mind when I'm building. As long as the PCB remains the same, I can try these modifications.
--- There will also be a resistor in parallel with L1 in the output. I guess that too will heat up.---
I was thinking that one simple solution would be to just tell whoever is using this amp to never switch it off. That way, the thermal cycles can be reduced a lot. I've always thought it's a generally good idea to keep audio electronic circuits powered on all the time.
 
jaycee said:
To be honest I don't think it's all that critical. I'm not sure what wire gauge I used....
Got it. :)

He does use the LED referenced current source, yes. You can also use two 1N4148's but they are noisier and more subject to temperature drift. I'd say use the LED.
I didn't know about the 1N4148 being noisy, thanks. I'll look towards the green light.

Check out LTSpice. This is rather good, free, and apparently runs very well on WINE.
Will try. In fact, I'm getting so sick of this two-OS problem that I'm seriously thinking of setting up VMWare on the laptop and getting a full-blown Windows system running on it, so that I can really do almost everything Windowy on it short of some devices.

Have you used Simetrix, BTW?
 
Hi John and thanks for dropping by. :)
john_ellis said:
The downside of using resistors in the base of the VAS current source is that this slows it down! I haven't seen too many Miller compensated amps. go unstable when the VAS saturates, but my preference is to add anti-saturation diodes. You can use 1N4148's up to about +/- 35V but higher voltages will need a higher voltage diode (still needs to have very low capacitance).
Can you give me a part number?

And at a conceptual level, why does the CCS need to be fast-reacting if it's feeding some circuits lower down which are operating in Class A anyway? Won't the load circuits draw a steady time-invariant current? (I'm totally a beginner, so please be patient. :) )

One reason not many people seem to favour the CFP is because it can only let the output transistors switch off slowly because the base is discharging through a 100 ohm or thereabouts resistor. The capacitance of a forward biased base junction is substantial, and really could do with some help from the drivers. The fully-complementary configuration enables the drivers to pull some of the charge out instead of letting it discharge through a resistor.
By fully complementary, do you mean a mirror image topology, with the VAS and input stages mirrored? If you're referring to output topology, I know only of CF and EF... I didn't know there was anything called fully complementary OPS.

Does capacitive slowness of the OPS devices become better handled if I use driver transistors with very low Cob? I'm probably mixing things up, but I thought I'd ask.

And is this capacitive slowness that you're referring to going to contribute to switching distortion, where the OPS devices don't switch off the moment that drive current is removed from their bases? If I've understood right, this is tackled by adding a cap connecting the two bases of the P and N devices in an EF topology, isn't it? Or am I mixing things up again?

But you have chosen to use high speed 2SA 1302 etc outputs. These should n't have so much problem as the older slower devices. On the other hand, in a CFP they may tend to oscillate.
Actually, I intend to use 2SA1943 and 2SC5200 devices, because I have them. I'm told even these have gone out of production. I was also thinking of using MJ15003/004 devices (TO3), since Slone said they could be used.

I am not a big fan of Miller compensation even with two pole. You've put a 2k load on the VAS at high frequencies. If you are going to make sure your VAS can drive this it needs at least 15 mA (depending on your supply voltage) just to dump current into this load. This is the reason I think many amplifiers are designed with too little VAS current.
And if have understood the circuit right, the VAS current has been set to about 6mA by using 100 Ohms of R14. That's a few times lower than what you're talking about.

I'd recommend that you set the VAS current to about 25-30 mA.
This will mean setting R14 to about 18-22 Ohms and R22 to about 10 Ohms, right? The VAS transistors are already heatsinked, so the heat shouldn't be too much of a problem.

What are your suggestions for the VAS devices? Should I use the 669/649 set of devices or switch to any of the newer low-Cob devices?

Two pole compensation often puts a spike in the frequency response around the upper -3dB point. Have you checked the effect on your design?
I have actually done no checking... I'm just going with whatever was there in Randy Slone's schematic. I'll have to do the entire thing of setting up a simulation program and getting all the device models before I can check these things. :(
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.