Of course not Bob. I had difficulty to consider class D as "a sampled-data system". Despite it use a switching frequency, or, more exactly square waves pulses, the modulation of them by the analog signal is linear and continuously variable, from 0 to 100%. No datas, no samples, on my point of view. Just the comparator transform a voltage to a time in an analog (proportional) way.
Not so different from a microphone, which transform an air pressure to a voltage.
Not far from a FM, or the AM modulation of a tape recorder. Am-i wrong ?
Changes of scales, changes of support for information, but analog informations.
Very different from a power DAC.
Last edited:
I always thought that the day will come when it will be common place that it will be all digital except the I/F to the speakers, data into the PA and out comes analog at the very end of the chain, analog pre-amps/PA's will be a thing of the past, in with the TT collection, my nostalgia room, my Pink Floyd wall covering.
Slight over sight, live music will still be analog however, why digitize the real deal, unless you want it for play back.
TI is there already with tas5614la, tas5708, it will just keep on getting better from here. No worries/discussions of MOSFET linear region.
That's it folks, a i2s and a i2c i/f that's it and some coding
I wonder when Gibson will come out with a S/PDIF i/f? on their e-guitars
Slight over sight, live music will still be analog however, why digitize the real deal, unless you want it for play back.
TI is there already with tas5614la, tas5708, it will just keep on getting better from here. No worries/discussions of MOSFET linear region.
That's it folks, a i2s and a i2c i/f that's it and some coding
I wonder when Gibson will come out with a S/PDIF i/f? on their e-guitars
Last edited:
Bob, I seem to remember an AES preprint in really Jurassic times that analyzed oscillator bias along exactly these lines .. 1970's or even earlier.
Hi Esperado,
Class D audio amplifiers are indeed sampled-data systems, and you are not alone in not realizing this is the case.
Consider the simple PWM modulator. The audio signal is put in on one side of a comparator and a sawtooth ramp is put in on the other side. The frequency of the sawtooth is the PWM carrier frequency. The ONLY time the incoming audio signal has any influence on the output and workings of the modulator is when the comparator makes a decision whether to have its output be high or low. This is at the time when the amplitude of the sawtooth is the same as the amplitude of the input signal.
Thus, what the input signal is doing between those times is not looked at by the system. This is thus a sampled data system, albeit one that many are not used to.
Cheers,
Bob
To transport the signal, with no added noise (hum etc...) and for trivial reasons ?
Digital mixing desks are more powerfull in regard to sound processing, more compact and lighter. Easier to automatize too. Reason why more and more PA mixing desks are digital nowadays.
Still unable to move my mind. (let me time to think deeper at your input, Bob.)
I don't agree for the moment: A class D amp is a servo system, with an analog feedback. So, "looked at by the system" (always from my point of view;-). More than that, there is an infinite number of different switching times to reflect the signal.
For me digital is different by nature: we make steps measurements (of levels or differences of levels), at the same regular intervals, and we transport (and process) this abstract (finite) data. The digital data can be anything, even numbers written by hand on a piece of paper.
This interest-me a lot, as it is the first time i have the feeling to think (or feel) different from you about an electronic assembly.
Last edited:
Ok, Bob, i see your point.
You are looking to class D amps from the inside (there is only one measurement during the pulse) and i'm looking at it from outside of the loop (infinite analog numbers, analog feedback).
I have moved one horizontal step in your direction ;-)
Can-you move one vertical step in my direction, so we can drink together ?
BTW: i agree now with your "Sampled Datas" definition, witch says nothing about the nature of the datas.
You are looking to class D amps from the inside (there is only one measurement during the pulse) and i'm looking at it from outside of the loop (infinite analog numbers, analog feedback).
I have moved one horizontal step in your direction ;-)
Can-you move one vertical step in my direction, so we can drink together ?
BTW: i agree now with your "Sampled Datas" definition, witch says nothing about the nature of the datas.
Last edited:
On the topic of the thread, if not the current discussion: Doing my first scratch design with the philosophy of fix the biggest problem first, I wound up with a terribly conventional amp that could have been taken out of the book. Almost. ( lateral MOSFET drivers) A few things I noticed we could use more detail on, or at least what left me wanting:
A subject I don't seem to have seen covered in either this or Bob's book are some of the SOP values for things like the driver stage standing current and it's effects or the ratio between the bias servo transistor or the bias biasing resistors and their effect on servo "gain". Great to do all the thermal resistive calculations, but how much do I have to overbias for what ration to prevent me from runaway and not get underbiased. Is this really a prototype lab question that can't be really explored?
Some of the protection methods are discussed, but kind of limited to the output stage. I have seen so many more in designs, but the subtleties of each are hard to understand. I also, by simulation, see most of these things to terribly ugly things on limiting. Maybe a more forceful message from these two authors on " obviously big enough" would be helpful to us beginners.
I would like to see some discussion of clipping and recovery behavior in more detail and what can be done to make it happier. Can you actually do anything to improve driver or output rail sticking other than not clipping? I noticed in Spice at least, behavior at clipping was improved when I separated the references for the IPS and VAS stage ccs. Ugly spikes not quite as ugly. Again, topology linked to objective measurement that stands a chance of being relevant to subjective hearing.
Both books go through the same basic discussion of SOA, but I found when I pulled up data sheets, I am first faced with log plots not linear as the chapters discuss. It is all fine and dandy to draw several loci on a graph, but no help if we are left without the starting graph. At least I can draw an old school load line for a tube from what the good Mr. Jones taught me.
Maybe it is to much "audiophile" ( dare I use that word in Dr. Self's presence?) but some trade-offs of tangible things could have been discussed. One place I had too much interest is in the behavior of a ccs. Some are far stiffer, some far more linear remaining at a more consistent impedance far higher in frequency. Real parameters we can model. Nothing about philosophy of how they sound or their measurable effects on distortion. Is a simple JFET/Diode because it remains consistent to a much higher frequency going to sound better than a CFP that is far stiffer, but falls off much earlier? Or are they all really more than good enough so pick your poison? ( yea, .00001 or .00002, who cares?)
Based on my wife's super critical hearing, correlating to many commercial amps over the years, I wonder if Douglas or Bob put enough emphasis on decreasing higher order harmonics and what evil things they do to less that stellar tweeters. This is the objective parameter I can correlate to her pure subjective hearing. Amps whose harmonics do not fall of rapidly uniformly sound worse than amps who do. ( the 5th and 7th are at the same level as the third, etc)
Oh, Bob, on page 396 you hint at the conversions from noise density to SNR, but not quite a big enough hint for someone trained as a tech, not an engineer. Nit picking, as a decent book on SPICE would be bigger that the both amp books together. Left me confused. Two book reviews for the price of one!
A subject I don't seem to have seen covered in either this or Bob's book are some of the SOP values for things like the driver stage standing current and it's effects or the ratio between the bias servo transistor or the bias biasing resistors and their effect on servo "gain". Great to do all the thermal resistive calculations, but how much do I have to overbias for what ration to prevent me from runaway and not get underbiased. Is this really a prototype lab question that can't be really explored?
Some of the protection methods are discussed, but kind of limited to the output stage. I have seen so many more in designs, but the subtleties of each are hard to understand. I also, by simulation, see most of these things to terribly ugly things on limiting. Maybe a more forceful message from these two authors on " obviously big enough" would be helpful to us beginners.
I would like to see some discussion of clipping and recovery behavior in more detail and what can be done to make it happier. Can you actually do anything to improve driver or output rail sticking other than not clipping? I noticed in Spice at least, behavior at clipping was improved when I separated the references for the IPS and VAS stage ccs. Ugly spikes not quite as ugly. Again, topology linked to objective measurement that stands a chance of being relevant to subjective hearing.
Both books go through the same basic discussion of SOA, but I found when I pulled up data sheets, I am first faced with log plots not linear as the chapters discuss. It is all fine and dandy to draw several loci on a graph, but no help if we are left without the starting graph. At least I can draw an old school load line for a tube from what the good Mr. Jones taught me.
Maybe it is to much "audiophile" ( dare I use that word in Dr. Self's presence?) but some trade-offs of tangible things could have been discussed. One place I had too much interest is in the behavior of a ccs. Some are far stiffer, some far more linear remaining at a more consistent impedance far higher in frequency. Real parameters we can model. Nothing about philosophy of how they sound or their measurable effects on distortion. Is a simple JFET/Diode because it remains consistent to a much higher frequency going to sound better than a CFP that is far stiffer, but falls off much earlier? Or are they all really more than good enough so pick your poison? ( yea, .00001 or .00002, who cares?)
Based on my wife's super critical hearing, correlating to many commercial amps over the years, I wonder if Douglas or Bob put enough emphasis on decreasing higher order harmonics and what evil things they do to less that stellar tweeters. This is the objective parameter I can correlate to her pure subjective hearing. Amps whose harmonics do not fall of rapidly uniformly sound worse than amps who do. ( the 5th and 7th are at the same level as the third, etc)
Oh, Bob, on page 396 you hint at the conversions from noise density to SNR, but not quite a big enough hint for someone trained as a tech, not an engineer. Nit picking, as a decent book on SPICE would be bigger that the both amp books together. Left me confused. Two book reviews for the price of one!
Hi tvrgeek,
Thanks for this detailed post. It is very helpful to me. I am actively working on the Second Edition of Designing Audio Power Amplifiers and this is the kind of feedback I need to make a better book with more clear coverage of the existing material while adding new material. Sometimes only a single sentence or a small added paragraph will greatly enhance the understanding of a point being made. I'll add this post to my file of things to address in the new edition.
Feel free to send me an email or PM with further detail on what might help.
Cheers,
Bob
Ok, Bob, i see your point.
You are looking to class D amps from the inside (there is only one measurement during the pulse) and i'm looking at it from outside of the loop (infinite analog numbers, analog feedback).
I have moved one horizontal step in your direction ;-)
Can-you move one vertical step in my direction, so we can drink together ?
BTW: i agree now with your "Sampled Datas" definition, witch says nothing about the nature of the datas.
First and foremost, I will always be happy to drink together!
I think that an open-loop PWM class D amplifier, as I described, is a sampled-data system. However, when we put negative feedback around such an amplifier, there would seem to be a chance that the NFB might mitigate some of the effects of aliasing that might occur.
Interestingly, on the output side, the pulse width of a PWM modulator is an analog quantity in time. However, the output of a sigma-delta class D modulator, which is pulse density modulation, is quantized in pulse density - there either is a pulse at a given time or not.
Cheers,
Bob
Some help if you can . About 2 years ago a friend updated an amp I made using soft recovery diodes . I was mildly alarmed as my experience with Schottky diodes was that I could hear a difference and didn't like it . Reading articles on snubbers for conventional diodes left me confused . If the inductance etc isn't factored in forget even trying . The wisest advice seemed to be 10 nF plus 100 R probably does more good than harm and preferable to just 10 nF which cause ripples down shifted in frequency . The amp is in Vienna and me in Oxford so I never got the chance to listen .
The two diodes being considered are FFPF15S60 STU 15A 150A surge 600V
SBYV28-200-E3 -/54 3.5A 90A surge 200 V . The amplifier is 2 x Hypex UCD 180 . I am a little bit tempted to try the 200 V type for everything ( they are for a pre amp ) . Usually I use a 35 amp conventional bridge and never give it a second thought . I was surprised to see 200 A surge is about a maximum with most devices even in the > 15 A ranges ( 50 A , 340 A surge ) . Looking at 22 000 uF 45 V I assume that is about 240 A @ 60 Hz ( 4.17 mS . i = C ( dv/dt ) . The transformer coil resistance is about 0.5R which should limit current to about 60 A . A friend did the exact calculations and came to 52A allowing for all factors . The last time I did this in detail was 40 years ago . Since then generally what worked last time has been good enough .
I would be interested to now if there is any virtue in using two bridge rectifiers when marginal on current . One per cap rather than a centre tapped +/- supply . The easy answer is don't do any of this , please not that one .
Any thoughts especially if the Hypex is adverse to either type diode .
These are soft recovery diodes . Many conflicting opinions . Some say yes they produce very little noise . Others say worse noise of all .
Seeing as this must be as interesting as class D dead-band and the like someone here must know .
The two diodes being considered are FFPF15S60 STU 15A 150A surge 600V
SBYV28-200-E3 -/54 3.5A 90A surge 200 V . The amplifier is 2 x Hypex UCD 180 . I am a little bit tempted to try the 200 V type for everything ( they are for a pre amp ) . Usually I use a 35 amp conventional bridge and never give it a second thought . I was surprised to see 200 A surge is about a maximum with most devices even in the > 15 A ranges ( 50 A , 340 A surge ) . Looking at 22 000 uF 45 V I assume that is about 240 A @ 60 Hz ( 4.17 mS . i = C ( dv/dt ) . The transformer coil resistance is about 0.5R which should limit current to about 60 A . A friend did the exact calculations and came to 52A allowing for all factors . The last time I did this in detail was 40 years ago . Since then generally what worked last time has been good enough .
I would be interested to now if there is any virtue in using two bridge rectifiers when marginal on current . One per cap rather than a centre tapped +/- supply . The easy answer is don't do any of this , please not that one .
Any thoughts especially if the Hypex is adverse to either type diode .
These are soft recovery diodes . Many conflicting opinions . Some say yes they produce very little noise . Others say worse noise of all .
Seeing as this must be as interesting as class D dead-band and the like someone here must know .
Last edited:
While I'm probably a bit late to this thread, I have some questions regarding Chapter 8 of v06 of Audio Power Amplifier Design (push-pull VAS). I'm working on a modern design to replace a completely obsolete power amplifier in a Fisher 600-T receiver. I have looked through the books by Self and Cordell as well as designs from Marshall Leach and was looking at the twin complimentary IPS and push-pull VAS as it would be symmetrical. However, Self, after analysis, eventually reverts back to a single VAS and current source.
When I simulate this (cheap and dirty Java simulator), there are issues at startup which don't appear in the P-P version, though they probably would with imperfect components. The issue that does not seem to be addressed is what happens when these circuits are first powered up. If the VAS and its current source do not come to life at the same time, there is significant DC on the output for a period of time until these circuits all stabilize. While a delayed speaker relay would solve this, most amplifiers don't have speaker relays.
I have a Harman-Kardon Citation 24 (same circuit as Citation 22) that is full complementary IPS and P-P VAS that uses a delayed muting circuit (same circuit is used for protection) that essentially shuts down the VAS until the circuits have had time to stabilize. It appears that this could be done in the Self or Cordell circuits using a transistor in parallel with the VAS current limit transistor along with an equivalent shunt transistor in the current source as long as both of these were turned off at the same instant by a single circuit.
Again, I've seen no discussion of what happens at power up with voltages rising over time. Due to capacitor tolerances, the + and - rails are not guaranteed to track during power up, adding to the startup problem.
Additionally, the Citation has eliminated the capacitor in the feedback loop and DC is zeroed by a resistor network and trimpot injecting a correction voltage to the IPS. Eliminating this capacitor is also part of my plan.
Discussion?
When I simulate this (cheap and dirty Java simulator), there are issues at startup which don't appear in the P-P version, though they probably would with imperfect components. The issue that does not seem to be addressed is what happens when these circuits are first powered up. If the VAS and its current source do not come to life at the same time, there is significant DC on the output for a period of time until these circuits all stabilize. While a delayed speaker relay would solve this, most amplifiers don't have speaker relays.
I have a Harman-Kardon Citation 24 (same circuit as Citation 22) that is full complementary IPS and P-P VAS that uses a delayed muting circuit (same circuit is used for protection) that essentially shuts down the VAS until the circuits have had time to stabilize. It appears that this could be done in the Self or Cordell circuits using a transistor in parallel with the VAS current limit transistor along with an equivalent shunt transistor in the current source as long as both of these were turned off at the same instant by a single circuit.
Again, I've seen no discussion of what happens at power up with voltages rising over time. Due to capacitor tolerances, the + and - rails are not guaranteed to track during power up, adding to the startup problem.
Additionally, the Citation has eliminated the capacitor in the feedback loop and DC is zeroed by a resistor network and trimpot injecting a correction voltage to the IPS. Eliminating this capacitor is also part of my plan.
Discussion?
Forgot this thread was in the amp forum. I started a new set of comments over in the lounge on the latest edition I have been studying.
I will summarize:
I too was interested soft recovery diodes as they model well (low leakage?), have higher C, but should not produce as much junk on switching. Problem is, models are only that. I know how a HEXFRED worked as a rectifier in a real amp, but that does not tell me how they would work as bias servo flying clamps or a Baker clamp.
I liked the additional explanations using SPICE, but I think the good Dr. trusts we are more proficient than we may be. I look forward to the next Linear Audio for some more thoughts on simulating stability with a Tian probe and TMC.
Edition is only a few months out, and already some of the devices mentioned have been EOL'd.
I much appreciated the discussions of the wonder patents from the 80's. They look good on paper, but the question remains if they work better, or if they were never kept in production as it does not matter with MP3 and won't fit in a 9 channel AVR. I was also interested in all the differential mirrored IPS and VAS concepts that seems to proliferate on these pages. My modeling could not find any advantage, but a few amps I remember fondly ( Acrus and all my Parasounds) use such topology.
Comments on WHY a specific transistor was selected for different uses was quite useful.
I agree with Fred, I have worries about power up and off that simulations won't really tell us. I don't know enough to model a transformer I may choose, so I am assuming instant Vee.
More could be mentioned about various places where additional clamping might want to be implemented and why (or why not). I see a lot out there but don't always know what the designer was thinking. I usually can't find a reason that seems probable. Just too much copy of SOP?
In all, one more helpful step.
I will summarize:
I too was interested soft recovery diodes as they model well (low leakage?), have higher C, but should not produce as much junk on switching. Problem is, models are only that. I know how a HEXFRED worked as a rectifier in a real amp, but that does not tell me how they would work as bias servo flying clamps or a Baker clamp.
I liked the additional explanations using SPICE, but I think the good Dr. trusts we are more proficient than we may be. I look forward to the next Linear Audio for some more thoughts on simulating stability with a Tian probe and TMC.
Edition is only a few months out, and already some of the devices mentioned have been EOL'd.
I much appreciated the discussions of the wonder patents from the 80's. They look good on paper, but the question remains if they work better, or if they were never kept in production as it does not matter with MP3 and won't fit in a 9 channel AVR. I was also interested in all the differential mirrored IPS and VAS concepts that seems to proliferate on these pages. My modeling could not find any advantage, but a few amps I remember fondly ( Acrus and all my Parasounds) use such topology.
Comments on WHY a specific transistor was selected for different uses was quite useful.
I agree with Fred, I have worries about power up and off that simulations won't really tell us. I don't know enough to model a transformer I may choose, so I am assuming instant Vee.
More could be mentioned about various places where additional clamping might want to be implemented and why (or why not). I see a lot out there but don't always know what the designer was thinking. I usually can't find a reason that seems probable. Just too much copy of SOP?
In all, one more helpful step.
There is a good file with comments to DS book on Samuel Groner's web:
http://www.sg-acoustics.ch/analogue_audio/power_amplifiers/pdf/audio_power_amp_design_comments.pdf
SG-Acoustics · Samuel Groner · Power Amplifiers
http://www.sg-acoustics.ch/analogue_audio/power_amplifiers/pdf/audio_power_amp_design_comments.pdf
SG-Acoustics · Samuel Groner · Power Amplifiers
This is a fact... not conjecture. I have made a good living researching this issue and on designing products to filter out this noise on the ac lines from getting into everything. The spectrum of noise is very wide.
Large screen Video/TV and computers (large box types especially) put a lot of noise directly onto the line which is hard to remove. But every digital product today uses a form of smp in them and so, yes, the power pollution and its side effects are getting worse.
Thx-RNMarsh
Last edited:
In most cases, constant Vee is fine but in the case of a power amplifier, the power supply is very much a part of the amplifier. Here is how I do it: In my case, I was able to measure the DC resistance of the transformer windings. Set up an AC source for each half of a center tapped winding, add the resistance in series to each half, then connect through rectifier diodes and filter capacitors. Model any rail decoupling RC circuits as well. That's where the startup problems become apparent.
For my Fisher 600-T example, the power transformer is probably around 250 VA, the main winding is 30-0-30 V with no load, the DC resistance is 0.31 ohms per half. It is 28-0-28 V with a 4 amp load (measurement errors cause the calculations to be inconsistent). The DC output is 39 V per rail open circuit, 33 V at 3 A, 30 V at 5 A. These numbers can be used to check the simulation and tweak as necessary. Other similar VA transformers will probably be close and this should be good enough for simulation.
For a stereo amplifier driven from a single power supply, double the resistor values and cut the filter capacitors in half.
For low frequency testing, use 19 Hz or 21 Hz, NOT 20 Hz. You don't want the input signal to always be in phase with the 120 Hz ripple frequency (or 100 Hz on the other side of the pond) because it might mask some problems.
Andrew (or anyone else), got evidence of this?
I'm with you on mains getting worse but I think PFC helps make things better .. maybe even better than conventional rectifier cap supplies.
Mr. Cordell?
Hi kgrlee,
The next edition of my book will have a whole chapter devoted to SMPS for power amplifiers, including in-depth discussion of power factor correction (PFC). I have been working on that subject for some time now, including as part of another chapter that I am writing as part of a multi-author Handbook.
A key thing to bear in mind is that conventional SMPS starts with a rectifier-capacitor right on the mains to create DC of about 340V for a subsequent switching DC-DC converter. So this approach has all of the usual rectifier spike nasties of a conventional supply, but right on the mains side, not isolated by a 60Hz transformer whose poor HF response might actually mitigate these nasties a bit. At the same time, there is opportunity to do some HF filtering on the 340V intermediate DC bus to isolate the switching DC-DC converter noise from flowing back into the mains. Often, not much is done in this regard, unfortunately.
In a PFC SMPS, the front-end at the mains is different. The PFC SMPS starts with a boost converter supply. It creates a DC intermediate bus voltage usually on the order of 385Vdc. The mains goes into a full-wave bridge first, but that bridge does not have a filter capacitor. Instead, the full-wave-rectified output (called a haversine) feeds the boost converter. The boost converter is a dynamic boost converter that takes any instantaneous voltage on the haversine waveform and boosts it up to the 385Vdc value - the boost ratio is constantly changing with time. The average current pulled from the line by the boost converter is controlled by feedback in real time to have a current waveform that emulates that of the current of a resistor - thus the near-unity power factor.
The PFC thus pretty much eliminates the line-frequency impulsive rectifier current nasties of a conventional supply, but does have the potential to have a switching converter topologically closer to the mains. So for HF switching noise, the POTENTIAL exists in principle for it to be worse. The extent to which PFC vs no PFC SMPS differ in HF EMI injection onto the main really is a strong function of the design, however.
Overall, my view is that PFC SMPS have the potential to be far superior in lower emissions if they are conscientiously designed. Virtually all PC power supplies incorporate PFC. The PFC switching controller function is fairly complex, but it is implemented in inexpensive ICs.
A great many pro-sound amplifiers incorporate PFC not just to be good neighbors in the emissions sense (or to satisfy international governmental regulations), but also because PFC allows them to extract more power from a given mains source with a given amount of impedance, sag and waveform distortion on it.
Cheers,
Bob
Bob, I think PFC goes a long way to getting V and I in phase. But, the problem we still have is the wide band hash. IEC specs set a minimum spec and not a 'lets kill the noise' standard unfortunately.
I have to confess, despite being a marketer in a company that's involved in shipping upwards of a quarter of a billion SMPS controllers a year, when it comes to audio I am a conventionalist who believes in iron and big caps.
I have to confess, despite being a marketer in a company that's involved in shipping upwards of a quarter of a billion SMPS controllers a year, when it comes to audio I am a conventionalist who believes in iron and big caps.
Yes, I pointed out above that PFC does not necessarily do anything for the HF wideband hash. Perhaps I did not make myself clear.
BTW, the boost converter topology starts off with an inductor that is followed by a MOSFET switch to ground. As a result, the switching action is a bit isolated by this series inductor that is fundamental to the operation of the boost converter. The quality and self-capacitance of this inductor can make a significant difference in the amount of emissions conducted back to the mains. Of course all of the SMPS is preceded by an EMI filter whose quality really matters. As you know, sometimes less expensive EMI filters are incorporated right into the IEC mains input connector.
Cheers,
Bob
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.