Hi, I have a basic newbie question.
A bit of background first. I have never designed an amplifier. I have read books on audio amplifier design by Randy Slone, Doug Self and Bob Cordell, and done some DIY circuit building, designed some PCBs from existing schematics.
Now for my question. 25, 30 years ago, Randy Slone and Doug Self used to design amplifiers whose THD went from, say, 0.004% at 1KHz to perhaps 0.05% at 20KHz. These were considered top of their lines. I'm focusing only on SS Class B here.
Now I am seeing discrete Class B amps routinely getting into triple-zero THD territory. The DIYaudio Wolverine. The Benchmark AHB2. The Topping B100/B200. And others. How are they able to do this?
I am aware of composite amplifiers which use a precision opamp in the feedback loop of a larger Class B amp. Neurochrome and Kaltecs.com do this. But both of these designers have done it only with chip amps. The Wolverine does not use an opamp in the feedback path. And I am told that it's damn hard to design a composite amp which does not oscillate, so I'm assuming that this approach is usable with a lot of effort with chip amps but unviable with a classic three-stage discrete design. I have heard the term TMC. What is it? I am aware of Miller compensation capacitors from Randy Slone's books, and he hesitated to use a CRC pi filter for stability concerns, he'd usually use just a single capacitor. Is the TMC I hear about fundamentally different from the stuff he explained? I am also aware that regulated supply rails for the input and VAS stages of an amp may deliver some improvements in performance, though I haven't seen it being used by Slone or Self, no idea why. Does this help?
So my question is basically: what are all the techniques modern designers are using to get to triple-zero territory? They are using parts which do not seem to be very exotic or unique, nor are their designs so narrow that we get a very unique set of circumstances which give us the magic. What changed in 20 years? Is this achieved by obsessing over details (like using an optical connection from front panel power switch to the rear, instead of an electrical cable, to keep mains hum out, or like feeding the supply rails of the actual amp with the output of a regulated massive power supply, like I'm told Halcro does), or by advances in topology and semiconductor parts quality?
(I am aware that my question is more an expression of my ignorance and misunderstanding than a simple search for some facts. I am hoping for some patient answers from some of you which will complete my education and clear my misunderstandings.)
My question is only about Class B SS amps.
A bit of background first. I have never designed an amplifier. I have read books on audio amplifier design by Randy Slone, Doug Self and Bob Cordell, and done some DIY circuit building, designed some PCBs from existing schematics.
Now for my question. 25, 30 years ago, Randy Slone and Doug Self used to design amplifiers whose THD went from, say, 0.004% at 1KHz to perhaps 0.05% at 20KHz. These were considered top of their lines. I'm focusing only on SS Class B here.
Now I am seeing discrete Class B amps routinely getting into triple-zero THD territory. The DIYaudio Wolverine. The Benchmark AHB2. The Topping B100/B200. And others. How are they able to do this?
I am aware of composite amplifiers which use a precision opamp in the feedback loop of a larger Class B amp. Neurochrome and Kaltecs.com do this. But both of these designers have done it only with chip amps. The Wolverine does not use an opamp in the feedback path. And I am told that it's damn hard to design a composite amp which does not oscillate, so I'm assuming that this approach is usable with a lot of effort with chip amps but unviable with a classic three-stage discrete design. I have heard the term TMC. What is it? I am aware of Miller compensation capacitors from Randy Slone's books, and he hesitated to use a CRC pi filter for stability concerns, he'd usually use just a single capacitor. Is the TMC I hear about fundamentally different from the stuff he explained? I am also aware that regulated supply rails for the input and VAS stages of an amp may deliver some improvements in performance, though I haven't seen it being used by Slone or Self, no idea why. Does this help?
So my question is basically: what are all the techniques modern designers are using to get to triple-zero territory? They are using parts which do not seem to be very exotic or unique, nor are their designs so narrow that we get a very unique set of circumstances which give us the magic. What changed in 20 years? Is this achieved by obsessing over details (like using an optical connection from front panel power switch to the rear, instead of an electrical cable, to keep mains hum out, or like feeding the supply rails of the actual amp with the output of a regulated massive power supply, like I'm told Halcro does), or by advances in topology and semiconductor parts quality?
(I am aware that my question is more an expression of my ignorance and misunderstanding than a simple search for some facts. I am hoping for some patient answers from some of you which will complete my education and clear my misunderstandings.)
My question is only about Class B SS amps.
Read their product literature and marketing materials. In amongst the fluff you may find a nugget or two of actual technical information.
For example, Topping's B200 web page proudly proclaims that it uses something called NFCA . Scour their material and find out what they mean by NFCA, and what its likely technical impact might be.
Benchmark's product literature discusses their use of feedforward circuitry and patents are mentioned. Dig into this and do some research.
The Wolverine's full and complete schematic is posted here on these diyAudio Forums! It is not hiding any secrets from you! Study the schematic and determine whether it contains new techniques not found in your 20 year old textbooks. For example, look at jumper "J1" on the Wolverine schematic, which selects between "TMC" and "TPC" . . . . you'll want to know what those actually mean.
For example, Topping's B200 web page proudly proclaims that it uses something called NFCA . Scour their material and find out what they mean by NFCA, and what its likely technical impact might be.
Benchmark's product literature discusses their use of feedforward circuitry and patents are mentioned. Dig into this and do some research.
The Wolverine's full and complete schematic is posted here on these diyAudio Forums! It is not hiding any secrets from you! Study the schematic and determine whether it contains new techniques not found in your 20 year old textbooks. For example, look at jumper "J1" on the Wolverine schematic, which selects between "TMC" and "TPC" . . . . you'll want to know what those actually mean.
Power transistors went from Ft=3MHz to Ft=30MHz. That enables more negative feedback.
Thermal control has gotten better. Emitter resistors can be made smaller, resulting in lower distortion.
Circuit simulation has enabled frequency compensation methods beyond one-pole.
BTW, 0.05% THD at 20KHz is better than most solid-state amplifiers that Stereophile measures. Most manufacturers realize that the THD wars have gotten silly.
Ed
Thermal control has gotten better. Emitter resistors can be made smaller, resulting in lower distortion.
Circuit simulation has enabled frequency compensation methods beyond one-pole.
BTW, 0.05% THD at 20KHz is better than most solid-state amplifiers that Stereophile measures. Most manufacturers realize that the THD wars have gotten silly.
Ed
Thanks. I too have been seeing the faster discrete devices, lower capacitance (Cob?), didn't know about emittor resistors.
Have circuit simulators and device models improved? Self and Slone both used simulation extensively, but device models were a subject of discussions, I've witnessed some of those here.
And about Stereophile, 0.05% at 20KHz, etc, yes I've seen this for a long time. In fact, I've seen how some audiophiles prefer coloured sounding amplifiers and get upset with the antiseptic cleanness of Self's recent preamp's sound. Therefore amps selling at prices where a handful of extra components would not make the slightest dent on their gross margins will still report 0.2% THD, where by a slight increase in circuit complexity, they could have brought it down by an order of magnitude. That's a separate path to follow, I'm not curious why they do it. I'm a devotee of ASR.
Have circuit simulators and device models improved? Self and Slone both used simulation extensively, but device models were a subject of discussions, I've witnessed some of those here.
And about Stereophile, 0.05% at 20KHz, etc, yes I've seen this for a long time. In fact, I've seen how some audiophiles prefer coloured sounding amplifiers and get upset with the antiseptic cleanness of Self's recent preamp's sound. Therefore amps selling at prices where a handful of extra components would not make the slightest dent on their gross margins will still report 0.2% THD, where by a slight increase in circuit complexity, they could have brought it down by an order of magnitude. That's a separate path to follow, I'm not curious why they do it. I'm a devotee of ASR.
TMC/TPC improve THD and IMD substancialy but it can be at the expense of stability if used too agressively,
wich is the case for most amps using this compensation wich are stable on capacitive loads only thanks to an output inductance.
For the rest 0.05% THD at 20kHz is meaningless since the first residual is at 40kHz, in old mags they used 10kHz instead.
Agree on the whole but methink that the novelty in THD matter of the 2SA1302/2SC3281 was mainly the sustained gain at high currents, wich indeed yield more GNFB at high power.
I thought that Self threw everything he discussed into his Blameless design? And that design's THD is an order of magnitude away from triple-zero?
My question is actually about the THD gap between the Blameless (and its derivatives designed by Randy Slone) and the modern designs which are hitting triple-zero THD.
I personally think it's okay to insist on the inductance -- do whatever it takes but make the amp stable in 99% situations.
I understand that the harmonics of 20KHz may be well outside the audible range, but I am beginning to believe that there is a correlation between the HD-vs-frequency graph and the multi-tone IMD performance. And IMHO, multi-tone IMD differences are audible from amp to amp than THD differences. Just an opinion, not proven. (To prove it, I'd need a set of amps with the same THD profile but with different multi-tone IMD profiles, which is hard to put together.)
For your information, there are two people on this forum who organized double-blind listening tests on amplifiers. In their experience, the only things that are sometimes bad enough to be audible are frequency response errors with an actual loudspeaker load and clipping behaviour, if the amplifier is driven into clipping. Their observations are hidden somewhere in this thread: https://www.diyaudio.com/community/...overall-amplifier-quality.407222/post-7551743
Take a look at the impressive results achieved
with just a few components :
https://www.angelfire.com/ab3/mjramp/index.html
https://www.angelfire.com/ab3/mjramp/mjr7-mk5.html
with just a few components :
https://www.angelfire.com/ab3/mjramp/index.html
https://www.angelfire.com/ab3/mjramp/mjr7-mk5.html
1. 0.05% thd at 20k has different means, at full power or 1W. Usually, at 1W the amp is still in class A region. It is common to reach 0.005% at 20k in class a. I found 5W is more useful figure. It is big enough to get out of class a region and it is most useful power band for home use.
2. To improve thd, use the best output device, lateral mosfet. They are designed for audio. They are just the best for the task.
3. Get as much as negative feedback that you can get. TMC(transitional miller compensation) is the most effective way to add NFB. Another thing, conventionally, people design amplifiers with Unit Loop Gain bandwidth less than 1MHz. It makes sense for slow devices such as 2n3055, but it is too conservative for modern output devices that with 30MHz ft or more. You could bump it to around 2MHz with minimal drawback.
4. At some point, background noise/hum is more important than THD. Learn how to layout the ground.
2. To improve thd, use the best output device, lateral mosfet. They are designed for audio. They are just the best for the task.
3. Get as much as negative feedback that you can get. TMC(transitional miller compensation) is the most effective way to add NFB. Another thing, conventionally, people design amplifiers with Unit Loop Gain bandwidth less than 1MHz. It makes sense for slow devices such as 2n3055, but it is too conservative for modern output devices that with 30MHz ft or more. You could bump it to around 2MHz with minimal drawback.
4. At some point, background noise/hum is more important than THD. Learn how to layout the ground.
I didn't find any measurements of HD
If I understand Slone correctly, good low distiortion amps with BJT OPS devices are never in Class A... they are always in Class B, even with very low power. If they have a period of overlap between the upper rail OPS and lower rail OPS devices, this gives rise to higher distortion. With MOSFET OPS devices, this overlap is possible without distortion increase, so we can call them Class AB amps -- they can be over-biased.
TS talks about a triple-zero THD. So, that's not about listening but about objective measuring.
Such low THD is not only about schematic design but also about the design of the whole thing. I mean that's about a war of signal and noise (or interference), so it's the physical placement of ALL wires too. That's why I would recommend reading some great Bibles such as Henry W. Ott's "Noise Reduction Techniques in Electronic Systems" and similar books, like, High-Speed Signal Propagation: Advanced Black Magic by Howard Johnson and Martin Graham.
Last edited:
Regarding layout, the currents through the output devices in a class-(A)B amplifier are by definition grossly distorted. When they couple magnetically or via common impedances into the signal path, that can affect distortion a lot, particularly when they couple to something that doesn't get corrected by feedback. For example, magnetic coupling to the input or output wiring or the output inductor.
See Edward Cherry, "A new distortion mechanism in class-B amplifiers",
https://linearaudio.net/sites/linearaudio.net/files/Cherry mutual inductance distortion_0.pdf
Last edited:
The first thing you have to do is look at the output spectrum of an amplifier for THD and IMD. That tells the story, not a number. I guess the THD or IMD numbers merely tell you if you should look deeper. Yes, the lower numbers are worth something and manufacturers ought to chase them. Again, not just the number but the performance across the spectrum.
To answer a couple other questions. Simulators have become better, as have models. They will still give you a rosy picture of performance. You need to include component variances and make certain your models are good. PCB layout can change everything as affordable simulators do not model that (and that means you need to design the PCB).
Parts have improved as well as our understanding of what is important.
I recently measured the performance of a very expensive home amplifier that didn't sound very good. It didn't measure very well either (and this company does not disclose service information). The correct measurements do in fact correlate with how the electronics sound.
The Blameless amplifier. A paid, top designer will not give away his best work. He simply demonstrated a basic design without mistakes.
To answer a couple other questions. Simulators have become better, as have models. They will still give you a rosy picture of performance. You need to include component variances and make certain your models are good. PCB layout can change everything as affordable simulators do not model that (and that means you need to design the PCB).
Parts have improved as well as our understanding of what is important.
I recently measured the performance of a very expensive home amplifier that didn't sound very good. It didn't measure very well either (and this company does not disclose service information). The correct measurements do in fact correlate with how the electronics sound.
The Blameless amplifier. A paid, top designer will not give away his best work. He simply demonstrated a basic design without mistakes.
Are you accusing me of listening and of polishing wires now? I don't believe in audible distortion differences between well-designed amplifiers, but judging by post #8, the thread starter does (for multitone IMD anyway).
First, in class A, no matter BJT or MOSFET, you get lower distortion. For example, 0.005% 20KHz at 1W in Class A vs 0.05% 20KHz at 5W from the same amp.
Yes, BJT has an optimal bias point for Class B region. I remember usually below 100mA. Even though, the THD in class B is always worse than that in class A region.
Actually, you can have both at the same time. You could parallel 2 or 3 pairs output devices with optimal bias each. The ideal is to get a clean first watt in class A.
Regarding MOSFET, they have wider shut off transition region than BJT. Thus, the transition from class A to class B is less disruptive. Here, I am talking about "Lateral MOSFET", such as 2SK1058.
Don't use power switch type MOSFET like IRFP240. They are no better than some widely available BJTs like MJL3281 or 2SC5200.
Gee jxdking,
My Symasym amplifiers worked best with approx 5 mA bias current. Parts were matched. Most Marantz and others were 20 mA, 30 mA on the high side. Other designs like CFP ran hotter slightly, but there was no reason to go beyond 50 mA (and I question that high bias level) unless the design wasn't very good. Mosfets need higher standing current to control bias more than anything else. They don't current share until they hit the negative tempco.
As for multiple output designs, they work better with outputs matched. At higher currents the emitter resistors determine current sharing with good devices. This is decades of empirical measurements and auditions.
Most Mosfets that aren't lateral suffer from a gate charge issue. I'll take bipolar output sections over Mosfets any day, serviced all of them for decades.
My Symasym amplifiers worked best with approx 5 mA bias current. Parts were matched. Most Marantz and others were 20 mA, 30 mA on the high side. Other designs like CFP ran hotter slightly, but there was no reason to go beyond 50 mA (and I question that high bias level) unless the design wasn't very good. Mosfets need higher standing current to control bias more than anything else. They don't current share until they hit the negative tempco.
As for multiple output designs, they work better with outputs matched. At higher currents the emitter resistors determine current sharing with good devices. This is decades of empirical measurements and auditions.
Most Mosfets that aren't lateral suffer from a gate charge issue. I'll take bipolar output sections over Mosfets any day, serviced all of them for decades.