http://www.cambridgeaudio.com/assets/documents/HFWMay06840ADougSelfiviewweb.pdf
Hi Arthur, Sorry I am going mention Mr Self again. Because in the link above he mentions the value of THD sinewave testing WRT perceived audio quality and this is actually IMO a subtle change in his position from the 1990's book writing years, where the AP System 1 was king.
Intermodulation and the high order distortion components 'manufactured' by feedback in particular.
I suspect the class XD output stage is more of a gimmick than practical value, (sorry again but you have to appreciate manufacturers are always after a sales tool and with modern semiconductors a class B output is pretty good) but the reviews point to this being a fine sounding amplifier and this hints that in 2008 and beyond Mr Self may be using his System 1 (or more likely System 2) but is also using his ears more.
Fantastic!
More power to his elbow!! And possibly hope to those searching for topologies (and compensation schemes) which sound musically correct and engaging.
Hi Arthur, Sorry I am going mention Mr Self again. Because in the link above he mentions the value of THD sinewave testing WRT perceived audio quality and this is actually IMO a subtle change in his position from the 1990's book writing years, where the AP System 1 was king.
Intermodulation and the high order distortion components 'manufactured' by feedback in particular.
I suspect the class XD output stage is more of a gimmick than practical value, (sorry again but you have to appreciate manufacturers are always after a sales tool and with modern semiconductors a class B output is pretty good) but the reviews point to this being a fine sounding amplifier and this hints that in 2008 and beyond Mr Self may be using his System 1 (or more likely System 2) but is also using his ears more.
Fantastic!
More power to his elbow!! And possibly hope to those searching for topologies (and compensation schemes) which sound musically correct and engaging.
Thank you therefore.
Who knows brands and models of commercial products, where is inside this topology?
Pioneer Amp
Not as far as I know!
Funny you should post this today, I got the prototype amps out at the weekend. Now i can actually measure things (!) I might give them a run up. Problem last time was that it was a completely new design, which worked, didn't sound that good, but with no test eqipment other than a very old 'scope it was hard to see what was going on.
I kinda got side tracked on to some other projects, not to mention moving from Sydney to Melbourne.
Not as far as I know!
Funny you should post this today, I got the prototype amps out at the weekend. Now i can actually measure things (!) I might give them a run up. Problem last time was that it was a completely new design, which worked, didn't sound that good, but with no test eqipment other than a very old 'scope it was hard to see what was going on.
I kinda got side tracked on to some other projects, not to mention moving from Sydney to Melbourne.
I am still looking for such brands.
About the Italian URL
HiFi-ForumLibero • Leggi argomento - AMPLI: un 3d con una dedica precisa
there are mentioned additional approaches.
Hi Guys
I know this is an old thread but I very much admire the work of Mr. Cherry!
I think the most important aspect of NDFL is that the nested loops return to the output so that the output stage - which requires the most correction - has the most correction applied to it. How I interpret the ETI articles is that Cherry felt that in doing this even the bias level of the output stage became less critical to overall performance.
Because the nested loops refer to the output rather than the input, the notion presented in post 73(?) is incorrect. There, a corollary of cascaded low-pass filter stages was used. As an input-referred description, the cascaded filter conclusion is correct, but it is incorrect for an output-referred system. besides, it is more like parallel filters than series filters.
In 1980 I used complimentary Rush stages in a design without realising they were something already designed. The idea was to buffer the outputs of complimentary input diff-amps and provide a means to tie in a floating adjustable current source to bias the output stage, which is complimentary Darlingtons with collector outputs. The circuit worked and sounded okay although I never had a means to measure it.
Have fun
Kevin O'Connor
londonpower.com
I know this is an old thread but I very much admire the work of Mr. Cherry!
I think the most important aspect of NDFL is that the nested loops return to the output so that the output stage - which requires the most correction - has the most correction applied to it. How I interpret the ETI articles is that Cherry felt that in doing this even the bias level of the output stage became less critical to overall performance.
Because the nested loops refer to the output rather than the input, the notion presented in post 73(?) is incorrect. There, a corollary of cascaded low-pass filter stages was used. As an input-referred description, the cascaded filter conclusion is correct, but it is incorrect for an output-referred system. besides, it is more like parallel filters than series filters.
In 1980 I used complimentary Rush stages in a design without realising they were something already designed. The idea was to buffer the outputs of complimentary input diff-amps and provide a means to tie in a floating adjustable current source to bias the output stage, which is complimentary Darlingtons with collector outputs. The circuit worked and sounded okay although I never had a means to measure it.
Have fun
Kevin O'Connor
londonpower.com
check out this URL:
http://www.diyaudio.com/forums/pass-labs/181552-b1-preamp-build-thread.html
no vintage or currently available amplifiers on the market?
Hi Guys
Just a comment about Doug Self's crossover displacement method, since it was brought up here:
The purpose of XD is to move the crossover distortion that cannot be eradicated from a class-B amplifier to a point in the signal swing where it might be less objectionable. The submilliwatt level where normal crossover distortion occurs places this disturbance right where most people listen most often - low SPL levels - where XD moves the distortion up to a few-watts level where the SPL is much higher and can mask the THD.
The numbers in Doug's PA book suggest XD cuts the THD at high frequencies by about 50%. This is because large signal nonlinearity is reduced and the class-A region of the amp is made a bit larger. The latter point would suggest to most builders that simply increasing the bias to enlarge the class-A region might have a similar effect. Certainly this would push the transition to class-B to a higher power level where the signal is less likely to traverse.
This brings up a detail about power requirements. It is surprising how many watts people think they need for their amp or their system. Most listening levels happen with less than 1W in total even with inefficient speakers. Most of the elaborated designs in Doug's book are 20-50W. The class-A region in the class-B amps is typically only about 2mW for the CFP output with a single pair and about 100mW for an EF-1 with single pair. These have 15mA and 115mA idle respectively. Leach's amp idles around 100mA and has a similar class-A region, even though at +/-50V the Qs are dissipating 10W.
With respect to the Cherry amp and Self/Lin amps sounding "diabolical", I would suggest that there are other issues in the playback system if a low-THD amplifier sounds "bad". Most people have never heard truly low-THD sound and it is impossible to correctly adjust a class-B amp for lowest distortion without a distortion analyser. These amps may represent an extreme of performance you don't like. At the other end are things like the Zen amps and similar that only sound good with simple music - they mud out with complex signals, just as the 10W JLH amp does.
There is a happy place in between where wide-ish bandwidth and low THD combine for great sound.
Have fun
Kevin O'Connor
londonpower.com
Just a comment about Doug Self's crossover displacement method, since it was brought up here:
The purpose of XD is to move the crossover distortion that cannot be eradicated from a class-B amplifier to a point in the signal swing where it might be less objectionable. The submilliwatt level where normal crossover distortion occurs places this disturbance right where most people listen most often - low SPL levels - where XD moves the distortion up to a few-watts level where the SPL is much higher and can mask the THD.
The numbers in Doug's PA book suggest XD cuts the THD at high frequencies by about 50%. This is because large signal nonlinearity is reduced and the class-A region of the amp is made a bit larger. The latter point would suggest to most builders that simply increasing the bias to enlarge the class-A region might have a similar effect. Certainly this would push the transition to class-B to a higher power level where the signal is less likely to traverse.
This brings up a detail about power requirements. It is surprising how many watts people think they need for their amp or their system. Most listening levels happen with less than 1W in total even with inefficient speakers. Most of the elaborated designs in Doug's book are 20-50W. The class-A region in the class-B amps is typically only about 2mW for the CFP output with a single pair and about 100mW for an EF-1 with single pair. These have 15mA and 115mA idle respectively. Leach's amp idles around 100mA and has a similar class-A region, even though at +/-50V the Qs are dissipating 10W.
With respect to the Cherry amp and Self/Lin amps sounding "diabolical", I would suggest that there are other issues in the playback system if a low-THD amplifier sounds "bad". Most people have never heard truly low-THD sound and it is impossible to correctly adjust a class-B amp for lowest distortion without a distortion analyser. These amps may represent an extreme of performance you don't like. At the other end are things like the Zen amps and similar that only sound good with simple music - they mud out with complex signals, just as the 10W JLH amp does.
There is a happy place in between where wide-ish bandwidth and low THD combine for great sound.
Have fun
Kevin O'Connor
londonpower.com
The easy way to increase the ClassA output current is to over-bias the ClassAB output stage. But as shown by Self and others, that increases the crossover distortion cf. the optimal bias ClassAB output stage.
There is an alternative way to achieve higher ClassA current from a ClassAB output stage.
Choose the optimal bias voltage for the output stage.
Use this bias voltage and increase the output pairs from 1 to 2. This quadruples the ClassA output current available.
Using a 3pair output stage increases the ClassA output current by a factor of 9.
These increased ClassA currents are available from the same optimally biased transistors for least crossover distortion.
A further alternative that can be used alone, or in conjunction with the added output pairs, is to lower the emitter resistor value and adjust the Vbias to suit this lower emitter resistor.
eg. swapping out Re=0r47 to be replaced with 0r22 increases the ClassA current by approximately 95%.
Reducing to Re to 0r1 increases the ClassA current by approximately 300% ref 0r47.
Combining these two methods from a 1pr Re=0r47 optimally biased stage gives ~49mW of ClassA, whereas a 4pr Re=0r1 optimally biased stage gives ~9W of ClassA.
There is an alternative way to achieve higher ClassA current from a ClassAB output stage.
Choose the optimal bias voltage for the output stage.
Use this bias voltage and increase the output pairs from 1 to 2. This quadruples the ClassA output current available.
Using a 3pair output stage increases the ClassA output current by a factor of 9.
These increased ClassA currents are available from the same optimally biased transistors for least crossover distortion.
A further alternative that can be used alone, or in conjunction with the added output pairs, is to lower the emitter resistor value and adjust the Vbias to suit this lower emitter resistor.
eg. swapping out Re=0r47 to be replaced with 0r22 increases the ClassA current by approximately 95%.
Reducing to Re to 0r1 increases the ClassA current by approximately 300% ref 0r47.
Combining these two methods from a 1pr Re=0r47 optimally biased stage gives ~49mW of ClassA, whereas a 4pr Re=0r1 optimally biased stage gives ~9W of ClassA.
While we are mentioning the work of JLH a relevant design that sought to address Kevin's point was the Class AB unit that he published a year or so after the original class A 10 watt amp'. It was in the 15-20 watt range and was designed so that it ran about200 milliamps through the output stage. The subsequent critics in the Letters Section of WW said that this worsened the THD (which it did) but JLH's reply was that they had not understood the article and that the whole purpose was to get class quality for the first few watts and still have some reserve power. The stated goals were to get class A in the region where most listening was done (1-2 watts @ 8 ohms) and yet not have the penalty of the extra heat, expense etc of the original 10 watter.
Friend built it in 1976 and was very happy.
Cheers, Jonathan
Friend built it in 1976 and was very happy.
Cheers, Jonathan
Last edited:
Hi Guys
Jonathan, that is a very good point about optimising for the first few watts.
I've measured the power used in my system and it is pretty low. I have fairly efficient speakers and a smallish listening space, and don't listen to music too loud - sometimes you have to crank it but I want to be able to listen to music forty years from now, too!
I would like to see more optimisation of low-level performance. Class-A is certainly an easy route to that end. In my case, 10W would be "super power" but easily provides all the headroom I'd ever need.
It is always a little disconcerting when you see the THD vs power graphs and the rising noise/THD at low power. Of course, noise is more or less constant and as a ratio to the signal looks larger as signal level drops, so just a ratio thing that looks bad on a graph.
It was interesting to see a graph that Self put in his book that showed the EF to have lower high frequency distortion than a CFP at low levels. (figs. 6-44,45,46)
It seems like noise is the ultimate limiter to low-level performance.
Have fun
Kevin O'Connor
londonpower.com
Jonathan, that is a very good point about optimising for the first few watts.
I've measured the power used in my system and it is pretty low. I have fairly efficient speakers and a smallish listening space, and don't listen to music too loud - sometimes you have to crank it but I want to be able to listen to music forty years from now, too!
I would like to see more optimisation of low-level performance. Class-A is certainly an easy route to that end. In my case, 10W would be "super power" but easily provides all the headroom I'd ever need.
It is always a little disconcerting when you see the THD vs power graphs and the rising noise/THD at low power. Of course, noise is more or less constant and as a ratio to the signal looks larger as signal level drops, so just a ratio thing that looks bad on a graph.
It was interesting to see a graph that Self put in his book that showed the EF to have lower high frequency distortion than a CFP at low levels. (figs. 6-44,45,46)
It seems like noise is the ultimate limiter to low-level performance.
Have fun
Kevin O'Connor
londonpower.com
I'm afraid don't have a copy of Self's book so can't really comment on your references. I have gained a lot from him in some areas but I was never happy with (A) his blanket rejection of listening tests and (B) a lack of square wave results. I think the obsession with low THD is commendable in one respect but that approach is possibly missing out on other criteria that may well be useful.
That's why I like JLH's work (I have a class A amp) because he knew his fundamental electronics (better than most), measured things and yet was open to the mystery of hearing and open to new mechanisms.
That's why I like JLH's work (I have a class A amp) because he knew his fundamental electronics (better than most), measured things and yet was open to the mystery of hearing and open to new mechanisms.
Hi Guys
Self has earned his living satisfying the needs of the companies he's worked for. Verifiable performance is a safe and easy mark to works towards. Of course, if his client says it does not "sound" right and to change it up, no doubt Doug can erode the circuit performance to what will please the boss. I don't think he completely dismisses qualitative judgement, but rather prefers to emphasize quantitative judgement and train his ear to hear what 'better' sounds like.
Everyone ultimately listens to their equipment. If you are not inspired to play more music with it then something is wrong.
Decades ago, the National Research Council of Canada conducted now famous blind testing of audio equipment to determine a correlation between measured performance and listener response. For all equipment types, including speakers and electronics, a flat frequency amplitude response was consistently preferred, as was lowest broadband distortion and IM.
If you hear accurate sound from equipment with flat response it can sound lifeless compared to everything that colours the sound. But that colour is distortion in excess. It may be argued that 'some' profile of distortion is pleasing with some music and at some times, and this fits in with the nostalgaic quest we all end up on searching for the same experience of hearing exciting music for the first time.
Have fun
Kevin O'Connor
londonpower.com
Self has earned his living satisfying the needs of the companies he's worked for. Verifiable performance is a safe and easy mark to works towards. Of course, if his client says it does not "sound" right and to change it up, no doubt Doug can erode the circuit performance to what will please the boss. I don't think he completely dismisses qualitative judgement, but rather prefers to emphasize quantitative judgement and train his ear to hear what 'better' sounds like.
Everyone ultimately listens to their equipment. If you are not inspired to play more music with it then something is wrong.
Decades ago, the National Research Council of Canada conducted now famous blind testing of audio equipment to determine a correlation between measured performance and listener response. For all equipment types, including speakers and electronics, a flat frequency amplitude response was consistently preferred, as was lowest broadband distortion and IM.
If you hear accurate sound from equipment with flat response it can sound lifeless compared to everything that colours the sound. But that colour is distortion in excess. It may be argued that 'some' profile of distortion is pleasing with some music and at some times, and this fits in with the nostalgaic quest we all end up on searching for the same experience of hearing exciting music for the first time.
Have fun
Kevin O'Connor
londonpower.com
I have edition2 with copies of the latest modifications/additions added to keep me sort of up to date.
Fig6.xx only go to 6.22, but I find fig5.44 referring to CFP in the previous chapter.
This appears to be the fig you are referring to, but I don't have 5.45 nor 5.46 (must have missed copying those)
The comment alongside figure 5.44 is
You have misquoted what Self says and this is probably because you have not understood the Self definition of "his ClassB".
Optimal ClassAB bias is where the crossover distortion is minimised by adjusting the output bias voltage measured across the emitter resistors. For a CFP those emitter resistors are referred to the drivers emitter since the output emitters are tied to the supply rails. Optimal ClassAB is what Self has defined as "his ClassB". It is important you recognise his non standard use of the definition.
What the fig5.44 comment is actually saying (my interpretation):- if one over biases or under biases the CFP output stage, then the crossover distortion increases. This situation is exactly the same with an EF stage. Optimal ClassAB bias minimises the crossover distortion. Move the bias to either side of the optimal and the crossover distortion is worse. Cordell says exactly the same.
Last edited:
The standard definition of classes would be better described as being traditionnal. And it is perferctly ambiguous. Self tried to correct it.
For many people, class B means that each half has a conduction of 180°, no less, no more. This can't be really achieved and maintained for long using common amp schemes.
The non-standard of Self's way of looking at things has the huge advantage to be very clear.