yea, for a gain of 20, 8.5MHz/23=370KHz. DUH!
Still, a decade higher towards unity gain is 3.7MHz, and he used to use a GBP of 10.
Still, a decade higher towards unity gain is 3.7MHz, and he used to use a GBP of 10.
Re: Post #251
Re: Post #251
http://www.diyaudio.com/forums/showthread.php?postid=641377#post641377
Jens,
Can you explain why the FB is taken prior to the output inductor? I am not questioning your knowledge, I am seeking to learn the reasoning.
Regards,
John L. Males
Willowdale, Ontario
Canada
22 November 2006 11:56
Re: Post #251
http://www.diyaudio.com/forums/showthread.php?postid=641377#post641377
JensRasmussen said:
Jens,
Can you explain why the FB is taken prior to the output inductor? I am not questioning your knowledge, I am seeking to learn the reasoning.
Regards,
John L. Males
Willowdale, Ontario
Canada
22 November 2006 11:56
Hi,
the feedback point takes output signal straight into the inverting input of the amplifier.
Any rubbish on that feedback signal will be amplified and fed to the speakers, or worse screw up the stability of the amplifer's stages.
The Thiel network is a filter system that loads up the output stage and helps keep it stable but just as importantly it keeps speaker lead contamination out of the inverting input.
For the filter to work, the feedback point must be on the speaker output but before the filter.
There is an exception or two.
Dr. Cherry proposes that the Zobel part of the Thiel network be resistor to capacitor to speaker return. That then permits the feed back tapping to be placed between the resistor and capacitor and turns the single pole filter into a 2pole filter.
The other exception is the feedback taken one stage earlier and this effectively places a buffer between the speaker contamination and the inverting input. One or two have used this. Leach uses a modified version of this, but two stages earlier and frequency selective.
the feedback point takes output signal straight into the inverting input of the amplifier.
Any rubbish on that feedback signal will be amplified and fed to the speakers, or worse screw up the stability of the amplifer's stages.
The Thiel network is a filter system that loads up the output stage and helps keep it stable but just as importantly it keeps speaker lead contamination out of the inverting input.
For the filter to work, the feedback point must be on the speaker output but before the filter.
There is an exception or two.
Dr. Cherry proposes that the Zobel part of the Thiel network be resistor to capacitor to speaker return. That then permits the feed back tapping to be placed between the resistor and capacitor and turns the single pole filter into a 2pole filter.
The other exception is the feedback taken one stage earlier and this effectively places a buffer between the speaker contamination and the inverting input. One or two have used this. Leach uses a modified version of this, but two stages earlier and frequency selective.
Hi Andrew,
Thanks for your reply. As always you are most informative.
This makes sense. I had no idea if the reason may have been to take after the coil to feedback some of the speaker back EMF as one way to deal with the speaker back EMF. As you likely know, many designs appear to take the feedback after the coil. When I saw Jen's comment and many agreing I knew I needed to increase my understanding.
I assume the ideal point to take the feedback from the speaker output end is where the the speaker output RC connects to the trace for the output devices?
Regards,
John L. Males
Willowdale, Ontario
Canada
22 November 2006 14:23
Thanks for your reply. As always you are most informative.
This makes sense. I had no idea if the reason may have been to take after the coil to feedback some of the speaker back EMF as one way to deal with the speaker back EMF. As you likely know, many designs appear to take the feedback after the coil. When I saw Jen's comment and many agreing I knew I needed to increase my understanding.
I assume the ideal point to take the feedback from the speaker output end is where the the speaker output RC connects to the trace for the output devices?
Regards,
John L. Males
Willowdale, Ontario
Canada
22 November 2006 14:23
Hi Key,
don't take it from the trace BETWEEN the output devices.
find a convenient point along the trace from that point between the output devices and to where the Thiel network starts. The closer the where the Thiel network starts the better.
If the Thiel network is fitted to the speaker output terminals then the feedback could come from there as well. It then takes account of losses in the output cabling.
Now if we went to 4 core cabling and fitted the Thiel at the speaker, then we could take the feedback from the speaker end of the speaker cables. But I suspect that would be quite a challenge to achieve stability.
don't take it from the trace BETWEEN the output devices.
find a convenient point along the trace from that point between the output devices and to where the Thiel network starts. The closer the where the Thiel network starts the better.
If the Thiel network is fitted to the speaker output terminals then the feedback could come from there as well. It then takes account of losses in the output cabling.
Now if we went to 4 core cabling and fitted the Thiel at the speaker, then we could take the feedback from the speaker end of the speaker cables. But I suspect that would be quite a challenge to achieve stability.
Jens,
Can you explain how you calculated the Break Frequency and values for the Feedback network for your version of the Leach Amplifier? So far I am struggling in trying to understand the Leach Feedback Network beyond how to calculate the gain.
Regards,
John L. Males
Willowdale, Ontario
Canada
28 November 2006 02:32
Can you explain how you calculated the Break Frequency and values for the Feedback network for your version of the Leach Amplifier? So far I am struggling in trying to understand the Leach Feedback Network beyond how to calculate the gain.
Regards,
John L. Males
Willowdale, Ontario
Canada
28 November 2006 02:32
Hi Andrew,
Before posting the question I believe I had reviewed all of the papers from Professor Leach's web pages.
I had closely reviewed the Low TIM and Super Leach amplifier pages. I found most of the values the same as the Smaller Leach, but some of the key values are different I have determined. One value difference I would say is not important, but a couple of the other Smaller Leach values are different wherein both versions by Professor Leach are exactly the same, but not the same as the Smaller Leach.
I found a web page and and one of the "Selected Publications" on the various pages/links of Professor Leach's site concerning the Feedforward Feedback Network Compensation used in the Leach Amplifiers. One simply mirrors the schematic part numbers and references the Low TIM amplifier for the respective parts values. The other document is filled with all sorts of theory and formulas as well as a SPICE model. In the latter case neither the content details nor the related SPICE model have the same values as either of the Leach amplifiers parts list on Professor Leach's site nor as used in the Smaller Leach design.
That said, as far as my prior research done before posting I would be interested in what Leach paper you would be referring to that shows the Smaller LEach using the same values as Professor Leach?
I am in no way questing Jen's design. Jen is very thorough in Jen's thought and execution of a design. From my careful comparisons between different parts of information available it appears to me there are a few values that are different with the Smaller Leach (bypass capacitors differences aside) that I thought Jens may be able to explain, assuming I am still correct in the specific value differences of significance. Again explain as in how calculated, not explain to justify rational for any design values such as the "Break Frequency" or Time Constant chosen.
I am simply trying to figure out how the capacitance values of Professor Leach's Feedforward Compensation Feedback Network are calculated. The gain calculation is easy and straight forward. The capacitance related calculations seem to be a nest of V, T, and k related values (of course there are R and C values) I cannot seem to find an initial starting point, nor how to translate whichever value I need to determine or set the "Break Frequency". In a couple different places in the Leach materials I have found there is reference to the "Break Frequency" being 100 KHz and in another document 150 KHz. Neither case examples how these "Break Frequencies" are determined, nor what, if any, frequency overlap there is between the HF and LF elements of the Leach Feedforward Feedback Network Compensation.
I am also trying to determine the relationship of the Leach Feedforward Feedback Network Compensation resistor values on the "Break Frequency" to determine what design differences I would run through the calculations to better understand the Leach Feedforward Feedback Network Compensation.
Regards,
John L. Males
Willowdale, Ontario
Canada
28 November 2006 09:40
Before posting the question I believe I had reviewed all of the papers from Professor Leach's web pages.
I had closely reviewed the Low TIM and Super Leach amplifier pages. I found most of the values the same as the Smaller Leach, but some of the key values are different I have determined. One value difference I would say is not important, but a couple of the other Smaller Leach values are different wherein both versions by Professor Leach are exactly the same, but not the same as the Smaller Leach.
I found a web page and and one of the "Selected Publications" on the various pages/links of Professor Leach's site concerning the Feedforward Feedback Network Compensation used in the Leach Amplifiers. One simply mirrors the schematic part numbers and references the Low TIM amplifier for the respective parts values. The other document is filled with all sorts of theory and formulas as well as a SPICE model. In the latter case neither the content details nor the related SPICE model have the same values as either of the Leach amplifiers parts list on Professor Leach's site nor as used in the Smaller Leach design.
That said, as far as my prior research done before posting I would be interested in what Leach paper you would be referring to that shows the Smaller LEach using the same values as Professor Leach?
I am in no way questing Jen's design. Jen is very thorough in Jen's thought and execution of a design. From my careful comparisons between different parts of information available it appears to me there are a few values that are different with the Smaller Leach (bypass capacitors differences aside) that I thought Jens may be able to explain, assuming I am still correct in the specific value differences of significance. Again explain as in how calculated, not explain to justify rational for any design values such as the "Break Frequency" or Time Constant chosen.
I am simply trying to figure out how the capacitance values of Professor Leach's Feedforward Compensation Feedback Network are calculated. The gain calculation is easy and straight forward. The capacitance related calculations seem to be a nest of V, T, and k related values (of course there are R and C values) I cannot seem to find an initial starting point, nor how to translate whichever value I need to determine or set the "Break Frequency". In a couple different places in the Leach materials I have found there is reference to the "Break Frequency" being 100 KHz and in another document 150 KHz. Neither case examples how these "Break Frequencies" are determined, nor what, if any, frequency overlap there is between the HF and LF elements of the Leach Feedforward Feedback Network Compensation.
I am also trying to determine the relationship of the Leach Feedforward Feedback Network Compensation resistor values on the "Break Frequency" to determine what design differences I would run through the calculations to better understand the Leach Feedforward Feedback Network Compensation.
Regards,
John L. Males
Willowdale, Ontario
Canada
28 November 2006 09:40
AndrewT said:
100% correct for parts noted from Leach Amp.
The same respective part values from the Smaller Leach Amp are?
What are the other different values between the two feedback networks?
Regards,
John L. Males
Willowdale, Ontario
Canada
28 November 2006 15:12
The 3 dB corner of 22K and 47 pf is 150K, as is 5.5K (11K//11K) and 180 pf. f=1/(2*PI*R*C). What you have is a high pass filter in R20/C9 and a low pass formed by R17//R18 and C8.
I remember somewhere along the line Jens commented that he reduced C9 to 39 pf for better phase margin.
Bob,
Thank you so much. I was very aware of the quote you took from the noted link of Professor Leach. I was assuming it was high and low pass filters, but I was not able to recreate the noted frequencies mostly because a paper on the Feedforward Feedback Network Compensation paper of Professor Leach was confusing me more than helping me. I also did not realise one had to calculate the LP with the resistors as a parallel resistance. The paper that was confusing me trying to figure out due to my lack of EE knowledge was:
W. M. Leach, Jr., Feedforward Compensation of the Amplifier Output Stage for Improved Stability with Capacitave Loads," IEEE Transactions on Consumer Electronics, Vol. 34, No. 2, pp. 334-338, May 1988.
http://users.ece.gatech.edu/~mleach/papers/Feedforward.pdf
(I do not know how to make the text reference be a hyperlink on the forum)
I was hoping, which you have confirmed, the -3db point is supposed to be the same for the HP and LP points. When I read what you quoted from the Professor Leach's page I was not so certain the HP and LP -3db points were same. I was not sure if there was a bandpass frequency region being implied in way things were worded form the web page you quoted from.
I had tried searching the thread in hopes of finding why Jens choose 39pF vs 47pF in the HP element of the Feedforward Feedback Network Compensation. I could not find in searching, but perhaps I was not using the right search keyword on the thread.
Based on the formula the -3db points are:
HP 153,921.6 Hz (47pF)
HP 185,495.3 Hz (39pF)
LP 160,762.6 Hz (180 pF)
The choice of 39pF by Jens you believe was for improved phase margin seems to result in about bit more than a 25KHz bandpass region. Correct? Might this improved phase margin be related to time constant difference of the HP? I assume the 25Khz bandpass in the 170Khz region is not significant from a sonic perspective?
Do you know why 470uF (470ms) was chosen over the 220uF (242 ms) Professor Leach uses for the RC network to signal ground after the HP/LP Feedforward Feedback Network Compensation? My limited understanding of this RC to signal ground should have a time constant of no more than 100ms. It appears the smaller Leach design doubled the time constant of Professor Leach's design. I have seen designs where it was suggested reducing the RC capacitance from 200uF to 100uF which also was using a 1K0 resistance.
Bob, thanks so much for your answer and citing one of the references I had read trying to figure out the answer to my question.
Regards,
John L. Males
Willowdale, Ontario
Canada
29 November 2006 00:21
Thank you so much. I was very aware of the quote you took from the noted link of Professor Leach. I was assuming it was high and low pass filters, but I was not able to recreate the noted frequencies mostly because a paper on the Feedforward Feedback Network Compensation paper of Professor Leach was confusing me more than helping me. I also did not realise one had to calculate the LP with the resistors as a parallel resistance. The paper that was confusing me trying to figure out due to my lack of EE knowledge was:
W. M. Leach, Jr., Feedforward Compensation of the Amplifier Output Stage for Improved Stability with Capacitave Loads," IEEE Transactions on Consumer Electronics, Vol. 34, No. 2, pp. 334-338, May 1988.
http://users.ece.gatech.edu/~mleach/papers/Feedforward.pdf
(I do not know how to make the text reference be a hyperlink on the forum)
I was hoping, which you have confirmed, the -3db point is supposed to be the same for the HP and LP points. When I read what you quoted from the Professor Leach's page I was not so certain the HP and LP -3db points were same. I was not sure if there was a bandpass frequency region being implied in way things were worded form the web page you quoted from.
I had tried searching the thread in hopes of finding why Jens choose 39pF vs 47pF in the HP element of the Feedforward Feedback Network Compensation. I could not find in searching, but perhaps I was not using the right search keyword on the thread.
Based on the formula the -3db points are:
HP 153,921.6 Hz (47pF)
HP 185,495.3 Hz (39pF)
LP 160,762.6 Hz (180 pF)
The choice of 39pF by Jens you believe was for improved phase margin seems to result in about bit more than a 25KHz bandpass region. Correct? Might this improved phase margin be related to time constant difference of the HP? I assume the 25Khz bandpass in the 170Khz region is not significant from a sonic perspective?
Do you know why 470uF (470ms) was chosen over the 220uF (242 ms) Professor Leach uses for the RC network to signal ground after the HP/LP Feedforward Feedback Network Compensation? My limited understanding of this RC to signal ground should have a time constant of no more than 100ms. It appears the smaller Leach design doubled the time constant of Professor Leach's design. I have seen designs where it was suggested reducing the RC capacitance from 200uF to 100uF which also was using a 1K0 resistance.
Bob, thanks so much for your answer and citing one of the references I had read trying to figure out the answer to my question.
Regards,
John L. Males
Willowdale, Ontario
Canada
29 November 2006 00:21
I'm not a great fan of this forum's search engine. I often have better luck looking in the threads a Google search turns up. Perhaps a search on 39pf or "39 pf" will work here.
Edit due to accidental posting:
I once tried to wade through the Leach paper you cited, but don't have the background or enough motivation to really understand it.
I think if you look at it again you'll see a bit of a spread rather than a bandpass. Of course with first order filters you've got quite a bit of overlap. I haven't crunched the numbers, but I suspect that if you did, you would find that in Jens' version the sum of the feedback loops is flatter. With first order filters you get a bit of a hump in total response at XO frequency, spreading the frequencies reduces or eliminates the hump.
Of course the real question is how do you determine where you need to put the changeover? Professor Leach probably explains in the cited paper, but it's over my head.
As for the big electrolytic in the feedback loop, a larger capacitor gives a lower cutoff frequency. Too small is a problem, but bigger generally isn't and it ensures that you don't start rolling off in the audio band.
Edit due to accidental posting:
I once tried to wade through the Leach paper you cited, but don't have the background or enough motivation to really understand it.
I think if you look at it again you'll see a bit of a spread rather than a bandpass. Of course with first order filters you've got quite a bit of overlap. I haven't crunched the numbers, but I suspect that if you did, you would find that in Jens' version the sum of the feedback loops is flatter. With first order filters you get a bit of a hump in total response at XO frequency, spreading the frequencies reduces or eliminates the hump.
Of course the real question is how do you determine where you need to put the changeover? Professor Leach probably explains in the cited paper, but it's over my head.
As for the big electrolytic in the feedback loop, a larger capacitor gives a lower cutoff frequency. Too small is a problem, but bigger generally isn't and it ensures that you don't start rolling off in the audio band.
BobEllis said:
Hi Bob,
I have heard that comment about the diyAudio search engine before. I would normally read the whole thread, but at moment my time is very limited. I know the thread is worth reading for many reasons. I will when I have time.
I have patience and will try to read and at least understand as much as I need to for the reason(s) I read such papers. The one I cited and you seemed to have had a challenge understanding as I have.
Opps, sorry, I am bit tired here, you are correct it is a 25KHz spread, not overlap.
Oh thanks Bob, it has been a couple years since I have spent alot of time reading and understanding filters. You are right on money that first order filters with same -3db HP/LP points do in fact have a hump in the region of the crossover frequency. So shifting a -3db point can help reduce the hump.
Professor Leach calls what you term "changeover frequency" as "Break Frequency". If that cited paper gives any idea how to choose the "Break Frequency" it is certainly not clear. Personally I think it is determined by real world experience and possibility refined based on the specific amplifier design/layout.
Your comment about the electrolytic is interesting. I think I will make an effort to learn more about this. I know in one case a design had "experts" make a case to change from 200uF to 100uF. If I am deducing correctly 100uF translates to 10Hz, 200uF translates to 5Hz, and 470uF translate to 2.1Hz.
Regards,
John L. Males
Willowdale, Ontario
Canada
29 November 2006 02:14
Hi,
I don't know how Leach arrived at the exact values he chose for the swap between global feedback and the 3stage feedback.
I do know that he chose the swap method to eliminate the possibility of high frequency phase reversal becoming oscillation in the GFB loop. He presumably chose the frequency to suit his devices and PCB layout and hard-wiring interconnections to the output devices. I do not know if he took sound quality into account.
Re: the DC blocking cap in the lower leg of the NFB loop.
A lowish value that gives a turnover point (-3db) close to the audio band is very noticeable as loss of bass signal in the final output.
There is some evidence that distortion in the signal is also very significantly increased if the cap begins to generate significant voltage from the AC signal passing through it.
Both these effects can be minimised by ensuring that the turnover frequency is well below the pass band required from the amplifier.
In addition the input filter should be at least half an octave above the NFB filter.
I have used the NFB filter set at between 120 to 150mS and that seems to give good bass and adequate treble but I often parallel the electrolytic with a PES or PP across the pins (about 1uF) to help the high frequency pass easily.
Many commercial amps are quoting -1db @ 4Hz. This implies an input filter set at 2Hz. If the other filters are set in the ratio 1:1.4:2 then the NFB filter will be near 112mS.
I don't know how Leach arrived at the exact values he chose for the swap between global feedback and the 3stage feedback.
I do know that he chose the swap method to eliminate the possibility of high frequency phase reversal becoming oscillation in the GFB loop. He presumably chose the frequency to suit his devices and PCB layout and hard-wiring interconnections to the output devices. I do not know if he took sound quality into account.
Re: the DC blocking cap in the lower leg of the NFB loop.
A lowish value that gives a turnover point (-3db) close to the audio band is very noticeable as loss of bass signal in the final output.
There is some evidence that distortion in the signal is also very significantly increased if the cap begins to generate significant voltage from the AC signal passing through it.
Both these effects can be minimised by ensuring that the turnover frequency is well below the pass band required from the amplifier.
In addition the input filter should be at least half an octave above the NFB filter.
I have used the NFB filter set at between 120 to 150mS and that seems to give good bass and adequate treble but I often parallel the electrolytic with a PES or PP across the pins (about 1uF) to help the high frequency pass easily.
Many commercial amps are quoting -1db @ 4Hz. This implies an input filter set at 2Hz. If the other filters are set in the ratio 1:1.4:2 then the NFB filter will be near 112mS.
Jens,
are you here?
I have just seen your posting and pic showing the diodes mounted on top of an output device.
http://www.diyaudio.com/forums/showthread.php?postid=50284#post50284
How successful was this?
are you here?
I have just seen your posting and pic showing the diodes mounted on top of an output device.
http://www.diyaudio.com/forums/showthread.php?postid=50284#post50284
How successful was this?
My guess is it would be a bit overcompensated unless fewer diodes are used.
I did something similar when I reduced the bias of my A75s. I mounted the Vgs multiplier on top of one of the output devices. It now takes several minutes to come up to usable bias from a cold start.* When the Vgs multiplier was not in contact with the outputs it only took a few seconds. Someday I will open it up again and mount the Vgs multiplier on the output heat sink to provide a bit looser coupling.
* The warm up process is an interesting demonstration of what crossover distortion sounds like. I now notice it at lower levels than I did before.
I did something similar when I reduced the bias of my A75s. I mounted the Vgs multiplier on top of one of the output devices. It now takes several minutes to come up to usable bias from a cold start.* When the Vgs multiplier was not in contact with the outputs it only took a few seconds. Someday I will open it up again and mount the Vgs multiplier on the output heat sink to provide a bit looser coupling.
* The warm up process is an interesting demonstration of what crossover distortion sounds like. I now notice it at lower levels than I did before.
AndrewT said:
There seems to be little information in this regard based on my search attempts on the internet. I have seen only one person recently on diyAudio take about applying Leach Feedforward Feedback Network Compensation in other amps. With Bob's information it seems the one other example of the network values has chosen a slightly lower HF point and a even higher LP point resulting in about a 80KHz spread between the two points.
I think, as Bob pointed out, this is perhaps why Jens choose a different HP point to address the hump effect a first order filter has when both HP and LP -3db points are same or close to same. This may be the motivation one other person choose an 80KHz spread.
Do I use the same 1/(2*p*iR*C) formula to calculate the -3db point for the NFB lower leg RC?
Is this why Jens has a 470Uf||1uF input capacitor and the same on the NFB lower leg? I typically see the I/P coupling in the 1u0 - 2u2 and 22k0-33k0 range and often using a 100u0 and 1k0 lower NFB lower leg. This suggests many designs may not follow this guideline? I would say many amps I have seen would have about a 2-4Hz I/P coupling -3db and about a 15Hz NFB lower leg -3db. I appears Jens has chosen both to be the same for the Smaller Leach, assuming I understand the math correctly.
It appears Jens has chosen a 471ms for the NFB filter.
I have noticed how Jens has used a 1u0 film cap in parallel with an electrolytic capacitor where in many designs a 0u1 is used instead and none with the NFB lower leg capacitor. Many amps I have seen generally have about five 0u1 bypass capacitors in usual places, whereas Jens has used more expensive and larger case PE or PP 1u0 capacitors. Is the use of the 0u1 more a case of significant cost/space ratio vs lessor sonic improvement using the 1u0 in these instances (input coupling capacitor aside)? I am not suggesting the 1u0 film bypass capacitor does not offer a sonic improvment over the 0u1, just not as much vs the extra cost and PCB space.
Maybe I should use a 6.8 UltraCap Polypropylene in parallel with the lower leg NFB electrolytic.
Is this ratio order Input:NFB Lower Leg:NFB?
Regards,
John L. Males
Willowdale, Ontario
Canada
29 November 2006 10:01
BobEllis said:
Bob,
I have read various pros/cons for the Vgs or Vbe to be mounted on top of one of the output devices or on the heatsink. The sense from what I have read suggests the Vgs/Vbe should be on the heatsink. I think the motivation for placing the Vgs/Vbe on an output device was to elimiate the thermal resistance of the output device(s) to the heat sink. While well intended, I suspect the transient nature of music can cause a more sensitive Vgs/Vbe fluxing when mounted on an output device that could result in inferior sonic byproducts in the output as well as NFB signals.
What might be really more appropriate ontop of the output device is a thermal cutoff sensing device to shut down the amp when the Tc rises above the temperature limit chosen by the designer. Of course such a thermal sensing device can be mounted on the heatsink as well, but one has to offset the tripping temperature to account for the thermal lag/difference of the output device vs heatsink/thermal resistance interface of output device to heatsink.
Regards,
John L. Males
Willowdale, Ontario
Canada
29 November 2006 10:24
- Status
- Not open for further replies.