So be it then, a single board.
Are going to use the wall wart as discussed or a PCB mounted transformer or a separate transformer wired to the board.
Quescient current is 40 mA per amp if we go 100 mA max then a 24VAC, 5VA transformer should do the job.
Gareth/Hugh if you can start sending me partnumbers and manufacturer or better still PDF data sheets of the major parts, I can start creating the parts library which takes most of the time since I draw these in autoCAD which is mechanically a far more accurate than any PCB program.
Are going to use the wall wart as discussed or a PCB mounted transformer or a separate transformer wired to the board.
Quescient current is 40 mA per amp if we go 100 mA max then a 24VAC, 5VA transformer should do the job.
Gareth/Hugh if you can start sending me partnumbers and manufacturer or better still PDF data sheets of the major parts, I can start creating the parts library which takes most of the time since I draw these in autoCAD which is mechanically a far more accurate than any PCB program.
One more thing, choosing aheat sinks that is internationally easily available is paramount since it would probably have to be bolted to the PCB.
This amp can cost a pretty penny if all the components are imported. I think what I should do is post a PDF of the PCB and people can see the physical sizes of the componets and can place what they already have to see if they fit the footprint of those specified.
This will be double sided PCB since there will be a lot of tracks running across channels to get to the comon points such as power supply, input, output, control switches and pots.
This amp can cost a pretty penny if all the components are imported. I think what I should do is post a PDF of the PCB and people can see the physical sizes of the componets and can place what they already have to see if they fit the footprint of those specified.
This will be double sided PCB since there will be a lot of tracks running across channels to get to the comon points such as power supply, input, output, control switches and pots.
Nico,
1. Power Supply. It seems that AC supplies are harder to find than the world mains smps DC ones. Maybe it's worth considering a relatively accessible 18Vdc 500mA supply, regulating it further to 15V with a LM317, then splitting the 15V single rail supply into two 7.5V rails using the virtual earth topology.
This would be inexpensive, and available off the shelf. I do not see a need for an output to exceed 12Vpp, so this would fit nicely. To me, the wallwart is the way to go; cheap, compact, easy to source, mains powered from 70-250Vac, removes AC induction from the HP amp, and reduces internal complexity.
2. Let's drop the chokes/CMCs. This will save much expense, and in light of the smps, ripple will be extremely, indeed vanishingly small anyway.
3. If we use a single board, mounting controls one side and power transistors the other, we are prescribing a specific case size which limits assembly from the DIY junk box. It makes wonderful sense to use the base of the enclosure for the hot devices, and the controls, input/output/power plugs can all be on one side (or pcb mount controls one side, flying leads for all the connections on the other). The C4793 and A1837 outputs I specify do NOT require wafer insulators, simplifying assembly. This approach means many different enclosure sizes could be utilised, as long as they are metallic. I favor diecast alloy boxes; they make outstanding heatsinks, and this saves on buying discrete heatsinks, which then have to convect their heat to the outside of the enclosure anyway. Inefficient, and undesirable.
4. The dimension of the enclosure will largely be dictated by the size of the board. I suggest we wait until you have some idea of the size during the layout process.
I will do my utmost to supply the information concerning plugs, couplers, devices, pots, etc this weekend.
Thanks again for your willingness to contribute so significantly. The layout is far and away the biggest part of this job (with documentation not far behind!).
Cheers,
Hugh
1. Power Supply. It seems that AC supplies are harder to find than the world mains smps DC ones. Maybe it's worth considering a relatively accessible 18Vdc 500mA supply, regulating it further to 15V with a LM317, then splitting the 15V single rail supply into two 7.5V rails using the virtual earth topology.
This would be inexpensive, and available off the shelf. I do not see a need for an output to exceed 12Vpp, so this would fit nicely. To me, the wallwart is the way to go; cheap, compact, easy to source, mains powered from 70-250Vac, removes AC induction from the HP amp, and reduces internal complexity.
2. Let's drop the chokes/CMCs. This will save much expense, and in light of the smps, ripple will be extremely, indeed vanishingly small anyway.
3. If we use a single board, mounting controls one side and power transistors the other, we are prescribing a specific case size which limits assembly from the DIY junk box. It makes wonderful sense to use the base of the enclosure for the hot devices, and the controls, input/output/power plugs can all be on one side (or pcb mount controls one side, flying leads for all the connections on the other). The C4793 and A1837 outputs I specify do NOT require wafer insulators, simplifying assembly. This approach means many different enclosure sizes could be utilised, as long as they are metallic. I favor diecast alloy boxes; they make outstanding heatsinks, and this saves on buying discrete heatsinks, which then have to convect their heat to the outside of the enclosure anyway. Inefficient, and undesirable.
4. The dimension of the enclosure will largely be dictated by the size of the board. I suggest we wait until you have some idea of the size during the layout process.
I will do my utmost to supply the information concerning plugs, couplers, devices, pots, etc this weekend.
Thanks again for your willingness to contribute so significantly. The layout is far and away the biggest part of this job (with documentation not far behind!).
Cheers,
Hugh
smps concept
Hugh,
Recently I decided to test some of my Switch Mode Power Supply (SMPS) plugpacks:
- My 12volt router power supply had about 50mV or ripple at about 15kHz,
- A Cisco 48volt plugpack (from a Wireless Access Point) had about 5mV of ripple at about 150kHz.
- The smallest open frame SMPS from Jaycar specifies 100mV (rms) of ripple.
I tried simulating with LTspice a variety of power supply ideas to clean this ripple up. I finally decided that a Capacitance Multiplier was (just) better than either a simple series or shunt regulator.
As 24 volt SMPS seem to be universally available (my test is if Mouser has them, then anyone can get them), I suggest you go with as high a voltage as possible (ie 24 volts) as this gives you the most options on filtering the ripple and noise out. Attached is a concept diagram that has some simple input filtering, Nico's virtual earth circuit (but slightly modified), followed by capacitance multipliers (Using separate capacitance multipliers for each channel and rail).
If you are interested in a non inverting Baxandel style tone control circuit, I have a nice circuit from an old Crown preamp. It has gain, and is switch defeatable. Email me if interested and I'll send you a copy.
Paul Bysouth, October 2009
Hugh,
Recently I decided to test some of my Switch Mode Power Supply (SMPS) plugpacks:
- My 12volt router power supply had about 50mV or ripple at about 15kHz,
- A Cisco 48volt plugpack (from a Wireless Access Point) had about 5mV of ripple at about 150kHz.
- The smallest open frame SMPS from Jaycar specifies 100mV (rms) of ripple.
I tried simulating with LTspice a variety of power supply ideas to clean this ripple up. I finally decided that a Capacitance Multiplier was (just) better than either a simple series or shunt regulator.
As 24 volt SMPS seem to be universally available (my test is if Mouser has them, then anyone can get them), I suggest you go with as high a voltage as possible (ie 24 volts) as this gives you the most options on filtering the ripple and noise out. Attached is a concept diagram that has some simple input filtering, Nico's virtual earth circuit (but slightly modified), followed by capacitance multipliers (Using separate capacitance multipliers for each channel and rail).
If you are interested in a non inverting Baxandel style tone control circuit, I have a nice circuit from an old Crown preamp. It has gain, and is switch defeatable. Email me if interested and I'll send you a copy.
Paul Bysouth, October 2009
Attachments
Thanks Paul,
Much appreciated. These are good figures, and your circuit both incorporates the VE and is very simple and effective. It does seem that R1 and R2 could gainfully be replaced by a ferrite CMC (nice and cheap from Rockby!) and this would offer just a tad more voltage on the rails.
Yes, do please email to me, most appreciative.
Hugh
Much appreciated. These are good figures, and your circuit both incorporates the VE and is very simple and effective. It does seem that R1 and R2 could gainfully be replaced by a ferrite CMC (nice and cheap from Rockby!) and this would offer just a tad more voltage on the rails.
Yes, do please email to me, most appreciative.
Hugh
Nico,
My thinking was that you didn't have any emitter resistors (R5,R6 in my diagram) and that didn't define or limit the current though Q1,Q2 (in my diagram), so I added emitter resistors. Then I decided that one zener was easier to wire than two diodes, so I increased R5,R6.
I did some spice testing with both even and uneven loads and it looked OK, given that the headphone amp loads aren't going to be too uneven. Do you think that two diodes rather than a zener and small (or even zero) R5,R6 values would be better?
Paul Bysouth, Oct 2009.
My thinking was that you didn't have any emitter resistors (R5,R6 in my diagram) and that didn't define or limit the current though Q1,Q2 (in my diagram), so I added emitter resistors. Then I decided that one zener was easier to wire than two diodes, so I increased R5,R6.
I did some spice testing with both even and uneven loads and it looked OK, given that the headphone amp loads aren't going to be too uneven. Do you think that two diodes rather than a zener and small (or even zero) R5,R6 values would be better?
Paul Bysouth, Oct 2009.
Paul,
I support the capacitance multiplier, I use it in my own head phone amplifier and it does clean up the rails nicely. It introduces a slow start-up to eliminate pops at switch on as well.
I think Hugh has frozen the design and every new post suggesting an alternative will drag it out - some posts will contradict others and the design will go in circles.
There is of course nothing wrong with eventually ending with an ultimate design, but each design has to be tested, listened to and compared, not just simulated.
Simulation is a good start and may speed up the initial design. But like everything else it is not the end all.
Think about it if you only simulated the chroma, hue and contrast drivers for TV and go into production with the simulated results. Can one guarantee that the colours will be absolutely true and natural from the outset. I think this is were simulators fall very short.
If one does not prototype and test the results you can wind up in trouble especially in a commercial environment. In DIY it is easy, you just call it quits and bin it.
My headphone amplifier although simulated in the initial design and verified again and again as the design matured went through twelve iterations (prototypes) and took seven months from kick-off to finalisation - the development cost me around US$ 6000 in real cash, (excluding my time ) to end up with my current headphone amp.
Kind regards
Nico
I support the capacitance multiplier, I use it in my own head phone amplifier and it does clean up the rails nicely. It introduces a slow start-up to eliminate pops at switch on as well.
I think Hugh has frozen the design and every new post suggesting an alternative will drag it out - some posts will contradict others and the design will go in circles.
There is of course nothing wrong with eventually ending with an ultimate design, but each design has to be tested, listened to and compared, not just simulated.
Simulation is a good start and may speed up the initial design. But like everything else it is not the end all.
Think about it if you only simulated the chroma, hue and contrast drivers for TV and go into production with the simulated results. Can one guarantee that the colours will be absolutely true and natural from the outset. I think this is were simulators fall very short.
If one does not prototype and test the results you can wind up in trouble especially in a commercial environment. In DIY it is easy, you just call it quits and bin it.
My headphone amplifier although simulated in the initial design and verified again and again as the design matured went through twelve iterations (prototypes) and took seven months from kick-off to finalisation - the development cost me around US$ 6000 in real cash, (excluding my time ) to end up with my current headphone amp.
Kind regards
Nico
Paul,
Nico is well qualified (I won't embarrass him by telling you his qualifications or his commercial successes in electronics!) but I have come to greatly respect his point of view. One ignores Nico at one's peril!
OTOH, I agree about defining current in the VE devices. I like your idea better than Nico's, however, it has to be said that any resistor in the emitters will greatly increase Zout, and this militates against the function we seek.
But I agree with Nico that a couple of 1N4148s is a good bias generator, even if we use 0.22R emitter resistors. re will limit bias current through these devices as there is only DC to contend with, no AC signal at all.
We do have some wiggle room with the design. It's not frozen just yet, but it's getting very cold...... this weekend will see it in deep freeze!
Cheers,
Hugh
Nico is well qualified (I won't embarrass him by telling you his qualifications or his commercial successes in electronics!) but I have come to greatly respect his point of view. One ignores Nico at one's peril!
OTOH, I agree about defining current in the VE devices. I like your idea better than Nico's, however, it has to be said that any resistor in the emitters will greatly increase Zout, and this militates against the function we seek.
But I agree with Nico that a couple of 1N4148s is a good bias generator, even if we use 0.22R emitter resistors. re will limit bias current through these devices as there is only DC to contend with, no AC signal at all.
We do have some wiggle room with the design. It's not frozen just yet, but it's getting very cold...... this weekend will see it in deep freeze!
Cheers,
Hugh
Nico,
You right, there is too much current going through my virtual circuit. These things tend to slip through when your working on 3 power supplies and trying to build a voltage regulator for a 1965 Volkswagen all at the same time (and I've just burnt my finger on the soldering iron).
I've also just run some more simulations, and I agree that two resistors would also be OK to define the virtual earth, as there is very little current through them (and that's almost entirely DC).
Paul Bysouth
You right, there is too much current going through my virtual circuit. These things tend to slip through when your working on 3 power supplies and trying to build a voltage regulator for a 1965 Volkswagen all at the same time (and I've just burnt my finger on the soldering iron).
I've also just run some more simulations, and I agree that two resistors would also be OK to define the virtual earth, as there is very little current through them (and that's almost entirely DC).
Paul Bysouth
In reply to post 550:
The C-multiplier might be better because of higher BW, etc. But we don't use one in the first place because of the Zout. So to do this we would put the multiplier in front of the regulator.
The performance of the multiplier as a simple noise isolator is increased a heck of a lot if you cascode the pass device (and then the device on the right can be made a lower power version since it only has 1-2V across it). After that, load rejection is our only worry. Ask if you need a schematic.
And perhaps the zener could be made an LED or a string of them?
With regard to Zout, there are two factors:
1: Emitter resistors. We can decrease Zout by decreasing these, but we would have to decrease bias voltage too.
2: EF transconductance. This depends on the bias current of the EF. A higher bias current will translate to lower Zout because of transconductance curves. Emitter resistors are likely to swamp this factor though since it equates to somewhere around 200mohms depending on which class.
I agree that two resistors should be enough for the virtual earth. With those 1000uF bypass caps, any sound will be shunted from the virtual earth. I suppose the point is what is our maximum value we are willing to use for those caps, because it will determine whether an EF virtual earth is necessary.
- keantoken
The C-multiplier might be better because of higher BW, etc. But we don't use one in the first place because of the Zout. So to do this we would put the multiplier in front of the regulator.
The performance of the multiplier as a simple noise isolator is increased a heck of a lot if you cascode the pass device (and then the device on the right can be made a lower power version since it only has 1-2V across it). After that, load rejection is our only worry. Ask if you need a schematic.
And perhaps the zener could be made an LED or a string of them?
With regard to Zout, there are two factors:
1: Emitter resistors. We can decrease Zout by decreasing these, but we would have to decrease bias voltage too.
2: EF transconductance. This depends on the bias current of the EF. A higher bias current will translate to lower Zout because of transconductance curves. Emitter resistors are likely to swamp this factor though since it equates to somewhere around 200mohms depending on which class.
I agree that two resistors should be enough for the virtual earth. With those 1000uF bypass caps, any sound will be shunted from the virtual earth. I suppose the point is what is our maximum value we are willing to use for those caps, because it will determine whether an EF virtual earth is necessary.
- keantoken
Hi KT,
I'd have to say your job here is not yet finished..... keep on truckin'!
Your comments about Zout of the output stage, and the effects of the highish 10R emitter resistors, are well made. However, Zout of an emitter follower is approximatley 26/mA, so if it's passing say 40mA at idle, and operating at around this point in Class A, Zout is typically about 0.65R, pretty low. Add to this the high emitter resistor, and we have 10.65R, but this is reduced dynamically by feedback, which is around 40dB at audio frequencies. So, in truth, the Zout is around 100 times less, that is, 0.10R approximately. Of course, if we insert even a 10R series resistor, let alone a 120R as recommend in the IEC standard, we throw out this advantage utterly. So, really, it's almost academic.
The VE topology is a simple, convenient way to completely scotch hum and noise. If headphones can do over 90dBA at the ear with just one milliwatt, we are talking of signals of 80mVp into 32R. The situation is ameliorated if a series resistor of 120R is used, as this will voltage divide down the noise by a factor of six times for 24R cans, being 15.5dB, a worthwhile improvement.
But consider what an output of 80mVp means in terms of noise, if it can generate 90dBA at the ear. Let's look at 1mV of noise, or 38dB down. This would be at 52dBA at the ear, which is highly audible..... and therefore totally unacceptable.
Let us try for 100uV of noise, 20dB better, or 58dB down. This would give us the equivalent of 32dBA noise at the ear, which, if the listener were not already deaf from his vigorous session at 90dBA, would be on the verge of audibility. (110dBA will cause permanent ear damage at 1m, and much more damage again if applied directly to the ear.)
Thus, we really need better than about 50uV, a very stringent standard indeed.
There are two significant factors to consider. What frequency is this noise? Secondly, can we use some sort of noise cancellation technique, so that we can avoid throwing lashes of capacitance at the problem and reducing it through brute force?
Nominally, with mains powered equipment, the noise is at 100Hz (50Hz mains) or 120Hz (60Hz mains). These are highly audible, as they fall roughly an octave below middle C (256.12Hz). But the noise need not be at mains frequency if we use a switch mode supply.
These typically operate over the range 15KHz to 100KHz. Clearly, caps work better at higher frequencies, as their capacitive reactance (recip2pifC) decreases with increasing f. This is convenient, because it means we we have ultrasonic ripple artefacts, easy to remove with small caps, and thus our only requirement for large caps will be the earth return for the output stage currents, that is, audio. They are more problematic, of course, as they extend down to approximately 25Hz, depending on the recording.
I therefore do not feel that a cap multiplier is either needed or desirable, because it introduces the non-linearities of the active pass device. Furthermore, we don't need slow switch on, because until the rails settle, the offset of the amp topology chosen will be quite high, and we don't want our cans to suffer for a long period. This is also at the heart of the worthy suggestion for a mute switch, which is convenient for other reasons as well.
As for noise cancellation, assume we have identical common mode noise on each rail, but in antiphase. If we transfer this noise via caps to a single point from each rail, our virtual earth, then the antiphase common mode noise will elegantly cancel and give us an inky black earth point, which can then be used for all audio input and output connections. It is true that a proper earth - copper plate buried in the garden - is quiet, but it's not so practical, and this is a far smarter way to do it because it can be done in complete isolation from the real world.
So, here's the plan: Single rail 24V smps supply, some small high frequency film caps to remove any output ripple, LM317 regulator to further remove any grunge, a couple of small caps, one of them electrolytic, as a reservoir, and then our virtual earth.
I will put up these diagrams very soon.
Cheers,
Hugh
I'd have to say your job here is not yet finished..... keep on truckin'!
Your comments about Zout of the output stage, and the effects of the highish 10R emitter resistors, are well made. However, Zout of an emitter follower is approximatley 26/mA, so if it's passing say 40mA at idle, and operating at around this point in Class A, Zout is typically about 0.65R, pretty low. Add to this the high emitter resistor, and we have 10.65R, but this is reduced dynamically by feedback, which is around 40dB at audio frequencies. So, in truth, the Zout is around 100 times less, that is, 0.10R approximately. Of course, if we insert even a 10R series resistor, let alone a 120R as recommend in the IEC standard, we throw out this advantage utterly. So, really, it's almost academic.
The VE topology is a simple, convenient way to completely scotch hum and noise. If headphones can do over 90dBA at the ear with just one milliwatt, we are talking of signals of 80mVp into 32R. The situation is ameliorated if a series resistor of 120R is used, as this will voltage divide down the noise by a factor of six times for 24R cans, being 15.5dB, a worthwhile improvement.
But consider what an output of 80mVp means in terms of noise, if it can generate 90dBA at the ear. Let's look at 1mV of noise, or 38dB down. This would be at 52dBA at the ear, which is highly audible..... and therefore totally unacceptable.
Let us try for 100uV of noise, 20dB better, or 58dB down. This would give us the equivalent of 32dBA noise at the ear, which, if the listener were not already deaf from his vigorous session at 90dBA, would be on the verge of audibility. (110dBA will cause permanent ear damage at 1m, and much more damage again if applied directly to the ear.)
Thus, we really need better than about 50uV, a very stringent standard indeed.
There are two significant factors to consider. What frequency is this noise? Secondly, can we use some sort of noise cancellation technique, so that we can avoid throwing lashes of capacitance at the problem and reducing it through brute force?
Nominally, with mains powered equipment, the noise is at 100Hz (50Hz mains) or 120Hz (60Hz mains). These are highly audible, as they fall roughly an octave below middle C (256.12Hz). But the noise need not be at mains frequency if we use a switch mode supply.
These typically operate over the range 15KHz to 100KHz. Clearly, caps work better at higher frequencies, as their capacitive reactance (recip2pifC) decreases with increasing f. This is convenient, because it means we we have ultrasonic ripple artefacts, easy to remove with small caps, and thus our only requirement for large caps will be the earth return for the output stage currents, that is, audio. They are more problematic, of course, as they extend down to approximately 25Hz, depending on the recording.
I therefore do not feel that a cap multiplier is either needed or desirable, because it introduces the non-linearities of the active pass device. Furthermore, we don't need slow switch on, because until the rails settle, the offset of the amp topology chosen will be quite high, and we don't want our cans to suffer for a long period. This is also at the heart of the worthy suggestion for a mute switch, which is convenient for other reasons as well.
As for noise cancellation, assume we have identical common mode noise on each rail, but in antiphase. If we transfer this noise via caps to a single point from each rail, our virtual earth, then the antiphase common mode noise will elegantly cancel and give us an inky black earth point, which can then be used for all audio input and output connections. It is true that a proper earth - copper plate buried in the garden - is quiet, but it's not so practical, and this is a far smarter way to do it because it can be done in complete isolation from the real world.
So, here's the plan: Single rail 24V smps supply, some small high frequency film caps to remove any output ripple, LM317 regulator to further remove any grunge, a couple of small caps, one of them electrolytic, as a reservoir, and then our virtual earth.
I will put up these diagrams very soon.
Cheers,
Hugh
Last edited: