I think I understand. If the source impedence to the input of the amp is 150 ohms, then using a diff. pair, each current source load resistor must be roughly 300 ohms (give or take re'). So if I lower the value of the input resistor (assuming this is the only series resistance) to about 50 ohms I can keep the constant gm degeneration resistors at 100 ohms, is this correct?
I see why preamp/amps from different manufacturers can make poor combinations based on the output of the pre stage feeding the input stage of the power amp.
I see why preamp/amps from different manufacturers can make poor combinations based on the output of the pre stage feeding the input stage of the power amp.
Hello Brannon,
Looks like you've been busy with Doug Self's book
Although I first read it years ago, I've been studying the third edition for the last few months. I recognise this circuit as a fairly complete collection of his thoughts and ideas...
I'm interested because I'm thinking about using a triple EF stage like yours. I notice that DS has written more about them in the 3rd edition - he didn't go into much detail in the earlier versions (mainly for space reasons, I think). I like the idea of triples for several reasons - not least because they are a good way to avoid Distortion 4
Couple of questions. In his analysis of slew rate (when he suggested adding C6), he ends by making the point that slew rate is significantly affected by speaker impedance - in other words, the optimal value of C6 for a 8 ohm load is different for a 4 ohm load. So, in your simulation experiments, have you found that using a triple output stage "solves" this problem? I'm curious to know this - I wasn't going to use C6 because it isn't really a complete solution for CPF and standard EF output stages.
As you're using C6, I wonder if it is worth making a separate current source for the input tail? The collector of Q6 will have an AC component on it when slewing which will modulate the tail current. I've no idea how much this is in practice, wonder if it's worth simulating?
One thing that I'm not quite clear about from the diagram - are C16 and C12 in parallel?
See you're using input bootstrapping. Having never tried before, I must admit to feeling slightly uneasy about this - I'd be interested to hear how you get on with it...
Another point - hopefully obvious. The feedback resistor R5 (I think) will dissipate some heat at high output levels - around 1/2W by my quick reckoning. It might be worth considering using 2 or more larger-value resistors in parallel - this means you can use standard 1/4W resistors. But - more importantly - film resistors tend to fail in an open-circuit fashion - meaning the amp will go open-loop. Not good for your speakers But if there are several in parallel, chances are only one will fail at any particular time - the only effect you'll notice is an increase in gain - easy to spot in a stereo setup...
I've been thinking about your last post (the noise issue). I'm not clear because I think might be confusing the degeneration resistors with the resistor which sets the current in the current source?
Someone will correct me, but slightly changing the source impedance as you suggest won't change the gm of the LTP and certainly not the gain of the amp (apart from any small potential-divider effects because of the bootstrapped value of R40(?)). Changing the degeneration resistors will obviously change the gm of the LTP, which is why you adjust the tail current to bring the gm back to its original value (I'm sure you realise that)
I might be wrong, but is John talking about the noise generated by the current source itself? This is something that I haven't studied in depth, but I guess it's not a big issue here. As long as the impedances are low (<2K), the noise contribution here should be insignificant. But, I'd welcome input here
Overall - good work Just one final thing - is there any chance of a higher-resolution schematic? Some of the component values and designations are hard to read.
Best regards,
Mark
Looks like you've been busy with Doug Self's book
Although I first read it years ago, I've been studying the third edition for the last few months. I recognise this circuit as a fairly complete collection of his thoughts and ideas...I'm interested because I'm thinking about using a triple EF stage like yours. I notice that DS has written more about them in the 3rd edition - he didn't go into much detail in the earlier versions (mainly for space reasons, I think). I like the idea of triples for several reasons - not least because they are a good way to avoid Distortion 4
Couple of questions. In his analysis of slew rate (when he suggested adding C6), he ends by making the point that slew rate is significantly affected by speaker impedance - in other words, the optimal value of C6 for a 8 ohm load is different for a 4 ohm load. So, in your simulation experiments, have you found that using a triple output stage "solves" this problem? I'm curious to know this - I wasn't going to use C6 because it isn't really a complete solution for CPF and standard EF output stages.
As you're using C6, I wonder if it is worth making a separate current source for the input tail? The collector of Q6 will have an AC component on it when slewing which will modulate the tail current. I've no idea how much this is in practice, wonder if it's worth simulating?
One thing that I'm not quite clear about from the diagram - are C16 and C12 in parallel?
See you're using input bootstrapping. Having never tried before, I must admit to feeling slightly uneasy about this - I'd be interested to hear how you get on with it...
Another point - hopefully obvious. The feedback resistor R5 (I think) will dissipate some heat at high output levels - around 1/2W by my quick reckoning. It might be worth considering using 2 or more larger-value resistors in parallel - this means you can use standard 1/4W resistors. But - more importantly - film resistors tend to fail in an open-circuit fashion - meaning the amp will go open-loop. Not good for your speakers
But if there are several in parallel, chances are only one will fail at any particular time - the only effect you'll notice is an increase in gain - easy to spot in a stereo setup...I've been thinking about your last post (the noise issue). I'm not clear because I think might be confusing the degeneration resistors with the resistor which sets the current in the current source?
Someone will correct me, but slightly changing the source impedance as you suggest won't change the gm of the LTP and certainly not the gain of the amp (apart from any small potential-divider effects because of the bootstrapped value of R40(?)). Changing the degeneration resistors will obviously change the gm of the LTP, which is why you adjust the tail current to bring the gm back to its original value (I'm sure you realise that)
I might be wrong, but is John talking about the noise generated by the current source itself? This is something that I haven't studied in depth, but I guess it's not a big issue here. As long as the impedances are low (<2K), the noise contribution here should be insignificant. But, I'd welcome input here
Overall - good work
Just one final thing - is there any chance of a higher-resolution schematic? Some of the component values and designations are hard to read.Best regards,
Mark
Self created a world of trouble for himself by linking the input tail current source and the VAS current source, anyone who can afford a few small signal Qs should separate them. (but it has been amusing following Self’s acrobatics over the years as he tries to patch it up without adding a 20 cent transistor)
The positive feedback provided by C6 can cancel some of Q9 miller C but 2 questions need to be answered; what determines (and how stable is) the gain multiplying the positive feedback and how well does the positive feedback match the Q9 Cbc (which varies with Vcb.) Trimming C6 seems to be necessary for good results, but what are you going to measure to optimize the adjustment?
I think positive feedback in amplifier circuits has only been useful where it cancels an undesired(?!), local negative feedback, explaining the popularity of “regeneration” with early tubes having large internal plate resistance
I would avoid the whole morass by cascoding and using stiffer Vref for the current sources, separate higher voltage input stage supplies are helpful to allow full output swing while dropping several V in cascodes and increased degeneration, R12 should be larger (~100 Ohms seems to work in andy_c’s sims), it improves stability and the other VAS gain increasing steps I suggest more than make up for the gm reduction
Cherry’s compensation modification involves moving the right end of C8 closer to the output, putting the driver/output stages inside the VAS/compensation feedback loop where all of the VAS gain is available to reduce their distortion. The smallest (and presumably safest) step is to connect C8 to Q30 emitter (and increase Q29,30 bias current), the pre-driver Q should be sufficiently faster than the output devices that the added phase shift within the compensation stage has no effect on global stability and the reduced loading further increases VAS gain. The low impedance driving the compensation network lets you move the 2nd pole up by reducing R14 without the increased load reducing VAS gain. Cherry insists the global stability holds even when the Ccomp is moved to the final output Q and any difficulties are local HF instabilities
http://www.diyaudio.com/forums/showthread.php?postid=220909#post220909
I design low level, low noise circuits and have used quite complicated compensation in composite op amp and discrete Q circuits, I don’t have direct experience compensating power output stages and certainly eagerly anticipate and carefully consider any contributions from experienced practicing audio amp designers – an important part of engineering is knowing what should work in theory and persisting in trying to reconcile theory and practice, sometimes in pursuing a theoretical improvement you find “there is no there there” due to erroneous application or incomplete theory, or that you forgot to solder a joint
(*.png files are smaller, I think Word can create them if you save your doc as .htm)
The positive feedback provided by C6 can cancel some of Q9 miller C but 2 questions need to be answered; what determines (and how stable is) the gain multiplying the positive feedback and how well does the positive feedback match the Q9 Cbc (which varies with Vcb.) Trimming C6 seems to be necessary for good results, but what are you going to measure to optimize the adjustment?
I think positive feedback in amplifier circuits has only been useful where it cancels an undesired(?!), local negative feedback, explaining the popularity of “regeneration” with early tubes having large internal plate resistance
I would avoid the whole morass by cascoding and using stiffer Vref for the current sources, separate higher voltage input stage supplies are helpful to allow full output swing while dropping several V in cascodes and increased degeneration, R12 should be larger (~100 Ohms seems to work in andy_c’s sims), it improves stability and the other VAS gain increasing steps I suggest more than make up for the gm reduction
Cherry’s compensation modification involves moving the right end of C8 closer to the output, putting the driver/output stages inside the VAS/compensation feedback loop where all of the VAS gain is available to reduce their distortion. The smallest (and presumably safest) step is to connect C8 to Q30 emitter (and increase Q29,30 bias current), the pre-driver Q should be sufficiently faster than the output devices that the added phase shift within the compensation stage has no effect on global stability and the reduced loading further increases VAS gain. The low impedance driving the compensation network lets you move the 2nd pole up by reducing R14 without the increased load reducing VAS gain. Cherry insists the global stability holds even when the Ccomp is moved to the final output Q and any difficulties are local HF instabilities
http://www.diyaudio.com/forums/showthread.php?postid=220909#post220909
I design low level, low noise circuits and have used quite complicated compensation in composite op amp and discrete Q circuits, I don’t have direct experience compensating power output stages and certainly eagerly anticipate and carefully consider any contributions from experienced practicing audio amp designers – an important part of engineering is knowing what should work in theory and persisting in trying to reconcile theory and practice, sometimes in pursuing a theoretical improvement you find “there is no there there” due to erroneous application or incomplete theory, or that you forgot to solder a joint
(*.png files are smaller, I think Word can create them if you save your doc as .htm)
jcx said:
Interesting because in the book he does recomend seperating the sources.
He sugggests other ways around it if you really don't want to spend the extra pennies on seperate sources. So, is this a hangover from his previous work and recomendations?
jcx said:
Can I just clairify something:
C6 is to cancel the stray capacitance of the CB junction - but this isn't Miller. I know that this will appear to be pedantic, but to fair to newbies, it worth pointing out that the so-called Miller effect is when this Cbc is increased by a factor equal to the gain of the stage.
jcx said:
I share your concerns here. To make the adjustment you simply measure the slew rate - adjust C6 for symetrical slewing. DS uses a simple RC differentiator setup for this.
In the past I've tried some interesting tricks to massively speed up these sorts of circuits. Not for audio, I should add
jcx said:
I've wondered about cascoding the I source too - I really wouldn't really mind losing a volt or two of swing...
Sorry - couldn't identify R12 from the diagram
jcx said:
Instinctively, this worries me - like you, I'd love to hear from anyone who's done this for real
Cheers,
Mark
Jcx, let's not let the issue go too easily. First, we should understand what makes a noise contribution to an amplifier stage.
In a transistor, it is the second stage shot noise which is effectively r(e)/2+ Rbb' (which is a base resistivity common to all transistor bases). Rbb' can range from 2 ohms to 400 ohms. Selection of a high Rbb' transistor will spoil things immediately. The only real reason to parallel input devices is to lower the effective Rbb' with a 'transistor array'.
The second important contribution is the input resistor, put there for RFI, as a rolloff, or just for fun. This will completely compromise the input noise, if you are not careful.
The third contribution will be the effective NOISE GAIN of the current sources used as a load.
Another contribution can be from the differential pair current source, which will add its noise to the input, UNLESS the second stage has common mode rejection.
These are points to ponder. I have not done a full analysis of the circuit in question on this thread, but it would be worth a computer simulation to see each and every of these effects. Good designing
In a transistor, it is the second stage shot noise which is effectively r(e)/2+ Rbb' (which is a base resistivity common to all transistor bases). Rbb' can range from 2 ohms to 400 ohms. Selection of a high Rbb' transistor will spoil things immediately. The only real reason to parallel input devices is to lower the effective Rbb' with a 'transistor array'.
The second important contribution is the input resistor, put there for RFI, as a rolloff, or just for fun. This will completely compromise the input noise, if you are not careful.
The third contribution will be the effective NOISE GAIN of the current sources used as a load.
Another contribution can be from the differential pair current source, which will add its noise to the input, UNLESS the second stage has common mode rejection.
These are points to ponder. I have not done a full analysis of the circuit in question on this thread, but it would be worth a computer simulation to see each and every of these effects. Good designing
hi John;
I think we agree on bipolar Q noise theory and your point about the mirror degeneration resistance is a subtle potential “gothca”, increasing the mirror degeneration resistors seems good from several perspectives.
PNP input Qs have the built in advantage of lower base resistance due to the higher conductivity of N doped Si and many part # should be acceptable, particularly in a audio amplifier where noise weighting reduces 1/f concerns. I quite agree that a really poor Q choice could compromise noise performance and that the input, feedback and emitter degeneration resistors are too high to make pursuing the lowest available rbb input Qs useful – achieving <4 nv/sqrt(Hz) total input stage noise density shouldn’t be hard though which is comfortably below the 11 nV spot noise that gives a 120 dB dynamic range (unweighted, 20 KHz).
The current mirror can be relied on to provide 30-40 dB of CMR , reducing the tail current source noise contribution at lower frequencies. AC imbalance does cause increasing noise at higher frequencies, again depending on the (noise high pass) corner frequency perceptual noise weighting reduces the impact.
mhennessy; R12 is the 22 Ohm VAS degeneration R
Enjoy your weekend, I’m spending mine away from computers
I think we agree on bipolar Q noise theory and your point about the mirror degeneration resistance is a subtle potential “gothca”, increasing the mirror degeneration resistors seems good from several perspectives.
PNP input Qs have the built in advantage of lower base resistance due to the higher conductivity of N doped Si and many part # should be acceptable, particularly in a audio amplifier where noise weighting reduces 1/f concerns. I quite agree that a really poor Q choice could compromise noise performance and that the input, feedback and emitter degeneration resistors are too high to make pursuing the lowest available rbb input Qs useful – achieving <4 nv/sqrt(Hz) total input stage noise density shouldn’t be hard though which is comfortably below the 11 nV spot noise that gives a 120 dB dynamic range (unweighted, 20 KHz).
The current mirror can be relied on to provide 30-40 dB of CMR , reducing the tail current source noise contribution at lower frequencies. AC imbalance does cause increasing noise at higher frequencies, again depending on the (noise high pass) corner frequency perceptual noise weighting reduces the impact.
mhennessy; R12 is the 22 Ohm VAS degeneration R
Enjoy your weekend, I’m spending mine away from computers
First, I would like to thank everybody that has posted a reply to my thread. I have gained insight and knowledge in discrete amplifier design from people I thought would never have the opportunity to meet much less discuss the attributes of different amplifier topologies.
It may sound like I'm brown-nosing, but the truth is I was that kid that grew up reading Audio and Stereophile, reading of the latest offerings from Mark Levinson, Parasound, Threshold, Pass Labs, etc. drooling over the latest designs by "Audio Giants" like John Curl and Nelson Pass. Driving to L.A. on the weekends to browse high end audio shops and listen to equipment I could never afford. When I joined this forum, I thought I was dreaming when I saw these gentlemen were actually members of a DIY audio forum. (JCX, I know you have done some impressive work in some realm of engineering, I can tell in the depth of your knowledge from your posts, maybe not audio?) This is what separates people that are engineers only for profit (which is not a bad thing) and people that are truly passionate about this field. Although I lack the knowledge of most prominent engineers out there, I am extremely passionate. There is something very magical in the way a good hi-fi system sounds. Although I spent many hours daydreaming about Mark Levinson equipment, there was no way I could ever afford it. My first "real hi-fi" system I bought with money acquired from graduation presents after completing high school. My first system consisted of a Pioneer
A-717 integrated amplifier (it actually used 1302/3281 outputs), one of the "lower end" Sony ES single CD player, an AIWA 3 head tape deck (no evil auto-reverse) and a set of Acoustic Research
TSW-510 three way acoustic suspension loudspeakers. Everything from this original system is still intact minus the AR's (foamrot attacked the woofers). I am still amazed at the quality of choices I made with my limited funds (some of the only good choices I made at that age).
Now I'm older, my fascination with electronics seemed genetic (my father an aerospace engineer, my grandfather an RF engineer). I went back to school and obtained my bachelors and hope to pursue a masters. I seem to be somewhat afflicted by the audio virus, as far as doing it professionally, it seems like a pretty tough gig to break into professionally. One of my cousins was a QC engineer for Harmon Int'l in Northridge (JBL) and he put me in touch with an HR lady there, but nothing materialized (economy is pretty stagnant here in SoCal) Maybe I'll appreciate audio more if I don't ever do it professionally.
Anyway, to make a long story short (I think) some time ago (age 16 or 17) I realized that the weakest link in the audio chain were the loudspeakers. They generated the most distortion ( even higher than some tube amps...lol) I started reading books by Vance Dickason and articles from pioneers like Linkwitz, Bullock and D'Appolito. I learned of a new PC based loudspeaker design called LEAP (Loudspeaker Enclosure Analysis Program)and it's companion measurement system LMS (Loudspeaker Measurement System) developed by another brilliant guy named Chris Strahm. I dropped over $2k on this system which for a 19 year old, was a lot of money. Some friends and family members thought I was insane for spending that kind of money on a test system but they soon changed their 'tune' after building them some nice loudspeakers, it was a very rewarding (addicting) experience.
That takes me to the next step. Designing your own CD or DVD player doesn't seem very practical (maybe a D/A, but not the rest of it) So the next obvious step was into preamp/power amp design. I've built several line level audio preamps and one mic preamp based on some of the excellent opamps that are available (NE5532/4, OPA627/637, AD797). I must say that I find the NE5532 to be about the best compromise of opamp designs that are tailored towards audio. In Electronics, which is sector of Physics, there is no free lunch. Designs are typically about compromise. You obtain one attribute only to give up another and I feel the NE5532, although a dated design, is very, very hard to beat. The AD797 will only beat it with low source impedences at high gains (touchy near unity gain). The OPA627 is a good all around performer with the added benefit of low offsets (which I find is of little benefit in small signal audio) and beats the NE only with high source impedence. It doesn't have the desirable common mode characteristics of the NE5532 either.
Speaking of the NE5532, that brings me to Douglas Self. Yes, another one of my heroes. His books are straight-forward, practical and insightful. Douglas Self has spent a large portion of his life de-mystifying distortions and non-linearities that have plagued audio design for years, he is truly passionate about his work. His reasons for disliking MOSFETS, exotic opamps and the like are based in solid reasoning (MOSFET=expensive, lower gm, NE5532 overall outperforms 99% of opamps in the audio spectrum ). This may sound like non-esoteric 'drivel' to some but I really believe that any alteration of an input signal is bad (ok, maybe a nice mesa guitar amp does sound good, but in my opinion a guitar amplifier is an integral part of the total instrument). I subscribe to "an audio amplifier should be a straight wire with gain" camp. Harmonics of any order are just that: 2nd, 3rd, 5th it doesn't matter, they are deviations of the input, bottom line. If the "sonic truth" is out there, we should strive to eliminate them (until the law of diminishing returns sets in) and maybe that's where Douglas Self finds himself (no pun intended) with the whole argument for a cascode in the VAS to reduce the capacitive feedthrough effects of the VAS current source. Maybe the benefit will be so small it can safely be considered
negligible!
eloquance in simplicity backed with very tight design principles, I guess that's what Doug Self means to me!
Sorry to get on a rant. JCX, don't put down Doug! He is a hero to simpletons like myself. Let's design a ultra-clean Lin 3 stage!!!
It may sound like I'm brown-nosing, but the truth is I was that kid that grew up reading Audio and Stereophile, reading of the latest offerings from Mark Levinson, Parasound, Threshold, Pass Labs, etc. drooling over the latest designs by "Audio Giants" like John Curl and Nelson Pass. Driving to L.A. on the weekends to browse high end audio shops and listen to equipment I could never afford. When I joined this forum, I thought I was dreaming when I saw these gentlemen were actually members of a DIY audio forum. (JCX, I know you have done some impressive work in some realm of engineering, I can tell in the depth of your knowledge from your posts, maybe not audio?) This is what separates people that are engineers only for profit (which is not a bad thing) and people that are truly passionate about this field. Although I lack the knowledge of most prominent engineers out there, I am extremely passionate. There is something very magical in the way a good hi-fi system sounds. Although I spent many hours daydreaming about Mark Levinson equipment, there was no way I could ever afford it. My first "real hi-fi" system I bought with money acquired from graduation presents after completing high school. My first system consisted of a Pioneer
A-717 integrated amplifier (it actually used 1302/3281 outputs), one of the "lower end" Sony ES single CD player, an AIWA 3 head tape deck (no evil auto-reverse) and a set of Acoustic Research
TSW-510 three way acoustic suspension loudspeakers. Everything from this original system is still intact minus the AR's (foamrot attacked the woofers). I am still amazed at the quality of choices I made with my limited funds (some of the only good choices I made at that age).
Now I'm older, my fascination with electronics seemed genetic (my father an aerospace engineer, my grandfather an RF engineer). I went back to school and obtained my bachelors and hope to pursue a masters. I seem to be somewhat afflicted by the audio virus, as far as doing it professionally, it seems like a pretty tough gig to break into professionally. One of my cousins was a QC engineer for Harmon Int'l in Northridge (JBL) and he put me in touch with an HR lady there, but nothing materialized (economy is pretty stagnant here in SoCal) Maybe I'll appreciate audio more if I don't ever do it professionally.
Anyway, to make a long story short (I think) some time ago (age 16 or 17) I realized that the weakest link in the audio chain were the loudspeakers. They generated the most distortion ( even higher than some tube amps...lol) I started reading books by Vance Dickason and articles from pioneers like Linkwitz, Bullock and D'Appolito. I learned of a new PC based loudspeaker design called LEAP (Loudspeaker Enclosure Analysis Program)and it's companion measurement system LMS (Loudspeaker Measurement System) developed by another brilliant guy named Chris Strahm. I dropped over $2k on this system which for a 19 year old, was a lot of money. Some friends and family members thought I was insane for spending that kind of money on a test system but they soon changed their 'tune' after building them some nice loudspeakers, it was a very rewarding (addicting) experience.
That takes me to the next step. Designing your own CD or DVD player doesn't seem very practical (maybe a D/A, but not the rest of it) So the next obvious step was into preamp/power amp design. I've built several line level audio preamps and one mic preamp based on some of the excellent opamps that are available (NE5532/4, OPA627/637, AD797). I must say that I find the NE5532 to be about the best compromise of opamp designs that are tailored towards audio. In Electronics, which is sector of Physics, there is no free lunch. Designs are typically about compromise. You obtain one attribute only to give up another and I feel the NE5532, although a dated design, is very, very hard to beat. The AD797 will only beat it with low source impedences at high gains (touchy near unity gain). The OPA627 is a good all around performer with the added benefit of low offsets (which I find is of little benefit in small signal audio) and beats the NE only with high source impedence. It doesn't have the desirable common mode characteristics of the NE5532 either.
Speaking of the NE5532, that brings me to Douglas Self. Yes, another one of my heroes. His books are straight-forward, practical and insightful. Douglas Self has spent a large portion of his life de-mystifying distortions and non-linearities that have plagued audio design for years, he is truly passionate about his work. His reasons for disliking MOSFETS, exotic opamps and the like are based in solid reasoning (MOSFET=expensive, lower gm, NE5532 overall outperforms 99% of opamps in the audio spectrum ). This may sound like non-esoteric 'drivel' to some but I really believe that any alteration of an input signal is bad (ok, maybe a nice mesa guitar amp does sound good, but in my opinion a guitar amplifier is an integral part of the total instrument). I subscribe to "an audio amplifier should be a straight wire with gain" camp. Harmonics of any order are just that: 2nd, 3rd, 5th it doesn't matter, they are deviations of the input, bottom line. If the "sonic truth" is out there, we should strive to eliminate them (until the law of diminishing returns sets in) and maybe that's where Douglas Self finds himself (no pun intended) with the whole argument for a cascode in the VAS to reduce the capacitive feedthrough effects of the VAS current source. Maybe the benefit will be so small it can safely be considered
negligible!
eloquance in simplicity backed with very tight design principles, I guess that's what Doug Self means to me!
Sorry to get on a rant. JCX, don't put down Doug! He is a hero to simpletons like myself. Let's design a ultra-clean Lin 3 stage!!!
For the record, I usually design at 0.4nV/rt Hz, 20 dB below what is considered OK here. An example of one of my older designs, the Vendetta Research phono preamp, which I stopped production on about 12 years ago, is designed to 0.4nV/rt Hz and is apparently in 'Hi Fi Plus' in a recent issue. Even the Levinson JC-1 had the same essential noise level as the Vendetta Design, and first came out 30 years ago.
A: I put a Self-like circuit in to an LT-Spice sim not long ago just to mess with it and better understand the book. One thing I noted was that no mater what way I looked at it C6 doesn't do anything unless you make it too big and whole thing becomes unstable. The clipping asymetry was very minimal in any case. I'm a simple minded sort and I've come to the notion that as long as one is building it oneself the easiest way to deal with clipping is more watts. If itn't doesn't clip who care it the clipping isn't pretty!
B: Previous comments abourt connecting the two sources seem valid. If serarate you can worry about one at a time and get each right on its own. A minor bonus is the PCB is slightly easier to lay out - but then I find PCB lay out only slightly less irritating than Rubik's Cube.
C: NE5532/4 may be old but it does the job. However, I have found a weakness that maybe worth knowing for some one. I was trying to layout a simple circuit that took three signala, A, B, and C and divided B and summed it to A and C. Three in two out plus a virtual B. I tried using the NE5532/4 (I had some sitting around doing nothing useful) as a buffer to avoid cross talk between A and C an almost expired from frustration untill I tried using a TL071 instead. The cross talk dropped to insignificant levels. This showed up in both SPICE and real life. It looked like it shouldn't be possible but appearently there is some mechanism by which a signal can move "through" a NE5532/4 in a reverse direction. I'm guessing it had something to do with the feed back loop. Keep in mind, just sticking TL071's in the SIM or the PCB cured it; the circuit was not altered in any other way.
B: Previous comments abourt connecting the two sources seem valid. If serarate you can worry about one at a time and get each right on its own. A minor bonus is the PCB is slightly easier to lay out - but then I find PCB lay out only slightly less irritating than Rubik's Cube.
C: NE5532/4 may be old but it does the job. However, I have found a weakness that maybe worth knowing for some one. I was trying to layout a simple circuit that took three signala, A, B, and C and divided B and summed it to A and C. Three in two out plus a virtual B. I tried using the NE5532/4 (I had some sitting around doing nothing useful) as a buffer to avoid cross talk between A and C an almost expired from frustration untill I tried using a TL071 instead. The cross talk dropped to insignificant levels. This showed up in both SPICE and real life. It looked like it shouldn't be possible but appearently there is some mechanism by which a signal can move "through" a NE5532/4 in a reverse direction. I'm guessing it had something to do with the feed back loop. Keep in mind, just sticking TL071's in the SIM or the PCB cured it; the circuit was not altered in any other way.
Mr. Pass,
Thank-You for the suggestion on including base resistors on the drivers and outputs. Do you have a suggestion for starting values (probably <20ohms is my guess)? I will simulate the response with the base resistors in place and see how the AC Analysis changes. If there is a positive effect (which I'm sure there will be)
they will be incorporated into the prototype as it is developed.
Also, I can assure you that in the statement "Almighty Nelson Pass" I was not being facicious. You should read a little further down the thread.
Sincerely,
Brannon
Thank-You for the suggestion on including base resistors on the drivers and outputs. Do you have a suggestion for starting values (probably <20ohms is my guess)? I will simulate the response with the base resistors in place and see how the AC Analysis changes. If there is a positive effect (which I'm sure there will be)
they will be incorporated into the prototype as it is developed.
Also, I can assure you that in the statement "Almighty Nelson Pass" I was not being facicious. You should read a little further down the thread.
Sincerely,
Brannon
Brannon,
SPICE .AC analysis can be misleading, it linearizes the circuit at the operating point and just plots the linear transfer function response, no simulation is going on, the signal levels can be anything without regard to ps voltage/current limits, device saturation, ect.
To find potential output stage oscillations I would bang on the output with a current step or add a small series voltage step source in the forward path with a very small max time step in a .tran analysis
Output stage oscillation is more likely to appear with unmodeled circuit components such as ouput C and power wiring series L (and sometimes mutual coupling), it is quite an art to estimate and include these off-schematic RF frequency parasitics that are a consequence of real world circuit construction
SPICE .AC analysis can be misleading, it linearizes the circuit at the operating point and just plots the linear transfer function response, no simulation is going on, the signal levels can be anything without regard to ps voltage/current limits, device saturation, ect.
To find potential output stage oscillations I would bang on the output with a current step or add a small series voltage step source in the forward path with a very small max time step in a .tran analysis
Output stage oscillation is more likely to appear with unmodeled circuit components such as ouput C and power wiring series L (and sometimes mutual coupling), it is quite an art to estimate and include these off-schematic RF frequency parasitics that are a consequence of real world circuit construction
Brannon,
Let me add my 2c to this, based on my studies of the sonics alone, without too much regard for distortion measurements.
1. As commented earlier, reduce input impedance to scotch noise. 47K better, 33K better again, and 22K best of the lot. For modern sources you really don't need input stage bootstrapping, not sure why you have it implemented here. Of course, any change to the input impedance will need to be reflected in the feedback loop in order to trim offset to the minimum.
2. You are using a high input stage current (6.5mA?) which gives high transconductance in the input stage, which you then clobber with high levels of emitter degeneration. This will change the distortion spectrum of the input stage and it won't be good. For a little more H3, giving a sharper sound with more impact, choose 1mA stage current with about 10R of degeneration. And don't use that current mirror; use a resistor on the collector of Q1 only.
3. A reference voltage of just 0.65V on the input stage current source, as JC has identified, will cause lots of CCS noise. Use a quiet LED in place of Q6, or other low noise source (you choose!)as the voltage reference, and considerably increase the value of this resistor. If you use a 2V LED, then for 1mA stage current this resistor increases to 1.3K. You could also use a couple of LEDs in series for 4V and increase it further still to 3K3 for even lower noise. Better still, scotch the CCS entirely and feed the diff stage with a resistor, but decouple this supply from the rest of the rail.
4. Keep the CCS voltage references completely separate, also identified here, for minimal interaction between these stages.
5. I've tried both two stage compensation (briefly) and emitter follower buffers at the VAS. I've not noticed any improvement in sonics at all over a simple VAS (more properly referred to as a transimpedance amplifier since it is current in for voltage out) driven straight from the collector of Q1. Try to keep the lag compensation capacitor as low as possible. It has much impact on sound quality.
6. Don't use emitter degeneration on Q24, your 'VAS'.
7. Reroute C6 from the base of Q6 to the base of Q3. This will give phase compensation for capacitive loads, something required with most speaker loads. And then remove R21/C11.
8. Bias generator is very good, temp compensated, elegant.
9. The output protection needs careful calibration, and it's possible the zeners might introduce noise. Bypass them??
10. Triplets for output stages are probably not necessary. A double emitter follower is fine, with the existing charge suckout network (R30/C10), but base stoppers are a very good idea, as NP remarked.
11. Output device emitter resistors are too small. 0R22 is normal, and 0R47 (with corresponding changes to VI protection) are even better as they force much more even current sharing and make the amp tolerant of relatively loose output matching.
12. I'd decouple the positive rail for the VAS and diff pair supplies. Contrary to folklore CCSs are very sensitive to noise at their supplies.
13. I'm being controversial, but input stage bandwidth limiting is a mixed blessing sonically; try it without to see if you discern a difference.
14. You need a 10R/0.1uF Zobel, also identified by others.
Hope this helps,
Cheers,
Hugh
Let me add my 2c to this, based on my studies of the sonics alone, without too much regard for distortion measurements.
1. As commented earlier, reduce input impedance to scotch noise. 47K better, 33K better again, and 22K best of the lot. For modern sources you really don't need input stage bootstrapping, not sure why you have it implemented here. Of course, any change to the input impedance will need to be reflected in the feedback loop in order to trim offset to the minimum.
2. You are using a high input stage current (6.5mA?) which gives high transconductance in the input stage, which you then clobber with high levels of emitter degeneration. This will change the distortion spectrum of the input stage and it won't be good. For a little more H3, giving a sharper sound with more impact, choose 1mA stage current with about 10R of degeneration. And don't use that current mirror; use a resistor on the collector of Q1 only.
3. A reference voltage of just 0.65V on the input stage current source, as JC has identified, will cause lots of CCS noise. Use a quiet LED in place of Q6, or other low noise source (you choose!)as the voltage reference, and considerably increase the value of this resistor. If you use a 2V LED, then for 1mA stage current this resistor increases to 1.3K. You could also use a couple of LEDs in series for 4V and increase it further still to 3K3 for even lower noise. Better still, scotch the CCS entirely and feed the diff stage with a resistor, but decouple this supply from the rest of the rail.
4. Keep the CCS voltage references completely separate, also identified here, for minimal interaction between these stages.
5. I've tried both two stage compensation (briefly) and emitter follower buffers at the VAS. I've not noticed any improvement in sonics at all over a simple VAS (more properly referred to as a transimpedance amplifier since it is current in for voltage out) driven straight from the collector of Q1. Try to keep the lag compensation capacitor as low as possible. It has much impact on sound quality.
6. Don't use emitter degeneration on Q24, your 'VAS'.
7. Reroute C6 from the base of Q6 to the base of Q3. This will give phase compensation for capacitive loads, something required with most speaker loads. And then remove R21/C11.
8. Bias generator is very good, temp compensated, elegant.
9. The output protection needs careful calibration, and it's possible the zeners might introduce noise. Bypass them??
10. Triplets for output stages are probably not necessary. A double emitter follower is fine, with the existing charge suckout network (R30/C10), but base stoppers are a very good idea, as NP remarked.
11. Output device emitter resistors are too small. 0R22 is normal, and 0R47 (with corresponding changes to VI protection) are even better as they force much more even current sharing and make the amp tolerant of relatively loose output matching.
12. I'd decouple the positive rail for the VAS and diff pair supplies. Contrary to folklore CCSs are very sensitive to noise at their supplies.
13. I'm being controversial, but input stage bandwidth limiting is a mixed blessing sonically; try it without to see if you discern a difference.
14. You need a 10R/0.1uF Zobel, also identified by others.
Hope this helps,
Cheers,
Hugh
To add to what jcx has mentioned, Self's book has some info about output stage local VHF instability in connection with the Zobel network on page 195 of the most recent edition of his book. He mentions an inductance value of 0.47 uH in series with 8 Ohms caused local VHF instability in his output stage when no Zobel network was present.
I can relate a bit of anecdotal evidence about getting parasitic oscillation (or at least instability) to show up in SPICE. I was working on a +/-90Vdc Jung-style regulator for my power amp (minus output stage). This regulator has a two-stage Darlington pass element. For a load, I was using an ideal DC current source, so AC load impedance was high. When testing the response to a step of load current, I noticed some high-frequency ringing at the step transition of the output. Checking the period showed it to be around 50-75 MHz. Since the unity loop gain frequency of the global feedback loop was a little less than 1 MHz, this ruled out a problem with overall loop instability. I cured it with a combination of Zobel network at the output and series resistor in the base of the first transistor of the Darlington pair. So sometimes SPICE really does detect these things, but it seems to take a transient analysis to do it. I now use the Middlebrook technique with AC analysis for global feedback stability analysis, and transient analysis for local stability verification.
I can relate a bit of anecdotal evidence about getting parasitic oscillation (or at least instability) to show up in SPICE. I was working on a +/-90Vdc Jung-style regulator for my power amp (minus output stage). This regulator has a two-stage Darlington pass element. For a load, I was using an ideal DC current source, so AC load impedance was high. When testing the response to a step of load current, I noticed some high-frequency ringing at the step transition of the output. Checking the period showed it to be around 50-75 MHz. Since the unity loop gain frequency of the global feedback loop was a little less than 1 MHz, this ruled out a problem with overall loop instability. I cured it with a combination of Zobel network at the output and series resistor in the base of the first transistor of the Darlington pair. So sometimes SPICE really does detect these things, but it seems to take a transient analysis to do it. I now use the Middlebrook technique with AC analysis for global feedback stability analysis, and transient analysis for local stability verification.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.