Hi,
Here's a thought I had. While we now know how to adjust for DC offset in this circuit..
Many posts ago I had suggested putting a resistor in each leg of the current mirror, which would allow the use of a trip pot to adjust for DC offset. Bruno said that didn't work, because after a second or two the resistors would heat up enough for the DC offset to be right back where it was.
So, I wonder if it would be possible to use a positive temp co thermistor in each leg of the current mirror, to balance the DC offset automatically.
It wouldn't be good at start up, but the idea is with say a 5 second warm up period, which is not exactly unheard of, DC offset would be auto balanced by the time the output stage is switched on.
What do you guys think?
Here's a thought I had. While we now know how to adjust for DC offset in this circuit..
Many posts ago I had suggested putting a resistor in each leg of the current mirror, which would allow the use of a trip pot to adjust for DC offset. Bruno said that didn't work, because after a second or two the resistors would heat up enough for the DC offset to be right back where it was.
So, I wonder if it would be possible to use a positive temp co thermistor in each leg of the current mirror, to balance the DC offset automatically.
It wouldn't be good at start up, but the idea is with say a 5 second warm up period, which is not exactly unheard of, DC offset would be auto balanced by the time the output stage is switched on.
What do you guys think?
I use Amidon Iron powered toroidal cores “Type 2” material (red cores), losses are very low even at 1MHz, while distortion OPEN loop is just 0.0015% - 120W 8Ohms 500KHz, with no need to mess about with air-gaps. Distortion components are predominately ODD order due to change of permeability of the core across the audio cycle due to “non even air gaps” within the Iron Powered core.
For lower Open Loop distortion with Iron Powered cores “Gapping” reduces these distortion components even lower.
With Iron powder core’s at low flux densities, the flux tends to concentrate around the “easiest” paths (lowest reluctance) where the magnetic particles are in close proximity – Gapping concretes the “Flux” in a “linear” gap.
“As the flux density increases, the easier path’s are the first to saturate. Those portions of the magnetic particles that have saturated become non-magnetic, making their paths more difficult. Incremental flux increase shifts to adjacent paths, where the magnetic material has not yet saturated, but where the gap is somewhat wider.
This process continues – effectively widening the distributed gap with increasing flux density and as a result, the permeability (and inductance) is progressively reduced – the result is observed as rounding of the B-H curve.”
My thanks to Jaka Racman for providing the leads to this distortion mechanism.
For the DC drift issue, the simples way to resolve the problem is to use thermally coupled transistors for the Input pair & current mirror. Zetex provide dual packaged high voltage transistor – which are also VERY low noise.
Note: - the Zetex duals are dual die, so electrical parameters are not matched (although most likely from the same Wafer batch – if not Wafer) – but thermally they are very closely coupled.
John
For lower Open Loop distortion with Iron Powered cores “Gapping” reduces these distortion components even lower.
With Iron powder core’s at low flux densities, the flux tends to concentrate around the “easiest” paths (lowest reluctance) where the magnetic particles are in close proximity – Gapping concretes the “Flux” in a “linear” gap.
“As the flux density increases, the easier path’s are the first to saturate. Those portions of the magnetic particles that have saturated become non-magnetic, making their paths more difficult. Incremental flux increase shifts to adjacent paths, where the magnetic material has not yet saturated, but where the gap is somewhat wider.
This process continues – effectively widening the distributed gap with increasing flux density and as a result, the permeability (and inductance) is progressively reduced – the result is observed as rounding of the B-H curve.”
My thanks to Jaka Racman for providing the leads to this distortion mechanism.
For the DC drift issue, the simples way to resolve the problem is to use thermally coupled transistors for the Input pair & current mirror. Zetex provide dual packaged high voltage transistor – which are also VERY low noise.
Note: - the Zetex duals are dual die, so electrical parameters are not matched (although most likely from the same Wafer batch – if not Wafer) – but thermally they are very closely coupled.
John
Hi,
I just got some audio out of my prototype. Cleanest noise I've ever heard off of a plastic protoboard!
I can only listen to it for seconds at a time, zero heatsinks on the mosfets as of yet and they get hot fast~! I have no scope to work with though so I'll likely have to settle for that, but I can add some heatsinks.
John, that's excellent information, thank you! How does those type 2 cores you have differ from the average iron powder, if at all? As I do have a few IP cores here..
Thanks,
Chris
I just got some audio out of my prototype. Cleanest noise I've ever heard off of a plastic protoboard!
I can only listen to it for seconds at a time, zero heatsinks on the mosfets as of yet and they get hot fast~! I have no scope to work with though so I'll likely have to settle for that, but I can add some heatsinks.
John, that's excellent information, thank you! How does those type 2 cores you have differ from the average iron powder, if at all? As I do have a few IP cores here..
Thanks,
Chris
Hello,
I just thought I would throw in a few words about the red/uncolored(clear) material #2 cores from Amidon. Those are the same ones as from CWS ByteMark. I was surprised to hear that they are useful in class D filters as I have observed them doing well as power supply output chokes too.
I just thought I would throw in a few words about the red/uncolored(clear) material #2 cores from Amidon. Those are the same ones as from CWS ByteMark. I was surprised to hear that they are useful in class D filters as I have observed them doing well as power supply output chokes too.
Chris,
By comparision, the Pulse eng. cores designed for the Zetex Class D, are very poor with 0.1% OL 100W 8ohms. Also the Pulse cores suffer from large switching losses.
Other "unknowns" also seem to measure about the same i.e. 0.1% 100W
Todate, I havn't found a better core to the "Amidon" type II cores. I no longer source my type II cores from Amidon, but from CWS (Coil Winding services) Sanata Ana CA.
Note: - the results are open loop, so with post inductor FB, the THD will be lower.
John
By comparision, the Pulse eng. cores designed for the Zetex Class D, are very poor with 0.1% OL 100W 8ohms. Also the Pulse cores suffer from large switching losses.
Other "unknowns" also seem to measure about the same i.e. 0.1% 100W
Todate, I havn't found a better core to the "Amidon" type II cores. I no longer source my type II cores from Amidon, but from CWS (Coil Winding services) Sanata Ana CA.
Note: - the results are open loop, so with post inductor FB, the THD will be lower.
John
Talking about output inductor, Micrometals cores also perform very well. I think their type 2 must be very similar to Amidon.
Btw, just for curiosity, does anyone know how does the LC audio Zap pulse perform in this arena? The output filter seems to be 2nd order, and the L has a RM ferrite core. What is the output ripple level, more or less, at typical supply voltages (say, +/-40V)? It is in the 1Vpp order, isn't it?
I have asked LCaudio ,but they seem to be always at holidays, so they don't answer.
Best regards
Btw, just for curiosity, does anyone know how does the LC audio Zap pulse perform in this arena? The output filter seems to be 2nd order, and the L has a RM ferrite core. What is the output ripple level, more or less, at typical supply voltages (say, +/-40V)? It is in the 1Vpp order, isn't it?
I have asked LCaudio ,but they seem to be always at holidays, so they don't answer.
Best regards
TriPath originally used the Amidon Type II then recommended the MicroMetals version so I guess there the same.
The original version was designed by James Lau at Amidon for TriPath - James now owns and runs CWS - has manufacturing in China so very good prices.
Also noticed other Class D manufacturers using them on their Class D Eval. boards.
John
The original version was designed by James Lau at Amidon for TriPath - James now owns and runs CWS - has manufacturing in China so very good prices.
Also noticed other Class D manufacturers using them on their Class D Eval. boards.
John
Hi John,
Thanks for the reply.
I was impressed by the open loop figures and if they perform as you say (I'm sure they do) I see no reason why not to use them over anything else.
I have more questions but I think I'll read up on them before I go on with what is likely "basics".
With that in mind you seem to also have an impressive source of information I was wondering if you wouldn't mind posting a link to it?
Tripath, I think, also switches alot faster than we're interested in here, so that's good.
I may be blind at this hour of the day, but I was just going through the Zetex site, they don't seem to list any noise ratings in their data sheets?
The only reference I've found there regarding low noise is that they seem to prefer their medium power transistors over small signals ones, but that wasn't in regards to their dual die models. While their dual die medium power seem to be for switching, yet have 150+ ns rise and fall times? Again, no noise figures on them at all. It's got me a bit confused.
Must sleep.
Thanks,
Chris
Thanks for the reply.
I was impressed by the open loop figures and if they perform as you say (I'm sure they do) I see no reason why not to use them over anything else.
I have more questions but I think I'll read up on them before I go on with what is likely "basics".
With that in mind you seem to also have an impressive source of information I was wondering if you wouldn't mind posting a link to it?
Tripath, I think, also switches alot faster than we're interested in here, so that's good.
I may be blind at this hour of the day, but I was just going through the Zetex site, they don't seem to list any noise ratings in their data sheets?
The only reference I've found there regarding low noise is that they seem to prefer their medium power transistors over small signals ones, but that wasn't in regards to their dual die models. While their dual die medium power seem to be for switching, yet have 150+ ns rise and fall times? Again, no noise figures on them at all. It's got me a bit confused.
Must sleep.
Thanks,
Chris
Zetex don’t publish noise figures for their device’s, as it’s not there target market. Transistor noise is a function of Base resistance and is heavily dependent on operating current. That’s why low noise head amps use parallel small signal transistors or medium power devices. The lowest noise input stage I’ve built uses ZDT758 (PNP transistors have lower noise then there direct NPN counterparts); they have a noise figure of less then 0.25dB per device. My experience so far, is that all Zetex medium / high power devices have very low noise.
I’ve no direct working experience of Bruno’s UCD circuit topology, it’s my understanding (but not confirmed), that the UCD180 modules use BF821 for the input pairs, and BC847 for the current mirrors. The BF821 are not known for there switching speed or low noise performance. This would seem to suggest that the ZETEX medium power dual die devices should work well as input devices and possibly current mirrors.
I’m afraid my sources of information have been collated over many years, and from my own research. I believe that the information about “Rounding of the B-H characteristics” originally come from a Unitrode / Motorola paper titled “Magnetic Core Characteristics” – Jaka Racman originally gave me the link (once again thanks Jaka)….
After research that basically involved taking the standard Iron powered mixture used to manufacture the Type II cores, and “Rumbling” it for longer (1 week extra!) to reduce and smooth the particles / grain size, I can confirm that the smoother / smaller particle size iron powered cores performed much better, with open loop THD of 0.0006% 120W 8Ohms 500KHz compared to the control batch manufactured from the standard pre “extra rumbled mix” of 0.0015% under the same test conditions.
John
I’ve no direct working experience of Bruno’s UCD circuit topology, it’s my understanding (but not confirmed), that the UCD180 modules use BF821 for the input pairs, and BC847 for the current mirrors. The BF821 are not known for there switching speed or low noise performance. This would seem to suggest that the ZETEX medium power dual die devices should work well as input devices and possibly current mirrors.
I’m afraid my sources of information have been collated over many years, and from my own research. I believe that the information about “Rounding of the B-H characteristics” originally come from a Unitrode / Motorola paper titled “Magnetic Core Characteristics” – Jaka Racman originally gave me the link (once again thanks Jaka)….
After research that basically involved taking the standard Iron powered mixture used to manufacture the Type II cores, and “Rumbling” it for longer (1 week extra!) to reduce and smooth the particles / grain size, I can confirm that the smoother / smaller particle size iron powered cores performed much better, with open loop THD of 0.0006% 120W 8Ohms 500KHz compared to the control batch manufactured from the standard pre “extra rumbled mix” of 0.0015% under the same test conditions.
John
Hi,
Once again John, many thanks. It's nice to know what they do differently to produce better IP cores.
Ahhh, I just tripped over a great little part for the full bridge design, to the tune of a quad matched set! Perfect.
Differential full bridge:
Comming soon.
Regards,
Chris
Once again John, many thanks. It's nice to know what they do differently to produce better IP cores.
Ahhh, I just tripped over a great little part for the full bridge design, to the tune of a quad matched set! Perfect.
Differential full bridge:
Comming soon.
Regards,
Chris
That's only if your amp has two input stages. If you build a full bridge amp with only one comparator (as one would normally do), a single matched pair will do. How to convert a discrete HB UcD into FB by putting two extra current outputs on the comparator stage has been posted somewhere round here.classd4sure said:
Re BF821 (Johnw): This is a provision for voltages over +/-60V. Below that a dead normal pair of BC856B will do fine.
The input transistors don't switch (though they may cut off when the differential input voltage is big enough). The actual switching happens in the "thing below" that isn't exactly a current mirror, "current comparator" would be more correct.
Even then the transistors used there don't switch in the sense of "saturation", so whether the transistors are marketed as "switching" is not an issue.
Hi Bruno, Welcome back I do hope you had a great time.
I'm aware of how to get a full bridge with one comparator going, Johan and I worked that out a week or so ago. The driver pairs just needed more current in order to account for the greater loading of having doubled everything. We had realized current hogging could be an issue with such an arrangment though, that's in fact where the quad transistor comes into play 😀
So, I'm now using some Analog Devices transistors, the SSM-2220 as the input pair, and the MAT-04 as the quad floating outputs. They're only 35V and 40V parts, but for a full bridge with +-30Volt rails, should make a fair bit of power I think, 500W according to simulation, for a 4ohm load, but that will depend heavily on the power supply. They already have models in pspice, offer free samples... I'm happy with them.
I think I fought with my full bridge circuit most of the night over nothing, the amplitude of the feedback signals decays a great deal, almost to nothing. However I realized after that the difference between the two feedback signals, remains unchanged. So I now think it's OK. No circuit changes seem to affect the decay anyway.
I've had some good results with my full bridge up to a 1000Watts, however I've made some changes, so I've a few rough edges to smooth over and then I'll be posting a fully differential full bridge for everyone.
Re. the input pair not switching, it seems like they do. Constantly switching between the potential of "the thing below". Though I guess the input pair in question is working in a linear fashion, you'd at least want them of sufficient bandwidth to be able to reproduce the switching faithfully. Right? Ft>>>>Fs
I've found it a pure shame, I can smooth the output out completly and achieve amazing results THD wise, by increasing the value of the output filter cap to say double what it should be. That would no longer be a proper filter though would it. I wonder if that actually matters with this type of feedback/modulation scheme?
Regards,
Chris
I'm aware of how to get a full bridge with one comparator going, Johan and I worked that out a week or so ago. The driver pairs just needed more current in order to account for the greater loading of having doubled everything. We had realized current hogging could be an issue with such an arrangment though, that's in fact where the quad transistor comes into play 😀
So, I'm now using some Analog Devices transistors, the SSM-2220 as the input pair, and the MAT-04 as the quad floating outputs. They're only 35V and 40V parts, but for a full bridge with +-30Volt rails, should make a fair bit of power I think, 500W according to simulation, for a 4ohm load, but that will depend heavily on the power supply. They already have models in pspice, offer free samples... I'm happy with them.
I think I fought with my full bridge circuit most of the night over nothing, the amplitude of the feedback signals decays a great deal, almost to nothing. However I realized after that the difference between the two feedback signals, remains unchanged. So I now think it's OK. No circuit changes seem to affect the decay anyway.
I've had some good results with my full bridge up to a 1000Watts, however I've made some changes, so I've a few rough edges to smooth over and then I'll be posting a fully differential full bridge for everyone.
Re. the input pair not switching, it seems like they do. Constantly switching between the potential of "the thing below". Though I guess the input pair in question is working in a linear fashion, you'd at least want them of sufficient bandwidth to be able to reproduce the switching faithfully. Right? Ft>>>>Fs
I've found it a pure shame, I can smooth the output out completly and achieve amazing results THD wise, by increasing the value of the output filter cap to say double what it should be. That would no longer be a proper filter though would it. I wonder if that actually matters with this type of feedback/modulation scheme?
Regards,
Chris
Hi Chris,
Did you try emitter resistors on the output transistors of the bridge comparator to alleviate current hogging?
Did you try emitter resistors on the output transistors of the bridge comparator to alleviate current hogging?
Hi Subwo1,
That's a good suggestion that I hadn't thought of. Would that eliminate it or help it a great deal? What if one resistor got hotter than the others? I guess you'd need one of those multi resistor jobs?
I'm not sure that would be at all necessary with something like a Mat-04 on the job however.
Regards,
Chris
EDIT:
I doubt spice will model current hogging phenomena, but it's of course of great concern for anyone who wants to build it! So I think anyone wanting to build it should keep this in mind, I'm not going to add any to my diagram, but I think adding some could make it a bit more robust, which never hurts at all. I'll leave it up to them.
That's a good suggestion that I hadn't thought of. Would that eliminate it or help it a great deal? What if one resistor got hotter than the others? I guess you'd need one of those multi resistor jobs?
I'm not sure that would be at all necessary with something like a Mat-04 on the job however.
Regards,
Chris
EDIT:
I doubt spice will model current hogging phenomena, but it's of course of great concern for anyone who wants to build it! So I think anyone wanting to build it should keep this in mind, I'm not going to add any to my diagram, but I think adding some could make it a bit more robust, which never hurts at all. I'll leave it up to them.
I can sympathize with you in omitting them in the diagram. I think they help a lot because they tend to change the output current of a transistor from being a function of base current times beta to voltage drop across resistor / value of resistor.
Now I get it - I didn't realise it was the current output transistors that you were talking about. I forgot that I solved that problem long ago, using an electrical trick, not a thermal one.
I hope this sets you in the right direction.
FWIW, before I found that solution I tried some two-crystal duals, but the thermal coupling turned out to be rather poor, because the coupling was only made through the packaging resin. Another reason why we couldn't use them was because wave soldering does not reliably produce good results with 6-pin SOT23 packages. Anyhow, the whole thing is very satisfactorily solved now.
I hope this sets you in the right direction.
FWIW, before I found that solution I tried some two-crystal duals, but the thermal coupling turned out to be rather poor, because the coupling was only made through the packaging resin. Another reason why we couldn't use them was because wave soldering does not reliably produce good results with 6-pin SOT23 packages. Anyhow, the whole thing is very satisfactorily solved now.
In a discrete design, adding Emitter resistors to the “Current Comparator” transistors connected to the Neg. Supply rail is mandatory to resolve “Current hogging” – just as with discrete Current Mirrors used in AB designs.
I would still add the Emitter resistors even if you use monolithic pairs such as the MAT04's
John
I would still add the Emitter resistors even if you use monolithic pairs such as the MAT04's
John
Hi,
Siiiigh, I decided on that about an hour ago but I'm still in the denial stage about it. Thanks for swaying me to the dark side.
Rb in my case is now 100ohm so I can afford to put in something less than that.
I'm not yet sure if something less than that would be sufficient though. I'd like to keep them as small as possible yet realize if I go too small they'll become ineffective.
Regards,
Chris
Siiiigh, I decided on that about an hour ago but I'm still in the denial stage about it. Thanks for swaying me to the dark side.
Rb in my case is now 100ohm so I can afford to put in something less than that.
I'm not yet sure if something less than that would be sufficient though. I'd like to keep them as small as possible yet realize if I go too small they'll become ineffective.
Regards,
Chris
- Status
- Not open for further replies.